978 3 662 03750 8

Studies in Economic Theory
Editors
Charalambos D. Aliprantis
Purdue University
Department of Economics
West Lafayette, IN 47907-1310
USA
Nicholas C. Yannelis
University of Illinois
Champaign, IL 61820
USA
Springer-Verlag Berlin Heidelberg GmbH

Titles in the Series
M. Ali Khan and Nicholas C. Yannelis (Eds.)

Equilibrium Theory in Infinite Dimensional Spaces
Charalambos D. Aliprantis, Kim C. Border

and Wilhelmus A.J. Luxemburg (Eds.)
Positive Operators, Riesz Spaces,
and Economics
Donald G. Saari
Geometry of Voting
Charalambos D. Aliprantis and Kim C. Border

Infinite Dimensional Analysis
Jean-Pierre Aubin
Dynamic Economic Theory
Mordecai Kurz (Ed.)

Endogenous Economic Fluctuations
Jean-Fran~ois Laslier
Tournament Solutions and Majority Voting
Ahmet Alkan
Charalambos D. Aliprantis
Nicholas C. Yannelis
Editors
Current Trends
in Economics
Theory and Applications
Proceedings of the Third International Meeting

of the Society for the Advancement of Economic Theory,
Antalya, Turkey, June 1997
With 21 Figures
and 55 Tables
i Springer
Prof. Ahmet Alkan
Bogazici University
Department of Management
Bebek 80815
Istanbul
Turkey
Prof. Charalambos D. Aliprantis

Purdue University
West Lafayette, IN 47907-1310
USA
Prof. Nicholas C. Yannelis

University of Illinois at Urbana Champaign
Champaign, IL 61820-6271
USA
Library of Congress Cataloging-in-Publication Data

Die Deutsche Bibliothek - CIP-Einheitsaufnahme
Current trends in economics: theory and applications; Antalya, Turkey,
June 1997; with 55 tables / Ahmet Alkan ... ed. - Berlin; Heidelberg;
New York; Barcelona; Hong Kong; London; Milan; Paris; Singapore; Tokyo:
Springer, 1999
(Proceedings of the ... international meeting of the Society for the Ad-
vancement of Economic Theory ...; 3) (Studies in economic theory; 8)
ISBN 978-3-642-08471-3 ISBN 978-3-662-03750-8 (eBook)
DOI 10.1007/978-3-662-03750-8
This work is subject to copyright. All rights are reserved, whether the
whole or part of the material is concerned, specifically the rights of
translation, reprinting, reuse of illustrations, recitation, broadcasting, re-
production on microfilm or in any other way, and storage in data banks.
Duplication of this publication or parts thereof is permitted only under
the provisions of the German Copyright Law of September 9, 1965, in
its current version, and permission for use must always be obtained
from Springer-Veriag. Violations are liable for prosecution under the
German Copyright Law.
© Springer-Verlag Berlin Heidelberg)999
Originally published by Springer-Verlag Berlin Heidelberg New York in 1999.
Softcover reprint of the hardcover 1st edition 1999
The use of general descriptive names, registered names, trademarks, etc.
in this publication does not imply, even in the absence of a specific
statement, that such names are exempt from the relevant protective laws
and regulations and therefore free for general use.
Hardcover design: Erich Kirchner, Heidelberg
SPIN 10655629 4212202-5 4 3 2 1 0 - Printed on acid-free paper
Foreword
In 1990, the Society for the Advancement of Economic Theory

(SAET) was founded with the main purpose to advance our knowledge in
theoretical economics and to facilitate communication among researchers
in economics, mathematics, game theory and any other field which is po-
tentially useful to economic theory. To achieve these goals, SAET sponsors
the research journal Economic Theory published by Springer-Verlag and
holds international conferences every other year. The first two conferences
of SAET took place in the island of Cephalonia, Greece, in the summers of
1993 and 1995. In the summer of 1997, the conference was held in Antalya,
Turkey.
The twenty-nine papers in this volume are mostly by participants in the
Antalya meeting of SAET and form a broad sample of the 150 papers pre-
sented there. Topics covered include cooperative and noncooperative games,
social choice and welfare, bargaining, matchings, auctions, mechanism de-
sign, general equilibrium, general equilibrium with finance, industrial or-
ganization, macroeconomics, and experimental economics. We have chosen
to present the papers according to the alphabetical order of first author
names instead of grouping them by topic or theme. We have appended a
complete listing of the sessions in the conference together with a list of
program committee members and of sponsors at the end of the volume.
We thank all the authors and the participants of the Antalya meeting
for making this volume possible. We fully express our gratitude to the
referees for their help in the selection process and our indebtedness to
Sumru Altug, Kemal Badur, Mehmet Barlo as well as to Qaglar Erol, Kenan
Tata, Ali Gungoraydinoglu, Serkan Albayrak for their painstaking efforts
in the compilation and preparation of this volume.
A. Alkan, C. D. Aliprantis and N. C. Yannelis

October 1998
Contents
Page
Foreword ............................................................. v
Pseudo-supermodular games: theory and an application ................ 1

Elettra Agliardi
A degree of manipulability of known social choice procedures ......... 13

Fitad Aleskerov and Eldeniz K urbanov
On the properties of stable many-to-many matchings under responsive

preferences ........................................................... 29
Ahmet Alkan
Cost uncertainty, taxation, and irreversible investment ................ 41

Sumru Altug, Fanny S. Demers and Michel Demers
The significance of the market portfolio: theory and evidence ......... 73

Stefano Athanasoulis and Robert J. Shiller
Does money matter? A deterministic model with cash-in-advance

constraints in factor markets ......................................... 107
Erdem BG.§91, and Ismail Saglam
Welcoming the middlemen: restricting competition in auctions by

excluding the consumers ............................................. 119
Subir Bose and George Deltas
Acyclicity of fuzzy preferences ....................................... 133
Manabendra Dasgupta and Rajat Deb
On the connection between correlated equilibria and sunspot

equilibria ........................................................... 141
Julio Ddvila
Ex post individually rational trading mechanisms .................... 157

Fra~oise Forges
Allocation among multi-member households: issues, cores, and

equilibria ........................................................... 177
Hans Gersbach and Hans Halter
Valuation equilibrium revisited ...................................... 201

Peter Hammond and Antonio Villar
viii
The equal-distance rule in allocation problems with single-peaked

preferences .......................................................... 215
Carmen Herrero and Antonio Villar
The existence of the satisfactory point ............................... 225
Adam Idzik
Optimal entry and the marginal contribution of a player ............. 233

K unio Kawamata
On an Edgeworth characterization of rational expectations equilihria
in atomless asset market economies .................................. 255
Leonidas C. Koutsougeras
A characterization of the equal income market equilibrium choice
correspondence based on average envy freeness ...................... 265
Bombed Lahiri
Non-keynesian effects of fiscal contractions: theory and applications

for Germany ........................................................ 271
Bernd Lucke
Incentives and risk sharing in a stock market equilibrium ............ 293
Michael Magill and Martine Quinzii
Mergers and acquisitions, bargaining, and synergy traps ............. 325
Maria-Angels Oliva and Luis A. Rivera-Batiz
Protectionism versus non-protectionism under cost uncertainty in
Cournot and Stackelberg markets ................................... 347
Benan Zeki Orbay
The canonical extensive form of a game form: symmetries ........... 367
Bezalel Peleg, Joachim Rosenmiiller and Peter Sudh6iter
Bargaining in a changing environment ............................... 389
Clam Ponsat{ and J6zsef Sakovics
An extension of the nonatomic assignment model .................... 405

Domiswamy Ramachandran and L. Riischendorf
Testing decision rules for multiattrihute decision making ............. 413

Christian Beidl and Stefan Traub
Collusion and distribution of profits under differential information .,. 455

Konstantinos Serfes and Nicholas C. Yannelis
ix
Predicting proposal configurations in cooperative games and

exchange economies ................................................. .475
Anton Stefanescu
International financial equilibrium with risk sharing and private
information .......................................................... 491
Barl Taub
Remark on extended price equilibria ................................. 519
Rabee Tourky
Program of the Ill. International Conference of the Society for the
Advancement of Economic Theory .................................. 533
Pseudo-supermodular games: theory and an
application
Elettra Agliardi
Department of Economics, University of Bologna, 40126 Bologna, Italy
Abstract. In this paper we introduce the notion of pseudo-supermodular

games and analyze the existence and order structure of equilibria using
lattice-theoretical methods. Moreover, we provide, as an application, an
analysis of equilibrium solutions for evolutionary games with monotonic
selection dynamics, identifying the implications for pseudo-supermodular
games.
JEL Classification Numbers: C60, C70, C73

Keywords: Supermodularity, evolutionary games, lattice
1 Introduction
In this paper we consider games in which the set of feasible joint decisions is
a lattice and exploit order and monotonicity properties of the games using
lattice-theoretical methods to analyze the existence and order structure of
equilibria. The class of games in which lattice-theoretical methods are most
powerful are supermodular games. They have been first introduced by Top-
kis (1979) and further analyzed in a number of works including Lippman,
Mamer and McCardle (1987), Vives (1990), Milgrom and Roberts (1990,
1991, 1994a, 1994b), Milgrom and Shannon (1994), DeGraba (1995). Super-
modular games are characterized by the fact that each player's strategy set
is partially ordered, the marginal returns to increasing one's strategy rise
with increases in the competitors' strategies and, in the case of multidimen-
sional strategies, the marginal returns to anyone component of the player's
strategy rise with increases in the other components. Broadly speaking,
these games exhibit "strategic complementarities" which yield monotone
increasing best replies; as a result, many of most important economic ap-
plications of noncooperative game theory are encompassed by this class
of games (see, for example, Bulow, Geanakoplos and Klemperer (1985),
Cooper and John (1988), Dybvig and Spatt (1983), Farrell and Saloner
(1985), Katz and Shapiro (1986), Agliardi and Bebbington (1994,1997)).
In this paper we propose a generalization of the concept of supermodular
games and introduce the concept of pseudo-supermodular games. In Section
2
2 we introduce the relevant mathematical preliminaries and present the

basic theory of pseudo-supermodular games, including the main proposition
on the existence of equilibria. Moreover, we show that the condition of
pseudo-supermodularity is weaker than supermodularity and alternative
to Milgrom and Shannon (1994) notion of quasi-supermodularity.
Our analysis of pseudo-supermodular games also unearths a further re-
sult of independent interest. In Section 3 we provide, as an application,
an analysis of evolutionary games with monotonic selection dynamics and
identify the implications for pseudo-supermodular games. Our finding is a
contribution in the study of the connections between evolutionary dynamics
and Nash equilibrium analysis.
For general k x k symmetric games there are already some results con-
cerning such connection. Bomze (1986) and Weibull (1995) show that any
population state which is stable in the replicator dynamics corresponds to
a symmetric Nash equilibrium and a strategy which is in Nash equilibrium
with itself is stationary in this dynamics. Put in another way, such results
assert that a stable outcome under the replicator dynamics must be a Nash
equilibrium, producing a "stability implies Nash" theorem in evolutionary
game theory. Nachbar (1990) extends the analysis beyond the replicator dy-
namics to show that if a dynamic process satisfies a monotonicity condition
then a limiting outcome of a converging process must be a Nash equilibrium.
Notice, however, that all these results, which hold for symmetric games, re-
quire convergence, but there is nothing inherent in the evolutionary process
which guarantees its convergence. On the other side, Samuelson and Zhang
(1991) provide a result which holds for asymmetric games too, and regard-
less of whether the process converges, but it does not correspond to a Nash
equilibrium; indeed, they show that if the evolutionary process satisfies a
monotonicity condition, then any strategy which fails to survive the iter-
ated elimination of pure strategies which are strictly dominated by other
pure strategies will be eliminated from the population. Consequently, even
if the evolutionary selection process fails to converge, in the long run virtu-
ally no individual will behave irrationally in the sense of playing strategies
which are never best replies, but coordination on Nash equilibrium may be
lost along non-convergent evolutionary paths. Our finding in Section 3 is
that for a large class of games which contain a variety of plausible social
evolutionary processes and economically meaningful situations rationality
and coordination on Nash equilibrium are not lost although without con-
vergence. Indeed, we show that if the game is pseudo-supermodular, then
we obtain Nash equilibrium solutions although without another "stability
implies Nash" theorem. For those who are troubled by the strength of as-
sumptions such as the common knowledge of rationality or the consistency
of beliefs this may be an appealing alternative.
3
2 Monotonicity of optimal solutions and

pseudo-supermodular games
Let L be a partially ordered set, that is, a set with a partial order ~ that
is reflexive, antisymmetric and transitive. For x and y elements of L, let
x V y denote the least upper bound, or "join", of x and y in L, if it exists,
and let x /\ y denote the greatest lower bound, or "meet", of x and y in L,
if it exists. A lattice is a partially ordered set L any two of whose elements
have a meet x /\ y and a join x V y in the set. A lattice (L, ~) is complete
if every non-empty subset of L has a meet and a join in L. IT a subset
8 of L has the property that x E 8 and Y E 8 imply that x V y E 8
and x /\ Y E 8, then 8 is a sublattice of 1. In many applications, L will
be Rn with the component-wise order. IT x and y are real numbers, then
x V y = max{ x, y} and x /\ Y = min{ x, y}. If x and y are n-vectors, then
xVy = (Xl VYl, .... xn VYn) and x /\ Y = (Xl /\ Yl, .... xn /\ Yn). For a lattice L
with the given relation ~, with 8 and 8' subsets of L, we say that 8 ~ 8'
if, for every x E 8 and x' E 8', x /\ x' E 8 and x V x' E 8'.
Consider lex) as a real-valued function on a lattice 1. We define pseudo-
supermodularity in the following way:
Definition. lex) is pseudo-supermodular on L if (i) I (x)V !(x') ~ I(x/\x')
implies !(x V x') ~ ! (x) /\ !(x') and (ii) ! (x) V!(x') > !(x /\ x') implies
!(x V x') > ! (x) /\ lex'), for all x E L, x' E L.
Note that when L = lR every function is pseudo-supermodular as the
order operations meet and join are then trivial. When the choice space
is of greater dimension, pseudo-supermodularity involves additional mul-
tivariate restrictions as well. First, it is clear from the definition that any
strictly increasing or decreasing function is pseudo-supermodular. Further-
more, pseudo-supermodularity expresses a weak kind of complementarity
between the choice variables. It is indeed a weaker condition than super-
modularity. We recall here that a function I : L -> lR is supermodular if
!(x)+ !(x' ) ~ I(x/\x')+ !(xVx') for all x E L, x' E L. A function I: L -> lR
is quasi-supermodular if (i) I(x) ~ I(x /\ x') implies that !(x V x') ~ !(x')
and (ii) I(x) > !(x/\x') implies that !(xVx') > f(x'), for all x EL, x' EL.
The notion of quasi-supermodularity has been introduced by Milgrom and
Shannon (1994) and is a generalization of supermodularity.
Proposition 1. IT ! is supermodular then it is pseudo-supermodular.
Proof. Assume that! is not pseudo-supermodular, that is, there exist
x, x' E L such that! (x) V !(x' ) ~ !(x /\ x') and !(x V x') < ! (x) /\ lex').
Since! is supermodular we get lex) + I(x') ~ !(x /\ x') + !(x V x') <
! (x) /\ !(X') + ! (x) V lex'), which yields lex) + !(x') < !(x) + !(x'), a
contradiction. •
Proposition 2. IT! is quasi-supermodular, then it is pseudo-supermodu-
lar.
4
Proof. Suppose that f is not pseudo-supermodular, that is, there exist

x, x' E L such that f (x) V f(x') 2:: f(x A x') and f(x V x') < f (x) A
f(x'). Since f is quasi-supermodular, then, if f(x) 2:: f(x A x'), we have
f (x) V f(x') 2:: f(x) 2:: f(x A x') implies f(x V x') 2:: f(x'); therefore, we
get f (x) A f(x') ~ f(x') ~ f(x V x') and f (x) A f(x') > f(x V x'), a
contradiction. If f(x) < f(xAx'), then f(x') 2:: f(xAx') implies f(xVx') 2::
f(x) 2:: f (x) A f(x'), contradicting f(x V x') < f (x) A f(x'). Thus, it is
proved that if f is quasi-supermodular, then it is pseudo-supermodular as
for definition (i); similarly, it can be proved for (ii) .•
Remark 1. Quasi-supermodularity is not a necessary condition for pseudo-
supermodularity, which implies that our results cannot be captured by the
analysis in Milgrom and Shannon (1994). We give here an example of a
pseudo-supermodular function that is not quasi-supermodular.
Consider the following lattice L = {O, a, b, c}, where 0 = (0,0), a = (1,0),
b = (0,1), c = (1,1), with the usual partial ordering. Suppose that f : L ~
JR is such that f(O) < f(a) < fee) < f(b). It can be easily verified that
f is pseudo-supermodular. However, it is not quasi-supermodular. Indeed,
taking x = a, x' = b we have f(x) > f(x A x') but f(x V x') < f(x').
Let T be a partially ordered set and f : L x T ~ JR. Then we say that f
satisfies the single crossing (SC) property in (x, t) iffor x' > x" and t' > t",
(i) f(x',t") 2:: f(x",t") implies f(x',t') 2:: f(x",t') and (ii) f(x',t") >
f(x",t") implies f(x',t') > f(x",t'). The term "single crossing property"
is used by Milgrom and Shannon (1994) because the expression f(x', t) -
f(x, t), regarded as a function of t, crosses zero only once and only from
below, when these conditions hold. Furthermore, we say that f satisfies the
additional (A) property in (x, t) if for t' > t and for x, x' E L such that
f(x', t) > f(x, t), (i) f(x V x', t) 2:: f(x, t) implies f(x V x', t') 2:: f(x', t')
and (ii) f(x V x',t) > f(x,t) implies f(x V x',t') > f(x',t'). Condition
(A) implies that increasing a parameter (t) more than raises the marginal
returns in activities.
Now we can state our first main result.
Proposition 3. Let f : L x T ~ JR, where L is a lattice, T a partially
ordered set and Se L, and let 'P(t, S) be the set of optimal solutions to the
problem maxxES f(x, t). Then 'P(t, S) is monotone nondecreasing in (t, S) if
f is pseudo-supermodular in x and satisfies the (SC) and (A) properties in
(x, t). Pseudo-supermodularity in x and the (SC) property in (x, t) are also
necessary conditions for 'P(t, S) to be monotone nondecreasing in (t, S).
Proof. Let S' 2:: S, t' 2:: t, x E 'P(t, S) and x' E 'P(t', S'). Consider x V x';
since x E 'P(t,S) and S ~ S', then f(x,t) 2:: f(x Ax',t).
By pseudo-supermodularity and the definitions of join and meet, we ob-
tain the following string of inequalities:
f (x, t) V f (x' , t) 2:: f (x, t) 2:: f (x A x', t) => f (x V x', t) 2:: f (x, t) A f (x' , t).
5
By the (SC) and (A) properties we get
f(x V x', t) ~ f(x, t) 1\ f(x', t) => f(x V x', t') ~ f(x', t'),
hence x V x' E <p(t', S').

Similarly, we can show that x 1\ x' E <p(t, S). Indeed, consider x 1\ x';
since x' E <p(t',S') and S::; S', then f(x',t') ~ f(x V x',t'). By the (SC)
and (A) properties, we get f(xv x',t)::; f(x',t) 1\ f(x,t), and by pseudo-
supermodularity;
f (x V x', t) ::; f (x, t) 1\ f (x', t) = > f (x, t) ::; f (x, t) V f (x' , t) ::; f (x 1\ x' , t),
hence, x 1\ x' E <p(t, S).

Pseudo-supermodularity is also a necessary condition. Suppose t is fixed,
x, x' ELand let x be such that f(x, t) ~ f(x', t). Suppose that pseudo-
supermodularity does not hold, that is, there exist x, x' such that f(x 1\
x', t) ::; (or <)f(x, t) V f(x', t) = f(x, t) and f(x V x', t) < (or ::;)f(x, t) 1\
f(x', t) = f(x', t). Let S = x, x 1\ x' and S' = x', x V x'; then S::; S'. Thus
argmaxsf(x, t) ::; argmaxs' f(x, t). However, x ::; x' cannot hold, because
it yields xl\x' = x, xVx' = x' and therefore f(xVx', t) = f(x', t) < f(x', t),
(or f(x 1\ x', t) = f(x, t) < f(x, t)), a contradiction.
Finally, the (SC) property is also a necessary condition. Indeed, let S =
{x, x"} with x" ~ x. Then f(x",t) - f(x,t) ~ (or »0 implies that x" E
<p(t,S) ::; <p(t",S) for t" ~ t, so f(x",t") - f(x,t") ~ (or »0 for every
t" ~ t .•
Remark 2. Let us give an example of a function that satisfies pseudo-
supermodularity and the (SC) and (A) properties, but fails quasi-
supermodularity, and whose optimal solutions have the monotonicity prop-
erties as stated in Proposition 3.
Consider the lattice L introduced in Remark 1, let T = 1,2, and take
the function F: L x T --+ JR, F(x, t) = {i<~»/ x~b , where f(O) < f(a) <
f(c) < f(b) < 2f(c), f(b) > 2f(0), for every t E T.
It can be easily verified that F is pseudo-supermodular in x, satisfies the
(SC) and (A) properties in (x, t) and that the set of optimal solutions to
the problem maxF(x, t), s.t. x E S CL is nondecreasing in (t, S), as stated
in Proposition 3.
However, F is not quasi-supermodular in x. Indeed, taking x = a, x' = b,
we have F(x, 1) > F(x 1\ x', 1) but F(x V x', 1) < F(x', 1).
Proposition 3 gives a characterization of the set of optimizers. Let us now
develop the analysis for equilibrium problems and introduce the notion of
pseudo-supermodular games.
Consider a non-cooperative n-person game with the players denoted by
i = 1, ... , n. The decision of player i is an mi-vector Xi. The joint decision
is x = (Xl, ... , x n ) and the set of feasible joint decisions is a subset L
6
of ]Rm where m = L:::~=l mi. Elements of L are called strategy profiles,

L = LI Ln. Each strategy set Li is partially ordered by 2: and the
X ... X
strategy profiles are endowed with the product order, that is, x 2: x, means
Xi 2: x~ for all i. We assume that (Li, 2:) is a complete lattice for all i. The
feasible decisions for a given player may depend on the decisions chosen by
the other players. Let fi(X) be the payoff player i gets as a result of a joint
decision x E L, where Ji(x) is a real-valued function on L for i = 1, .. n. Let
(X-i, Yi) = (Xl, ... , Xi-I, Yi, Xi+t, ... , xn). A feasible joint decision X E L
is an equilibrium point if Ii(x) 2: fi(X-i, Yi) for all Yi E Li(X), where
Li(X) = {Yi : (X-i, Yi) E L} and i = 1, ... , n.
Definition. A game is a pseudo-supermodular game if for each i, Li is a
complete lattice, fi(X) is pseudo-supermodular in Xi, and fi(X) satisfies the
(SC) and (A) properties in (Xi, x-d.
We can now state our second main result.
Proposition 4. Suppose a game is pseudo-supermodular and, for each i, Li
is compact, Ii is upper semi-continuous in Xi for fixed X-i, and continuous
in X-i for fixed Xi. Then for each i there exist strategies ~i and Xi which
are the smallest and largest strategies which survive strict pure iterated
admissibility, and the pure strategy profiles ~= (~i' i = 1, ... , n) and X =
(Xi, i = 1, ... , n) are Nash equilibria.
Proof. We recall here that the strategies which survive strict pure iterated
admissibility are the strategies that remain after iterated elimination of
pure strategies that are strictly dominated by other pure strategies. For
S c L, let U(S) = (Ui(S), i = 1, ... , n), where Ui(S) = {Xi E LilVx~ E
Li,:JX E S such that Ii(Xi, X-i) 2: fi(X~' X-in, is the set of strategies of
player i that are not strictly dominated when i plays against strategies in
S. The proof is done in two steps.
STEP 1. Take ~, z E L with ~ ::; z. The first step amounts at showing
that sup U([~, zl) = B(z) and inf U([~, zl) = B(~), where B(z) is the
largest best response to z and B(~) the smallest to~.
Certainly, [B(~), B(z)] c [inf U([~, zl), sup U([~, z])], by the definition
of B and B. We have to show the converse, that is, if Z rt. [B(~), B(z)] then
Z rt. U([~, zl).
Consider first the case Zi not::; rh = B(z) for some i, and let X E [~, z]. In
this case we can show that Zi /\ fh strictly dominates Zi against every X E [~,
z]. Indeed, if fi(Zi, X-i) - fi(Zi /\ fh, X-i) 2: 0, by pseudo-supermodularity
we can write the following chain of inequalities:
fi (Zi , X_i)V fi(Yi, X-i) 2: fi(Zi, X-i) 2: li(ziNh,·x-i) => li(ziVfh, X-i) 2:
fi(Zi, X-i) /\ fi (t}i , X-i)' By the (SC) and (A) properties we get fi(Zi V fh,
Z-i) ~ Ii(fh, Z-i). But Zi Vih > ih, since Zi not::; fh, and by the definition
of ih, Ji (Zi Vih, Z-i) < Ii (fh, Z-i), a contradiction. Therefore, Zi /\ rh strictly
dominates Zi against every X E [~, z], implying that Z rt. U([~, zl).
7
Consider now the case Zi not 2: J!..i = B(gJ for some i, and let x E [~, iJ.
In this case we can show that Zi V y. strictly dominates Zi against every
x E [~, iJ. Indeed, suppose it is not true, that is, !i(ZiVJ!..i' X-i) ::; fi(Zi, X-i).
-~
Then by the (SC) and (A) properties we can write fi(Zi V J!..i' ~-i) ::; !i(Zi,
~-i)/\ !i(J!..i' ~-i) and by pseudo-supermodularity !i(J!..i,Li) ::; fi(Zi'~_i)V
fi(J!..i' ~-i) ::; !i(Zi /\ J!..i'~-i)· But this yields a contradiction because by the
definition of -Yi and by Zi /\ -~ y. < -1.
y. (since Zi not 2: -1.
y.), we should have
fi(Y., ~-i) > fi(Zi /\ -~ Y·, ~-i)' Therefore !i(Zi Vy., x-d > fi(Zi, X-i) and
Z tJ. U([~, iD. It follows that [B(~), B(i)J = [inf U(~, iD, sup U([~, i])].
-~ -~
STEP 2. The proof proceeds as in Milgrom and Roberts (1990), using

the result of the previous step 1, the definition of strategies which survive
strict pure iterated admissibility, the continuity of f and the definition of
Nash equilibrium. •
Proposition 4 states that all strategies which survive strict pure iterated
admissibility lie in an interval (.2:, x) whose maximum and minimum points
are the largest and smallest Nash equilibria. Obviously, if the game has a
unique pure Nash equilibrium it follows that it is dominance solvable.
Remark 3. The set of Nash equilibria of a pseudo-supermodular game is

a non-empty complete lattice. A parallel result is proved by Zhou (1994)
for supermodular games. Since the set of Nash equilibria is the set of fixed
points of the best response correspondence B(x), where B(x) = nr=lBi(X),
in which Bi(x) = argmaxxiELi!i(Xi, X-i), the claim follows from an ap-
plication of a generalization of Tarski's (1955) fixed point theorem for a
monotone non-decreasing correspondence on a complete lattice (see The-
orem 1, Zhou, 1994). Because B is the product of B:s, we only need to
show that Bi satisfies the conditions of the above-mentioned fixed point
theorem. First, take any X-i 2: x'.-i' Zi E Bi(X) and z: E Bi(X'); we have
Zi V z: E Bi(X) and Zi /\ z: E Bi(x') by Proposition 3, that is, Bi is
monotone non-decreasing. Second, take Zi , z: E Bi(X); by Proposition 3
we have Zi V z: E Bi(X) and Zi /\ z: E Bi(X), that is, Bi is a sublattice of
L i . Finally, since fi is upper semi-continuous in Xi for X-i and continuous
in X-i for fixed Xi, then Bi is compact, hence complete (see Birkhoff, 1948).
These conditions allow us to conclude that the set of fixed points of B is a
non-empty complete lattice.
Notice, however, that the set of Nash equilibria of a pseudo-supermodular
game need not be a sublattice of 1. Here is an example adapted from
Topkis (1979). Let n = 3, ml = m2 = m3 = 1, Li = [O,lJ for each i,
and fl(X) = h(x) = h(x) = XIX2X3. Then (1,0,0) and (0,1,0) are Nash
equilibria but (1,1,0) = (1,0,0) V (0, 1,0) is not a Nash equilibrium, so the
set of equilibrium points is not a sublattice of L.
8
3 Pseudo-supermodular games and selection

dynamics: an application
In this section we consider, as an application, an analysis of equilibrium
solutions for a large class of dynamic models of evolutionary selection
processes on games.
r
Denote by a pseudo-supermodular game in which Li is compact, fi
is upper semi-continuous in Xi for fixed X-i, and continuous in X-i for
fixed Xi, for each i, as it is required by Proposition 4. Suppose that the
n players are repeatedly randomly matched to play single repetitions of
the game r. Each player i plays a pure strategy Xi, where Xi is an mi-
vector in L i . Denote by D.(Li) the set of probabilities distributions on Li
(mixed strategies). The distribution of strategies among players is given by
W E b.(Ll) .x ... , ' x D.(Ln), D.(Li) C ]Rrni, which identifies the proportions
of players playing each of the pure strategies. Over time, such proportions
adjust in response to payoff differences. These changes can be described in
the following way:
Definition. A function g assigns to each population state w, player pop-
ulationi, and pure strategy Xi available to player i, the growth rate
gixi(w) of the associated population share Wix, = (w~~~, .. "W~:i»). Write
g(w) = (gl(W), ... ,gn(w)), Then, for all w, and Xi, i = 1, .. " n,
(1) wix,
-(j)
= gix, ( W) WiXi'
(j) , 1
J = , .. " mi
is a regular selection dynamics if it satisfies for any w

(i) g is Lipschitz continuous;
. ( ) (j) - 0
") ",rni
( 11 L....j=l glX, W wixi - ,z. -- 1, .. " n,
We make use of the following property:
Definition. g is payoff monotonic if, for all population states w, player
positions i and pure strategies Xi, X~
Monotonicity requires that if pure strategy Xi receives a higher payoff

than x~ and if both Wix, and Wix' are positive, then WiXi grows faster than
Wix'. . Denote by cm the class of payoff-monotonic g-functions. Such a class
enc~mpasses several dynamics which have received attention by economists;
in fact, the imitation dynamics, the standard replicator dynamics and con-
tamination dynamics are examples of it (Weibull, 1995).
Now we have the following result:
Proposition 5. For any g E cm and pseudo-supermodular game r, all
plays lie eventually in the set of strategies that survive strict pure iterated
admissibility. Such a set is bounded above and below by the largest and
smallest Nash equilibrium strategy profiles.
9
In particular, for games that have only one strategy profile which survive
strict pure iterated admissibility, every process consistent with the above
dynamics converges to such unique strategy profile, which is Nash.
Proof. The proof consists of two steps.
STEP 1. The first step amounts at showing that if Xi E Li does not sur-
vive pure strict iterated admissibility, then its population share converges
to zero along any interior initial solution path to (1). Such a proof is a
straightforward generalization of Samuelson and Zhang (1992) for the case
of multipopulations.
We recall here that a strategy Xi E Li survives pure strict iterated ad-
missibility if there exist sequences of the form Li = XiO, Xil, ... , Xit, with
Xik ~ Li and where Xik+l is the set of player i's pure strategies that are
not strictly dominated by any pure strategies in Li given that the other
players j choose strategies from Xjk ~ Lj, j = 1, .'" i - 1, i + 1, .. " n, for
k = 1, .. " T - 1, and with Xi E Xit = Xit+l. Let Sio ~ Li be the set of
player i's strategies which do not survive pure strict iterated admissibility
and are not eliminated in the limit by the selection process.
Now, suppose that the statement above - that is, the population share
associated with a strategy which does not survive pure strict iterated ad-
missibility, converges to zero along any interior initial solution path to (1)
- is false; then U Sio =1= 0,
i
For all Y E USio let k(y) be such that y E Xik(y) \ X ik (y)+1 if Y E Sio
i
and let Yo be the minimizer of key) on USio and put k = k(yo). Let
i
Yo E Sio. Then there exists Yl E Li such that h(yo, X-i) < h(Yb X-i)
for all X-i E TI Xjk. Since k minimizes key), we have limt-+co W-i(t) = 0
#i
for all X-i tf- TI Xjk, Then, as t - t 00, fi(Yo, W_i(t)) - h(Yl, W-i(t)) goes
#i
to limt-+co,x_iE IT X jk [h(yo, X-i) - h(Yb X_i)]W_i(t), which is negative.
j#i
By the definition of regular and monotonic selection dynamics, there exists
8 < 0 such that
iih yo (t) _ WiYl (t) < 8 for all t sufficiently large,

Wiyo (t) WiYl (t)
which yields
Wiyo (t) ~ GeM, for some G,
WiYl (t)
Since limt-+co Wi!lQ~tt~
W"'Yl
= 0, it follows that limt-+co Wiy 0 (t) = 0, contradicting
the definition of Yo' Therefore, USio = 0,
i
STEP 2. The second step is to show that the set of strategies that survive
strict pure iterated admissibility is bounded above and below by the largest
10
and smallest Nash equilibrium strategy profiles. Such a step follows from
Proposition 4. Finally, also the last assertion of Proposition 5 follows from
Proposition 4. •
Proposition 5 is a further contribution within the research area examin-
ing the links between equilibrium concepts and the outcomes of dynamic
evolutionary processes.
References
Agliardi, E. and M. Bebbington (1994), Self-reinforcing Mechanisms and
Interactive Behaviour, Economics Letters, 46, 281-287
Agliardi, E. and M. Bebbington (1997), Self-reinforcing Mechanisms and
Market Information, European Journal of Operational Research, 96, 444-
454
Birkhoff, G. (1948), Lattice Theory, American Mathematical Society, New
York
Bomze, I. (1986), Non-Cooperative Two-Person Games in Biology: a Clas-
sification, International Journal of Game Theory, 15, 31-57
Bulow, J., J. Geanakoplos and P. Klemperer (1985), Multimarket
Oligopoly: Strategic Substitutes and Complements, Journal of Political
Economy, 93, 488-511
Cooper, R. and A. John (1988), Coordinating Coordination Failures in
Keynesian Models, Quarterly Journal of Economics, 103, 441-463
DeGraba, P. (1995), Characterizing Solutions of Supermodular Games; In-
tuitive Comparative Statics, and Unique Equilibria, Economic Theory,
5, 181-188
Dybvig, P. and C. Spatt (1983), Adoption Externalities as Public Goods,
Journal of Public Economics, 20, 231-247
Farrell, J. and G. Saloner (1985), Standardization, Compatibility and In-
novation, Rand Journal of Economics, 16, 70-83
Katz, M. and C. Shapiro (1986), Technology Adoption in the Presence of
Network Externalities, Journal of Political Economy, 94, 822-841
Lippman, S., J. Mamer and K. McCardle (1987), Comparative Statics in
Non-Cooperative Games via Transfinitely Repeated Play, Journal of Eco-
nomic Theory, 41, 288-303
Milgrom, P. and J. Roberts (1990), Rationalizability, Learning and Equi-
librium in Games with Strategic Complementarities, Econometrica, 58,
1255-1277
Milgrom, P. and J. Roberts (1991), Adaptive and Sophisticated Learning
in Normal Form Games, Games and Economic Behaviour, 3, 82-100
Milgrom, P. and J. Roberts (1994a), Comparing Equilibria, The American
Economic Review, 84,441-459
Milgrom, P. and J. Roberts (1994b), The Economics of Modern Manufac-
turing: Technology, Strategy and Organization, The American Economic
11
Review, 80, 511-527

Milgrom, P. and C. Shannon (1994), Monotone Comparative Statics,
Econometrica, 62, 157-180
Nachbar, J.H. (1990), Evolutionary Selection Dynamics in Games: Conver-
gence and Limit Properties, International Journal of Game Theory, 19,
59-90
Samuelson, L. and J. Zhang (1992), Evolutionary Stability in Asymmetric
Games, Journal of Economic Theory, 57, 363-391
Shannon, C. (1995), Weak and Strong Monotone Comparative Statics, Eco-
nomic Theory, 5, 209-227
Tarski, A. (1955), A Lattice-theoretical Fixed Point Theorem and its Ap-
plications, Pacific Journal of Mathematics, 5, 285-309
Topkis, D. (1979), Equilibrium Points in Nonzero-sum n-Person Submod-
ular Games, Siam Journal of Control and Optimization, 17, 773-787
Vives, X. (1990), Nash Equilibrium with Strategic Complementarities,
Journal of Mathematical Economics, 19, 305-321
Weibull, J.W. (1995), Evolutionary Game Theory, The MIT Press, Cam-
bridge MA
Zhou, L. (1994), The Set of Nash Equilibria of a Supermodular is a Com-
plete Lattice, Games and Economic Behaviour, 7, 295-300
Degree of manipulability of social choice
procedures'
Fuad Aleskerov 1 and Eldeniz Kurbanov2
1 Department of Economics, Bo~azi~i University, Bebek, Istanbul 80815 Turkey
and Institute of Control Sciences, Moscow, 117806 Russia

2 Department of Computer Sciences, Bogazici University, Bebek, Istanbul 80815
Turkey
Abstract. 26 known and new social choice rules are studied via computa-
tional experiments to reveal to which extent these rules are manipulable. 4
indices of manipulability are considered.
JEL Classification Number: D71
Keywords: Social choice rules, degree of manipulation, efficiency of ma-

nipulation
1 Introduction
Theoretical investigation of manipulation problem shows that in a very
wide framework any rule transforming agents' preferences into social choice
is either manipulable or dictatorial [5,11]. To which extent social choice cor-
respondences are manipulable were studied in [6,13]. More concretely, in [6]
the degree of manipulability of choice procedures was introduced. A degree
of manipulability was measured as the ratio of manipulable profiles, i.e.,
profiles, in which at least one agent can manipulate, to the total number of
profiles. Unlike [6], in [13] already several measures of manipulation were
studied. The present work continues [6,13]. We investigate a degree and an
efficiency of manipulation of 26 known and new social choice procedures,
'Yapl Kredi Bank (Turkey) providcc\ t.wo personal cOlllput.ers for carrying out. t.llC
(:omput,at.ions. Our (:olleagues Profs. C. Akc;ay, E. Alper, Y. ArM" S. (hmucur, A.
Garkoglu, M. Eder from Bogazic;i Universit.y (Ist.anbul, Turkey) gave us access t.o t.heir
pers(lllal (:omput.ers for t.his work. Prof. G. Alpay kiudly gave his perlllission t.o use t.hc
(:Oluput.er lahorat.ory (10 PCs) present.ed t.o Boga,r,ic;i Universit.y hy Int.erhallk. Prof. H.
Erse! support.ed t.his work from t.he very heginlling. Mrs. B. Bl\rekc;i gave us valuahle
t.edlllieal assiAt,ance.
The work of F. Aleskerov was part.ially support.ed hy Russian FOllndat.ion of Basic
Research (grant. 95-<ll00057A), NATO Research Program (1995-1996), and t.he grant. of
European COlllmunit.y (INTAS Project 'Measurement. and Aggregat.ion of Preferences').
We express ollr t.hanks t.o all t.hese eolleagues and organi,r,ations.
14
including those which were studied in [13J. Articles [6,13J and the present
paper also have another common feature. Since the problem of theoreti-
cal investigation of the degree of manipulability of social choice rules has
not been solved yet, in the present work, like in [6,13J, the analysis has
been done by computer computations. The computations have been car-
ried out in two ways. With small number of agents n and alternatives m
the computations were exhaustive, Le., included complete examination of
manipulability of all profiles. In those cases when such examination re-
quired a great amount of time (for some rules already with 5 agents and 4
alternatives), statistical method has been applied.
The structure of the paper is as follows. In Section 2 measures of ma-
nipulability are described. In Section 3 the social choice rules under study
are given. In Section 4 the computation scheme is given. In Section 5 the
results are discussed and concluding remarks are given.
2 The framework
The following problem is considered. A finite set A of alternatives is given
which consists of m alternatives (m > 2). Agents from finite set N
{I, ... , n}, (n > 1), have preferences over alternatives from the set A.
These preferences are assumed to be linear orders, that is
- irreflexive (\Ix xPx),
- transitive (\lx,y,z xPy&ypz::::} xpz), and
- connected (\Ix, y, x f y either xPy or ypx) binary relations.
An ordered n-tuple of individual binary relations is called a profile. A
social decision is considered to be a subset of the set A. The set of all linear
orders on A is denoted as C. Then the social choice rule under the question
is defined as F: cn --t A.
The effect of manipulability can be described as follows. Let
be the profile of agents' true preferences, while
be a profile in which all agents but i-th reveal their true preferences. Let
---7 ---7
C( P), C( P -i) denote (single valued) social choice with respect to profile
--t ---7 ---7--t
P and profile P -i, respectively. Then C( P -i)PiC( P), which is the out-
come when i-th agent deviates from her true preference, is more preferable
for her than in the case when she expresses her sincere opinion. We study
the effect ofmanipulability as in [6,13J under the condition that social choice
always consists of only one alternative. Hence the following common way
of how to switch from a multiple choice (which is defined by the rule under
15
--+
the given P) to a single valued choice is accepted: in the case of multiple
choice the outcome has been chosen with respect to the alphabetical order
--+
of chosen alternatives. For instance, if the choice is C(P) = {X2,X3,X5},
then final social choice is defined to be X2.
3 Indices of manipulability
The number of alternatives being m, the total number of possible linear
orders is obviously equal to m!, and total number of profiles with n agents
is equal to (m!)n. In [6J, to measure a degree of manipulability of social
choice rules, the following index was introduced (we call it Kelly's index
and denote as K):
do
K = (m!)n'
where do is the number of profiles in which manipulation takes place! .

Before defining the next criterion of manipulability let us note that for an
agent there are (m! -1) linear orders to use instead of her sincere preference.
But in how many cases would her manipulation be successful? The answer
on this question leads to the notion of "freedom" of manipulation of a given
agent in a given profile. Denote as Ki (i = 1, ... , nj 0 :S Ki :S m! - 1) the
number of such orderings. Dividing it to (m! - 1) one can find the freedom
of manipulation of an agent i in this profile. Summing up this index over all
agents and dividing it to n one can find the average freedom of manipulation
in the given profile. Summing the index over all profiles and dividing this
sum to (m!)n we obtain our next index
,,(m!)n
~j=l
"n
~i=l Ki
h= (m!)n . n . (m! - 1)
.
Thus, the index h defines an average freedom of manipulation of a given
social choice function.
These indices K and h (as well as index J) measure the degree of ma-
nipulability in terms of the share of manipulable profiles or the share of
orderings using which an agent can manipulate.
The following two indices show the efficiency of manipulation, i.e., to
which extent an agent can be better off via manipulating her sincere or-
--+
dering. Let social decision be, under a profile P, an alternative x which
stands at k-th place from the top in the sincere ordering of i-th agent. Let
after her manipulation the social decision be an alternative y which stands
1 We also studied an extended version of Kelly's index. Denote by ~k the number of

profiles in which exactly k voters can manipulate. Construct index Jk = (~~n which
shows the share of profiles in which exactly k voters can manipulate. Obviously. K =
Jl + J2 + ... + I n . Then we consider the vectorial index J = (JIt J2 •.... I n ).
16
in the sincere ordering of the i-th agent at j-th place from the top, and let
j < k. Then 0 = j - k shows how the i-th agent is better off. Let us sum
up 0 for all advantageous orderings /'i,i (defined above), and let us divide
the obtained value to /'i,i. Denote this index through Zi, which shows an
average "benefit" of the agent i gained via manipulation /'i,i orderings from
(rn! - 1). Summing up this index over all agents and over all profiles, we
obtain the index under study
J.2 - L):in L~=l Zi

- (rn!)n·n .
The indices It and 12 were also used in [13].

The next criterion 13 is a modification of 12 , Instead of evaluating the
"average" benefit Zi for i-th agent, we evaluate the value
Ziax = max(Zl, ... , ZKJ.

In other words, the value Ziax shows the maximal benefit which can be
obtained by agent i. Summing up this index over all agents and over all
profiles, we obtain our next index
The indices K,11,12,I3,J have been calculated for each of the rules in-
troduced in the next section.
4 Social choice rules

Let us introduce several important notions and definitions, which are per-
manently used hereafter. The concepts and rules used below can be found
in [1-4,8,9,13,14].
--t
Definition 1. Majority relation for a given profile P is a binary relation
IL which is constructed as follows:
XILY {:} card{i E N I XPiY} > card{i E N I yPix}.

--t
Definition 2. Condorcet winner CW in the profile P is an element un-
dominated in the majority relation IL (constructed according to the profile),
i.e.,
CW = {a I a: 3x EA, XfLa}.
--t
Definition 3. A contruction of a profile P onto the set X ~ A is a profile
--t
P /X = {Pt/X, ... ,Pn/X}, Pi/X = Pi n (X x X).
17
Definition 4. Upper contour set of an alternative x in the relation P is

the set D{ x) such that
D(x) = {y E A I yPx}.
Lower contour set of x in the relation P is the set L{ x) such that
L{x) = {z E A I xPz}.
Let us describe now the social choice rules. The rules under study can
be divided into several groups.
a) Scoring rules;
b) Rules using majority relation;
c) Rules using value function;
d) Rules using tournament matrix;
e) q-Paretian rules.
4.1 Scoring rules

1. Plurality Rule.
Choose alternatives that have been admitted to be the best by the max-
imum number of agents, i.e.,
where n+(a, P) = card{i E N I Vy E A aPiy}.

---t
Consider the example with 5 agents and 4 alternatives given below.
PI P2 P3 P4 P5
a b d b a
d a ace
b c bad
c d c d b
In this example according to the plurality rule alternatives a and b will

be chosen. Then, the single valued choice will be a.
2. Approval Voting.
Let us define
n+(a, P,q) = card{i E N I card{Di(a)}::; q},

---t
i.e., n+ (a, P, q) means the number of agents for which a is placed on q+ 1'st
place in their orderings. Thus, if q = 0, then a is the first best alternative
18
for i-th voter; if q = 1, then a is either first best or second best option, etc.
The integer q can be called as degree of procedure.
Now we can define Approval Voting Procedure with degree q
--+ --+--+
a E C(P) {:} [Vx EA n+(a, P ,q) 2 n+(x, P ,q)],
i.e., the alternatives which have been admitted to be between q + 1 best by

the maximum number of agents are chosen.
It can be easily seen that Approval Voting Procedure is a direct gener-
alization of Plurality Rule; for the latter q = O.
Remark. Approval Voting Procedure is in general defined in such a way
that a voter can choose any number of alternatives from the top, i.e., q is
defined not for all voters equally [4].
3. Run-off Procedure.
First, a simple majority rule is used (an alternative with 50%+1 votes
will be chosen). If such an alternative exists the procedure stops; otherwise,
two alternatives with maximal number of votes are taken. Assuming that
preferences of agents about these two alternatives do not change, again
simple majority rule is applied. Since agents' preferences are linear orders,
a single winner (if number of agents is odd) always exists.
4. Hare's Procedure.
Firstly simple majority rule is used. If such an alternative exists, the
procedure stops; otherwise, the alternative a with the minimum number of
votes is omitted. Then the procedure is again applied to the set X = A \ {a}
--+
and the proffie P / X.
5. Inverse Plurality Rule.
The alternative, which is regarded as the worst by the minimum number
of agents, is chosen, i.e.,
--+
where n-(a, P) = card{i E N I Vy E A YPia}.
Consider the previous example.
--+ --+ --+ --+
In this example n-(a, P) = O,n-(b, P) = l,n-(c, P) = n-(d, P) = 2.
--+
Therefore C( P) = {a}.
6. Borda's Rule.
--+
Put to each x E A into correspondence a number ri(x, P) which is equal
--+
to the cardinality of the lower contour set of x in Pi E P, i.e.,
19
The sum of those numbers over all i E N is called Borda's count for
alternative x,
n
r(a, P) = LT\(a,Pi).
i=l
The alternative with maximum Borda's count is chosen, i.e.,

--f' r ~ ----1'
a E C(P) ~ [Vb E A,r(a, P) 2 r(b, P)].
7. Black's Procedure.
If Condorcet winner exists, it is to be chosen. Otherwise, Borda's Rule
is applied.
8. Inverse Borda's Procedure.
For each alternative Borda's count is calculated. Then the alternative a
with minimum count is omitted. Borda's counts are re-calculated for profile
-+
P / X, X = A \ {a}, and the procedure is repeated until choice is found.
9. Nanson's Procedure.
For each alternative Borda's count is calculated. Then the average count
is calculated, r = (I: r(a, P)) /IAI ,
aEA
and alternatives c E A are omitted
-+ -+
for which r(c, P) < r. Then the set X = {a EA: r(a, P) 2 r} is consid-
-+
ered, and the procedure is applied to the profile P / X. Such procedure is
repeated until choice is not empty.
10. Coombs' Procedure.
The alternative which is the worst for maximwn nwnber of agents is
omitted. Then profile is contracted to the set X, and the procedure is
repeated until choice is not empty.
4.2 Rules using majority relation

11. Minimal Dominant Set.
A set Q is called a dominant one if each alternative in Q dominates each
alternative outside Q via majority relation. Otherwise speaking,
x E Q {:} [Vy E A\Q, xILyJ.

Then a dominant set Q is called the minimal one if none of its proper
-+
subsets is a dominant set. Then social choice C( P) = Q. The final choice
consists of the alternative which is the first in alphabetical order in the
chosen set.
20
12. ~linimal Undominated Set.

A set Q is called an undominated one if no alternative outside Q dom-
inates some alternative in Q via majority relation. Undominated set Q is
called the minimal set if none of its proper subsets is an undominated set.
--+
Then social choice C( P) = Q. If minimal undominated set is not unique,
then social choice is defined as the union of these sets.
13. Minimal Weakly Stable Set.
A set Q ~ A is called weakly stable if it has the following property:
x E Q iff [3y E A \ Q S.t. yJ.LX => 3z E Q s.t. ZJ.Ly],
Le., x belongs to Q iff (if there exists an alternative y E A\Q which dom-
inates x via majority relation J.L, then there exists an alternative Z E Q,
which dominates y, Le., ZJ.LY.)
A set Q ~ A is called the minimal weakly stable set if none of its proper
subsets is a weakly stable set.
--+
Then social choice C( P) = Q.
This rule is an ordinal counterpart of Maschler's bargaining set [7].
14. Fishburn's Rule.
Construct upper contour set D(x) of relation J.L and binary relation, as
follows:
x,y {:=> D(x) C D(y).
Then undominated alternatives on , are chosen2 , i.e.,
15. Uncovered Set 1.

Construct lower contour set L(x) of relation J.L and binary relation 8 as
follows:
xoy {:=> L(x) :J L(y).
Then undominated alternatives on 8 are chosen, Le.,
--+
x E C(P) {:=> (13y E A I y8x).
16. Uncovered Set H.

An alternative x is said to B-dominate an alternative y (denoted as xBy)
if XJ.Ly and D(x) ~ D(y), where D(x) is the upper contour set of x in J.L.
Social choice consists of B-undominated alternatives, Le.,
xE CC?) {:=> (l3y E A I yBx).

2Note that 'Y is a strict partial order, i.e., irrefiexive and transitive binary relation.
21
17. Richelson's Rule.

Construct lower and upper contour sets D(x) and L(x) for each x E A
in the relation jt, and binary relation er as follows:
xery ~ [L(x) :2 L(y) /\ D(x) ~ D(y)/\
([L(x) =:J L(y)] V [D(x) c D(y)])].

Then undominated alternatives on er are chosen, i.e.,
--t
X E C(P) ~ (l3y E A I yerx).
4.3 Rules using value function

18. Copeland's Rule l.
Construct function u(x), which is equal to the difference of cardinalities
of upper and lower contour sets of alternative x in majority relation jt, i.e.,
u(x) = IL(x)I-ID(x)l. Then social choice is defined by maximization of u,
that is,
--t
X E C(P) ~ [Vy E A, u(x) ~ u(y)].
19. Copeland's Rule 2.

FUnction u( x) is defined by cardinality of lower contour set of alternative
x in majority relation jt. Social choice is defined by maximization of u(·).
20. Copeland's Rule 3.
Function u( x) is constructed by cardinality of upper contour set of alter-
native x in majority relation jt. Social choice is defined by minimization of
u(·).
21. Young's Procedure.
--t
If for profile P there exists Condorcet Winner, then it is chosen. If there
is no such alternative, then partial Condorcet winner CW (j) is defined
on coalition w, Iwl = i < n. Construct the function u(x) as cardinality of
maximal coalition in which x is the Condorcet Winner. Then the alternative
with maximal value u(x) is chosen, i.e.,
--t
a E F( P) <=} (a is CW(jo) for io = m~ {j}).
l:$J:$n
4.4 Rules using tournament matrix

22. Simpson's Procedure (Maxmin Procedure).
Construct matrix S+ such that
S+ = {n(a, b)}, n(a, b) = card{ i E N I aPib}, n(a, a) = 00.

22
Social choice is defined as

---+
x E C( P) {:} x = argmax{min{
aEN bEA
n(a, b)}}.
23. Minmax Procedure.

Construct matrix S- such that
\la, bE X, S- = {n(a, b)}, n(a, a) = -00.

---+
x E C( P) {:} x = argmin{max{
bEA aEA
n(a, b)}}.
4.5 q-Paretian rules

The class of the rules considered below was introduced in [1] and studied
in [2].
24. Strong q-Paretian Simple Majority Rule.
---+
Let f(P; {i},q) = {x EA: IDi(x)1 ~ q}. Let I = {l eN: III = r~l}
be the family of simple majority coalitions. Then social choice is defined as
C(A) = un f(?;
IEIiEI
{i},O),
i.e., the alternative which is in between top alternatives for each voter in at
least one simple majority coalition is chosen. If there is no such alternative,
then the choice is considered over simple majority coalitions with q=l, q=2,
etc., until the choice is not empty. Again, if more than one alternative is
chosen, the final choice is defined by alphabetical rule of tie-breaking.
25. Strong q-Paretian Plurality Rule.
This rule is a counterpart of the rule 24 with the following addition. If
several options are chosen, then for each alternative is counted how many
coalitions choose this alternative. Then the alternative with maximal value
of this index is chosen. The rule is studied in [12].
26. Strongest q-Paretian Simple Majority Rule.
C(A) = n
IEI
J(P;l,q),
where f(P;l,q) = {x EA: !nDi(X)! ~ q}, III = r~l The alternative is

iEI
chosen if it is Pareto optimal in each simple majority coalition with q=O.
If there are no such alternatives, then q=l, q=2, etc., is considered until
the choice is not empty.
23
5 Computation scheme
With small values of parameters m and n the exhaustive study of ma-
nipulability of the rules has been made. This was done via analysis of all
proffies and checking all possible orderings for all agents. For each voter
(m! - 1) number of orderings are checked. The complexity is evaluated
as (m! - 1) . n . (m!)n . S, where S is a complexity of choice algorithm.
With m=5 and n=5 the approximate lower bound of complexity is 10 15
which makes that exhaustive study practically impossible3 • Hence we use
---t
the statistical scheme [13]. If we introduce a random variable 1/( P) which
---t ---t
shows whether the proffie P is manipulable in Kelly's sense, then 1/( P)
has binomial distribution with maximal variance
1
(j2 = N. p. (1 - p) = N· ("2)2 = 0.25· N,
where N = (m!)n is the number of proffies. Binomial distribution is ap-
proximated by normal distribution and 95% confidence interval is given
as
( 1.96· (j
jL- m 1.96· (j)
,jL+ m .
Choosing 0.02 confidence interval we obtain the evaluation of the number
of profiles
N = [2. 1.96 . 0.25] 2 ~ 240000
0.002 .
This number has been used for all indices.
6 Results of experiments
The complete tables with the results of the computational experiments can
be obtained via Internet, http://www.econ.boun.edu.tr/papers. Below we
provide only main observations. The indices K,h,I2,h, and J have been
evaluated for all combinations of values m = 3,4, 5, and n = 2 - 10. The
indices K, h, h, h have been evaluated for the case m = 5 and n = 51 as
well.
If one compares the rules in each group 4a-4e and chooses the rule which
has minimal value of the corresponding index on each pair (m, n) with
m = 3,5 and n = 3,5,7,9, i.e., with odd number of alternatives and agents,
then the following Table 1 can be obtained. In the cells of Table 1 the
numbers of rules are given.
It is immediately seen that the difference between rules is obtained on the
rules from the first group, namely, on the scoring rules. Hare's rule is less
3For some rules we made such an exhaustive study. For example, for Coombs's rule
the calculations for the case m = n = 5 took 2 months of work of PC-133.
24
Choice Rules K 11 12 13
Scoring Rules 4 7 2, with q=2 2, with q=2
by Majority Relation 14-17 14-17 14-17 14-17
by Value Function 18-20 18-20 18-20 18-20
by Tournament Matrix 22,23 22,23 22,23 22,23
q-Paretian Rules 26 26 26 26
TABLE 1.
manipulable in terms of Kelly's index, but if the index 11 is considered then

the less manipulable rule is Black's rule. The approval voting rule with q = 2
is the least manipulable rule when indices 12 and 13 are considered. In other
words, the approval voting rule is less beneficial in terms of manipulation.
To compare the rules in terms of degree and efficiency of manipulation
Figures 1-4 are given. In these Figures the values of indices K, It, 12 and
h are given. In each Figure one can see 6 graphs for the plurality, Bor-
da's, Hare's, Black's, Fishburn's and approval voting with q = 2 rules
constructed for the case m = 5 and n = 3,5,7,9,51. Since the correspond-
ing values for Simpson's rule and Copeland's rule are very close to those
of Fishburn's rule, those rules from groups 4c and 4d are not shown in the
Figures. 4
Note now that Hare's rule for Kelly's index has the least value, a bit
greater value has Fishburn's procedure. On the other hand, both Plurality
rule and Borda's rule have much higher values. For the index It Plurality
rule dominates all other rules, and the least value is observed for Black's
rule.
Now evaluating the efficiency of manipulation by indices 12 and 13 , one
can see that the least efficient one is the Approval Voting rule with q = 2.
For the case of 12 the values for Fishburn's rule is very close to those of
the approval voting rule, and for the case of 13 it is seen that the Approval
Voting rule is the least efficient one, i.e., it provides the least benefit (in
average) of manipulation. This fact gives one more reason of the practical
application of this rule. Note that both well known rules, Plurality rule
and Borda's rule, are very highly manipulable with respect to both of these
indices.
From our point of view since we can separate now the rules with respect
to their degree and efficiency of manipulation, the results presented can
be used in real situations of decision making. At the same time this study
4For the case m = 5 and n = 51 the total number of profiles is approximately equal
to 10 100 , among which we check only 240,000 profiles. So, we do not discuss the results
obtained in this case.
25
opens several problems. The first one, naturally, is how to obtain theoretical
estimations of the indices introduced above which are not restricted by
small values of m and n. Obtaining such estimations will provide the basis
of using different social choice rules in different practical situations.
The next question is 'Can we construct other indices to find different
properties of rules under study which deal with degree and efficiency of their
manipulation?'. Finally, in the study above we choose the alphabetical rule
of tie-breaking. However, it might be more realistic to make the analogous
study considering non single-valued choice, and using generalized concept
of manipulability (see, e.g., [10]).
References
1. Aleskerov, F. "Procedures of Multicriterial Choice." Preprints of the
IFAC/IFORS Conference on Control Science and Technology for Devel-
opment, Beijing, China, 1985.
2. Aleskerov, F. "Relational-Functional Voting Operators." California In-
stitute of Technology, Social Science Working Paper, n. 818, 1992.
3. Banks, J. "Sophisticated Voting Outcomes and Agenda Control." Social
Choice and We(fare, v.1, 1985.
4. Brams, S. and P.Fishburn (1983) Approval Voting. Boston, Basil Black-
well.
5. Gibbard, A. "Manipulation of Voting Schemes: A General Result."
Econometrica, v.41, 1973.
6. Kelly, J. "Almost All Social Choice Rules are Highly Manipulable, but
Few aren't." Social Choice and We~fare, v.10, 1993.
7. Maschler, M. "The Bargaining Set, Kernel, and Nucleoulus." in Hand-
book of Game Theory, (eds. R.J. Aumann and S. Hart), v.1, 1992.
8. Moulin, H. Axioms of Cooperative Decision Making. Cambridge, Cam-
bridge University Press, 1987.
9. Nurmi, H. Comparing Voting Systems. D. Reidel Publishing Comp., Dor-
drecht, 1987.
10. Pattanaik, P. Voting and Collective Choice. Cambridge, Cambridge Uni-
versity Press, 1976.
11. Satterthwaite, M.A. "Strategy-proofness and Arrow's Conditions: Ex-
istence and Correspondence Theorems for Voting Procedures and Social
Welfare Functions." Journal of Economic Theory, v.lO, 1975.
12. Sertel, M. and E.Kalaycioglu, "Tlirkiye i~in bir Se~im Y6ntemi
Tasanmma Dogru." TOSIAD, Istanbul, 1995 (in Thrkish).
13. Smith, D. "Manipulability Measures of Common Social Choice Func-
tions." Social Choice and We~fare, (forthcoming)
14. Volsky, V. and Z. Lezina, "Voting in Small Groups." Moscow. Nauka,
1991 (in Russian).
26
Figure 1.
07K~__________________________~
0:6 ".,.".,.,. ,." . ."".,. .,. . . ,.,. . .,. ,)(.. ,. ,. ,. ..." "" '.x"
0,5
--+--1 Plurality Rule
.,. ·0 ··· · 2 Approval Voting q=2
0,4 - . -. - - 4 Hare's Rule
·.. ·· .... H ..·····• 6 Borda's Rule
0,3 . __. ___ . _._--- - ----.- - -' -- - - -.-----
----.---- 7 Black's Rule
0,2 ---+-14 Fishburn's Rule
0,1
n
O+-------~----~------~----~
3 5 7 9 51
FIGURE 1.
0,045
0,04
0,035 --+--1 Plurality Rule
. ... 0 ·· .. 2 Approval Voting q=2
0,03
---B _. 4 Hare's Rule
0,025
·....... ·H·..·.. •· 6 Borda's Rule
0 ,02 ----.---·7 Black's Rule
0,015 ---+-14 Fishburn's Rule
0,01
0,005
n
0
3 5 7 9 51
FIGURE 2.
27
Figure 3.
0,45 ~12,---_ _ _ _ _ _ _ _ _ _ _ _-,
0,4 ............
0,35 ....• -.,.".,)(,.-.., '"
...•." .•..,.,..)(.,.,..,.•. , ..,.,.......
---+--1 Plurality Rule
0,3 .... 0 ···· 2 Approval Voting q=2
-.... ...• .. 4 Hare's Rule

0,25
~-.----------.--.---~---- ---~--- .~-- ........ ·K· ......· 6 Borda's Rule
0,2 . .......... -':' _ ... --_ .... .
.---...--- 7 Black's Rule
0,15
~ 14 Fishburn's Rule
0,1
0,05
n
°3 5 7 9 51
FIGURE 3.
Figure 4.
0,5 !;'IJ'---_ _ _ _ _ _ _ _ _ _ _---,
0,45
0,4
0,35
.,_ ,-
... .............., . . ..•................• ---+--1 Plurality Rule
.. ·· 0 .... 2 Approval Voting q=2
0,3
.. .• . . 4 Hare's Rule
o,25 ....- -
.........)(........ 6 Borda's Rule
0,2 ~......................~;,..;.;.,;d~~_: ----a---- 7 BI ac k's Rul e
0,15 ~ 14 Flshburn's Rule
0,1
0,05
n
o+-----~-----+----~----~
3 5 7 9 51
FIGURE 4.
On the properties of stable many-to-many
matchings under responsive preferences
Ahmet Alkan
Department of Management, Bogazi~i University, Bebek, Istanbul 80815, Turkey
Abstract. We show that the set of stable many-to-many matchings under

responsive preferences is a complete distributive lattice but does not possess
some of the other nice properties - monotonicity and strategyproofness -
associated with the set of stable one-to-one matchings.
JEL Classification Numbers: C78
Keywords: Matching, lattice
1 Introduction
Job matching when workers may contract with several firms and student
enrollment in courses are two examples of many-to-many matching situa-
tions that arise.
We show in this paper that the set of stable outcomes in a many-to-
many matching market has all the structural properties known for one-to-
one matching, when preferences are responsive (Roth) [6], that is to say,
when preferences over subsets of agents are consistent with complete strict
preference orders on the individual agents. We then give examples to show
that some of the other impressive properties that hold on the one-to-one
domain no longer hold for "polygamous" agents, namely those who are to
be matched with more than one agent from the opposite side, even when
preferences are responsive.
Roth and Sotomayor [7] had shown for many-to-one matchings under
responsive preferences that the sets of mates a polygamous agent may get
under alternative stable matchings are always such a selection of sets that
he has a complete strict preference order on them. They then showed that
the set of stable matchings, as in one-to-one matching, is a complete and
distributive lattice with respect to the partial orders induced by the com-
mon preferences of all agents on the same side of the market, with the
duality property that what is better for agents on one side of the market
coincides with what is worse for those on the other side. In particular, there
exists a best stable matching for agents on (any) one side of the market
which happens at the same time to be the worst stable matching for agents
30
on the other side.

We show here that all of the above properties also hold for many-to-
many matching. Let us point out that these results obtain here in a rather
different way than in Roth and Sotomayor [7], where, in fact, a many-
to-one market is first transformed into an equivalent one-to-one market
by the cloning of polygamous agents and then results known to hold for
one-to-one matching are applied. This method appears to be unyielding
in many-to-many matching. We here start by defining the supremum of
any two matchings as the matching where each agent on one side of the
market selects, according to his preferences on individuals, the best mates
to fill his quota from the mates he has in one or the other or both of the
initial two matchings. We define the infimum, not entirely analogously, as
that matching where each agent accepts the mates whom he has in both
of the initial two matchings and to fill his quota then selects the worst
among the remaining ones. We show that, on the set of stable matchings,
the supremum is a stable matching itself and that the supremum with
respect to one side of the market coincides with the infimum with respect
to the other side. We then deduce that the sets of mates an agent gets at
two different stable matchings are always such that he prefers any mate
in one but not the other to every mate in the latter but not the former.
Whatever his (responsive) preferences on subsets of agents are, every agent
thus has a complete strict preference order on the set of stable matchings.
That the stable set is a complete distributive lattice (under responsive
preferences) and other properties then follow as corollaries. Let us mention
that, as shown by Blair [1], stable many-to-many matchings do in fact
form a complete lattice for the broader class of substitutable preferences,
but then some other properties that obtain for responsive preferences no
longer hold, in particular, the lattice is not distributive.
The similarity in structure between many-to-many and one-to-one
matching is all the more interesting in view of the fundamental differences
that abound in strategic and other properties associated. We provide sim-
ple examples to show that the polygamous-side optimal stable matching
mechanism is manipulable by the polygamous agents themselves, that a
blocking pair of agents may each be worse off in every stable matching
than in the one they are blocking, that a polygamous agent may be hurt in
consequence of being more preferred than before by an opposite-side agent
- none of which can occur in one-to-one matching - and that an agent
who gets his most preferred mate in one stable matching may no longer do
so when agents are allowed to have one more mate each. We provide in the
Appendix a statement of the many-to-many version of the Gale-Shapley
Deferred Acceptance Procedure to which we refer in the examples.
31
2 Stable many-to-many matchings are a complete

distributive lattice
A many-to-many matching market M = (lw, W, k, >-) consists of two finite
sets lvI, W of agents, say men and women, where each man m has a (positive
integer) quota km and a complete strict preference order >-m on W U {0},
each woman w has a quota kw and a complete strict preference order >-w on
Mu {0}, and 0 denotes a non-agent being matched with whom will stand
for being unmatched. Call an agent, say w E W, acceptable for m E A1,
if w >-m 0. We assume wlog throughout that a woman w is acceptable for
man m if and only if m is acceptable for w.
A matching is an assignment of at most km distinct acceptable women
to each man, and at most kw distinct acceptable men to each woman, such
that woman w is a mate (assignee) of m if and only if m is a mate of w.
Formally, a" matching is a mapping J.L from M to 2w such that
1J.L(m) I ~ km for every m E lvI,
I
IJ.L-l(w) = I{m E lvI I wE J.L(m)} I ~ kw for every wE W,
and w E J.L( m) only if w is acceptable for m. We let J.L stand for the inverse
matching J.L- l as well, and alternatively write mw E J.L whenever w E J.L(m)
for any pair mw E j\l[ x W.
Let M = (.lld, l-V, k, >-) be a matching market. For any agent, say, man
m and for any subset 8 of women, let xm(8) stand for the km'th most
preferable woman of m in 8 when 181 ~ km and let xm(8) stand for the
non-agent 0 when 181 < km. In case J.L is a matching, we write xm(J.L) for
Xm (J.L( m)). Our first definition is standard.
Definition 1 A matching J.L is said to be
(i) blocked by a pair mw if mw 1 J.L but w >-m xm(J.L) and m >-w xw{J.L),
(ii) stable if it is not blocked by any pair mw . •
We now introduce the main definition of the paper.
Definition 2 Let J.Ll' J.L2 be any two matchings.
Define the male supremum J.L = J.Ll V M J.L2 by letting J.L(m) = 'Hm(8) for
every m, where 8 = J.Ll (m) U J.L2 (m), and the set 'Hm (8) consists of the
highest km women in 8 in the order >-m of m when 181 ~ km and is equal
to 8 itself when 181 < km.
Define the male infimum J.L = J.Ll 1\ AI J.L2 by letting J.L( m) = /-Ll (m) n
.e
J.L2(m)) U m(8) for every m, where 8 = /-Ll (m) U J.L2(m)"'-(J.Ll (m) n /-L2(m)) ,
and the set .em (8) consists of the lowest Im = km - I/-Ll (m) n /-L2 (m) I women
in 8 in the order >-m of m when 181 ~ Im and is equal to 8 itself when
181 < lm. (Define female supremum and infimum analogously.) •
32
Notation In any context where two women w, w' need not be distinct, we
shall sometimes use the symbol tm for the preferences of a man m. Thus
w tm w' and w i- w' implies w )-m w'.
Proposition 1 The supremum of any two stable matchings is a stable

matching.
Proof. Let 1£1, 1£2 be any two stable matchings and consider (say) the male
supremum 1£ = 1£1 VM 1£2'
To see that 1£ is a matching, assume the contrary that IJ.t(w)1 > kw for
some woman w. Then the set
is nonempty. Let m be the most preferable man for w in S. The pair mw

belongs to either 1£1 or 1£2 but not both. Say mw E 1£2"1£1' Note that
for otherwise S ~ J.t(w) ~ J.t1(W) U J.t2(W) ~ J.t2(W) implying J.t(w) = S U

(1£1 (w) n 1£2 (w)) ~ 1£2 (w) contradicting the fact that 11£2 (w) 1 ~ kw while
11£(w) 1 > kw by assumption. Then w prefers m to at least one man in
J.t1(W) and so m )-w x w(J.t1)' Since mw (~ 1£1) does not block 1£1, then
x m (J.t1) )-m w. But then IJ.t1(m)1 = km and by definition of the supremum
w
notinJ.t( m). Contradiction.
To see that 1£ is stable, assume the contrary and pick any (blocking) pair
mw ~ JL such that
Note first that mw ~ JL1 U 1£2 for otherwise by our assumption mw would
belong to the supremum 1£. Now wlog say x w(J.t2) tw x w(J.t1)· Since J.t(w) ~
1£1 (w) U 1£2 (w), trivially Xw (1£) t Xw (1£1) so by assumption m )- w Xw (1£1)'
Since mw (~ 1£1) does not block 1£1, then x m (J.t1) )-m w, and therefore by
assumption again Xm (J.t1) )-m xm(J.t). On the other hand, x m (J.t1) )-m w
also gives 11£1 (m) 1= km which implies Xm (1£) tm Xm (1£1)' Contradiction. •
Call an agent, say man m, exposed in a matching 1£ if IJ.t(m)1 < km. The
Proposition below which is of interest itself is frequently used in the proofs
to come.
Proposition 2 (i) If an agent is exposed in a stable matching then he has

the same mates in every stable matching.
(ii) If an agent has an exposed mate in a stable matching then he has
that mate in every stable matching.
33
Proof. (ii) follows from (i). To prove (i), suppose to the contrary that
J.t' (m') -:F J.t( m') for two stable matchings J.t, J.t' and a man m' who is exposed
in J.t. It is straightforward to see that then there exists a woman w who has
at least one less mate in J.t than she has in the supremum Ji = J.t VM J.t'.
(See that IJi(m) 1 ~ 1J.t(m) 1 for all m and IJi(m') 1 > 1J.t(m') 1 in particular.
Using the equality Em 1Ji(m) 1 = Ew IJi(w) 1 which holds by definition for
any matching Ji, then Ew IJi(w)1 = Em IJi(m)1 > Em 1J.t(m) 1 = Ew 1J.t(w) 1
and so indeed IJi(w) 1 > 1J.t(w) 1for some woman w.)
Now take any m E Ji(w)~(w) and note m ~w 0 =xw(J.t). On the other
hand, w E Ji(m)"-J.t(m) which by definition of the supremum implies w ~m
xm(J.t). Thus mw blocks J.t. Contradiction. •
Proposition 3 The male supremum and the female infimum of any two
stable matchings are identical.
Proof. Consider any two stable matchings J.t1, J.t2 and let!!:.. = J.t1 /\w J.t2, Ji =
J.t1 V M J.t2·
We first show that Ji ~ J.t. Take any mw E Ji, i.e., w E Ji(m). We will
show that m E J.t(w). If wE J.t1(m) n J.t2(m) then already by definition of
the infimum mw
E !!:... So wlog say
wE J.t2(m)"-J.t1(m), i.e., mE J.t2(W)"-J.t1(W).

Note that 1J.t1(m)1 = 1J.t2(m) 1 = km (for otherwise J.t1(m) = J.t2(m) by
Proposition 2(i).) Then
w ~m X m(J.t1)
(for otherwise X m(J.t1) ~m W contradicting w E Ji(m).) Since mw (rj. J.t1)
does not block J.t1, then
X w(J.t1) ~w m.
So, if contrary to what we want to show, m does not belong to J.t(w), then
I
neither does any man in J.t1(w)"-J.t2(w), but then I!!:..(w) < 1J.t2(w) 1 ~ kw,
which implies w is exposed (because, for w unexposed, 1J.t1(W) 1= 1J.t2(w) 1=
kw hence by definition of the infimum 1J.t(w) 1= kw.) By Proposition 2(i),
then, J.t1 (w) = J.t2 (w) contradicting m 1.-J.t1 (w). Therefore m does belong
to J.t( w) and so Ji ~ J.t.
I
Now suppose Ji ~-!!:... Then IJi(w) 1 < I!!:..(w) ~ kw for some woman w
who (thus) is exposed. By Proposition 2(i) again, J.t1(W) = J.t2(W) and so
!!:..(w) = Ji(w). Contradiction. Thus Ji =!!:... •
Proposition 4 Given any two stable matchings J.t1, J.t2 and any agent (say)
m, let 8 i = J.ti(m)"-(J.t1 (m) n J.t2(m)), i = 1,2. Then either m prefers any
woman in 8 1 to every woman in 82 or m prefers any woman in 82 to every
woman in 8 1.
Proof. Take any two stable matchings J.t1, J.t2 and any man m such that
J.t1(m) -:F J.t2(m) (otherwise, by Proposition 2(i), there is nothing to prove).
34
By Proposition 2(i), m is unexposed and so 8 i i= 0, i = 1,2. Denote wt

and wi respectively the most and the least preferred woman for m in 8 i ,
i = 1, 2. Say wlog that m prefers wi to w2. We show below that m prefers
wi to wt which proves the Proposition:
Suppose to the contrary that m prefers wt to wi. Recalling m is unex-
posed and the definition of infimum, then mwi belongs to JLII\M JL2' So by
Proposition 3, mwi belongs to JLI Vw JL2' Recall m ~ JL2(wi) and check
from Proposition 2( ii) that wi is unexposed. By definition of the supre-
mum then wi prefers m to x w 1- (JL2)' Since m prefers wi to wi hence
(trivially) to X m (JL2), then mwi blocks JL2' Contradiction. •
We shall now consider matching markets M = (M, W, k,~) where pref-
erences ~ are over subsets of agents. For any agent, say m, we denote ~m
the preferences of m. For any subsets 8, 8' of women, we write 8 ~m 8' to
mean "m finds 8 at least as good as 8'" and write 8 ~m 8' to mean "m
prefers 8 to 8' strictly unless 8 = 8'." Preferences ~m are said to be re-
sponsive if m has a complete strict preference order >-m on the set of women
W such that, for any subsets 8,8' of women when 8 = 8' "'- {w'} u {w}
and w' E 8',
8 ~m 8' if and only if w >-m w'.
We shall refer to >- as the preferences that underlie ~.
Stability with respect to ~ is defined in the natural way: Say that a pair
mw blocks a matching JL if mw ~ JL and m prefers to JL(m) some subset
of JL( m) U {w} with cardinality at most km while w prefers to JL( w) some
subset of /-l(w) U {m} with cardinality at most kw' A matching is stable if
it is blocked by no pair of agents. One sees immediately that a matching
is stable with respect to the (responsive) preferences ~ if and only if it is
stable with respect to the preferences >- that underlie ~.
Corollary 5 Every agent has a complete preference order on the set of
stable matchings.
Proof. Follows immediately from Proposition 4 . •

It is interesting to note that Corollary 6 below is all that corresponds to
the so-called Decomposition Lemma for one-to-one matching which starts
off the proofs of many of the results in that domain.
Corollary 6 If a man and a woman prefer the same matching between any
two stable matchings, then they are matched to each other either in both or
in neither of the two matchings.
Proof. Suppose JLI' JL2 are both stable and the pair mw both prefer JLI to
JL2' Then mw does not belong to JL2 "'-JLI for otherwise, by Propositions 4
and 3, mw E JLI I\w JL2 = JLI VM JL2 contradicting the assumptionthat m
prefers JLl to JL2' On the other hand, neither does mw belong to JLI "'-JL2 for
otherwise, by Proposition 4, mw would block JL2' •
35
Now, to bring our results together, note that, if we define the supremum
of any two matchings as that matching where each agent is assigned his
more preferred set of mates between the two matchings, then by Proposi-
tion 4 and Corollary 5 the supremum so-defined coincides with the supre-
mum given in Definition 2 and so, by Proposition 3, is stable itself. The
same hold upon replacing supremum with infimum and more with less
in the previous sentence. The following Corollary then speaks for either
definition of supremum and infimum and is easily checked to follow from
Propositions 1,3,4 and Corollary 5. (Definition: The lattice is distributive
if J.Ll VM (J.L2 AM J.L3) = (J.Ll VM J.L2) AM (J.Ll VM J.L2)·)
Corollary 7 The set of stable matchings is a complete and distributive

lattice under the operations of supremum and infimum. The supremum with
respect to one side of the market coincides with the infimum with respect
to the other side. •
3 Monotonicity and strategyproofness

Our first example shows that an individual may only suffer from a change
in preferences which might at first sight appear to be beneficial for him.
A change thus occurs in the preferences of a woman, in Example 1 below,
such that she reverses her preferences between two men (all else remains
the same) and this change only hurts the now favorite man, so long as
stability is the rule.
Example 1. Monotonicity with respect to opposite-side prefer-

ences: Consider the matching market with two men and three women
whose preferences are given below, where the first number in each cell is
the rank of the associated woman in the preference order of the man and
the second number is the rank of the man in the preference order of the
woman. Thus, m ranks w' second and w' ranks m first. Suppose all quotas
are equal to 1 except only that m' has a quota of 2.
w w' w"
m 1,1 2,1 3,1
m' 2,2 1,2 3,2
Then there exists a unique stable matching J.L where m gets w and m'
gets w' w". (Check by the Deferred Acceptance Procedure that J.L is both
the male optimal and the female optimal stable matching.)
Suppose w who ranks m first and m' second now is to rank m' first and
m second, while all other preferences remain the same, as stated below:
36
w w' w"
m 1,2 2,1 3,1
m' 2,1 1,2 3,2
In the unique stable matching that exists now, m gets w' and m' gets
ww". Note that m', who prefers w' to w, is worse off.
Remark Example 1 shows in particular that the mono tonicity property
which the optimal stable matching is known to have (Gale and Sotomayor
[4]), namely that if an individual is placed higher than before in the pref-
erence order of an opposite-side agent then that individual is never worse
off at the stable matching which is optimal for his side, does not hold for
polygamous agents. •
Our next example shows that the stable set fails in being "monotonic"
with respect to quota expansions.
Example 2. Monotonicity with respect to quota expansions: Con-

sider the market
w w'
m 1,1 2,1
m' 2,2 1,2
See simply that m' gets his most preferred woman w when all quotas
equal 1, but that he gets no one when the quota of m (alone) increases to
2.
A similar loss may occur even when all quotas increase simultaneously
as in the case of the following market:
w w' w"
m 1,3 2,3 3,3
m' 2,1 1,1 3,2
m" 2,2 3,2 1,1
Check that, in the unique stable matching when all quotas equal 1, m
gets his most preferred woman w, while in the unique stable matching when
all quotas equal 2, m loses wand gets w' w".
Example 3. Optimality, Strong Stability, Strategyproofness: Con-

sider the market below with the same agents and preferences in Example
1 except only that m now has quota equal to 2 while m' has quota 1:
37
w w' w"
m 1,2 2,1 3,1
m' 2,1 1,2 3,2
Check that the matching J.L where m gets w' w" and m' gets w is the only
stable matching. Observe that
(i) the pair mw' blocks the alternate matching J.L' where m gets ww" and
m' gets w', and that
(ii) both m, m' and (the blocking woman) w' prefer J.L' to J.L (as each
thereby replaces his/her second choice with his/her first choice).
We thus reach the following two conclusions:
Fact 1. An optimal stable matching may be strongly inefficient for the
agents for whom it is optimal. This cannot happen for single-quota agents
as stated in the so-called Weak Pareto Optimality Theorem (Roth and
Sotomayor [8]).
Fact 2. There may exist a blocking pair of agents who are each strictly
worse off in every stable matching. That this cannot happen in one-to-one
matching is given by the so-called Strong Stability Theorem in Demange,
Gale and Sotomayor [2].
Observe further that, if m (alone) were to reverse his preferences between
w' and w", as stated below, then m gets ww" and is better off (as is m'
who gets w') at the new male optimal stable matching:
w w' w"
m 1,2 3,1 2,1
m' 2,1 1,2 3,2
We thus obtain
Fact 3 The optimal stable matching mechanism is manipulable by those
agents for which it is optimal. That this cannot happen for single-quota
agents is established by the Strategyproofness Theorem (Dubins and Freed-
man [3], Roth [6]).
Remark All the conclusions reached in Example 3 have already been noted
by Roth [6]. We have chosen to state it here primarily for the sake of
completeness and secondarily for the reason that it is simpler than the
example in [6] which involves 3 agents on one side and 4 on the other (see,
e.g., Theorem 5.10 in Roth and Sotomayor [8]) . •
Appendix
We give here a statement of the Deferred Accepted Procedure which is just
the natural extension ofthe Gale-Shapley Procedure ([5]) to many-to-many
38
matching and which terminates at optimal stable matchings. The reader

may find it useful in checking our assertions in the examples in Section 3.
Deferred Acceptance Procedure Initially, each agent on one side of the
market, say each man, proposes to his most preferred acceptable women,
no more than his quota in number. Then, each woman accepts her most
preferred acceptable men, no more than her quota in number, among those
who have proposed to her, and rejects everyone of those whom she does not
accept. In each round thereafter, each man proposes to his most preferred
acceptable women, no more than his quota in number, among those whom
he has not previously proposed to and been rejected by, and then each
woman accepts/rejects as above.
Proposition 8 Deferred Acceptance Procedure stops at the optimal stable
matching.
Proof. The argument is a reproduction of the proof of Gale and Shapley [5]:
Suppose the contrary and let m be the first man throughout the Procedure
who is rejected by an achievable woman (say) w, that is to say a woman
with whom he is matched under some stable matching, say J.L. Then, by
assumption, none of the (achievable) women with whom m is matched
in J.L has (in the Procedure) rejected m before he was rejected by w. In
particular, there must be one among them, say w', whom he had until then
not proposed to and who must therefore be lower than w in his preference
order. Thus m prefers w to w' therefore also to his least preferred woman
in J.L. On the other hand, w must have preferred to m each of a quota-full
of men whom she had when she rejected m, at least one of whom (say) m'
is not her mate in J.L. But then m' w blocks J.L. •
References
[1] Blair, C.: The lattice structure of the set of stable matchings with mul-
tiple partners. Mathematics of Operations Research 13, 619-28 (1988)
[2] Demange, G., Gale, D., Sotomayor, M.: A further note on the stable
matching problem. Discrete Applied Mathematics 16, 217-22 (1987)
[3] Dubins, L.E., Freedman, D.A.: Machiavelli and the Gale-Shapley algo-
rithm. American Mathematical Monthly 88, 485-94 (1981)
[4] Gale, D., Sotomayor, M.: Some remarks on the stable matching prob-
lem. Discrete Applied Mathematics 11, 223-32 (1985)
[5] Gale, D., Shapley, L.: College admissions and the stability of marriage.
American Mathematical Monthly 69, 9-15 (1962)
[6] Roth, A.E., The college admissions problem is not equivalent to the
marriage problem. Journal of Economic Theory 36, 277-88 (1985)
39
[7] Roth, A.E., Sotomayor, M.: The college admissions problem revisited.
Econometrica 57, 559-70 (1989)
[8] Roth, A.E., Sotomayor, M.: Two-sided Matching. Cambridge: Cam-
bridge University Press (1990)
Cost uncertainty, taxation, and irreversible
investment *
Sumru Altug 1 , Fanny s. Demers2, and Michel Demers2

1 Kor; University, Istinye 80860, Istanbul, Turkey and Centre for Economic
Policy Research, London, England
2 Carleton University, Colonel By Drive, Ottawa, Ontario KlS 5B6, Canada
Abstract. We examine the impact of learning about the unknown costs

of investment on irreversible investment decisions, and show that the pres-
ence of learning increases the endogenous cost of adjustment and depresses
investment. We demonstrate convergence of the state of information and
capital stock to the ergodic set. Once learning is complete, in contrast to
the exogenous cost-of-adjustment model, a mean-preserving increase in risk
raises the endogenous marginal adjustment cost, reducing investment and
the steady-state capital stock.
We use data on the U.S. economy to study the impact of uncertainty
and risk in the determinants of the costs of investing. Among our salient
findings is that increases in uncertainty have a much larger impact quan-
titatively on investment than increases in risk. Thus, if firms are unsure
about various aspects of the stochastic environment that they face, the re-
duction in investment is much larger compared to the case in which there
are increases in the riskiness in the price of capital or other determinants
of the costs of investing.
JEL Classification Number: E22
Keywords: Irreversible investment, costs of adjustment, learning, risk
1 Introduction
The literature on investment distinguishes between the actual and the de-
sired (or optimal) capital stock, where the latter is determined by factors
such as output and input prices, technology and interest rates, among oth-
ers. A prevalent explanation of the gradual adjustment of the capital stock
"The first draft of this paper was written while Fanny and Michel Demers were visiting
scholars at the CEPREMAP. Accordingly, they would like to thank their colleagues at
the CEPREMAP for their hospitality, and are grateful to Jean-Pascal B(massy, Pierre-
Yves Renin, Pierre Malgrange, and seminar participants for helpful comment.s.
42
to the desired level is the hypothesis of convex costs of adjustment (see

Eisner and Strotz, 1963). Abel (1979) and Hayashi (1982) show that the
standard adjustment cost model leads to a Q-theory of investment under
perfect competition and a constant returns to scale production technology.
An alternative explanation assumes that it takes time to build productive
capacity (see Kydland and Prescott, 1982; Altug, 1989, 1993). A number of
authors have recently emphasized irreversibility and uncertainty as impor-
tant factors underlying the gradual adjustment of the capital stock; see, for
example, Nickell (1977, 1978), Demers (1985, 1991), Bean (1989), Bertola
(1989), Pindyck (1988), and Bertola and Caballero (1994).
The effects of increases in risk on current investment is a topic that
has been discussed extensively in the investment literature. A number of
papers have studied the effect of output price risk on investment in the cost
of adjustment and irreversible investment models. Hartman (1972, 1973),
Pindyck (1982), and Abel (1983, 1985) provide analyses of this issue in the
cost of adjustment model, and conclude that for competitive firms operating
under constant returns to scale, greater risk in the output price and in the
prices of variable factors increases investment. If the production function
exhibits nonconstant returns to scale, however, the effect of an increase in
risk is ambiguous. In terms of the irreversible investment model, Demers
(1985, 1991) show how output price uncertainty reduces the investment of
a Bayesian firm, while working with Brownian motions without learning,
Bertola (1989), Caballero (1991), and Pindyck (1988) show that the impact
of output price risk on investment is negative. Pindyck (1993) and Caballero
and Pindyck (1992) analyze a model of the market with entry and exit and
show that even under perfect competition, risk lowers investment.
By contrast, there are very few analytical results with respect to the price
of capital equipment. For the adjustment cost model, Hartman concludes
that a mean-preserving spread has no impact while Pindyck (1982) and
Abel (1983, 1985) examine randomness in the cost of adjustment function,
and show that if the third derivative of the cost of adjustment function is
positive, investment increases as a result of greater risk. Yet, this topic is
particularly important in view of Jones' (1994) finding of a strong negative
relationship between growth and machinery price.
In this paper, we consider the profit-maximization problem of a risk
neutral monopolistically competitive firm with irreversible investment and
learning about the unknown costs of investment. While our framework is
similar to Demers (1991), we assume that the cost of investing may be
imperfectly known due to uncertainty about the investment tax credit or
about the price of machinery. Taxation is introduced in the manner of Hall
and Jorgenson (1967), Abel (1982), and Hayashi (1982).
In our model, the firm is uncertain about the permanent component of
the purchase price of capital equipment or the investment tax credit. It
uses noisy observations on prices and tax rates, as well as observations on
other informative variables, to make inferences about the permanent values
43
using a Bayesian updating rule. Following the approach in Demers (1991),

we show that uncertainty about the future costs of investment unambigu-
ously lowers investment. Uncertainty and irreversibility lead to a marginal
adjustment cost which arises endogenously with the learning process of
the firm. The prospect of obtaining better future information increases the
marginal adjustment cost and depresses current investment. Next, we prove
the convergence of the firm's state of information and of its capital stock,
which constitute a joint Markov process, to the steady-state. Whereas in
Demers (1991) the steady-state capital stock is a singleton, in our paper,
the desired capital stock is an ergodic set.
We also examine the impact of greater cost riskiness in the purchase
price of capital and the investment tax credit. Increases in risk are defined
in the sense of first-order stochastic dominance as well as in the sense of a
mean-preserving spread. When the shocks are independently distributed,
a mean-preserving increase in the riskiness of the cost of investing raises
the endogenous marginal adjustment cost, thereby reducing investment and
lowering the steady-state capital stock. This is in contrast to the results
obtained for the exogenous cost-of-adjustment model. In the serially cor-
related case, we provide some mild conditions under which a MPS for the
price of capital reduces investment and the steady-state capital stock.
To determine the quantitative significance of these effects, we present
simulation results obtained by numerically solving the model. These simu-
lations use aggregate and manufacturing data for the U.S. as well as infor-
mation on the determinants of the cost of capital contained in Jorgenson
and Sullivan (1981), and Jorgenson and Yun (1991). The numerical so-
lutions of the model can be used to examine the time path of investment
with learning by the firm when the riskiness of its environment is changing.
By contrast, analytical determination of the effects of such factors requires
that the impact of uncertainty and learning be considered separately from
the impact of the increase in risk. Such numerical solutions can also be
used to study the effects of increases in risk under more general conditions
than those required to derive analytical results.
Section 2 describes the irreversibility model, Section 3 analyzes the im-
plications of uncertainty and Section 4 establishes convergence. Section 5
investigates the impact of greater risk. Section 6 describes the simulation
procedure while Section 7 presents the simulation results. All proofs are
contained in the appendix.
2 The model
We consider a monopolistically competitive risk neutral firm. Each period,
it makes variable input and investment decisions. At time t it produces
output Vi using capital Kt (which is predetermined at t), and variable
inputs Lt. The firm's production function F (Kt, Lt) is twice continuously
differentiable, increasing, concave, and satisfies the Inada conditions.
44
Let Pt denote the output price. We assume a constant elasticity demand

function. 1 The inverse market demand function is given by
(1)
where e < -1 is the price elasticity of demand, and at is a known parameter
representing the state of demand. Denote by Wt the nonstochastic variable
input price vector.
The optimal choice of variable factors involves static optimization un-
der certainty. We can define the short-run profit function at time t,
n (Kt, at, wd, as
= max{ptF (Kt, Lt) - Wt . Lt}
Lt>O
= a-;1/e F (Kt, L* (Kt, at, Wt))(!+e)/e (2)
-Wt· L* (Kt, at, Wt)
where L* (Kt, at, Wt) denotes the optimal choice of variable factors. The
short-run profit function n (Kt, at, wt} is continuous in Kt, at, and Wt,
increasing in Kt and at, decreasing in Wt, and strictly concave in Kt. We
assume that n (Kt, at, Wt) is bounded for finite Kt, at, and Wt. The firm's
after-tax cash flow at time t, Rt , is defined as
T
Rt = (1 - Tt) n (Kt, at, Wt) + Tt L Dx,t-xptx1t-x - (1 - 'Yt) P: It, (3)
x=1
where It is the firm's rate of gross investment measured in physical units,

and pf the purchase price of investment goods, Tt is the corporate tax
rate at time t, 'Yt is the investment tax credit at time t as a percentage of
the price of the investment good, Dx,t-x is the depreciation allowance per
dollar invested for tax purposes for capital equipment of age x on the basis
ofthe tax law effective at time t-x, and T is the life of the equipment. Let r
denote the real rate of interest. For future reference, define Zt as the present
value of tax deductions on new investment and pI
as the tax-adjusted price
of investment goods, where
T
Zt = LTHnDn,t (1 + r)-n (4)
n=1
and
(5)
We will assume that n
= fi + Ukt where pk is an unknown parameter
representing the permanent component of n,
and Ukt is a random variable
lOur theoretical results do not depend on this assumption. We only use this specifi-
cation of the demand function in order to perform our quantitative analysis.
45
with a known distribution function. In later sections, we consider the case

when Ukt is independently and identically distributed for all t, and when it
is serially correlated. As a result, we can write p{ = pI + UIt, where p{ =
(1 - 'Yt - Zt) P:is the permanent component of the tax-adjusted purchase
price and UIt = (1 - 'Yt - zt) Ukt is the transitory component. 2
Let 8, 0 < 8 < 1, be the deterministic depreciation rate. The law of
motion of the capital stock is
(6)
We assume that investment is irreversible;
(7)
Now, turning to the informational structure of the model, we note that

the firm knows the possible states of prices but does not know with certainty
which one is the true state. Let P == {p~, ... ,p~} be the set of possible
states. The firm has a prior probability distribution about the true state
pk denoted by \(10, where \(10 = ["p~, ... ,,,p~l,,,p? > O,i = 1, ... ,n, and
L~=l "p? = 1; "p? is the prior probability that pk = pf, i = 1, ... ,n.
Each period the firm observes the realization of ~ and the realization of
a non-price signal bt which is informative about pk and which may consist
of various pieces of information such as the monetary growth rate, financial
surveys, etc .. We denote by H the set of all possible signals ht == (P:, bt ).
Letting /j(Xt I pn,j = 1,2, be the density of Xt, conditional on pf being
the true state, Xt =~, bt , we can define >"(ht I pf) == fI(P: I pf)h(bt I pf)
as the likelihood of observing ht given that the true state is pf. The firm
revises its prior probability distribution \(10 by applying Bayes' law. That
is,
\(It (h t ) = ["pi (h t ) , ... ,,,p~ (h t )] represents the firm's state of information

at time t. For all t, \(It E D(P) is the set of all probability distributions on
P. The evolution of the information state can be expressed as
(9)
where g denotes the law of motion of the information state and updating
according to Bayes' law. Note that g does not have a time subscript since
2 Alternatively, the tax credit could be stochastic and p~ could be known for all t; in
this case, 'it = 'Y+u1't, where u1't is a random variable drawn from a known distribution
pI
for all t. Thus, = (l-'Y-zt)p~ would be the permanent component of the tax-adjusted
price of capital, and UIt = -U1'tP~ would be the transitory component.
46
the revision process is time-invariant. Furthermore, the evolution of the

information state has the Markov property since the past information states
q;l, ... ,q;t-2 are irrelevant to the revision process at time t as long as
q;t-l is known. In other words, q;t-l completely describes the state of
information at time t - l.
The firm solves
max {(I - Tt) IT (Kl' al, Wl) - p{It + (10)

{ft}~l
~J3t-l it-l [(1 - Tt) IT (Kt, at, Wt) - p{It18(h~ lq;l)dhD
subject to (6), (7), (9), Kl and q;l given, where J3 = (1 + r)-l, 0 < J3 < 1 is
the discount factor and 8( MIq; l) is the predictive density of the sequence
of observations h~ defined as h~ == {h2' h3, ... , h t }. Thus, the firm chooses
a sequence of strategies for future investment plans {It} ~l which specify
future investment as a function of the state of information q;t and the
capital stock Kt to maximize the expected present value of future net cash
flows conditional on the state of information, subject to the irreversibility
constraint and the laws of motion for the capital stock and the state of
information, K 1 , q;l given. Using dynamic programming, we can express
the problem recursively as
V(Kt, 'lit) = max{(1- Tt} IT (Kt, at, Wt) - p{It +
i
It
J3 V(Kt+ b q;t+1 (ht+1))8(ht+l Iq;t)dht+l} (11)
subject to (6), (7), (9), Kt, q;t given.

Let VK denote the partial derivative of V with respect to K. The first-
order necessary and sufficient condition for the optimization problem at
time t is
=0 if 1;>0
:s; 0 if I; = O.
(12)
We can use the envelope theorem to find the partial derivative of the val-
uation function V (Kt+1 , q;t+l(ht+1)) with respect to Kt+1 for t = 0,1,2, ...
as
47
(1 - Tt+1) ITK (Kt+b O!Hb wt+d + (1 - 6) pI+1

if 1;+1> 0
(1 - Tt+1) ITK (Kt+1, O!Hb Wt+1) + (1 - 6){3x
IH VK((1 - 6)Kt+1,g(wt+1, ht+2»8(ht+2 I WH1 )dht+2
if 1;+1 = o.
More compactly,
+ (1 - 6) x
i
VK(Kt+1. wt+1 (ht+1» = (1 - Tt+d ITK (Kt+1, O!t+1, Wt+1)
min [p{+1,{3 VK ((1- 6) Kt+1,g (WH1, ht+2)) 8(ht+2 I wt+1)dhH2]
(13)
where VK(Kt+1, WH1 (h H1 )) is the shadow value of capital. Let Et denote
the expectation operator using the predictive density 8(ht+1 I wt). Also
assume that an interior solution obtains in period t. After substituting for
the shadow price and for {3, the first-order condition (12) for time t can be
rearranged as
(1- Tt+1)ITK (KH1.O!HbWt+1) = et; + (1- 6) {Etir{+1-
Et min [P{+1' (1 + r)-1 Et+1 VK ((1 - 6) Kt+1,g (w t+1, h..t+2))]}, (14)
where ITK is the partial derivative of IT with respect to K t+1, Et denotes
the expectation operator using the predictive density 8(ht+1 I wt), and
et; = p{ (r + 6)-(1 - 6) (Et p{+1-p{) is the firm's cost of capital in the sense
of Jorgenson (1963).3 That is, the cost of one unit of capital is expressed
as net of expected capital gains on the undepreciated portion of the unit.
If investment were reversible, equation (14) would reduce to
(1 - Tt+1)ITK (KHb O!t, wHd = et;,
which is the optimality equation for a firm purchasing or selling capital
services one period in advance before the uncertainty about the future
price of investment goods is resolved. Note that the choice of investment is
no longer a dynamic problem when investment is reversible.
The second term on the right-hand side of equation (14) is a risk pre-
mium that the firm requires for the loss of flexibility that it incurs since
it cannot disinvest. It represents a marginal cost arising endogenously due
to the irreversibility constraint, and varies through time with the firm's
state of information. This contrasts with the standard model where costs
of adjustment are exogenous and non-timevarying.
3See also Nickell (1978), chapters 2, 8 and 9.

48
3 The impact of uncertainty

At time t the firm chooses its optimal investment, It, on the basis of its
expectations of the unknown permanent price of investment goods. The
latter depend on the firm's subjective beliefs w~ch are updated after ob-
serving a particular realization h t of the signal ht . At the same time, the
firm anticipates receiving in future periods a sequence of messages hnl
whose realization is not yet known. The learning behaviour of the firm can
be likened to conducting an "experiment" which is "sufficient" or "more
informative" in the sense of Blackwell (1951, 1953). We can say that the
firm expects to perform an "experiment" which is "more informative" if,
for example, it expects the resolution in the near future of some current
political uncertainty or of uncertainty with respect to some aspects of mo~-
etary or fiscal policy. Accordin&-to Blackwell's definition, an experiment ht
is sufficient for the experiment ht if there exists a positive stochastic trans-
formation", ( ht, ht ) defined on H x jj such that
pk E P, ht EH, (15)
where -X(htJpk) == fl(pt Jp )h(btJp ) and iH ",(ht, ht)dht = 1 for all ht E H,

.-.-- -.. k-k . . . . . -k -.. ..... -..
and ",(ht,ht} does not depend on the unknown parameter pk. We will say
-00
that ht+l is m~re informative in the sense of Blackwell than hnl if at le~t
one element, hi, is more informative than hi, i = t + 1, t + 2, ... and if hj
is as informative as hj for all j =I=- i.
We examine the impact of varying the informativeness of the non-price
::::?O
signals bt+l' while keeping the riskiness faced by the firm constant. In
Section 7, we will consider the impact on the firm's investment of varying
-koo
the informativeness of the price signal Pt+! while also changing the riskiness
faced by the firm. The following lemma and theorem will permit us to obtain
the main results of this section.
Lemma 1 VK (Kt, 'lit) is concave in 'lit for all Kt E ~+.
Theorem 2 (Marschak and ~yasawa, De Groot) If the experiment ht+!

is sufficient for the experiment ht+! in the sense of Blackwell, then, for any
concave function cjJ defined on D [P] and any probability vector 'lit E D [P],
Proposition 3 The maryinal cost of adjustment increases with the antic-

ipation of receiving more informative signals in the sense of Blackwell.
49
Proposition 3 reveals that the anticipation of receiving infonnative

signals and of learning increases the endogenous marginal adjustment
cost. To see that lower current investment ensues, consider the follow-
ing. Let Et denote expectation with respect to the predictive density e
and Et expectation with respect to the predictive density S. Also, let
~-I
Ct; = pI (r+6) - (1-6)(EpH1 -p{) denote the Hall-Jorgenson cost of capital
when there is learning and more informative signals are anticipated. Denote
by I; and h the optimal investment levels for S and e,
respectively. We
have
~-I
Ct; + (1 - 6){ EtPt+1 -
~
Etmin [-I~t+1 VK((l- 6) 2Kt + (1- 6)I;,g('If
Pt+1,E ~H1 ,hH2))(1
;:::: + r)- 1] }
~ Ct + (1 - 6){Etp{+1 -
Et min [p{+1' EH1 VK ((1 - 6)2 Kt + (1 - 6)1; ,g('Ift+1, h H2 )) (1 + r)-1]}
= (1 - Tt+1)IIK((l - 6) Kt + I;, O!t+1, WH1),
where the inequality follows by Proposition 3 and the equality by the op-
timality of I; given the anticipation of more informative signals. Hence,
since V is concave in K, we have I; ~ h. We can state
Proposition 4 Denote by I; and h, respectively, the unique optimal ir-

reversible investment levels which solve (10) when the sequences hn1 and
-00 -00
ht+1 are anticipated. Ifh t+1 is more informative than hn1 in the sense of
Blackwell, then I; > h.
Proposition 4 indicates that the anticipation of ''more informative" sig-
nals in the future reduces the optimal (current) investment of the risk
neutral firm. Hence, a firm which anticipates to learn the permanent com-
ponent will only gradually adjust its capital stock to the steady-state.
4 Convergence to the desired stock of capital

The firm's desired stock of capital is the capital stock achieved by the
firm when learning is complete and the firm's state of information has
converged to a vector 'If* which assigns probability one to the true state
pk. In view of the irreducible purchase price risk Ult, the desired capital
stock is stochastic. Thus, the vector (K*, 'If*) is an element of the ergodic
set r == {(K, 'If) I K E [KL, KH] ,'If = 'If*}, where KL and KH solve
pI +UI = {3EVk [(1- 6)K + I(K, 'If * , pI +UI)] , (16)

pI + 1!I = {3EVk [(1- 6) K + I(K, 'If* ,pI + 1!I)] , (17)
50
where E denotes expectation with respect to knowledge of the true distri-

bution and where 1!:.J and '1h denote the lowest and highest values of the
random component of the investment price.
Let s = [K, W] denote the current state of the firm, with s E 8, where
8 == [0, KU] x D [P], and where 8 = B ([0, KU] x D [P]) is its a-algebra
and KU denotes an upperbound for K.4 Hence, (S, 8) is the state space. The
measurable space of events is (H, B (H)), where B (H) is the a -algebra
of H. Define a stochastic kernel Q : 8 x B (H) ---t [0,1] as Q (s, B) =
fB ( (h'ls) dh' where ((h'ls) is the predictive density of h' conditional on
the state s.5
The evolution of the state of the firm can be described as follows. If the
current state of the firm is s E 8, an event h E H is realized according
to the stochastic kernel Q(s, .). Next period's state is then determined by
the laws of motion for K and W'. Hence, we can express the dynamic
evolution of the firm's state as s' = tP (s, h')
where tP : 8 x H
---t 8. It can
be shown that tP (-, h') is continuous in s for all h' E H and that tP (s, .)
is measurable with respect to B(H) for all s E 8. As a result, we have
Lemma 5 Define tP- 1 (M)s == {h' EH I s' E M} ,M E 8, and P: 8 x

8 ---t (0,1) by
P(s,M) = iM ((tP- 1 (M)s Is) ds'. (18)
Then pes, M) is a transition probability on the state space (S, S).

Define C(8) as the set of all continuous and bounded 8-measurable real-
valued functions on 8. C(8) is a Banach space with the sup norm Iml ==
sUPsES Im (s)l· As a result of Lemma 5, a continuous linear transformation
T : C(S) ---t C(S) can be defined by (Tm)(s) = f m(x)P(s, dx). T is a
Markov operator associated with the transition probability P. The adjoint
T* ofthe Markov operator T is defined by (T*I1)(L) = f P(x, L)(dx) where
11 is a bounded countably additive set function. T* : ca(s) ---t ca(s) where
ca( s) is the Banach space of bounded countably additive set functions on
(8,8) and is a dual space to C(8). The space ca(s) has as its norm the total
variation norm, defined as 1111 = sup 2:~=1 III (Ni)1 where the supremum is
taken over all finite partitions of 8 into disjoint subsets {Ni}. If 11 (N) is
the probability that the firm's state is in the set N at time t, then (T*)(N)
is the probability that the state is in N at time t + 1.
4We simplify the notation and let the variables without time subscript (or time
superscript) such as K and W denote the current period's values and primes denote next
period's values.
5Define X(wt,B) == f B 8(ht+llw t )dht+l. The stochastic kernel Q(st,B) induces
the same measure on CH, B CH)) given St as X (wt, B) given wt. That is, Q (St, B) ==
X(wt,B).
51
The steady state of the stochastic dynamic process defined by T is an

invariant probability measure f-l* such that T* f-l* = f-l*.
Theorem 6 There exists a unique invariant probability measure f-l* satis-
fying T* f-l* = f-l* .
Theorem 7 The sequence U=~==-Ol T*if-l* In} converges weakly to the in-
variant probability measure f-l*. The converyence is uniform.
The last two theorems guarantee the existence of a unique steady state
and the convergence of the firm's initial information state and capital stock.
Thus, in the long run once learning is complete, and the true state pk is
known with certainty, the firm's capital stock reaches the desired level.
5 Increases in risk
In this section, we assume that pk is known, and we examine the impact
of greater risk on irreversible investment. We drop the assumption that :u:
takes values in a compact set and instead we assume that it takes values in
R+. Let G(P:H I P:) denote the (objective) distribution of P:+l conditional
on P:.We will be interested in examining the effect of distribution func-
tions which are stochastically larger in the sense of first-order stochastic
dominance (FSD)6 and riskier than G in the sense of a mean-preserving
spread (MPS) 7 •
-k
Definition 1 Pt+l with distribution function G(P:H I P:) is stochastically
~
larger than nHwith distribution function G(P:+l I pn

in the sense of FSD
if and only if G(P:+l I pn
~ G(P:+I I P:) for all P:H·
-k ~
Definition 2 PtH with distribution function G(P:+l I P:) is a MPS of
n+l with distribution function G(P:H I P:) if and only if
We can rewrite the shadow price of capital as
VK(Kt+1.P{H) = (1 - Tt+l) ilK (KtH' atH, Wt+l)
+ (1 - 8) min [P{+l',B J VK ((1 - 8) K t+1.P{+2) dG(p:+2 I P:H)] .(19)

We first consider the case where n is independently distributed.
6 SeeHadar and Russell (1971).
7See Rothschild and Stiglitz (1971).
52
Lemma 8 Assume that it:

and, therefore,
for all t. Then, VK is concave in P:.
n is independently distributed
Proposition 9 Assume that n
is independently distributed for all t. A
MPS in n+l increases the firm's marginal risk premium at time t and
reduces It.
The proof follows from Lemma 8. Hence, when investment is irreversible,

a riskier price of capital goods (in the sense of a MPS) depresses the ex-
pected future marginal value of capital and raises the marginal endogenous
n
adjustment cost, thereby reducing current investment. We now turn to the
general case where may be correlated through time.
Lemma 10 Assume that the following conditions hold:
(1) it: and, therefore, n is correlated through time for all t;

(2) G(P:+l I P: + €) dominates G(P:+1 I P:) according to FSD, i.e.
G(P:+1 I P: + €) ~ G(P:+l I P:) for all P:+l' whenever € ~ 0, for
all t.
Then VK is increasing in P: .
The second condition in Lemma 10 can be interpreted to mean that the
future resembles the present. The higher the current value of pf, the higher
is the probability of observing a high value of the purchase price of capital
goods next period.
Proposition 11 Suppose the assumptions of Lemma 10 hold. If G(pf+1 I

pf) dominates G(pf+l I pf) by FSD, then the marginal endogenous cost of
adjustment falls and investment increases under G.
The prooffollows from Lemma 10. Proposition 11 shows that ifthe future
resembles the present when prices are serially correlated, then a FSD shift in
G(P:+l I P:) which lowers the probability of low values of P:+l and indicates
the prospect of facing higher purchase prices for investment goods in the
future will lower the marginal endogenous cost and induce the firm to raise
current investment.
Lemma 12 Suppose that the assumptions of Lemma 10 hold and, in ad-

dition, assume
for € ~ 0, h ~ 0, for all t. Then VK is concave in P:.

53
The second and third conditions of Lemma 12 reveal that while larger
values of pf induce stochastically dominating shifts in the conditional dis-
tribution of ~+1' they do so at a nonincreasing rate. 8 For example, if
pf+1 = ppf + Uf+l' where 0 < p::; 1, which includes the first-order autore-
gressive process and the random walk, then the second and third conditions
are satisfied.
Proposition 13 Suppose that assumptions of Lemma 12 hold. An MPS
in ~+1 increases the marginal endogenous cost of adjustment and reduces
current investment.
The proof follows from Lemma 12.
Proposition 14 Suppose that the assumptions of Lemma 12 are satisfied
by the steady-state conditional distribution of~+l. An MPS in ~+1 implies
that the lower and upper bounds of the ergodic set of capital stocks fall,
leading to a lower capital stock in the steady-state.
Proposition 14 indicates that firms facing a more variable (in the sense of
MPS) distribution of the tax-adjusted price of investment goods will have
a lower capital stock in the steady-state.
6 Simulation procedure
In this section, we provide a quantitative characterization of the model with
irreversible investment and learning. This is achieved by numerically solving
for the optimal investment policy function under a variety of assumptions
about the stochastic environment facing firms. Section 6.1 describes the
numerical solution procedure while Section 6.2 describes how the parameter
values are determined.
6.1 Numerically solving for the optimal investment policy

function
For the purpose of the numerical results, we assume that the production
function has the Cobb-Douglas form
yt = F(Kt,L t ) = AKiL:-'1, (20)
where 0 < TJ < 1 and A is a scale parameter. Under these assumptions, the
firm's short-run profit function (which has been optimized over the choice
of the variable factors) is given by
IT (Kt, at, Wt) == VtKf a:-J.L,
8 This condition is reminiscent of an assumption imposed in the principal-agent liter-

ature: in order to satisfy the second-order condition in the context of moral-hazard, one
needs to impose that the cumulative distribution function of 9 conditional on the action
a, F(9Ia) be concave in a. See Holmstrom (1979).
54
where a = (1-1})(1 +e:) and b = 1-1}(1 +e:). Then Vt = AIL/'fIw;/b(N- a/ b _

N-c/bJ > 0, N = (1 + e:)c 1 (1 -1}) < 1, and 0 < JL = -1}(1 + e:)/(1-1}(1 +
e:)) < 1.
We solve for the optimal investment policy function using numerical dy-
namic programming with value iteration. (See Bertsekas, 1976.) To see how
value function iteration is implemented, for any V E C(8), define the map-
ping (TV)(Kt, wt) from the right-side of (11) subject to (6), (7), (9), Kt,
wt given. It is straightforward to show that the mapping T satisfies the
contraction property. (See, for example, Demers, 1991.) Thus, iterations of
the form TnvO, where VO E C(8), will converge to the true value function
V* as n goes to infinity. Since evaluation of the mapping that defines the
valuation function V involves maximization with respect to the level of
investment, the investment policy function is found as a by-product of de-
termining the function V. To account for the presence of the irreversibility
constraint, we convert the choice problem into one of choosing next peri-
od's capital stock Kt+l to satisfy the constraint Kt+! 2: (1 - 8)Kt . Since
both Kt and Kt+! must belong to the finite set of feasible points denoted
K, the elements of the set K are chosen such that (1- 8)K also belongs to
this set. 9
6.2 Parameter values

Table 1 shows the common set of parameter values that are used in the
simulations. The firm in our model is assumed to be monopolistically com-
petitive and to operate under constant returns to scale. These assumptions
are consistent with recent evidence obtained using sectoral U.S. data. For
example, Morrison (1992) and Roeger (1995) report evidence for small but
significant markups in U.S. manufacturing while Basu and Fernald (1997)
and Burnside (1996) have argued that returns to scale are approximately
constant at the 2-digit industry level.
The value of the elasticity of demand e: is determined using the markup
estimates reported by Morrison (1992) and Roeger (1995).1° Morrison's
(1992) estimates imply an average markup of 1.162 for U.S. manufacturing
9This is accomplished by choosing the elements of lC as Km-HI = (1 - 6)i/nK,

K are chosen so that the capital
j = 1, ... ,rn, where n is a positive integer, and rn and
stock grid covers the support of the stationary distribution for the capital stock. (See
Sargent, 1980.) Recall, however, that the lower and upper bounds of the ergodic set for
capital depend on knowledge about the true value function V. Hence, we solve for an
upper bound on the firms's capital stock denoted K max by assuming that the firm in
period 1 knows that the true state of cost is equal to the lowest value of t/ and the
purely random component of p{ always takes on its lowest value :!!{.
lOIn a static version of the firm's problem, the markup of price over marginal cost is
defined as MRKP = Pt/MCt, where MCt is the marginal cost of producing an additional
unit of output. Under profit maximization, marginal cost equals marginal revenue. Mak-
ing use of this fact and the form of the inverse demand function in equation (1), the
markup can be expressed as MRKP = e:/(1 + e:).
55
Table 1
Parameter Values
EO 'TJ (3 6 T Z W
-3.607 0.346 0.947 0.102 0.5043 0.3018 8.006
Table 2
Sample Properties of Exogenous Series
Sample Period Mean Std. Deviation

Wt 1960:1-1993:3 8.006 2.957
p~ 1947:1-1993:4 1.048 0.0577
z; 1948:1-1992:4 0.5985 0.0237
"It 1948: 1-1992:4 0.0178 0.0203
over the period 1960-1985 while Roeger's (1995) estimates imply an average
markup of 1.6054 using U.S. 2-digit manufacturing industries for the period
1953-1984. We use a simple average of Morrison's and Roeger's estimates,
and set the value of c equal to c = -3.607.
The value of the elasticity of output with respect to capital 'fJ is also
determined using the implications of the firm's profit maximization problem
under imperfect competition. According to this problem, 'fJ can be measured
by the share of capital in total factor costs, CK, which, in turn, equals
CK = {MRKP/<iJ)SK, where SK is the revenue share of capital, <iJ is the
degree of returns to scale and MRKP is the markup. Under constant returns
to scale, <iJ = 1. The direct measurement of the revenue share of capital SK
depends on the definition of the capital stock and of national income, with
estimates ranging from 0.25 to 0.43 (see Cooley, 1995, Chapters 1 and 6).
Since capital in our model corresponds more closely to a narrower definition
of capital such as the sum of producers' durable equipment and structures,
we set S K = 0.25 to obtain the value of 'fJ used in our study.
The values of the corporate income tax rate Tt, the depreciation rate
8, the present value of depreciation allowances Zt, and the investment tax
credit It are derived from Jorgenson and Sullivan (1981) and Jorgenson and
Yun (1991). In the simulations, we assume that Tt, Zt, and It are constant.
56
We set Tt equal to the sample average of T q reported by Jorgenson and

Yun, Table 3.1, where Tq is defined as Tq = Tt + T~ - Tt T~, and Tt and T~
denote the federal and state corporate tax rates. We calculate the annual
depreciation rate 0 as a weighted average of the average depreciation rates
on 27 categories of equipment and 23 categories of nonresidential structures
and other assets calculated by Jorgenson and Yun, 1991, Table 3.19. (See
also Jorgenson and Sullivan, 1981.) The constant value of z used in the
simulations is obtained as the product of the sample mean of z; from Table
2 and the corporate tax rate T.11
The discount factor (3 is determined to imply an annual real interest
rate of r = 5.6% so that (3 = 0.947. We measure the real wage rate as
the ratio of the seasonally adjusted series on the nominal wage, defined
as gross average hourly earnings of production or nonsupervisory workers
on private nonagricultural payrolls from the Establishment Survey, and
the deflator for gross domestic product. The (constant) value of w used
in the simulations is determined as the sample average of the real wage
series. The real purchase price of capital p:
is measured as the ratio of the
deflator for business fixed nonresidential investment from Table 7.1 of the
National Income and Product Accounts and the GDP deflator. The value
of the scale parameter A and the constant value of a merely affect the units
in which we measure the quantity and price of output. In what follows, we
set a equal to one, and assign an arbitrary value to A for the purpose of
the numerical calculations. Table 2 describes the statistical properties of
the time-varying variables.
7 Simulation results
This section derives the numerical results obtained by simulating the model
under alternative assumptions. The first two subsections consider the ver-
sion of the model without learning and show the effects of increases in risk
with i.i.d. and serially correlated shocks. Section 7.3 allows for learning
about the permanent component of the price of capital.
7.1 Independently and identically distributed shocks

In this specification, we assume that there is no learning and that the
investment tax credit is equal to its sample average reported in Table 2.
Thus, the firm knows the permanent component of the price of capital and
also observes the realization of the price of capital in period t. Although
it does not observe the future realizations of ~ at time t, it knows that
they are identically and independently distributed. For computational ease,
11 The present value of depreciation allowances and the effective investment tax credit
rate are calculated using unpublished data that were kindly provided to us by Dale
Jorgenson.
57
we assume that n= exp(pk + Ukt) is lognormally distributed so that

logcn) = pk + Ukt is governed by the normal distribution with mean pk
and variance O"~.
To determine the values of pk and O"~ used in the simulations, we use the
expressions for the mean and variance of a lognormally distributed random
variables together with the sample mean and variance of the price of capital
reported in Table 2. Since the observed price of capital pf is assumed to be
the realizations from a lognormally distributed random variable, we equate
its sample mean and variance to the expressions for its population mean
and variance, defined as exp(pk +O"V2) and exp[2(pk +0"~)1-exp[2pk +O"~],
respectively, to solve for pk and O"~. This yields the values pk = 0.0454 and
O"~ = 0.0033.
The solution of the model involves finding the optimal policy function for
all points in the discretized state space. In the absence of uncertainty and
learning, the state variables of the model are given by (Kt, p[). We discretize
the current price of investment by assuming that it takes on five values.
These values are obtained by evaluating the expression p[ = (1-,,(- Z )pf for
pf E {exp(pk - 20"k) , ... ,exp(pk + 20"k)}. We choose the capital stock grid
to include 100 points. As a result, there are 500 points in the discretized
state space. 12
When the shocks have a continuous distribution, it is necessary to ap-
proximate the expectation of the future valuation function appearing on
the right-side of (11) as part of the value function iteration used to obtain
the solution of the model. We use simple Monte Carlo integration for this
purpose as described, for example, by Keane and Wolpin (1994). Let vn
denote the n'th iterate of the value function. Notice that the expectation
on the right-side of (11) has the form
EV n +1 (Kt+l,ii{+1) = E {max [IJ(a, w, Kt+1)-

Kt+2
p{+1(Kt+2 - (1- 8)Kt+1) + {1EV n (Kt+2,p{+2)]},

where the expectation is now taken with respect to the true distribution of
P{+l (as opposed to the predictive distribution e as in equation 13)). Given
knowledge of EV n (Kt+2,p[+2) from the previous iteration, the Monte Carlo
simulation approach estimates EV n +1(Kt+1,p{+1) by simulating D draws
of the future price of capital p{+1 using the distribution of finding the n,
maximum of the expression in square brackets for each simulated value of
P{+l, and approximating the expectation by the average of the maximum
12 To obtain sufficient numerical variation in the capital grid, we scaled up all variables
by setting the scale parameter A = 4. The resulting value of the steady-state capital
stock is 0.3915. When calculating the maximum sustainable capital stock, the choice of
the lowest price of capital is somewhat arbitrary; if pi is evaluated at pk - lOuk, then
K maz = 1.062. We defined the capital grid on the interval [0.0554,1.1577].
58
over the D draws. We generate D = 150 draws from the normal distribution
with mean pk and variance O'~ to form the simulated average.
One characteristic of the model's solution is that investment is decreasing
in the quantity of capital as well as in the price of investment. At the
lowest price of investment, the firm only stops investing when its capital
stock exceeds the deterministic steady-state capital stock by 66%. At higher
values of p{, the irreversibility constraint becomes binding at smaller values
of the capital stock. For example, at the highest price of capital, the firm
stops investing in new capital when its capital stock is less than one-third of
the deterministic steady-state capital stock. Furthermore, the average level
of investment, conditional on investing, is also monotonically decreasing in
the price of capital.
As a way of describing the model's overall behavior, Table 3 reports the
number of times the irreversibility constraint is binding for the (K, p) pairs
in the entire state space and the average level of investment, conditional on
investment occurring. When the variance of the price of capital is equal to
the sample variance of n, that is, O'k = 0.0033, the irreversibility constraint
binds 189 times out of a total of 500 states, or 37.8% of the time, and the
average level of investment when investment is positive is 0.3919.
We can conduct a mean-preserving increase in risk for this specification
by increasing the variance of the transitory shock to the price of capital.
In part (l-ii) of Table 3, we report the consequences for investment be-
havior when O'~ is doubled. As Proposition 9 predicts, the incidence of the
binding irreversibility constraint increases and the average level of invest-
ment falls. It is interesting to note that while there is only a 2% increase
in the incidence of the binding irreversibility constraint, the average level
of investment falls by more than 10%.
7.2 Serially correlated shocks

Next, we turn to the case when the transitory component of the price of
capital is a serially correlated process. We consider the effects of a mean-
preserving increase in the riskiness of the price of capital when n follows
a first-order autoregressive process. Since p{ is equal to a constant times
the price of capital, p{ also follows a first-order autoregressive process.
For computational ease, we assume that log~) evolves as log(n+1) =
pk + plog(Pf) + Ukt, where 0 < p < 1 and Ukt rv N(O, O'~). The solution
procedure is the same as in Section 7.1, where we considered normal shocks
that were independently distributed. In this case, we need to account for
the serial correlation in log(n) by calculating the expectation of the future
valuation function conditional on each realization of the price of investment
in period t, that is, E [V(Kt +1,P{+l)lp{].
The results in Table 3, parts (2-i) and (2-ii), show the effect of an increase
in the variance of Ukt on investment. As in the serially independent case, a
mean-preserving increase in the riskiness of the price of capital leads to a
59
Table 3
Number Incidence Average Level
of States of It =0 of Investment
Independently and Identically Distributed Shocks
(I-i) a% = 0.0033 500 189 0.3919
(l-ii) a% = 0.0066 500 193 0.3521
Autoregressive Shocks
(2-i) a% = 0.0033 500 181 0.1934
(2-ii)a% = 0.0066 500 189 0.1793
greater incidence of the binding irreversibility constraint and a lower level

of investment. It is also interesting to compare the results in parts (I-i) and
(l-ii) with those in parts (2-i) and (2-ii), respectively. When the price of
capital is positively auto correlated, a low price of investment today signals
a low price of investment tomorrow so that firms find it optimal to wait.
As a consequence, the average level of investment is lower when the shocks
are serially correlated compared to the case in which they are independent.
7.3 Learning
As in Section 7.1, we assume that the price of capital n is lognormally
distributed so that log(n) = pk + Ukt follows a normal distribution with
mean pk and variance O'~. In this specification, the firm does not know with
certainty which value pk takes on. However, it has a prior probability dis-
tribution about the true state pk denoted by wo, where WO = [-,p~, ... ,-,p~j,
where -,p?r, is the prior probability that pk = p~, m = 1, ... , n. We assume
that the unknown permanent component of the logarithm of the price of
capital equals the mean of the logarithm of the actual price of capital and
values that are ±0.2 away from it. Thus, the set of possible values for pk is
defined as P == {p~, p~, pn = {-0.1546,0.0454,0.2454}. In the simulations,
we initially consider the case with O'~ = 0.0033.
The solution of the learning model is complicated by the fact that firms
form their expectation of the future by using the predictive density for
-k
PtH == 10g(nH)' conditional on the value of their information vector at
time t, which shows the posterior probabilities of the possible values of pk.
Using the notation of Section 2, this can be expressed as
n
e~H I wt) =
m=l
L f(f/;+l I p~, O'~)-,p;".
Thus, to simulate E[V(Kt+l' wt+dlwtj, we need to use draws from the
60
-le
predictive density of Pt+I. However, the predictive density is a mixture of
normals and is not normal. To do the simulations, we draw a total of D
normal random variables with mean zero and unit variance. We know that
-le
a fraction 1/1'!n of the draws on Pt+1 must come from the normal distribution
with mean p~ and variance O"~ for m = 1,2,3. Thus, for each value of 1/1'!n
at date t, we generate Dm = D . 1/1'!n draws on the future price of capital
for m = 1,2,3 and evaluate the price of investment for each of these draws
as P~.i = p~.i(1 - 'Y - z), i = 1, ... , Dm. Given knowledge of the value
function from the n'th iteration denoted by E [V n (Kt +2 , q,t+2)1q,t+I] , we
can calculate the maximum of the period t + 1 return for each value of P~.i
as
max {II(a, w, Kt+d - p~.i(Kt+2 - (1- 6)Kt+1)+

Kt+2
.8E[vn(Kt+2' q,t+ 2)Iq,t+1J}, i = 1, ... ,Dm.
Averaging the resulting maximum function over the total number of draws
Dm and taking the weighted average over m = 1,2,3 using the posterior
probabilities 1/1'!n yields the simulated value for E[vn+1(Kt+I. q,t+1)Iq,t]
that is used at the next value function iteration.
As before, we discretize the state space by assuming that the price of
investment takes on five values. Since the posterior probabilities of the true
state pf for i = 1,2,3 at any date depend on the realization of the price of
capital at that date, the posterior probabilities must also be evaluated for
the five possible values of log(K) at each date. AB before, we draw a total
of D = 150 draws to calculate E[V(Kt+I, q,t+1)Iq,tj at each iteration and
choose the capital stock grid to include 100 points.
Table 4 reports the results with learning. In parts (i) and (ii), we consider
a uniform prior distribution that places probability one-third on each value
of ple, and an informative prior that places 0.8 probability on p~ = 0.0454
and 0.1 probability on values that ±0.2 away from it, respectively. With
a uniform prior, the irreversibility constraint binds 351 times out of 500
states or 70.2% of the time and the average level of investment is reduced
to 0.123. Compared to the case without uncertainty and learning reported
in part (I-i) of Table 3, this is an 85.7% increase in the incidence of the
binding irreversibility constraint and a 68.6% reduction in the average level
of investment.
The results in Table 4 show that the impact of uncertainty about the
permanent component to the price of capital is to induce a dramatic decline
in investment expenditures and a much greater frequency with which firms
prefer to delay current investment. This occurs even if the transitory shocks
to the price of capital are independently and identically distributed with
the same variance as in part (I-i) of Table 3. Even with an informative
prior, the decline in total investment is still very substantial compared
to part (I-i) of Table 3. (See part (ii) of Table 4.) These results indicate
61
Table 4
Number Incidence A verage Level
Specification of States of It = 0 of Investment
(i) Uniform Prior 500 351 0.123
(ii) Informative Prior 500 324 0.198
(iii) Overestimation 500 327 0.0825
(iv) Underestimation 500 387 0.0789
(v) Increase in Risk 500 434 0.0209
that firms reduce their investment level in anticipation of more information

in the future that will dissipate their current uncertainty about the true
distribution ofthe price of investment. Taken together, the results in Tables
3 and 4 suggest that policies that are aimed at reducing uncertainty about
the unknown costs of investing will have a greater impact in increasing
investment than policies that are aimed at reducing the cyclical variation
in the price of capital or other components of the costs of investing.
Next, we consider the quantitative significance of optimism and pes-
simism by the firm, modelled here as underestimation and overestimation
of the permanent component of the price of capital, respectively.13 Part (iii)
of Table 4 considers the effects of pessimism on the part of the firm while
part (iv) considers the effects of optimism. 14 In part (iii), the incidence of
the binding irreversibility constraint is 65% and roughly equal to that in
part (ii), where we assume an informative prior. However, the average level
of investment is lower in part (iii) than it is in part (ii). Thus, when the firm
overestimates the costs of investing, the noisy observations that it receives
on the permanent component of the price of capital tend to reinforce its
prior beliefs, and it refrains from investing compared to the case in which
it has an informative prior.
An interesting result that emerges from Table 4 is that when the firm
underestimates the costs of investing, investment actually falls compared
to the case in which it overestimates them. As part (iv) of Table 4 shows,
the irreversibility constraint binds 77.4% of the time, and the average level
of investment is reduced to 0.0789. This result arises as follows. Since the
simulations are done holding all the remaining parameters constant, the
subjective distribution which underestimates the true state has a lower
13Fanny Demers (1985) shows how lack of confidence, underestimation and overesti-
mation of productivity and monetary shocks lead to cyclical effects.
14The prior distribution in the former case places the probablities 'I/l~ = 0.1, 'I/lg = 0.1
and "!jI~ = 0.8 for the three possible values of pk while in the latter case, the prior
distribution is 'I/l~ = 0.8, "!jig = 0.1 and 'I/l~ = 0.1.
62
subjective mean than the one which overestimates the true state. It is
therefore FSD dominated. Denote by lunder the optimal investment level
given underestimation. Since VK is increasing in pk, we have from the first-
order condition E[VKloverestimation] > E[VKlunderestimation] = pk(l_
, - z), where we evaluate VK at the same level of capital stock, namely,
(1 - 8)Kt + lunder, but under different distributions. By concavity of V,
lover > lunder where lover is the optimal investment given overestimation.
When the firm underestimates the future cost of investment and thus
expects a lower future cost than the true value, it believes that there is
no particular advantage of investing now. In this case the irreversibility of
investment takes primacy in the firm's evaluation, and it decides to invest
less today. On the other hand, when the firm overestimates the future cost
of investment, it sees an opportunity in investing now since it will be more
costly to invest at a later date. This effect partly mitigates irreversibility
considerations, and the "unduly pessimistic" firm invests more than the
"unduly optimistic" firm.
A final experiment that we perform is to change the riskiness of the price
of capital while the firm is still learning. For this purpose, we consider a
uniform prior distribution over possible values of pk and assume that the
variance of the underlying signal is double the value that it has in part
(i) of Table 4. As the results in part (v) of Table 4 show, an increase in
risk when the firm is uncertain about the permanent component pk has
a greater impact than an increase in risk that takes place in the absence
of such uncertainty and learning. When the variance of the observed price
of capital increases, firms find it more difficult to make inferences about
the unknown permanent component of capital, and refrain from investment
when irreversibility is present. This is the so-called information effect of an
increase in the variance of the price of capital, and the above results show
that it has a substantial negative impact on investment.
8 Conclusions
In this paper, we examined the impact of learning about the unknown costs
of investment on the risk neutral monopolistic firm's irreversible investment
decision. The costs of investing may be imperfectly known due to uncer-
tainty about the investment tax credits or about the price of investment
goods. We showed that the prospect of obtaining better future information
and of learning increases the endogenous cost of adjustment and depresses
current investment. Second, once learning is complete and the firm knows
the permanent component, we investigated the impact of greater cost risk-
iness. We showed that, in contrast to the exogenous cost-of-adjustment
model, a mean-preserving increase in the riskiness of the cost of investing
raises the endogenous marginal adjustment cost, thereby reducing invest-
ment and lowering the steady-state capital stock. Hence, in terms of the
well-known Knightian distinction, we established a negative relationship
63
between uncertainty and investment and risk and investment.

In our simulations, we illustrated the quantitative impact of increases in
uncertainty and risk for the costs of investment. While merely suggestive of
the types of effects that can be generated from the irreversible investment
model, our simulations indicate the rich set of results that can be obtained
about changes in various aspects of the stochastic environment facing firms.
Combined with an explicit estimation strategy that systematically matches
the model with the data, the quantitative predictions of the irreversible
investment model ought to prove useful in assessing the impact of various
tax policies, or policies aimed at changing other aspects of the costs of
investing at the firm level or at the sectoral level, where considerations of
irreversibility may play an important role.
In this paper, we have abstracted from general equilibrium effects which
may arise. For example, learning and increases in risk may affect investment
in the aggregate, thereby altering the price for investment goods and feeding
back on a further investment response. We leave this type of issue for future
work. We have also not explored in a thorough manner the implications of
our model for the effects of alternative tax policies. For example, we have
not discussed uncertainty with respect to the investment tax credit or the
corporate tax rate either theoretically or quantitatively; the latter affects
the after-tax marginal value of capital as well as the tax-adjusted price of
investment goods. We leave these topics for future research as well. (See
Altug, Demers, Demers, 1998a, 1998b.) Finally, another topic that we will
consider in future work is the estimation of the model's parameters taking
explicit account of the restrictions implied by the irreversible investment
model.
Appendix
Proof of Lemma 1. The proof is by backward induction and uses the fact
that V can be viewed as the limit of a sequence of value functions corre-
sponding to finite horizon problems. Consider a sequence of value functions
{VN(K, w)} defined by VO == 0, and V N = T [V N- 1 (K, w)], where a su-
perscript on V, K, I denotes the number of time periods left until the end
of the horizon. Observe that Vi == (l-r)II(Kl) where Kl = (1-8)K2+I2.
Therefore, 11 == O. Also,
V2 = max{{l- r)II(K2) - pk2I2 + ,BE[Vl(Kt, w1 )]}.
12
Now, vi = (1- r)IIK(K2) + (1- 8) min {pk2,,B(1- r)IIK ((1- 8)K2)}.

Thus, Vi is trivially concave in w. Next,
vI = (1- r)IIK(K 3) + (1- 8) min {pk3,,BE[V2(K2, w2 )J}.
The proof proceeds along the same lines as that of Lemma 2 in Demers
(1991) to show that VI(K, Wit) ~ JiVI(K, wi ) + (1 - Ji)VI(K, wi) where
64
wl' = /l-Wi + (1 - /l-)W i , 0 :::; /l- :::; 1. Then, assuming that vt, ... ,vll-
1 are
concave in W, we can show that V/l is concave in w. Since the sequence

of value functions {VN (K, w)} converges uniformly to the limit function
V(K, w), the value function V is differentiable almost everywhere in K,
and the sequence {v/l} converges uniformly to the limit function VK, VK
is a limit of concave functions in w.
Proof of Proposition 3. From Lemma 1 and Theorem 2,
Et+l VK ((1- 8) Kt+l, 9 (wt+l, ht+2)) (1 + r)-l

falls with the anticipation of more informative signals. As a result, the
likelihood of being constrained at t + 1 increases (that is, the likelihood
that
increases). Hence, the marginal adjustment cost increases.
Proof of Proposition 4. From an application of Lemma 1 and the first-

order conditions for It, we have
ke(ht+ll~t)VK ((1- 8)Kt + I;, ~t+l(ht+l)) dht+l :::;
i 8(ht+llwt)VK ((1- 8)Kt + I;, wt+ 1 (ht+d) dht+l :::; PI.
By the concavity of V in K t+1, it :::; It.

Proof of Lemma 5. First, one needs to show that for all 8 E 8, P(8,·)
is a probability on (8, S). Note that (8,·) is a measurable function of hi.
Hence, -l(M)s == B E B(H) for all 8 E 8, M E S. Observe that the
expression on the right-hand side of equation (18) is simply
In other words, given the state 8 at time t, the stochastic kernel Q induces
the same measure in B, as the measure induced by the transition probability
Pin M. Notice that Q(8,·) is a probability measure. Taking inverse images
under preserves all the necessary set theoretic operations. Therefore,
P(8,·) is a probability measure on (8,S). Second, one needs to show that
for all M E 8, PC M) is measurable with respect to S. This follows since
QC M) is measurable with respect to S. Thus, P is a transition probability
on (8,S).
Proof of Theorem 6. Before we turn to the proof we need the following

definition.
65
Definition 3 T is irreducible in the sense of Rosenblatt {1961} if for any

continuous function m> 0, m :f- 0, mE C(S) and any s E S, there exists
a positive integer N(m, s) such that L~l (Tim) (s) > 0.
Since S is compact, T is irreducible (See Rosenblatt, 1967, p. 476). The
ergodic set r is a compact subset of S and is such that P(s, E) = 1 for
sEE. By Mendelssohn and Sobel (1980), Theorem 6.1 (iii), there exists a
unique invariant measure JL*.
Proof of Theorem 7. The convergence of any initial measure to a unique

invariant measure JL* is demonstrated by analyzing the properties of the
Markov operator. Before we prove Theorem 7, we need the following defi-
nition and lemma.
Definition 4 A Markov operator T is
{i} uniformly mean stable if the sequence of continuous functions
{L~~Ol Ti(m)jn} is uniformly convergent for each m E C(S);
{ii} tight if for each € > 0, there is an integer n and a compact set F€
such that pn(s, F€) > 1 - € for all s;
{iii} equicontinuous if it is uniformly mean stable and tight.
Lemma 3.A The Markov operator T is equicontinuous.
Before we proceed with the proof of Lemma 3.A, we need the follow-
ing definitions. Let X denote a Banach space and bX the unit ball of X.
Lin(X), the space of all continuous linear maps X --t X, is a Banach space
under the operator norm ITI == sUPxEbX ITxl.
Definition 5
{1} T is compact if the image of bX under T has compact closure in X.
T is quasi-compact if there is a compact operator L and an integer n
such that ITn - LI < 1.
{2} Let T be in Lin(X). T is weakly compact if the image under T of
bX has compact closure in the weak topology on X. T is quasi-weakly
compact if there is a weakly compact operator L and an integer n such
that ITn - LI < 1.
Definition 6 T is stable if Tm is continuous and bounded whenever m E
C(S) is.
Proof of Lemma 3.A By Futia's Theorem 3.3 T is equicontinuous if it

is stable and quasi-compact. We need to show that T is stable and quasi-
compact.
(i) T is stable:
Q(s, B) E ca(H) for all s E S. We need to show that the mapping S --t
ca(H) defined by s --t Q(s,·) is continuous. Let s = [K, W] and s# =

66
[K#, W#] and recall that Q(s, B) = X(W, B) == iE 8(h' I W)dh'. Then,
IQ(s,B) -Q(s#,B)1 IX(W,B) -X(w#,B)1
= Il[e(h' I'll) - 8(h' I W#)]dh'l
= l t, >. I pk) (h' dh' [7fi - 7ft]

= I>.· [w-w#]1
< 11>.11 IIw - w#II
for all B E B (H). Therefore, the mapping s - t Q(s,·) is Lipschitz with
constant not exceeding 211>'11. Applying Futia's Proposition 5.6, if m E
0(8), then Tm E 0(8). Therefore, T is stable.
(ii) T is quasi-compact:
We first show that T is weakly compact. By Futia's Proposition 4.6, T

is weakly compact if the transition density is bounded. To establish the
latter, observe that the transition probability P(s, M) defined in equation
(18) has a continuous density p(s, x) : 8 x 8 - t R. Since 8 is compact and
a continuous function on a compact set is bounded, the transition density
is bounded and T is weakly compact. This implies that T is quasi-weakly
compact which in turn, by Futia's Proposition 4.4, implies that T is quasi-
compact.
Consequently, T is equicontinuous since it is stable and quasi-compact.
By Lemma 3.A T is equicontinuous. It follows by Futia's Theorem 2.10
that the sequence { I:~==-OI T*i f.L* / n}
converges weakly to only one invariant
probability f.L* .15
Proof of Lemma 8. The proof is by backward induction and uses the fact
that V can be viewed as the limit of a sequence of value functions corre-
sponding to finite horizon problems. Consider a sequence of value functions
{VN(K,pk)} defined by V O== 0, and V N = T [VN-l(K,pk)] , where a su-
perscript on V, K, [ or on pk denotes the number of time periods left
until the end of the horizon. Observe that VI == (1 - T)II(Kl), where
Kl = (1 - 8)K2 + [2. Therefore, J1 == O. Also,
V2 = max{(1 _ T)II(K2) _ pk2 [2 + ,6E[Vl(Kl ,pkl)]}.
12
15 An alternative proof of existence of an invariant probability measure J.L*is as follows.

Since S is compact, T is tight. In the proof of Lemma 3.A we showed that T is stable. By
applying Futia's Theorem 2.9, given that T is stable and tight, there exists an invariant
probability measure J.L*.
67
Now, vk = (1 - T)IIK(K2) + (1 - 8) min {pk2, ,6(1 - T)IIK (1 - 8)K2)}

is piecewise linear and, therefore, concave in pk2. Assume that v1b i =
3, ... , N - 1 is concave in pki. Notice that
is piecewise linear and therefore concave in pkN. Since the sequence of

value functions {VN (K,pk)} converges uniformly to the limit function
V(K,pk), the value function V is differentiable almost everywhere in K,
and the sequence {Vf!} converges uniformly to the limit function VK, VK
is a limit of concave functions in pk.
Proof of Proposition 9. Let It be the optimal investment level given the
distribution function G and h the optimal investment level given Gwhere
G is a MPS of G. We have
J VK«l- 8)Kt + I;,p{+1)dGW+1)
: :; J
VK«l- 8)Kt + I;,p{+1)dG(P:+1) :::; p{,
where the first inequality follows by Lemma 8 and the second by the Kuhn-
Tucker condition for It. Since V is concave in K H1 , h :::; It.
Proof of Lemma 10. As in the proof of Lemma 8, we consider a se-
quence of value functions {VN(K,pk)} defined by VD == 0, and V N =
T [VN-l(K,pk)] , where VI == (1- T)II(K1) where K1 = (1- 8)K2 + 12.
Therefore, 11 == O. Also,
and
vk = (1 - T )IIK(K2) + (1 - 8),6EVk [(1 - 8)K2 + 12, pk1) I pk2]

= (1 - T)IIK(K2) + (1 - 8),6(1 - T)IIK (1 - 8)K2 + 12 (pk2) ) .
We need to consider two cases: 12(pk2) = 0, and 12 (Pk2) > 0. If 12 (pk2) = 0,

vk is trivially increasing in pk2. If 12 (pk2) > 0, then
dI 2 1
where dpk2 ,6(1 - T)IIKK'
Therefore, Vk p k2 = (1 - 8) > 0, which implies that Vk is increasing in pk2.

Now, given that K2 = (1- 8)K3 + 13,
68
and
where
dI3 1-,BlimE_o~Jvk[dG(pk2Ipk3+€)-dG(pk2Ipk3)]. 3
dpk3 = ,BE[VkK I pk3] if I > 0
::k: = 0 if 3= O.
1
We need to consider two cases: 13 = 0 and 13 > O.

(a) Now, suppose that 13 > O. Then Vi p k3 reduces to
Vi p k3 = (1 - 8) {1 -
,B lim! J vk ((1 - 8) K3 + I 3,pk2) [dG(pk2 I pk3 + €) - dG(pk2 I pk3)] +
J
e-O €
,B lirn! vk (1 - 8) K3 + I 3,pk2) [dG(pk2 I pk3 + €) - dG(pk2 I pk3)]}

e-O €
= (1 - 8) ~ 0 if G(pk2 I pk3 + €) ~ G(pk2 I pk3).
(b) Suppose next that 13 = O. Then Vi p k3 reduces to
Vi p k3 = (1 - 8) x
,B lirn!
e-O €
J vk (1 - 8) K 3,p2k) [dG(pk2 I pk3 + €) - dG(pk2 I pk3)] ~0
if G(pk2 I pk3 + €) ~ G(pk2 I pk3). That is, since Vk is increasing in pk2,

Vi p k3 ~ 0 if G(pk2 I pk3 + €) dominates G(pk2 I pk3) by FSD, whether
13 > 0 or 13 = O.
. .m pk3 . Now, assume t hat VK"'"
Thus, VK3"IS mcreasmg VKN- 1 are m-
V:
4 .
creasing in pk. Proceeding analogously, we can show that is increasing

in pk. Therefore, the limit function VK is increasing in pk.
Proof of Proposition 11. Proceed analogously as in the proof of Propo-

sition 9 and apply Lemma 10.
Proof of Lemma 12. As in the proof of Lemma 8, we consider a se-

quence of value functions {V''V (K,pk)} defined by V O == 0, and VN =
T [VN-l(K,pk)) , where VI == (1 - r)I1(Kl) and Kl = (1 - 8)K2 + 12,
Therefore, 11 == 0. Note that
69
Since E[Vk ((1- 8)K2,pk1) Ipk2] == (1 - r)IIK ((1- 8)K2), the second
term is not a function of pk2. Thus, VI is piecewise linear and concave
in pk2. Now, given that K2 = (1 - 8)K3 + 13,
V 3 = (1 - r)II(K 3) + (1 - 8){3EV2 [((1 - 8K 3 + 13,pk2) I pk3] ,

and
vi = (1 - r)II K (K 3) + (1 - 8) min {pk3, {3E[Vl ((1 - 8)K3,pk2) Ipk3]} .
Now let Tk == E[VI ((1 - 8)K 3,pk2) Ipk3]. Then

T3K p k3 p k3 --
(1 - 8) {3lim lim hI
€--+Oh--+O 10
Jvi ((1 - 8) K 3,p2k) {[dG(pk2 I pk3 + h + f)
_dG(pk2 I pk3 + f)] - [dG(pk2 I pk3 + f) _ dG(pk2 I pk3)]}.
Since vi is increasing in pk2, we know that the first and second terms in
brackets are both positive since by assumption G(pk2 I pk3 + f) dominates
G (Pk2 I pk3) by FSD and G (pk2 I pk3 + 10 + h) dominates G (Pk2 I pk3 + f) by
FSD. Therefore, Tt- p k3 p k3 :S 0 if [G(pk2 I pk3 + h + f) - G(pk2 I pk3 + f)] :S
[G(pk2 I pk3 + f) - G(pk2 I pk3)], that is, if successive increases in the
observed value of pk3 induce FSD shifts in the conditional distribution of
next period's pk, but they do so at a decreasing (or constant) rate.
Since vi is the minimum of concave functions, it is concave in pk3. As-
sume that VR-' ... ,V:-1 are increasing and concave in pk. Proceeding anal-
ogously, we can show that the sequence of value functions {V:- (K,pk)}
converge uniformly to the limit function V(K,pk), that the value function
V is differentiable almost everywhere in K, and that the sequence {V:-}
converges uniformly to the limit function VK given by equation (13). Since
VK is a limit of concave functions, it is concave in pk.
Proof of Proposition 13. Proceed analogously as in the proof of Propo-
sition 9 and apply Lemma 12.
Proof of Proposition 14. The proof follows from Proposition 13.
References
Abel, A. (1979). Investment and the Value of Capital. (New York: Garland
Publishing).
Abel, A. (1982). "Dynamic Effects of Permanent and Temporary Tax Poli-
cies in a q Model of Investment," Journal of Monetary Economics, 9,
353-73.
Abel, A. (1983). "Optimal Investment under Uncertainty," American Eco-
nomic Review, 73, 228-33.
70
Abel, A. (1985). "A Stochastic Model of Investment, Marginal q and the

Market Value of the Firm," International Economic Review, 26, 305-29.
Altug, S. (1989). "Time-to-Build and Aggregate Fluctuations: Some New
Evidence," International Economic Review 30, 889-920.
Altug, S. (1993). "Time-to-Build, Delivery Lags, and the Equilibrium Pric-
ing of Capital Goods," Journal of Money, Credit, and Banking 25, 301-
319.
Altug, S., M. Demers, and F. Demers (1998a). "Tax Policy and Irreversible
Investment," Unpublished manuscript.
Altug, S., M. Demers, and F. Demers (1998b). "Political Risk and Irre-
versible Investment," Unpublished manuscript.
Basu, S. and J. Fernald (1997). "Returns to Scale in U.S. Production:
Estimates and Implications." Journal of Political Economy, 105, 249-
283.
Bean, C. (1989). "Capital Shortage and Persistent Unemployment," Eco-
nomic Policy, 8, 11-53.
Bertola, G. (1989). "Dynamic Programming, Option Pricing and Irre-
versible Investment," MIT, mimeo.
Bertola, G. and R. Caballero (1994). "Irreversibility and Aggregate Invest-
ment," Review of Economic Studies, 61, 223-46.
Bertsekas, D. (1976). Dynamic Programming and Stochastic Control. (New
York: Academic Press.)
Blackwell, D. (1951). "Comparison of Experiments" , Proceedings of the 2nd
Berkeley Symposium on Mathematical Statistics and Probability (Berke-
ley: University of California Press).
Blackwell, D. (1953). "Equivalent Comparison of Experiments," Annals of
Mathematical Statistics, 24, 265-73.
Burnside, C. (1996). "Production FUnction Regressions, Returns to Scale,
and Externalities," Journal of Monetary Economics, 37, 177-20l.
Caballero, R. (1991). "Competition and the Non-Robustness of the
Investment-Uncertainty Relationship," American Economic Review, 81,
279-88.
Caballero, R. and R. Pindyck (1992). "Uncertainty, Investment and Indus-
try Evolution," NBER Working Paper no. 4160.
Cooley, T. (1995). Frontiers of Business Cycle Research. (Princeton:
Princeton University Press.)
DeGroot, M. (1970). Optimal Statistical Decisions. (New York: McGraw-
Hill).
Demers, F. (1985). "Bayesian Learning as a Propagating Mechanism in a
Dynamic General Equilibrium Model of Business Cycles," Ph.D. disser-
tation, The Johns Hopkins University.
Demers, M. (1985). "Investment under Uncertainty, Irreversibility and the
Arrival of Information over Time," Essay 1, Ph.D. Dissertation, The
Johns Hopkins University.
Demers, M. (1991). "Investment under Uncertainty, Irreversibility and the
71
Arrival of Information Over Time," Review of Economic Studies, 58, 333-

50.
Eisner, R. and R. H. Strotz (1963). "Determinants of Business Investment,"
in Commission on Money and Credit, Impacts of Monetary Policy. (En-
glewood Cliffs: Prentice Hall).
Dixit, A. and R. Pindyck (1994). Investment Under Uncertainty (Prince-
ton: Princeton University Press).
Futia, C. (1982). "Invariant Distributions and the Limiting Behaviour of
Markovian Economic Models," Econometrica, 50, 377-408.
Hadar, J. and W. E. Russell (1971). "Stochastic Dominance and Diversifi-
cation," Journal of Economic Theory, 3, 288-305.
Hall, R. and D. Jorgenson (1967). "Tax Policy and Investment Behav-
ior," American Economic Review, 57, 391-414.
Hartman, R. (1972). "The Effects of Price and Cost Uncertainty on Invest-
ment," Journal of Economic Theory, 5, 258-66.
Hartman, R. (1973). "Adjustment Costs, Price and Wage Uncertainty and
Investment," Review of Economic Studies, 40, 259-67.
Hayashi, F. (1982). "Tobin's Marginal q and Average q: A Neoclassical
Interpretation," Econometrica, 50, 213-24.
Holmstrom, B. (1979). "Moral Hazard and Observability," Bell Journal of
Economics, 19, 74-91.
Jones, C. (1994). "Economic Growth and the Relative Price of Capital,"
Journal of Monetary Economics, 34, 359-82.
Jorgenson, D. (1963). "Capital Theory and Investment Behavior," Ameri-
can Economic Review, 53, 47-56.
Jorgenson, D. and M. Sullivan (1981). "Inflation and Corporate Capital
Recovery," in C. Hulten (ed.) Depreciation, Inflation, and the Taxation
of Income from Capital. (Washington: The Urban Institute.)
Jorgenson, D. and K-Y. Yun (1991). Tax Reform and the Cost of Capital.
(Oxford: Clarendon Press.)
Keane, M. and K. Wolpin (1994). "The Solution and Estimation of Discrete
Dynamic Programming Models by Simulation and Interpolation: Monte
Carlo Evidence," Review of Economics and Statistics, 76, 648-672.
Kydland, F. and E. Prescott (1982). "Time-to-Build and Aggregate Fluc-
tuations," Econometrica, 50, 1345-1370.
Marschak, J. and K. Miyasawa (1968). "Economic Comparability of Eco-
nomic Systems," International Economic Review, 9, 137-74.
Mendelssohn R, and M.J. Sobel (1980). "Capital Accumulation and the Op-
timization of Renewable Resource Models," Journal of Economic Theory,
23, 243-60.
Morrison, C. (1992). "Markups in U.S. and Japanese Manufacturing: A
Short-Run Econometric Analysis," Journal of Business and Economic
Statistics, 10, 51-63.
Nickell, S. (1977). "Uncertainty and Lags in the Investment Decisions of
Firms," Review of Economic Studies, 44, 249-63.
72
Nickell, S. (1978). The Investment Decisions of Firms. (Cambridge: Cam-

bridge University Press).
Pindyck, R. (1982). "Adjustment Costs, Uncertainty and the Behaviour of
the Firm," American Economic Review, 72,415-27.
Pindyck, R. (1988). "Irreversible Investment, Capacity Choice and the
Value of the Firm," American Economic Review,78, 969-85.
Pindyck, R. (1993). "A Note on Competitive Investment Under Uncer-
tainty," American Economic Review, 83,273-77.
Roeger, W. (1995). "Can Imperfect Competition Explain the Difference
Between the Primal and Dual Productivity Measure? Estimates for U.S.
Manufacturing," Journal of Political Economy 21, 316-330.
Rosenblatt. M. (1967). "Transition Probability Operators," Proceedings of
the 5th Berkeley Symposium in Mathematics, Statistics and Probability,
vol. 11, part 2, 473-83.
Rothschild, M. and Joseph E. Stiglitz (1971). "Increasing Risk: I, A Defi-
nition.," Journal of Economic Theory, 2, 66-84.
Sargent, Thomas J. (1980). " 'Tobin's q' and the Rate of Investment in
General Equilibrium," Carnegie-Rochester Conference Series on Public
Policy 12, 107-154.
The significance of the market portfolio:
theory and evidence*
Stefano Athanasoulis 1 and Robert J. Shiller2
1 Iowa State University, Department of Economics, Heady Hall, Ames, lA
50011, U.S.A.
2 Yale University, Department of Economics, Cowles Foundation 30, Hillhouse
Ave., New Haven, CT 06511, U.S.A.
Abstract. The market portfolio (world portfolio) is in one sense a least im-
portant portfolio to provide to investors; there is always a better portfolio
for social planners to make available to them. In a J-agent one-period sto-
chastic endowment economy, where preferences are quadratic, the market
portfolio is never spanned by the optimal markets a social planner would
create. With identical preferences, the market portfolio is orthogonal to all
J - 1 portfolios which achieve a first best solution. These conclusions rely
on the assumption that the social planner has perfect information about
agents' utilities. We also show that as the contract designer's information
about agents' utilities becomes more imperfect, the optimal contracts ap-
proach contracts that weight individual endowments in proportion to el-
ements of eigenvectors of the variance matrix of endowments. If there is
a substantial market component to endowments then a social planner, for
reasons of robustness and simplicity, may conclude that creating a contract
to allow trading the market portfolio would be a significant innovation.
To gauge the empirical relevance of the world portfolio, we estimate the
optimal contracts, using Maddison [1995] data on per capita income, for
Canada, France, Germany, Italy, UK and the USA. We find that when the
contract designer has no information about utilities, the first two contracts
nearly span the world portfolio.
JEL Classification Numbers: Gll, G12

Keywords: Market portfolio, spanning
·We thank Peter Bosaerts, Subir Bose, John Geanakoplos, Chiaki Ham, Jesus San-
tos, Paul Willen, Eric van Wincoop, participants at the Universit.y of Chicago Finance
Workshop, t.he Economic Theory Conference in Turkey (1997), Summer Econometrics
Societ.y Meet.ings in Pasadena (1997) and participant.s at. the NBER. conference on As-
set. Pricing (1997) for helpful comment.s. The authors are responsible for any remaining
errors.
74
1 Introduction
The "market portfolio," the portfolio of all endowments in the world, has
great significance in the capital asset pricing model (CAPM) in finance. The
Sharpe-Lintner CAPM characterization of optimal risk sharing implies that
in equilibrium no one will be subject to a random shock that is not shared
by everyone else. 1 Thus, the CAPM gives us the "mutual fund theorem,"
which asserts that only one risky portfolio need be available to individual
investors, the mutual fund that holds the market portfolio. In this paper we
seek further clarification of the significance of the market portfolio beyond
the bounds of the restrictive assumptions of the CAPM.
The original version of the CAPM was designed to describe how agents
should invest in existing financial assets. Thus each agent has some stock of
wealth and she must choose how much of her wealth to invest in each asset.
There is some zero cost intermediary that allows the agents to purchase the
assets. One of the key insights of the CAPM is that each agent needs only
the market portfolio (the portfolio of all the financial assets in the world)
and the risk free bond to be available to them to trade in so that they
obtain their optimal allocation of risk. It is in this sense that the market
portfolio is so important in the CAPM.
In our analysis, we will drop the (highly unrealistic) assumption of the
CAP M that all risks are tradable; we include in our model non-financial
endowments such as labor income. Thus in general no one will be able to
hold the market portfolio (the portfolio of all the endowments of the world)
unless unprecedented new institutional arrangements are made to permit
it to be traded. 2 We instead develop a CAPM-type model in which each
individual has a random endowment that is initially not marketable, and
we will consider adding one, two, or more contracts that make it possible to
buy or sell portfolios of claims on the endowments. We assume that these
contracts are to be traded in markets open to everyone, and a market price
will be generated such that total excess demand by all agents is zero. Thus,
by creating these contracts, we are creating new markets for portfolios of
endowments, making a risk tradable that had not been so before.
We confine our attention to designing N contracts, where N is small, in
order to prescribe in simple terms the most important risk management
actions that should be taken by groups of people and to ask if the "market
portfolio" (the world portfolio) is the most important contract, or is even in
the span of the most important contracts. Most people take no more than
simple prescriptions from existing models. Practitioners usually do not use
1 CAPM will refer to the Sharpe-Lintner version unless specified otherwise.

2lt. is an historical accident that the portfolio of all endowments in the world is called
the "market portfolio," a term that sounds odd when we consider the fact that one
CHllllot in practice buy a share in the entire world which includes claims to all countries
future income streams. We persist in this old terminology.
75
the CAPM to arrive at precise definitions of optimal portfolios, but merely

refer to the conclusion of the CAPM that investors should hold the market
portfolio of investable assets. The indexed funds that are now commonplace
were designed with this objective. But this common prescription disregards
the correlation of portfolio returns with other endowments.
It is very important, at the time financial innovation takes place, to
consider what are conceptually the most important markets. We cannot
have liquid markets for everything, and history shows that markets that
are not sufficiently valuable to participants will not succeed, and markets
will sometimes disappear when better markets are created.
It seems intuitive that it might be a good idea to create a market for all
the endowments of the world, not just the financial assets, and that such a
market portfolio might be the most important portfolio of all. It is possible
some day that the market portfolio and other major aggregates will be
traded. New derivative contracts cash-settled on income or price indexes
can achieve this goal. Methods of creating cash- settled futures contracts for
long-term claims on indexes of national income or of occupational income
are discussed in Shiller [1993]; see also Shiller and Athanasoulis [1995]. This
paper examines the theoretical arguments for setting up a market for the
portfolio of world endowments. We also conduct an empirical investigation
and find empirical justification for constructing the world portfolio. 3
In our model it is immediate that, regardless of the number or kind of
markets created, whether or not a market for the market portfolio (world
portfolio) is created, risk premia, represented by prices of our contracts
here, are as in the CAPM determined exclusively by covariances with the
market portfolio. The market portfolio is also in another sense the most
important market; with a normalization rule that we define in the paper,
the market portfolio is the portfolio which would carry the highest risk
premium (highest absolute value of price), Theorem 3 below. In fact all
other portfolios uncorrelated with the market portfolio will have a zero
risk premium.
And yet we find, curiously, that of all possible markets to create, a market
for claims on the market portfolio would be, by a social welfare criterion, a
least important market to create, not a most important market, Theorem
2 below. If we are in the business of creating markets for endowments that
are not tradable, then there is a natural order to creating such markets.
There is a most important market to create, and then, after this, a market
that would be the next best market to create, and so on. The market for
the market portfolio turns out to be a completely unimportant market in
this ordering, still not spanned by all the other markets when we get to the
end of the ordering, and then the welfare gain to creating it is zero. This
is not to say that a market for the market portfolio would not be useful to
3 Tlw empirical applicat.ion is conduct.ed for income indexed cont.ract.s.

76
people if it were created first, or if it were created second or third, only that
there would always be something better to do instead. This result may be
regarded as, in a sense, the very antithesis of the mutual fund theorem.
Neither will we ever want to create markets for individual endowments or
for portfolios weighting all endowments with the same sign. Optimal con-
tracts will always involve portfolios of risky endowments with both positive
and negative quantities and their weighted sum is zero. The optimal con-
tracts are thus always essentially swaps, i.e., one side trades the negatives
for the positives. This result may be regarded as in a sense the apotheosis
of swaps.
The results that there is no need for a market for the market portfolio
and that only swaps will be created rest on the assumption that the con-
tract designer who is creating the new markets knows everything about
utilities. We show one representation of lack of knowledge on the part of
the market designer that brings the market portfolio back to some potential
significance, Theorem 7 below. If lack of knowledge is high and if there is a
strong market component to endowments, then something approximating
the market portfolio may well be of first importance. Creating the market
portfolio makes possible more robust definitions of subsequent markets.
Theorem 1 below is part of a framework developed in Athanasoulis [1995]
and Shiller and Athanasoulis [1995]; it was developed independently by De-
mange and Laroque [1995b]. A related analysis is found in Duffie and Jack-
son [1989] and Willen [1997]. See Geanakoplos [1990] for an introduction to
General Equilibrium with incomplete markets. Cass, Chichilnisky and Wu
[1996] show how the number of assets needed to obtain a complete markets
solution can be greatly reduced by constructing a set of mutual insurance
contracts and a smaller set of Arrow securities when compared to an Arrow-
Debreu world. This is related to our results as we only need assets far less
than the number of states of the world to obtain a first best solution. We
however consider which assets are best to construct if we do not complete
assets markets. Demange and Laroque show [1995a] that in an economy
with general utilities (not necessarily quadratic), when all residual risk is
hedged, the only important assets remaining to construct in the economy
are non-linear assets, such as options, whose realizations depend exclusively
on the realization of the market portfolio. Our results are complementary
to this Demange and Laroque result rather than being a competing result.
Our analysis here starts from no markets at all, and studies a sequence
of markets to allow linear spanning of the original endowments; Demange
and Laroque [1995a] are considering moving yet beyond the linear span-
ning, and it is in the subsequent nonlinear markets alone that the market
portfolio has (under their assumptions) such importance.
In the empirical section, we estimate the optimal contracts in three differ-
ent cases; when the contract designer is perfectly informed about utilities,
when she has less than perfect information and when she has no informa-
tion. Our estimation is conducted for Canada, France, Germany, Italy, UK
77
and the USA using the Maddison [1995] data on per capita income, 1880-
1992.4 Analyzing the first two contracts, we find the first contract is a swap
of risk between the US and the European Union, while the second contract
is a swap of risk within the European Union. This is true for the cases
where the contract designer has perfect information and a little less than
perfect information. However, we find that when the contract designer has
no information about utilities, then the world portfolio is nearly spanned
by the first two optimal contracts. The first contract is a contract for US
risk and the second contract is a contract for European Union risk. Thus
there is an empirical justification, as well as theoretical, for constructing
and marketing the world portfolio.
The paper is organized as follows. We first layout the assumptions of the
general equilibrium model and then solve the agent's problem for a given
set of available contracts. The resulting expressions for equilibrium prices
and quantities will be used in all subsequent parts of this paper. We then
go through several variations on the maximization problem faced by the
contract designer, differing in assumptions about pre-existing markets and
about the information available to the designer. An empirical investigation
ensues to answer whether the world portfolio is an important portfolio. We
then conclude with some practical advice for contract designers.
2 The model
There are J agents in this economy indexed by j = 1, ... , J, each repre-
senting an individual except in Section 11, where each agent represents a
large number of individuals. All random variables are defined on a complete
probability space (n, F, P), where n is the set of states of the world and
wEn is the state of the world. F is a a-algebra of subsets of n known as
events and P : F ---t[0,1] satisfying P(0)=0 and p(n)=l is a probability
measure on (n, F) held commonly by all agents in the economy.
There is a single good in the economy which is consumed. Each agent j
has an endowment Xj E L2(n, F, P) where L2 (n, F, P) is the set of random
variables which are square integrable, i.e., have finite mean and variance.
We will denote the demeaned stochastic endowment as Xj = Xj - E(xj).
Define x to be the 1 x J vector of random endowments in the economy and
similarly let x be the 1 x J vector of demeaned stochastic endowments.
Then E(x'x) = L: is the J x J covariance matrix of the endowments in the
economy. Define E(x'xj) = L:j and E(xjxj) = L: jj .
The N S J contracts indexed by n = 1, ... , N designed in this paper
are futures contracts. Let In E L2(n, F, P) be the risky transfer made in
the nth futures contract resulting in In(w) units of consumption contingent
4 These are the set of G-7 couutries for which all the observations are available, 1880-
1992.
78
on state W E O. To purchase contract n, the agent must promise today to

pay a riskless price Pn E n in the period where the state of the world is
resolved. Thus if the state W E 0 is realized, agents who take a long position
in contract n receive In(w) - Pn, those who take a short position pay this
amount. Define I to be the N x 1 vector, whose nth element is In and P
to be the N x 1 vector whose nth element is Pn. Without loss of generality
we construct the futures contracts such that E(f) = 0 and E(f 1') = IN,
where IN is the N x N identity matrix. These are two normalizations that
have no effect on the economy. For example, if E(fn) = 1, then we need
only increase the price Pn by one. So the equilibrium is invariant to these
linear transformations. If var(fn)=2 then we need only increase the price
of contract n by the square root of 2.
Given that we restrict our attention to the set of linear equilibria, i.e., we
use quadratic utility, it must be that the optimally chosen risky transfers,
I, are in the space spanned by the initial endowment risks, X. This is a
point shown by Demange and Laroque [1995]. Furthermore it must be that
since I is in the space spanned by x and we are studying linear equilibria,
the optimal contracts will be linear combinations of the elements of X.
Consequently we define I = A'x', where A is a J x N matrix and A~x' E
L2(0, F, P) , n = 1, ... , N and An is the nth column of A . Therefore,
according to our notation, E(fl') = A'~A = IN.
3 Agents
Each representative agent has a utility function Uj : L2(0" F, P} ---> R. We
make the simplifying assumption that each agent has mean-variance utility
as follows:
(1)
where Cj is the consumption of agent j, the same as the endowment plus
proceeds from hedging. Each agent j takes the risky transfers contracts I
which is a vector L2(0, F, P) process and the futures prices P E RN as
given and solves for her optimal futures positions, qj, as
qj = argmaxqjE'R.N {UjlCj = Xj + q/(f - P)}. (2)
We can rewrite this in a simpler form as
qj = argmaxqjE'R. N {E(Xj) - q/p - ~ (~jj + q/A'~Aqj + 2q/A'~j)}.

(3)
Remembering that A'~A = IN, the optimal demand for this agent is
qj = -~P - Cov(f, Xj) = -~P - A'~j. (4)

Ij Ij
This demand curve tells us that agent j will purchase more of a security as
its price declines. She will purchase less of the security the more it covaries
79
with her endowment since it provides less hedging services. To help the
exposition of this paper it is convenient to form the N x J matrix Q whose
lh column is qj and rewrite (4) as
= -pt'r- 1 - Cov(j,x) = -pt'r- 1 - A'1:,
Q (5)
where r is the J x J diagonal matrix with the lh diagonal element equal
to 'Yj and t the J x 1 unit vector.
4 Equilibrium
The equilibrium condition in this economy is simply that the futures con-
tracts are in zero net supply. We can represent equilibrium in this economy
as
(6)
From equilibrium condition (6) we can derive the equilibrium pricing equa-
tion
P = -A'1:t (t'r-1tr1 . (7)
Definition The market portfolio is defined by its dividend
fm =xm, (8)
where m is a scaled unit vector
t
m=== . (9)
(t'1:t) .5
If we multiply and divide the right hand side of (7) by (t'1:t)·5 then the
price of a contract n depends on A~1:t(t'1:I)-·5 === A~1:m, the covariance of
contract n with the market. Thus we can derive the CAPM pricing equation
from equation (7). If the covariance of a contract with the market is zero,
as for example with a risk free asset, then the price of this asset is PI = O.
The price of the market portfolio is Pm = - (::i?~(. It follows that
cov(jn, fm)
Pn - PI = (f) (Pm - PI) ,
var m
(10)
which is the familiar CAPM pricing equation and Co:a~fj::)') is the famil-
iar beta of the CAPM model. Similar results are obtained by Magill and
Quinzii [1996], Duffie and Jackson [1989], Oh [1996] and Mayers[1972].5
Substituting (7) into (5) we also obtain
Q = -A'1:M (11)
r. F!'Olll cquation (7) it. tUl'llS out thl\t. Pt = 0 and t.lmH Pn is just the risk premium
from the CAPM pricing cCjuation.
80
and we define M == IJ - u'r- l (i'r-1i)-1 and A == [AI: A2 : ... : AN],

where An is the nth column of A. These are the equilibrium demands in
matrix form. Looking closely at the above expressions we see that "EM
is the J x J matrix, whose lh column is the amount of each risk agent
j wants to sell off at market-clearing prices. In the end, if markets are
complete, each agent will hold the inverse of her own risk aversion times
the harmonic mean of all individuals' risk aversion, of the market. This
result will be recognizable to those familiar with the CAPM economy, see
Huang and Litzenberger [1988].
5 Contract design
The contract designer's problem is to maximize welfare, total utility, in the
economy given she is constrained to choose N ~ J contracts. The contract
designer will choose the J x N matrix A to maximize the sum of utilities
in the economy. From (3) we know that each agent's utility is given by
If we sum over all J agents, drop E(xj), and put this in matrix form we
obtain
tr ( -Q' Pi' - ~r ("E + Q'Q + 2Q' A'"E)) , (13)
where tr denotes the trace. If we substitute (11) and (7) into (13) we obtain
tr (~rM'"EAA'"EM - ~r"E) , (14)
where the term ~r"E has no effect on the contract designer's decision. Thus,
using tr(AB) = tr(BA), the contract designer's problem simplifies to
A E argmaxAnER.J .n=I ..... N {tr (A'"EMr M'"EA) IA'"EA = IN}. (15)
This leads to a fundamental theorem shown separately by Demange and

Laroque [1995b]and by Shiller and Athanasoulis [1995]:
Theorem 1 The A matrix that solves (15) has columns corresponding to
the N eigenvectors with highest eigenvalues of
MrM'"E. (16)
Proof. We may write the Lagrangian as
I:- = A~"EMrM'"EAI + ... + A~"EMrM'"EAN-

(17)
Al (Ai"EAI -1) + ... + AN (A~"EAN -1).
81
We are requiring in this problem that the diagonal of the matrix A'~A is
equal to to The first order conditions can be written as
Vn = 1,···,N (18)
and
A~~An = 1 Vn = 1,··· ,N. (19)
If we define A to be the N x N diagonal matrix with the nth diagonal
element to be An we can combine the first order conditions to obtain
(20)
and
diag(A'~A) = L. (21)
Thus taking the inverse of ~ through equation (20) gives us the result. Fi-
nally if one premultiplies equation (20) by A', one obtains A'~Mr M'~A =
A. The trace of the left hand side of this is the objective function the plan-
ner is trying to maximize. Since this equals A, it is diagonal and as such
the planner will choose the N eigenvectors corresponding to the N largest
eigenvalues. 6 0
Note that if we take a Cholesky decomposition of the variance matrix ~,
~ = C'C, and premultiply through equation (20) by C,-1, then CMr M'C'
is positive semidefinite and symmetric with eigenvectors CA. The eigenval-
ues of a positive semidefinite symmetric matrix are all real and nonnegative,
and these are the same as the eigenvalues of Mr M'~. Since the rank of M
is J - 1, there are only J - 1 nonzero eigenvalues, and hence only J - 1
contracts are of any value. Thus, there is no point in creating all J possible
contracts, at most J - 1 are needed and A need have no more than J - 1
columns. If there is a fixed cost to creating markets, then N, the number
of markets created, can be chosen optimally. We create all markets whose
eigenvalues (divided by two) are greater than this cost.
One will notice in the above problem that we did not constrain the off
diagonal elements of A'~A to be zero. Notice however that A is diagonal
and since C' Mr M' C is positive semidefinite and symmetric with eigenvec-
tors CA, it follows that A'~MrM'~A is diagonal. Since A'~MrM'~A =
A'~AA it must be that A'~A is diagonal. Thus the contraint that the off
diagonal elements are zero is satisfied in the unconstrained problem. This
was shown by Darroch [1965] and by Okamoto and Kanazawa [1968].
6 The market portfolio is least important

It is now very easy to prove our featured result that the market portfolio
is in a sense a least important portfolio to allow trading in.
6The second order condit.ions that we have a maximum are satisfied.

82
Theorem 2 The A matrix that solves problem (15) is orthogonal to g ==

L/r-l(L/r-l~r-IL)-·5, and all N S J -1 markets together do not span the
market portfolio.
Proof. By (20) it follows t.hat A = MrM/~AA-I. Since gM = 0, it
follows t.hat gA = 0, i.e., the A matrix is orthogonal to g. We can then
show by contradiction that all N S J - 1 contracts do not span the market
portfolio: if there exists a vector v such that Av = m, then gAv = gm =
L/r-It{t/~L)-·5(L/r-l~r-IL)-·5 =f. 0 which is a contradiction. 0
Lemma 1 In the case where agents have the same risk aversion, 'Y j' the
market portfolio is orthogonal to all optimal contracts.
Proof. By Theorem 2, g is orthogonal to all optimal contracts and g =m
here. 0
This lemma is particularly important since symmetry of risk aversions is
likely to be assumed when designing new contracts.
For any J, the result that gA = 0 means that no linear combinat.ion
of the N optimal contracts can be constructed with all positive elements.
Only "swaps" between endowments can be constructed by portfolios of the
optimal contracts. No matter how many markets we choose to create (re-
gardless of N), it will be impossible to renormalize these markets, define
different markets as linear combinations of them, so that. any market is
not a swap. All possible portfolios constructed from the optimal portfolios
represent exchanges of endowments for ot.her endowments. Since the mar-
ket portfolio holds positive quantities of all endowments, it is an example
of a market that cannot be constructed from the optimal contracts con-
structed from the above method. Consider the case where all agents have
the same risk aversion, Lemma 1, so that r is proportional to the identity
matrix. It then follows from this theorem and lemma that in all possible
portfolios constructed from the optimal contracts defined by A, the sum of
the portfolio weights in terms of endowments are zero and the portfolios
are orthogonal to the market portfolio. Furthermore, if N = J - 1, then
the resulting equilibrium is Pareto optimal. See Magill and Quinzii [1996]
P.181 for pareto optimality of the CAP M equilibrium. This follows here
since the case with N = J - 1 contracts results in the CAPM equilibrium,
i.e., CAPM allocation of risk.
To understand these results better let us consider a two-agent exam-
ple. A two-agent example ignores some of the complexity that the optimal
market solution method is supposed t.o handle, but it will make some ba-
sic concepts more transparent. We can then illust.rate the solution to the
contract designer's problem on a simple two-dimensional graph, Figures 1
and 2, with the first element of AI, aI, on the horizontal axis and the sec-
ond element of Ab a2, on the vertical axis. On t.hese figures the constraint
A~ ~AI = 1 is that t.he Al vector must end somewhere on the ellipse shown.
The ellipse shown illustrates a case of positive correlation between the two
83
a2
FIGURE 1. Illustration of Optimal portfolio weights when both agents have same
risk aversion, iwc is an iso-welfare curve, nc is the normalization constraint.
endowments, where both endowments have the same variance and a corre-
lation coefficient of one half. On each figure, iso-welfare curves are parallel
straight lines (one pair of which is shown); the further from the origin the
higher the welfare.
The optimal vector Al must be orthogonal to g, which means that the
vector is in the upper left quadrant (or lower right), and is not the in the
same quadrant as the market portfolio vector m. In Figure 1, the case is
shown where all the "I's are one, and so 9 equals the market portfolio vector.
Each agent will use the optimal contract to swap half of her endowment
risk for half of the other's, and both agents will end up holding a share of
the world. In this case, the optimal contract is orthogonal to the market
portfolio, and the market portfolio contract would be utterly useless to
the agents if it were created instead of the optimal contract. The optimal
contract is found on the graph by finding the highest iso-welfare curve, iwc,
that satisfies the constraint, tangent to the ellipse. Clearly in this symmetric
situation there is no value to being able to trade the market portfolio for
these agents, as they would both like to take the same position.
In Figure 2, the case is shown where "11 equals 3 and "12 equals 1. Now,
the 9 vector no longer coincides with the market portfolio vector, m, and
the optimal Al vector results in an unequal swap. In the swap, the more
risk averse agent gives up three times as much of the risky component of
her endowment to the other agent, and pays a price to the other agent
for doing so. After the swap, the more risk averse agent is bearing only
one quarter of world endowment risk, the less risk averse agent is bearing
three quarters. This is the Pareto optimal outcome: there are no more risk
sharing opportunities, and each agent is bearing world endowment risk in
84
a2
a1
FIGURE 2. Illustration of Optimal portfolio weights when agent 1 is three times

more risk averse than agent 2, iwc is an iso-welfare curve, nc is the normalization
constraint.
accordance with own risk preferences. Note that in this case had we instead
created the market portfolio first, it would have been of some use though
it would touch an iso-welfare curve that is closer to the origin. In both
figures, the isoquants for the objective function in (15) are parallel straight
lines with just such a slope that the tangency between them and the ellipse
A~ I;A 1 - 1 = 0 occurs at a point defining a vector perpendicular to g.
7 The market portfolio has the highest absolute

value of price
Even though, as we have concluded, the market portfolio is in this model an
unimportant contract to trade, it remains true that the market portfolio
is the most important market by a different measure: it is the contract
(subject to our normalization) that has the highest possible absolute value
of price. Let us change the objective of the contract designer in designing
the first market to maximize the absolute value of price of the contract,
in designing the second market to maximize the absolute value of price
subject to zero covariance with the first market, and so on. Using equation
(7) the problem becomes
Theorem 3 The contract that satisfies (22) is the market portfolio, i.e.,
Al =m.
85
Proof. The first order conditions can be written as
(23)
and
A~I:AI = 1. (24)
Solving these we obtain the result. 0
Thus as in the CAPM, the only insurance which costs anything is to insure
oneself against the market.
It may seem puzzling that the market is completely unimportant to con-
struct by a social welfare criterion and yet has the highest absolute value of
price. But note that the expression to be maximized by the social welfare
criterion (15) is the trace of A'I:Mf MI:A, and
A'I:fI:A - A'I:u'I:A ( i'f- l i) -1 (25)

A'I:fI:A - pp' (i'f- l i) .
The trace of the first term in the above expression is the expression to
be maximized by a contract designer choosing A to minimize a ",(-weighted
sum of variances of agents' endowments subject to the normalization con-
straint and subject to the constraint that all prices are zero, See Shiller and
Athanasoulis [1995], and the trace of the second term is proportional to mi-
nus the sum of squared prices. Higher absolute value of price counts against
welfare gain because higher price means more of the risk is redistributed
(market risk) rather than pooled (residual risk).
8 Pre-existing markets
The above theorems take no account of pre-existing markets, markets for
some endowments or linear combinations of endowments that already ex-
ist before the contract designer begins to define new markets (contracts).
Suppose that we modify problem (15) to represent that there is a single
pre-existing contract, where the coefficients of the endowments in the lin-
ear combination that defines this pre-existing contract are given by the
J x 1 vector AI, the first column of A, which, without loss of generality,
we normalize so that A~ I:AI = 1. (It is trivial to extend our results to
more than one pre-existing contract.) The contract designer will then de-
sign N* = N - 1 markets, choose A* = [Ai A 2··· AN>], the remaining
columns of A, (A = [AI: A*]) subject to the normalization rule A'I:A = I.
Then A * is defined by
A* E argmaxA:'ER",n>=I, ... ,N> {tr (A*'I:lldTM'L:A*) IA*'L:A* = IN>,

A*'I:A I = O}.
(26)
86
Theorem 4 The A* matrix that solves (26) has columns corresponding to

the N* eigenvectors with highest eigenvalues of
Mf M'''E, (27)
where == I J - A1A~'E.
Proof. We can write the Lagrangian as
£ = Ai''EMfM''EAi + ... + A~*'EMfM''EAN*

-Al (Ai''EAi -1) + ... - AN* (A~*'EAN_ -1) - (28)
blAi''E Al + ... - bNA~_'E Al.
The first order conditions are
2'EMf M''EA n- - 2An- 'EAn--

(29)
bn-'E Al = 0 Vn* = 1"", N*
and
A;:_ 'EA~_ - 1 = 0 Vn* = 1"", N* (30)
A;:_'EAl = 0 Vn* = 1"" ,N*. (31)
If we premultiply equation (29) by A~, then we obtain bn *
2A~ 'EMf M''EA n-. If we substitute bn - into equation (29), form the N*
equations n * = 1,,'" N* into a matrix and rearrange, we arrive at an
equation in terms of eigenvectors of (27). If we premultiply equation (29)
by A;;'_ then we obtain A;;'_ 'EMf M''EA;;'_ = An- which is the element of
the expression of the function the planner is trying to maximize. Thus the
planner chooses the columns of A * as the N* eigenvectors with the highest
eigenvalues of (27). 0
An example can be constructed that illustrates that with one pre-existing
market, if we are to create only one more market optimally as we have
defined, then the resulting two markets may span the market portfolio.
Suppose that Al is a column of zeros except for the first element, which is
strictly positive. The pre-existing market is just a market for the endow-
ment of the first agent. Suppose, for simplicity, that 'E equals the identity
matrix and that f also equals the identity matrix except that the upper left
element is not one but a "very" large number; the first agent is very risk
averse. With these assumptions, if there had been no pre-existing market,
the first market to create would have been a swap between the first agent
and the rest of the world, with all other agents receiving equal weight in the
contract. With the pre-existing market, the optimal A* will be proportional
to a column of ones with the first element replaced with zero; creating this
market will enable the first agent to swap her endowment risk for the rest
87
of the world's, by shorting the first market and going long the second. In
this example Al and A* together also span the market portfolio.
The result that pre-existing markets may cause the contract designer
optimally to create contracts that allow spanning of the market portfolio
does not mean that the market portfolio is in any real sense important.
In the above example, the agents use the two markets to construct a swap
between the first agent's endowment and world endowment, not to take a
position in world endowment. Had the contract designer, in constructing
the contract represented by A *, ignored orthogonality with the pre-existing
market and just created the contract defined as the solution to (15), thereby
directly creating the swap between the first agent's endowment and the rest-
of-the-world endowment, then almost all the welfare improvement available
to hedgers would be available just by using the second market. One may
suppose that if the welfare gain available through the pre-existing market
is small enough then it might well disappear after the second market is
created.
9 The market portfolio as a pre-existing market

It is instructive to consider the problem for the contract designer with
the constraint that the first market is the market for the market (world)
portfolio, that is, assuming that Al = m. While a market for the world
portfolio of endowments does not now exist, we shall see that there may
be reasons to construct it. At the very least, as we shall see in this section,
these markets are conceptually relatively simple to understand, and such
simplicity might promote more effective use of the markets.
Lemma 2 If the market portfolio exists (Le., if Al = m) then all other

contracts (constructed so that our normalization A'~A = IN holds) will
necessarily have a zero price.
Proof. If the first contract is the market then it must be the case that
the rest of the contracts A~., n* = 1,···, N* are constructed such that
A~.~Al = Ji;~: = O. If this is the case, then A~.~t = 0 and from
equation (7), the result follows. 0
Let us define the N* x J matrix Q* such that its lh column is the

demand vector for agent j of the N* contracts. We then have
Theorem 5 When Al = m the A* matrix that solves (26) has the property
that Q* = - A *'~M has columns corresponding to the N* eigenvectors with
highest eigenvalues of
(32)
Proof. Using equation (25) and Lemma 2, the problem the social planner
88
solves is
A* E argmaxA;,ERJ,n-=l, ... ,N- {tr (A*'~r~A*) IA*'~A* = I N -,

(33)
A*'~Al = O,A l = m}
Proceed as with Theorem 4. 0
'~ is the variance matrix of residuals when the endowments are re-
gressed on the world endowment. If r = I J , i.e., if everyone has the same
risk aversion, then the optimal markets are defined in terms of eigenvectors
of this simple variance matrix. Moreover, since Q*' = - ~A *, the posi-
tion that agent j holds of the nth contract is just the regression coefficient
corresponding to the nth contract when the endowment of that agent is re-
gressed on the vector of contract payoffs xA *. These results, coupled with
the above-noted zero prices for all contracts other than the market con-
tract, make this equilibrium a simple one to understand. Once the market
portfolio is traded, the problem agents face for orthogonal contracts is a
variance minimization problem.
10 Uncertainty about preferences

The preceding analysis assumed great knowledge on the part of the contract
designer: the designer was assumed to know perfectly all utility functions.
The unrealism of this assumption would appear to be an issue if we try
to apply this analysis to the design of actual markets. We show that the
relaxation of this assumption may restore the importance of the market
portfolio.
Uncertainty about preferences poses a real problem to the contract de-
signer since we cannot assume that agents have the same uncertainty about
their own preference parameters that the contract designer does. Agents
have perfect knowledge about their own preference parameters and maxi-
mize their expected utility knowing their 'Yj. The above analysis of market
equilibrium, equations (6)-(11), must be done for the agents' true risk pref-
erences. When we arrive at the contract designer's problem, (15), we face
the problem that the contract designer does not know the true M and
r matrices. The only reason the contract designer does not know agent's
demands is because she does not know the coefficients of risk aversions.
Supposing now that the true elements of r are unknown to the market de-
signer, we will suppose that the market designer chooses N :S J contracts
to solve a maximization problem which is the same as (15) but replacing
the unknown value to be maximized in (15) with its expected value:
A E argmaxAnERJ,n=l, ... ,N {tr [E (A'~MrM'~A)lIA'~A = IN} (34)

Note that since M is a function of r, the expression involves expectations
of a nonlinear function ofr. In order to deal with (34), we rewrite the matrix
89
A'L:Mr M'L:A as
A'L:MrM'L:A == A'L:rL:A - A'L:u'L:A (L'r- 1L)-1. (35)
One obtains equation (35) by substituting in for M.
Theorem 6 The A matrix that solves (34) has columns corresponding to
the N eigenvectors with highest eigenvalues of
(36)
Proof. Substitute (35) into (34) and proceed as in Theorem 1. 0

Note that, unless E [r - U'(L'r- 1L)-1] is singular, the matrix (36) is
generally non-singular, and so our conclusion above that only J - 1 mar-
kets are needed no longer holds. If there is no constraint on the number
of markets constructed, the contract designer will create all J contracts,
and then the contracts will span the market portfolio. Let us assume the
,",(/s for all j = 1, ... , J are iid. This assumption represents a symmetric
state of knowledge of all individuals' risk aversion parameters. With this
assumption we can rescale (36) as
L: - cu'L:, (37)
E(L'r- 1L)-1
where c = Eh)
With (37) we can easily take account of specific distributional assump-
tions about r. We need only derive the expected value and expected value
of the harmonic mean of the elements of r, to define the scalar c.
The limiting case of this problem, when the variance of '"'( increases to
infinity, is particularly interesting. This is the case where the contract de-
signer's information is becoming more diffuse.
Theorem 7 If '"'( j' j = 1, ... , J are iid lognormal variates, then as the vari-
ance, (T2, of In( '"'( j) goes to infinity, the A matrix that solves (34) approaches
a matrix whose columns are N eigenvectors of L: with the corresponding
highest eigenvalues.
Proof. Define the geometric mean of risk aversion parameters to be G =
(TIf=l'"'(j)-1 and the harmonic mean as H = (,2:.f=1'"'(j1)-1. Under
the lognormal assumption ~t~? = e::/I-'~ a;. 2: C
= exp ( _(T2 J2J?») ).
Therefore lima2->oo ~t~] = O. Since H ~ G everywhere, (see for example
Hardy et. al. [1964]' p. 26) then lima2->oo ~l~l = lima2->oo c = O. Thus,
the limit of the matrix (37) as (T2 goes to infinity is L:. Since the solution of
problem (34) is a continuous function of the elements of the matrix (37),
and since the limit of a continuous function is the function of the limit, the
theorem follows. 0
90
If one is going to construct some contract given she knows nothing about
the utilities in the economy, what should the contract be? One wants to
somehow maximize the probability that their contract will have the highest
welfare improvement in the economy. As such the contract designer should
construct the contract that markets the largest component of risk in the
economy. This is exactly the result of Theorem 7. Given we know nothing
about risk aversion, we have the best chance of welfare improvement in
the economy by allowing agents to hedge the most risk possible. The first
principal component of I; is unrestricted by our theory. It could have all
positive elements and could approximate the market portfolio.
If the first principal component of I; is approximately the market port-
folio and its eigenvalue is large, then people have a substantial covariance
with the market. Among those agents with similar market exposures, those
who are more risk averse can sell a share of the market portfolio to less risk
averse agents, thereby reducing their risk. We do not need to know who is
more risk averse in setting up markets to make this possible.
Let us return to the two-agent examples that were plotted in Figures 1
and 2. If we do not know which agent is the more risk averse, then this
maximization problem facing the contract designer is not as simple as it
appeared from that figure. We do not know the position of the vector g,
that is whether Figure 1, Figure 2 or some other figure is relevant. Thus
the position of the optimal Al vector cannot be determined.
We plot instead in Figures 3 and 4 the expected iso-welfare-curve to
the maximization problem (34). These are not parallel straight lines but
ellipses. If we have only a little uncertainty about risk aversion, see for ex-
ample Figure 3 where c=.49, the expected iso-welfare curves are elongated
and near the origin resemble the parallel straight lines of Figure 1. But if
our uncertainty about risk aversion is large, see Figure 4 where c=O, the
expected iso-welfare curves are elongated in the perpendicular direction. In
the extreme case, where the uncertainty about agents' risk aversion makes
it very probable that one is much more risk averse than the other, then,
not knowing which is the more risk averse, the best contract we can design
in this example is simply a market for the market portfolio.
With very little uncertainty in these terms about the "('s, the optimal Al
for our two-agent example with i.i.d. "('s will still be a vector perpendicular
to the market portfolio, a vector with a slope of minus one. Note that
Figure 3, where c=0.49, resembles Figure 1 in the vicinity of the origin.
Figure 1 corresponds to c=O.5. However, even a small amount of uncertainty
means that there will still be a reason to create a second market, and A2 will
be the market portfolio vector, in the first quadrant, with slope of plus one.
As the uncertainty about the "('s increases, the eigenvalue corresponding to
Al shrinks relative to the eigenvalue corresponding to A 2 , and at some point
becomes the lower; at this point we must switch the order of the columns
of A, and the market portfolio becomes the best portfolio to create. What
has happened finally is that uncertainty about the "('s has become so great
91
a2
FIGURE 3. Illustration of Optimal portfolio weights when risk aversions are iid
and c=.49. eiwc is an expected iso-welfare curve and nc is the normalization
constraint.
that we can no longer predict what kinds of swaps will be useful to agents.
The market portfolio may still be useful if either agent is more risk averse
than the other; that agent can sell part of the market component of her
endowment to the other.
Note that this conclusion using the lognormal assumption might be gen-
eralized to other distributions but it is not true of all distributions of Ij > 0
with finite means. The important point of the theorem is that the contract
designer's information about agents' utilities becomes more diffuse. If for
some reason, as the variance approaches infinity, the contract designer's
information becomes less diffuse, then the contract designer can better
construct contracts since she has more information which results in more
welfare improvement.
Consider, for example, a case where Ij can only take on two values, Ijl
and Ij2' Ijl is fixed, the mean t is fixed and we vary Ij2' The probability
we observe Ijl or Ij2 are prl and PT2 respectively. Thus we have
(38)
and
(39)
We increase the variance of Ij by moving the higher value Ij2 towards
infinity. As we do this we reduce the probability pr2 that risk aversion for
person j equals Ij2' It is easy to show that in the limit, as the variance is
increased to infinity, i.e., as 'Yj2 ~ 00, the expected value of the harmonic
mean of J values approaches Ijl' In the limit, the probability approaches
92
a2
a1
------~----~~--_P~-----
FIGURE 4. Illustration of Optimal portfolio weights when risk aversions are

iid and c=O. eiwc is an expected iso-welfare curve and nc is the normalization
constraint.
one that all J values are the same so that the probability approaches one
that the expected value equals the harmonic mean of the J values. This
example shows that as all peoples risk aversion approaches "tj1 in the limit
and thus as the variance goes to infinity, the contract designer becomes
more informed.
11 Each agent represents K people

We now suppose that each of the J "agents" is a group of K people who
share the same endowment, but may differ from each other in terms of risk
tolerances as measured by "t. Each "agent" may represent a country or an
occupational group.
Allowing multiple individuals per "agent" is important, since in practice
we are likely to want to apply the methods for contract design not to
data on individual endowments but to data on endowments of groupings
of individuals. It is also important to consider multiple individuals per
"agent" since our uncertainty about risk aversion may be better thought of
as recognition of diversity of risk aversions within each group, rather than
as uncertainty about the average risk aversion of all people in each group.
Assuming that all individuals in a group share the same endowment, the
variance matrix of individual endowments is now E = E 0 (~~') where
o denotes the kronecker product and ~ is a K-element column vector of
ones. Assuming that all individuals risk parameters "t are iid regardless of
the "agent" group to which the individual belongs, we suppose that the
contract designer desires to maximize total utility of all individuals, i.e., to
93
find the matrix A that solves

AE argmaxA"ER.J,n=l, ... ,N {tr (A':E:EA - cA':Eu':EA) IA':EA = IN},
(40)
where c == E(t:/~~~)n-1 and where Y = ~ IX! K and f is r IX! I K .
Theorem 8 The A that solves (40) equals A IX! K where A solves (34). 7
Proof. To prove this we use the multiplication rule for Kronecker products,
(Ai !XI B i )(A2 IX! B2) = (AiA2) IX! (B i B 2). Note that c = 7<, where c ==
E(L/~~~)L)-l . We have H == :E-cu':E = ~®(KK')- ~ ((U')IX!(KK'))(~IX!KK') =
~ !XI (KK') - 7< (( U'~) !XI (KK') (KK')) = ~ IX! (KK') - c( ii'~ IX! (KK')) =
H!XI(KK'), where H == ~-C(u')~. Now, from above we know that HA = AA.
H(A!XI K) = (H IX! (KK'))(AIX! K) = (HA) IX! (KK'K) = KH A IX! K. We can also
show that (AA) !XI K = (A IX! K)A. Hence H(A IX! K) = (A IX! K)(KA). Thus,
the same set of eigenvectors that solve (34) solve (40), with eigenvalues
multiplied by K. D
Thus, the bigger problem of designing optimal markets for all N K people
collapses to the simpler problem discussed in the preceding section. Note
that since H is the same rank as H, there are no more nonzero eigenvalues,
the presence of K individuals per "agent" does not introduce the need for
any more than J markets.
12 Uncertainty about preferences with a pre-existing

market portfolio
We have seen in an earlier section that if the pre-existing market is the mar-
ket portfolio, then all remaining contracts constructed, such that A'~A = I,
have a zero price, Lemma 2. It is interesting to ask what the optimal con-
tracts are if there is uncertainty about ,,'s and the market portfolio already
exists. The contract designer chooses A * to
A* E argmaxA;,En.J,n*=l, ... ,N*{Etr(A*'~MrM'~A*) IA*'~A* = IN"

A*'~Ai = 0, Ai = m}. (41)
Using (35) and noting from the contraints that A*'~~ = 0, we may rewrite
the contract designer's problem as
A* E argmaxA~En.J ,n*=l, ... ,N* {tr (A*'~E(r)~A*) IA*'~A* = I N*,

A*'~Ai = O,Ai = m}.
(42)
7 There arc an infinit.e numher of A's which will solve (40) of which A ® It is onc. All
re~mlt ill t.he Hamc equilihrium.
94
Theorem 9 The A* matrix that solves (42) has the property that Q* =
- A *'~ has columns corresponding to the N* eigenvectors with highest
eigenvalues of
'DI>E(f). (43)
Proof. Proceed as with Theorem 5.0
This theorem shows that given the expectations of f, uncertainty about
the ),'s does not affect the optimal markets when the market portfolio is
a pre-existing market. We know that the amount of uncertainty, Theorem
6, or diversity, Theorem 8, of the ),'s affects the optimal contracts if the
market portfolio is not pre-existing. As such one reason to construct the
market portfolio first is that the remaining markets' definitions are robust
to misspecification of the uncertainty or diversity of )"s.
13 Empirical application
13.1 Estimation method
The method for estimating our contracts comes from Shiller and Athana-
soulis [1995] NBER paper no.5095 and the theory presented above. We
assume that each representative agent j has the utility function
T
" Utj (44)
Uj = L...t (1 )t'
t=l +p
We define felicity Utj as a function of mean and variance
Ytj )'j var(Ytj)

(45)
Utj = (YOj(l + g)t)"Yj - '2 (YOj(l + g)t)hi+1) '
where Ytj is the mean, the expectation conditional on information at year
o (the contract date) of country j's real per capita income and var(Ytj) is
the variance conditional on information at year 0 of country j's real per
capita income at year t. )'j is the coefficient of relative risk aversion of the
agents in country j./3
The contract designer chooses the optimal contracts to construct in the
economy given she knows the agent's optimal decisions. In order for the
theory above to fit with this empirical application, we must take into ac-
count the population of each country and the intertemporal utility function
(44). To do this, the contract designer's problem is slightly augmented. The
contract designer's problem is
AEargmaxA n €RJ,n=l,oo.,N {tr E(A'~MfPOPM'~A) IA'~A = IN}, (46)
'" One can obtain this hy taking a second order Taylor expansion of a constant relative
risk aversion utility function and making some simplifying assumpt.ions.
95
where POP is the JxJ diagonal matrix with the population of the lh
country in the lh diagonal position. This problem is derived by the contract
designer maximizing the sum of the utilities of the agents in the world. The
A matrix is the J x N matrix whose nth column weights the demeaned per
capita incomes of the countries in the analysis. The intertemporal part of
this problem is taken into account in the covariance matrix and is described
in the Estimation section. The basic first order condition to choose the A
matrix is
E [~MPOprM'~A] = ~AA, (47)
where M == I - POP u'r-1 (t' PO pr- 1t) -1 and ~ is the covariance matrix
of per capita incomes. We may rewrite the first order condition as
~POPE(r)~A - ~POPtt' POP~AE [(t'r- 1 POPt)-1] = ~AA. (48)
r includes the coefficients of relative risk aversion for each country on the
diagonal. It is assumed that the ~ matrix, the covariance matrix of per
capita incomes, is known with certainty.
13.2 Estimation
We estimate the ~ matrix as per Shiller and Athanasoulis [1995] where we
let ~ = L~~1 ~tht where h = ((1+p)(1+g)1'+1 )-t.9 Thus we are estimating
a 20 year covariance matrix under the assumptions that all the coefficients
of relative risk aversion are the same. We also make the assumption that the
rate of discount is known. We let the rate of time preference, p = .02, and
the rate of relative risk aversion, , = 3. We assume that the average yearly
growth rate of the six countries is g = .02 and for sensitivity analysis we let
g = 0 to make sure the contracts are robust to different rates of discount
and growth.
Our data source on per capita incomes comes from Maddison [1995],
1870-1992, for Canada, Germany, France, Italy, UK and USA. The data
are constructed in real terms, in international Geary-Khamis dollars, so
that international comparisons can be accurately made. In particular it
takes into account the differences in prices of non-traded goods to measure
real income differentials.
We assume that, is lognormally distributed with E(r) = 3 and Var(r) =
0,1 and 00. To obtain E [(t'r- 1pOPt)-1] when the Var(r) = 1, we solved
numerically by drawing, from a lognormal 100,000 times and calculating
the expectation. Thus we produce six tables for estimates of the A matrix:.
Three for g = .02 and the variance of, evaluated at 0, 1 and 00 and three
for g = 0 and the variance of , evaluated at 0, 1 and 00.
9This comes directly from the utility function (44) and substituting in (45).
96
13.3 Discussion
The results are shown in the tables section at the end of the paper. First
we see that the US has the largest per capita income in the set of coun-
tries we use (in international Geary-Khamis dollars), Table 1 column l.
Furthermore we can see from the populations of the countries that the US
dominates the others, Table 1 column 2. Thus if we were to construct the
market portfolio, and we will discuss the case when 9 = .02, column three
of Table 1, the weight would be very high for the US, and almost equal
weighting for the European Union and a small weighting for Canada. From
the variance matrix, Table 2, we notice that the UK has the smallest vari-
ance of per capita income. While France, Germany and Italy have larger
variances than the US, the size of the US (population) causes the variance
of the US to dominate. We also see (from the correlation matrix, Table 3)
that the US correlates negatively (almost zero) with both Italy and France
as well as having a small (almost zero) correlation with Germany so there
should be large risk sharing opportunities there. Canada also has similar
correlations with these three European countries but its high correlation
with the United States means that it will be a small part of this risk sharing
arrangement. While France has a low correlation with the UK, the fact that
the UK's per capita income variance is low and the pOPulations of the two
countries are similar implies there will probably not be a great risk sharing
opportunity there. In short the UK from the point of view of pooling risk
is irrelevant.
The three Tables representing the A matrix are very revealing. We discuss
the results for 9 = .02, Tables 6, 7, and 8, though we find the results are
robust (i.e., the optimal contracts do not change significantly when 9 = 0).
We discuss the first two contracts for the first two Tables, 6 and 7, and
discuss the first three when the uncertainty about agents risk aversions
is infinite, Table 8. When the contract designer has no uncertainty about
the risk aversions the first contract is a swap of risk between the U.S. on
one side and France, Germany and Italy on the other side. This we can
see once again from the correlation matrix that the US has almost zero
correlation with these countries and much of this risk can be pooled. The
second contract is a swap of risk between France and Germany. France's
correlation with Germany is .37 which is higher than her correlation with
the UK which is .04. However one must remember that the UK's overall
variance of per capita income is very small, insignificant from a pooling
standpoint. As such France is much better off pooling risk with Germany
even though her correlation with Germany is higher.
When the contract designer has some uncertainty about the coefficients
of risk aversion, Table 7, the variance of the ,),'s is 1, the results of the first
two contracts are almost identical to the case where the contract designer
has no uncertainty. The first contract is a swap between the US on one side
and France, Germany and Italy on the other. The second swap is between
97
France and Germany. What is happening here is that though there is some
uncertainty about the coefficients of risk aversion, the pooling opportuni-
ties within these groups outweighs the possibility of making some error in
judgement about the risk aversions of agents. Thus making a small error
in judgement about agent's risk aversion is not very costly here since the
pooling opportunities are so large.
Finally the contracts when the contract designer has no information
about agent's coefficients of risk aversion are quite different, Table 8. The
first contract is a contract for US risk. Given no information, the contract
designer decides on a contract which will potentially affect the most people
in the world. US population dominates the set of countries in our study.
The second contract is one for France, Germany and Italy. One will notice
that if one subtracts contract two from contract one, one gets the swap of
the US for France, Germany and Italy. Thus the lack of knowledge on the
contract designer's part causes her to split up the first contract in the com-
plete information case into two contracts in the incomplete information.
Since this is the largest risk sharing opportunity in the world, it allows
agent's in the world to swap this risk as well as allow for the possibility
that the designer is making an error in judgement. It is also true that if one
were to add contracts one and two together, one would approximately get
the market portfolio. The third contract is similar to the second contract
in the complete information case, a swap between France and Germany.
Thus we see that as we move from the complete information case to the
complete lack of information case, the first contract is decomposed into
the first two, then the other contracts (generally) are ordered the same.
Furthermore the market portfolio may be spanned within the first few con-
tracts, and appears to be in this case. These results also seem to be robust
as changing the discount rate (growth rate to 9 = 0) does not signifi-
cantly alter the qualitative results. One will also notice that since the first
two contracts in the complete lack of information case summed together is
approximately the market portfolio, all other contracts are approximately
insurance contracts. That is the cost to purchase those other contracts (the
premium) must be approximately zero.
14 Conclusion and practical implications for contract

design
We have presented several alternative maximization problems for contract
designers to define optimal risk management contracts. Thus, we have sev-
eral alternative definitions of the optimal markets to create.
The simplest maximization problem, (15), is the most restrictive: it as-
sumes no preexisting markets and no uncertainty about preferences. It
yielded the striking conclusion that the contracts created would never allow
trading the market portfolio, and no linear combination of the portfolios
98
defined in the contracts could even have non-negative quantities of all en-
dowments. The question is, how restrictive are the assumptions in (15)?
Of course, we are not in a situation where there are no pre-existing mar-
kets, and so one might conclude that the alternative maximization problem
that accounts for these, (26), is the more relevant. We are, however, some-
what inclined against this view. We should not automatically assume that
we are constrained by pre-existing markets. History shows that pre-existing
derivative markets actually do sometimes wither away when another deriv-
ative market appears that serves hedgers better. lO
A more important issue is uncertainty about preferences which leads us
to problem (34), or if there are K individuals per agent, problem (40).
These lead to the same solution and so our maximization problem (34)
may be the most relevant. As a matter of historical fact, market designers
have found it very difficult to predict in advance of creating a new market
who will want to take positions in the new market. Our representation of
uncertainty about preference parameters can be regarded as a metaphor
for our difficulty in predicting investor behavior.
Thus, taking account of this uncertainty as in (34) would be of great
practical importance for contract designers. If contract designers assumed
enormous uncertainty about preferences, so that the limiting case described
in Theorem 7 applies, then, if there is a substantial market component in
the economy, one might think that something approximating the market
portfolio would be the most important market.
There are reasons to suspect that endowment (income) shocks have a sub-
stantial world component; technology shocks proliferate around the world
and economies are linked through trade. We find this to be empirically
true here. If we do not know E accurately but believe there is a strong
world component to endowments, we may want to impose on our estimate
of E a prior that this component is important. This might lead a contract
designer to construct the market portfolio. It may be noted that a pos-
sible outcome of estimating E, using (34) and specifying moderate prior
uncertainty about risk parameters would be a conclusion that something
approximating the market portfolio is not the most important new market
to create, but still one of the more important markets.
Actual markets we create should be easy to describe and understand to
ensure their success. Despite fundamental uncertainty about future vari-
ances and risk aversions, we think it is safe to advise that a market for
the market portfolio should be constructed. Even though it may not be
precisely on the list of most important markets with estimated E, it al-
lows for more robust contract definition and enhances the simplicity and
understandability of equilibrium.
10 An example of this is the demise of the GNMA CDR futures resulting from the
formation of the Treasury-Bond futures, see Johnston and McConnell [1989J.
99
Table 1
1992 Per Capita Income (in 1990 Geary-Khamis Dollars and)
Population (in OOO's) and Market Portfolio Weights (x10- s )
Country P.C. Income Population Market Market

(g = .02) (g = 0)
Canada 18159 28436 0.75 0.46
France 17959 57372 1.52 0.93
Germany 19351 64846 1.72 1.05
Italy 16229 57900 1.53 0.94
UK 15738 57848 1.53 0.94
USA 21558 255610 6.76 4.15
Table 2
20 year Covariance Matrix of P. C. Incomes (x105 ), 9 = .02
Country Canada France Germany Italy UK USA
Canada 949
France -121 1927
Germany 278 842 2732
Italy 182 1178 1238 1542
UK 257 23 428 248 222
USA 855 -498 246 -170 212 1234
100
Table 3
20 year Correlation Matrix of P. C. Incomes, 9 = .02
Canada 1.00
France -0.09 1.00
Germany 0.17 0.37 1.00
Italy 0.15 0.68 0.60 1.00
UK 0.56 0.04 0.55 0.42 1.00
USA 0.79 -0.32 0.13 -0.12 0.40 1.00
Table 4
20 year Covariance Matrix of P. C. Incomes (x105 ), 9 = 0
Canada 2487
France -184 5458
Germany 717 2843 7719
Italy 622 3700 3878 4722
UK 689 202 1229 752 594
USA 2144 -1145 540 -343 584 2931
Table 5
20 year Correlation Matrix of P. C. Incomes, 9 = 0

Canada 1.00
France -0.05 1.00
Germany 0.16 0.44 1.00
Italy 0.18 0.73 0.64 1.00
UK 0.57 0.11 0.57 0.45 1.00
USA 0.79 -0.29 0.11 -0.09 0.42 1.00
101
Table 6
A matrix with Varh') = 0, 'Y = 3, g = .02 and p = .02.
Numbers are x 10- 5
Country 1 2 3 4 5
Canada 0.30 -0.57 1.93 0.83 -19.40
France -2.17 -4.26 -8.31 1.00 -0.74
Germany -1.81 6.78 -3.20 0.63 -2.65
Italy -1.79 -1.16 9.29 -9.56 4.41
UK -0.22 -0.64 5.50 15.57 6.43
USA 5.69 -0.16 -5.20 -8.47 11.96
Table 7
A matrix with Var('Y) = 1, E'Y = 3, g = .02 and p = .02.
Numbers are x 10- 5
Country 1 2 3 4 5 6
Canada 0.31 0.45 -0.77 1.97 1.70 -11.07
France -2.15 3.77 7.74 -4.02 2.06 4.70
Germany -1.78 -6.67 1.41 -3.12 1.56 -4.18
Italy -1.77 0.84 -3.42 11.80 -6.76 -2.25
UK -0.22 0.70 -8.00 -3.97 4.93 31.11
USA 5.75 -0.56 8.17 2.96 -12.33 6.15
102
Table 8
A matrix with Var(-y) = 00, ET' = 3, 9 = .02 and p = .02.
Numbers are xW- 5
Country 1 2 3 4 5 6
Canada 0.61 0.24 1.06 -2.17 13.06 -15.45
France -0.99 1.99 6.01 7.75 2.73 3.56
Germany 0.33 3.67 -5.83 3.55 0.61 -4.32
Italy -0.42 2.28 2.65 -10.87 -8.54 -1.34
UK 0.31 0.55 -0.51 -3.51 14.08 29.34
USA 7.92 0.79 4.28 3.31 -11.99 7.89
Table 9
A matrix with Var(-y) = 0, ET' = 3, 9 = 0 and p = .02.
Numbers are xW- 5
Country 1 2 3 4 5
Canada 0.21 0.34 1.26 0.46 12.19
France -1.28 2.74 -5.31 0.04 0.71
Germany -1.20 -4.16 -1.96 0.04 1.78
Italy -1.10 0.86 6.42 -4.63 -3.18
UK -0.06 0.35 2.51 9.94 -3.67
USA 3.42 -0.13 -2.92 -5.86 -7.82
103
Table 10
A matrix with Var('Y) = 1, E'Y = 3, 9 = 0 and p = .02.
Numbers are x 10- 5
Country 1 2 3 4 5 6
Canada 0.21 0.26 -0.16 1.37 10.52 7.25
France -1.27 2.47 -3.09 -4.60 1.42 -2.57
Germany -1.19 -4.07 0.11 -2.07 0.96 2.84
Italy -1.10 0.66 -0.99 7.33 -4.32 0.98
UK -0.06 0.04 5.42 0.25 3.56 -19.33
USA 3.43 -0.64 -5.70 -0.88 -8.01 -4.03
Table 11
A matrix with VarC'Y) = 00, El' = 3, 9 = 0 and p = .02.
Numbers are xlO- 5
Country 1 2 3 4 5 6
Canada 0.35 0.27 0.66 -1.31 7.98 9.96
France -0.97 0.93 3.48 5.19 1.71 -1.82
Germany -0.44 2.01 -3.77 2.02 0.24 2.95
Italy -0.66 1.22 1.72 -6.75 -5.05 0.20
UK 0.11 0.37 -0.31 -1.93 9.27 -18.05
USA 4.69 2.12 2.46 2.40 -7.78 -5.21
104
References
Athanasoulis, Stefano, 1995, "Essays in the Analysis of Assets Markets,
Derivatives Design, International Risk Sharing and Macroeconomic Pol-
icy" Ph.D. Dissertation, Yale University.
Cass, David, Graciela Chichilnisky and Ho-Mou Wu, 1996, "Individual Risk
and Mutual Insurance", Econometrica, 64, 333-342.
Darroch, J.N., 1965, "An Optimal Property of Principal Components",
Annals of Mathematical Statistics, 36, 1579-1582.
Demange, Gabrielle, and Guy Laroque, 1995(a), "Efficiency and Options
on the Market Index," document N. 95-17, DELTA (Joint Research Unit
CNRS-EHESS-ENS) Paris.
Demange, Gabrielle, and Guy Laroque, 1995(b), "Optimality of Incom-
pelete Markets," Journal of Economic Theory, 65, 218-232.
Duffie, Darrell, and Matthew O. Jackson, 1989, "Optimal Innovation of
FUtures Contracts," Review of Financial Studies, 2, 275-296.
Duffie, Darrell, and Rohit Rahi, 1995, "Financial Market Innovation and
Security Design: An Introduction," Journal of Economic Theory, 65, 1-
42.
Geanakoplos, John, 1990 "An Introduction to General Equilibrium with
Incomplete Markets," Journal of Mathematical Economics, 19, 1-38.
Hardy, G. H., J. E. Littlewood, and G. Polya, 1964, Inequalities, Cambridge,
Cambridge University Press.
Huang, Chi-FU and Robert H. Litzenberger, 1988, Foundations for Finan-
cial Economics, New York, North-Holland.
Johnston, Elizabeth Tashjian and John J. McConnell, 1989, "Requiem for a
Market: An Analysis of the Rise and Fall of a Financial FUtures Market,"
Review of Financial Studies, 2, 1-23.
Lintner, J., 1965, "The Valuation of Risky Assets and the Selection of
Risky Investments in Stock Portfolios and Capital Budgets," Review of
Economics and Statistics, 47, 13-37.
Magill, Michael and Martine Quinzii, 1996, Theory of Incomplete Markets,
MIT Press, Cambridge, Massachusetts.
Maddison, Angus, 1995, Monitoring the World Economy 1820-1992,
OECD, Paris.
Mayers, David, 1972 "Non-MarkeTable Assets and Capital Market Equilib-
rium under Uncertainty," in Michael Jensen, editor, Studies in the Theory
of Capital Markets, Praeger, New York, 223-248.
Oh, Gyutaeg, 1996 "Some Results in the CAPM with Nontraded Endow-
ments," Management Science, 42(2), 286-93.
Okamoto Masashi and Mitsuyo Kanazawa, 1968, "Minimization of Eigen-
values of a Matrix and Optimality of Principal Components," The Annals
of Mathematical Statistics, 39, 859-863.
Sharpe, W.F., 1964, "Capital Asset Prices: A Theory of Market Equilibrium
Under Conditions of Risk," The Journal of Finance, 19,425-442.
105
Shiller, Robert J., 1993, Macro Markets: Creating Institutions for Managing
Society's Largest Economic Risks, Oxford University Press.
Shiller, Robert, and Stefano Athanasoulis, 1995, "World Income Compo-
nents," NBER Working Paper.
Willen, Paul, 1997, From Wall Street to Main Street: The Consequences of
Financial Innovation on International Trade and Individual Risk-bearing,
unpublished Dissertation, Yale University.
Does money matter? A deterministic model
with cash-in-advance constraints in factor
markets*
Erdem B8.§<;ll and ismail Saglam2
1 Department of Economics, Bilkent University, Bilkent, Ankara 06533, Turkey

2 Department of Economics, Princeton University, Princeton, NJ 08544-1021,
U.S.A.
Abstract. This paper points to the importance of finance constraints in

affecting competitive outcomes in economies that operate with money. We
study a simple dynamic economy that operates through the use of fiat
money under cash-in-advance constraints in all markets. We assume that
the labor market opens before the goods market and the total money stock
changes at a constant rate.
After describing the economy, we define and characterize its stationary
monetary competitive equilibria (Sl\ICE). In all cases where SMCE exist,
the real wage is observed to be below the marginal product of labor and
negatively related to the money growth rate, leaving positive pure profits
to the firms. Only under Friedman's optimal money supply rule, zero pure
profits are obtained.
Neutrality of money is also studied in this context. In cases where the
SMCE is unique, the equilibrium is Pareto efficient and money is neutral.
In the case of continuum of equilibria and under sticky wages, which is
consistent with market clearing in this model, the level of money stock can
affect levels of output, employment, and welfare.
JEL Classification Numbers: D52, D9, E41

Keywords: Cash-in-advance, labor markets, neutrality of money
·The second author gratefully acknowledges the hospitality of t.he Economics Depart-
ments of the University of Michigan and Princeton University, the grant awarded by the
Scientific and Technical Research Council of Turkey (TUBITAK) in scope of the NATO
Science Fellowship Programme, supports from Bilkent University and from the Cent er
for Economic Design of Bogazi~i University. A previous versiun has been presented at
the Ill. International Conference on Economic Theory and Applications, Antalya, .June
1997, and at Bilkent and Bogazi!;i Universities. We are in particular grateful to Selahat-
tin imrohoroglu, Ivan Pastine, SUbidey Togan and especially an anonymous referee for
their useful comments.
108
1 Introduction
Classical results such as Pareto efficiency of competitive equilibrium, sweep-
ing out of pure profits in case of constant returns production technologies,
and real wage of a worker being equated to its marginal product have all
been derived under the assumption that the market structure is complete.
Such a structure would prevail if either the commodity markets (Debreu,
1959, Chapter 7), or the security markets are complete (Arrow, 1964). Un-
der either institutional arrangement there is no role to be played by fiat
money. Even if the government were to create such outside money, its mar-
ket value would be zero.
One way of introducing valued fiat money in a model would be to im-
pose the requirement that at least for some goods no purchase is possible
without a specific piece of paper, and that the time horizon is infinite.
The use of such cash-in-advance constraints in macroeconomic models was
proposed by Clower (1967) and was popularized especially with the pa-
pers by Lucas (1980, 1984, 1990) and Lucas and Stokey (1983, 1987). In
these papers, cash-in-advance constraints are imposed on the consumers'
purchases of a subset of commodities or assets. However the firm, an ar-
tificial entity which has a constant returns production function, does not
face any cash constraint. So the interpretation would be that they sell their
to-be-produced goods in return for cash to consumers and pay the wage
and rental bills (partially) with this money. In that case, the classical result
of zero pure profits with marginal products being equal to factor returns
follows. An exception is a paper by Fuerst (1992) where cash-in-advance
constraints are imposed on all transactions. Fuerst (1992), following Lucas
(1990), considers a stochastic model with representative gigantic families,
members of which share the same objective of maximizing family welfare.
A firm is a part of the family and has access to money markets to borrow
their cash needs for the puchase of labor services. As a result, a deviation
of real wages from marginal product of labor takes place.
In contrast, we study a much simpler deterministic model but with two
different types of agents who do not cooperate. We too observe a deviation
of the equilibrium real wage from the marginal product of labor. This ob-
servation points to the importance of sequencing of transactions and the
corresponding nature of the cash-in-advance constraints. We assume that
first the factor markets and then the commodity markets open. In such
an environment, producers face a cash-in-advance constraint in their factor
payments. Money needed by producers here could be interpreted as working
capital. After the factor payments are made, the commodity markets open
and the factor income can be spent here, again under a cash-in-advance
constraint. This minor change in the sequencing of transactions is shown
to have an important effect on the equilibrium prices. For typical parame-
ter values, real rental rate for the single factor that is subject to a cash
constraint turns out to be strictly below its marginal product. This gap is
109
observed to widen as the money growth rate is increased. Therefore, in this

model we observe (pure) income distribution effects of inflation.
Also, under a very special case, a continuum of equilibria which can be
Pareto ranked is shown to exist. In such a case, non-neutrality of the level
of money stock itself becomes a possibility, despite the fact that we have
a deterministic model with perfect foresight individual optimization and
market clearing.
Section 2 introduces the details of the simplest possible model we could
think of, to capture the setting described above. Section 3 characterizes the
set of equilibria for various parameter values on technology and preferences.
The final section concludes by making connections to the literature on the
optimal money supply rule.
2 The model
We consider an economy in which there are two commodities at each time
t, labor Lt and a nonstorable consumption good (apple) qt. There are
two types of agents indexed by i = 1,2. Neither one values leisure,1 and
their preferences over the lifetime consumption are the same and given by
'L-:of3 t U(Cit), where f3 E (0,1) is the common discount factor, Cit is the
consumption of agent i at time t and U(.) is the common utility function for
both agents showing the instantaneous satisfaction derived from consuming
apples at each time. We assume that U is twice continuously differentiable,
U' (.) > 0 and U" (.) < O.
Agent 1 has a labor endowment L and a constant returns technology
f1(L) = L to convert lab or into apples with marginal product of labor
equal to one. Agent 2 has no labor endowment, but a better technology
fz(L) = "(L which is also constant returns to scale with marginal product
oflabor greater than one b > 1). Other than these production possibilities,
there are no endowment of apples.
We assume that real wage contracts that promise delivery of apples af-
ter the harvest are not enforceable. Therefore the economy operates with
money under cash-in-advance constraints in both labor and apple markets.
Money is perfectly storable and Mi,t denotes the money holding of agent
i at time t. We assume that initially all currency in the economy, M, is
owned by agent 2, that is, M 1 ,o = 0 and M2,O = M.2
The total currency in the economy that we denote by M t is assumed to
1 Since leisure does not enter the utility function, the factor of production in the model
could, perhaps, be better interpreted as, for example, water (for irrigation) or even bees
(for pollination). We thank an anonymous referee for suggesting these interpretations.
2 Stokey and Lucas with Prescott (1989, Exercise 5.17) indicate that even if initial
money stock were not zero, after a finite number of periods, our agent 1 would choose to
rull it down to zero. This would be the case whenever the gross real return from savings
is low compared to the gross rate of time preference 1/(3.
110
be altered every period by the government at a constant rate () > -1 so

that for all t
Mt+1 = (1 + ())Mt .
Now let Wt and Pt, respectively, denote the money wage rate and the price
of apples at time t.
The timing of transactions is as follows: Agent i starts period t with Mi,t
units of currency. Then government transfers agent 2 by paying him ()Mt
units of currency (Note here that government taxes agent 2 when () < 0).
We assume that agent 2 regards this transfer as given and not as a function
of his initial money balances. 3 First the labor market opens where labor
can be sold (by agent 1 to agent 2 due to the endowment structure) in
return for money. Then apple production takes place with the purchased
labor. After the harvest of apples, goods market opens where apple can be
sold (normally by agent 2 to agent 1 for earning money to be used next
period) for money.
Given the endowment structure described above, and the prices
(Wt,Pt)~o, we can write the infinite-horizon utility maximization problelllS
of the two agents as follows:
Agent 1
Lj3tU(CI,t)
00
(PI) max
t=O
subject to, for all t
_ - s d
ct,t-(L-Lt)+qt,
L: ~ L,
MI,t+! = MI,t + Wt L: - Ptqt,
MI,t, CI,t,qt, L: ~ 0, and
MI,o =0 is given.
Agent 2
Lj3tU(C2,t)
00
(P2) max
t=O
subject to, for all t
= 'Y L dt - qt,s
C2,t
wtLf ~ M2,t + ()Mt ,
M2,t+1 = M2,t + ()Mt - wtLf + ptqf,
3This assumption makes full sense only when there is a large number of agents in the
economy, and when the government follows a lump-sum distribution policy, rather than
a proportional one.
111
M2,t, C2,t, q: ,Lt 2: 0, and

M2,O = M is given.
The cash-in-advance constraint of agent 2 is present in its labor market
transactions and is given by the second constraint in (P2). For agent 1, the
non-negativity requirement for Ml,t, together with the third constraint in
(PI), is in effect a cash-in-advance constraint on apple purchases.
We will assume that there are large and equal numbers of each type
of agents and hence be concerned with monetary competitive equilibrium
(MCE). Under the aforementioned assumption that initially all currency
in the economy is owned by agent 2 (the entrepreneur), MCE consists of
an infinite sequence of apple prices, money wages, labor demands, labor
supplies, apple demands, apple supplies and money holdings by the two
agents such that at each date demands, supplies and money holdings are
optimal under the given wage and price sequences, demand equals supply
in both labor and apple markets, and money holdings sum up to the total
money supply at each time.
Formally, we say that (Pt,wt,Lt,Lt,qt,qt,Ml,t,M2,t)~o is a MCE if
(i) (Lt, qt, Ml,t)~O solves (PI) and (Lt, qt, M2,t)~O solves (P2)
under (Wt,Pt)~o,
(ii) Lt = Lt and qt = qt, for all t, and

(iii) Ml,t + M2,t = M t == (1 + (W M, for all t.
In addition to the MCE concept, we define the stationary monetary com-
petitive equilibrium (SMCE), where the prices, wages, demands, supplies,
money holdings are consistent with dynamic optimization and market clear-
ing, and also satisfy for all t;
(i) = wt+dWt = Mi,t+d Mi,t = (1 + 0) for all i, and
Pt+1/Pt
(ii) Lt+1/Lt = Lt+dLf = qt+1/qt = qt+dqt = 1.
In the rest of the paper we will try to characterize the set of SMCE.
3 Results
In both (PI) and (P2), we first eliminate the equality constraints by substi-
tuting for Cl ,t, qt and C2,t, qt, respectively. Now, based on the observation
that only contemporaneous values of Lt appear in the instantaneous util-
ity function at all times, it is straightforward to obtain labor demand and
supply as a function of real wage as shown in Figure 1.
If at any time t the real wage wt/Pt is less than ,,(, agent 2 will spend
all of his money on purchasing labor so that the cash-in-advance (CIA)
constraint will be binding. If wt/Pt = ,,(, labor demanded by agent 2 will
be between zero and (M2,t + OMt}/pt. If wt/Pt > ,,(, labor demanded will
be zero.
112
Lt,L't
o Pt ,,(Pt Wt
FIGURE 1. Labor supply and demand functions for a given Pt and M2,t.
So the labor demand function will be
(M2,t + ()Mt}/wt if wt/Pt < ,,(,

Lt(wt/Pt) = { ~ [0, (M2,t + ()Mt)/Wtl if Wt/Pt = ,,(, (1)
otherwise.
For agent 1, the comparison of real wage is with his own marginal product,
which is 1. If at any time t the real wage wt/Pt is higher than 1, he will
supply L, if equal to 1 he will be indifferent between 0 and L and if less
than 1 he will supply zero labor. The labor supply function can then be
written as
~
if wt/Pt > 1,
Li(w';p,) { : [0,11 if wt/Pt = 1, (2)
otherwise.
The following proposition characterizes stationary monetary competitive

equilibrium over the parameter space of ((3, "(, ()).
Proposition 1 Assume that (3 < (1 + ()). There exists a SMCE if and only
~f (3"( 2: 1 + () and is given by (3)-(7) for each t:
113
Mt+dL if f3'Y > 1 + B

(3)
(1 + B)tiiJ E [Mt+d L, 00) if f3'Y = 1 + B
(1 + B)Wt
Pt = f3'Y ' (4)
Lt = L: = Mt+dWt, (5)
qf = qt = Mt+1/Pt, (6)
M1,t = 0, M2,t = Mt == (1 + B)t M. (7)
Proof. We observe that if any two of the money, lab or and apple markets
clear, then the third one will also clear. This follows from the third equality
constraints in both (PI) and (P2). Therefore, we can eliminate qt and ql
in the problems (PI) and (P2), respectively. Similarly, eliminating C1,t and
C2,t in (PI) and (P2), respectively, as well and rewriting the inequality
constraints accordingly, we obtain
(PI') max Lf3

00
t -
U( L+ ( -Wt - 1) Lt
s 1
+ -(M1,t - M1,t+1) )
t=O Pt Pt
subject to, for all t,
L + (Wt -
Pt
1) Lt + Pt"
~(M1 t - M1 t+1) 2: 0,
S
L: ::; L,
M1,t+1 ::; Ml,t + Wt L:,
M 1 ,t, L: 2: 0, and
M 1,0 = 0.
(P2')
subject to, for all t,
( 'Y - -Wt) Ltd 1

+ -(M2 t + BMt - M2 t+1) 2: 0,
Pt Pt' ,
M2,t+1 2: M2,t + BMt - WtLt,
WtLt ::; M2,t + BMt ,
M 2 ,t, Lt 2: 0, and
M 2,0 =M.
114
For money market clearing, we must have M 2,t = M t == (1 + O)t M in

SMCE since Ml,t = 0 for all t. Let us show that such a money holding plan
is optimal for agent 2.
First note that (1) and (2) imply that the labor market is in equilibrium
only if 1 ::; wt!Pt ::; ,.
We will first show there is no SMCE with wt!Pt =,.
Wt/Pt = , into the objective function of (P2') to obtain
To see this, insert
The Euler equation for agent 2, associated with control (l/pt)M2,Hl, is

given by
U'(C2,t} = f3~U'(C2,t+1)' for all t. (8)
PHl
At a stationary equilibrium, C2,t+1 = C2,t for all t. Since U'(.) > 0 and
f3pt!Pt+1 = f3/(1 + 0) < 1 in SMCE (by the assumption f3 < 1 + 0),
it follows that (8) cannot hold in SMCE, contradicting that wt!Pt = ,
supports SMCE.
Now suppose Wt/Pt E [1, ,). The CIA constraint in the labor market will
be binding for agent 2, so we can substitute Lt = (M2,t + 0M t )/ Wt into the
objective function of (P2') to obtain
max ~
L.)3 t U(M2
,'t + OMt + -(M2,t
1 + OMt - M2,t+l) ) .
t=O Wt Pt
In this case, the Euler equation, associated with control (1/pt)M2,Hl, is

given by
(9)
At a stationary equilibrium, C2,Hl = C2,t must hold for all t. Since U'(.) > 0,
it then follows that wt+dpt = f3, in SMCE. Using WHl = (1 + O)Wt, we
obtain wt!Pt = f3,/(1 + 0). Note that wt!Pt <, in SMCE, since f3 < 1 + 0
by assumption.
If f3, > 1 + 0 then wt!Pt > 1 and Ll = Lt = L, implying that
Wt = (M2,t + OMt)/L = Mt+dL for all t. It then follows that Pt =
(1 + O)wt!(f3,) = M H 2/(f3,L) for all t, and qt = qt = (M2,t + OMt)/pt =
f3,L/(1 + 0) for all t.
If f3, = 1 + 0 then Lt E [0, L]. SO, for any wage sequence Wt = (1 + O)tw
with w E [(1 + O)M/L, oo) the labor market will be in equilibrium with
Lt = Ll = Mt+dWt = (1 + O)M/w for all t. The apple price will be
Pt = (1 + O)tw and the goods market will be in equilibrium with qt = qt =
MHdwt = (1 + O)M/w for all t.
115
Finally, note that the transversality condition for agent 2 is satisfied,

since we have
= lim 3 t -yLt U'

t-oo' Alt+l
((1 - -L)
1+B
-yL d)
t
= lim ( (3)t (1 -yLt U' ((1 (3) Ld )

t-oo 1 TB + B)i\;! - 1 + 0 -y t
= 0,
using the fact that in SMCE Lt+l = Lt ~ L for all t and the assumption
that f3 < 1 + B.
The Euler equation (9) together with the transversality condition above
verifies that agent 2 is maximized at the described equilibrium. One can
easily check that the inequality constraints in both (PI) and (P2) are also
satisfied, which completes the proof. 0
The most striking feature of this result is that in the whole class of SMCE
proposed by (3) - (7) the real wage rate f3-y/(1 + B) is below the marginal
product of lab or -y in the superior plant. So, both agent 1 (the owner of
the inferior technology, and de facto the worker) and agent 2 (the owner
of the superior technology, and de facto the enterpreneur) are better off in
the monetary equilibria relative to an autarky. This result, however, rests
upon the assumption that the contracts on apples are not enforceable. In
an economy where commodity contracts are enforceable, the entrepreneur
would be left with zero profits, since competition under the absence of
finance constraints would equalize the real wage rate with the marginal
(and in this case also the average) product of labor.
Another observation is that the apple price in the equilibrium is decreas-
ing in the marginal product of labor -y and in the subjective discount rate
f3 and increasing in the inflation rate. That is, in societies that are more
patient or more productive, workers buy the apple at a lower price and
consume more of it. Similarly, inflation is observed to affect the workers
welfare adversely by reducing equilibrium real wages. If inflation is too
high or if there is too much discounting or insufficient productivity, then
the monetary equilibrium may break-down.
A final observation is about the neutrality of the initial level of money
stock. In situations where f3-y/(1 + 0) > 1, so that wt/Pt > 1, money
is trivially seen to be neutral. However, in the case of wt/Pt = 1, labor
supply becomes infinitely elastic, and hence the nature of the stationary
monetary competitive equilibria allows for situations where money is not
neutral as well as situations where it is. That is, when the nominal wages
Wt are exogeneously determined in the interval of [M2,t+1/L, 00), satisfying
Wt+1/Wt = (1 + 0), an increase in the initial money stock, M, in the econ-
omy gives rise to an equal amount of increase in lab or transacted and hence
quantity of apples produced, which can be seen from equations (5) and (6).
116
But, since the simple economic environment that we consider in this paper
does not identify an equilibrium selection mechanism for nominal wage de-
termination when wt!Pt = 1, one may equally be justified in claiming that
nominal wages and prices move together with money holdings, leaving the
real side of the economy unaffected. So, we have a situation of indetermi-
nacy, which certainly involves an extreme yet an interesting possibility for
non-neutrality of money in the context of a model with maximizing agents,
perfect foresight and market clearing.
4 Conclusions
In this paper, we examined a very simple model in which cash-in-advance
requirements for production affect the functional distribution of income.
We introduced some heterogeneity by allowing for two types of representa-
tive agents, a worker type and an entrepreneur type. In order to attain the
stationarity of equilibrium, we distributed all the initial fiat money to the
entrepreneurs. As a consequence, these agents are observed to become em-
ployers in equilibrium and to enjoy higher profits than what would prevail
in an Arrow-Debreu economy.
When there is money growth in the form of lump-sum transfers to the
enterpreneurs, the equilibrium real wage is observed to be adversely affected
by the resulting inflation rate. Similarly, the real wage is adversely and
favorably affected by the subjective discount factor and the productivity
of labor respectively. If these three factors tend to draw the equilibrium
below the reservation real wage of the workers, the monetary equilibrium
breaks down and an inefficient situation of autarky may prevail. Such a
case would be observed, for example, under very high inflation rates.
When there is deflation through lump-sum taxation of the entrepreneurs,
however, the real wages become higher. And in fact under the optimal
money supply rule, the Arrow-Debreu equilibrium is restored. This rule,
which consists of equating the rate of monetary contraction with the sub-
jective discount rate (i.e. () = (3 - 1), was suggested by Friedman (1969)
and was later studied in the context of cash-in-advance economies where
good markets open first, but leisure enters the utility function (see, for ex-
ample, the survey by Woodford, 1990). Here we show that this rule is also
applicable in economies where factor markets open first.
We also observe an indeterminacy of equilibria under very specific para-
meter values. This corresponds to the case where real wage is as low as the
reservation wage of workers, so that the labor supply is infinitely elastic.
This allows for situations where unexpected jumps in the nominal money
supply may not be neutral, and may favorably affect total output.
The results of the paper are sensitive to the sequencing of markets. If
we were to distribute all of the initial money to workers, and to let the
goods market open first, the Arrow-Debreu allocations would be obtained.
However, we would like to argue that the "factor markets first" scenario is
117
more realistic, since sales of goods before production takes place (via for-
ward commodity contracts) are observed very rarely in real life economies.
References
Arrow, K.: The role of securities in the optimal allocation of risk-bearing.
Review of Economic Studies 31, 91-96 (1964)
Ba§~l, E.: Markets as institutions. Unpublished Ph.D. dissertation, Bilkent
University, Department of Economics (1995)
Clower, R.W.: A reconsideration of the microfoundations of monetary the-
ory. Western Economic Journal 6, 1-8 (1967)
Debreu, G.: Theory of value. New York: WHey (1959)
Friedman, M.: The optimum quantity of money and other essays. Chicago:
Aldine (1969)
Fuerst, T.S.: Liquidity, loanable funds and real activity. Journal of Mone-
tary Economics 29, 3-24 (1992)
Lucas, R.E., Jr.: Equilibrium in a pure currency economy. Economic Inquiry
18, 203-20 (1980)
Lucas, R.E., Jr.: Money in a theory of finance. Carnegie-Rochester Confer-
ence Series on Public Policy 21, 9-46 (1984)
Lucas, R.E., Jr.: Liquidity and interest rates. Journal of Economic Theory
50, 237-64 (1990)
Lucas, R.E., Jr. and Stokey, N.L.: Optimal fiscal and monetary policy in
an economy without capital. Journal of Monetary Economics 12, 55-93
(1983)
Lucas, R.E., Jr. and Stokey, N.L.: Money and interest in a cash-in-advance
economy. Econometrica 55, 491-513 (1987)
Stokey, N.L. and Lucas, R.E., Jr. with E.C Prescott: Recursive methods in
economic dynamics. Cambridge: Harvard University Press (1989)
Woodford, M.: The optimum quantity of money. In: Friedman, B.M., Hahn
F.H. (eds.) Handbook of monetary economics. Amsterdam, New York:
North-Holland (1990)
Welcoming the middlemen: restricting
competition in auctions by excluding the
consumers
Subir Bose l and George Deltas2
1Department of Economics, Iowa State University, Ames, lA 50511, U.S.A.

2Department of Economics, University of Illinois at U.-C., 330 Commerce West
Building, 1206 South SixthStreet, Champaign, IL 61820, U.S.A.
Abstract. We study an auction with two distinct types of potential bid-

ders: consumers who wish to purchase the item for their own consumption
and middlemen who wish to purchase the item for the purpose of reselling it
to the final consumers. Typically, the behavior of the former is studied un-
der the private values paradigm, while the behavior of the latter is studied
under the common values paradigm. We consider the possibility that both
types of bidders compete in the same auction. We show that if the middle-
men have access to a larger market of consumers than the auctioneer, then
the auctioneer may prefer to prevent the consumers from participating in
the auction.
The intuition for this result is that the presence of consumers in the auc-
tion creates a "winner's curse" effect for the middlemen. In equilibrium, the
latter win when part of t.heir customer base has relatively low valuations.
This effect makes the middlemen more conservative in their bidding when
they compete with consumers.
In the model we consider, middlemen can access a market of N consumers
by spending a marketing cost c. In the auction that the auctioneer arranges,
apart from the middlemen, only one randomly chosen consumer shows up.
We show that as long as c > 0, the auctioneer prefers the restricted auction,
under which the consumer is prevented from participating.
JEL Classification Numbers: D44, D43.
Keywords: Auctions with resale
1 Introduction
Bidders typically consist of two types: Consumers who wish to purchase the
item for their own consumption, and middlemen (who we will also refer to
as 'intermediaries') who want to purchase the item in order to resell it to one
of the consumers. In this paper we show that it may be in the interest of the
120
auctioneer to design an auction so as to systematically exclude a particular

class of bidders. This seems counter-intuitive, since extra competition is
usually beneficial to the auctioneer. More interestingly perhaps, the class
that the auctioneer prefers to exclude consists of the final consumers.
The phenomenon described in this paper is actually observed in auctions
of certain items. For example, in many real estate auctions, where the items
being sold are clearly substitutes, and hence the standard explanation in
terms of complementarity of values does not apply, properties that could
be sold separately are sold in large batches. While one could explain this
on the basis of economizing on transactions costs, in this paper we provide
an alternative explanation. 1 This explanation is based on the observation
that the presence of final consumers in the auction makes the middlemen
less aggressive in their bidding.
The crucial feature of our model is the assumption that the auctioneer
is unable to access the entire market of consumers without incurring a pro-
hibitively high cost. An intermediary, on the other hand, upon spending a
marketing cost, can attract all the consumers to any auction he might or-
ganize. If the auctioneer does not prevent anyone from participating, some,
but not all, of the consumers will show up and compete with the interme-
diaries for the item. 2 When the intermediaries have a positive marketing
cost, it is shown that their optimal strategy results in high valuation con-
sumers winning the auction. The intermediaries do not outbid a consumer
with a sufficiently high valuation because the expected resale value does
not cover the price and the marketing cost. The intuition is that the pres-
ence of consumers in the auction creates a "winner's curse" effect for the
middlemen. In equilibrium, the latter win when part of their customer base
has relatively low valuations. For the middlemen, winning is "bad news" .
This effect underlies the result of our model which states that forbidding
consumer participation always yields more revenue to the auctioneer.
There is a recent flurry of activity on models of auctions which incorpo-
rate the possibility of profitable resale. One strand of the literature pursues
explanations for resale that are based on the existence of further gains from
trade available after the object is sold via an auction. A potential source
of these gains from trade can arise from the non-participation of a subset
of the buyers in the first auction. The winner in the first auction may then
find it profitable to re-auction the item [see Haile (1996)].
Another potential source of gains from trade arises from the existence of
uncertainty about the private value of the object which is resolved after the
1 In t.his model we consider t.he case of a single it.em being sold. This allows us t.o
ahstract. completely from the effect of transactions cost.s and possible complementarity
of values.
2 In t.his paper wc concentrate on the special case where there is a single part.icipat.ing
consumer. This makes dearer the adverse effect on the auctioneer's expect.ed revenues
when int.ermediaries have t.o compete wit.h t.he final consumers.
121
first auction. A bidder who expects to value the object most highly wins
it in the first auction. When uncertainty about private values is resolved,
he may find it profitable to sell the object to one of the other bidders [see
Haile (1997)].
A third possibility of gains from trade occurs if bidder asymmetry results
in an inefficient allocation in the first price auction. The winner may find
it profitable to resell the item to one of the losing bidders [see Gupta and
Lebrun (1997)].
In all of the above models all participants are potentially final consumers:
they directly value the object that is put up for sale. Another strand of
the literature considers participants of two types: the first type consists of
participants in the first auction and are competing for the object with the
intention of re-selling them. The second type consists of the final consumers
who compete among themselves to purchase the items from the re-sellers.
Bikhchandani and Huang (1989), for instance, analyze a common value
auction where an exogenous number of bidders compete in a multiple object
auction to acquire items that will then be resold to consumers. In that
paper, the policy question is which auction format to use to award the
objects to the resellers rather than to exclude or not to exclude the final
consumers from the auction.
This paper adopts the exogenous partition of the players into re-sellers
and final consumers. Unlike the model in Bikhchandani and Huang (1989),
ours is a private values model where some of the final consumers can be
present in the first auction. Furthermore, the policy question we analyze
is whether or not the seller should allow the participation of the final con-
sumers. To the best of our knowledge, the only other paper (excluding
related work by these authors described in the conclusion) where the prof-
itability of such an exclusion is explored is an example in Haile (1996). In
that example, where the bidder types are "almost common knowledge" , it is
shown that the seller can profitably exclude one of the buyers from the auc-
tion. In this example, the competing buyers know each other's type with
probability that becomes arbitrarily close to 1 and all competing buyers
have some intrinsic value for the auctioned off object.
The rest of the paper is organized as follows: In Section 2, we describe
the model and the results. The last section concludes with a brief discussion
of possible extensions and future research.
2 The model
2.1 Modeling framework
An auctioneer has one unit of an indivisible item for sale. We assume the
auctioneer's valuation to be zero; this normalization allows us to use profit
maximization and revenue maximization interchangeably.
There are N potential consumers of the item, where N is an integer
122
strictly greater than 2.3 Each consumer i values the item at a monetary
value of Vi, where Vi is a random draw from the distribution F(v). F(v)
has continuous and non-zero density f (v) over the compact interval [Q, v] .
The valuation Vi of each consumer i is consumer i's private information.
In addition, there are M intermediaries. We assume that M is an integer
weakly greater than 2. The intermediaries are not interested in consuming
the item themselves but are willing to buy the item if they can then resell
it to some consumer, in hope of making a profit.
As we have mentioned above, a crucial assumption of the model is the
auctioneer's inability to access the entire market of consumers. More specif-
ically, we assume that if the auctioneer allows unrestricted participation in
the auction that he organizes, then all of the M intermediaries and a single,
randomly chosen consumer show up for participation. This fact is assumed
to be common knowledge. On the other hand, if an intermediary arranges
an auction, he can get all the N consumers to participate. However, to
do so, the intermediary has to spend a marketing cost equal to c dollars.
Notice that the marketing cost is incurred only after the intermediary is in
possession of the item and is going to organize an auction. We assume that
the marketing cost is the same for all intermediaries, in other words, the
marketing cost Cj for all intermediaries j is equal to c. Instead of organizing
the "unrestricted auction" described above, the auctioneer can organize a
"restricted auction". In a restricted auction, only the intermediaries are
allowed to participate and the consumer is excluded.
Both restricted and unrestricted auctions are assumed to be oral English
auctions without any reserve. Furthermore, we assume that, in the event
one of the middlemen wins the auction, he will also sell the item via an
oral English auction without reserve. That is, both the auctioneer and
the middlemen have the same selling "technology". They only differ in
their marketing "technology". We will frequently refer to the middlemen's
auction as the "second" auction.
We first analyze the auctioneer's restricted auction.
2.2 Restricted auction

The bidding strategy of middleman j is to bid up to his "valuation" of the
item. Of course, his valuation is equal to the expected resale value of the
item minus his marketing cost c.
The expected resale value of the item is equal to the expected value of
the second highest consumer valuation, E[V(2)], where
E[V(2») = LV N [N - 1) V F(v)N-2 [1- F(v)] f(v) dv.
3We cOlIlment, throughout the paper, on how the results are affected in the special
case of N = 2.
123
Hence, the bidding strategy of the middleman j is to bid until the price
in the auction reaches P, where P is given by
P = LV N [N - 1] v F(v)N-2 [1 - F(v)] f(v) dv - c.
If the marketing cost is sufficiently high then the intermediaries will

choose not to participate in the auction. This critical value of the marketing
cost, CR, is given by
CR = LV N [N - 1] v F(v)N-2 [1 - F(v)] f(v) dv.
Since all the intermediaries have the same information about the resale
value of the item and since they all have the same marketing cost, the
expected revenue of the auctioneer, ERR, is given by
ERR = LV N [N - 1] v F(v)N-2 [1 - F(v)] f(v) dv - c,
if c < CR and zero otherwise. We next analyze the unrestricted auction.
2.3 Consumer bidding strategies in the unrestricted auction

Let the valuation of the participating consumer be v. Recall that it is com-
mon knowledge that there is a single consumer bidding in the auctioneer's
auction. Therefore, the participating consumer knows that, in the event
that he fails to win the item, he will be able to compete again for it when
the winning middleman resells it. This implies that he should not bid up
to his true value v in the auctioneer's auction, even though it is an English
auction.
Lemma 1 Suppose the participating consumer's valuation is v. Then his
dominant strategy is to drop out of the auction when the price in the auction
reaches r(v), where r(v) is given by
r(v) = v - LV [v - y] [N - 1] F(y)N-2 f(y) dy.
Proof. Suppose the participating consumer has lost in the first auction.
Since he is the only participating consumer, the winner must be an in-
termediary. Then, the current consumer can bid for the same item in the
auction organized by the winning intermediary. Since this auction is also
assumed to be an English auction without reserve, the price the consumer
will pay in the event that he wins will be equal to the highest valuation of
the remaining N - 1 consumers. His expected surplus from participating
in that second auction, ES(v), is given by
ES(v) = LV [v - y] [N - 1] F(y)N-2 f(y) dy.

124
Therefore, if he wins at the auctioneer's auction and pays a price p, where

p > rev), his expected surplus will be lower than ES(v), his expected
surplus if he had stopped bidding at a price rev) and let the intermediaries
win. A similar argument shows that dropping out at a price p where p <
r( v) is dominated by bidding up to r( v). •
Notice that one can infer the consumer's valuation by observing the price
at which he stops bidding. This can be seen by demonstrating that r( v) is
an increasing function of v in [Q, vJ. Taking the derivative of rev), we have,
dr(v)
dv
1- LV [N - 1J F(y)N-2 fey) dy
- [v - vJ [N -lJ F(v)N-2 f(v)

1 - F(v)N-l
> 0
& > 0 for v E ~,v).
Thus, rev) is an invertible function of v. 4
2.4 Bidding strategies of intermediaries in the unrestricted

auction
Define R( v) to be the expected resale value of the item if the participating
consumer's valuation was known to be v. Since the second auction is also
an English auction without any reserve, RC v) is the expected value of the
second highest value of N random variables, N -1 of which are distributed
i.i.d. with cdf F(v), and one of which has a degenerate distribution that
takes the value of v with probability 1. We then have
R(v) = LV [N - 1J y F(y)N-2 fey) dy
+v [N -lJ [1 - F(v)J F(v)N-2
+ LV [N -1] [N - 2] y F(y)N-3 [1- F(y)J fey) dy.

The first term on the right hand side is the contribution to the revenue
when all of the remaining N -1 consumers have valuations less than v. The
second term accounts for the revenue when only one of the N -1 consumers
4Note that we here use the assumption that all audions are without any reserve
price. The part.icipating consumer would not be willing to reveal his true valuation if
there was a reserve price in the middlemen's auction, since the information revealed
in the firHt auction could be uHed by the middlemen to Het the ff',serve price in their
auction. Furthermore, a reserve in the auctioneer's auction would not, ill general, allow
the inference of v since there would be a pooling of types at the reserve.
125
have valuation more than v. In this case the participating consumer loses
in the second auction, but is the one who determines the price. The third
term is the contribution to revenue when two or more of the N - 1 buyers
have valuations greater than v.
To ease our exposition, we define the function w(v; c) as follows:
w(v; c) == R(v) - r(v) - c.
This is the expected profit of an intermediary upon winning the object
when the participating consumer has valuation v and the intermediary pays
a price equal to r(v). Define v* to be the value of v, such that w(v*; c) = O.
The importance of v* is that it determines the price r( v*) at which the
intermediaries drop out of the auction, making the participating consumer
the winner.
Lemma 2 '11 (v; c) is a decreasing continuous function of v. Furthermore,
~f '11(11.; c) > 0 and c > 0, then there exists a unique v* E (11., v) such that
w(v*;c) =0
Proof. Observe that r(v) can be rewritten as
r(v) v - LV [v - y] [N - 1] F(y)N-2 f(y) dy

v - v LV [N - 1] F(y)N-2 f(y) dy
+ LV y [N - 1] F(y)N-2 f(y) dy
v [1 - F(v)N-l] + LV y [N - 1] F(y)N-2 f(y) dy.
Notice that since R(v) = r(v),

w(v; c) = R(v) - r(v) - c = -c < O.
Since w(v;c) is a continuous function of v, the existence of v* when
W(Q; c) > 0 follows from the Intermediate Value Theorem.
Substituting the above expression for r(v) into the definition ofw(v;c)
we get
w(v;c) v [N - 1] [1 - F(v)] F(v)N-2 +
LV [N - 1] [N - 2] y F(y)N-3 [1 - F(y)] f(y) dy

v [1 - F(v)N-l] - C,
which, by expanding the first term, simplifies to

w(v; c) = [N - 1] v F(v)N-2 - [N - 1] v F(v)N-l +
126
l V
[N - 1] [N - 2] y F(y)N-3 [1- F(y)] fey) dy -
v [1- F(v)N-l] - C.
Taking the derivative of w( Vj c) with respect to v yields
dW(vjc)
= [N - 1] [N - 2] v F(v)N-3 f(v) + [N - 1] F(v)N-2 -
dv
[N - If v F(v)N-2 f(v) - [N - 1] F(v)N-l -
[N - 1] [N - 2] v F(v)N-3 f(v) +
[N - 1] [N - 2] v F(v)N-2 f(v) -
1 + F(v)N-l + [N - 1] v F(v)N-2 f(v),
which, by cancelling the first and sixth terms, can be written as
dw(vj c)
= [N - 1] F(v)N-2 - [N - 1]2 v F(v)N-2 f(v)
dv
- [N - 1] F(v)N-l + [N - 1] [N - 2] v F(v)N-2 f(v)
- 1 + F(v)N-l + [N -1] v F(v)N-2 f(v).
As shown below, the 2nd, 4th, and 6th terms cancel out.
- [N - 1]2 v F(v)N-2 f(v) + [N - 1] [N - 2] v F(v)N-2 f(v)

+ [N - 1] v F(v)N-2 f(v)
= [v F(v)N-2 f(v)] [(N - 1) + (N - 1) (N - 2) - (N - 1)2]
v F(v)N-2 f(v) [N -1 + N 2 - N + 2 - 2N - N - 1 + 2N]
2
= v F(v)N-2 f(v) [0]
= o.
We can now simplify the above expression for the derivative of w( Vj c)
to get
dW(vj c)
= [N - I]F(v)N-2 - [N - I]F(v)N-l - 1 + F(v)N-l
dv
[N - I]F(v)N-2[1 - F(v)]- [1 - F(v)N-l]
= [N -1]F(v)N-2[1 - F(v)]
-[1 - F(v)][1 + F(v) + ... + F(v)N-2]
[1- F(v)] [ {(N -1)F(v)N-2}
- {I + F(v) + ... + F(v)N-2} ].
When v = v, dWJ~;c) = OJ since F(v) = 1.

127
When v < v,F(v) < 1, and, therefore, we have dq,J~;c) < O. This is
because since F(v) < 1, F(v)K < F(v) for all positive integers K greater
than or equal to 2, which means 1+F(v)+ ... +F(v)N-2 > (N -1)F(v)N-2.
Finally, since W(v; c) is a decreasing function of v, v* is unique. •
The following proposition states the crucial result that it is not in the
interest of the intermediaries to always try to outbid the consumer.
Proposition 3 Suppose N > 2 and the valuation of the participating

consumer is equal to v. If WelL; c) > 0, then the intermediaries do par-
ticipate in the unrestricted auction, but drop out when the price reaches
P = min{R(v) - c, r(v*)}. Ifw(lL;C) < 0, then the intermediaries do not
participate in the unrestricted auction.
Proof. Suppose first that WelL; c) > O. Consider the case in which the price
in the auction has reached the value r( v*) and that the consumer is still
bidding. If the intermediaries continue bidding but drop out later when the
price reaches a higher value, they get a payoff of zero, which they will get
even if they drop out at the current price of r(v*). If the consumer drops
r v
out at a price = reV) where > v*, then the intermediary who wins pays
r.
a price which is no less than But by paying a price in the auction which
is at least r, the intermediary's expected profit is equal to R(V) - c r-
which is strictly less than zero. In other words, by bidding beyond a price
r(v*) to win the item, the intermediary's expected profit turns out to be
negative. Since by dropping out at a price r(v*), the intermediary gets
payoff of zero, the intermediaries should not continue bidding beyond a
price equal to r(v*). Consider next the case in which the consumer drops
out at a price rev) < r(v*). The intermediaries will then bid the price
up to R(v) - c < r(v*) where the inequality follows from the fact that
R(v) is monotonic in v and R(v*) - c - r(v*) = 0 by definition. A similar
reasoning shows that dropping out at a price strictly less than r( v*) is
weakly dominated by dropping out at a price equal to r(v*).
The second part of the proposition follows from the monotonicity of the
function W(v;c). If W(lL;C) < 0, then the winning intermediary's expected
profit is always negative. 5 Therefore, the winning intermediary will lose
money no matter what the valuation of the participating consumer is. It is
better, then, for the intermediaries not to participate in the auction at all
and get a payoff of zero. •
Remark 1 Note that it is here where we use the assumption that N is

strictly greater than 2. If N were equal to 2, then w(v; c) would be equal
to -c for all values of c. In other words, if N = 2, the intermediaries
would not participate at all in the unrestricted auction. In that case, the
auctioneer's revenue .from the unrestricted auction would be zero.
5Reeall that r(:Q) =:Q, and that 'I'(Vj c) is a non-increasing function of v.

128
Define CUR to be the value of the marketing cost for which W(:!Z.; Cu R) = O.
Observing that r(1!.) = 1!., the critical value of the marketing cost at which
the intermediaries are indifferent between participating and not participat-
ing in the unrestricted auction is equal to
CUR = LV [N - 1] [N - 2] v F{v)N-3 [1 - F{v)] f{v) dv -1!..

Notice that CUR is positive for N > 2. When N = 2, CUR = -1!. < O. This
is another illustration of the fact that when N = 2 the intermediaries do
not participate at all in the auction. Corollary 4 below follows immediately
from the comparison of the expressions for CR and CUR.
Corollary 4 The critical value of the marketing cost at which the inter-
mediaries will not participate in the unrestricted auction is lower than the
critical value of the marketing cost at which they will not participate in the
restricted auction.
Remark 2 Consider a shift in the distribution of F(v) such that the val-
uations of all consumers increase by A. Then, CUR remains unaffected. In
contrast, CR increases by A.
Finally, by observing that w{v; 0) = 0 which implies that when c = 0,

v* = V, we can state
Corollary 5 When the marketing cost, c, is equal to zero, the intermedi-
aries will always outbid the participating consumer.
The above Corollary indicates that when the marketing cost is equal to
zero the middlemen always win the unrestricted auction.
2.5 Revenue comparison of auction formats

Given the optimal bidding strategies derived above, the auctioneer's rev-
enue in the unrestricted auction, ERuR, is given by
ERuR = Lv'
[R{v) - c] f{v) dv + r{v*) [1 - F{v*)]
The first term is the contribution to the revenue when the intermediaries
win, i.e., when the consumer turns out to have valuation less than v*.
The second term is the revenue when the intermediaries drop out and the
participating consumer wins and pays a price equal to r{v*).
We are now ready to state the main result of this paper.
Proposition 6 For any level of the marketing cost, c, ERR ~ ERuR with
strict inequality holding when c E (O, CR).
129
Proof. Consider first the case when c 2: CR. It then follows from the
definition of CR and Corollary 4: that the expected revenue of both formats
is equal to zero. We next turn to the case when CUR < c < CR. For this
range of the marketing cost the intermediaries will not participate in the
unrestricted auction, but will participate in the restricted auction and bid
a positive price. It immediately follows that ERR> ERu R. Finally, let us
consider the case when c ~ CUR. Recall that for all c E [0, CUR], v* E [Q, vl.
The expected revenue from the unrestricted auction is
ERuR = LV' [R(v) - c]J(v)dv + r(v*; N)[1 - F(v*)l
= LV' [R(v) - clf(v)dv + 1~ f(v)r(v*j N)dv.
Since R(v) is an increasing function of v, and R(v*) - c = r(v*), we have
ERuR ~ LV[R(V) - c]J(v)dv
with strict inequality holding for c =J O. Notice that
LV[R(V) - c]J(v)dv = LV R(v)f(v)dv - c.
Define
A(v) = v[N - 1][1 - F(v)lF(v)N-2.
Then,
LV R(v)f(v)dv = LV[R(V) - A(v)]f(v)dv + LV A(v)f(v)dv
[R(v) _ A(v)]F(v)IV _ (V d[R(v) - A(v)] F(v)dv

~ J~ dv
+ LV A(v)f(v)dv.
Expanding the Right Hand Side of above expression yields
LV R(v)f(v)dv = LV[N - IlyF(y)N-2 f(y)dy
- LV {[N - l]vF(v)N-2 f(v)
- [N -1][N - 2]vF(v)N-3[1 - F(v)]J(v) }F(v)dv

130
+ LV v[N -1][1- F(v)]F(v)N-2f(v)dv
= LV[N - 1]vF(v)N-2 f(v)dv-
LV[N - l]vF(v)N-l f(v) +
LV [N -l][N - 2]vF(v)N-2[1 - F(v)lJ(v)dv
+ LV v[N - 1][1- F(v)]F(v)N-2 f(v)dv.
CombininK terms yields
LV R(v)f(v)dv = LV N[N - 1]vF(v)N-2 f(v)dv-

LV N[N - l]vF(v)N-l f(v)dv.
Finally, by factoring out the above expression we get
LV R(v)f(v)dv = LV N[N - 1]vF(v)N-2[1- F(v)]f(v)dv.
Therefore,
LV[R(v) - clJ(v)dv = ERR. (**)
Putting together equations (*) and (**), we have our result . •

We are now able to compare the revenues derived from the two auction
formats for all values of the marketing cost. When c 2 CR both auction
formats yield zero revenue. For CUR < c < CR the unrestricted auction
yields zero revenue but the restricted auction yields positive revenue. For
o < c ~ CUR both formats yields positive revenue, with the revenue from
restricted auction exceeding that from the unrestricted auction. Finally,
when c = 0 both auction formats yield the same revenue.
3 Conclusion and future research

In this paper we have shown that if an auctioneer has a choice between
having an auction with both middlemen and final consumers and an auc-
tion with only the middlemen participating, then, provided the middlemen
have access to a larger market of consumers, the auctioneer may prefer the
restricted auction. In fact in the paricular model we consider here, we show
131
that if the unrestricted auction results in one randomly chosen consumer

participating, then the auctioneer always prefers the restricted auction if
the middlemen have some positive marketing cost.
Bose and Deltas (1998b) derive similar results in the context of sec-
ond price sealed bid auctions, while Bose and Deltas (1998a) obtain the
conditions under which an auctioneer would prefer forbidding consumer
participation when consumer and intermediaries are probabilistically par-
ticipating in the auction. However, the most interesting question would be
to investigate to what extent the intuition of this and the other related pa-
pers carries over to a wider set of selling mechanisms. For example, it might
be worthwhile to explore under what conditions in a bargaining situation,
restricting one's outside options can be advantageous to a bargainer.6
A second interesting topic would be to study "factory outlets" j to see
when and under what conditions manufacturers choose to have a factory
outlet. We believe the intuition of the results we report might help to
explain why manufacturers usually strictly prefer to sell their items through
retail stores and let the retail stores do the final selling to the consumers,
when it is clear that the manufacturers could sell to some final consumers
as well if they wished to do so.
References
Bikhchandani, Shushil and Chi-fu Huang (1989). "Auctions with Resale
Markets: An Exploratory Model of Treasure Bill Markets," Review of
Financial Studies, 3, 311-339.
Bose, Subir, and George Deltas (1998a). "Auctions with Probabilistically
Participating Resellers and Final Consumers," University of Illinois at
Urbana-Champaign, CCBA Office of Research Working Paper No. 98-
0108.
Bose, Subir, and George Deltas (1998b). "A Sealed-bid Auction with Con-
sumers and Middlemen," manuscript.
Gupta, Madhurima and Bernard Lebrun (1997). "A Simple Model of First-
Price Auction with Resale," manuscript, Universite Laval.
Haile, Philip A. (1996). "Auctions with Resale," manuscript, University of
Wisconsin-Madison.
Haile, Philip A. (1997). "Auctions with Private Uncertainty and Resale
Opportunities," manuscript, University of Wisconsin-Madison.
Vincent, Daniel R. (1990). "Dynamic Auctions," Review of Economic Stud-
ies, 57, 49-61.
6Vinccllt (1990) provides an example of this when the seller has private information
about the value of the item that he sells.
Acyclicity of fuzzy preferences
Manabendra Dasgupta1 and Rajat Deb2
1 Department of Economics, University of Alabama at Birmingham,
Birmingham, AL 35295, U.S.A.

2 Department of Economics, Southern Methodist University, Dallas, TX 75275,
U.S.A.
Abstract. In traditional choice theory, acyclicity of preference is both nec-

essary and sufficient for existence of maximal sets defined over finite sets
and, if preferences are smooth, over compact (possibly infinite) sets. This
paper, using the intuition suggested by traditional choice theory, provides
a counterpart of the acyclicity condition when preferences are fuzzy. We
show that just as in the case of precise (nonfuzzy) preferences, the pro-
posed condition is necessary and sufficient for the existence of maximal
sets defined over finite sets and, with smooth preferences, over all compact
(possibly infinite) sets.
JEL Classification Numbers: C60, D71
Keywords: Acyclicity, maximal set, compact set, max-min transitivity
1 Introduction
In economics the theory of fuzzy sets has found applications in the areas of
consumer behavior and revealed preference (Panda and Pattanaik (1986),
Basu (1984)), measurement of inequality (Basu (1987)), and in the the-
ory of social choice (Barrett, Pattanaik and Salles (1992)), Dutta (1987),
Dasgupta and Deb (1994)). To make such applications possible, there is
an emerging ancillary literature devoted to developing fuzzy counterparts
to well known precise concepts of the type used in the type of rational
optimizing exercises that arise in economics. Two well known exact (non-
fuzzy) concepts, that of transitivity of a preference relation (representing
the objective of a rational agent) and that of a choice function (the kind
of rule that a rational agent should adopt for choosing in the presence of
fuzzy preferences), have been extended to the theory of choice with fuzzy
preferences and the analysis of their fuzzy counterparts have received con-
siderable attention in the literature (Barrett, Pattanaik and Salles (1990),
Dutta, Panda and Pattanaik (1986), Dasgupta and Deb (1991, 1996)).
Transitivity and choice are related concepts in that, for finite sets (and
in the presence of continuity conditions on preferences for infinite compact
134
sets), transitivity ensures the existence of maximal elements. However, it is

well known (see, for instance, Sen (1970), Bergstrom (1975) and Mukherjee
(1977)) that, for exact preferences, while transitivity is a sufficient but not
necessary condition for the existence of maximal elements in finite sets. A
much weaker condition, acyclicity, is both necessary and sufficient for max-
imal sets to be nonempty over finite sets and, if preferences are smooth,
over compact (possibly infinite) sets. Given that violations of transitivity
arise in many different types of problems in economics, the analysis of this
minimal requirement for "demand correspondences" 1 to exist is of some
significance. To our knowledge, no counterpart of acyclicity exists in the
standard2 literature on choice with fuzzy preferences. The purpose of the
paper is to develop a concept of fuzzy acyclicity, to compare the intuition
underlying the concept to its precise counterpart and prove that just as
in the case of exact preferences, the proposed condition is necesssary and
sufficient for maximal sets to be defined over finite sets and with smooth
preferences for maximal sets to exist for all compact (possibly infinite) sets.
2 Preliminaries
Let X a nonempty subset of a metric space be the set of alternatives. Let
n be the set of all nonempty compact subsets of X. A fuzzy binary weak
preference relation (FWPR) is a real valued function R : X2 -+ [0,1].
For all x, y EX, the relation between x and y is exact or nonfuzzy iff
{R(x,y),R(y,x)} ~ {0,1}. lithe relation between some x,y E X is exact
we will denote this by xRy if R(x, y) = 1 and not xRy if R(x, y) = O. The
asymmetric component of an exact weak preference relation over any pair
will be denoted by P.
In the standard (exact) theory of choice there are two approaches to
choice: the R-greatest approach under which for any A E n: C(A, R) = {x :
for all YEA, xRy} is picked and the maximal approach under which for
any A E n: M(A,R) = {x : for all YEA, not yRx}.3 We have argued
elsewhere (Dasgupta and Deb (1991), (1996)) in favor of the two following
rules for making exact choices in the presence of fuzzy preferences. The first
rule is based on the maximal approach (Definition (1.3) and the second on
the R-greatest approach (Definition (2.2)). We have shown (Dasgupta and
1 It is well known that, if preferences and budget sets are compact and convex, acyclic-
ity is not needed for the existence of maximal elements.
2By "standard" we mean the usual "cardinal" fuzzy set theory. In a different frame-
work, for ordinally fuzzy striet preference relations, an acycJicity condition has been
proposed by Barrett, Pattanaik and Salles (1990). It has been used in the context. of
sodal choke (rather than in the context of rational choice theory) to show the existence
of a vetoer.
3 While for exact connected preferences these two sets coincide, adopting these two ap-
proaches in the context of fuzzy preferences may give different results even if preferences
are "connected" (Dasgupta and Deb (1991)).
135
Deb (1991)) that these rules represent intuitively appealing ways of making
"fuzzy maximal" and "fuzzy R-greatest" sets exact:
Definition 1 Let R be a FWPR and () E [O,IJ.
(1.1) For all x,y E X and () E (0,1]' xPOy iJJ(R(x,y) - R(y,x)) 2: ().
(1.2) For all x,y E X and () = 0, xPOy iJJ R(x,y) - R(y,x) > 0.
(1.3) F?! all x E X and all A E n, x E Mo(A,R) iJJ for all yEA, not
yPox.
Remark. One may view min {(R( x, y) - R(y, x) ), O} as a ..:neasure of strict

domination of x over y and letting "weak domination" Ro be defined by
not yPox for all x, yE X, we can view Mo(A, R) := M(A, Ro) as being the
set of undominated alternatives for some tolerance level (). As in the exact
case, for all FWPR R and for all A E n, IAI ::; 2 Mo(A, R) i= 0. On larger
sets "cycles" in preferences can cause this set to be empty.
Definition 2 Let R be a FWPR and () E (O,lJ.
(2.1) For all x, yE X, xRoy iJJ R(x, y) 2: ().
(2.2) For all x E X and all A E n, x E CoCA, R) iJJ for all yEA, xRoy.
(2.3) Po is asymmetric component of Ro.

Remark. For exact sets, the R-greatest approach selects those elements
which are at least as good as anything else that is available. CoCA, R) :=
C(A, Ro ) recommends the selection of those elements which are at least ()
as good as anything else for some prespecified tolerance level ().4
As in the case of C(A, R), the set CoCA, R) may be empty for two reasons:
(a) preferences are cyclical (b) preferences are not connected. In standard
fuzzy set theory, an FW P R R is connected iff for all x, y EX, R( x, y) +
R(y,x) 2: 1. If we restrict the preferences to be exact, this reduces to the
usual condition: for all x, yE X, either xRy or yRx. Like in the exact case,
the standard notion of fuzzy connectedness is enough to ensure that for
all A E n and IAI ::; 2, CoCA, R) i= 0 for all () E [0, .5J. This condition,
however, cannot ensure that CoCA, R) i= 0 for IAI = 2 if () > .5. It is easy to
check that, for all R and all A E n, if IAI ::; 2 then the following condition
is necessary and sufficient for CoCA, R) i= 0.
Definition 3 For all () E (0,1], an FWPR R is ()-connected if, and only

if, for all x, y E X either R(x, y) 2: () or R(y, x) 2: ().
4Jt is easy to show that in general XP9Y implies xRoy and that, for () > .5, XP9Y
implies XP9Y' The converse of these results is, in general, not true.
136
Imposing the above condition, which is also equivalent to the usual con-
dition "for all x, yE X, either xRy or yRx" when preferences are exact, will
allow us to focus on the role played by acyclicity in ensuring the existence
of a nonempty R-greatest set.
3 Fuzzy acyclicity
While a number of alternative definitions of transitivity have been proposed
and analyzed in the literature (see Dasgupta and Deb (1996)), the most
extensively used is the notion of "maxmin" transitivity. The FW P R, R, is
maxmin transitive iff for all x, y, z EX, R(x, z) 2: min{R(x, y), R(y, z) }.5
If R is exact, it is well known (Sen (1970)) that transitivity is not neces-
sary for the nonemptiness of the R-greatest or maximal sets. The weaker
condition "acyclicity" suffices. A finite sequence Xl, X2, ... , Xk E A such
that XlPX2PX3, ... PXkPXl is a cycle of length k in P. For preference rela-
tions R, the P-acyclicity of R is the absence of cycles of length k in P for all
finite positive integers k. Alternatively, letting P* be the transitive closure6
of P, P-acyclicity of R may be defined as: for all x, y E X, xP*y implies
not yP*x. For exact relations R, this condition is necessary and sufficient
for the nonemptiness of M(A, R) for all finite A E n. Moreover, for con-
nected exact preference relations R, since M(A, R) = C(A, R), P-acyclicity
is necessary and sufficient for C(A, R) =I 0 for all finite A E n.
In extending the notion of P-acyclicity of R to fuzzy preference relations
R we will use an intuition underlying acyclicity in the exact case. In this
(exact) case the condition tells us that if there exists a chain of "strict dom-
inance" connecting Xl to Xk (i.e., XlPX2PX3,." PXk) then either R(Xl, Xk)
is "large" enough
or R(Xk, Xl) is "small" enough
The two following definitions of fuzzy acyclicity try to capture this intuition.
Definition 4 Let R be an FWPR and () E [0,1). R is Po(k)-acyclic iff for

all A E n, IAI = k,
either for all Xl,X2, ... ,Xk E A XlPOX2POX3, ... POXk implies
R(Xl,Xk) > min{R(x2,Xl),R(X3,X2), ... ,R(Xk,Xk-l)}
50bserve that if R is restricted to being exact this reduces to the usual definition of
transitivity for exact binary relations.
6 This is the smallest transitive relation having P as a sub relation and is given by: for
all X,y E X, xp·y iff there exists a positive integer k and Xl,X2 .... ,Xk E X such that
X1PX2PX3.··.PXk·
137
or for all Xl> X2, ... ,Xk EA: XIPOX2POX3, ... POXk implies R(Xk' Xl) <
max{R(xl> X2), R(X2' X3), ... , R(Xk-l> Xk)}.
We will say that R is fo-acyclic iff R _is Pe(k)-acyclic for all positive
integers k and that it is P -acyclic if it is Po -acyclic for () = 0.
Remark. Notice that if the "either" part of the condition is satisfied for
some k-tuple, the "or" part may be satisfied over some other k or m-tuple
of alternatives. Let X = {XI,X2,X3,X4} and R(x, x) = 1 for all X E X.
Also, let R(xl> X2) = R(X2' X3) = R(XI' X3) = R(X3, xt} = R(X4' Xl) =
°
R(X3, X4) = R(X4' X3) = .8 and R(X2' Xl) = R(X3, X2) = R(XI' X4) =
R(x~ X4) = R(X4' X2) = .6. One can check that for () = the relation sati-
fies Pe-acyclicity without satisfying either the "either" part or the "or" part
of the above condition everywhere. The "either" part is satisfied nontrivially
over the ordered triple (Xl> X2, X3), the ordered quadruple (X4' Xl, X2, X3)
and the "or" part over the ordered triple (X4,XI,X2). Elsewhere, the con-
dition is trivially satisfied because the hypothesis does not hold.
Definition 5 Let R be an FWPR and () E [0,1). R is Pe(k)-acyclic iff

for all A E n, IAI = k,
either for all Xl, X2,"" Xk EA: XIPOX2POX3, ... POXk implies
R(XI,Xk) > min{R(x2,xt},R(x3,x2), ... ,R(Xk,Xk-d}
or for all Xl, X2,"" Xk EA: XIPOX2POX3, ... POXk implies R(Xk' xd <
max{ R(XI' X2), R(X2' X3), ... , R(Xk-l, Xk)}.
We will say that R is Po-acyclic iff R is Po (k)-acyclic for all positive
integers k.
The reader can easily verify that both the definitions of acyclicity given
above are weaker than maxmin transitivity.
Theorem 1 Let R be an FWPR, () E [0,1]. For all finite A E n,

Me(A, R) -=10 iff R is Pe-acyclic.
Theorem 2 Let R be a ()-connected FW PR, () E (0, 1]. For all finite A E
n, Co(A, R) -=10 iff R
is Po-acyclic.
We will prove Theorem 1. The proof of Theorem 2 is similar and is

omitted.
Lemma 3 Let A E n,IAI = k and R be an exact binary relation on X.

If R is P-acyclic then for all Xl> X2, ... ,Xk E A and YI, Y2, ... , Yk E A,
XIPX2P... PXk and yIPY2P... PYk imply Xl = YI,X2 = Y2,··· ,Xk = Yk.
Proof. Let N = {j : j E {l, 2, ... , k} and Xj -=I yj}, assume to the contrary
that N -=I f/J and let
j* = minj.
jEN
138
By the choice of j*, for all positive integers m < j*, Xm = Ym. Since
Xj_ "# Yj- and {XI,X2, ... ,xd = {YI,Y2, ... ,yd, for all positive integers
m < j*, Xm = Ym and P-acyclicity imply Xj- E {Yj-+1,Yj-+2, ... ,yd. This
implies Yj- P*Xj-. Arguing similarly, we have Xj- P*Yj-. This contradicts the
P acyclicity of R . •
Proof of Theorem 1. We will first show that R is P9-acyclic implies that

M9(A, R) "# 0 for all finite A E n. Assume to the contrary that there exists
A E n such that M9(A,R) = 0. By the definition M9(A,R) := M(A,~)
this implies that there exist {Xl, X2, ... , Xk}
<;;; A and XIP9X2P9X3, ... P9Xk and XkP9XI. P9-acyclicity implies that
either
[(R(xI, Xk) > min{R(x2, Xl)' R(X3, X2), ... , R(Xk, Xk-l)} (1)
and for all j E {I, 2, .. , k - I}
R(xj+I, Xj) > min{ {R(X2, xd, R(X3, X2), ... , R(Xk, Xk-l), R(xI, Xk)}
-{R(xj+1, Xj)}}]
or
[(R(Xk, Xl) < max{R(XI, X2), R(X2, X3), ... , R(Xk-l, Xk)})
and for all j E {1,2, .. ,k-1},
R(Xj,xj+d < max{ {R(XI, X2), R(X2, X3), ... , R(Xk-l, Xk), R(Xk, Xl)}
-{R(Xj, Xj+l)} }].
We will argue that if the first of these possibilities (i.e., (1)) is true, then
we would get a contradiction. The other case is similar and its proof is
omitted.
Let 1t = [{(X2, Xl)' (X3,X2), ... , (Xk-I, Xk-2), (Xk, Xk-d} U{(XI, Xk)}].
There are two possible subcases:
(a) minR(x, y) for (x, y) E 1t is given by R(XI, Xk).
(b) minR(x,y) for (x,y) E 1t is given by R(Xj+I,Xj) for some j E
{1,2, ... ,k-1}.
(a) contradicts the first part and (b) the second part of (1).
To complete the proof we need to show that if M9(A, R) "# 0 for all
finite A E n then R is P9-acyclic. If P9-acyclicity is violated, then there
exists an A E n, IAI_= kl.- Xl, X2, ..,..:, Xk E A, XIP9X2P9X3, ... , P9Xk and
YI, Y2,·· ., Yk EA, yIP9Y2 P9Y3, ... , P9Yk such that
and
139
By its definition 7 , Re is an exact binary relation on X. Since Mo (A, R) -#

o for all finite A E n, Re is Po-acyclic. By Lemma 3, Xj = Yj for all
j E {1,2 ... ,k}. Thus, (3) can be written as
R(Xk,XI) > max{R(xI,X2),R(X2,X3), ... ,R(Xk-I,Xk)}

> max{R(x2, Xl) + 0, R(X3, X2) + 0, ... , R(Xk, Xk-l) + O}
> min{R(x2, xd + 0, R(X3, X2) + 0, ... , R(Xk, Xk-l) + O}
> R(x!, Xk) + 0,
wh~e the last inequality follows from (2). If 0 = 0, then by the definition
of Po, the second inequality in the above set of inequalities is strict and
hence we would get !!-(Xk, Xl) > R(:.!, Xk) implying Xk POXI' If 0 > 0, then
by the definition of Po, we have Xk POXI' Thus in either case, since we have
XIPOX2POX3, ... POXk and Xk POXI, our assumption that Mo(A, R) "# 0 for
all finite A E n is violated for A = {Xl, X2,.··, xd . •
To extend our existence results to infinite sets we will assume that R is
smooth in the following sense:
Assumption 1 The set W(x):= {y: R(x,y) > R(y,x)} is open in X for
all X EX.
Theorem 4 Let R be an FW PR, satisfying Assumption 1 and 0 E [0, 1].

For all compact A E n, Mo(A, R) -# 0 if! R is Po-acyclic.
The above result follows immediately from the finite intersection prop-
erty of compact sets by observing that by Theorem 1 for all finite sets
{Xl, X2, ... , Xk} which are a subset of an arbitrary compact set A, the set
n;=dNW(xj) n A] 2 MO({XI,X2, ... ,xd,R) -# 0, where NW(x) is the
complement of the set W(x) in X.
°
Observe that (a) M01(A,R) ~ M0 2 (A,R) for all ()2 ~ ()l and (b) for
02-connected R, Mo] (A, R) ~ CO 2 (A, R) for ()l = and for all ()2 E (0,1]8.
Thus, using Theorem 4, we get the following result.
Theorem 5 Let R be an FW PR, satisfying Assumption 1.
7For all x,y E x, xReY iff not yPex.

8 Assume to the contrary that ih = 0, x E Me] (A, R) and R(x, y) < (}2 for some yEA.
(}1 = 0, x E Me) (A, R) implies R(x, y) 2 R(y, x) for all yEA. By (}2-connectedness,
R(x, y) < (}2 implies R(y, x) 2 rh > R(x, y). This contradicts R(x, y) 2 R(x, y) for all
yEA.
140
(i) Mo(A, R) =J. 0 for all compact A E n and all 8 E (0,1] iff R is P-
acyclic.
(ii) Co(A, R) =J. 0 for all compact A E n and all 8 E (0,1] iff R is
P-acyclic and R is 8-connected.
References
Barrett, C.R., Pattanaik P.K. and M.Salles (1990), "On choosing rationally
when preferences are fuzzy". Fuzzy Sets and Systems 34: 197-212.
Barrett, C.R., Pattanaik P.K. and M.Salles (1992), "Rationality and aggre-
gation of preferences in an ordinally fuzzy framework". Fuzzy Sets and
Systems 49: 9-13.
Basu, K. (1984), "Fuzzy revealed preference theory". Journal of Economic
Theory 32: 212-227.
Basu, K. (1987), "Axioms for a fuzzy measure of inequality". Social Choice
and Welfare 32: 275-288.
Bergstrom, T. (1975), "Maximal elements of acyclic relations on compact
sets". Journal of Economic Theory 10: 403-404.
Dasgupta, M. and R. Deb (1991), "Fuzzy choice functions". Social Choice
and Welfare 8: 171-182.
Dasgupta, M. and R. Deb (1994), "F-decomposable social aggregation rule
and acyclic choice". Journal of Mathematical Economics 23: 33-44.
Dasgupta, M. and R. Deb (1996), "Transitivity and fuzzy preferences".
Social Choice and Welfare 13: 305-318.
Dutta, B.(1987) "Fuzzy preference and social choice". Mathematical Social
Sciences 13: 215-229.
Dutta, B., S. Panda and P.K. Pattanaik (1986) "Exact choice and fuzzy
preferences". Mathematical Social Sciences 11: 53-68.
Mukherjee, A. (1977), "The existence of choice functions". Econometrica
45: 889-894.
Panda, S.C. and P.K. Pattanaik (1988) "Demand theorem when preferences
are fuzzy". Indian Journal of American Studies 18: 139-142.
Sen, A.K. (1970), Collective Choice and Social Welfare, Oliver and Boyd,
London.
On the connection between correlated
equilibria and sunspot equilibria*
Julio Davila
Departamento de Economfa e Historia Econ6mica, Universidad Aut6noma de

Barcelona, 08193 Bellaterra, Spain
Abstract. The conjecture stated in Maskin-Tirole (1987) about the exis-

tence of a close connection between correlated equilibria of market games
modelling economies with asymmetric information about an extrinsic un-
certainty and finite markovian stationary sunspot equilibria of related over-
lapping generations economies is reconsidered in this paper. It is shown that
there is no robust example of an overlapping generations economy with a
non-negligible set of finite markovian stationary sunspot equilibria which
can be identified with correlated equilibria of the corresponding economy
a la Maskin-Tirole, and vice versa. This result qualifies the conjectured
connection between the two equilibrium concepts in these two frameworks.
JEL Classification Numbers: D52, ElO
Keywords: Correlated equilibrium, sunspot equilibrium, overlapping gen-

erations economy
1 Introduction
The concepts of correlated equilibrium [Aumann (1974, 1987)J and sun-
spot equilibrium [Shell (1977), Cass-Shell (1983)J have been put in relation
with each other very often in the economics literature since actually both
cope with the role of extrinsic uncertainty on the outcomes of simultaneous
optimizing behavior. As pointed out in Forges-Peck (1995), the difficulty for
the assessment of the precise link between them comes from their original
development in different frameworks: on the one hand a game-theoretic one
for the correlated equilibrium concept and on the other hand a competitive
• I t.hank Eric Maskin, Piero Gott.ardi, .Tames Bergill, Karl Shell and .Tallies Peck
for t.heir remarks and discussions on t.his subject., awl Andnm Mas-CoIell for useful
comlllent.s. This wsearch was st.art.ed while visit.iug Harvard Universit.y as Post.-doctoral
Fellow awl has since t.hen benefit.ed from remarks from at.t.endant.s at. several conferences
mul seminars. FUlHling from t.he Spanish M illist.ry of Educat.ion and Science from t.he
research projects PB92-0120-C02-091, PB96-1160-C02-02 and a post.-doctoral fellowship
is grat.dully acknowledged.
142
one for the sunspot equilibrium idea. Comparisons between them have been
made by means of translating one of the two equilibrium concepts into
the world of the other, either considering a game-theoretic avatar of the
sunspot equilibrium in market games a la Shapley-Shubik (1977) [see Peck-
Shell (1991), Forges-Peck (1995)J or considering a competitive version of the
correlated equilibrium idea [see Maskin-Tirole (1987)J.
In Maskin-Tirole (1987) it was conjectured that the concepts of a (finite-
states markovian stationary) sunspot equilibrium in a simple overlapping
generations economy and a competitive version of a correlated equilibrium
in the economy with asymmetric information they study may well be closely
connected. Specifically, under some conditions, a sunspot equilibrium of
the former economy could be seen as a correlated equilibrium of the corre-
sponding latter economy and vice versa. The goal of this paper is to make
precise what the conditions are under which this is so and, accordingly, to
assess the degree of equivalence between the two concepts in the particular
frameworks of overlapping generations economies and the one considered
in Maskin-Tirole (1987).
The investigation of this issue closest to the one developed in this pa-
per is Forges-Peck (1995). In that paper the authors study the connection
between the two equilibrium concepts by means of a market game a la
Shapley-Shubik which mimics the overlapping generations economy. They
get an equivalence result under the assumption of a continuum of players at
each stage and a general sunspot process which may not be markovian. As
for markovian sunspot equilibria, the scope of the equivalence shortens but
is still shown to hold for sunspots fluctuating between two values. Interest-
ingly enough, the case of sunspots taking just two values is the only one in
which the lack of robustness of the equivalence of markovian sunspot equi-
libria to correlated equilibria shown in Section 3 does not hold for reasons
which will be apparent in the proofs.
The main statements of this paper are in Propositions 1 and 2, in Sections
3 and 4, respectively. Their message can be summarized as follows: arbitrar-
ily close to any overlapping generations economy, there is another economy
for which the set of markovian stationary sunspot equilibria, equivalent to
some correlated equilibrium of a Maskin-Tirole economy is negligible, and
conversely. Auxiliary lemmas are collected in the Appendix.
2 The OG economy and the MT economy

The discussion that follows compares sunspot equilibria1 of an overlapping
generations economy a la Samuelson (1958) and correlated equilibria of the
economy with asymmetric information studied in Maskin-Tirole (1987).
Specifically, the overlapping generations economy considered is the sim-
1 More precisely, finit.e states markovian stat.ionary sunspot equilibria.

143
plest one: it consists of countably many generations of a representative

agent, each living for two consecutive periods, one consumption commod-
ity per period which can be produced with the leisure endowment of the
contemporary "young" agent by means of a linear technology 2 with pro-
ductivity 1 and which is consumed by the contemporary "old" generation
in exchange of their "money" holdings. Besides, at each period a sunspot
signal (i.e., carrying no information on the fundamentals of the economy)
takes one out of k possible values following a first order Markov chain and
is publicly observed by the contemporary agents who, according to a shared
belief on a perfect correlation between sunspots and prices, may make their
decisions contingent to the future values of the sunspot (see, for instance,
Azariadis-Guesnerie (1986) for details) and their expectations on future
prices turn out to be nonetheless rational at a sunspot equilibrium.
The economy in Maskin-Tirole (1987) consists of two agents who have
to decide how much to produce of a commodity (each of them producing
a distinct commodity) with their leisure endowment in order to exchange
it for the commodity produced by the other agent 3 , with imperfect in-
formation about which is the state of the world. The state of the world
is completely characterized by the values taken by two signals taking ran-
domly a finite number k of values with a joint probability distribution. Each
agent observes one signal distinct from that observed by the other agent.
Even though the uncertainty is extrinsic (i.e., innocuous for the fundamen-
tals of the economy), since the exchange value of the commodity produced
by one agent depends on the quantity produced by the other agent of his
own good, if the agents make their decisions contingent on the state of
the world, the prices and allocation may turn out to be contingent on the
extrinsic uncertainty and thus their expectations rational at a correlated
equilibrium (see Maskin-Tirole (1987) for details).
3 From sunspot equilibria to correlated equilibria

Let U denote the class of C 2 , strictly concave real-valued functions u on
lR 2 + monotone in the sense 4 that D 1 u(x) < 0 and D 2 u(x) > 0 at any x E
lR!+, endowed with the topology of C2 uniform convergence on compacta.
Given a utility function u E U, if h, l2 > 0 distinct and a Markov matrix
mu
M= ( (1)
m21
2 Production is actually inessential. The economy is formally equivalent to an exchange

economy.
3Their ut.ility depends only 011 their consumption of the good produced by the other
agent and t.heir OWIl leisure.
4 In this sense, the first argument of a utility funcion of this class should he seen as
labour supply and the second one as consumption.
144
satisfy the two equations 5
mll (DIU(h, ld + t7-D2U(h, h))+

(2.1)
mI2(D1u(h, l2) + t7D2U(h, l2)) = 0,
m21 (DIU(l2' h) + t;D2u(l2,h))+

(2.2)
m22(Dlu(l2' l2) + ~D2U(l2' l2)) = 0,
then (l, M) E lR~+ x (0,1)2 constitute a (2-states markovian stationary)

sunspot equilibrium -a 2-SSE from now onwards- of a simple overlapping
generations economy -an OG economy henceforth- (see Azariadis and
Guesnerie (1986) for details and characterizations of its existence 6) with
a representative agent with utility function u and li being the amount of
lab or supplied by the young agent when the sunspot signal takes the value
i = 1,2. The k equations analogous to equations (2) but defining a k-SSE
for any integer k 2: 3 can straightforwardly be written (in general, the
support of a k-SSE has to be an off-diagonal point of lRi+, i.e., with at
least two components distinct). Notice that for the sake of determining its
k-SSE, an OG economy is completely characterized by the utility function
u, thus we will speak of a u-OG economy in what follows.
As it was remarked in section 5 of Maskin-Tirole (1987), these conditions
are very similar to those that a correlated equilibrium of the economy they
consider has to satisfy. Specifically, W, l2) E lR~ x lR~+ and the joint
probability distribution matrix for two signals sI, /i,
taking one out of two
values each,
(3)
is a correlated equilibrium W, l2, IT) -a 2-CE henceforward- of the 2

periods, 2 commodities economy with asymmetric information in Maskin-
Tirole (1987) -the MT economy, from now onwards- with two agents
i = 1,2, each with a utility function u i of the class U and supplying an
amount l!i
of labor whenever agent i (privately) observes the value Si = 1,2
of the signal he receives about the state of the world, if they satisfy the
four equations (see Maskin-Tirole (1987) for details and characterizations
5 Here, as well as in what follows, ratios trivially equal to 1 are made explicit for the
symmetry of the equations to become more apparent.
6Equat.ions 2.1-2.2 result from the utility maximization first order conditions and the
feasibility of the equilibrium allocation condition.
145
of its existence 7)
71"11 (DIU 1(lL m+ ~D2Ul(lL m)+

1 (4.1)
7I"12(Dlul(lt,l~) + #D2U
1
1(lL l~)) = 0
71"21 m
(DIUl(l~, + ~D2Ul(l~,ln)+
2 (4.2)
7I"22(DIUI(l~,l~) + #D2Ul(l~,l~))
2
=0
7I"11(DIU2(l~,tD + ~D2u2(l~,lD)+
1 (5.1)
7I"21(DIU2(l~,l~) + #D2u2(l~,l~))
1
=0
7I"12(DIU2(l~, lD + ~D2U2(l~, lD)+

2 (5.2)
7I"22(DIU2(l~,l~) + aD2u2(l~,l~)) = o.
Again the 2k equations analogous to equations (4)-(5) but defining a k-
CE for any integer k ~ 3 can straightforwardly be written too. Equations
(4) (respectively, equations (5)) seem at first sight to be the equilibrium
equations characterizing a sunspot equilibrium of an overlapping genera-
tions economy with identical agents with utility function u 1 (respectively,
u 2 ) and a Markov matrix for the sunspot signal given by the conditional
distribution of 8 2 given 8 1 (respectively, 8 1 given 8 2 ). Notice, nevertheless,
that a careful inspection of the equations reveals that this is not exactly
so, and that they are different from equations (2) characterizing a sunspot
equilibrium by the fact that the supply of labor is now indexed not only by
the value of the signal observed by each agent, but by the identity of the
agent who observes it as well. Notice too that for the sake of determining
its k-CE, an MT economy is completely characterized by the utility func-
tions u 1 and u 2 . Thus we will speak of a (ul,u 2)-MT economy from now
onwards.
A natural question arises from the comparison of the two sets of con-
ditions (2) and (4)-(5): Is there a correlated equilibrium corresponding or
"equivalent" to any sunspot equilibrium given by equations (2) as conjec-
tured in Maskin-Tirole (1987)? To begin with, the question itself needs to
be clarified. Notice that for our purposes a u-OG economy can be iden-
tified with a function u of the class U, while a (u 1,u2)-MT economy can
be equally identified with a couple of such functions ut, u2 • Thus, what is
7Equations (4) result from the utility maximization first order conditions of the agent
"observing the rows" of IT taking into account the feasibility of the equilibrium allocation
condition. Similarly for equations (5) and the agent "observing the columns". Notice that
the denominators have been dropped with no harm in the computation of conditional
probabilities.
146
exactly meant by the "equivalence" between the equilibria of two such dif-
ferent economies? If we are given a sunspot equilibrium of a u-OG economy,
which is the (u 1 , u 2 )-MT economy one of whose correlated equilibria is the
"equivalent one"? If we interpret an overlapping generations economy as
a sequence of agents which only trade with those two who are contiguous
to them, the representative agent cares only about the rate at which he
gives away, say, right-handed good in exchange for the left-handed good,
and he does not bother at all whether the hand which receives his right-
handed good belongs to a distinct agent than the hand which gives him
left-handed good, as it is indeed the case in an OG economy, or to the
same agent. The same is true for any of the two agents of the Maskin-
Tirole economy, they just care about the terms of trade. This allows to
identify, as it was implicitly done in Maskin-Tirole (1987), any u-OG econ-
omy with the corresponding (u, u)-MT economy, i.e. the one where both u 1
and u 2 are equal to u. Therefore, by "a correlated equilibrium equivalent
to a given sunspot equilibrium" of a u-OG economy will be meant one of
the (u,u)-MT economy such that its allocation of resources is related to
the allocation of the sunspot equilibrium as follows: l1 = h and l~ = l2, for
all i = 1,2, i.e., such that the correlated equilibrium treats symmetrically
both agents. We will speak of two such equilibria as "giving the same allo-
cation of resources" , although we should keep in mind that, as a matter of
fact, the two economies, and hence their allocations of resources, are quite
distinct.
An answer to the question above is given by the next proposition.
Proposition 1 There is no robust overlapping generations economy whose

markovian stationary sunspot equilibria that both fluctuate between more
than 2 states, and are equivalent to some correlated equilibrium of the corre-
sponding Maskin- Tirole economy, constitute a non-negligible subset of that
class of sunspot equilibria8 •
Before making this claim precise, let us define a regular k-SSE (l, M) E
~~+ x (0, 1)k2-k of a u-OG economy to be a k-SSE such that the perfect
foresight steady state is not in its support. Notice that with the assumptions
made on u this implies that for all i = 1, ... , k, DIU(li' li) + D2U(li' li) f 0,
otherwise li would be the steady state labor supply. Proposition 1 above is
but the interpretation of the following fact 9 •
Proposition 1 (restated) For any u in a dense subset of the set of func-
tions in U for which the u-OG economy has 3-SSE, the set of 3-SSE equiva-
8There is a good reason for not. to try to simplify stating t.he seemingly equivalent
statement t.hat typically such set is negligible. The reason is that the usual sense given
to "typically" would mean in this case "for any economy in a dense and open subset of
the space of economies". Density is what is proved in Proposition 1, and not openness.
9 St.at.ed specifically for k-SSE with k = 3. As for k > 3 see the concluding remarks.
147
lent to some 3-CE of the corresponding (u, u)-MT economy is null measure
in the set of all 3-SSE in a neighborhood of any regular 3-SSE.
The limitation of scope in the restatement of Proposition 1 to neigh-
borhoods of regular 3-SSE is not very constraining, since most 3-SSE are
regular in the sense that, while the set of regular 3-SSE is locally a 6-
dimensional manifold, that of non-regular 3-SSE is locally contained in a
finite union of 5-dimensional manifolds (see Lemma 7).
Proof. Let u be of the class U and (l, M) be a regular 3-SSE of u-OG. Then
the set of 3-SSE is locally a manifold of dimension 6 in a neighborhood of
(l, M) (see Lemma 3). For this 3-SSE to be equivalent to a 3-CE (l, l, Il)
of (u, u)-MT, its correlation matrix Il should generate, either by rows or
by columns, the Markov matrix M and satisfy the equilibrium conditions
for a 3-CE not already implied by those of the 3-SSE. That is to say, there
should exist 7rij E (0,1), for all i,j = 1,2,3 such that
3
L 7rij =1 (6)
i,j=1
(7)
3
L 7rjdij(l) = 0, Vi = 1,2,3, (8)
j=1
where fij is defined on Rt+ and such that
Should there be such a Il, the previous system in 7rij would have to be
linearly dependent, which can be proved to be equivalent to the singularity
of the matrix A = (mjdij(l)). But letting B = (mij/ij(l)) (notice the dif-
ferent order in the indices of the probabilities of transition in the entries of
A and B), the equilibrium equations of the 3-SSE guarantee the singularity
of this matrix 10, which implies that the singularity of A is equivalent to
the fulfillment of
(m13 m 21 m 32 - m12m 23m31)(J12(l)f23(l)h1(l) - i13(l)f21(l)h2(l)) = 0.

(9)
That is to say, should (l, M) be a 3-SSE of u-OG equivalent to a 3-CE of
(u, u)-MT, then either
(10)
lOSince they actually state that (1,1,1) is in the null space of B.

148
or
(11)
But equations (10) and (11) are both transversal at (I, M) to the equations
defining the 3-SSE for any u in a dense subset of the functions in U for
which u-OG has 3-SSE (see Lemmas 4 and 5). Therefore, the subset of
3-SSE equivalent to a 3-CE is contained in the union of two manifolds of
dimension 5 embedded in the manifold of dimension 6 constituted by the
3-SSE in a neighborhood of (I, M). The null measure of 3-SSE equivalent
to a 3-CE follows immediately. •
4 From correlated equilibria to sunspot equilibria

Going the other way round, in order to obtain a sunspot equilibrium (specif-
ically, a 2-SSE of some u-OG economy) from a given correlated equilibrium
of a (u 1, u 2 )-MT economy, the latter must give the same allocation to both
agents, as can easily be seen by comparing equations (2) and equations (4)
(or (5» in Section 3.
Thus, in order to interpret equations (4) as the conditions characterizing
a 2-SSE of the u1-OG economy necessarily the equations It = l~ and l~ =
l~ must hold. The same is true if equations (5) are to be interpreted as
the conditions of a sunspot equilibrium of the u 2 -OG economy. Hence,
there are two sunspot equilibria associated to a (u 1, u 2 )-MT economy (one
corresponding to equations (4) and the other to equations (5» if and only
if li = li and l§ = l~ hold for a given correlated equilibrium of it. Notice
that, as a matter of fact, if u 1 and u 2 are distinct utility functions each
of these two sunspot equilibria is an equilibrium of a different overlapping
generations economy, either one with identical agents with utility function
u 1 or another one with identical agents with utility function u 2 . The same
argument applies for any k-SSE with k ~ 2. As a conclusion, the only
correlated equilibria of a (u 1, u 2 )-MT economy that can be interpreted
as finite markovian sunspot equilibria of some u-OG economy are those
which treat symmetrically both agents. This allows us to state the next
proposition.
Proposition 2 There is no robust Maskin- Tirole economy whose finite

correlated equilibria which are equivalent to finite markovian stationary
sunspot equilibria of a corresponding overlapping generations economy con-
stitute a non-negligible subset of that class of equilibria.
The previous proposition is but the interpretation of the following fact. 11
Proposition 2 (restated) For u 1 , u 2 in a dense subset of the set in U x U
11 Stated specifically for k-CE with k = 2. As for k > 2, see the concluding remarks.
149
for which the (u l , u 2 )-MT economy has 2-CE, the set of symmetric 2-CE
is null measure in the set of all 2-CE.
Proof. Let ut, u 2 be of the class U and let (l, l, II) be a symmetric 2-CE
of (ut,u 2 )-MT. Then the set of 2-CE is locally a manifold of dimension 3
in a neighborhood of (l, l, II) which is moreover, for any (u l , u 2 ) in a dense
subset of U 2 , transversal at (l, l, II) to the linearly independent equations
if = q, for i = 1,2, satisfied by symmetric 2-CE (see Lemma 6). Thus
the set of symmetric 2-CE of (ut, u 2 )-MT is a manifold of dimension 1
embedded in the manifold of dimension 3 of 2-CE in a neighborhood of
(l, l, II). The null measure follows immediately. •
As a final remark, notice that if both utility functions of (u l ,u2 )-MT
are the same, then the two k-SSE associated to any symmetric k-CE are
equilibria of the same overlapping generations economy. Nevertheless, these
two sunspot equilibria need not be the same: although they necessarily have
the same allocation, their Markov matrices may be distinct. In general,
these two Markov matrices coincide if and only if the joint distribution is
symmetric and, only in this case, both sunspot equilibria are the same.
5 Conclusion
This paper states the conditions under which sunspot equilibria of a sim-
ple overlapping generations economy and correlated equilibria of a related
economy with asymmetric information about some extrinsic uncertainty as
in Maskin-Tirole (1987) can be considered to be equivalent in the sense that
any sunspot equilibrium of the first economy can be naturally interpreted
as a correlated equilibrium of the second one and vice versa. Specifically, it
turns out that the translation of sunspot equilibria (3-SSE) to correlated
equilibria (3-CE) is not robust whenever possible (and vice versa for 2-CE
and 2-SSE) in these particular frameworks.
It is worth to be mentioned that the propositions have been stated for
the minimum integer k ~ 0 which makes each of them hold. Actually,
Proposition 1 does not hold for 2-SSE since in that case the singularity of
matrix A is guaranteed by that of matrix B, which is not the case for k = 3
nor is it very likely for any k > 3, although this remains a conjecture (an
obvious pattern for matrix B in the general case becomes easily apparent
and makes this conjecture quite sensible, since for it not to be true some
conditions on the values of the second order partial derivatives of u at the
equilibrium allocation would have to be fulfilled, which needs not be the
case). The same remark applies to the extension of Proposition 2 to any
k-CE.
Propositions 1 and 2 may help to complete the understanding that the
previous literature provided us about the links between both equilibrium
concepts.
150
Appendix
Lemma 3 Let U be any function in U such that 12
3
83 = {(x,M) E lR!+ \~3 x (0,1)61 Vi E {1,2,3}'Lmijfij(x) = O}:I 0
j=l
(12)
(hj being such that hj(x) = D 1u(Xi,Xj)+;:D2u(Xi,Xj) and ~3 being the
diagonal of R3 ) and let (a, N) E 8 3 be such that Vi E {1, 2, 3}, hi(a) :10.
Then 83 is locally a manifold of dimension 6 in a neighborhood of (a, N).
Proof. Given such u, let F : lR~+ \ ~3 X (0,1)6 --+ lR 3 be such that
L:J=l m1jf1j(X) )
F(x, M) = ( L:~=1 m2jf2j(x)
L:j=l m3jhj(x)
and let (a, N) E F- 1(0) = 8 3 . The Jacobian of F at (a, N), DF(a, N), is13
L:J=l n1jDdlj(a) n12D2!12 (a) n13 D 3f13(a)

( n21Dd21 (a) L:J=l n2jD2f2j(a) n23D 3f23(a)
n31Dd31(a) n32D2!32 (a) L:J=l n3j D 3hj(a)
!I2(a) - fn(a) !I3(a) - fn(a)

0 0 (13)
0 0
0 0
f21(a) - f22(a) f23(a) - f22(a)
0 0
0 0
0
hl(a) - h3(a)
0
h2(a) - h3(a)
)
12To prove the existence of such functions and their characterization is actually
the mathematical toil of the literature on k-SSE in OG economies (see Chiappori-
Guesnerie(1989) and Grandmont(1986) among others).
13 A 3 x 3 Markov matrix M is thought of as (ml2. ml3. m2l. m23. m3l. m32) E (0.1)6.
151
Should DF(a,N) not be full rank, then necessarily fi1(a) = fi2(a) =

fi3(a) for some i E {I, 2, 3}, but then (a, N) E F- 1(O) would imply fii(a) =
0. Thus, if for all i = 1,2,3, fii(a) =1= 0, then DF(a,N) is full rank and
therefore 8 3 constitutes a manifold of dimension 6 in a neighborhood of
(a,N) . •
Lemma 4 Let U ' be the subset of those junctions in U such that
8 3= {(x, M) ER~+ \ ~3 X(0,1)6 I Vi E{I, 2, 3}, 2:;=1 mijfij(x) = 0,

ft2(X)!23(X)!31(X) - ft3(X)!21(X)!32(X) = O} =1= 0
(14)
(Jij being as in Lemma 3). The set of junctions u E U' such that 8 3 is
locally a manifold of dimension 5 in a neighborhood of any (a, N) E 8 3
satisfying fii(a) =1= 0, for all i E {I,2,3}, is dense in U' .
Proof. Let u E U ' , let G : R~+ \ ~3 X (0,1)6 --t R4 be such that
2:;=1 m1jf1j(X)
2:;=1 m2j!2j(X)
G(X, M) = (15)
2:;=1 m3j!3j(X)
f12(X)!23(X)!31(X) - ft3(X)!21(X)!32(X)
and let (a, N) E G- 1 (O) = 8 3 be such that fii(X) =1= 0, for all i E {I, 2, 3}.
Then DG(a, N) has as its three first rows DF(a, N) computed in the proof
of Lemma 3 and as fourth row a 9-tuple whose three first entries are ex-
pressions in terms of the second order partial derivatives of u at the points
(ai,aj), i,j = 1,2,3, and the six last entries are zero. Now fii(a) =1= 0, for
all i E {I, 2, 3}, guarantees that the six last entries of the three first rows
of DG(a, N) are full rank and therefore should the forth row be a linear
combination of the others, it would necessarily be the trivial one with zero
scalars. Thus the three first entries of the forth row would have to be zero
too, which requires the second order partial derivatives of u at the points
(ai, aj) to satisfy three equations. But arbitrarily close to u there is another
C 2 real-valued function u in U' with the same gradients as u at the points
(ai,aj) (guaranteeing thus that (a,N) E a-1(O) for the corresponding j)
but not satisfying the conditions on second order partial derivatives neces-
sary to prevent full rank of Da(a, N) and, thus, such that a- 1 (O) = 83 is
locally a manifold of dimension 5 in a neighborhood of (a, N) .•
Lemma 5 Let U" be the subset of those functions in U such that
8g = {(x, M) E lR~+ \ ~3 X(0,1)6 I Vi E {I, 2, 3}, 2:;=1 mijlij(x) = 0,

m13m 21 m32 - m12m23m31 = O} =1= 0
(16)
152
(Jij being as in Lemma 3). The set of functions u E U" such that sg is
locally a manifold of dimension 5 in a neighborhood of any (a, N) E sg
satisfying fii(x) =1= 0, for all i E {1,2,3}, is dense in U".
Proof. Let u E U", let H: lRt+ \ ~3 x (0,1)6 ---t lR4 be such that
2:;=1 mljiIj(x)
H(x,M) =
2:;=1 m2jhj(x) (17)
2:;=1 m3jhj(x)
and let (a, N) E H-l(O) = sg be such that fii(x) =1= 0, for all i E {I, 2, 3}.
Then DH(a, N) has as its three first rows DF(a, N) computed in the proof
of Lemma 3 and as fourth row a 9-tuple whose three first entries are zeros
and the six last entries are
Should the three first columns of DF(a, N) be linearly dependent, then

arbitrarily close to u there is a function u in U" with the same gradients at
the points (ai,aj) (guaranteeing that (a,N) E iI- 1 (0) for the correspond-
ing j) making them linearly independent. Thus the forth row of DiI(a, N)
is necessarily the trivial linear combination of the three first rows with
scalars 0, which cannot be since N E (0,1)6. Therefore, DH(a, N) is full
rank and iI- 1 (0) = sg is a manifold of dimension 5 in a neighborhood of
(a,N) . •
Lemma 6 Let W be the subset of those ordered pairs of functions in U
such that 14
C2 = {(x 1 ,x2,p) E lR!+ \ ~2 x lR!+ \ ~2 x (0,1)41 L.i,jPij = 1,

Vi E {I, 2}, 2:~=1 PijfMx) = 0, 2:~=1 pjd{j(x) = O} =1= 0
(19)
h'
(Jijh bemgsuchthatfij(x
. h)
=D1u h (h
Xi,Xjh' ) +?;:"D2U
x. h (h hi
Xi,Xj), whereh I =
E
2 if h = 1 and vice versa). The set of (ut, u 2) W such that
(1) C2 is locally a manifold of dimension 3 in a neighborhood of any

(at, a 2, II) E C2,
(2) the linearly independent equations x} - x; = 0, for all i = 1,2, are

transversal to such manifold at any intersection point,
14Maskin-Tirole(1987) provides a necessary characterization for such functions.

153
is dense in W.
Proof. Let (u l , U 2 ) E W, let <p : lR~+ \ /).2 X lR~+ \ /).2 X (0,1)4 --t lR5 be
such that
2:i,j Pij - 1
2:~=l Plj!1j(X)
<p(xl,x 2,P) = 2:~=lP2jfij(x) (20)
2:~=l pjdlj(x)
2:~=l Pj2!?j(X)
and let (a l ,a2,IJ) E <p-l(O) = C2. Then D<p(a l ,a2,IJ) is
0 0
7f'llDdll (a) + 7f'l2 D l !12 (a) 0
0 7f'2lDdil (a) + 7f'22Ddi2(a)
7f'l1 D2!tl(a) 7f'2lD2!f2(a)
7f'l2 D2!?l(a) 7f' 22 D2!i2 (a )
0 0
7f'l1D2!ll (a) 7f'l2D2!f2(a)
7f'2lD2!il(a) 7f'22D2!i2(a) ... (21)
7f'u D d'A(a) + 7f'21Ddf2(a) 0
0 7f'l2Ddil (a) + 7f'22Ddi2(a)
1 1 1 1
ffl(a) !l2(a) 0 0
0 0 fil (a) fi2(a)
!tl(a) 0 !t2(a) 0
0 fil (a) 0 !?2(a)
Should the lower left 4 x 4 submatrix in terms of the second order partial
derivatives of u l and u 2 not be regular, there exist (ul, u2 ) in Warbitrar-
ily close to (ul, u 2 ) in the product topology with the same gradients at
points (ai\aj), i,j,m,n = 1,2 (so that (a l ,a2,IJ) E cl>-l(O) = (;2 for the
corresponding j's) which make this submatrix regular and such that the
154
coordinates of (1,0, -1,0) and (0,1,0, -1) with respect to its rows are not
in the (non trivial 15) null space of the lower right 4 x 4 submatrix 16. This
has two consequences:
(1) First, D<1>(a 1 ,a2 ,IT) is full rank (for it not to be so, looking at the
first four columns we would conclude that the first row would have
to be the trivial linear combination with zero scalars of the others,
which clearly cannot be considering the last four columns). Hence,
<1>-1(0) = 62 is locally a manifold of dimension 3 in a neighborhood
of (a 1 ,a2 ,IT).
(2) Moreover, should (a I , a 2 , IT) satisfy the equations xi - x~ = 0 too, for

all i = 1,2, these would be transversal to <1>-1(0) = 62 at (a 1 ,a2 ,IT)
since their gradients are
-1 o 0000). (22)
o -1 0 0 0 0
Thus the subset of 62 satisfying xi - x~ = 0, for all i = 1,2, is locally a

manifold of dimension 1 in a neighborhood of (aI, a2 , IT) embedded in the
3-dimensional manifold of <1>-1(0) .•
Lemma 7 Let U* be the subset of those junctions in U such that

3
8 3 = {(x, M) E R!+ \ ~3 x (0,1)6 I Vi E {l, 2, 3}, L mijfij(x) = O} -::f- 0
j=l
(23)
(fij being as in Lemma 3) and for any u E U* and i E {1, 2, 3} let
Di = {(x, M) E R!+ \ ~3 x (0,1)6 I fii(x) = O}. (24)
Then, for any i E {1, 2, 3}, the set of functions u E U* such that 83 and
Di have a transversal intersection whenever nonempty is dense in U* .
Proof: Let u be in U* such that, for some i = 1,2,3, 83 n Di -::f- 0, and let
15The lion null vector (7rll, 7rl2, 7r2l, 7r22) is in it.

16Let A and B be respectively the lower left and right 4 x 4 submatrices of
D4>(a l , a 2 , IT), and perturb the second order partial derivatives of u l and u 2 at all
(at, aj) and (a~, a;) respectively in order to have A regular. The coordinates of
c = t1,0,-1,0) and c' = (0,1,0,-1) in the basis constituted by the rows of A are
(At)-lc and (At)-lc' respectively. The mappings associating (At)-lc and (At)-lc' to
(At)-l are snrjeet.ive and hence (At)-l can be perturbed in order to put both (At)-lc
and (At)-lc' out of the null space of B in case this was not so yet. By continuity At
will still be regular.
155
(a, N) E 8 3 n Di . Let Wi : ~t+ \ ~3 x (0,1)6 -+ ~4 be such that

2:;=1 m1j!1j(X)
2:;=1 m2j!2j(X)
(25)
2:;=1 m3j!3j(X)
fii(X)
Then (a,N) E Wi1(0) and the fourth row of DWi(a,N) is such that all its
entries other than the i-th are zero. Should the i-th one, in terms of the
second order partial derivatives of u at (ai, ai), be zero too, then arbitrarily
close to u there is uin U* with the same gradient at the points (ai, aj) (and
thus ~a, N) E ~i1(0) for the corresponding i), such that the forth row
of DWi(a, N) is non-null and hence this Jacobian is necessarily full rank
(otherwise, if the fourth row of D~i(a, N) were linearly dependent of the
others, necessarily for all h = 1,2,3, ih1 (a) = ih2(a) = ih3(a), which from
(a, N) E ~i1(0) implies ihh(a) = 0, i.e. - g~~ :::::
= 1 for all h = 1,2,3.
But the strict quasi-concavity and monotonicity of u guarantees that there
is a unique (a, a) satisfying this condition. Thus, necessarily, ah = a and
hence a would not be off-diagonal.) The conclusion follows immediately.
References
Aumann, R.J., Subjectivity and correlation in randomized strategies, Jour-
nal of Mathematical Economics, 1974, Vol. 1, 67-96.
Aumann, R.J., Correlated equilibrium as an expression of Bayesian ratio-
nality, Econometrica, 1987, Vol. 55, 1-18.
Azariadis, C. and R. Guesnerie, Sunspots and cycles, Review of Economic
Studies, 1986, Vol. 53, 725-736.
Cass, D. and K. Shell, Do sunspots matter?, Journal of Political Economy,
1983, Vol. 91, 193-227.
Chiappori, P.-A. and R. Guesnerie, On stationary sunspot equilibria of
order k, in Economic Complexity: Chaos, Sunspots, Bubbles and Nonlin-
earity, W. Barnett, J. Geweke and K. Shell (Eds.), 1989.
Forges, F., Sunspot equilibrium as a game theoretical solution concept, in
Proceedings of the Sixth International Symposium in Economic Theory
and Econometrics, Barnett, W. (Ed.), 1991.
Forges, F. and J. Peck, Correlated equilibrium and sunspot equilibrium,
Economic Theory, 1995, Vol. 5, 33-50.
Grandmont, J.-M., Stabilizing competitive business cycles, Journal of Eco-
nomic Theory, 1986, Vol. 40, 57-76.
Maskin, E. and J. Tirole, Correlated equilibria and sunspots, Journal of
Economic Theory, 1987, Vol. 43, 364-373.
156
Peck, J. and K. Shell, Market uncertainty: correlated and sunspot equilibria

in imperfect competitive economies, Review of Economic Studies, 1991,
Vol. 58, 1011-29.
Samuelson, P.A., An exact consumption-loan model of interest with or
without the social contrivance of money, Journal of Political Economy,
1958, Vol. 66,467-482
Shapley, L. and M. Shubik, Trade using one commodity as medium of
payment, Journal of Political Economy, 1977, Vol. 85,937-968.
Shell, K., Monnaie et allocation intertemporelle, mimeo, 1977.
Ex post individually rational trading
mechanisIDS*
Fran<;oise Forges
THEMA, Universite de Cergy-Pontoise, 33, Boulevard du Port, 95001

Cergy-Pontoise, France and Institut Universitaire de France.
Abstract. We model the trading possibilities of one seller and n potential

buyers as a strategic game, which extends the two-person sealed-bid double
auction. We introduce particular communication equilibria for this game,
the self-fulfilling equilibria. In the case of arbitrary informational external-
ities, we show that self-fulfilling equilibria are equivalent to veto-incentive
compatible mechanisms. As a corollary, they are ex post individually ratio-
nal. In the case of private values, we show that all communication equilibria
have these properties and can furthermore be achieved through unmediated
preplay communication.
JEL Classification Numbers: C70, C78
Keywords: Communications equilibria, trading mechanism, veto-incentive

compatible mechanism
1 Introduction
The simplest model of trade involves one seller and one buyer, who both
privately know their own value for a single object initially owned by the for-
mer agent. The sealed-bid double auction mechanism has been extensively
used to solve this simple collective choice problem (see, e.g., [2], [11], [14],
[15] and [21]). More general trading mechanisms have also been proposed
(see, e.g., [4], Section 3, [5], [8], [9], [15] and [19]). In particular, Matthews
and Postlewaite ([15]) have investigated whether the double auction bid-
ding game was rich enough to reach as many solutions as general trading
rules. They have proved that the outcomes of a large class of trading mech-
anisms could be achieved as Nash equilibria of the double auction game
preceded by (interim) preplay communication. The mechanisms that can
*1 wish t.o thank S. Matthews for helpful remarks on an earlier version of this paper
(namely, [5]). Several articles by T. Gresik and my own work with E. Minelli motivated
the present version, which benefited from the comments of R. Gary-Bobo and Y. Tau-
man. I also grat.efully acknowledge the financial support of TMR network contract ERB
FMRX eT 96 0055, which enabled me to present this paper in Alicante and Tel-Aviv.
158
be simulated as communication equilibria of the bidding game satisfy a

strong form of incentive compatibility (which will be called veto-incentive
compatibility in the sequel), which implies ex post individual rationality.
The importance of ex post individually rational trading rules has been
stressed in a series of papers by Gresik ([8], [9] and [11]). In [8], he considers
a bilateral trade model with independent, private values. He illustrates on
an example that the trading rules introduced in [19] may fail to be ex
post individually rational. First, they may use entry fees and participation
subsidies, but even more, when trade occurs, the buyer sometimes has to
pay more than his valuation and the seller sometimes receives less than
his. Gresik suggests that the limited success of general mechanism design
(typically represented by [17] and [19]) partly comes from the fact that this
method concentrates on interim individual rationality, without requiring
any form of ex post individual rationality. He notes that "in most markets,
each trader has the ability to refuse to trade when the 'best' negotiated
terms give him negative utility. That is the rules of trade used by many
institutions, including those Wilson (see [25]) mentions [simple rules such
as auctions, double auctions, bid-ask markets, etc.], are ex post individually
rational" .
In the present paper, we extend the approach proposed in [15] to the
case of several buyers and identify "simple trading rules" which achieve a
large set of outcomes and are always ex post individually rational when
values are private. However, our construction hardly survives if the latter
assumption is relaxed, since in this case, even the standard two-person
sealed-bid auction may fail to be ex post individually rational (we provide
an example below). In particular, the nice property of Wilson's "simple
trading institutions" does not necessarily hold for non-private values.
We consider one seller, with a single object for sale, and n potential
buyers; all agents have private information, and (except in Section 5) the
informational externalities are arbitrary!. The basic framework is described
in details in Section 2; it is slightly more general than in [17], where the
seller does not have any private information and the informational exter-
nalities are restricted to specific revision effects.
In Section 3, we first characterize a large set of trading rules, in which
the traders fully commit themselves at the interim stage. As usual (see
[15], [17], [19], [25], etc.), we impose two requirements on trading rules:
incentive compatibility and interim individual rationality. As in [8], [9] and
[11], we observe that the corresponding outcomes will rarely be feasible
in practice, because they are not necessarily ex post individually rational.
We thus impose further restrictions on trading rules. In order to achieve
ex post individual rationality in a way that is consistent with incentive
1 Notice that most papers concentrate on private values (with some exceptions, like
[10], and to some extent, [17]).
159
compatibility, we consider mechanisms which first select a successful trader,

a price and possible transfers as a function of the types reported by the
agents (this is standard) and then submit these transactions to the agents'
approval. Such a mechanism is veto-incentive compatible if telling the truth
and accepting the transactions is a Nash equilibrium of the induced game.
We show (Proposition 1) that veto-incentive compatible rules are incentive
compatible (in the usual sense) and ex post individually rational. Except
for the fact that we allow for random mechanisms (which turns out to be
useful for some results), our approach is an extension of the one followed in
[15J to the case of several buyers and arbitrary informational externalities2 •
The point of Section 3 is to introduce a sensible set M; of outcomes for
the basic trading problem described above; M; contains all interim payoffs
that can be achieved through a veto-incentive compatible mechanism. In
many situations, it seems reasonable to look for an (ex ante or interim) effi-
cient mechanism among the veto-incentive compatible ones. Gresik (see [8J,
[9J and [11]) has shown that, in a number of benchmark two-person trad-
ing problems with private values, imposing ex post individual rationality
generates no loss of efficiency.
In Section 4, we investigate the implementation3 of veto-incentive com-
patible mechanisms in a decentralized world. We introduce a contracting
game (denoted as Gli , where 8 is a parameter in [0,1]), which represents
the last stage of a negotiation process between the traders, and reduces
to the 8-sealed-bid auction in the two-person case. More precisely, Gli is
described as follows: all traders simultaneously make bids, while the seller
also selects the trader who gets the object (possibly himself). Sale only oc-
curs if the seller decides not to keep the object and bids a smaller amount
than the winning buyer; in this case, the price is determined as a con-
vex combination (according to 8) of the seller's and the buyer's bids. By
relying on a general revelation principle (see [3J, [4] and [18]), we obtain
a simple description of all equilibria which can be achieved by allowing
(ex ante and interim) preplay communication in Gli • We denote as C(Gli )
the corresponding set of payoffs. A fully satisfactory implementation result
would be that M; = C(Gli ). Such an equality would mean that any veto-
incentive compatible rule, in which the traders delegate the organization of
2 As far as interim properties are concerned, one can dispense with random mecha-
nisms. If one cares about the agents' well-being after the effect of the mechanism, it is
natural to consider the very final stage in which all lotteries have been performed. The
description of such a stage requires random mechanisms. With this formulation, we shall
in fact be interested in posterior individual rationality (every agent gains fwm partic-
ipating in the game given the information revealed by the mechanism) rather than ex
post individual rationality (every agent gains from participating in the game given the
full description of the state of nature). A similar distinction was introduced in [6]. In the
case of private values, the two forms of individual rationality are essentially equivalent.
3Implementation is understood here in a wide sense, and in any case refers to weak
implementation. We shall come back to this below.
160
the transactions to a planner, can be realized as a non-cooperative commu-

nication equilibrium of the contracting game, in which preplay negotiation
is not binding. Unfortunately, we show by a counter-example that the pre-
vious equality does not hold in general. In the case of non-private values,
thanks to a refinement of the communication equilibrium introduced in [7],
we establish (in Proposition 2) that M; coincides with the set SF(CIi ) of
self-fulfilling equilibrium payoffs of C li • To put the result differently, not all
communication equilibria of C Ii , but only the self-fulfilling ones, are ex post
individually rational. This seems to indicate a weakness of the 8-sealed-bid
auction if values are not private (and this even in the widely considered
two-person case).
We establish in Section 5 that the desired equality M; = C(CIi) holds
if values are private (Proposition 3); as a corollary, this set also coincides
with SF( CIi). A similar property was obtained in [15] in the case of a single
buyer. We generalize another result from [15] and strengthen the (weak)
implementation interpretation mentioned above by establishing that (an
interesting class of) solutions in M; can be achieved through unmediated
communication in C Ii • If communication is performed without the help of
a mediator, the preplay phase really represents the (unbinding) negotia-
tion between the traders. In our view, the fact that unmediated preplay
communication enables the traders to reach a large class of outcomes is a
desirable property, because it implies that the traders can dispense with a
planner. On the other hand, as emphasized in [15] and [20], preplay commu-
nication dramatically worsens the full (or unique) implementation problem.
In [20], Palfrey and Srivastava solve this issue in the case of independent
types and private values by designing for any interim efficient allocation, a
prep lay communication proof mechanism, which uniquely implements this
allocation.
2 Basic model
We shall consider trade between a seller (agent n + 1), who owns a
single indivisible object, and n potential buyers (agents 1,2, ... , n). Let
I = {I, 2, ... , n+1}. We denote as Ti the set oftypes of agent i (i E 1). Ti will
typically be a finite set or a compact real interval. Let us set T = I1iEI Ti
and let F be a joint distribution on T. If the sets T i , i E I, are finite,
we assume that F is such that every ti E Ti has positive probability. In
the compact real case, we assume that F has a positive density over the
underlying intervals.
Given a vector of types t = (ti)iEI, Vi(t) denotes the value of the object
to agent i at the state of nature t. We assume that Vi(t) E [0, m] where m
is some positive real number. Values are private if
In this case, the value of the object to a given agent only depends on his
161
own type.
In the next Sections, the above quantities will constitute the basic ingre-
dients of several Bayesian games, which will model possible trading scenar-
ios.
3 Incentive compatible mechanisms

A standard way of associating a game with our basic framework is to in-
troduce a mechanism (see, e.g., [15], [17], [19], [24] and [25]). While the
literature usually focuses on deterministic mechanisms, we allow for possi-
bly random ones. The reason for doing so will be clear later on.
A mechanism chooses the agent w (the "winner") who acquires (or pos-
7r = °
sibly keeps) the object, a sale price 7r E [0, m] (with the convention that
if w = n + 1) and a vector of transfers 4 c = (cih::;i::;n E ]Rn. The
following definition is motivated by the interpretation that a mechanism
selects (w, 7r, c) as a function of types reported by the agents.
Definition 1 A random (resp., deterministic) mechanism /1 is a transition
probability (resp., a mapping) from T to I x [0, m] x]Rn.
The basic ingredients introduced in Section 2, together with a mechanism

/1, define the following Bayesian game:
(1) Nature chooses t = (ti)iEI in T according to F; agent i is only in-

formed of ti (i E I).
(2) The agents are invited to simultaneously report their types; let Si be
agent i's report (i E I).
(3) The mechanism /1 selects (w, 7r, c) in I x [0, m] x]Rn, which determines
the payoffs as follows:
Vi(t, w, 7r, c) = [Vi(t) - 7r] I [w = i] - Ci i = 1, ... , n (1)

n
Vn+l(t, w, 7r, c) = [7r - Vn+1(t)] I [w # n + 1] + .l::>i (2)
i=l
where I[.] denotes the indicator function.
3.1 Interim individually rational incentive compatible

mechanisms
One usually imposes two requirements on mechanisms: incentive compati-
bility (LC.) and interim individual rationality (INT.LR.). Both properties
4 As usual, Ci can model a fee to be paid by buyer i to the seller, or a subsidy paid
by the seller to agent i.
162
are formulated at the interim stage of the above Bayesian game5 , i.e., at
stage 2. I.e. means that no agent can gain by lying unilaterally about his
type (i.e., that telling the truth is a Nash equilibrium of the above game).
INT.I.R. means that at stage 2, all agents expect nonnegative expected
utility from participating in the game. Since both properties relate to stage
2, they can be formulated in terms of conditional expectations (given ti for
agent i). Hence, as far as these properties are concerned, one can restrict
on particular deterministic mechanisms without loss of generality. More
precisely, let us consider the conditional distribution that is induced by F
and J.L on T-i x I x [0, mJ x ~n when agent i's type is ti and agent i's report
is Si (possibly different from ti). According to (1), the conditional expected
payoff of agent i (i = 1, ... , n), with respect to this distribution, given ti,
Si, and the other agents' types Li is6
Vi (ti' Li) EJ.L [I ['Ill = iJlsi, LiJ
-EJ.L [ir1 ['Ill = iJ + cilsi, LiJ.
where EJ.L denotes the expectation with respect to J.L. From (2), one can
proceed similarly for agent n + 1. Let us denote as Ui(J.Llti) the expected
utility of agent i (i E 1), given his type ti, at the truthful equilibrium,
namely
(3)
Definition 2 A mechanism J.L is 1. C. if, and only if
Ui(J.Llti) 2: E [Vi (ti' L i , 'Ill, ir, c) Iti' Si] Vi E I, Vti, Si E Ti.
J.L is 1NT.1.R. if, and only if,
Let us set, for every sET and i = 1, ... , n,

PieS) EJ.L [I [w = i]ls] , (4)
Xi(s) EJ.L [ir1 [w = i] + cils]. (5)
Pi (s) is the probability that agent i (i = 1, ... , n) gets the object when the
reported types are s while Xi (s) is the total expected amount that buyer
i (i = 1, ... , n) pays to the seller. Let us denote as Pn+1(s) (resp., Xn+l(S))
the probability that the seller sells the object (resp., the total expected
amount received by the seller), namely,
n
(6)
5 The interpretation is that from the interim stage on, the agents fully delegate their
decision power to a planner.
6When confusion call arise, we write random variables as w, fr, etc.
163
(7)
(p, x) = (pi, Xi)iEI corresponds to a standard description of (determin-

istic) mechanisms in the literature (see, e.g., [9], [10], [17] and [19]). The
I.e. and INT.I.R. properties of J1, can be expressed in terms of (p, x) by
observing that
for every i = 1, ... , nand ti, Si E Ti. Similarly,
E [Vn+l (tn+l' L(n+l), ill, if", c) Itn+l' Sn+l] =

E [Xn+l (Sn+b L(n+l)) - Vn+l (tn+l, L(n+1))Pn+l (Sn+b L(n+l) )ltn+1]
3.2 Veto-incentive compatible mechanisms

Obviously, I.e., INT.I.R. mechanisms need not be ex post individually ra-
tional, i.e., they may involve transactions that yield a negative payoff to
some agents at stage 3. An easy explanation is that the mechanisms consid-
ered above may use entry fees and participation subsidies. This is illustrated
by Gresik ([8], p. 177) in the case of bilateral trade with independent private
values. He provides an example of an incentive efficient mechanism that
is not ex post individually rational. Myerson ([17], Section 7) constructs
an incentive efficient auction mechanism for a specific trading problem in
which two buyers have stochastically dependent private values. This opti-
mal mechanism is not ex post I.R., but even more, as Myerson explains,
in the underlying example, there is no optimal auction in which the seller
does not sometimes pay money to the bidders. This negative result con-
trasts with the ones obtained by Gresik (see [8], [9] and [U]) under suitable
assumptions.
The next simple example illustrates that even in the absence of entry
fees and participation subsidies, and even in the case of private values,
I.e., INT.I.R. mechanisms may fail to be ex post individually rational,
because the sale price may be below the seller's reservation price or above
the successful buyer's reservation price.
Example 1 Let n = 1 (one buyer, one seller), Tl = {5, 7}, T2 = {I, 3}, F
be uniform, Vi(ti) = ti and consider the mechanism
J1,(5, 1) = J1,(7, 3) = (1,2,0), J1,(5, 3) = J1,(7, 1) = (1,6,0)
(the object is always sold, at price 2 (resp., 6) if the reported values are
(5,1) or (7,3) (resp., (5,3) or (7,1)).
164
In order to eliminate mechanisms that are not ex post individually ra-

tional, we add a fourth stage to the game introduced previously:
4. The outcome (w, 7r, c) selected by the mechanism f..L is revealed to all
agents; every agent has the option of refusing the transaction and getting
a zero payoff.
In this new game, we focus on Nash equilibria in which the agents
truthfully reveal their types (at stage 2) and accept the transaction (at
stage 4). A mechanism satisfying the associated equilibrium conditions will
be called veto-incentive compatible (VETO-I. C. ).
Definition 3 A mechanism f..L is VETO- 1. C. iff
for every i E I and every ti, Si E Ti.
The interpretation of VETO-I.C. is that the agents can refuse unprof-

itable transactions and fully anticipate this possibility when they report
their type. VETO-I.C. is very close to the I.C.* condition which was in-
troduced by Matthews and Postlewaite ([15]) in the particular case of two
traders and private values. The main difference is that LC.* is expressed
for deterministic mechanisms of the form ((4),(5)), which may still hide a
final lottery 7. Random mechanisms allow us to require that every agent
gets a nonnegative payoff at the very last stage of the game (namely, stage
4), after all lotteries have taken place.
We shall establish that VETO-LC. mechanisms satisfy a form of ex post
LR., which we call a posteriori LR.
Definition 4 A mechanism f..L is a posteriori 1.R. iff
for every i E I and every ti E Ti'
This property means that at equilibrium (i.e., if all players truthfully

reveal their types and accept the transactions), every agent gets a nonneg-
ative expected payoff at stage 4, given the information that he has at that
stage (Le., his type and the outcome (w, 7r, c) selected by the mechanism).
A posteriori I.R. may thus differ from ex post I.R. as far as the latter usu-
ally entails that the whole state of nature t is revealed to the agents (see
[6] for a similar distinction between ex post and a posteriori efficiency)8.
7More precisely, in the model considered in [15], VETO-I.C. of the original random
mechanism implies VETO-I.C. of the expected mechanism (p, x), which in turn is equiv-
alent to I.C.* combined with ex post I.R. (to be formally defined below).
x Observe that a posteriori I.R. differs from ex post .,.eg1Y~t, which appears for instance
165
In the case of private values, by recalling (1) and (2), a posteriori I.R.
takes the following form:
n
[1f-Vn+1(tn+1)]I[w#n+1]+L~ > 0 a.s.
i=l
for every ti E Ti (i E I). These inequalities imply the standard ex post I.R.
property considered in the literature (see, e.g., [8], [9], [10] and [15]), which
is expressed in terms of the mechanism (p, x) (see (4),(5».
Proposition 1 Let {£ be a VETO-I.e. mechanism. Then {£ is I.e. and a
posteriori I.R. {and thus, INT.I.R.}.
Proof: It is obvious that VETO-I.C. implies I.C. To check that VETO-
I.C. implies a posteriori I.R., recall (3), take Si = ti in (8) and use that, for
every random variable X,
E[max{O,x}] ~ E[x] => x ~ 0 a.s.
It is not difficult to construct examples which show that I.C. and a pos-
•
teriori I.R. do not imply VETO-I.C. (see, e.g., [15], example 1).
Observe that an a posteriori I.R. mechanism cannot select (positive)
entry fees to be paid by unsuccessful potential buyers, while any transfer
between the successful buyer and the seller can be included in the sale price.
Hence, the only possible transfers that can be selected by an a posteriori
I.R. mechanism must correspond to participation subsidies offered by the
seller to unsuccessful potential buyers (provided that such transfers are
profitable to the seller). In particular, if there is only one buyer and one
seller, no transfer can occur. Given this state of affairs, we shall from now
on focus on mechanisms which only select a winner and a price (i.e., such
that c = 0 with probability one).
Definition 5 M; denotes the set of all interim payofJs that can be achieved
through a VETO-I.e. mechanism without transfers.
Remark: In the light of the previous analysis, one is tempted to search for
an (ex ante or interim) efficient mechanism among the VETO-I.C. ones. In
the case of bilateral trade with private values, Gresik (see [8], [9] and [11])
identifies conditions which guarantee that the restriction to ex post I.R.
mechanisms entails no loss of efficiency.
in some interpretations of the winner's curse in common value auctions (sce [16] and
[26]). The standard approach, which entails that each bidder anticipates the information
revealed by a winning bid, is typically a posteriori I.R. (given the winner and the price),
but is not necessarily immune to ex post regret (given the realized value of the random
variable of interest).
166
4 Communication in a contracting game

4.1 A contracting game
Let us introduce another Bayesian game, denoted as GO, where 8 is a para-
meter varying in [0,1]. The players and the types are exactly as in Section
2. Every potential buyer i chooses a bid bi in [0, m] (i = 1, ... , n), while the
seller simultaneously chooses a bid bn +1 in [0, m] together with a "winner"
k in I. Let us set b = (biLEI and
The payoff functions in GO are defined as follows:
uf(t, k, b) [Vi(t)-pO(k,b)]I[i=k,bk~bn+l], i=I, ... ,n (9)

u~+1 (t, k, b) [pO (k, b) - Vn+1 (t)] I [k :/= n + 1, bk ~ bn+1]' (10)
The interpretation is that all traders make bids, while the seller also
chooses the trader k who gets the object (possibly himself). If the seller
decides not to keep the object, trade can only occur if the winning buyer
(k) makes a higher bid than the seller.
In the case of a single buyer (n = 1), the contracting game GO reduces
to the extensively studied sealed-bid double auction (see, e.g., [2], [11], [14],
[15] and [21]). Obviously, with several buyers, many other generalizations
are conceivable (see, e.g., [24] and [25]). In particular, the framework is
one in which the seller could organize a standard auction (see [26]). How-
ever, such procedures require some transmission of information between the
agents, for instance, to determine the highest bid. Here, we distinguish the
communication phase from the final contract. The main justification for
proceeding in this way is that once communication is allowed, it can take
a variety of forms, which are difficult to control. We thus want to analyze
the effects of all conceivable scenarios of communication.
As suggested in the previous comments, the contracting game must be
interpreted as the "reduced form" of the last step of a negotiation process,
so that the simultaneous proposals of the seller and the buyers are only ap-
parent. For instance, if 8 = 0, the game is essentially equivalent (in strategic
form) to the following one: the seller selects a buyer (k) and proposes him
a price (b n +1 ); buyer k accepts or refuses the proposal. In the first case,
he pays bn +1 and gets the object. Otherwise, the seller keeps the object 9 •
This new game exactly models the final stage of a negotiation process. It
is sensible if it is preceded by a phase of communication between the seller
and the buyers. This preliminary phase enables the seller to fix the price
bn +1 as a function of messages from the buyers. For instance, as noted
9The strategic form obtained by restricting buyers to monotonic strategies in the

second scenario is the same as the original strategic form.
167
above, the preplay communication phase can consist in a standard (first

price, second price, etc.) auction.
To sum up, we view G 6 as a simple representation of the final stage
in which a transaction possibly takes place. The game makes sense if the
players are allowed to communicate before this final stage.
4.2 Communication equilibria in the contracting game

In order to model the traders' exchange of information, we shall use com-
munication devices, which select signals for the players as a function of
messages transmitted by the players.
Definition 6 A communication device for G 6 consists of a space of mes-

sages Mi and a space of signals Si for every player i E I, together with a
transition probability v from M = TIiEI Mi to S = TIiEI Si.
Given a communication device d = ((Mi' Si)iE[, v), G6 can be extended

into a new Bayesian game G6 (d) which is played as follows:
(1) Nature chooses (ti)iEI in T according to Pj agent i is only informed

of ti (i El).
(2) Every agent i sends a message mi E Mi to the communication device.

These choices are made simultaneously.
(3) The communication device selects (Si)iEI E S according to

v(.I(mi)iEI) and transmits Si to agent i.
(4) The agents make simultaneous bids and the seller also chooses a win-
ner, as in G 6 • The payoffs are as in G 6 •
Definition 7 A communication equilibrium of G 6 is a Nash equilibrium

of G 6 (d), for some communication device d. C(G 6 ) denotes the set of all
interim payoffs from communication equilibria in G 6 .
As shown in [3], [4] and [18] (chapter 6), by a general revelation principle,
C( G 6 ) contains all equilibrium payoffs that can be achieved through any
conceivable scheme of (preplay and interim) communication in G 6 ; this
result still holds if one restricts to "truthful and obedient" communication
equilibria, which are associated to communication devices such that Mi =
T i , i E I, and Si = [0, m], i = 1, ... , n, Sn+l = [0, m] xl (namely, the agents
are asked to report their types and get a recommendation on how to bid
in G 6 ).
Let us illustrate that communication equilibria of G 6 may fail to be ex
post individually rational.
168
Example 2 Let n = 1 (one buyer, one seller), TI

{t~, tn, F be uniform and
VI (tL t~) = 4, v2(tL t~) =1

vI(tLt~) = 5, v2(tL t~) = 2
VI(t~, t~) = 5, V2(t~, t~) = 2
VI(t~, t~) = 10, V2(t~, t~) = 3.
Let 8 = 0.5; the following bids
bl(ti) = ~(t~) = 3.5

bl(t~) = ~(t~) = 7.5
form a Nash equilibrium (and thus a communication equilibrium) of G ti • In

state (t~, t~), sale occurs at price 5.5, which is above the buyer's valuation
of the object (VI(t~,t~) =5).
This example shows that in general, C(Gti ) will be a strict subset of M;
(see definition 5). In Section 5, we shall prove that if values are private,
C( G ti ) = M;. Before that, we shall identify the subset of C( G ti ) which
coincides with M;.
4.3 Self-fulfilling equilibria in the contracting game

We shall now focus on particular communication equilibria, which were
introduced in [7].
Definition 8 A self-fulfilling equilibrium of G ti is a communication equi-

librium of Gti satisfying the following requirements:
(1) The underlying communication device d = ((Mi, Si)iE!, v) is such

that Mi = T i , Si = I x [0, m] for every i E I and v is a mechanism
without transfers (namely, v selects the same signal in I x [0, m] for
all agentslO ).
(2) Every player truthfully reports his type at stage 2 of Gti(d).
(3) For every signal (w, 71'") E I x [0, m] selected by v, the strategic choices
of the players at stage 4 of Gti(d) induce w as winner and 71'" as price
(given (w,71'"), the seller chooses k = k(w,71'") = w, and ifw i= n + 1,
the bids bw = bw (w,71'") and bn +1 = bn +1(w,71'") satisfy bw (w,71'") 2:
bn +1 (w, 71'") and pti (k( w, 71'"), b( w, 71'")) = 71'").
SF( Gti ) denotes the set of all interim payofJs from self-fulfilling equilibria
in G ti .
10 As in Section 2, we adopt the convention that 7r = 0 if k = n + 1.

169
The next proposition states that the self-fulfilling equilibria of GO are

ex post LR. in the strong sense introduced in Section 3 (namely, they can
be achieved through a VETO-LC. mechanism) and that, conversely, every
outcome of a VETO-LC. mechanism can be implemented as a self-fulfilling
equilibrium of GO, i.e., as a particular kind of communication equilibrium
of GO.
Proposition 2 SF(GO) = M;.
Proof. Consider a self-fulfilling equilibrium of GO and let v be the un-
derlying communication device. By definition, v is a mechanism without
transfers (see definition 1 and the end of Section 2). We have to show that
v is VETO-LC. Observe that the revelation of types is similar in GO(v)
and in the scenario associated with a VETO-LC. mechanism. Let us as-
sume that v selects the public signal (w,7I"). In GO(v), given (w,7I"), every
buyer i can guarantee a nonnegative payoff by bidding bi = 0 (i = 1, ... , n),
and similarly for the seller, who can bid bn +1 = m. By the self-fulfilling
equilibrium conditions, if (w, 71") induces a sale (i.e., w -I n + 1), the traders
make bids in order to accept the transaction. They thus also accept it in
the mechanism scenario, in which they have less strategic possibilities.
Conversely, let v be a VETO-LC. mechanism. v can be viewed as a com-
munication device for GO. Let us consider particular strategies in GO (v).
At stage 2, every agent tells the truth. At stage 4, given the signal (w, 71")
selected by v, the seller chooses k( w, 71") = wand bn +1 = 71"; if w -I n + 1,
buyer w chooses bw(w, 71") = 71"; every buyer i -I w chooses bi = O. It is easily
checked that the VETO-LC. property of v implies that the previous strate-
gies form an equilibrium of GO(v) and satisfy the self-fulfilling property. •
ReIllarks:
(1) The previous proposition holds for every 8 E [0,1] and shows in par-
ticular that the set SF(GO) is independent of 8. The same will be
true for the results in Section 5.
(2) We have considered public mechanisms, which send the signal (w, 71")
to all players. It would seem more natural that only the seller and
the winning buyer (w) get this information. Proposition 2 holds in
this case, provided that the same definition of mechanism is used
for VETO-LC. and self-fulfilling equilibria. However, if values are
not private, the mechanism conveys information on the value of the
object. Hence, any change in the signals will modify the VETO-LC.
condition (namely, (8)). On the other hand, as argued in [12] and [13],
public mechanisms may be preferred to private ones, for instance,
because their messages are verifiable.
(3) The proposition and the example above show that, in general, solu-
tions in M; correspond to a strict subset of C(GO). From this, one
170
might be tempted to conclude that even if n = 1, the sealed-bid dou-

ble auction is not appropriate if values are not private. In the next
Section, we shall see that the communication equilibria of G{j behave
nicely when values are private.
5 Private values
From now on, we assume that every agent knows the value of the object to
himself, namely that the value functions satisfy Vi(t) = Vi(t i ) for every i E I
and ti E Ti. [15] contains a result which is similar to the next proposition
in the case n = 1.
Proposition 3 If values are private, C(G{j) ~ M;.

Proof. Let us consider a communication equilibrium. By a general rev-
elation principle (see [3] and [18], chapter 6), we can assume without loss
of generality that the communication equilibrium is canonical. This means
that the set of messages of every agent is a copy of his set of types, the set of
signals to every buyer (resp., seller) is a copy of the set of bids [0, m] (resp.,
I x [0, m]) and the communication equilibrium strategies are truthful and
obedient.
The basic idea of the proof is the following: a canonical communication
equilibrium generates a mechanism which selects a winner and a price.
Since this mechanism simulates the communication equilibrium, it should
be VETO-I.e. The only problem with this reasoning (which makes that it
is not valid if values are not private) is that the mechanism just constructed
may not convey the same information as the original communication device.
In the case of private values, the argument works because, once the winner
and the price are known, further information on the traders' types does not
matter.
In order to formalize the previous idea, let us construct a mechanism from
the given canonical communication device. Let (k, b) = (k, b1 , •.• , bn+ 1 ) be a
recommendation selected by the communication device. We construct the
mapping (w, 7i") as follows:
w(k, b) = k and 7i"(k, b) = p{j(k, b) if k E {I, ... , n} and bk ~ bn+1 ,

w(k, b) = n + 1 and 7i"(k, b) = 0 otherwise.
Let IL be the mechanism induced by the initial communication device and

the previous mapping. By using (9), (10) and the private values assumption;
one can show that
Uf(ti,k,b) [Vi(ti) - 7i"(k, b)] I [w(k, b) = i] , i = 1, ... , n (11)

u~+ 1 ( tn+! , k , b) [7i"(k, b) - Vn+l (tn+l)] I [w(k, b) -# n + 1].
171
In order to show that f..L is VETO-I.C., consider buyer i and assume his
type is ti. If he reports Si, possibly different from ti, the communication
device makes his recommendations given (Si, Ld. Let bi be the bid which
is recommended to agent i. The communication equilibrium conditions for
this agent can be written as
for every ti and Si in T i , where f3i ranges over all possible bidding strate-
gies of buyer i (in the game with communication). By choosing f3 i =
min {bi , Vi(ti)}, player i guarantees
Uf(ti, k, f3i' bn +1) ~ max { 0, uf (ti' k, bi , bn +1 ) } . (13)
It is easily checked that (11), (12) and (13) imply VETO-I.C. One can
obviously proceed in a similar way for the seller . •
Propositions 2 and 3, together with the fact that SF(G Ii ) ~ C(G Ii ), yield
the following corollary:
Corollary 4 If values are private, C( G Ii ) = SF( G Ii ) = M;.
From Definition 7, communication equilibria in GIi are achieved with the
help of a communication device, which may select signals in a complex
way. In practice, one expects that communication will have the form of a
simple conversation between the traders, without the help of any media-
tor. The next proposition states that if values are private, a large class of
communication equilibria of G6 can be achieved through a simple scheme
of unmediated communication. A siInilar result was obtained in [15] in the
case n = 1 (see the remark below).
Proposition 5 If values are private, every payoff in M; which can be
achieved by means of a mechanism with finite support is a fully revealing
Nash equilibrium payoff of the following unmediated communication bidding
game G~:
(1) Nature chooses (ti)iEI in T according to F; agent i is only informed

ofti (i E I).
(2) Every agent i chooses a message Si E 7';,. Agent 1 (resp. n+1) also se-
lects Zl (resp. Zn+l) in [0,1]. These choices are made simultaneously,
and then revealed to all players.
(3) The agents make simultaneous bids and the seller also chooses a win-
ner, as in G 6 . The payoffs are as in G Ii •
172
Proof. Let us fix a payoff in M;, which can be achieved by means of a

mechanism v with finite support. We shall construct an equilibrium of G~,
which yields the same payoff.
First, let us specify the strategies of the agents at stage 2; every agent
i reveals his true type, i.e., Si = ti, while agent 1 and agent n + 1 further
choose Zl and Zn+l respectively, independently of each other, uniformly in
[0,1].
Zl and Zn+1 generate a new variable Z E [0, 1] in the following way:
the rth term of the binary expansion of Z is 1 (resp., 0) whenever the rth
terms of the binary expansions of Zl and zn+1 coincide (resp., differ). This
procedure is known as a jointly controlled lottery, because Z is uniformly
distributed as soon as (at least) one of the agents, 1 or n + 1, selects his
own variable uniformly.
In order to define the strategies at stage 4, we associate a (finite) partition
Pt of [0,1] with every t E T:
Pt = {Jt(w, 7I")I(w, 71") E S(v)}
where S(v) is the support of v and the subintervals Jt (w,7I") of [0,1] are
constructed in such a way that their respective lengths are v(w,7I"It).
All agents can evaluate Z (from Zl and Zn+1) and given S (the types
announced at stage 2), they can identify the element J s(w, 7I") of Ps which
contains z. The idea is that if Js(w, 71") 3 z, they should mimic the scenario
of M; when v selects (w, 71"). More precisely, the bids at stage 4 are defined
as follows:
if 71" ::; Vi(ti) and w = i

otherwise
for every i = 1, ... , n, and
k(t n +1, w, 71")

{:
W.
if 71" ~ Vn +l(tn +1) and w -=I n
otherwise
+1
It is straightforward to check that the strategies defined above form an

equilibrium of G~, and yield the original payoff in M; .•
Remarks:
(1) Private values are crucial in the previous proof; this assumption guar-
antees that once the winner and the price are settled, knowledge of
the types is not useful. Similar results are available in the general
case, but they require a much more intricate proof (see, e.g., [1], [4],
[22] and [23]).
173
(2) The use of random mechanisms in the definition of M; simplifies

the previous proof. As noted in Section 3, Matthews and Postlewaite
( [15]) use deterministic mechanisms (p, x) of the form (4), (5) , (6)
and (7) in the case n = 1 (single buyer). They establish a result
which is similar to Proposition 5, namely that every payoff from an
ex post 1.R., 1.C.* mechanism (p, x) can be realized as a fully revealing
equilibrium of G~. They construct an equilibrium of G~ which yields
the desired payoff in M;, but does not exactly mimic the scenario of
the original mechanism. Let us show that this proof does not extend
in the case of several buyers. Matthews and Postlewaite define z as
above and identify bidding strategies which guarantee that the object
is sold to buyer i (i = 1, ... , n) with probability Pi(t), at price p,-
xi((~1
if Pi(t) # O. A necessary equilibrium condition is that for every i =
1, ... , n and every t E T such that Pi(t) # 0,
(14)
However, the ex post individual rationality of (p, x) only implies that
Xn+l(t) - [1- Pn+1(t)]Vn+l(tn+l) ~ 0,

which is weaker than (14) as soon as n ~ 2. Starting with ran-
dom mechanisms, we require that the original mechanism satisfies
a stronger form of ex post individual rationality, which allows us to
construct a payoff equivalent Nash equilibrium in the unmediated
bidding game. Nevertheless, our construction only applies to mecha-
nisms with finite support.
(3) As observed by many authors (e.g., [15] and [20]) it is usually im-
possible to prevent traders from communicating in a non-binding,
unmediated way before making decisions. Proposition 5 shows that
under private values, elementary preplay communication considerably
enlarges the set of feasible outcomes and thus makes the full imple-
mentation problem harder. In order to achieve full implementation
in environments where preplay communication cannot be precluded,
Palfrey and Srivastava [20J construct mechanisms that are immune to
preplay communication. These mechanisms are designed to reach a
specific, interim efficient allocation; unlike the ones considered above,
they are not universal (see [4]).
References
[1] Baniny, 1., 1992, Fair distribution protocols or how the players replace
fortune, Mathematics of Operations Research 17, 327-340.
[2J Chatterjee, K. and W. Samuelson, 1983, Bargaining under incomplete
information, Operations Research 31, 835-851.
174
[3] Forges, F., 1986, An approach to communication equilibria, Econo-

metrica 54, 1375-1385.
[4] Forges, F., 1990, Universal mechanisms, Econometrica 58, 1341-1364.
[5] Forges, F., 1990, Trading games with asymmetric information, Discus-
sion Paper 880, The Center for Mathematical Studies in Economics
and Management Science, Northwestern University.
[6] Forges, F., 1994, Posterior efficiency, Games and Economic Behavior
6,238-261.
[7] Forges, F. and E. Minelli, 1997, Self-fulfilling mechanisms and rational
expectations, Journal of Economic Theory 75, 388-406.
[8] Gresik, T., 1991, The efficiency of linear equilibria of sealed-bid double
auctions, Journal of Economic Theory 53, 173-184.
[9] Gresik, T., 1991, Ex ante efficient, ex post individually rational trade,
Journal of Economic Theory 53, 131-145.
[10] Gresik, T., 1991, Ex ante incentive efficient trading mechanisms with-
out the private value restriction, Journal of Economic Theory 55, 41-
63.
[11] Gresik, T., 1996, Incentive-efficient equilibria of tw~party sealed-bid
bargaining games, Journal of Economic Theory 68, 26-48.
[12] Lehrer, E., 1996, Mediated talk, International Journal of Game Theory
25, 177-188.
[13] Lehrer, E. and S. Sorin, 1997, One-shot public mediated talk, Games
and Economic Behavior 20, 131-148.
[14] Leininger, W., P. Linhart and R. Radner, 1989, Equilibria of the
sealed-bid mechanism for bargaining with incomplete information,
Journal of Economic Theory 48, 63-106.
[15] Matthews, S. and A. Postlewaite, 1989, Preplay communication in
tw~person sealed-bid double auctions, Journal of Economic Theory
48, 238-263.
[16] Mougeot, M. and F. Naegelen, 1991, Malediction du vainqueur et ra-
tionalite economique, Revue Economique 42, 29-50.
[17] Myerson, R., 1981, Optimal auction design, Mathematics of Operations
Research 6, 58-73.
[18] Myerson, R., 1991, Game theory: analysis of conflict, Harvard Univer-
sity Press, Cambridge.
175
[19] Myerson, R. and M. Satterthwaite, 1983, Efficient mechanisms for bi-

lateral trading, Journal of Economic Theory 29, 265-28l.
[20] Palfrey, T. and S. Srivastava, 1991, Efficient trading mechanisms with
pre-play communication, Journal of Economic Theory 55, 17-40.
[21] Satterthwaite, M. and S. Williams, 1989, Bilateral trade with the
sealed-bid k-double auction: existence and efficiency, Journal of Eco-
nomic Theory 48, 107-133.
[22] Urbano, A. and J. E. Vila, 1997, Pre-play communication and coordi-

nation in two-player games, mimeo, Valencian Institute of Economic
Research.
[23] Urbano, A. and J. E. Vila, 1997, Unmediated talk under incomplete
information, mimeo, Valencian Institute of Economic Research.
[24] Wilson, R., 1985, Efficient trading, in: G. Feiwel (ed.), Issues in Con-
temporary Economics and Welfare, Macmillan, New York, 169-208.
[25] Wilson, R., 1987, Game-theoretic analyses of trading processes, in: T.

Bewley (ed.), Advances in Economic Theory, Fifth World Congress,
Cambridge University Press, Cambridge, 33-70.
[26] Wilson, R., 1992, Strategic analysis of auctions, in: R. Aumann and S.
Hart (eds), Handbook of Game Theory with Economic Applications,
North Holland, Amsterdam, 227-279.
Allocation among multi-member households:
issues, cores and equilibria
Hans Gersbach 1 and Hans Haller 2
1 Alfred-Weber-Institut, University of Heidelberg, Grabengasse 14, D-69117
Heidelberg, Germany.
2 Department of Economics, Virginia Polytechnic Institute and State
University, Blacksburg, VA 24061-0316, U.S.A.
Abstract. In a general equilibrium model, we allow for households with

several, typically heterogeneous, members; households that make (efficient)
collective consumption decisions where different households may use differ-
ent collective decision mechanisms; yet households that operate within a
competitive market environment. While raising other main issues like the
optimality and decentralization properties of the corresponding equilib-
ria, this paper deals primarily with the existence of competitive equilibria
among multi-member households and with core allocations for economies
with multi-member households.
JEL Classification Numbers: D51
Keywords: Multi-member households, competitive equilibria, core
1 Issues
The concept of general equilibrium among multi-member households in-
corporates the fact that the allocation of resources among consumers and
the ensuing welfare properties are affected by the specifics of a pre-existing
partition of the population into households. Conversely, the formation of
households can - partly or fully - be driven by economic considerations,
by the anticipated effects of the emerging household structure on the alloca-
tion of economic resources. See Becker (1978, 1993) for the most prominent
voice on endogenous household formation. When we consider households in
the sequel, we depart from traditional economic theory which has, with few
exceptions, treated households as if they were single consumers. We allow
for households with several, typically heterogeneous, members; households
that make (efficient) collective consumption decisions where different house-
holds may use different collective decision mechanisms; yet households that
operate within a competitive market environment. This departure from the
traditional market model permits us to investigate the interplay of dual
178
roles of households, households as collective decision making units on the

one hand and households as competitive market participants on the other
hand. While we maintain the term "household" throughout, the broader
interpretation as socio-economic group or simply group would be appropri-
ate in many instances, in particular since we do not impose restrictions on
household or group size, respectively. In subsections 1.1 and 1.2, we moti-
vate and report on previous work on the optimality and decentralization
properties of equilibria. In subsections 1.3 and 1.4, we outline current and
future research, including the innovations of this paper.
1.1 Fixed household structure

A particular line of research originates with Chiappori (1988, 1992) who,
in contrast to traditional economic theory, suggests a model of collective
rationality of households as an alternative to the neoclassical model where
households are treated like single consumers. Inspired by Chiappori's work,
Haller (1995) analyzes the implications of collective household decisions in
a general equilibrium model with an exogenously given, fixed household
structure. HaIler assumes that household members have individual pref-
erences. Taking a market price system as given, the household makes an
efficient consumption choice (in terms of the preferences of its members)
subject to its budget constraint. This behavioral assumption leads to the
notion of a competitive equilibrium among households. Alternatively,
the household might assign endowment or income shares to its members
and let them operate individually in the market. That behavioral assump-
tion corresponds to the concept of a competitive equilibrium among
individuals.
The general question is whether it makes any difference who partici-
pate in the market, households or individuals. Specifically, Haller (1995)
addresses the normative issue of optimality (efficiency), i.e., the ques-
tion whether competitive exchange among households as entities leads to
a Pareto-optimal allocation. He finds that the answer is in the affirmative
as long as each household makes an optimal (efficient) choice subject to its
budget constraint and, by doing so, exhausts its budget. This conclusion
can be generalized to a core inclusion statement. To this end, Haller intro-
duces the notion of the H-core (household core) which reflects the fact that
only households are market participants.
Of course, non-optimal equilibrium allocations can occur even in
economies consisting exclusively of one-person households, provided that
some consumers possess satiation points in the interior of their budget
sets whereas other consumers have non-satiated preferences and exhaust
their budgets. With multi-person households rather than individuals par-
ticipating in the market, this phenomenon is more likely, however. Namely,
a household with negative intra-household externalities may have a bliss
point despite the fact that each household member has monotonic prefer-
179
ences with respect to her individual consumption. Just imagine a household

composed of two smokers. Each household member may individually pre-
fer to always smoke more, since the additional nicotine intake more than
compensates for the deterioration of air quality it causes. Nevertheless, the
negative externalities due to air pollution can be such that the two smokers
agree on an unconstrained optimum consumption for the household. There-
fore, it is not too surprising that certain externalities lead to sub-optimal
equilibrium allocations. Then the major contribution of HaIler (1995) con-
sists in identifying externalities that do not hinder Pareto-optimality of
equilibrium outcomes: Each household, by internalizing its intra-household
externalities, furthers global efficiency. Equilibrium efficiency is obtained,
if each household makes an efficient choice under its budget constraint and
the nature of consumption externalities among household members is such
that an efficient household choice implies budget exhaustion.
Regarding the general question whether it makes any difference who par-
ticipate in the market, households or individuals, HaIler (1995) addresses
further the positive issue of individual decentralization: Does competi-
tive exchange among households lead to outcomes that can also be attained
via competitive exchange among individuals? In other words: Given a com-
petitive equilibrium allocation with only households participating in the
market, can this allocation also be attained as a competitive equilibrium al-
location where the individual household members participate in the market
- after being allotted suitable income or endowment shares? The answer
is in the affirmative in the absence of any externalities and with standard
monotonicity and smoothness conditions. When intra-household externali-
ties are present, individual decentralization of equilibrium outcomes among
households is still possible in exceptional cases. But as a rule, individual
market participants do not fully internalize intra-household externalities
whereas a household does it by assumption.
1.2 Variable household structure

The present paper is part of a broader research project aimed at studying
the allocation of commodities and consumers through the interaction of
two allocation mechanisms: collective decisions and markets. In Gersbach
and HaIler (1997), we are investigating further the efficiency properties and
the decentralization possibilities of the dual allocation mechanism. Tran-
scending the limitations of HaIler (1995), the most important innovation
in Gersbach and HaIler (1997) is a model with a variable household
structure. One of the central issues ought to be how well the market is
performing under such conditions. The performance standard, that is which
allocations qualify as "optimal" or "efficient", should depend on how much
freedom a social planner is granted to allocate resources and people. Ac-
cordingly, we consider two notions of Pareto optimality: constrained Pareto
optimality and full Pareto optimality. The attribute "constrained" refers
180
to a fixed household structure whereas "full" refers to a variable household

structure.
The chosen framework makes it possible to address, in principle, socio-
economic questions such as the positive question which households will form
and the normative question which households should form. The framework
also allows to address related welfare policy issues like who should be the
designated recipients of transfers, households or individuals.
Gersbach and HaIler (1997) find that in the absence of externalities or
the presence of special externalities, details of household formation and
(efficient) collective decisions within households dp not affect Pareto ef-
ficiency, but do influence the allocation of commodities. Reliance on the
market and efficient choices within households need not result in fully op-
timal allocations. When consumption externalities within households are
not too positive, every fully Pareto optimal allocation of commodities and
consumers can be potentially decentralized by a competitive equilibrium
among households after rearranging households and suitably redistributing
endowments via lump-sum transfers. The market may fail, if the household
as entity or a distinguished household member are designated as recipient
of the aggregate transfer to the household.
Like Pareto optimal allocations, core allocations can be defined without
recourse to market prices. Which allocations qualify as core allocations
depends on which coalitions can form and what these coalitions can achieve
on their own. Here we suggest two core concepts: a constrained core concept
(the H-core or household core of HaIler (1995)) and a full core concept.
Again, the attribute "constrained" refers to a fixed household structure
whereas "full" refers to a variable household structure. The former requires
that a coalition is the union of existing households. The latter requires that
a coalition can only form a sub-economy of its own, if it comes up with a
household structure of its own.
The basic analysis of the constrained core is straightforward. The core
inclusion result of HaIler (1995) combined with the equilibrium existence
results derived in the present paper yields sufficient conditions for the non-
emptiness of the constrained core. The study of the full core proves more
intriguing. We present an example where the full core is empty under stan-
dard assumptions. Under very restrictive conditions, non-emptiness of the
full core can be shown. It is also interesting to note that the distinction
between clubs and households becomes blurred when we apply the full
core concept that allows any coalition to form and exercise blocking power.
There are, however, questions of individual power that suggest themselves
more convincingly with respect to households than with respect to clubs.
Take for example the effect of exogenously given intra-household bargain-
ing power on resource allocation. Or the allocative impact of the threat
to leave (individual rationality). In preliminary work, Gersbach and HaIler
(1998) pursue this sort of questions.
181
1.3 This paper

In the following sections, we present novel results. We first introduce general
notation and definitions. We then state and demonstrate existence results
for general equilibria among households, with a fixed household structure.
Finally, we turn to core allocations for economies with multi-member house-
holds.
1.4 Future research

The basic framework developed here and elsewhere can be extended in
several meaningful ways. To begin with, households in our model could
be reinterpreted as firms if we assume that individuals supply inelastically
labor and the household endowments are the result of the production of
firms. Moving beyond this reinterpretation by imposing further structure
on the technology of potential firms could allow to examine the boundaries
of firms in a general equilibrium setting. Secondly, as already suggested in
1.2, one can study the household formation process in more detail. Thirdly,
the approach might be useful to study changes in the socio-economic struc-
ture. Fourthly, the optimal design of social transfer systems depends on the
nature of collective decisions within households and on incentives to form
households. The design of transfer systems should take into account not
only the impact on the well-being of individuals under the given household
structure, but also the impact on the household structure itself.
To conclude, the formal integration of collective decision and competi-
tive market models and the investigation of related normative and organi-
zational issues should be viewed as a major research task. The work has
just begun.
2 Allocation of consumers and commodities

In this section, we describe the primitive data of the model: commodities,
consumers, and household structures, and furthermore our main innova-
tion, the concept of a feasible allocation of commodities and consumers.
We then define the concept of competitive equilibrium among households.
Finally, we show for a fixed household structure the existence of competitive
equilibria among multi-member households.
2.0 General notation and terminology

In order to simplify the formal presentation and to facilitate the exposition
of several examples, we first introduce a general and convenient notation
for the comparison of vectors.
Consider any L E N, a = (at. ... ,ad, b = (b 1 , ... , bd E ]RL, and f :
]RL ~ R. By a 2 b, we mean at 2 bt for all i = 1, ... , L. By a » b we
mean al > bt for alIi. Finally, a > b stands for a 2 b, ai-b. The function f
182
is called non-decreasing, if for any a, b E ~L, a ~ b implies f(a) ~ f(b). It

is increasing, if for any a,b E ~L, a ~ b implies f(a) > f(b). It is strictly
increasing, if for any a,b E ~L, a> b implies f(a) > f(b). If F is a finite
set and XF = (Xj)jEF E (~L)F, then the aggregate of the vector XF over
the "population" F is denoted XF and defined as XF = "£jEF Xj E ~L.
We deal with a finite set of consumers or individuals, represented
by a set I. A generic consumer is denoted i or j. The population I is
partitioned into households, i.e., there exists a partition P of I into non-
empty subsets. We call any such partition P a household structure in I.
Let P denote the set of all household structures in I. At times, we consider
a fixed, possibly exogenously given household structure. At other times,
we treat the household structure as an object of endogenous choice and,
hence, consider variable household structures. Notation differs accordingly.
However, h serves as a symbol for a "household" throughout this paper.
2.1 Commodity allocations

There exists a finite number f ~ 1 of commodities. Thus the commodity
space is ~l. Each commodity is formally treated as a private good, possibly
with externalities in consumption. Each consumer i E I has
• consumption set Xi = ~~
so that the commodity allocation space is X == I1 j EI Xj' Let x = (Xi),

Y = (Yi) denote generic elements of X.
2.2 Consumer allocations

We distinguish between a fixed and a variable household structure. The
general presumption is that the consumer population is divided into house-
holds. Therefore the consumer allocation space is P. Fixed household
structure means that there is a given household structure PEP, parti-
tioning the consumer population I into households. Variable household
structure means that households are endogenously formed so that some
household structure PEP is ultimately realized.
2.3 Feasible allocation of commodities and consumers

An allocation is a pair (x; P) E X X P specifying the consumption bundle
and household membership of each consumer .
• The ALLOCATION OF COMMODITIES has the form x = (Xi) =
(Xi)iEI, meaning that consumption bundle Xi E Xi is assigned to individual
i.
• The ALLOCATION OF CONSUMERS assumes the form P =
{PI,"" PH} so that H = IPI and every individual i belongs to exactly
183
one partition element Pn . The household structure P sorts consumers into

households .
• HOUSEHOLD CONSUMPTION. For a potential household h ~ I,
h -# 0, set Xh = I1iEh Xi, the consumption set for household h. Xh has
generic elements Xh = (Xi)iEh.
If x E X is a commodity allocation, then consumption for household h is
the restriction of x = (Xi)iEI to h, Xh = (Xi)iEh. If (x; P) is an allocation,
then a household h E P attains the household consumption Xh E Xh .
• FEASIBILITY. The economic units endowed with resources are house-
holds rather than individuals. Notice, however, that in an environment
with endogenous household formation, each singleton {i} is a potential
one-person household with its own endowment.
For a potential household h ~ I, h -# 0, its endowment is a commodity
bundle W h E Ri, W 2 O. A special case is
(IPR) Individual Property Rights: Wh = LiEh W{i} for each house-

hold h.
In general, the social endowment with resources depends on the household

structure. Namely, if the household structure P E 'P is in place, then the
social endowment is
Wp= LWh.
hEP
A different household structure can yield a different social endowment. Al-
lowing the endowment of a household to differ from the sum of endowments
of the potential one-person households formed by its members can be in-
terpreted as resource costs of setting up households or, in the opposite
direction, as economies of scale enjoyed by larger households.
We call an allocation (x; P) E X X 'P feasible, if
LXi =Wp. (1)

iEI
With free disposal of resources, the feasibility requirement could and should
be relaxed. We call an allocation (x; P) E X x 'P weakly feasible, if
(2)
2.4 Consumer preferences

In principle, a consumer might have preferences on the allocation space
X x'P and care about each and every detail of an allocation. For individual
i El, we assume that i has preferences on X x 'P represented by a
• utility function Ui : X x 'P --+ R.
184
In the sequel, we shall make the general assumption that an individual

does not care about the features of an allocation beyond the boundaries
of his own household. That is given a particular household structure, he
is indifferent about the affiliation and consumption of individuals not be-
longing to his own household. Condition HSP is a formal expression of this
assumption.
(HSP) Household-Specific Preferences: Ui((Xi P)) = Ui((Xhi h))

for i E h, hE P, (Xi P) E X x 'P.
If a fixed household structure P is given, then the arguments P and h

of the utility function may be omitted and HSP reduces to the condition
of intra-household externalities employed in HaIler (1995).
The general assumption HSP is justifiable on the grounds that we want

to design a model where households - composed of one or several per-
sons - play a significant role in the allocative process. HSP still admits a
lot of flexibility. For example, it allows for pure group externalities which
solely depend on the persons belonging to a household. It also permits var-
ious kinds of consumption externalities. Consumption externalities can be
anonymous. An individual cares only about its own consumption and aggre-
gate consumption in the household, not the composition of the household
or who consumes exactly what among fellow household members. Con-
sumption externalities can also be personal and therefore the extent of the
externalities does not only depend on the level of consumption, but also on
the specific persons who consume in the household. For instance, a person
may dislike smoking by other members of a group, but he may suffer less
if a specific person smokes.
To formulate some of these externalities, we need a bit more notation.
For i E I, put Hi == {h <;;;; Ili Eh}. Hi is the set of potential households
of which i would be a member. If hE Hi and Xh E Xh, then we can write
Xh = (Xi,Xh\i) where h\i serves as shorthand for h\{i} and
Xh\i E Xh\i = IT Xj
jEh\i
describes the consumption of household members j other than i. Now we
are ready to formulate externalities.
(MON) Monotonicity: Ui(Xi, Xh\i) is increasing in Xi for all i E I, h E

Hi.
Strict Monotonicity: Ui(Xi, Xh\i) is strictly increasing in Xi for all i E
I,h E Hi.
(NNE) Non-Negative Externalities: Ui(Xi, Xh\i) is non-decreasing in
xh\i for all i E I, hE Hi.
185
(NPE) Non-Positive Externalities: Ui(Xi, Xh\i) is non-increasing in

Xh\i for all i E I, h E Hi.
(PGE) Pure Group Externalities: For each consumer i, there exist

functions U? : Xi -+ Rand Uf : Hi -+ R such that Ui(Xh; h) =
U?(Xi) + Uf(h) for Xh E X h, hE Hi.
A special case of PGE is group size externality where Uf(h) =

Uf (Ihl). A very special case of PGE is the absence of externalities, cor-
responding to Uf = O.
PGE assumes that one can additively separate the pure consumption
effect UiC ( Xi) from the pure group effect Uf (h). In general, this is impos-
sible. Separation with respect to the consumption of individual household
members may be possible instead. Finally, as a polar case to individual
separability, a consumer may only care about the aggregate consumption
of his fellow household members. This gives rise to notions of local and
global anonymity.
(SEP) Separable Externalities: Ui((Xi,Xh\i); h) =

LjEh\ivt(Xj) for i E I,h E Hi.
(LAN) Local Anonymity: Ui( (Xi, Xh\i); h) Ui( (Xi, Xh\i); h) for i E
I,h E Hi'
(GAN) Global Anonymity: Ui((Xi, Xh\i); h) = Ui((Xi, Xh\i)) for i E

I,h E Hi'
3 Competitive equilibrium among households

We first define the concept of competitive equilibrium among households.
Secondly we show, for a fixed household structure, the existence of com-
petitive equilibria among households.
3.1 Definition
In an equilibrium among households, a household chooses an efficient con-
sumption schedule for its members, subject to the household budget con-
straint. Throughout this and the following subsection, we take a household
structure PEP as given. First, we consider a household h E P and a price
system p E lRe. For
denote
186
Then h's budget set is defined as
Bh(p) = {Xb E Xh : P*Xb ~P·Wh}.
We define next the efficient budget set EBh(p) by
Xh = (Xi)iEh E EBh(p) if and only if
(1) Xb E Bh(P) and

(2) there is!!Q Yb E Bh(P) such that
Ui(Yb; h) ~ Ui(Xb; h) for all i E hj

Ui(Yh; h) > Ui(Xb; h) for some i Eh.
Finally, a competitive equilibrium among households given
household structure P or a P-equilibrium for short is defined as a
price system P together with an allocation x = (Xi) satisfying
(1) Xh E EBh(p) for all h E P and

(2) Li Xi = wp.
Thus in a competitive equilibrium, each household makes an efficient choice
under its budget constraint and markets clear. Efficient choice by the house-
hold refers to the individual consumption and welfare of its members, not
merely to the aggregate consumption bundle of the household. At times, we
replace the exact market clearing condition by the following free disposal
condition
(3) Lixi~WP.
The following budget exhaustion property proves useful in the welfare
analysis of HaIler (1995) and Gersbach and HaIler (1997). It is also hy-
pothesized in our first existence result, Proposition 1.
(BE) Budget Exhaustion: For each household h E P, any household

consumption profile Xb E Xh, and any price system P E Ri,
3.2 Existence
We maintain the assumption of a given household structure P. It turns out
that a proof of existence of a P-equilibrium is similar to proving existence
of a competitive equilibrium among individuals. In both cases, one uses a
routine argument after demonstrating suitable properties of market excess
demand.
187
In our first approach to the existence problem, we follow the path out-
lined by Ellickson (1993) and many others who, in the succession of Debreu
(1952), rely on Kakutani's Fixed Point Theorem.
For the time being, suppose that household endowments are strictly pos-
itive: Wh »0 for each h E P. Choose k > 0 such that the social endowment
Wp belongs to the cube I( = [0, k]l. Set K = [0, 2kJf. Suppose further that
for all h E P, i E h, "the utility function Ui (Xh; h) is continuous and con-
cave in xh."Next consider a household hE P which maximizes, for each
p E ~ == {p E R~ : L,j Pj = I}, its aggregate welfare Wh given as
Wh(Xh) = I: Ui(Xh; h)
iEh
on its truncated budget set Bh(p) n Kh. Since Bh(P) n Kh is compact

and non-empty and Wh is continuous, Dh (p) == the set of aggregate welfare
maximizers, is non-empty and compact. Since Bh(p)nK h is convex and Wh
is concave, Dh(P) is convex. Since Wh » 0, the constraint correspondence
BhO nKh is continuous. Therefore, by the Maximum Theorem (Ellickson
(1993; Th. 5.47)), the demand correspondence DhO is u.h.c.
Suppose in addition (BE). Then the market excess demand correspon-
dence cP : ~ --£> [-k, 2IIlk]l, given by
cp(p) == I:[Dh(P) - Wh]

h
is convex-valued, u.h.c., and satisfies the strong form ofWalras Law. Hence
by Th. 6.37 of Ellickson (1993), there exists a pair (p*, z*) E ~x [-k, 2lIlk]l
s.t.
(a) z* E cp(p*) and
(b) z* :S 0 and zj = 0 whenever p* »0.
This result yields the following
Proposition 1 Suppose BE and
(i) Wh »0 for all h E P;
(ii) Ui(Xh; h) continuous and concave in Xh for all h E P, i E h.
Then there exists a P-equilibrium with free disposal (p; x). If, moreover,
there is at least one household whose members have strictly monotonic pref-
erences and enjoy non-negative externalities, then exact market clearing
prevails at the equilibrium (p; x).
Proof. Take (p*, z*) E ~ X [-k,2IIlk]l with (a) and (b) whose existence
has already been established. z* E cp(p*) means
188
where xb = (Xt)iEh E Dh(P*) for each h E P. Furthermore, z* ~ 0 implies

Eh xb ~ Wp and consequently xi ~ Wp for all i. Therefore xb E Kh for
all h.
First, we assert that xb maximizes W h on the non-truncated budget set
Bh(P*). For suppose not. Then there exists Yh E Bh(P*) with Wh(Yh) >
Wh(X b ). Since xb E Bh(P*) n Kh, there exists oX E (0,1) with Axb +
(1 - oX)Yh E Bh(P*) n Kh. By concavity of Wh, Wh(oXX b + (1 - oX)Yh) ~
oXWh(Xb) + (l-oX)Wh(Yh) > Wh(xb),contradictingxb E Dh(P*). Hence to
the contrary, xb is an aggregate welfare maximizer on Bh(p*) as asserted.
Next we assert xb E EBh(P*), For suppose not. Then there exists Yh E
Bh(p*) with Ui(Yh; h) ~ Ui(X b ; h) for all i E hi Ui(Yh; h) > Ui(X b ; h) for
some i E h.
Therefore Wh(Yh) > Wh(X b), contradicting the fact that xb is an ag-
gregate welfare maximizer on Bh(P*), Hence to the contrary, the assertion
holds.
Both assertions combined mean that (P*; x*) is a P-equilibrium with free
disposal. Finally, if the members of some household h E P have strictly
monotonic preferences with non-negative externalities, then p. ~ 0 and
(b) renders z· = O. •
An alternative approach to the equilibrium existence problem exploits a
well known Excess Demand Lemma which is a consequence of Brouwer's
Fixed Point Theorem. To this end, we denote by ~ 0 the relative interior
of the unit price simplex ~ and by a~ the relative boundary of ~. For
Z : ~0 - Ri, we consider the following conditions:
Strong Walras Law: p. z(p) = 0 for all Z E ~ 0 •
Continuity Condition: z is continuous and bounded from below.

Boundary Condition: If p E a~ and (Pn)nElIl is a sequence in ~o with
Pn - p, then IIz(Pn)1I2 - 00.
Lemma 2 (Excess Demand Lemma) Let e~ 2.
Suppose Z : ~o _Re satisfies the Strong Walras Law and the above conti-
nuity and boundary conditions. Then there exists p* E ~o with z(p*) = O.
Remark. In the case e= 1, the assertion of the lemma holds trivially.
Namely, then ~ = ~o = {1} and 1· z(l) = 0 implies z(l) = O.
Modification of standard arguments reduces the proof of the next two
propositions to the excess demand lemma. Under the hypothesis of Propo-
sition 3 below, the excess demand of each household satisfies the Strong
Walras Law, and the continuity and boundary conditions. Hence aggre-
gate excess demand has the desired properties. Under the hypothesis of
Proposition 4, the excess demand of some household satisfies the strong
Walras Law, and the continuity and boundary conditions. The excess de-
mand Zh(P) of any household h satisfies p. Zh(P) ~ 0 for all p E ~o and
189
the continuity condition. Replace then Zh(p) by the artificial excess de-
mand Zh (p) = Zh (p) + ~p. Then the aggregate excess demand of the
economy has the desired properties.
Proposition 3 Suppose that wp » 0 and each consumer i satisfies the

following conditions:
• Ui is continuous and strictly concave.
• Ui is strictly monotonic and exhibits non-negative externalities.
Then a P-equilibrium (with exact market clearing) exists.
Proposition 4 Suppose that

(i) each consumer i satisfies:
• Ui is continuous and strictly concave;
(ii) at least one household h E P satisfies Wh » 0 and for i E h,
• Ui is strictly monotonic and exhibits non-negative externalities.
Then a P-equilibrium (with free disposal) exists.
4 Core allocations
We have reserved the symbol h for non-empty subsets of I interpreted as
"households". When we use instead the symbols C (or sometimes S) for a
non-empty subset of I, we want to indicate that we want to consider C as a
"coalition". The sub-population C is called a coalition, if it constitutes a
sub-economy. Feasible allocations for a coalition or sub-economy are defined
in analogy to feasible allocations for the full economy based on the entire
population I. Within a sub-economy -like in the full economy-
(a) households form;
(b) the available resources are allocated.
We thus can distinguish between four levels of social organization:
(1) Individuals.
(2) Households.
(3) Coalitions.
(4) Entire Population (Society, Grand Coalition).

190
Only individuals, households, and the entire population enter the descrip-
tion of a feasible allocation of commodities and consumers. Therefore,
households enjoy an operational status. In contrast, coalitions have a
hypothetical status. Given an allocation of resources and people, one
runs through a multitude of thought experiments where in each case the
allocation is tested against a feasible alternative for some coalition. If the
allocation passes all conceivable tests, it is distinguished as a core alloca-
tion. Since we are dealing with the allocation of commodities and people,
it is crucial what a coalition can and cannot do. As already indicated, a
feasible alternative for a coalition has two components:
(a) an allocation of its members into households, i.e., a household sub-
structure for the coalition;
(b) an allocation of its available resources (which may depend on the chosen
household structure) among its members.
Requirement (a), that a coalition has to come up with a household struc-
ture of its own, is non-traditional, but quite natural. It is a mild consistency
condition. Coalitions are subject to similar social constraints as society at
large. Consequently, if the household structure is variable, a coalition is
also flexible in the choice of its household structure. On the other hand, if
the population has to live with a fixed household structure, every coalition
is bound to conform with this household structure, too. Requirement (b)
is standard. The coalition has to rely on its own resources and further, to
be taken serious, it should have a practicable plan on how to use these
resources.
Next, we are going to express the feasibility conditions for a coalition in
a formal way. For a coalition C, let Qc denote the collection of partitions
of C into non-empty subsets. An element Q E Qc represents a household
structure for coalition C. The resources available to coalition C with
household structure Q E Qc equal
C is only a sub-economy in its own right, if it allocates consumers and

commodities under autarky. Accordingly, we define a feasible allocation
for C as a pair (ye; Q) E Xc x Qc such that
Ye =wQ.
where Ye = (Yi)iEe and Ye = EiEeYi'

Before we proceed to a formal description of core allocations, let us men-
tion an implicit assumption which is, perhaps, less innocuous in our context
than in traditional settings. In practice, the dissolution of households (mar-
riages) and coalitions can be painful and costly. Like in classical general
191
equilibrium theory and most of cooperative game theory, we assume zero

cost of dissolving existing groups. Two interpretations of a core allocation
are consistent with this assumption.
(1) There exists a status quo allocation. Individuals are free to regroup
themselves at no cost and to reallocate their resources. If no group (coali-
tion, household) can benefit from going its own way, the status quo is a
core allocation.
(2) There exists a proposed allocation at a stage where the allocation
of resources and people is to be decided de novo. Since no groups have
formed yet, no costs of dissolution can accrue. Instead of implementing
the proposal, individuals are free to pursue alternatives. If no group would
benefit from going its own way, the proposed allocation is a core allocation.
4.1 Constrained or household core

Suppose an immutable household structure P. We assume that the notion
of a household is legally, economically, or socially meaningful. Household
members or factions cannot entertain outside liaisons without the knowl-
edge, approval or at least tacit agreement of the rest of the household.
Therefore, if a household member belongs to a coalition, then implicitly
the entire household is part of that coalition. Therefore, since the house-
hold structure P is cemented, the only coalitions of interest here are those
composed of households in P. Let Q(P) denote the family of non-empty
subsets of P. For C E Q(P), define
G(C):= U h.
hEG
G(C) is the coalition consisting of all the constituents of all the households
in C. The fixed household structure P implies that C is the only possible
household structure for coalition G(C).
Definition 1 x E X belongs to the P-core or household core w.r.t. P
or constrained core w.r.t. P, denoted C(P), if
(i) x E X and
(ii) there is l!Q C E Q(P) and Y E X with:
(a) Ui(y; P) ~ Ui(x; P) for all i E G(C).
(b) Ui(y; P) > Ui(X; P) for some i E G(C).
(e) LiEC(G) Yi = Wo = LhWWh'
Notice that the P-core just defined is of the "strong core" variety. The
corresponding "weak core" is in general larger.
Definition 2 x E X belongs to the weak P-core, if

192
(i) x E X(P) and

(ii) there is 1!Q G E Q(P) and Y E X with:
(a) Ui(Y; P) > Ui(X; P) for all i E C(G).
(b) LiEC(G) Yi = Wc = LhEG Who
In general, existence of P-core allocations is not at issue, if P-equilibria
exist. The key result is due to HaIler (1995).
Proposition 5 If (p; x) is a P-equilibrium and BE holds, then x belongs

to the P-core.
Combining the existence result of Proposition 1 or 3 with the core inclu-

sion result of Proposition 5 establishes existence of P-core allocations.
4.2 Full core

The presumption here is that both commodities and consumers can be
freely allocated. Still, we posit that if a household member belongs to a
coalition, then the entire household is part of that coalition. But now we
allow for an endogenous household structure so that dissenting household
members have the option to secede, try to form new households and join
coalitions including such newly formed households.
Definition 3 An allocation (x; P) belongs to the {full} core, denoted C+,

if
(i) (x; P) is feasible and

(ii) there is no coalition C together with a feasible allocation for coalition
C, (YC;Q) E Xc x Qc such that
(a) Ui(Yh; h) 2:: Ui(Xh; h) for all i E h, hE Q,

(b) Ui(Yh; h) > Ui(Xh; h) for some i E h, hE Q.
Definition 4 An allocation (x; P) belongs to the weak core, denoted C-,

if
(i) (x; P) is feasible and

(ii) there is 1!Q coalition C together with a feasible allocation for coalition
C, (yc; Q) E Xc x Qc such that Ui(Yh; h) > Ui(Xh; h) for some
i E h, hE Q.
Our model is related to the club literature which for the most part pre-
sumes also a partition of the population into groups. Clubs and households
alike allow for intra-group externalities. The framework developed in this
193
paper differs from club theory in two respects. First, club theory empha-
sizes individual choices to participate in clubs whereas we focus on collective
decisions within groups. Second, clubs exist primarily, if not only, for the
provision of a local public good, also named "club good" or "public pro-
ject". In serving this purpose, the club incurs a resource cost which it tries
to recoup through the collection of (possibly personalized) admission fees.
However, the original claims to economic resources rest with the individuals
and the purchase of goods for private consumption is left to each member
who is subject to an individual budget constraint. In contrast, a household
forms an economic entity involved in the provision of local public goods
as well as the procurement of private consumption goods for its members.
It may be the household - rather than its members personally - who
is entitled to (endowed with) resources. And it is only the household, not
each member, who is subject to a budget constraint.
However, for the full core concept the different aspects of clubs and house-
holds are irrelevant. A feasible alternative for a coalition is an allocation
of its members into households or clubs and an allocation of its available
resources among its members. Hence, a feasible alternative for a coalition
has the same requirements for clubs and households. Thus, a full core allo-
cation in the framework of households is a full core allocation in the club
framework and vice versa if the same intra-group externalities are present.
Gilles and Scotch mer (1997) have shown that the core may be empty when
preferences are non-convex. In the next section, we provide an example that
the full core can be empty, even if preferences are convex.
4.3 Non-existence of full core allocations

We illustrate by means of the following example that the full core can be
empty, C+ = 0. Indeed, the example exhibits emptiness of the potentially
larger weak core: C- = 0. Moreover, it will become obvious from this ex-
ample that household formation can exhibit cycles. The latter can happen
in an ex ante homogeneous society that is one where all individuals have
the same endowments and preferences to begin with.
There is a single commodity, i.e., e
= 1. There are three consumers
labelled i = 1,2,3, with generic consumption bundles Xi. For a potential
household h C I, h f:. 0, its endowment is Wh = Ihl. In particular, each
individual has endowment Wi = W = 1. Moreover, we assume a weak form
of separability and GAN, i.e.,
Specifically, we assume
u® = rand v® = r - ~r2
9
=r - .!..r2 for r > 0,
a -
194
where 0: = 9/4 and va

= 3/2. Hence ex ante, individuals are completely
homogeneous with respect to endowments and utility functions. Towards
a detailed examination of the weak core, C- , we introduce a few auxiliary
functions:
• y2(y) = u(y) + v(2w - y) for y E [0,2] is the utility of an individual
belonging to a 2-player household, if this individual consumes y and the
other household member consumes the rest of their joint endowment 2w.
For the purpose of core analysis, only the case y2(y) E [1,2) is of interest.
Namely, y2(y) attains its maximum at y =
2 with value 2 in which case
the other household member enjoys utility v(2w) = v(2) = 2/9. However,
an individual can always secure a utility level of u(w) = u(l) = 1.
For U E [1,2), the equation y2(y) = U has two solutions. The smaller
one is y2(U) = 2 - )(2 - U)o:. It yields the highest utility to the other
household member, given the chosen member attains U .
• y3(y) = u(y) + v(3w - y) for y E [0,3] is the utility of an individual
belonging to a 3-player household, if this individual consumes y and the
other household members share the rest of the joint endowment 3w. For the
purpose of core analysis, only the case y3(y) E [1,3) is of interest. Namely,
y3(y) attains its maximum at y = 3 with value 3 in which case each of the
two other household members enjoys utility v(3w) = v(3) = -1. However,
an individual can always secure a utility level of u(w) = u(l) = 1.
For U E [1,3), the equation y3(y) = U has two solutions. The smaller one
is y3(U) = 3 - )(3 - U)o:. It is most favorable to the remaining household
members, given the chosen member attains U.
We now can make the following assertions - to be derived later.
(i) v(w)=I>O.
(ii) u(O) = v(O) = O.
(iii) u'(Y) > v'(z) for all y ~ 0 and z > O.
(iv) u" ~ 0 and v" ~ O.
u(3W-f(U)) +v(3w+f(U)) < u(2 - y2(U)) +v(y2(U))
(v)
for all U E [1,5/3].
(vi) u(3w-g 3(U)) +v(3w+g3(U)) < u(l) +v(l) for all U E (5/3,3).
The asserted properties allow to establish emptiness of the full core.
Proposition 6 For the economy as numerically specified in this sub-

section, C- = 0 and a fortiori C+ = 0.
Proof.
a.) The condition (i) implies that {x = (1,1,1); P = {{I}, {2}, {3}} is
195
not in C- , since forming a 2-person household would always benefit both

members, e.g., by allocating the resources equally within the household.
b.) We claim that conditions (ii)-(vi) imply that P = {I} is not part of a
weak core allocation.
First, we observe that (iii) and (iv) allow us to limit the test for core
allocations to cases where at least two individuals receive the same util-
ity levels. For suppose there were a feasible allocation with P = {I} and
three different utility levels, say UI > U2 > U3, and by (iii), correspond-
ing commodity bundles Xl > X2 > X3. Then keep P = {I}, but consider
the new commodity allocation 3w;-X3, 3w;-X3, X3 and the associated utility
a.
levels Ut, U2, U Clearly, the third individual still achieves the previous
a
utility level, U = U3 . Because of (iii), the first individual experiences a
decrease and the second individual experiences an increase in utility. Our
claim is that if the original allocation was in C- , then the new allocation
has also to be in C-. Because of a.), the only coalitions - and households
- to be considered are
a.) {1,2};
f3.) {2,3};
,.) {1,3}.
In case a.), we observe that Ut + U2 ~ UI + U2, since by (iv), the
functions u and v are concave and the particular redistribution among
individuals 1 and 2 can only increase their utility sum. Hence, if coalition
(household) {1, 2} was unable to improve upon the original allocation, then
it cannot improve upon the new allocation either. In case f3.), note that
a
U2 > U2, U = U3 and thus the analogue of the argument in case a.)
applies. In case ,.), the observation Uf = U2 allows a reduction to case
f3.). Hence the new allocation passes all crucial tests.
=
Secondly, we examine the case where P {I} and two individuals, with-
out loss of generality individuals 1 and 2, receive the same utility, UI = U2 •
By (iii), the respective commodity bundles must satisfy Xl = X2. Given
U3, the utility attained by individual 3, the best case scenario for the other
two individuals is X3 = y3(U3) and Ui = u(3w_~3(U3)) + ve w+ y23(U3)) for
i= 1,2.
<> If U3 < 1, then individual 3 would be better off forming her own
household.
<> If U3 E [1,5/3]' then in the 2-person household {1, 3}, individual 1 can
achieve the utility level Ui = u(2 - y2(U3)) + v(y2(U3)) while individual
3 maintains utility level U3. By (v), UI < Ui. By an additional small
reallocation from 1 to 3, coalition {1, 3} reaches an allocation that improves
upon the original one.
<> If U3 > 5/3, then by allocating resources equally within the 2-person
household {1, 2}, each individual i = 1,2 can achieve the utility level u( 1) +
196
v(l) which, by (vi), exceeds Ui. Therefore, the coalition {I, 2} can improve
upon the original allocation.
c.) Up to permutations, we are left with the case where individuals 1 and
2 are together in one household and individual 3 stays alone. We commence
with the special case where 1 and 2 achieve the same utility level, i.e., U1 =
U2 = u(l) + v(w). By (i), the latter exceeds the autonomous utility level
U3 = u(l). Therefore, by a slight deviation from equal distribution (within
the household {I, 3}) in favor of 1, the coalition {1,3} can improve upon
the given allocation. If the original utility levels of 1 and 2 are different,
then without restriction U1 < U2 and already equal distribution within
household {I, 3} yields an improvement for that coalition.
d.) So far, we have shown that (i)-(vi) imply emptiness of the weak core,
indeed. It remains to show (i)-(vi). (i)-(iv) are obvious. For U E [1,5/3]'
the inequality in (v) amounts to
9a + 12)(3 - U)a < 20 + 16)(2 - U)a + 3aU.

= =
It can be explicitly verified for U 1 and U 5/3. Next collect all variable
terms in U on the right hand side of the inequality:
F(U) 3aU + 16)(2 - U)a - 12)(3 - U)a

6.75U + 24)2 - U - 18)3 - U.
Setting JL = )2 - U, one obtains U = 2-JL2. The function 4>(JL) == F(2-JL2)

satisfies 4>1(1) < 0 and 4>" > O. Therefore 4>1 < 0 and FI> 0 in the relevant
range. Hence (v). Suppose U E (5/3,3]. Then
)(3 - U)a < /f1 =

_. -
3
9
4
V3.
Hence the left hand side (LHS) of the inequality of (vi) satisfies
LHS = 3 - 4~(6 - )(3 - U)a)2
On the other hand, u( 1) +v (1) = 194 . Direct comparison shows 1i (J3 -1) <
194 . Hence (vi) . •
Comment on Cycles. The preceding proof reveals also the existence of

cycles in the formation of households. Suppose one allows for sequential
coalition formation, beginning with any situation with a 2-person and a 1-
person household. This initial situation can be improved upon by a different
2-person household forcing the third person to go solo, and so forth.
197
Comment on Non-Emptiness. If the number of individuals is changed

from 3 to 4, then the full core is no longer empty: A core allocation con-
sists of two 2-person households with equal distribution of resources within
each household. The same is true for economies with an even number of
individuals while the full core remains empty if the number of individuals
is uneven.
4.4 Existence of core allocations

In the conventional setting where externalities are absent, one possible
proof of existence of weak core allocations proceeds in two steps. First,
one derives a "cooperative" or "coalitional" game from the pure exchange
economy. Secondly, one verifies a balancedness condition. We restrict our-
selves to a minimal elaboration of this procedure. All missing links can be
retrieved from Chapter 4 in Hildenbrand and Kirman (1988).
Formally, we define for each non-empty subset C of I, V(C) as the com-
prehensive hull of attainable utility allocations for coalition C as follows.
there exists a feasible (YC; Q) E Xc X Qc }

V(C) == { ~ = (~i)iEI E ]RI such that for all hE Q, i Eh:
~i ::; Ui(Yh; h)
A family B ~ 21 is called balanced, if there exist non-negative "balanc-

ing weights" >"s, SE B, such that for each i E I,
:L>"s=l,
SEl3;
where Bi == {S E Bli E S}.

Call the economy balanced, if for every balanced family B of coalitions,
n V(S) ~ V(I).
SEB
Due to the seminal contribution of Scarf (1967), we can state

Proposition 7 The weak core of a balanced economy is not empty.
Example 1. Assume "individual property rights", i.e., Wh = LiEh W{i} for

all potential households h. Further assume LAN, NPE, and Ui ( (Xi, Xh\i); h)
=Ui((Xi,Xh\i); Ihl) concave in Xh = (Xi,Xh\i) and weakly decreasing in
coalition size Ihl for i E I, hE 'Hi.
This economy is balanced. For let B be a balanced family of coalitions
with balancing weights >"s, S E B. Furthermore, let ~ E nSEB V(S). To
prove our claim, we have to show ~ E V(I). For every S E B, there exists
a feasible (y~,Qs) E Xs X Qs with ~i::; Ui(y~;h) for i E h,h E Qs.
198
Because of individual property rights, feasibility is not affected by split-

ting S into single--person households. Ceteris paribus, a person's utility can
only increase, if household size is reduced. Hence ~i ~ Ui (yf i {i}) for i E S
and LiES yf = LiES W{i}'
Now define an allocation x by setting
Xi =L >'s . yf.
SEB i
Since Xi is a convex combination of the yf, S E Bi , and Ui is concave in

consumption, ~i ~ Ui(Xii{i}) for i E S. Moreover, (Xi {{i}li E I}) is a
feasible allocation for the entire economy. Namely,
LiEI Xi = LiEI LSEB; >'s . yf LSEB >'s LiESyf

LSEB >'s LiES W{i} LiEI W{i} LSEBi >'s = LiEI W{i}'
This shows ~ E V(I) as asserted. Hence the economy is balanced and its
weak core is non-empty. In particular, there exits a weak core allocation
where the population is segmented into single--person households. 0
Example 2. Assume individual property rights. Further assume LAN,
NNE, and Ui((Xi,Xh\i)i h) = Ui((Xi,Xh\i)i Ihl) concave in Xh and weakly
increasing in coalition size Ihl for i E I, hE Hi' In analogy to the previous
example, one can show that this economy is balanced and its weak core
is non-empty. In particular, there exits a weak core allocation where the
whole population forms one big family. If MON is added to the assumed
properties, then individual property rights can be replaced by the weaker
condition of "weakly balanced property rights":
L >'sws ~
SEB
WI
for every balanced family B of coalitions with balancing weights >'s, SE B.

As is well known, for the latter condition to hold it is neither necessary nor
sufficient to have "super-additive property rights", i.e., Whuk ~ Wh +Wk for
any two disjoint potential households hand k. 0
Example 3. The two foregoing examples represent polar cases in the sense
that in the first example a segmentation into single--person households con-
stitutes a core outcome and in the second example formation of the biggest
household constitutes a core outcome. But less extreme possibilities exist.
Let us return to the counter-example of the previous sub-section. In that
example, we could observe cycles of household formation. But if the num-
ber of consumers in that example is changed from 3 to 4, then the strong
core is no longer empty. 0
199
5 Conclusion
Excess demands resulting from efficient collective decisions by multi-
member households tend to be structurally similar to the preference max-
imizing excess demands of individual consumers. This fact enables us to
show the existence of competitive equilibria among multi-member house-
holds. Equilibrium existence combined with a core inclusion result yields
non-emptiness of constrained cores. In contrast, the full core can be empty
under standard assumptions. The reasons for this are two-fold. First, any
coalition qualifies as a potential household. Secondly, intra-household ex-
ternalities are present. This points to two sets of restrictions guaranteeing
a non-empty full core. First, one can impose restrictions on the size and
structure of households. Examples of this kind can be found in Roth and
Sotomayor (1990) who show that the core in two-sided matching models
is non-empty and equals the set of stable matchings. Secondly, one can
impose restrictions on the nature of externalities. Our examples 1 and 2
are polar cases of certain types of externalities which yield non-empty full
cores.
References
Becker, G.S: "A Theory of Marriage", Chapter 11 in G.S. Becker: The
Economic Approacb to Human Behavior. The University of Chicago
Press: Chicago. Paperback edition, 1978; pp. 205-250.
Becker, G.S.: A Treatise on the Family. Enlarged Paper Edition. Harvard

University Press: Cambridge, MA. Harvard University Press, 1993.
Chiappori, P.-A.: "Rational Household Labor Supply", Econometrica 56,

1988,63-89
Chiappori, P.-A.: "Collective Labor Supply and Welfare," Journal of Po-

litical Economy 100, 1992, 437-467.
Debreu, G.: "A Social Equilibrium Existence Theorem", Proceedings of

the National Academy of Sciences of the U.S.A. 38, 1952, 886-893.
Ellickson, B.: Competitive Equilibrium. Cambridge University Press:

Cambridge, 1993.
Gersbach, H., and H. Haller: "Collective Household Decisions and Com-

petitive Markets", Virginia Polytechnic Institute and State Univer-
sity, WP E-97-07, 1997.
Gersbach, H., and H. HaIler: "Intra-Household Bargaining Power and

Equilibrium Allocation," mimeo, 1998.
200
Gilles, R.P., and S. Scotchmer: "Decentralization in Replicated Club

Economies with Multiple Private Goods", Journal of Economic The-
ory 72, 1997,363-387.
HaIler, H.: "Household Decisions and Equilibrium Efficiency", Virginia

Polytechnic Institute and State University, WP E-95-12, March 1995.
Hildenbrand, W., and A.P. Kirman: Equilibrium Analysis. North-Holland:
Amsterdam et al., 1988.
Roth, A.E., and M.A.O. Sotomayor: Two-Sided Matching: A Study in
Game-Theoretical Modelling and Analysis. Cambridge University
Press: Cambridge, 1990.
Scarf, H.E.: "The Core of an N-Person Game", Econometrica 35, 1967,
50-69.
Valuation equilibrium revisited"
Peter Hammond 1 and Antonio Villar 2
1 Department of Economics, Stanford University, RalphLandau Economics
Building, Stanford, CA 34305-6072, U.S.A

2 University of Alicante & IVIE Department of Economics, University of
Alicante, 03071 Alicante, Spain
Abstract. This paper extends the notion of valuation equilibrium which

applies to market economies involving the choice of a public environment.
Unlike some other recent work, it is assumed here that consumers and firms
evaluate alternative environments taking market prices as given (hence this
notion is closer to that of competitive equilibria). It is shown that valuation
equilibria with balanced tax schemes yield efficient allocations and that
efficient allocations can be decentralized as valuation equilibria, with tax
schemes that may be unbalanced.
JEL Classification Numbers: D51, H40
Keywords: Valuation equilibria, tax schemes
1 Introduction
Market economies can be regarded as collections of economic agents that
take individual decisions over their private feasible sets. Like budget sets
in the Walrasian model, these sets depend on prices and income from prof-
its. But there are also influenced by conditioning variables that may be
determined outside the market mechanism. Examples of these conditioning
variables are the prevailing social rules (e.g., laws, and in particular the
assignment of property rights), the provision of public goods by the pub-
lic sector (roads, health care, unemployment benefits), the regulation of
economic activities (the presence of quotas, the regulation of quality stan-
dards or labour conditions), the working of a tax system, etc. We can refer
to the collection of these conditioning variables as the public environment.
It is worth stressing that the concept of public environment allows many
different problems to be treated within one common setting. The cost of
'Thanks are due to Luis Corch6n, Ignacio Ortuno and Joaquim Silvestre for helpful
comments on an earlier version of this work. We wish also to acknowledge the financial
support provided by the DGICYT under project PB92-0342.
202
this generality is that little structure can be imposed on the set of public
environments. The chief question is the analysis of efficient allocations in
such a framework.
Mas-Colell's (1980) contribution to the pure theory of public goods pro-
poses the notion of valuation equilibria to address this problem. He con-
siders an economy involving a single private good (to be interpreted as a
Hicksian composite commodity), and the choice of a single public project
from a space where no linear structure is imposed. This equilibrium no-
tion may be regarded as an application to this abstract framework of the
Lindahlian approach to the provision of public goods. A valuation equi-
librium corresponds to a unanimously agreed choice of public project, the
role of the public sector being to design a mechanism that induces such
an efficient agreement. In a valuation equilibrium every consumer prefers
her consumption plan, consisting of some amount of the private good and a
public project, to any other which is affordable given the valuation function
(to be interpreted as tax system). Taking the private good as the numeraire,
Mas-Colell shows that valuation equilibria satisfy the standard properties
of competitive equilibria.
Later, Mas-Colell & Silvestre (1989) introduced the concept of cost-
share equilibrium which extends the work of Kaneko (1977) and Mas-Colell
(1980), to allow many public goods. Even though in some cases this model
allows for many private goods, the authors point out that ''the extension
that seems to us very difficult is to allow for produced intermediate goods."
(cf. p. 255) [see also Diamantaras & Wilkie (1994)].
Allowing several private goods offers a choice of how to extend the notion
of valuation equilibrium. One possibility is to assume that agents evaluate
alternative actions taking the equilibrium prices as given. The alternative is
to assume that they compute the new prices that would emerge in different
environments. In the first case, agents compare alternatives disregarding
the effect on prices that results from a change in the environment (as in
the standard competitive or Lindahl equilibrium model). In the second
case, they are highly sophisticated and able to calculate the set of condi-
tional equilibria associated with an alternative public environment (and to
coordinate on which one if there are several).
Diamantaras & Gilles (1996) and Hammond & Villar (1998) extend Mas-
Colell's work by taking the second avenue. They present slightly different
models with an abstract set of public projects and several private com-
modities that may be both inputs and outputs. Their notion of valuation
equilibria requires agents to maximize their payoff functions taking into
account the changes in the prices of private commodities resulting from
changes in the public environment. They show that these equilibria satisfy
the two fundamental welfare theorems.
The purpose of this paper is to discuss the efficiency and decentraliz-
ability of valuation equilibria in the case when prices of private commodi-
ties are treated as fixed. Besides completing the analysis, the interest of
203
this second approach is twofold. On the one hand, it allows us to analyse

the efficiency of equilibrium allocations with fewer informational require-
ments. On the other hand, it remains closer to the notion of competitive
and Lindahl equilibria and also permits several models to be encompassed
within one common framework, as will be illustrated later on. This does
not mean that the case of sophisticated consumers is uninteresting. On the
contrary, if there are significant nonconvexities in the provision of public
goods, changes in the public environment may induce substantial changes
in the prices of private goods that should not be neglected. What we claim
here is that both possibilities are worth investigating.
We consider an economy with an abstract set Z of public environments
lacking any formal structure. Agents' choice sets and payoff functions are
defined conditional on the public environment in such a way that, for each
given z in Z, there is a standard convex economy in the subspace of private
commodities. Agents' choices are also affected by a tax system that mod-
ifies net profits and budget sets depending on the public environment. A
valuation equilibrium is defined as a public environment, a feasible alloca-
tion, a price vector and a tax system such that: (a) Every consumer prefers
her consumption bundle and the public environment to any other which
is affordable given the equilibrium prices and the tax system; (b) Firms'
profits are maximized over their production sets; and (c) The aggregate net
taxes (Le., taxes less subsidies or environmental charges) add up to zero.
Our main findings are as follows:
(i) A valuation equilibrium is Pareto optimal (provided the tax system
is balanced).
(ii) Every Pareto optimal allocation can be decentralized as a (compen-
sated) valuation equilibrium.
(iii) Every valuation equilibrium yields a core allocation, provided the
tax system does not allow for intercoalitional transfers.
The rest of the paper proceeds as follows. Section 2 presents the model
(including the equilibrium notion and the assumptions). Section 3 contains
the results. Section 4 provides some examples of standard economic prob-
lems that can be considered as particular instances of valuation equilibria.
2 The model
Consider an economy with f private commodities. The vector W E Ri rep-
resents the aggregate initial endowments of private goods. There is an ab-
stract set Z whose members are vectors of those variables defining the public
environment. Each agent's feasible set and payoff function may be affected
by the values taken by these vectors z E Z. No particular structure will
be postulated on the set Z. We assume that there is some kind of public
agency or public sector that can determine these variables and/or affect
consumers' budget sets and firms' profit functions via taxes and subsidies.
204
Even though the tax system itself can be thought of as part of the public
environment, we find it more convenient to treat these variables separately.
For the sake of generality, agents' feasible sets will be defined as subsets of
]Ri X Z (where]Ri stands for the commodity space and Z for the public en-
vironment space), even though they may actually be independent of many
of the variables in Z. A point P E ]R~ is a price vector relative to private
commodities.
There are n firms in the economy. Let Yj C ]Ri X Z be the jth firm's
production set, and denote by Yj(z) the jth firm's conditional production
set when the environment variables take the value z E Z. That is,
Yj(z):= {Yj E]Ri I (Yj,z) E Yj}.
For each firm j = 1,2, ... , n, the mapping aj : ]R~ X Z -+]R is assumed to
represent the net subsidy this firm receives (or, if negative, pays as taxes),
as a function of prevailing prices and the value of the environment variables.
Thus for each given (p, z) in ]R~ x Z, the jth firm's conditional profits are
given by 7fj(p,z) = sup{PYj +aj(p,z) I Yj E Yj(z)}. In equilibrium this
supremum should be attained. This implies that, for each given (p, z) in
]R~ X Z, the jth firm selects a production plan Yj(p,z) that maximizes
(conditional) profits.
There are m consumers. The ith consumer is characterized by the col-
lection [Xi,Ui,Wi,(Oij)], where Xi C]Ri X Z, Ui : Xi -+]R, Wi E]Ri denote
the consumption set, utility function and initial endowments, respectively,
and Oij stands for the ith consumer's share in the jth firm's profits. By
definition, 0 ::; Oij ::; 1, and Ej=10ij = 1, for all i,j. For a given point
Z E Z the ith consumer's conditional consumption set is given by:
Evidently, a preference ordering is induced on each of these conditional

sets, which can be described by a conditional utility function ut : Xi(Z) -+
]R given by Ui(Xi) = Ui(Xi, z).
A mapping Ti : ]R~ X Z -+]R, for i = 1,2, ... , m, describes the ith con-
sumer's (net) tax mapping. Given a pair (p, z) E ]R~ X Z, the ith consumer's
behaviour Xi(P,Z) is obtained by solving the following program:
Max ut
s.t. Xi E Xi(Z)
n
PX i < PWi+ I: 0ij7fj(p,Z)+Ti(p,Z).
j=l
Consider now the following definitions. The first makes precise the notion
of a tax system, whereas the second specifies a relevant restriction:
Definition 1 A taz system is a collection of mappings T
[(aj)j=l' (Ti)~l], where each aj and Ti is a function from ]R~ X Z into
205
JR, and such that for every pair (p,z) E JR~ x Z one has Ej=1 Uj(p,z) +
E:1 Ti(p,Z) :$ O.
Definition 2 A tax system T is balanced if for any pair (p,z) E JR~ x Z
one has Ej=1 Uj(p, z) + E:1 Ti(P, z) = O.
A tax system is a collection of functions from JR~ x Z into JR such that

aggregate subsidies do not exceed aggregate taxes. This amounts to saying
that the economy is financially Viable in the sense that the tax system
cannot rely on resources from outside. Note that this definition is very
general as it can depend on many variables apart from private income and
gross profits. A tax system is balanced whenever aggregate taxes equal
aggregate subsidies (e.g., a cost-share system). In particular, the case of
Uj(p,z) = Ti(p,Z) = 0 for all i,j, each (p,z) in JR~ x Z constitutes a
(degenerate) balanced tax system. Note that balancedness ensures that
Walras Lawp[E:1 Xi(P,Z)- E:1 Wi- Ej=1Yj(P,Z)] = 0 holds, provided
consumers are not satiated.
The following defines the equilibrium notion (which is reminiscent of the
modern version of Lindahl's equilibrium):
Definition 3 A valuation equilibrium is a price vector p* in JR~, an en-

vironment z* E Z, an allocation [(xi), (Yj)] in IT:1 Xi(Z*) xITj=1 Yj(z*)
and a tax system T* such that
(i) For all i = 1,2, ... , m, (xi, z*) maxtmzzes Ui over the set of points
(Xi, z) in Xi that satisfyp*xi :$ P*Wi+ Ej=1 Oij7rj(P*, Z*)+Ti(P* ,z).
(ii) For all j = 1,2, ... ,n, p*yj +Uj(P*',z*) ~ P*Yj +Uj(P*,z) for all
(Yj,z) E Yj.
(,•••;,;,;) "m
L...ti=1 Xi* - "m
L...ti=1 Wi = "n *
L...tj=1 Yj·
That is, a valuation equilibrium is a price vector, a tax system and a

feasible allocation such that no agent finds it individually beneficial to
choose an alternative allocation that is affordable at given prices. Note that
we assume that every consumer computes her budget by taking prices of
private commodities and firms' profits as given (she only allows for changes
in the public environment through the effect on her own taxes).1 Also
observe that, by definition, aggregate taxes equal aggregate subsidies in
equilibrium. Indeed, in some cases the tax system may be thought of as
an incentive scheme that sustains the equilibrium environment, so that we
1 It can be easily shown that the results in section 3 remain valid if we consider
that consumers are more sophisticated, and can also compute the effect on their budget
constraint of changes in firm's profits.
206
can picture the situation as agents choosing the environment and the public
sector choosing the incentive scheme.
When utility maximization is relaxed to expenditure minimization one
gets the standard notion of compensated equilibrium, which will be used in
order to discuss the second welfare theorem.
Definition 4 A compensated valuation equilibrium is defined like a
valuation equilibrium except that condition (i) is replaced by
(i1 For every i =1,2, ... ,m, pOx; ::; P"Wi + 2:.";=10ij7rj(p",z") +
Ti(p" ,z*) and P*Xi ~ P"Wi + 2:.";=1 Oij7rj(P* ,z*) + Ti(P* ,z) whenever
(Xi,Z) E Xi withui(xi,Z) ~ Ui(X;,Z*).
3 Main results
The first main result says that a valuation equilibrium is Pareto optimal,
provided the tax system is balanced. The second establishes that any Pareto
efficient allocation can be decentralized as a valuation equilibrium. Finally,
it will be shown that a valuation equilibrium is in the core, provided that
inter-coalitional transfers are excluded.
Theorem 1 Let [p*, (xi), (yj), z*, T*] be a valuation equilibrium, and sup-
pose that consumers are locally non-satiated. 1fT" is a balanced tax system
the resulting allocation is Pareto optimal.
Proof. Let [P*, (xi), (yj), z", T"] be a valuation equilibrium, and suppose
that [(Xi), (Yj),z] together satisfy Xi E Xi(z) and Ui(Xi,Z) ~ Ui(xi,Z") for
all i, with strict inequality for at least one agent, and Yj E 1-j(z). Then
local non-satiation and the definition of valuation equilibrium imply that
m n m
Lp*xi > pOw + L [p*y; + O"j(P*,z*)] + LTi(P*'z)
i=1 j=1 i=1
and also that 2:.";=1 p*YJ + 2:.";=1 O"j(P*, Z*) ~ 2:.";=1 p"Yj + 2:.";=1 O"j(p" ,z).
These inequalities imply that
m n n m
LP"Xi > p"w+ LP"Yj + LO"j(P*,z) + LTi(p",Z)

i=1 j=1 j=1 i=1
But T" is balanced by assumption, so [2:.";=1 O"j(P*, z)+ 2:.::1 Ti(P", z)] = O.
Hence 2:.::1P"Xi > P*W+2:.";=1P*Yj, which implies that [(Xi),(Yj),z]
cannot be a feasible allocation satisfying 2:.::1 Xi::; W+ 2:.";=1 Yj. It follows
that no feasible allocation can be Pareto superior. •
Observe that an allocation may fail to be optimal if we drop balancedness.
Indeed it is easy to produce a tax system that induces a waste of resources
in equilibrium and an even greater waste out of equilibrium (e.g., the tax
system rewards a firm which destroys part of the initial endowments).
207
Remark 1 It follows from this theorem that, in a valuation equilibrium,

the tax system cannot be arbitrary once we put some structure on the set Z.
This is because, besides balancedness, the first-order necessary conditions
for efficiency must be satisfied in equilibrium. Informally, one can say that
consumers' taxes should be equivalent to a non-linear system of Lindahl
prices.
In order to obtain the second welfare theorem for this model we need
three assumptions:
Assumption 1 For every i = 1,2, ... , m and every z E Z:
(i) Xi(Z) is a non-empty, closed and convex subset of ]Ri;
(ii) ut : Xi(Z) --t ]R is continuous, quasi-concave, and satisfies local non-
satiation.
Assumption 2 For every j = 1,2, ... , m and each Z E Z, the set 1j(z) is
closed and convex in ]Ri, with 1j(z) n]R~ = {O}, and 1j(z) -]R~ C 1j(z).
Assumptions 1 and 2 essentially say that, for any given value of the
public environment, the resulting conditional economy is standard (except
in that we do not impose bounded consumption sets here). Note that this
is compatible with the presence of non-convexities (an example will be
discussed later). They also allow public goods to affect both production
and consumption possibilities.
Observe that these assumptions involve no restriction on the set Z whose
members may therefore contain all kind of variables.
We can prove now
Theorem 2 Under Assumptions 1 and 2, let [(xi), (Yj),z*] be a Pare to
optimal allocation. Then, there exist a price vector p* E lR~ - {O} and a
tax system T* such that [p*, (xi), (Yj), z*, T*] is a compensated valuation
equilibrium.
Proof. Take z* as given and apply the standard second welfare theorem
to the allocation [(xi), (yj)] in the resulting convex conditional economy.
This theorem ensures the existence of a price vector p* E ]R~ - {O} such
that [p*, (xi), (yj)] is a competitive equilibrium relative to the conditional
economy resulting from z*. It will be shown that [p*, (xi), (Yj), z*, T*] is
a valuation equilibrium for a tax system T*. This requires checking parts
(i), (ii) and (iv) of Definition 3, part (Hi) being satisfied by construction.
Define
O"j(P*, z) = inf{O, p*yj - 7rj(P*, z)}
Note that O"j(P*,z*) = 0 because 7rj(P*,z*) = p*yJ. Also, if (yj,z') E 1j
then:
*' (* ') <
PYj+O"jP,Z *' * * ~ ( * ') <
_PYj+PYj-7rjP,Z * * * * + O"jP,Z
_PYj=PYj (* *)
208
so that part (ii) of the definition is satisfied.

Taking p*, (xi ,z*) as given, define the compensation function
This is the income that individual i needs to spend on private goods in

order to be no worse off than at (xi, z*), when the environment changes to
z and prices are p •. Now let
By construction, these (Jj,Ti constitute a tax system, with (Jj(p·,z·)

Ti(p·, z·) = 0, for all j, i. In particular, part (iv) of the definition is satisfied.
Finally, take a consumer i and a consumption plan (Xi, z) E Xi such that
Ui(Xi,Z) ~ ui(xi ,z·). Then, p·Xi-Ti(p· ,z) ~ p·xi+p·xi-Ei(Z) ~ p·xi,
which implies that: 2
n
P·Xi ~ p·Wi + ~O:j7rj(p·,z·) +Ti(p·,Z).
j=1
Thus part (i') of the definition of compensated valuation equilibrium is also

satisfied and the proof is complete. •
Theorem 2 tells us that any efficient allocation can be decentralized as
a compensated valuation equilibrium. It also tells us that a compensated
valuation equilibrium with tmnsfers exists, provided that there is at least
one Pareto efficient allocation.
Observe that the tax system has been given an explicit form, which has
an easy and sensible interpretation. Each (Jj(p., z) tells us the tax paid by
the jth firm if the environment changes from z* to z. It is equal to the firm's
change in profits. Similarly, every Ti(P*, z) specifies the net amount that
the ith consumer will have to pay, if the environment is changed from z·
to z. It is equal to the equivalent variation in the sense of Hicks. Moreover,
it follows that (Jj (p., z·) = Ti(P·, z·) = 0, for all i,j.
Remark 2 A standard argument shows that the compensated valuation

equilibrium of Theorem 2 is also a valuation equilibrium provided that,
whenever (Xi, z) E Xi satisfies Ui(Xi, z) > Ui(xi, z·), there exists a cheaper
point X-i E Xi(Z) for which p.X-i < Ei(Z). This can be ensured by means
of the "essentiality condition" that appears in Mas-Colell (1980, p. 626),
Mas-Colell & Silvestre (1989, p. 250), or Diamantams & Gilles (1996, p.
855).
Our last result refers to the core, defined as follows:
2The scalars eij correspond here to a distribution of profits determined by the sepa-
ration argument.
209
Definition 5 A feasible allocation [(Xi), (Yj),Z] is in the core if there is

no coalition S c M = {I, 2, ... , m}, with an allocation [(xD, (Yj), z']' such
that:
(i) LiEs (x~ - Wi - L7=1 ()ijYj) = O.
(ii) ui(xLz') ~ Ui(Xi,Z), ViE S, with a strict inequality for some agent
in S.
A core allocation is thus one in which no coalition can re-arrange the
economy, using its own resources, so that the resulting allocation is pre-
ferred by all its members. Observe that profit shares are being interpreted
here as production shares (as usual when positive profits are possible).
The following definition helps characterize the tax systems which yield
valuation equilibria in the core.
Definition 6 Let Q* = [p', (xi), (YJ), z', T*] be a valuation equilibrium.

Say that T* satisfies the no-transfer condition relative to Q* if, for any
coalition S C M and each z E Z, one has
The no-transfer condition relative to a valuation equilibrium Q' tells us

that the restriction of T* over any coalition corresponds to a balanced tax
system. To understand better the extent of this requirement it is worth
thinking of the set Z as containing all non-empty subsets of M, so that the
tax system depends also on the coalition structure. Note that when there
are neither taxes nor subsidies this condition is automatically satisfied.
Another case in which this requirement holds is when the tax system is
given by Ti(P,Z) = - L7=1 ()ijOj(P,z) for all i (each individual i's net tax
is equal to i's share of the subsidies paid to the producers whose shares i
holds).
The following result is obtained:
Theorem 3 Let Q' = [p*,(xt),(YJ),z',T'] be a valuation equilibrium.

Suppose that consumers are locally non-satiated and that T' satisfies the
no-transfer condition relative to Q*. Then, the resulting allocation is in the
core.
Proof. For any S CM, consider any alternative allocation [( xD, (Yj), z']
satisfying (i) and (ii) of Definition 5. From the definition of valuation equi-
librium and the fact that preferences are locally non-satiated, it follows
that
n
p*x~ ~ P'Wi + L()ij[P*Yj +OJ(p',z*)] + Ti(P',Z') [1]
j=1
210
for all i E S. Summing over S and making use of the non-transfer condition
one gets
n n
> ~)P*Wi+ I:Oijp*Yj+ I : OijO'j(P*, z*) + Ti(P*,Z')]
iES j=1 j=1
n n
> 2:[P*Wi + 2: OijP*yj + 2: OijO'j(p', z') + Ti(P*, z')]
iES j=1 j=1
n
2:[P*Wi + 2: 0ijp*yj].
iES j=1
From (i) above it follows that both weak inequalities are equalities. Hence
[1] also holds with equality for all i E S. Because Q* is a valuation equi-
librium, ui(xi, z*) 2: Ui(X~, z') for all i E S. So S cannot be a blocking
coalition. Hence, the resulting allocation is in the core. •
4 Some examples
This section presents some specific models that can be interpreted as par-
ticular cases of the setting in sections 2 and 3. This illustrates how our
model is flexible enough to encompass several different situations. Interest-
ingly enough the existence of equilibrium is also guaranteed in all of these
particular cases.
In order to facilitate the discussion, let us denote by ro
the (degenerate)
balanced tax system given by O'j(p, z) = Ti(p, z) = 0, for all i,j, every
(p, z) in ~~ x Z. Consider the following assumption:
Assumption 3 For all z E Z one has
(i) Xi(Z) is bounded from below;
(ii) Wi E intXi(z) (i = 1,2, ... , m);

(iii) [y E Y(z) = 2:7=1 Yj(z), and Y -# 0] ==* -Y rf- Y(z).
This requires every consumer have bounded conditional consumption sets
with the endowment vector in the intersection of their interiors, and the
aggregate production set satisfy the irreversibility hypothesis.
4.1 Competitive equilibrium

The simplest instance of a valuation equilibrium is the standard competitive
case. To see this take Z to be a singleton (e.g., z describes the assignment
of property rights of a private ownership economy), let Yj, Xi be fixed
convex sets for all i,j, and let T = ra. The resulting economy is the
standard private ownership model of Arrow and Debreu (1954), and every
211
private competitive equilibrium corresponds to a valuation equilibrium.

Hence, under Assumptions 1, 2 and 3 these equilibria exist, are in the
core, and also every Pareto optimal allocation can be decentralized as a
competitive equilibrium.
A variant of this model is that in which externalities are allowed [e.g.,
Debreu (1952)]. It is well known that under standard conditions a competi-
tive equilibrium exists, though it may fail to be Pareto optimal. Optimality
can however be ensured if markets are combined with a suitable tax system
(or a system of personalized. prices). If one takes Z as the set of allocations
and makes the tax system equivalent to a Lindahl price system, a valu-
ation equilibrium corresponds to an efficient market equilibrium. In this
case our equilibrium notion corresponds to what Bergstrom (1970) calls a
distributive Lindahl equilibrium.
Another case worth mentioning is that of an economy in which com-
modities can be of different qualities [see for instance Dreze and Ragen
(1978)]. Members of the set Z can be thought of as different quality stan-
dards. Changes in these standards will typically affect consumers' feasible
sets and utilities, as well as firms' production possibilities. For an arbitrary
z E Z, and T = TJ, a private competitive equilibrium is well defined, and
exists under Assumptions 1,2 and 3. Yet it may be inefficient (e.g., some
consumers may be willing to pay for higher quality). The results in sec-
tion 3 show that any efficient allocation can be decentralized as a valuation
equilibrium. 3
4.2 Economies with public goods

Consider now an economy with f private commodities and k public goods.
There are n - 1 private firms and a publicly owned technology (the nth
firm) that produces the public goods using private goods as inputs. Let
Z = (Yn n Ri) X Rim, where Yn stands for the public firm, Ri for the
space of public goods, and Rim for the space of personalized. prices. Hence,
a point z = (s, q) E Z describes a vector of public goods s E Ri and a
vector of personalized prices q = (qt, ... , qm), with qi E Ri for all i. There
is a mapping c : R~ x Ri _Ri such that, for each given pair (p, s) in
R~ x Ri the private goods input vector c minimizes the cost of producing
s. Assume also that this is a well defined single-valued vector mapping, to
make things simpler.
The following balanced tax system can now be defined: For i = 1,2, ... , rn,
let Ti(P, z) = - E~=l q~8r, where 8r is the amount of the rth public good
supplied, and q~ the ith consumer's personalized. price for the rth public
good. And let lTn(p,z) = - E:'l Ti(p,Z) and lTj(p,z) = 0, for all j =f. n. A
Lindahl equilibrium for this economy (which exists under the assumptions
3See Corch6n (1994) for an analysis of the dual case in which prices are taken as the
public environment, and qualities take the role of balancing the markets.
212
of the model) is a valuation equilibrium Q* = [p*, (xi), (Yj), s*, q*, T*]
Wl·th L."i=l ~m Wi + ~n
~m xi* -_ L."i=l L."j=l Yj* - c (*
P ,s *) and - L."i=l
~m Ti (*
P ,z*)_ -
p*c(p*, s*).
4.3 Equilibrium with increasing returns to scale

Following the work of Scarf (1986), consider now a market economy with
a single firm having production set Y c ~e. Assume that there is a group
of commodities that can be identified a priori as pure inputs. The set of
commodity indices I = {I, 2, ... , e} is partitioned into two disjoint subsets,
IK = {I, 2, ... , k}, for some integer k < £, and IS (its complement). We
shall refer to goods in IK as capital goods. Commodities in IS are standard
commodities (i.e., they can be consumption goods, inputs and outputs). It is
useful to write production plans in the form: Y = (a, b), where a E - ~i is a
vector of capital goods, and b E ~e-k is a vector of standard commodities.
Consider now:
Assumption 4 For every a' E - ~i the set
B(a') == {b E ~e-kj I (a, b) E Y for some a 2: a'}

is convex.
This weakens the classical assumption of convex production sets. It says
that for any given vector of capital goods a', the projection on ~e-k of
those production plans not using more capital goods than those in a' is
a convex set. This allows us to interpret these special inputs as types of
fixed capital that give rise to non-convexities. Observe that this assump-
tion is compatible with the presence of firms with convex production sets,
increasing returns to scale, set-up costs or S-shaped production functions.
Let wK E ~i denote the initial endowments of capital goods, and define
Z as follows:
Z = {a E - ~i / a 2 _w K }.
The set Z describes the feasible allocations of the available capital goods.
Define Y(a') as the constrained production set, namely,
Y(a') = {y = (a, b) E Y with a 2 a'}.
Given a price vector p E~~ and a vector of capital goods a', the firm
chooses a profit maxilnizing production plan y* within its attainable set
-that is, it satisfies Py* 2 py, for all yE Y(a').
It can be shown [see Villar (1997)] that under Assumptions 1, 2, (i)
of 4 and 5 an equilibrium exists for this economy. Yet this equilibrium
may fail to be Pareto optimal. Indeed, Scarf (1986) shows that, under
general assumptions, one can always find economies with empty cores unless
all production sets are convex cones. A key element in that result is the
possibility that all commodities are consumed. This justifies the following
assumption:
213
Assumption 5 Capital goods are pure inputs, so they do not enter con-
sumers' preferences or feasible sets, and also a ?: a' implies B(a') C B(a) ..
Assumption 5 implies that the only z* which is a candidate for a valuation

equilibrium is that corresponding to a = -wK. Hence, making use of the
existence result mentioned above we obtain:
Theorem 4 Let E be an economy satisfying Assumptions 1, 2, (i) of 3, 4
and 5. Then there exists a valuation equilibrium with T = TO.
This equilibrium consists of a price vector and a feasible allocation in
which all agents are maximizing their payoff functions within their fea-
sible sets. These feasible sets correspond to budget sets, for the case of
consumers, and the production set subject to an input constraint, for the
firm.
Remark 3 Variants of this model can accommodate public goods, as shown
in Gines (1996). A generalized Lindahl equilibrium (which exists under
suitable assumptions) corresponds to a valuation equilibrium. See also Vega-
Redondo (1987) for a model with both externalities and non-convexities.
References
Arrow, K.J. & Debreu, G. (1954), Existence of Equilibrium for a Compet-
itive Economy, Econometrica, 22 : 265-290.
Bergstrom, T.C. (1970), A "Scandinavian Consensus" Solution for Effi-
cient Income Distribution among Nonmalevolent Consumers, Journal of
Economic Theory, 2 : 383-398.
CorcMn, L. (1994), Fixed Prices and QUality Signals, Mathematical Social
Sciences, 27 : 49-58.
Debreu, G. (1952), A Social Equilibrium Existence Theorem, Proceedings
of the National Academy of Sciences, 38 : 886-893.
Diamantaras, D. & Gilles. R.P. (1996), The Pure Theory of Public Goods:
Efficiency, Decentralization and the Core, International Economic Re-
view, 37 : 851-860.
Diamantaras, D. & Wilkie, S. (1994), A Generalization of Kaneko's Ratio
Equilibrium to Economies with Private and Public Goods, Journal of
Dreze, J.H. & Hagen, K. (1978), Choice of Product Quality: Equilibrium
and Efficiency, Econometrica, 46 : 493-513.
Gines, M. (1996), Core Selections in Economies with Increasing Returns
and Public Goods, mimeo, University of Alicante.
Hammond, P. & Villar, A. (1998), Efficiency with Nonconvexities: Extend-
ing the "Scandinavian Consensus" Approaches, The Scandinavian Jour-
nal of Economics, vol. 100, forthcoming.
214
Johansen, L. (1963), Some Notes on the Lindahl Theory of Determination

of Public Expenditures, International Economic Review, 4 : 346-358.
Kaneko, M. (1977), The Ratio Equilibrium and a Voting Game in a Public
Goods Economy, Journal of Economic Theory, 16 : 123-136.
Mas-Colell, A. (1980), Efficiency and Decentralization in the Pure Theory
of Public Goods, Quarterly Journal of Economics, 94 : 625-64l.
Mas-Colell, A. & Silvestre, J. (1989), Cost-Share Equilibria: A Lindahlian
Approach, Journal of Economic Theory, 47 : 239-256.
Scarf, H. (1986), Notes on the Core of a Productive Economy,. Ch. 21 in
W. Hildenbrand & A. Mas-Colell (Eds.) , Contributions to Mathemati-
cal Economics. In Honor of G~rard Debreu, North-Holland, Amsterdam,
1986.
Vega-Redondo, F. (1987), Efficiency and Nonlinear Pricing in Nonconvex
Environments with Externalities: A Generalization of the Lindahl Equi-
librium Concept, Journal of Economic Theory, 41 : 54-67.
Villar, A. (1997), Competitive Pricing in Economies with Nonconvex Pro-
duction Sets, mimeo, University of Alicante.
The equal-distance rule in allocation
problems with single-peaked preferences'
Carmen Herrero 1 and Antonio Villar 1
1 Department of Economics, University of Alicante, Apartadode Correos, 99,
03080 Alicante, Spain and IVIE
Abstract. We consider allocation problems with single-peaked preferences,

and focus on the equal-distance rule (Thomson, 1994). An axiomatic char-
acterization of the equal-distance rule is provided.
JEL Classification Numbers: C71, D61
Keywords: Allocation problems, single peaked preferences, equal-distance

rule
1 Introduction
Consider an allocation problem in which a group of agents has to perform
a common task that requires a given number of hours labour time, that
are paid at a given rate. Agents' supply of labour results from maximizing
their utility functions, that depends on income and leisure. Under standard
assumptions, each individual has an ideal amount of work to be performed,
called her peak, and having to work more or less decreases her utility. The
problem under consideration is that of devising a procedure to allocate
the task among the agents, in the understanding that when the aggregate
supply of labour differs from the amount of labour time required, agents
must be rationed.
This is an example of a family of problems consisting of the allocation of a
fixed amount of a divisible good among a group of agents whose preferences
are single-peaked [see Sprumont (1991), Ching (1994), Thomson (1994 a,b),
(1995), de Frutos and Mass6 (1995), Barbera, Jackson and Neme (1997),
Otten et al (1996), Dagan (1996), Herrero & Villar (1998a,b)]. The uni-
form rule is the best known procedure to solve these problems. This rule
recommends sharing the duties equally, as long as this is compatible with
'Thanks are due to William Thomson for his comments and suggestions. Finan-
cial support from the Direcci6n General de Investigaci6n Cientifica y Tecnica, under
project PB97-0120, and from the European Comission, FMRX-CT96-0055 are grate-
fully acknowledged.
216
efficiency and feasibility. This principle also appears in other contexts, as it

it the case for the constrained equal-awards rule in bankruptcy, in axiomatic
bargaining, and in the fiat tax in taxation.
The equal-distance rule (Thomson, 1994b) is an alternative procedure.
The main idea underlying the equal-distance rule is that the difference be-
tween the effort required and the effort supplied is a common responsibility.
Hence the rule proposes to divide this difference as equally as possible, as
long as this is compatible with efficiency and feasibility. The principle be-
hind the equal-distance rule has been also applied in other contexts as it
is the case of the constrained equal-losses rule in bankruptcy, in axiomatic
bargaining, and in the leveling tax in taxation.
An alternative allocation procedure, called the even rule, was introduced
in Herrero and Villar (1998a). This rule coincides with the equal-distance
rule when the aggregate supply of effort is smaller than the task, and with
the uniform rule otherwise.
These rules have in common a number of well-known properties, such as
efficiency, equal treatment of equals and consistency. And differ with re-
spect to some others, as strategy proofness and agenda-independence. Strat-
egy proofness says that agents have no incentives to misrepresent their
preferences. Agenda-independence requires the outcome to be the same
whether the task is allocated one shot or sequentially. The uniform rule
satisfies strategy proofness, whereas neither the equal-distance rule nor the
even rule exhibit this property. Both the equal-distance rule and the even
rule satisfy agenda-independence, but this is not the case for the uniform
rule.
The main difference between the equal-distance rule and the even rule
refers to the way in which they recommend to solve an allocation prob-
lem in which agents peaks are very asymmetric, in the case the task is
smaller than the aggregate supply of effort. They take opposite views in
this respect: whereas the even rule gives priority to the agents with small
peaks, the equal distance rule gives relatively more to those agents with
larger peaks. Indeed these properties characterize the two rules, together
with those already mentioned.
The even rule satisfies truncation, a property stating that all information
on agents' preferences beyond the task to be performed is irrelevant. As a
consequence, all agents with peaks above the task are considered indistin-
guishable. Herrero & Villar (1998a) show that the even rule is the only rule
that satisfies efficiency, equal-treatment of equals, agenda-independence
and truncation.
We shall show here that the equal-distance rule satisfies ar-truncation.
This property says that all information on preferences below the average ra-
tioning is irrelevant. As a consequence, agents with peaks below the average
rationing are treated as if they had their peaks at zero. The equal-distance
rule is the only rule that satisfies efficiency, equal treatment of equals, con-
sistency, agenda-independence and ar-truncation.
217
These 'dual' characterizations illuminate on the way in which these two

rules recommend to treat agents: whereas the even rule gives priority to
agents with small peaks, the equal-distance rule considers agents with peaks
small enough as negligible. In this way we have a good hint on which rule
to use depending on the type of problem at hand.
The paper is organized as follows. In Section 2 we present the formal
model and the definitions of the equal-distance, the uniform and the even
rules. In Section 3 we present several properties for rules. In Section 4 we
provide a characterization of the equal-distance rule.
2 The model
We follow Thomson (1995). Let N be an infinite population of potential
agents. For an agent i E N, let ~ be a preference relation over ~+, and
let ~ the associated strict preference. The preference relation is single-
peaked if there exists a number, p( Ri) E ~+, such that for all x, x' E ~+,
[(x' < x ~ p(Ri)) or (p(Ri) ~ x < x')] ===> x Pi x'.
Let § denote the family of single-peaked preference relations over ~+. The
preferences of all agents in N are drawn from §. Let N be the class of all
(non-empty) finite subsets of N. Given N EN, let §N denote the Cartesian
product of 1Nl copies of §, indexed by the members of N. Similarly, ~~
stands for the Cartesian product of 1Nl copies of ~+.
Given N EN, let R = (~)iEN E §N represent a profile of preferences for
N, and T E ~+ the total amount of time required to perform some task. A
problem is a pair e = (R, T). Let lEN be the family of all problems for N,
and let lE = UN lEN. A feasible allocation for e = (R, T) E lEN is a point
x E ~~ such that LXi = T. Let X (e) be the set of feasible allocations of
e.
A rule is a mapping F : lE ~ U~N, such that for all e = (R, T) E lE N ,
F(e) is a feasible allocation for e. The intended interpretation of F(e) is
that it is a desirable way of distributing T.
Next, we define three rules. All of them may be thought of as implement-

ing the idea of equal division subject to efficiency and feasibility constraints.
Uniform rule U : For all NE N, all e = (R, T) E lE N, and all i E N,
min{p(Ri) , A} if T ~ LP(~),
Ui (e) = {
max{p(Ri), A} otherwise.
where A is chosen so that U(e) E X(e).
Equal-distance rule D : For all N E N, all e = (R, T) E lE N, and all
i EN,
Di(e) = max{O, p(~) + A},
218
where>. is chosen so that D(e) E X(e).
The equal-distance rule selects that point in the feasible set that is closest
(according to the Euclidean distance) to the vector of preferred contribu-
tions (p(Ri))N. Note that if LP(~) :::; T, >. > 0, and otherwise>. < O. The
equal-distance rule makes the amounts of time as equal as possible from
their peaks, subject to the condition that all are assigned a nonnegative
share of the required effort.
Even rule E: For all N EN, all e = (R,T) E EN, and all i E N,
Ei(e) = { min{p(~), >'}

p(~) + ~ [T - LP(~)] otherwise.
where>. is chosen so that E(e) E X(e).
The even rule coincides with the uniform rule when T :::; LP(Ri) and
with the equal-distance rule otherwise. Hence it allows differences in the
rationing experienced by the agents when the required effort is smaller
than the overall supply, whereas it divides the undesired aggregate effort
equally.
3 Properties
Next, we consider several properties a rule may fulfill. The first property
is a way of formulating the idea of impartiality.
Equal treatment of equals: For all N EN, all e = (R, T) E EN, and all
= R j , then Fi(e) = Fj(e).
i,j E N, if Ri
Given e = (R, T) E EN, an allocation x E X (e) is efficient if there is no

Y E X(e) such that for all i E N, Yi ~ Fi(e), and for some j E N, Yj Pj
Fj (e). It is easy to prove that an allocation in X (e) is efficient if and only
if each agent consumes no more than her preferred bundle if T:::; LP(~)
and no less otherwise (Sprumont, 1991).The next property states that a
rule should select efficient allocations.
Efficiency: For all N E N and all e = (R, T) E EN, F(e) is efficient in

X(e).
Given a problem e = (R, T) E EN, assume that, after solving e by means

of a certain rule F, some agents leave with the amounts of time assigned
to them. Let Q C N be the set of the remaining agents, and LQ Fi (e) the
overall amount of time assigned to them. Let e~ = [R, LQ Fi(e)] E EQ
219
denote the reduction of e to Q, relative to F. A rule is consistent if all

agents in Q are asigned the same amount in the reduction of e to Q, relative
to F than they were assigned in the initial problem. Formally,
Consistency: For all N E N, all e = (R, T) E JE,N, all Q c N, and for all
i E Q, we have: Fi(e) = Fi(e~).
Before presenting the next property, let us introduce an operator on the

set §. Given a E ~+, let aa : § -+ § be defined as follows: For all R E §,
xaa(R)y iff (x + a) R (y + a). Given R E §, aa(R) is the shifting of ~
bya.
Consider now a group of agents N, with preferences R E §N that will face
a certain problem, and let us solve it by forecasting the amount of time
required to perform the task. Once this is done, let us suppose that the
actual amount of time required is greater than expected. Then two options
are open: either the previous distribution is cancelled altogether and the
actual problem is solved, or the rule is applied to the incremental time
required, after adjusting the preferences by shifting them by the amounts
initially assigned. Agenda-independence requires the recommendation
given by the rule to be independent of the chosen option. Formally,
Agenda-independence: For all N E N, all e = (R, T) E JE,N, all i E N,

and all T l , T2 E ~+ such that Tl +T2 = T, if el = (R, T l ), R~ = aPi(e1)(Ri),
and e2 = (R', T2), then F(e) = F(el) + F(e2)'
Agenda-independence says that the final allocation is the same no matter

how the global task is subdivided. This is a relevant property for those
problems in which the global amount of time required T is known a priori,
and a decision has to be taken about the scheduling (namely, whether to
distribute it all at once or in parts). When a rule is agenda-independent the
scheduling is immaterial with respect to agents' final contributions. Hence,
agenda-independence may be regarded as a procedural requirement.
The last property can be regarded as an instance of the general prin-
ciple of independence of irrelevant alternatives. 1 Given e = (R, T) E JE,N,
let ar(e) = E P1~il)-T .The number ar(e) is simply the average rationing
experienced by the agents involved in e. Note that ar(e) > 0 if and only if
L,P(Ri) > T. The property of ar-truncation states that any information
on the agents' preferences below ar(e) should be ignored. In consequence,
all those problems whose preferences coincide on [ar(e) , +00) are indistin-
guishable. Formally,
1 A different interpretation of that principle appears in Herrero and Villar (1998a),

with the property of truncation: For all N E N, all T E li+, and all R,R' E §N, if for
all i E N, Ri and R~ coincide on [0, T), then F(R, T) = F(R', T).
220
Ar-truncation: For all N E N, all T E lR+, and all R, R' E §N, if for all
i E N, Ri and R~ coincide on [ar{e),+oo), then, F{R,T) = F{R',T).
4 Characterization
Consider the following lemmata:
Lemma 1 (Herrero and Villar, 1998a). Let F be a rule satisfying ef-

ficiency, equal treatment of equals and agenda-independence. Then, for
all N E N, all (R, T) E EN such that LP{Ri ) ~ T, and all i E N,
Fi{R, T) = p{~) + ~[T - LP{Rj)].
Lemma 2 (Herrero and Villar, 1998a). Let F be a rule satisfying agenda-

independence. Then, for all N E N, all R E §N, and all i EN, Fi{R,.) is
continuous with respect to its second argument.
Lemma 3 Let F be a rule satisfying efficiency and ar-truncation. Then,

for all NE N, all e = (R, T) E EN, and all i EN such that P(Ri) < ar(e),
we have Fi(e) = o.
Proof. Given e = (R, T) E EN, let S(e) = {i EN: P{Ri) < ar(e)}.
Suppose that ar(e) > 0 (otherwise the result is obviously true). Thus,
LP{~) > T.
Let e' = (R',T) E EN, where for all i E N\S(e), R: = ~; for all
i E S(e), p(RD = O. Then, for all i E N, and for all x, Y E [ar(e), +00),
we have xRiy ~ xR~y. By efficiency, for all i E S(e), Fi(e') = 0, and by
ar-truncation, for all i E S{e), Fi{e) = o.•
Lemma 4 If a rule satisfies efficiency, equal treatment of equals, consis-
tency, agenda-independence, and ar-truncation. then for all N EN, all
(R,T) E EN, and alli,j EN, ifp{Ri) =p{Rj), then Fi{R,T) = Fj{R,T).
Proof. Let us prove the result by way of contradiction. Then, there exist
N E N, e = (R, T) E EN, and i,j E N, with P{Ri) = p{Rj) = p such
that Fi(R, T) =f Fj (R, T). First, note that T < LP(~). Without loss of
generality, assume that a = Fi(R, T) < Fj(R, T) = b, and let Tl = a + b.
Let e~,j} = (R, T 1 ) E E{i,j}. By consistency, Fi(e~,j}) = a p-b = p(Rj).
Let 8 > 0 such that p - a - 8> p - b. Let (R,Tl + 8), (R,T1 ), (R',8) E
E{i,j}. By agenda-independence, F(R, Tl + 8) = F(R, T 1 ) + F(R', 8). Let
e' = (R',8) E E{i,j}, and note that p(Rj) = p - b < ar(e'). By Lemma
3, Fj(e') = 0, and thus Fi(e') = 8. Consequently, for all 8 < b - a, and
all (R, Tl + 8) E E{i,j}, Fi(R, Tl + 8) = a + 8 and Fj (R, Tl + 8) = b. By
Lemma 2, and for 8 = b - a, Fj(R,2b) = b. Thus, for all 0 < € ~ b - a,
Fj(R, 2b - €) = b.
221
Suppose now that for some T2 < T l , Fj(R,T2) = d < b. By Lemma 2,

we may choose T2 such that b- d < ~. Reasoning as before, we obtain that
Fj(R,2d) = d. But, since 2b - 2d < c, it follows that Fj(R,2d) = b > d,
contradicting the hypothesis. Consequently, for all T ~ T l , Fj (R, T) = b.
But, this cannot be true for T < b. •
Theorem 5 The equal distance rule is the only rule satisfying efficiency,
equal treatment of equals, consistency, agenda-independence, and ar-trun-
cation.
Proof. The equal-distance rule trivially satisfies all the properties.

Let us see the converse implication. By Lemma 1, we only need to discuss
problems e = (R,T) E EN such that LP(Ri ) > T.
Let SI (e) the set of agents with the smallest peaks, S2 (e) the set of agents
with the second smallest peaks, ... , Sk(e) the set of agents with the highest
peaks. Let PI = p(Ri) for i E Sl(e), P2 = p(Ri) for i E S2(e), etc. Similarly,
let SI denote the cardinality of SI (e), S2 the cardinality of S2 (e) and so
forth.
Case 1. 0 ~ T ~ Sk(Pk-Pk-l)' Let Tl = LP(Ri)-npl, T2 = (S2P2+ .... +

SkPk) - (S2 + ... + Sk)P2, ... , Tk-l = Sk-lPk-l + SkPk - (Sk-l + Sk)Pk-l =
Sk(Pk - Pk-l) = T. Note that for all j = 1"" k - 1, we have that T < Tj.
Since T < Tl it follows that PI < ar(e). Thus, by Lemma 3, for all
i E Sl(e), Fi(e) = O. Let Ml = N\Sl(e), and let e~l' By consistency, for
all i E M l , we have Fi(e) = Fi(e~J.
Since T < T2 it follows that P2 < ar(e~J. Thus, by Lemma 3, for all
i E S2(e), Fi(e) = Fi(e~J = O. Let M2 = Ml \S2(e), and let e~2' By
consistency, for all i E M2, we have Fi(e) = Fi(e~2)'
After k - 2 steps, we have Mk-2 = M k - 3 \Sk-2 = Sk-l U Sk. Note that
whenever T < Sk(Pk - Pk-l), we also have Pk-l < ar(e~k_2)' and thus for
all i E Sk-l(e), we have Fi(e) = Fi(e~k_2) = O.
By Lemma 2, for all i E Sk-l (e), and T = Sk(Pk - Pk-l), we obtain that
Fi(e) = Fi(e~k_2) = O. Consequently, for all i E N\Sk(e), Fi(e) = O.
By consistency, peaks only, and equal treatment of equals, Fi(e)
Fi(e~k(e)) = ~. Therefore, F(e) = D(e).
Case 2. Sk(Pk - Pk-l) < T ~ SkPk + Sk-lPk-l - (Sk + Sk-l)Pk-2.

Let Tl = Sk(Pk - Pk-t} and el = (R, Tt}. For all i E N, let R~
aFi(ed(~) and e2 = (R', T-Tl)' By agenda-independence, F(e) = F(el~+
F(e2), and by Case 1, F(el) = D(el), namely, for all i E Sk(e) , Fi(el) = ;;,
and otherwise, Fi(el) = O. Consequently, for all i E Sk(e), p(RD = Pk -
1J..
Bk
= Pk-b and otherwise, p(RD = p(~).
Now, note that 0 ~ T-Tl ~ (Sk+Sk-l)(Pk-l -Pk-2). Thus, the problem
e2 = (R', T - T l ) is covered by Case 1. Thus, for all i E Sk(e) U Sk-l(e),
Fi(e2) = BkT;Tl
Sk-l
and otherwise, F i (e2) = O.
222
'
Consequently, for all i E Sk(e), Fi(e) = ~ + 8:+-8;:~1 for all i E Sk-l(e),
Fi(e) = BkT+S;:~l and otherwise, Fi(e) = 0, and thus, F(e) = D(e).
We can follow this procedure until all tasks smaller than or equal to
LP(Ri) are covered . •
The properties in Theorem 5 are independent, as the following examples

show. In each case we mention the property that is not satisfied.
(1) Equal treatment of equals. Select a particular agent i E N. For all
NE N, all e = (R, T), and all j E N, let F be the rule
Dj(R,T) if LP(Rk) ~T
Fj(R,T)= { P(Ri) ifj=i, LP(Rk)<T
Dj(R,T-p(~)) ifj#i, LP(Rk)<T.
(2) Efficiency. For all N EN, all e = (R, T), and all j E N, let H(e) =
{i EN: p(~) = maxN p(Rk)} and h = IH(e)l. Let F be the rule defined
by
( )= { t
Fj R,T
if j E H(e),
o otherwise.
(3) Consistency: Select a particular agent i E N. Let F be the following
rule: If i E N, 1Nl = 3, maxN p(Rt) > p(~) > minN p(Rt), LP(Rt} > T,
and T ~ p(~) - minNP(Rt ), then Fi(R,T) = T, and for all t E N\{i},
Ft(R,T) = O. F(e) = D(e) in any other case.
(4) Agenda-Independence: For all N EN, all e = (R, T), and all j EN, let
F be the rule
Uj (R, T) if T > LP(~),
Fj(RN,T) = {
Dj (R,T) ifT:SLP(~).
(5) ar-truncation: The even rule.
Remark 1 In dealing with the relationship between the equal-distance rule
and the even rule, see Herrero and Villar (1998a, Theorem 1), where
the even rule is characterized as the only rule satisfying efficiency, equal
treatment of equals, agenda-independence, and truncation. An alternative
characterization of the equal-distance rule appears in Herrero and Villar
(1998b).
References
Barbera, S., Jackson, M. and Neme, A. (1997), Strategy Proof Allotment
Rules, Games and Economics Behavior 18 : 1-21.
Ching, S. (1994), An Alternative Characterization of the Uniform Rule,
Social Choice and Welfare, 11 : 131-136.
Dagan, N. (1996), A Note on Thomson's Characterizations of the Uniform
Rule, Journal of Economic Theory, 69 : 255-261.
223
de Frutos, M.A. and Mass6, J. (1995), More on the Uniform Rule: Equality
and Consistency, mimeo, Universitat Autonoma de Barcelona.
Herrero, C. and Villar, A. (1998a), Agenda Independence in Allocation
Problems with Single-Peaked Preferences, Social Choice and Welfare,
forthcoming.
Herrero, C. and Villar, A. (1998b), An Alternative Characterization of the
Equal-Distance Rule in Allocation Problems with Single-Peaked Prefer-
ences, mimeo, Universidad de Alicante.
Otten, G.J., Peters, H. and Volij, O. (1996), Two Characterizations of
the Uniform Rule for Division Problems with Single-Peaked Preferences,
Sprumont, Y. (1991), The Division Problem with Single-Peaked Prefer-
ences: A Characterization of the Uniform Allocation Rule, Econometrica,
59 : 509-519.
Thomson, W. (1994a), Consistent Solutions to the Problem of Fair Division
when Preferences are Single-Peaked, Journal of Economic Theory, 63 :
219-245.
Thomson, W. (1994b), Resource-Monotonic Solutions to the Problem of
Fair Division when Preferences are Single-Peaked, Social Choice and Wel-
fare, 11 : 205-223.
Thomson, W. (1995), Population Monotonic Solutions to the Problem of
Fair Division when Preferences are Single-Peaked, Economic Theory, 5 :
229-246.
The existence of the satisfactory point *
Adam Idzik
Institute of Computer Science, Polish Academy of Sciences, ul. Ordona 21,

01-237 Warsaw, Poland
Abstract. An extension of Woodall's theorem on the satisfactory point to

arbitrary continuous functions, as well as a short proof of it, is presented.
An extension of Kulpa's theorem is also made. As application, proofs of the
existence of the core payoff in the matching game studied by Alkan and
Gale (1990), for arbitrary continuous functions (not only decreasing) and
Bapat's permutation-based generalization of Brouwer's fixed point theo-
rem, are given.
JEL Classification Numbers: D3, D4, D5, D6
Keywords: Covering of a simplex, satisfactory point, matching game
1 Introduction
Woodall (1980) proved the existence of the satisfactory point and then ap-
plied this result (Theorem 4) to show the existence of the fair division of
a cake. Also Kulpa (1994) proved the existence of a stable-like point and
applied his theorem (Lemma 1) to show the existence of rational divisions
of bounded Lebesgue measurable sets in Euclidean spaces. Some general-
izations of Woodall's results and Kulpa's results were presented by Idzik
(1995) and Ichiishi and Idzik (1996).
In this paper we apply some ideas of Ky Fan (1972, Theorem 3) derived
from his fundamental theorem on coincidence (Theorems 2.0 - 2.2) and
extend aforementioned theorems of Woodall and Kulpa to arbitrary conti-
nous functions and present very short proofs of them (original Woodall's
proof is four pages long!).
Next we apply our theorems to show the existence of the core payoff in the
matching game studied by Alkan and Gale (1990) for arbitrary continuous
functions (not only decreasing). We also derive, as a corollary, Bapat's
permutation-based generalization (1989) of Brouwer's fixed point theorem.
'The research reported in this paper is supported by the KBN Grant No. 2 P03A
017 11.
226
2 Extensions of Woodall's theorem and Kulpa's

theorem
Let N be a nonempty finite set. The cardinality of set N is denoted by #N.
Nand R stand for natural numbers and reals, respectively. Denote by RN
the (#N)-dimensional Euclidean space and by R~ the nonnegative orthant
of RN . Given a subset X of RN ,let co X denote the convex hull of X, int X
denote the interior of X, P(X) denote the family of all nonempty subsets
of X, aft' X denote the affine hull of X, ri X denote the relative interior
of X and oX denote the relative boundary of X. Let f : X -+ P(R N ) be
a function. We say that f is upper semicontinuous (u.s. c.) on X if the set
{x E X I f(x) C V} is open in X whenever V is an open subset of RN.
Functions f, g : X -+ P(R N) have a coincidence if there exists x E X such
that f(x) n g(x) =I- 0. The unit vectors of RN are denoted by ej, j E Nj
here e~ = 1 and e; = 0 for all i E N \ {j}. The unit simplex is the set
f).N := co {ej I j E N} and its faces are f).s := co {ej I j E S}, SeN.
The Euclidean inner product of two vectors x and y in RN is denoted
by X· y. We recall that a hyperplane H in RN is a set of the form H =
{x E RN I p' x = t}, where p ERN, P =I- 0, and t is a real number.
Given a compact convex set X in RN, a proper subset F of X is called
a proper face of X, if there exist p E RN \ {O} and t E R such that
F = X n {x E RN I p . x = t}, and p . x > t for every x E X \ F. In
this case, the hyperplane H = {x E RN I p. x = t} is called a supporting
hyperplane of X. A face G of a compact convex set X is an opposite face to
face F of X, if G = X n HI and F = X n H2 for some parallel supporting
hyperplanes HI and H2 of X.
Let us recall a corollary from the fundamental Ky Fan theorem (Theorem
2.0) and an extension of Gale's (1984) covering lemma and its dual version
(Theorem 2.1 and Theorem 2.2, respectively).
Now, let n:= #N, and define ms:= LjESej/(#S) for each SE P(N).
Choose a set K in RN such that
{ej I j E N} eKe aft' f). N, #K < 00;
the set K will be fixed throughout this section. A point v E aft' f). N is
uniquely expressed as an affine combination of the vertices of f). N, V =
Lj EN bj ej, bj E R, Lj EN bj = 1. The support of v is the set of j for which
bj =I- 0, and is denoted by supp v.
Theorem 2.0 (Ichiishi and Idzik (1996, Corollary 2.2A» Let X
be a nonempty compact convex subset of RN. Let f : X -+ p(RN) and
9 : X -+ P(X) be upper semicontinuous functions such that both f(x) and
g(x) are nonempty compact convex sets for each x E X. Let f transform
every face F of X in such a way that for each x E F, f(x) n aft' F =I- 0
(which would be the case, e.g., if f transforms every face F of X into
aft' F). Then f and g have a coincidence. In particular, X C f(X), and f
227
and g have jW;ed points.

Theorem 2.1 (Ichiishi and Idzik (1996, Theorem 2.3A» For each
i E N, let {Cf}vEK be a closed cover of fj"N satisfying
fj"T C U{Ci I v E K n aft' fj"T} for every T E P(N) \ {N}.
Then there exists a function 7r : N ---+ K such that
n
iEN
C;Ci) #0 and Usupp 7r(i) = N.
iEN
The closed covers considered in Theorem 2.1 were studied by Ichiishi

and Idzik (1990, Theorem 2.1) and Ichiishi and Idzik (1991, Theorem 3.1).
Theorem 2.1 reduces to Gale's (1984) lemma when K = {ej I j EN}. In
this case each cover {Cf}VEK is of the K-K-M type, that is, the boundary
condition is: fj"T C U jET C7i for every proper subset T of N. Our theorem
=
allows for covers of Shapley's (1973) type (K {ms I S E P(N)}); in this
case the boundary condition becomes: fj" T C USCT c;ns
for every proper
subset T of N.
Theorem 2.2 (Ichiishi and Idzik (1996, Theorem 2.3B» For each
i E N, let {Cf}vEK be a closed cover of fj"N satisfying
fj" N\ {j} C C7i for every j E N.
Then there exists a function 7r : N ---+ K such that
n
iEN
C;Ci) #0 and Usupp 7r(i) = N.
iEN
For the case K C fj" N, the type of closed covers considered in Theorem
2.2 was studied by Alexandrov and Pasynkov (1957) and by Scarf (1967).
Woodall (1980) proved the following
Theorem 2.3 (Woodall (1980, Theorem 4» Let fij : fj"N ---+ lR+ be
continuous functions such that fij(fj"N\{j}) = {O} for i,j EN. Then there
is a point x E fj"N (which we shall call a satisfactory point) and a bijection
7r : N ---+ N such that
(2.4) fi7rCi)(x) ~ fij(x) for i,j E N.
Observe that a more general theorem is true:
Theorem 2.5 Let fij : fj"N ---+ lR be continuous functions such that
fij(fj"N\{j}) = {O} for i,j EN and
(2.6) maXjEN fij(x) > 0 for x E 8/).N and i E N.
Then there is a point x E /). N and a permutation 7r : N ---+ N such that

228
(2.7) hrrCi)(X) = maXkEN fik(x) for i E N.

Observe that 8l1 N = UjEN l1N\{j} and that formulas (2.5) and (2.7) are
equivalent.
Proof of Theorem 2.5 Consider the sets
and apply Theorem 2.1 . o

Corollary 2.8 (Woodall (1980, Theorem 4), see Theorem 2.3) In
the case fij are nonnegative functions, we may assume the weak inequality
in (2.6).
Proof. To get this we take
-
fij(x) = /ij(x) +...J...
X·
m
(m E N)
and apply the limit procedure with m -+ 00. o

Kulpa (1994) proved the following
Theorem 2.9 (Kulpa (1994, Lemma 1» Let fj : l1N -+ ~+ be contin-
uous functions such that h(l1 N\{j}) = {O} for j EN. Then for any point
t E l1 N there is a point x E l1 N such that
(2.10) fj(x) = (2:kEN fk(x))tj for j E N.
We extend this theorem to
Theorem 2.11 Let fij : l1N -+ ~ be continuous functions such that
fij(l1 N\{j}) = {O} for i,j EN and
(2.12) 2: j EN fij(x) ~ 0 for x E 8l1 N and i EN
Then for any continuous functions ti : l1N -+ l1N;ti = {tij}jEN,i E N,
there is a point x E l1 N and a permutation 7r : N -+ N such that
(2.13) hrrCi)(X) ~ (2:kEN fik(x))tinCi)(x) for i E N.
Proof. Without loss of generality we may assume that (2: j EN fij(X)) >0
for x E 8l1 N (we can always consider
fij X +;;:
- (X) =fij () Xj (
X="
L...JXjejEl1 N ,mEN,jEN)
jEN
and apply a limit procedure with m -+ 00).
First, let us consider the case tij(X) > 0 for x E l1N and i,j E N. For
the sets
Cij = {x I fij(X) ~ (~ fik(X))tij(X)} for i,j EN,

kEN
229
we apply Theorem 2.1 and we are done. We get the general case by taking
- ()
ti" x = ...;.::.........,;--
tij(X)+€ for any € > 0, x E!l.N and (i,j EN),
3 1 +n€
and applying the limit procedure with € --+ O. 0
Theorem 2.14 Let gi : !l.N --+ Rand fij : !l.N --+ R be continuous func-
tions such that fij(!l.N\{j}) = {O} for i,j EN and
(2.15) gi(X) ~ 0 for x E 8!l.N and gi(X) ~ L,jEN fij(X)
for X E!l.N (i EN).
Then for any continuous functions ti : !l.N --+ !l.N;ti = {tij}jEN,i E N,
there is a point x E !l.N and a permutation 11" : N --+ N such that
(2.16) gi(X)t i7r (i)(X) ~ fi7r(i)(X) for i E N.
Proof. Without loss of generality we may assume that gi(X) >0 for x E
8!l.N (we can always consider
(x = LXjej E !l.N,m E N,i,j EN)

jEN
and apply a limit procedure with m --+ 00).

Now we consider the sets
and apply Theorem 2.2 to finish the proof. o

Remark 2.17 In the case fij = fik and tij = tik(i,j,k EN), Theorem 2.11
follows also from the theorem of Ichiishi and Idzik (1996, Corollary 2.2A)
(Theorem 2.0 in this paper). Furthermore, if the functions ti transform!l.N
to a fixed point t E!l.N (i EN), we get Kulpa's theorem.
Theorem 2.14 contains the following a permutation-based generalization
of Brouwer's fixed point theorem:
Corollary 2.18 (Bapat (1989, Theorem 4) Let fi : !l.N --+ !l.N be
continuous functions (i EN). Then there is Y E !l.N and a permutation
()" : N --+ N such that
Y<r(i) ~ fi<r(i)(Y).
And from Theorem 2.11 follows a dual theorem to Bapat's theorem:
Corollary 2.19 Let fi : !l.N --+!l.N be continuous functions (i EN). Then
there is x E !l.N and a permutation 11" : N --+ N such that
230
3 The matching game

The matching game is defined by two finite sets of players M and N
(w. 1. o. g., we assume N = M). Analogously as Alkan and Gale (1990),
we assume that a payoff set for a pair (i,j) of players is a graph of a
continuous real function fij defined on some interval 0 ~ x ~ Cij (w. 1. o.
g., we assume that Cij = 1).
Also we assume that one of the following cases holds for each i,j E N:
(3.1) fij(O) = 0 and fij(X) > 0 for 0 < x ~ 1,
(3.2) fij(l) = 0 and fij(X) > 0 for 0 ~ x< 1,
(3.3) fij(O) = C> 0 and hj{x) < C for 0 < x ~ 1,
(3.4) fij(l) = C> 0 and fij(X) < C for 0 ~ x < 1.
The cases (3.2) - (3.4) reduce to the case (3.1) by taking x = 1 - x and
lij(x) = C - fij(X), respectively. The matching 1r is a partition of the set
MUN into pairs (i, j), i E M, j E N and can be considered as a permutation
1r of the set N. A feasible payoff is a pair of vectors x = {xihEM, Y =
{Yj}jEN (x,y E RN) such that
(3.5) Yi = Atr(i)(X7r(i») for i E N and some permutation 1r : N ---4 N,

and is called stable if
(3.6) Yi 2: fij(Xj) for i,j E N.
Combining (3.5) and (3.6) we have
(3.7) fi7r(i)(X 7r (i») 2: fij(Xj) for i,j EN and some permutation 1r: N ---4
N
A core payoff of the matching game is a vector x = {XdiEN satisfying
(3.7). Observe that our definition of the core payoff differs from that of
Alkan and Gale (1990).
Theorem 3.8 There is a core payoff.
Proof. Define lij : ~ N ---4 R+ by
-
fij(X) = hj(xj) fen' X = L.J
~
Xjej E ~ N (i,j EN)
jEN
and apply Woodall's theorem (Theorem 2.3). o

Remark 3.9 In view of Woodall's theorem, we can prove a more general
result for the matching game than the existence of the core payoff. Namely
we can prove the existence of the stationary point for the matching. This
means that a feasible payoff may depend on externalities, i.e., on a choice
of the other players.
231
References
Alexandrov, P. and B. Pasynkov, 1957, Elementary proof of the essentiality
of the identical mapping of a simplex (in Russian), Uspekhi Matematicesk-
ikh Nauk 12, 175-179.
Alkan, A. and D. Gale, 1990, The core of the matching game, Games and
Economic Behavior 2, 203-212.
Bapat, R.B., 1989, A constructive proof of a permutation-based general-
ization of Sperner's lemma, Mathematical Programming 44, 113-120.
Fan, Ky, 1972, A Minimax inequality and applications, in: O. Shisha, ed.,
Inequalities Ill, Proceedings of the Third Symposium on Inequalities held
at the University of California, Los Angeles, September 1-9, 1969 (Aca-
demic Press, New York) 103-113.
Gale, D., 1984, Equilibrium in a discrete exchange economy with money,
International Journal of Game Theory 13, 61-64.
Idzik, A., 1995, Optimal divisions of the unit interval, Game Theory and
Applications, International Conference in Honor of Robert J. Aumann
on his 65th Birthday, Jerusalem.
Ichiishi, T. and A. Idzik, 1990, Theorems on closed coverings of a simplex
and their applications to cooperative game theory, Journal of Mathemat-
ical Analysis and Applications 146, 259-270.
Ichiishi, T. and A. Idzik, 1991, Closed covers of compact convex polyhedra,
International Journal of Game Theory 20, 161-169.
Ichiishi, T. and A. Idzik, 1996, Equitable allocation of divisible goods and
market allocation of indivisible goods, ICS PAS Reports, Warsaw.
Kulpa, W., 1994, Sandwich type theorems, Acta Universitatis Carolinae -
Mathematica et Physica 35, 45-50.
Quinzii, M., 1984, Core and competitive equilibria with indivisibilities, In-
ternational Journal of Game Theory 13, 41-60.
Scarf, H., 1967, The approximation of fixed-points of a continuous map-
pings, SIAM Journal of Applied Mathematics 15, 1328-1342.
Shapley, L. S., 1973, On balanced games without side payments, in: T.
C. Hu and S. M. Robinson, eds., Mathematical Programming, (Academic
Press, New York) 261-290.
Shapley, L. S. and H. Scarf, 1974, On cores and indivisibility, Journal of
Mathematical Economics 1, 23-37.
Thomson, W., 1995, The replacement principle in economies with indivis-
ible goods, Rochester Center for Economic Research Working Paper No.
403.
Woodall, D.R., 1980, Dividing a cake fairly, Journal of Mathematical Analy-
sis and Applications 78, 233-247.
Optimal entry and the marginal contribution
of a player
K unio Kawamata
Keio University 2-15-45 Mita, Minato-ku, Tokyo, Japan
Abstract. We introduce the concept of the marginal contribution of a

player (firm) and use it to derive conditions for optimal entry in various
industrial situations. It turns out that, in a competitive economy with
a finite number of goods but with a continuum of potential firms, the
marginal contribution of a firm coincides with the profit of the firm, and
so the optimal condition for entry is that the marginal firms should receive
zero profit. We also study the marginal contribution in the monopolistic
competition markets and establish the "excess entry theorem" in a new
setting.
JEL Classification Numbers: e71, D40

Keywords: Optimal entry, contribution of a player, convex game, profit
1 Introduction
The optimal 1 number of firms in an industry could be either one, two, or
many, depending on the market structure. The main purpose of this paper
is to introduce the concept of "the marginal contribution of a player( a firm
in the sequel)" and use it to derive conditions for optimal entry in various
industrial situations. We consider an economy with a finite number of goods
but with a continuum of potential firms. The "marginal contribution of a
firm" is defined roughly as (the limit, as the measure of the firm approaches
zero, of) the difference between the maximal welfare that the economy can
attain with the firm and without it. It turns out that, when there are
fixed costs, not all firms should produce positive outputs even if they have
the same production technology. Under perfect competition, the marginal
contribution of a firm coincides with the profit of the firm, and so the
optimal condition for entry is that the marginal firms should receive zero
lOur criterion of optimality here is the maximality of the Bergson-Samuelson type

social welfare function. We assume away the problems associated with imperfect in-
formation and suppose that the government can attain the optimum by some policy
means.
234
profit.
Our concept of the "marginal contribution of a firm" is closely related to
the idea which welfare economists, e.g., Kahn (1935) and Hicks (1939), had
in mind in discussing optimal industrial structure or the ''total condition-
s" for optimality. The game theoretic concept of Shapley value (see, e.g.,
Shapley (1953), Aumann and Shapley (1971)) is also related to the present
concept. But whereas the Shapley value is the "expected payoff" of the
game when all agents are arranged in random order, in our definition, firms
are ordered according to their productivity where productivity is defined
in a natural way. Using this concept we derive conditions for optimal en-
try which were obtained verbally or in a partial equilibrium framework by
Kahn (1935), Hicks (1939) and obtained in a general equilibrium framework
by Negishi (1962, 1972).2 See also Makowski (1980) and Ostroy (1980) for
related discussions.
Our analysis stands in contrast with previous studies in that the set of
agents are contained in a non-atomic measure space. The same approach
is also useful in analyzing the problems of the monopolistic competition
market, as we will show in Section 4. We establish a version of excess entry
theorem which conveys a message similar to the one Suzumura and Kiyono
(1987) established for the oligopolistic market. This approach, which follows
the procedure of Aumann (1964, 1975), has the advantage that the marginal
contribution of a firm can unambiguously be expressed in terms of the prices
and the allocation of the economy, and the convexity assumptions on prefer
ences and technologies can be relaxed to a certain extent.
2 A preliminary example
In order to clarify the nature of the problem and motivate the analysis in
the following sections, we first present a simple example and derive opti-
mal conditions for entry in this case. In this section all firms are treated
discretely, and the analysis is informal for reasons that will be explained
below.
Suppose that the welfare of an economy can be expressed by the utility
function
u = x·(a -I) (1)
of a representative consumer, where x is the amount of the consumption
good available to him, 1 is the amount of labor he supplies and a is a
positive number representing the maximal amount of labor that he can
2Negishi's theorems state that (i) if it is known that positive profit is impossible for
the new firm under prices ruling before entry, entry should not be made and that (ii)
if the new firm is running without a loss after entry, then the firm should have entered
after all (see Negishi (1972)). The last statement needs a careful interpretation if the
incumbent firms are not the most desirable from the welfare viewpoint.
235
supply in a fixed time (thus a - l represents consumption of leisure). Let

:l. = {1, 2, 3, ... } denote the set of firms in the economy that can potentially
produce the consumption good, and assume that the production function
of the j-th firm (j E :l.) can be written as
. _ { Vlj - bj if lj > bj (2)

xJ - 0 if l·J <
-
b·J '
where lj is the amount of labor, Xj is the amount of production, and bj is

a given non-negative number representing the fixed input of the j-th firm.
Let us first consider the situation where only firms in a subset J of:l. are
active (T his means that lj = 0 for all j E :l.\J). If some firms in J are not
producing positive outputs, then the consumer need not supply a positive
amount of labor to these firms. Hence in considering the social optimum we
may assume that all of the members of J are producing positive outputs.
We now formulate the problem (PJ ) for each such J c:l. as
(PJ) Maximize
u = x·(a -l)
subject to
X= LVlj -bj (3)
jEJ
and
(4)
jEJ
From this we easily obtain the familiar marginal conditions for optimal-
ity:
(5)
Hence, in view of (1), (3) and (4), we have the following optimal production
allocation:
X;(J) = Vlj - bj (j E J) (6)
l;(J) = (a- Lbi)/3n+bi (j E J), (7)
and the corresponding optimal utility
(8)
where the summations are over J , and n is the number of firms producing
positive outputs, i.e., the cardinality of J. (The above results show that J
must be chosen so that a - L bj > 0.)
In the next step we allow J to vary, and choose x; (J) and l; (J) so as to
maximize u*(J) . To simplify the analysis we shall suppose that the firms
are arranged so that
(9)
236
This implies that the production function of the j-th firm is uniformly
above that of the k-th firm for k > j. Thus, if the k-th firm is producing
positive outputs at the social optimum, then so should the j-th firm, for any
j < k. Hence in order to choose the optimal set of firms, J, it is enough
to determine the optimal number, n, of firms that will produce positive
output.
In the characteristic function form game (u", J) with the characteristic
function u" and the player set J, the marginalworthof aplayer j to coalition
S (S C J) is defined by
u"(S U {j}) - u"(S) for j rt S.

Hence writing u"[n] for u*(J) (where n is the cardinality of J) it may seem
natural to define the marginal worth of the n-th firm by u"[n]-u"[n-1] or,
supposing that u* [n] is defined for all real numbers, by duO /dn . Actually,
it turns out to be more convenient to define it by
duO /dn
(10)
Qu*/8l'
which is also independent of the choice of utility functions (The denomi-

nator represents the marginal disutility of labor evaluated at the optimum
allocation.). This corresponds to what we will later call the marginal con-
tribution of the firm.
In the special case where bj = b for all j, we have
u*[n] = 2vn(a - nb)i /3v3. (11)
Hence if we allow n to take on all positive values, we have
duO _ ya::-nb(a - 4nb)

(12)
dn - 3V3n
Since, by (5), aUl;' s are equal in this case, in view of (1), (3), (4) and (7),
we obtain
(13)
and
Ou"
QX = 2(a - nb)/3. (14)
Equations (12) and (13) then imply
QU" /dn a - 4nb

(15)
Ou" /8l 3n
237
Now if the price vector (-u;jui, 1) is used to evaluate the profit 7r of the
firm, we have, from (6), (7), (13), and (14)
7r = u; * -l·*
--x·
ui J J
a-4nb
= 3n
(16)
Comparing (15) with (16) we may conclude that the marginal contribution
of the firm is the profit of the firm.
The present analysis, which dealt with the case of a finite number of
firms, is somewhat informal because the marginal contribution was not
defined accurately. In the following sections, we shall rigorously establish
similar results in more general settings without restricting ourselves to the
special production functions and the utility function of this model.
Remarks
(a) In the special case where bj = b > 0 for all j, (12) shows that the
optimal number of the firms in the industry is given by aj 4b, if it is an
integer. This implies that not all firms should stay in the industry even if
they have the same technology.
(b) If, moreover, bj = 0 for all j, then u* [n] is an increasing function of
n, and there is no optimal number of firms for the economy.
(c) That the marginal worth is an increasing function with respect to
the coalition size is a characteristic feature of the convex game which has
been studied by Shapley (1971), Ichiishi (1981) and Topkis (1987), among
others. Remark(a) shows that the present model contains an example of a
non-convex game.
(d) As a model of entry in a free market, the discrete model must rely
seriously on the assumption that entry occurs in the order of superiority
in technology as expressed in (9). It is easy to construct an example in
which (i) a finite number of firms are making positive profits and that (ii)
a technologically superior firm incurs a loss should it enter the market.
To see this, slightly increase the parameter bi of an incumbent firm in the
model of Remark (a).
3 The marginal contribution and the efficiency price

In this section we consider two different models of an economy with a finite
number of goods but with a continuum of firms. There are no restrictions
on prices or quantities of the goods traded and monopolies are ruled out.
In both of these models firms are assumed to be arranged in a certain nat-
ural order, and we consider the overall effects of "another" firm joining the
industry. The performance of the economy is considered to be expressed by
a real valued function, which we may call an objective function or a welfare
238
function. The marginal contribution of a firm is then defined as the limit,

as the measure of the firm approaches zero, of the increase in maximum
welfare, divided by the marginal contribution to welfare (marginal utility)
of a numeraire, say, labor. (See also the discussion below.) Equation (10) is
the expression for this in the economic model of Section 2. For the defini-
tions of economic concepts not defined here we refer to Samuelson (1947),
Debreu (1959) and Arrow-Hahn (1971). The main result that we establish
in this section is
Theorem 1 In the classical 3 Arrow-Debreu competitive economy, the

marginal contribution of a firm is equal to the profit of the firm in terms
of the efficiency prices.
An efficiency price vector in terms of the numeraire good is the vector
of marginal rates of substitution when they exist. In general it is defined
by the normal vector of the separating hyperplane to, say, the production
set. Since competitive prices are also efficiency prices (cf. Debreu (1957) or
Arrow-Hahn (1971)), Theorem 1 implies
Theorem 2 Under the same assumption as in Theorem 1, the optimal
condition for the entry of firms is that the profit of the marginal firm should
equal to zero.
We will prove Theorem 1 under two slightly different sets of assumptions
in models A and B. Model A is a continuum analogue, extended in several
respects, of the example in Section 2. It is assumed that the welfare of the
economy is described by the utility function of a representative consumer.
Model B L'3 quite general in its treatment of production technology, but
consumers' demands for goods are assumed to be given exogenously.
Model A
Let us assume that the utility function of a representative consumer is
given by
(17)
where Xi (i = 1,2) denotes his consumption of good i, l is his labor supply
and a is a given positive number representing the maximum amount of
labor that he can supply. We make
Assumption A.I u(·) is increasing, strictly quasi-concave and twice con-
tinuously differentiable. The set of firms that can potentially be in industry
3This usually means the economic environment with convex preferences and convex
production technologies and with no externalities. However the term "classical" is used
here in a somewhat broader sense than usual. Firms may require a fixed amount of
inputs when they produce a positive amount of outputs although no inputs are required
when no outputs are produced. Hence the average cost curve is decreasing when output
levels are small.
239
i is represented by a bounded interval7i(i = 1,2). We suppose that Tt and

T2 are disjoint.
For each i(i = 1,2), let the production function of firm t be denoted by
.(t) _ {fi(li(t) - bi(t),t) if li(t) > bi(t)

0 if (18)
X, - li(t) ~ bi(t), (i=I,2)
where Xi (t) is the density of production of good i and li (t) (of which bi (t) >
o is a fixed amount) is the density of labor input for firm t in industry i. This
means that given li(t)dt oflabour the firm can produce fi(li(t)-bi(t), t)dt of
the product if li(t) > bi(t). (See, e.g., Aumann (1975) or Aumann-Shapley
(1971) for a related way of representing agents.) We make
Assumption A.2 For each t, fi(" t) is increasing, strictly concave,4 and
twice continuously differentiable. For each X, fi(x,,) is continuous except
possibly at a finite number of points, (i = 1,2).
The present model can be generalized to the case of any finite number of
goods. Model B allows the existence of intermediate goods. Let us denote
by 7i E 7i the set of firms actually producing positive outputs in industry
i(i = 1, 2). To simplify the analysis we make
Assumption A.a 7i is a disjoint union of a finite number of intervals
= 1, ... , ik)in Ti .
Tik (k
We may suppose (as was explained in Section 2) that all firms in Ti
are actually producing positive outputs. In the sequel we shall often write,
e.g., ITf instead of IT
fdt. The demands for goods are satisfied if
Xi ~ f (i = 1,2)
iT; Xi(t) (19)
and
(20)
Since u(·) is increasing, when finding the optimum, we may replace the
inequalities in (19) and (20) by equalities. And if we extend the definitions
of li(t) and Xi(t) , by setting them equal to zero outside Ti , we may replace
the domain of integration, 7i, by T = Tt U T2. Thus for each of Tt and T2,
we formulate the problem (PT) as
(PT) Maximize
u=u(l Xt(t),l x2(t),a-l(lt(t)+l2(t))) (21)
4We will argue below that this assumption is not practically important as is the case
in the discrete economy.
240
subject to
(22)
The existence of the maximum and some other related properties will
be discussed in Section 5 in a more general framework, in which we will
assume that Xi(t) and li(t) are Borel measurable functions. For the present
we assume that the maximum exists and impose the following conditions
on admissible functions.
Assumption A.4 li(t) and Xi(t) are continuously differentiable in the

interior of each of the sub-intervals Tik as defined in (A.3).
The problem(PT) can easily be solved by substituting (22) into (21).

Taking the variational derivative (see, e.g., Gelfand and Formin ((1963),
pp.27-28) of u with respect to li' we know that the conditions for the
extremum are
au afi + au = 0 (i = 1,2.) (23)
aXi Oli Ol
These are nothing but the familiar marginal conditions for optimality. We
next let Ti (i = 1,2) vary and consider the effects of the change on the
optimal solutions of (PT). To simplify the analysis we make
Assumption A.5 The left end-point of each sub-interval of Ti(i = 1,2),

as defined in (A.3), and the number of these sub-intervals are known.
The left end point represents (technologically) the most superior firm in
the industry. We may suppose that (A.5) is satisfied if there are only a finite
number of potential types of firms in an industry. More generally, (A.5) is
satisfied if it is possible to classify firms into a finite number of groups
in such a way that, within each of the groups, the production function
of one firm is uniformly above or below that of another. Owing to (A.5)
we need only consider changes in the right end-points of the sub-intervals
representing the most inferior firm. Let us consider the effects of a change
in a right end point, a = tis, of Tis.
Differentiating u(·), along the optimal path, with respect to a, we have
(denoting by j the index different from i)
du
da
(24)
Noticing that au/axi and au/Ol are independent of t, we have from (23),
and (24),
(25)
241
This means that the marginal contribution of firm a (the left hand side) is
equal to the profit of the firm in terms of the efficiency price vector,
auau
p= (--/-,1) (26)
aXi 8l
(the right hand side).
Model B
Let n be the number of goods in the economy and, for each i E N =
n
{1, 2, ... , n}, let be a bounded interval on the real line R. We consider
Ti to be the set of all potential firms in industry i. We assume that Ti
and T j are disjoint for i i j. If good i is not the product of any firm, we
take'!l to be empty. For all i,j E N with i i j, and each t E Ti let yj(t)
be the density of good j used (of which bj(t) is a fixed amount) in the
production of good i by firm t, and let Yi(t) be the density of its output.
For simplicity we assume that there are no joint outputs and we write the
firms' production functions as
for fi(t) 2:: bi(t)

(27)
otherwise (i E N, t E L) ,
whereili(t) = (yi(t), ... ,yLl(t),yt+1(t), ... ,y~(t)) and similarly for bi(t). We
make the following assumptions on the production technology.
Assumption B.l For each t, ft, t) is increasing and twice continuously
differentiable, and for each yi, fi(yi, .) is continuous except perhaps at a
finite number of points.
Assumptions must also be made on the asymptotic behavior of fi(., t),
in order to guarantee the existence of a maximum of the problem to be
formulated below. This point will be discussed in the Appendix so, for the
moment, we will not worry about the problem of existence. Let Ti C Ti
denote the set of firms producing positive outputs in industry i(i E N) and
impose
Assumption B.2 The same as Assumption A.3 in model A.
Let Ci (i = 1,2, ... , n - 1) denote the aggregate net demand for good i .
That demand will be satisfied if
Ci ~ 1. Yi(t)dt - L 1.111 (t)dt. (28)

Tt j#i TJ
Extending the definitions of yJ (t), Yi(t) and b;' (t), by defining them to be
equal to zero outside n, we may replace n in \28) by T = UTi' Now, for a
given (n) (i EN), we formulate the problem (PT) as
242
(PT) Maximize
(29)
subject to
Ci = ( ji(f/(t) - bi(t), t)dt - L { Y1 (t)dt

iT j~iiT
(i = 1, ... , n - 1) (30)
where Ci (i = 1, ... , n -1) and bi(t) (i = 1,2, ... , n) are assumed to be given.
A natural interpretation of the problem is that it is to minimize the sum
of the labor inputs of the economy on the conditon that specified demands
are satisfied. The equality in (30) is due to the assumption that ji(., t) is
increasing.
As in the previous model, the following conditions are imposed on the
admissible functions
Assumption B.a All yj(t) and Yi(t) are continuously differentiable in the
interior of each sub-interval, Tik' as defined in (A.3).
Following the standard procedure (cf. Gleaned and Forman (1963) pp.43-
46) we write the Lagrangian of the problem as
with Pn = 1 and Cn = 0, and obtain the Euler conditions for optimality

aJi(t)
Pi ay] = Pj (i,j ENt E 1i), (32)
where we set Ji(t) = ji(f/(t) - bi(t),t) . These are the familiar marginal
conditions for optimality. We next let (Ti) (i E N) vary and consider the
effects of the change on the optimal solutions of (PT). To simplify the
analysis we impose
Assumption B.4 The same as (A.5) in Model A.
With this assumption, we need only consider changes in the right end
points of the sub-intervals ns.
Differentiating (30) with respect to a ~ight end point a = tks, we have
(noticing that the optimal solutions of yf ,Yi and Pi are functions of a)
m (a ) = PkYk ()
a - '"'
L..JPiYik()
#k
a + h'"' ('"'
T iEN #i vYJ
~.~ ay]
L..JPi L..J aji(t) aa - '"'
L..J ay{)
j~i
aa dt.
(33)
243
But the terms in parenthesis cancel out since, by (32),
(34)
The last two relations imply
m(a) = PkYk(a) - LPiyf(a). (35)

i#
Since the marginal contribution of Yn to the objective function is Pn = 1,

we know that m(a) is the marginal contribution of firm a = tks. By (35),
it is equal to the profit of the firm in terms of the efficiency price vector
(Pl,P2, ···,Pn-l! 1).
4 The marginal contribution in a monopolistic model

In this section we apply the previous analysis to derive the marginal con-
tribution of a firm in a simple model of monopolistic competition. The
contribution to welfare of a monopolistically competitive firm is calculated
under the assumption that the behavior rule of other firms in the mar-
kets are unaltered. Another possible interpretation of the model will be
discussed below.
Model C
The basic framework of the model is the same as that of model A except
that there is only one industry in the present case. Using the previous
notation let
u= u(x) +a-l (36)
be the utility function of the representative consumer which is the sum of
utility from a consumption good u(x) and the leisure a - l. We assume
that u is an increasing concave function. The industry has a continuum
of potential firms, which we denote by T. We assume that T is a bounded
interval in the real line R . The production function of firm t of the industry
is denoted by
x(t) = {!(lo(t) - b(t), t) if l(t) > b(t) (37)

if l(t)::; b(t)
244
We will assume that f is concave in the region l(t) > b(t) We make As-
sumption (A.l) and Assumption (A.2) of Section 3 applied for the single
industry case.
The profit of the industry is expressed as
7r =h (px(t) - l(t))dt, (38)
where p is the price of the product in terms of the wage rate. We denote
the (inverse) demand function of the consumer as
p = P(x), P(x) = u'(x). (39)

We have
Assumption C.l
P'(x) < 0,
P(x) + xP'(x) > 0,
2P'(x) + xPII(x) < 0.
The first inequality means that the marginal utility of the good is decreas-
ing. The second and the third inequalities say that the marginal revenue
is positive and decreasing. It is also assumed that monopolistic firms in
the industry maximize their joint profits 7r with respect to l(·),xO, and T,
knowing the consumer's demand function for their good P. See, Remark (a)
below for another interpretation. We make assumptions (A.4) and (A.5).
We now consider the problem:
(P) Maximize
7r = l[p(x)x(t) -l(t)]dt (40)
subject to
x = h x(t)dt
= h f(l(t) - b(t), t)dt (41)
First we solve the problem considering that x and T are fixed. We set
the Lagrangean of the problem as
L = h[Pf(t) -l(t) - A(J(t) - ~)ldt (42)
where
f(t) = f(l(t) - b(t), t)
245
and
f3 = length of T.
From this we obtain the following Euler condition for optimality:
(P _ >/Jf(t) =1 (43)
8l
for all t. This implies that the marginal products of labor are equal for all
firms within the industry.
Next we vary x and a = ts (a right end point of a sub-interval). Assuming
that the solutions, still denoted l(·),x(·) etc., are unique and differentiable
with respect to x and a, we have
i[p' f(t) - «P - >.)f'(t) - 1) :~ + ~ldt = 0 (44)
and,
>.x
Pf(a) l(a) - >.(a)f(a) + If
+ f[«(P->')f'(t)-l)aal(.) - >'~ldt=O, (45)

iT a f3
where we have set
P ' _ dP d f'( ) _ af(t)
- dx an t - m(t) .
In view of (43) and (44), we then have
-£ P' f(t)dt = >.

1
P - f'(t) (t ET). (46)
We note that>. > 0 since P' > O. Hence, noticing that P' is independent
of t and using (41), we have
P(x) + xP'(x) = f'~t). (47)
On the other hand, (43) and (45) yield
f(a) = f'(a) (48)

l(a)
for each a = ts (s = 1,2, ... , Si). Since 1/ f'(t) is the marginal cost (MC) of
the product equation (47) may be expressed as
x dP
--. - = (p- MC)/p.
p dx
(49)
246
Combining (43), (47) and (48) we can state

Lemma 1 Under assumptions (A.l)-(A.5), the profit of each industry in
Model D is maximized if (i) marginal products of labor are equal for all
firms, (ii) the mark up ratio equals the elasticity of inverse demand function
for the product, and (iii) the marginal cost equals the average cost of the
marginal firm.
Next we consider a slightly different problem. Suppose that firms in the
industry maximize their joint profit, 7r as in the previous analysis, but
T is now under the control of government. T will be chosen so that the
utility of the representative consumer is maximized given the behavior of
the monopolistically competitive firms.
For each T E T let i(.), a{), and x be the solutions of the problem (P)
(hence these solutions satisfy (43), (47) and (48)). In the sequel, the tilde
sign over the functions will be deleted. We consider the following problem:
(P) Maximize
=
u(£ f(t)dt, a- £l(t)dt)
U u(x)+a-l
= (50)
with respect to T where l(t) is the solutions of the problem stated above.
Consider a change in a = ts, one of the right end points of the sub-
intervals in (A.5). Differentiating (50) along the optimal solution, with
respect to a, we have
dU
da
U'(x)f(a) + £ f'(t)! dt
+l(a) + J Ol
oa dt (51)
If the consumer maximizes utility at given market prices, then

du
dx =p, (52)
hence if we define
s = (p -
MC)/MC, (53)
where MC is the marginal cost of the industry, we find from (47) (noticing
that MC = 1/ /') that s is positive. Also (51) may be expressed as
dU
-=pf(a)-l(a)-s
cia
1T
dl.
-
da
(54)
Now differentiating (47) with respect to a and noticing that /,(t) is

independent of t we have
(2P'(x) + xP"(x)).-
ox
oa
247
f"(t) . dl(t)
= {J'(t))2 da
(55)
where
(56)
Finally to obtain a clear-cut conclusion, we assume that

Assumption C.2 The industry profit increases if there is an entry of a
marginal firm.
In view of (55) and assumptions on the sign of derivatives of functions
we can show that dl(t)/da > O. We have thus proved [see (54)]
Theorem 3 The marginal contribution of a firm in model D is equal to
the difference between (i) the profit of the firm and (ii) the increase in the
total costs of all monopolists each multiplied by the corresponding mark up
ratio, s. This second term takes on a positive value under assumption C.2.
Remarks
(a) Notice that if we denote the demand elasticity of the good (the recip-
rocal of the left side of (47)) bye, we have
1
S=--. (57)
e-1
Hence (54) is in accordance with the formula of Kahn ((1962) p.29), which
was obtained in a partial equilibrium framework. Notice that although he
did not assume the joint profit maximization, he did assume that the mark
up ratio is constant for all firms in the industry. As to the simplifying
assumption on which this result depends see McKenzie (1951).
(b) As an alternative interpretation of the present model, assume that
the industry is monopolized by a firm which has a continuum of potential
factories, T. Then the maximization of the profit of the monopolist can be
analyzed in exactly the same way as in the present model.
In the last interpretation, in view of (52), we have the following result:
Theorem 4a In the monopolistic market, if the firm operates its factories
until the last of them earn zero profit, the contribution is positive. Hence,
entry is excessive.
This corresponds to the content of the excess entry theorem in Suzu-
mura and Kiyono (1987), which was established for the homogeneous good
Cournot-type oligopoly model. Weizsacker(1980) analyses a heterogeneous
duopoly model with a quadratic utility function.
248
(c) The above framework may be interpreted as a model of monopolis-

tic competition, as formulated by Chamberlin (1933), with a large (non-
oligopolistic) group of suppliers of physically similar but economically dif-
ferentiated products. Bishop (1976)5 analyzed the welfare implication of
equilibrium of the market where, as in the Chamberlin's idealization, all
the actual and potential members of the group have the "same" costs and
face the "same" demands. He showed diagramatically that, in the monopo-
listic competition market, entry is excessive from the consumer's viewpoint.
The proposition was generalized in the present analysis to the case where
the production functions (cost functions) of firms in the industry may be
different.
Theorem 4b In the monopolistic competition model, where all the actual
and potential firms in the industry face the same demands, the optimal
product variety calls for production at a point short of minimum average
cost of the marginal firm.
This result is a direct consequence of (54). We need to interpret that
the domain of integration 1i now represents the variety of the (physically
identical) products in the industry.
5 Appendix to section 3: the existence of the

optimum
In this section we will prove the existence of solutions to problem (PT) in
models A and B, upon which our analyses depend heavily. Detailed proofs
will be given only for model B because the proofs for model A are essentially
the same and even simpler.
Model D
This is a modification of model B, with many of the technical assumptions
generalized. Let n be the number of goods in the economy. To simplify
the argument we assume that good n is labor. For i = 1,2, ... , n - 1, let
m, Bi, J.l) be a measure space where Ti is a bounded interval in the real line
~,Bi, is the a-algebra of Borel sets of 1i and J.l is the Lebesque measure.
As before Ti is the set of potential firms in industry i, and each member Ti
of Bi is interpreted as the set of firms that are actually producing positive
outputs in industry i.
Problem (PT) is formulated as in model B. But this time we choose Tn
to be empty (labor is never produced). Hence the problem reduces to
(PT) Maximize
(58)
51 owe to Professor Leonid Hurwicz for bringing this article to my attention.

249
subject to
Ci = r (fi(fi(t) - ii(t), t) - Ly~(t))dt (i = 1,2, ... , n - 1). (59)

iT j~n
Since fi is assumed to be increasing, (PT) is unaltered if we replace the
inequalities by equalities. Instead of (Bl) we make the following assump-
tion.
Assumption D.I For each i E N, (i) fi(fi(t), t) is continuous for almost
all (fji(t), t) E ~-l X Ti(IR~_l denotes the non-negative orthant of n - 1
dimensional Euclidean space) and (ii) fi(fji(t), t) is increasing in fji(t), for
almost all t E Ti.
For some of the arguments below it is enough to replace (i) by (i)'
for almost all t, fi(., t) is upper semi-continuous and, for almost all fji(t),
fi (fji (t), .) is measurable. Such a numerical function is usually referred to
as a Caratheodory function (in a minimization problem fie-, t) is assumed
to be lower semi-continuous). It is a special case of a normal integrand (see,
e.g., Ekeland-Temam (1976), pp. 231-234), all of which satisfy the condition
that (i) for alm~~t all t, fi(., tJ. is. upper ~emi-continuous and t~ere exists a
Borel function r r r
such that (fjt , .) = (fjt , .) for almost all fjt.
All functions that we consider are assumed to be integrable. We now
make
Assumption D.2 For each i = 1,2, ... , n-l, Ci > 0, and it is technologically
°
possible to satisfy net demand Ci + di , (i = 1,2, ... , n - 1) for some di >
(Le., (59) has solutions fji(t) 2: when each Ci is replaced by Ci + di )
°
The assumption on the sign of cis is made mainly for simplicity of ex-
position. It is very easy to cover the case where some of them are negative
(the case of primary factors of production).
In order to rule out the possibility that the production of a good will be
carried out by a negligibly small set of firms, we need a certain uniformity
assumption on the production technology. To simplify the argument we
assume that, given the set of active firms in an industry and the net final
demand for the good, there are lower bounds such that if the members of
a non-negligible set of firms are using inputs beyond any of the bounds,
then there exists a more efficient way of allocating resources within each
industry. More precisely, we make
Assumption D.3 (inefficiency of overconcentration) For each (i =
1,2, ... , n - 1) there exists ai E 1R+.- 1 (which may depend on Ci and T i )
such that if not fji(t) ~ ai for almost all t in some non-null set Si C Ti ,
there exists yet) E 1R+.- 1 such that yet) ~ ai for all Ti ,
250
and
This assumption is likely to be satisfied if firms in an industry can be

classified into a finite number of groups with positive measures, in such a
way that firms within each group are technologically the same or "simi-
lar" and there are "no increasing returns" to production. Because of this
assumption we may suppose that the optimal solution of (PT) lies in a
compact set defined by (ii (i = 1,2, ... , n - 1).
Finally we will give a simple definition. Let f and g be functions from
X x T to i. (the extended real line). We say that f is integrably dominated
by g if, for every e > 0, there exists a positive integrable function, e(t),
such that
f(x, t) ~ e(t) implies f(x, t) ~ e g(x, t).
We are now ready to state the following theorem.
Theorem 5 Under assumptions D.l, D.2, D.3, A.3, and A.S there exists
a solution to problem (PT) in model D, where the admissible solutions are
taken to be all measurable functions ..
The proof of Theorem 5 depends heavily on the following Theorem due
to Berliocchi and Lasry ((1973), pp.155-156), which is an extension of the
main theorem of Aumann and Pedes (1965). We can apply it to the present
case when Assumptions D.1 and D.2 are made beside the conditions stated
in the Theorem.
Theorem A. Let gn : lRn - 1 x T ~ lR be a Borel function such that x ~
gn(x,t) is upper semi-continuous almost everywhere and gi,g2, ... ,gn-l be
normal integrands of Rn-l x R+ ~ R. If (a) sup (O,gn) is integrably
dominated by gl + g2 + ... + gn-l and (b) lim (gl + g2 + ... + gn-l )(x, t) ~
00 as Ilxll ~ 00 almost everywhere, then the problem
(Q) maximize
subject to
irgi(x(t),t)dt ~ ki (i=1,2, ... ,n-1)
(where k i > 0) has a solution. If the domain of gn is S x T, where S is

compact, then the assumption on the asymptotic behavior of E gi can be
dropped.
Proof of Theorem 5. We define
(60)
251
gn(x(t),t) =- L.y~, (61)

#n
and for each i = 1,2, ... , n - 1 let gi(X(t), t) be defined by the right hand
side of (59). We may add a positive number to each of equalities in (58)
and (59) and assume that the right hand side of each of the equalities are
non-negative. This proves the existence of a solution to (PT).
The existence of a solution to problem (PT) in Model A can be proved
in a very similar way. The key to the proof is the following proposition of
Berliocchi and Lasry ((1973), p.155).
Theorem B. Let fi : XxT ~ lR (i = 1,2, ... , k) be Camthfodory functions
and gi : X x T ~ R (i = 1,2, ... ,n) be normal integmnds. If lim 'L,gi ~
00 almost everywhere and each Ifil is integmbly dominated by 'L,gi and u:
~k ~ ~ is continuous, then problem
(Q) Maximize
subject to
t/i(x(t),t)dt ~1 (i = 1,2, ... ,n)
has an optimal solution.

In Theorem 5, we gave conditions under which there exists a measurable
solution, yi(t)(i E N) to the problem (PT). Let us next give conditions
under which these functions are chosen to be continuous in each of the
subintervals ofTi. For each c5 i ,O < c5i < di, where di is defined in (D.2), we
consider a "perturbed problem":
(Po) Minimize
(62)
subject to
We set
and
h(c5) = inf(Po), (64)
namely, the infimum of problem Po.
252
We also set
n-1
L(x, t, 8*) = gn(x, t) + L: 8; gi (x, t), (65)
;=1
where we define x by (60) and gi(X, t) (i E N) by
gn(x,t) = L:Y~, (66)

#n
and
gi(x,t)=L:Y1-fi(fi-1i,t) (i=1,2, ... ,n-1). (67)
#i
We make
Assumption D.4 For every non-negative and non-zero 8* E Rn-1 and
almost all t E 11, there exists a single x E R(n-1)n such that L(x,t,8*) is a
minimum.
We notice that this assumption is satisfied if, for example, all functions
Ji(r/(t), t) are strictly concave in ii(t) (since then L in (65) is strictly
convex in x).
We will show
Proposition 1 Under Assumptions D.l, D.4, A.3, and A.5, problem (IT)
in model D has a solution which is continuous in t in each sub-unterval
defined in A.3.
The following proof depends heavily on the analysis in Ekeland and
Temam «1976, pp. 367-373). We write the Lagrangian of (P6) as
£ L(x(t), t, 8*)dt (68)
where L is defined in (65). By (D.3) we may assume that x(t) lies in a

compact set !C. Hence applying the measurable selection theorem (Ekeland
and Temam (1976), p. 236), we can find a measurable function 'Y(t, 8*) such
that
'Y(t,8*) = min{L(x,t,8)lx E!C} (69)
and
min £L(x(t), t, 8*)dt = £ 'Y(t, 8*)dt. (70)
We define x(t) by
L(x(t) , t, 8*) = 'Y(t, 8*)dt. (71)
It can be shown (Ekeland and Temam (1976), pp. 367-373) that 8* is
a sub-gradient of h(8), which is non-empty in the neighborhood of zero
because of (D.2). For fixed 8 (in particular, for 8 = 0), L(x,t,8*) is con-
tinuous in x and t. Hence, by the maximum theorem (see, e.g., Corollary
253
to Theorem 3 in B of Hildenbrand (1974)), x(t) is a non-empty and up-

per hemi-continuous set-valued mapping. Our uniqueness assumption (D.4)
then implies that x(t) (and hence, each fl(t)) is a continuous function.
Under the assumptions of the previous theorem, x( t) is continuous in each
of the sub-intervals of Ti . If these intervals are taken to be compact, x(t) is
a function of bounded variation (Dunford and Schwartz (1958)), and hence
is differentiable with respect to t almost everywhere in the sub-intervals.
This implies that under the assumptions of the theorem, we may assume
that all solution functions,yi(t) are differentiable almost everywhere. We
are thus in the realm of the ordinary theory of the calculus of variations,
and so the assumptions we made in model B are justified.
References
Arrow, K.J. and F.H. Hahn (1971): General Competitive Analysis, San
Francisco: Holden-Day.
Aumann, R.J. (1964): "Markets with a Continuum of Traders," Economet-
rica, 32, 39-50.
Aumann, R.J. (1975): "Values of Markets with a Continuum of Traders,"
Econometrica, 43, 611-647.
Aumann, R.J. and M. Perles (1965): "A Variational Problem Arising in
Economics," Journal of l'vfathematical Analysis and Applications, 11,488-
503.
Aumann, R.J. and 1. Shapley (1971): Values of Non-Atomic Games,
Princeton, NJ: Princeton University Press.
Berliocchi, H. and J.M. Lasry (1973): "Integrandes normales et measures
parametrees en calcul des variations," Bulletin de la Societe Mathema-
tique de Prance, 101, 129-184.
Bishop, R. (1967): "Monopolistic Competition and Welfare Economics", in
R.E. Kuenne (ed.), Monopolistic Competition Theory: Studies ,in Impact,
251-263.
Chamberlin, E.H. (1933): The Theory of Monopolostic Competition, Cam-
bridge, MA: Harvard University Press
Debreu, G. (1959): Theory of Value New York: John Wiley and Sons.
Ekeland, 1. and R. Temam (1976): Convex Analysis and Variational Prob-
lems, Amsterdam: North-Holland.
Gelfand, 1.M. and S.V. Fomin (1963): Calculus of Variations, Englewood
Cliffs, NJ: Prentice-Hall.
Hicks, J.R. (1939): "Foundation of Welfare Economics", Economic Journal,
49, 696-71l.
Hildenbrand, W. (1974): Core and Equilibria of A Large Economy, Prince-
ton, NJ: Princeton University Press.
Ichiishi, T. (1981): "Super-modularity: Applications to Convex Games and
to the Greedy Algorithm for LP", Journal of Economic Theory, 25, 283-
286.
254
Kahn, R.F. (1935): "Some Notes on Ideal Output", Economic Journal, 45,
1-35.
Makowski, L. (1980): "Perfect Competition, the Profit Criterion, and the
Organization of Economic Activity", Journal of Economic Theory, 22,
222-242.
Meade, J.E. (1937): Economic Analysis and Policy, Oxford University
Press.
McKenzie, L.W. (1951): "Ideal Putput and the Interdependence of Firms",
Economic Journal, 61, 785-803.
Negishi, T. (1962): "Entry and Optimal Number of Firms", Metroeconom-
ica, Vol. 14, 86-96.
Negishi, T. (1972): General Equilibrium Theory and International Trade,
Amsterdam: North-Holland.
Ostroy, J.M. (1980): "The No-Surplus Condition as a Characterization of
Perfectly Competitive Equilibrium", Journal of Economic Theory, 22,
183-207.
Samuelson, P.A.(1947): Foundations of Economic Analysis, Cambridge,
MA: Harvard University Press.
Shapley, L.S. (1953): "A Value for n-Person Games," in Contribution to the
Theory of Games Vol.II, ed. by H.W. Kuhn and A.W. Tucker, Princeton,
NJ: Princeton University Press.
Shapley, L.S. (1971): "Cores of Convex Games", International Journal of
Game Theory. 1, 11-26.
Suzumura, K. and K. Kiyono (1987): "Entry Barriers and Economic Wel-
fare", Review of Economic Studies, 54, 157-167.
Topkis, D.M. (1987): "Activity Optimization Games with Complementar-
ity", European Journal of Operations Research, 28, 358-368.
van Weizsacker, C.C. (1980). "A Welfare Analysis of Barriers to Entry",
Bell Journal of Economics, 12, 399-420.
On an Edgeworth characterization of rational
expectations equilibria in atomless asset
market economies'
Leonidas C. Koutsougeras
School of Economic Studies, University of Manchester, Room N.5.S, Dover

Street, Manchester M13 9PL, United Kingdom.
Abstract. We provide a definition of the rational expectations core in an

atom less economy with asset markets and show that allocations in the core
can be decentralized, via a system of asset and spot prices, as rational
expectations equilibria.
JEL Classification Numbers: D51
Keywords: Transient cooperation, asset markets, core
1 Introduction
The equivalence of the core with the set of competitive equilibria in purely
competitive economies provides a basic test of the price taking hypothesis
that underlies the definition of a competitive equilibrium. If such a result
failed to be true, then the Walrasian equilibrium would lose much of its
appeal as an analytical tool. Over the years there has been an extensive
literature on this subject that has provided a variety of results under alter-
native sets of assumptions. 1 In the context of asset markets, the price taking
hypothesis seems to be a fragile one. In such a framework, it is not clear
that deviations from price taking behavior provide negligible or no bene-
fit at all. For example, the Geanakoplos-Polemarchakis [4] sub-optimality
result suggests that if coalitions have a perception of their effect on spot
prices then they can (generically) find redistributions of assets so that the
effect on spot prices is to their advantage. We feel that in a sequential
framework, such as the asset market model, the price taking hypothesis
calls for some careful consideration, which would illuminate some subtle
'On different occasions I have benefited from discussions with R. Anderson, B. Gro-
dal, H. Hailer, M. Kaneko, J-F. Mertens, H. Polemarchakis, A. Rustichini and L. Zhou. I
would like to thank all of the above individuals for their insightful comments. Certainly,
I am responsible for any remaining errors.
lSee Anderson [lJ for a classification of alternative sets of assumptions and the asso-
ciated results.
256
issues that it entails.

The first task in this direction is to develop a core notion for dynamic
economies. Some basic references on the core in a dynamic framework
are Honkapohja [6J, Gale [3J and Repullo [13J. In the context of asset
market economies, the core has been studied in Koutsougeras [7J and in
Koutsougeras-Shafer [8J. It is evident in this literature that the develop-
ment of a time consistent core notion is not straightforward. In brief, the
problem that arises in the dynamic context is as follows. The notion of the
core as it has been formalized by Debreu and Scarf [2J involves blocking of
allocations, by groups of agents who trade commodities exclusively among
themselves. In the context of a dynamic economy with uncertainty this idea
entails coalitions that can enforce trades among its members, in all current
and future contingent markets at once. However, if the enforcability of fu-
ture trades is imperfect then it may not be possible to contract allocations
in the core of the contingent commodities economy ahead of time, as a di-
rect extension of the Debreu-Scarf definition would require. In such a case
one is forced to consider the possibility of transient cooperation. That is,
agents may form coalitions and cooperate in contingent markets where the
coalition can enforce trades, while trading with the rest of the economy in
all other markets. With the prospect of temporary cooperation the payoffs
that a coalition can guarantee to its members are no longer well defined
because they depend on the history of activities before the formation of the
coalition, as well as activities planned for the future beyond the horizon
of the coalition. Furthermore, blocking at any point in time introduces a
subordinate issue, namely, that blocking should involve beliefs about the
future trades that would follow as a consequence of a deviation.
In this paper we develop a core notion that embodies rationality of be-
liefs and provide an equivalence result with rational expectations equilibria
in the context of an atomless asset market economy. The key tool in the
formulation of rational expectations in the core of an asset markets econ-
omy, is the I-core introduced in Hammond et al [5J. The I-core is the set of
allocations that cannot be blocked by finite coalitions. It has been designed
for atomless economies with "widespread" externalities, namely economies
where the utility of each agent depends not only on his/her own allocation,
but also on the distribution of commodities throughout the economy. This
feature arises in our context in a natural way. Clearly, the core in each spot
economy depends on the distribution of endowments in that spot. How-
ever, in the presence of asset markets the endowments of agents in each
spot economy are their virtual endowments, i.e., their commodity holdings
after deliveries to asset claims are made. It is clear then, that the core in
each spot economy becomes a function of the distribution of assets through-
out the economy. Therefore, at each point in time there is an external effect
generated by the dependency of future spot trades on the current distrib-
ution of assets. In this way an atomless asset market economy provides a
natural environment for the I -core.
257
Besides its convenience in modeling a time consistent core notion, the use
of the f-core in the asset markets framework is appealing for conceptual
reasons as well. In that framework, spot prices and trades depend on the
distribution of activities in asset markets. Since coalitions of positive mea-
sure can affect the distribution of assets across the economy, they simply
cannot be negligible in the determination of future prices and trades. In
view of this fact, the only hope to justify price taking would be that the
effects of large coalitions on future prices and trades are not advantageous
to all their members. But this hope fails in most cases (generically), as
the constrained sub-optimality of rational expectations equilibria suggests.
Hence, we are left with the case of finite coalitions as the only alternative
to rationalize the price taking hypothesis.
Apart from its implications regarding the price taking hypothesis, our
equivalence result sheds light on the number of asset and spot markets
needed in an economy, so that coalitions can enforce any allocation in the
core of a contingent commodities economy. As one might suspect if there
were enough assets to enforce all possible future contingent deliveries then
individuals could arrange future deliveries at the beginning of time, in
such a way so that no coalition could block. In this case we would have
core allocations that correspond to a complete contingent market economy.
But how many assets are needed to achieve this? Clearly, if the span of
(real) assets with enforceable returns is at least equal to the number of
contingent commodities it would be possible to enforce any current and
future deliveries that cannot be blocked. But is this the minimal number
of assets required for this? The equivalence result that we provide here
answers this question.
We organize our approach as follows. In section two we develop the model
and the notation that we will use. In section three we introduce expectations
and we define the core of an economy with asset markets. Our main result
follows in section four and some concluding remarks in section five.
2 The model
In order to simplify the notation and the arguments, the results are devel-
oped in the context of a simple two period economy with assets which is
described as follows.
Let (A, A, JL) be a measure space where A is a Borel subset of a complete
separable metric space, denoting the set of agents. Uncertainty is described
by a finite set of states n. There are N commodity types available in each
=
state, so there are in totall N· n contingent commodities in the economy.
The consumption set of each agent is identified with ~l. Each individual
is characterized by a preference relation t,c ~l X ~l and a random initial
endowment ea E ~l. We impose the following assumptions on preferences:
• t, is strictly monotone, i.e., x ~y and x #y '* x >- y,

258
• J:: is C2 and strictly convex.

Let P be the space of preferences that satisfy the above assumptions,
endowed with the appropriate topology as in [11] (2, 2.6.7 A.3.4 pp. 96-
98). An economy is defined as a measurable mapping e : A -+ P X ]RI.
In order to hedge against the risks associated with the presence of uncer-
tainty, individuals are allowed to arrange future deliveries of commodities
by trading m (real) assets, available in zero net supply. An asset k is a vec-
tor (rk(w))WECl E ]R~, which specifies a promise for delivery of a bundle of
commodities in each state of nature. We refer to the collection R = {rk}k=l
as the asset structure. A portfolio of assets is represented2 by (J E ]Rm, with
the usual convention that if (J! < 0 ((J! > 0) then an agent a E A has a
claim (liability) on the k-th asset. In this way for a given asset allocation
(J : A -+]Rm, an individual is entitled to a dividend DOa = R· (Ja' An asset
market economy (GEl) can now be described as a pair {e,R}.
Remark 1 Notice that since in our discussion of the core there is no price
system involved, assets are means of enforcing deliveries of commodities
-rather than values- across states. In this framework, "incompleteness" of
the asset structure is understood to be in the sense that there may not be
enough assets to enforce all possible state contingent trades in the future.
For example, with any number of numeraire assets it will be impossible in
general to contract a core allocation ahead of time. In particular, if the
number of assets with linearly independent returns is less than the num-
ber contingent commodities -N . n- the asset structure does not allow for
complete span of the contingent consumption sets of individuals. In such a
case it is not clear at the outset, how many assets are needed to enforce an
allocation in the core of a full contingent markets economy. The analysis
that follows will illuminate this point.
Let L(A,]Rm) and L(A,]RI) denote the sets of measurable functions from
A to]Rm and ]RI respectively.3 An allocation for the asset market economy
defined above is a pair ((J,t) E L(A,]Rm) X L(A,]RI), so that fA (J(a) = 0
and fA t(a) = O. Note that, given an asset distribution, (J E L(A,]Rm) the
dividend mapping is a continuous function DOa : ]Rm -+ ]RI for each a E
A. Thus, the dividend mapping is measurable, viewed as a function of
A. In this way an allocation ((J, t) E L(A,]Rm) X L(A,]RI) gives rise to a
commodity allocation ea +Doa +ta E ]RI. A standard definition of a rational
expectations equilibrium for a GEl economy, that follows Radner [12], is
as follows:
Definition 1 A rational expectations equilibrium is a price system (q,p) E
]R+ x ]R~ and an allocation ((J, t) E L(A, ]Rm) x L(A, ]RI), such that
2In order to avoid confusion I will use Greek characters to denote asset holdings.
3Note that functions that differ on sets of measure zero are not identified.
259
For each a E A, [ea + D", + S >- ea + De", + taJ =>

[either q. 4J > 0 or p(w) . s(w) > 0 for some w E DJ.
According to the above definition, individuals are assumed to solve their
maximization problem, while taking current as well as expected prices as
given. In the next section we develop a core notion that characterizes such
equilibria.
3 The core
We start building our core notion for an asset market economy by for-
malizing expectations of future trades as follows. Let .c(A,Rm) denote the
quotient space of L(A,Rm), modulo sets of measure zero, i.e., the set of
equivalence classes of functions where two functions that differ on sets of
measure zero are identified. Spot trade expectations are given by a mapping
(1)
where for each 0 E .c(A, Rm), t(. )(0) is measurable on A.
The above specification has two important implications: First, that the
spot markets operate in an anonymous way, i.e., agents with the same
characteristics are assigned the same net trade. Second, that individual
actions in asset markets may have private but not social consequences in
spot markets, i.e, the net trade assignments in spot markets depend only
on the distribution of activities in asset markets. Both of those heuristic
assumptions draw from the hypothesis of a continuum of agents.
For a given asset distribution 0 E L(A,Rm), if an individual contemplates
holding a portfolio 4J E Rm then the final commodity allocation that this
agent expects is given by ea + D", + t~ (0). Notice that there is an external
effect on the expected consumption of each individual, which is due to the
dependency of (expected) spot trades on the asset distribution.
Given our specification of expectations we may view the way individuals
evaluate alternative portfolios as follows. A given 0 E L(A, Rm), induces a
preference relation ton Rm X Rm in a natural way
(2)
This induced preference relation inherits, for each 0 E L(A,Rm), the
following properties that will be useful to us in the sequel:
• t is strictly monotone, i.e., 4J ~ 't/J and 4J =I 't/J => 4J >- 't/J, 'V 4J, 't/J E
Rm.
• t is C2 and strictly convex.
260
• t is measurable on A.
•t is not affected by changes in the asset distribution on a null set of

agents, i.e., if O(a) = ((a) a.e.,then <P t 'IjJ {:} <P ~ 'IjJ.
This formulation of expectations suffices to discuss the core of our original

economy.
For a given asset distribution 0 E L(A,Rm) in E we associate the aug-
mented economy EJ(t e ), where each individual is characterized by the con-
sumption set Rm, the preference relation t and endowment 0 E Rm. We
give the following definition. 4
Definition 2 An asset distribution 0 E L(A,Rm) is in the core of the

augmented economy EJW),denoted by CJW), if
(i) i O(a) ::; 0
(ii) tI SeA and - Oa
aES
V aE S.
We also associate with our original economy the augmented economy

E} (0), where each individual is characterized by the consumption set RI,
the preference relation ~, and the virtual endowment ea + D(ja' For this
economy we have the following definition. 5
Definition 3 A spot trade t E L(A,RI) is in the core of the augmented

economy E}(O), denoted by C}(O), if
(i) fAt(a)::;O
(ii) V wEn, tI TeA, and s E RNITI, s.t. LaET Sa ::; 0 and
ea + D(ja + [t;:;-w, Sal >- ea + D(ja + ta V a E T.
where [t;:;-w, Sal denotes the net trade ta with the component ta(w) re-
placed by Sa'
Remark 2 There are conditions that guarantee the non emptiness of
CJW) and of Cj(O).
We are now ready to define the core for our original economy as follows.
4The careful reader will realize that this is the definition of the I-core applied to the
augmented economy eJ(t e ).
5 As before, this is the definition of the I-core applied to the augmented economy
ej(O).
261
Definition 4 An allocation (B,t) E L(A,Rm ) x L(A,R I ) is in the core of

the economy £ if (B,t) EC~W) xC}(B).
Notice that, according to this definition, it is possible that t!(B) =1= ta,
i.e., expectations need not be correct. The next definition requires that for
any given distribution of assets, anticipations of future trades are consistent
with the way the allocation mechanism works.
Definition 5 An allocation (B, t) E L(A, Rm) x L(A, RI) is in the mtional
expectations core of the economy £ if
(i) t!(B) = ta a.e.

(ii) (O,t) E C~W) x Cj(O).
With this definition we now proceed to the main result of this paper.
4 Equivalence
In this section we provide an argument in favor of the price taking hypoth-
esis in the model presented above, by proving that core allocations in the
sense of Definition 3 can be decentralized by a system of asset and spot
prices, as rational expectations equilibria.
Theorem 1 Let (B, t) E L(A, Rm) x L(A, RI) be an allocation in the m-
tional expectations core of the economy £. Then there exists a price system
(q,p) E Rm x R~ that supports (0, t) as a mtional expectations equilibrium.
Proof.
Consider the augmented economies £~W) and £j(O).
Step I. Since the induced preferences t on portfolios satisfy Properties 3.1
-3.4, Theorem 1 in [5] (pp.123) applies. In this way we can obtain q E Rm
such that
cp t O(a) => q. cp? 0 a.e.
From the feasibility of B, we have that q. Ba ~ 0 ae. Suppose now that
cp >- Ba and q. cp = o. Then by choosing .x < 1 sufficiently close to 1, so that
.xcp >- Ba, we obtain q . .xcp < q. cp = 0 which is a contradiction. Therefore,
we conclude that
cp >- Oa => q. cp > o. (3)
Step H. In each state WEn we have an Aumann type economy (except
that the consumption sets are not bounded below). The same theorem as
above applies, so we can obtain a price vector p(W) E R~ in each state
wEn so that
ea + Do" + [t;W,s] >- ea + Do" +ta => p(w)· s > 0, (4)

where [t~W, s] is the vector of net trades ta except in state w where the net
trade ta(w) has been replaced by s.
262
Step Ill. Suppose that there is a pair (cl>, s) E ~m X ~l such that for some
a E A we have that ea + Dc/> + s >- ea + Do .. + ta' By our assumptions on
preferences there is a unique hyperplane 7r E ~l through ea + Do .. + ta, so
that 7r' [ea + Dc/> + s] > 7r' [ea + Do .. +taJ, i.e., 7r' [Dc/> - Do .. + s - tal > O.
Therefore, it must be either 7r' [Dc/> - Do..] > 0 or 7r' [s - tal > o.
- If 7r' [Dc/> -- Do..] > 0 we can choose A > 0 sufficiently small, so that by
the uniqueness of 7r we obtain that ea + Do .. + ADc/>-o .. + ta >- ea + Do .. + ta'
Since the pair ((), t) is a rational expectations core allocation, we have that
ea + Do .. + ADc/>-o .. + t~((}) >- ea + Do .. + t~((}).
It follows then that (}a+A(cI>-(}a) >- (}a. Therefore, by (3) we conclude that
q. [(}a + A(cI> - (}a)] > 0, which implies that q. cl> > 0
-Ifontheotherhand7r·(s-ta ) > Oitmustbe7r(w)·(s(w)-ta(w)) > 0
for at least one wEn. Thus, 7r' [t;;-W, ta(W)+A(S(W)-ta(w)] > 7r·t a for each
A > O.It followsthenthat7r·(e a +Do .. +[t;;-W,ta (w)+A(s(w)-t a (w))]) > 7r'
(ea +Do .. +ta)' Using the uniqueness of 7r we have that for A sufficiently close
to zero it must be ea + Do .. + [t;;-W, ta(w) +A(S(W) -ta(w))] >- ea + Do .. +ta'
It follows from (4) that p(w)· [ta(w) + A(S(W) - ta(w))] > 0 which implies
that p(w) . s(w) > O.
Therefore, we conclude that for each a E A,
ea + Dc/> + s >- ea + Do .. + ta ~
either q. cl> > 0 or p(w) . s(w) > 0 for some wEn,

and this completes the proof of our theorem . •
Remark 3 In order to save unnecessary complications we have used a
setup where the consumption sets of agents are identified with the whole
of ~l. The usual definition of an economy and its corresponding rational
expectations equilibria involves ~~ as the consumption set. If one insists
on allocations to lie in ~~, then this can be guaranteed by assuming that
endowments are strictly positive and indifference surfaces through the en-
dowments do not intersect the axis. 6 Then the above proof still applies and
the individually rational region of the economy that contains all the alloca-
tions of interest would be contained in the interior of ~~.
Remark 4 As it is shown in Magill-Shafer [9J, if the number of assets with
linearly independent returns is greater than or equal to the number of states
then, generically, competitive allocations coincide with those of an economy
with complete contingent markets. Therefore, our result implies that the
core of an asset market economy coincides with that of an economy with
complete contingent commodity markets, if there are at least as many assets
6This assumption is standard in the existence of competitive equilibria in GEl models.

See for example [9).
263
with independent returns as there are states of nature. This correspondence

between the two models breaks down if the number of assets is less than the
number of states, and so does the correspondence between the cores of those
different economies. 7
5 Concluding remarks
In this paper we have proved an equivalence result that supports the price
taking hypothesis in GEl economies with a large number of agents. As
it is always the case, the equivalence of competitive equilibria with the
core reflects the premises of the core notion employed in the argument.
In developing our core notion we employed three premises: First, we in-
voked the usual rational expectations hypothesis, namely that individuals
can accurately predict future activities by observing current activities. This
hypothesis is natural since we wish to capture a rational expectations envi-
ronment. Second, we postulated that individuals do not believe that their
current activities can affect future (expected) trades. We do not regard this
hypothesis as unreasonable because in the context of an atomless economy
it follows from the previous hypothesis: future (expected) equilibrium ac-
tivities should depend on the distribution of current activities. Third, we
had to resort to the hypothesis that only 'small' coalitions can form at
any date. It should be emphasized that without this hypothesis the first
one does not make sense simply because there is no way to define rational
expectations. It is for this reason that we appealed to the f -core, a con-
struct which is based on the idea of 'small' coalitions, in order to develop
a time consistent core notion. As a consequence, our support of the price
taking hypothesis in GEl economies is conditional on whether or not large
coalitions are possible to form.
Apart from the problems in defining a time consistent core notion, the
size of blocking coalitions is crucial in the GEl context for yet another rea-
son. The effect of coalitions of positive measure on spot trades (or spot mar-
ket clearing prices) cannot be negligible. Furthermore, the sub-optimality
results in the GEl literature suggest that the grand coalition can affect
spot prices in a way that, generically, effects on future prices are to their
advantage. Therefore, if large coalitions (i.e., coalitions of positive mea-
sure) are indeed possible to form then we do not see how the price taking
hypothesis can be justified in this model. In conclusion, unless there is a
convincing reason to believe that only small coalitions can form, the price
taking hypothesis in GEl economies remains, to us, a delicate one.
Given that core equivalence is a necessary test of the price taking behav-
7Notice that this argument hinges on the equivalence result! In economies with a finite
number of agents, core allocations are not competitive in general. Thus, the argument is
invalid. In finite economies we are not aware of what is the minimum number of assets
needed for this equivalence to hold.
264
ior, the above considerations grant some importance to the investigation of

coalition formation in the study of GEl economies.
References
[1] Anderson, R (1991). "The Core in Perfectly Competitive Economies",
in Handbook of Game Theory, Vol.1, RJ. Aumann & S. Hart (eds) ,
Elsevier Science Publishers B.V., 1992.
[2] Debreu, G. & H. Scarf (1963). "A Limit Theorem on the Core of an
Economy", International Economic Review, 4, 236-246.
[3] Gale, D. (1978). "The Core of a Monetary Economy without Trust",
Journal of Economic Theory, 19, 456-49l.
[4] Geanakoplos, J. & H.M. Polemarchakis (1986). "Existence, Regularity
and Constrained Sub-optimality of Competitive Allocations when Mar-
kets are Incomplete", in Uncertainty, Information and Communication,
Essays in Honour of Kenneth Arrow, W.P. Heller, RM. Ross & D.A.
Starrett, eds., 3, Cambridge University Press, Cambridge.
[5] Hammond, P., M. Kaneko & M.H. Wooders (1989). "Continuum
Economies with Finite Coalitions: Core, Equilibria and Widespread Ex-
ternalities", Journal of Economic Theory, 44, 113-134.
[6] Honkapohja, S. (1977). "Money and the Core of a Sequence Economy
with Transactions Costs", European Economic Review, 10, 241-25l.
[7] Koutsougeras, L.C. (1995). "A Two Stage Core with Applications to
Asset Markets and Differential Information" Economies, Core Discus-
sion Paper No 9525.
[8] Koutsougeras, L.C & W.J. Shafer (1993). "The Rational Expectations
Core", manuscript University of Illinois.
[9] Magill, M. & W.J. Shafer (1991). "Incomplete Markets", in Handbook
of Mathematical Economics, W. Hildenbrand and H. Sonnenschein eds.,
IV, Elsevier Science Publishers B.V.
[10] Magill, M. & W.J. Shafer (1991). "Generically Complete Real Asset
Structures", Journal of Mathematical Economics, 19, 167-194.
[11] Mas-Collel, A. (1985). "The Theory of General Economic Equilibrium.
A Differentiable Approach", Cambridge University Press.
[12] Radner, R (1972). "Existence of Equilibrium Plans, Prices and Price
Expectations in a Sequence of Markets", Econometrica, 40,289-303.
[13] Repullo, R (1988). "The Core of an Economy with Transaction Costs" ,
Review of Economic Studies, LV, 447-458.
A characterization of the equal income
market equilibrium choice correspondence
based on average envy freeness'
Sombed Lahiri
Indian Institute of Management, Ahmedabad, 380 015 India
Abstract. A solution concept in the literature on fair division is the equal

income market equilibrium solution concept. We show that a solution which
satisfies consistency, replication invariance, efficiency, and envy freeness
must consist of equal income market equilibrium allocations. We offer a
further characterization of equal income market equilibrium with the strict
envy freeness property due to Zhou (1992) and without consistency.
JEL Classification Number: e71, D63

Keywords: Equal income market equilibrium, envy freeness, fair division
1 Introduction
In problems of fair division of a given bundle of resources, amongst a finite
number of agents, individual rationality from equal division plays a signif-
icant role. In a society, where all resources are socially owned, one cannot
argue in terms of equal ownership of the social endowment. One normally
takes the position that each agent has the right to veto any allocation,
which leaves him/her worse than equal division. Based on this premise,
individual rationality from equal division has been proposed as a minimal
requirement of distributive justice.
In Thomson (1982), we find an equity criterion called average envy-
freeness, which in the context of economies with convex preferences, implies
individual rationality from equal division. Average envy-freeness says that
no agent finds the average consumption of the other agents, superior to
his/her own consumption. This concept has been developed in the Foley
(1967) tradition of an envy-free allocation: no agent should find his/her
consumption inferior to the consumption of any other agent. We show in
*1 would like to thank Prof. Ahmet Alkan, for suggesting improvements, which have
made this paper more readable. The revision of this paper was undertaken while I was
visiting the Economic Research Unit, Indian Statistical Institute, Calcutta. I would like
to thank Satya Chakravarty and Diganta Mukherjee for suggestions and Rajyashree
Khushu-Lahiri for logistic suport, which made the revision possible.
266
this paper (the not too difficult result) that average envy-freeness does not
automatically imply individual rationality from equal division.
A solution concept which recurs with seeming regularity in the literature
of fair division is the equal income market equilibrium solution concept.
In a variable population framework Thomson (1988) provides an axiomatic
characterization, using the axiom of consistency. Consistency basically says
that the departure of some agents with their allocated consumption, should
not affect the consumption of the remaining agents, provided they operate
the same distribution mechanism as before. Lahiri (1997a, 1997b) uses this
same axiom to characterize the equal income market equilibrium choice
correspondence in convex and non-convex environments. Our main result
reported in this paper is similar to a Lahiri (1997b) result, although it
may not extend to the non-convex economies considered there. It is thus a
modest generalization of the Thomson (1988) result. We use consistency,
replication invariance, efficiency and average envy-freeness to show that if
a solution satisfies these properties, it must consist of equal income market
equilibrium allocations. Subsequently we drop consistency and arrive at
yet another characterization of subsolutions of equal income market equi-
librium choice correspondence using the strict envy-freeness property due
to Zhou (1992).
2 The model
Let lR denote the real line, lR+ the set of non-negative reals, and lR++ the
set of strictly positive reals. Let N denote the set of all strictly positive
integers. Given cp =I- Z c lR and Q any non-empty finite set, let ZQ denote
the set of all functions from Q to Z.
Let P be the collection of all non-empty finite sets Q, where to avoid
Russell's paradox, we assume that the statements "Q belongs to Q" and
"Q does not belong to Q" have no meaning. Given Q E P and kEN, let
Qk be the set Q X {1, ... , k}. Clearly Qk E P.
Let there be L 2: 2, (L E N) infinitely divisible goods in the economy.
The commodity space is lRL and the consumption set of any conceivable
agent (consumer) is lR~. Any Q( E P), is an agent set.
An economy E is a pair < (Ui)iEQ' w > satisfying the following proper-
ties:
(1) Q E P, is the agent set.
(2) w E lR~+, is the aggregate social endowment.
(3) Vi E Q, U i : lR~ ~ lR is a continuous and weakly increasing (Le.,
Xi, yi E lR~, Xi > > yi implies U i (Xi) > U i (yi)) utility function,
which is continuously differentiable in lR~+.
(4) Vi E Q, Xi, yi E lR~, Xi E lR~+ and Ui(Xi)
. L
= Ui(yi) implies
yt E lR++.
267
(5) Vi E Q, Ui : R~ --t R is semi-strictly quasi-concave (i.e. Xi, yi E

R~, t E (0,1), Ui(Xi) > Ui(yi) implies Ui(tXi + (1 - t)yi) >
U'(Y') ).
Let ~ be the set of all economies satisfying the above properties. Given
E =< (Ui)iEQ' W >, let A(E) = {(Xi)iEQ E (R~)Q j LiEQ Xi = w}.
A(E) is the set of all feasible allocations for E.
Given E as above a feasible allocation (Xi)iEQ is said to be Pareto Opti-
mal, ifthere does not exist (yi)iEQ E A(E) such that Ui(yi) > Ui(Xi) Vi E
Q. [This is actually the definition of Weak Pareto Optimality; however it
is easy to show that for our kind of economies Weak Pareto Optimal allo-
cations and Pareto Optimal allocations coincide.] Let P(E) denote the set
of all Pareto Optimal allocations. It is easy to show that P(E) -I- rp.
Given E as above let IR(E) = {(Xi)iEQ E A(E)jUi(Xi) ~
Ui (IQI) ViE Q}. IR(E) is the set of all allocations for E which are
individually rational from equal division.
Given E as above and kEN, Ek denotes the kth replica of E where Ek
is defined as follows:
Ek =< (Ui,j)(i,j)EQk,kw > where V(i,j) E Qk,Ui,j = U i .
Given E E ~ and (Xi)iEQ E A(E) (where Q is the agent set for E) and
kEN, let (y(i,j))(i,j)EQI be defined as y(i,j) = XiV(i,j) E Qk.
Given E E ~, a feasible allocation or)iEQ is said to be a price equilib-
rium if there exists p E R~\ {O} such that Vi E Q, Xi solves
s.t. p.Xi::; p.X', Xi E R~.

The following result is standard.
Proposition 1 Given E E ~
(i) every price equilibrium allocation is Pareto Optimal;

(ii) every Pareto Optimal allocation is a price equilibrium allocation
with respect to a unique price p E R~ \ {O}.
Let V c ~ be given.
A choice correspondence on V is a non-empty valued correspondence
F: V ~ UQEP(Rt)Q such that VE E V, F(E) c A(E).
A choice correspondence F on V is said to be efficient if F(E) c P(E)

VE E V; individually rational (from equal division) if F(E) c IR(E) VE E
V; consistent if VE =< (Ui)iEQ'W >E V, Vrp -I- M c Q, V(Xi)iEQ E
268
F(E), E' =< (Ui)iEM' LiEM Xi >E V implies (Xi)iEM E F(E')j repli-
cation invariant if "lE =< (Ui)iEQ,W >E V, Vk E N, (Xi)iEQ E F(E)
implies (y(i,j»)(i,j)EQk E F(Ek) provided Ek E Vj average envy-free if
"lE =< (Ui)iEQ,W >E V, Vi E Q,S = Q\{i} and (Xi)iEQ E F(E), we
have Ui(Xi) 2:: Ui (fsr LjES Xj) j and strictly envy-free if this holds for
all non-empty Se Q/{i}.
Observation 1. If F is individually rational from equal division, then F is

average envy-free. This result has been noted in Thomson [1982]. However,
there exist economies such that an allocation is averge envy-free but is not
individually rational from equal division.
Example. Let L =
2,Q = {1,2}, U1(XI,Xl) = xIXi, U2 (Xr,Xi)
= XrXi(Xr + Xi), for all (XLXi),(Xr,xi) E 1R!. Here Xj is amount
of commodity j consumed by agent i. Let W = (6 - y'5, 0J- 3 + 3-\75)'
(XL Xi) = (3 - y'5, 3-\75), (Xr, Xi) = (3, 0J-3). Neither agent envies
the other. Yet the allocation violates individual rationality from equal di-
vision (for both the agents). Note that, for two agent economies, every free
allocations and average every free allocations coincide.
Observation 2. Strict envy-freeness is a property due to Zhou [1992].

As observed in Thomson [1994], if we have a choice correspondence which
satisfies consistency and average envy-freeness then it is also strictly envy-
free.
We now define equal income market equilibrium choice correspondence

G as follows:
Let E =< (Ui)iEQ'W >E ~.(Xi)iEQ E A(E) is said to be an equal

income market equilibrium allocation if there exists p E IRt \ {O} such that
Vi E Q, ~ solves
.
s.t. p.X·:S 1, x·. E 1R+.
L
The following proposition is standard.
Proposition 2 Every E E ~ possesses an equal income market equilibrium

allocation.
In view of the above we can define G(E) to be the set of equal income
market equilibrium allocations for E E ~. G is called the equal income
market equilibrium choice correspondence. Further given
269
implies
Xi E R~+ Vi E Q.
The following proposition is easily verified:
Proposition 3: G satisfies Efficiency, Average Envy-Freeness, Strict Envy-

Freeness, Consistency and Replication Invariance.
3 The main results

We now present two variations of a theorem due to Thomson [1988].
Theorem 1 Let F be a Choice Correspondence on :E which satisfies Ef-

ficiency, Average Envy Freeness, Consistency and Replication Invariance.
Then
'liE E :E, F(E) c G(E).
Proof. Assume the conditions of the theorem for F. Let E =< (Ui)iEQ,
w > E :E and towards a contradiction assume,
-=i
(X )iEQ E F(E)\G(E).
By Proposition 1 and since F(E) C P(E), there exists p E R~\{O} such

that Vi E Q, T solves
. -=i . L
s.t. p.X~:S p.X , X~ E R+.
Without loss of generality, we may assume p.w = IQI. We would be done
if we could show p.T :s
fmVi E Q. Assume p.Y > fm for some i E Q.
. . -j -j-=i
Thus there eXIsts J E Q such that p.X < fm. Thus p.X < p.X . By the
smoothness assumption on preferences, there exists X E (0, 1) sufficiently
small so that Uj(X j + A(Y - Xi)) > Uj(X j ) VA E (0, X). Choose kEN
such that l!k < Xand consider Ek+1.
Let y(n,m) = r V( n, m) E Qk+l. By Replication Invariance
Let 5 C Qk+l be defined as follows:
5' = {( i, I)} U {(j, m) : m = 1, ... , k}
and 5 = 5' u {(j, k + I)}

Let E' =< (un)nES, (k + l)X i + T > .
270
By consistency, ((y(j,m»)::l' ~) E F(E). But kt:?{i = Xj +

k!l (~ - X j
) is the average consumption of agents in S'. Hence
uj (k~ :,X') > Uj(X j ) = Ui(yij,k+1))
This contradicts Average Envy-Freeness and proves the theorem. •

As a corollary to the above theorem it easily follows that the largest
choice correspondence which satisfies Efficiency, Average Envy-Freeness,
Consistency and Replication Invariance is G.
In the next theorem we replace consistency and average envy freeness by
strict envy freeness to obtain the following result:
Theorem 2 Let F be a choice correspondence on ~ which satisfies Ef-

ficiency, Strict Envy Freeness and Replication Invariance. Then VE E
~,F(E) c G(E).
Proof. The proof proceeds exactly as in the proof of Theorem 1, upto the
construction of S'. Then we skip the construction of E' and the reference
to consistency. From there on the analysis is once again the same as before,
except that the result now contradicts strict envy-freeness ((j, k + 1) envies
S'). •
We are thus able to characterize subcorrespondences of the equal income
market equilibrium correspondence in a variable population framework,
without appealing to consistency.
References
D. Foley [1967]: "Resource Allocation in the Public Sector," Yale Economic
Essays.
S. Lahiri [1997a]: "Axiomatic Characterization Of Market Equilibrium So-
lutions In Problems Of Fair Division," Mimeo.
S. Lahiri [1997b]: "Axiomatic Characterization Of The Equal Income Mar-
ket Equilibrium Choice Correspondence In Non-Convex Economics,"
Mimeo.
W. Thomson [1982]: "An Informationally Efficient Equity Criterion," Jour-
nal of Public Economics 18, 243-263.
W. Thomson [1988]: "A Study Of Choice Correspondences In Economies
With A Variable Number Of Agents," Journal of Economic Theory, Vol.
46, Pages 237-254.
W. Thomson [1994]: "Consistent Extensions," Mathematical Social Sci-
ences, Vol, 28, pages 35-49.
L. Zhou [1992]: "Strictly Fair Allocations in Large Exchange Economies,"
Journal of Economic Theory, 57, 158-175.
Non-keynesian effects of fiscal contractions:
theory and applications for Germany'
Bernd Lucke
Freie Universitiit Berlin, Boltzmannstr. 20 D-14195 Berlin, Germany
Abstract. Non-keynesian effects of fiscal contractions have attracted the

attention of economists throughout the 1980s. Most notably, cuts in gov-
ernment spending in Ireland, Denmark and Germany are known to have
coincided with increases in private consumption spending. Self-fulfilling ex-
pectations about the effects of stabilization policies may explain these and
a diversity of opposing experiences. This paper formulates a dynamic st~
chastic equilibrium model with equilibrium indeterminacy, which allows for
sunspot fluctuations as well as for systematic consumption effects of gov-
ernment policy in either direction. Cointegration tests and Euler equation
estimates suggest that the model is approximately in accord with Ger-
man data. Parameter estimates imply that the non-keynesian experiences
are not due to self-fulfilling expectations but to the productivity effects of
government-provided infrastructure services.
JEL Classification Number: El, E62
Keywords: Fiscal contractions, self-fulfilling expectations, equilibrium in-

determinacy, cointegration tests
1 Introduction
A number of European countries have abandoned the road of Keynesian
economic policies in the 1980s, basically on the grounds of disillusion about
the effects of deficit spending policies in the 1970s. Giavazzi and Pagano
(1990) illustrate the turnaround in public spending programs for major
OECD countries. They point out that the effects of fiscal contractions vary
substantially in cross-country samples, with Denmark (1983-1984) and Ire-
land (1987-1989) the front-runners on the non-keynesian side, i.e., expe-
riencing positive private consumption growth along with cuts in public
'Paper presented at the Society for the Advancement of Economic Theory, "Economic
Theory and Applications", Antalya, Turkey, 1997, and at the European Meeting of
the Econometrics Society, Toulouse, France, 1997. I thank Prof. JUrgen Wolters, an
anonymous referee, and seminar participants for helpful comments on earlier versions.
All remaining errors are mine.
272
spending, and Ireland (!) (1981-1984) and Spain (1986-1989) the leading
examples of keynesian effects, where public spending and private consump-
tion are positively correlated.
As a possible explanation Giavazzi and Pagano refer to a "German view"
of government stabilization policies as articulated by Hellwig und Neumann
(1987), who state that
"the direct impact of slower public expenditure growth is clearly nega-
tive... The indirect effect on aggregate demand of the initial reduction in
expenditure growth occurs through an improvement in expectations if the
measures taken are understood to be part of a credible medium-run progmm
of consolidation, designed to permanently reduce the share of government
in GDP... !and thus] taxation in the future. "
According to this view, benign effects of fiscal consolidation rely on the
perceived persistence of public spending cuts. If government spending is
expected to be permanently lower, then permanent income rises and the
thereby induced rise in private consumption may outweigh the depressing
effects of elementary keynesian multiplier mechanics. If, however, govern-
ment spending cuts are not long-run credible, then the latter effects domi-
nate and hence keynesian results emerge. Giavazzi and Pagano (1990, 1996)
argue that the credibility of fiscal contractions may rise with the severity
of the measures taken, i.e., large spending cuts are more likely to result in
increased private consumption than small.
Giavazzi and Pagano do not analyze the issue in a structural model; their
approach is entirely confined to reduced forms. In an on first sight unre-
lated line of research, real business cycle (RBC) analysts, however, have
established results that might prove helpful in explaining non-keynesian
phenomena of the said type. For a major focus of RBC theory has recently
been on models with equilibrium indeterminacy, see, e.g., Benhabib and
Farmer (1994) and Farmer and Guo (1994). In such models convergence
paths to the steady state are not unique, such that expectations, good
or bad, can be self-fulfilling. Moreover, a number of authors, e.g., Farmer
and Guo (1995) have presented econometric results that suggest an impor-
tant role for self-fulfilling expectations in macroeconomic fluctuations. The
central question to be addressed in this paper is hence whether the unsys-
tematic pattern of responses to fiscal consolidation efforts (and in particular
the non-keynesian experience found in German data) may be attributable
to multiple equilibrium paths.
For instance, assume that people increase their consumption because
they believe that a certain government policy will raise their permanent
incomes, while in fact the government policy does not have real effects at
all. In a standard RBC-model people will quickly realize that income does
not respond to the government policy and thus consumption plans will
be revised downwards. In an RBC-model with equilibrium indeterminacy,
however, the initial rise in consumption will cause an increase in income
which rectifies the initial beliefs, i.e., expectations are self-fulfilling. An
273
economy of this type is said to exhibit sunspot fluctuations, i.e., fluctua-

tions that do not depend on shocks to the fundamentals of the economy. 1
Needless to say, shocks in fundamentals, such as government policies with
real effects, may be amplified by sunspot fluctuations.
In order to analyze the scope for non-keynesian effects of fiscal contrac-
tions in the framework of RBC models with potential for equilibrium inde-
terminacy, I postulate the existence of production externalities which give
rise to increasing returns on the aggregate leve1. 2 This contributes to a novel
feature of the model as opposed to existing RBC models with equilibrium
indeterminacy: I assume the government to use tax proceeds for the produc-
tion of infrastructure services complementary to private production factors.
The assumption serves three purposes: First, it motivates the existence of
increasing returns in the aggregate despite constant returns for the indi-
vidual producer, and second, it ensures a more realistic role of government
activity than in standard RBC-models, where government expenditure is
often assumed to yield zero utility, cf. Christiano and Eichenbaum (1992)
or Johnson and Klein (1997).3 Third, however, and most importantly, it
opens two channels for government absorption to affect the productivity of
the private sector.
On the expenditure side, a fiscal contraction reduces infrastructure ser-
vices to private firms, hence productivity and permanent income tend to
decline. On the revenue side, however, lower tax rates signal increased per-
manent disposable income and thus stimulate consumption and private
capital accumulation. The net effect on income depends on the relative
strength of the productivity responses: If the marginal productivity of pri-
vate capital is larger than the marginal productivity of public infrastructure
services, then the fiscal consolidation will lead to increased permanent in-
come and private consumption.
Private consumption may thus be stimulated by two very distinct mech-
anisms: Self-fulfilling sunspot fluctuations irrespective of fundamentals on
lClearly, self-fulfilling expectations may have a negative impact on consumption if

the initial position of the public is pessimistic. Since sunspot models have the property
that the same fundamentals can give rise to different outcomes, they have a potential for
explaining the observed heterogeneity of responses to fiscal contractions in international
cross sections.
2Sunspot fluctuations can be analyzed in any kind of dynamic stochastic equilibrium
model. The model used in this paper is, in fact, slightly more general than the proto-
typical RBC model, since it allows for three stochastic trends that correspond to three
real shocks: Besides a productivity shock I model a tax rate shock and a net export
shock. I show that three stochastic trends is actually what the data require. Like most
RBC models, I do not explicitly model nominal effects on the real sector, although some
nominal influences might implicitly be included in the real shocks.
3The standard setup in which the only effect of government activity is a waste of re-
sources is inconsistent with rational political economy considerations: Why would people
vote for a government that is costly and useless?
274
the one hand and systematic productivity effects of fiscal consolidation

programs on the other hand. From an inferential point of view this is a
necessary design, since a valid test of the null hypothesis that private con-
sumption is driven by sunspot fluctuations is possible only if the model
provides an alternative hypothesis able to explain observed consumption
responses.
However, upward sloping labor demand is not sufficient for equilibrium
indeterminacy, and the same is true for increasing returns to production.
Rather, equilibrium indeterminacy implies further restrictions on the deep
parameters. I derive these following Wen's (1996) insightful analysis. I also
derive the parameter conditions under which changes in government spend-
ing systematically increase or decrease private consumption. Depending on
parameter magnitudes, the model thus allows for keynesian or for non-
keynesian effects of fiscal contractions and either sort of effect can be due
to systematic or to sunspot sources.
On the empirical side, I basically follow the estimation strategy laid out
by Farmer and Guo (1995). I use German quarterly national accounts data
to estimate an Euler-equation by instrumental variables methods. In addi-
tion, I derive the long run equilibrium properties from the Euler-equations
and check model adequacy by Johansen's (1988, 1995) method. Tests for
the rank of the cointegration space as well as parameter estimates from
the cointegrating vectors seem to indicate that the model is roughly in
accord with the data. Based on the parameter estimates obtained I reject
the possibility of non-keynesian sunspot fluctuations. Rather, my estimates
suggest that non-keynesian effects of Germany's fiscal contraction in the
1980s emerged from systematic sources, in particular from a marginal ben-
efit of infrastructure considerably lower than the marginal productivity of
private capital.
The rest of the paper is organized as follows: The theoretical model is
introduced in Section 2. Section 3 dicusses the data and derives prelimi-
nary parameter estimates using Hodrick-Prescott-filtered data. In Section
4 Johansen tests for cointegration and parameter estimates for an Euler
equation in error correction form are presented. Section 5 contains a sensi-
tivity analysis and Section 6 concludes.
2 The model
Suppose an economy is composed of many identical agents who maximize
expected utility under a number of constraints to be described below. Util-
ity depends positively on consumption etand negatively on hours worked
Lt. The capital stock Kt is predetermined so that the objective is to solve
275
where 0 < b < 1 is a discount factor and momentary utility u is defined

in consumption and labor units scaled by a growth component X t • Scaling
hours in such a way enables us to account for the negative growth trend
found in German hours per capita. Similar preferences have been specified
by Greenwood, Hercowitz and Huffman (1988) as well as Correia, Neves,
and Rebelo (1995); the standard case of stationary hours is retrieved by
setting 1/ = O. Note that we will expect 1/ to be negative in the case of
negatively trended hours, i.e., for conventional momentary utility functions
the marginal disutility of labor increases with the growth component.
Note that 1/7jJ is the intertemporal elasticity of substitution in labor.
The model thus includes the Hansen (1985)-Rogerson (1988) case of indi-
visible labor, since utility is linear in labor for 7jJ = 0, i.e., for an infinite
intertemporal elasticity of substitution.
Government expenditure G t is used to produce public infrastructure St
under a simple decreasing returns to scale production function
(2)
o < <P < 1. Public infrastructure is complementary to private production

factors and thus not only government investment but also government con-
sumption expenditure is counted as input in the production of infrastruc-
ture, since typical components of government expenditure such as public
education, enforcement of the law, and external security are clearly sup-
portive of private production. This view of government activity follows an
old line of reasoning in the public finance literature, see e.g., Littmann
(1957).4
Assuming Ricardian equivalence, see Seater (1993), all government ex-
penditure is tax financed. The tax proceeds are modelled as a simple linear
function of current output yt:
(3)
The log of the tax rate In Tt is assumed to be a random walk without drift
and with starting value In T at some distant point in the past 5
4Littmann divides total government expenses into government absorption (our at)
and transfers. Absorption consists of government production and government consump-
tion, where the latter is very narrowly defined and basically comprises expenditures for
representational purposes. While, admittedly, one might argue that not all government
absorption is used to provide public infrastructure, one might, on the other hand, point
out that even transfers are benefitial for private production. The German social security
system, for instance, is often credited for the relative absence of strikes and distributive
quarrels in the German society. Thus, under a wide definition of public infrastructure,
government infrastructure investment might even be larger than government absorption.
5Clearly, a tax rate cannot grow without bounds. The 1(1) assumption can still be a
reasonable approximation in finite samples.
276
Private production takes place under a Cobb-Douglas production func-

tion
(4)
where 0 < 0: < 1. Here production depends on the growth component X t ,
which is labor augmenting, on the level of public infrastructure St, and a
stationary technology shock At assumed to be 1 in the steady state. Since
the economy is inhabitated by many agents, the individual producer takes
St as given and does not recognize any link between St and his own actions.
In the aggregate, however, substituting (2) and (3) into (4) and solving for
per-capita output yields
(5)
hence in the aggregate the economy displays increasing returns to scale.

Net exports NX t are assumed to depend linearly on output,
(6)
where In Zt := In(l - Xt) is a random walk without drift and with starting
value zero at some distant point in the past. The law of motion for the
capital stock, finally, is given by
Kt = (1- 6)Kt- 1 + It-I, (7)

where It is private investment and 0 < 8 < 1 is a constant rate of depreci-
ation.
Under these conditions, the resource constraint can be written as
(8)
where Zt := 1 - Xt.
Solving (1) with respect to (4), (7), and (8) yields the following Euler-
equations:
CtL}+tP = 0: (Zt - Tt) ytX;(1+tP) (9)
Et [CHI] = Et [.BCt ((1- 6) + (Zt+1 - Tt+1) (1- 0:) ~:~) ] . (10)
To solve the model I use the technique proposed by King, Plosser and
Rebelo (1988). Specifying the growth process as
InXt+1 = In"Yx + InXt + €x,t+1, (11)
we can compute the steady state of the deterministic version of the economy
(obtained by setting the variances of all disturbances equal to zero). It is
a straightforward exercise to show that in the steady state hours Lt grow
277
at rate 1/ In~(x, while consumption, investment, government expenditure,

output and capital stock grow at rate In" where , is defined. as
,·-,x._ a(1+v)/(a-1»
.
Note that unless 1/ = 0, log hours are integrated of order one and coin-
tegrate with other model variables in a way to be derived below.
I derive a non-growing economy by dividing each variable by its growth
component. Denoting the growth-corrected variables by small letters we
have
Lt lit
it:= Xv' Yt:= a(l+v)/(a-1» ,
t Xt
etc. Substituting (5) into (9) and writing this in terms of the transformed
variables W(3 get
(l-q,)P+W) ~ 1 (l-dll(l+w) (l-n:)(l+'lIJ)
Ct=o:A t a (Zt- Tt)1't '" Yt a kt a • (12)
Similarly, (10) and (8) can be rewritten as
Et [ct+ll = Et [ ~~t
, exp €t+l
} ((1 - 8) + (Zt+l - Tt+d (1 - 0:) kYt +1 )]
HI
,
(13)
ct +,exp {€Hd kt+l - (1 - 8)kt = (Zt - Tt)Yt, (14)

respectively. In (13) and (14), I use the short-hand notation
Et := 0: (1 + 1/) /(0: - <p)€X,t.
Setting €t == 0 for all t and dropping time indices to denote (determinis-

tic) steady-state-values, (12), (13) and (14) can be solved for steady-state
output, consumption, and capital:
y=
[-0
T~ ((1- T)(l- 0:) ,B) a::-; ( 0: b - ~ (1- 8)) ) Ca

, - ~ (1 - 8) (1 - ~h + o:~ b - (1 - 8))
C = (1 - ~h + o:~ b - (1 - 8)) (1 _ T) Y
"'( - ~(1- 8)
(1 - 0:) ,B
k = ,_ ~ (1 _ 8) (1 - T) y.
Assuming that 0: > <p, a realistic assumption, and differentiating c with

respect to T, it is easy to see that steady-state consumption is maximized
278
for T = cp. Thus, if Tt < cp, a tax increase will have a positive systematic ef-
fect on consumption, since the marginal benefit of increasing infrastructure
outweighs the marginal tax burden, i.e., the marginal loss of private capital
productivity. However, if Tt > cp, shrinking government expenditures (and
taxes) will tend to increase private consumption.
To solve equations (12) - (14) analytically, I compute a log-linear Taylor
approximation of the system about its steady state. I use hatted variables
to denote percent deviations from steady-state-Ievels of the non-growing
economy, i.e.,
Yt := InYt -lny, Ct:= lnct -lnc,
etc. The log-linear analogues of (12),(13) and (14) are then given by
Ct = _1_ Zt + (CP(I+'Ij;) -~)Tt

I-T a I-T
+ (1 _ (1 - cp );1 + 'Ij;)) Yt
(l-a)(I+'Ij;)kA (l-cp)(I+'Ij;)AA
+ a
t + a
t (15)
(16)
(1 - (3) 'Y + a{3 ('Y - (1 - 8)) c + (k

'V +E ) _ (1 _ 8) k
(1 - a) {3 t I HI HI t
(17)
Here UHI is a zero mean error term which combines EHI and the Euler
equation error incurred by replacing expected values by their realizations.
Solving (14) for Yt and substituting into (15) and (16) yields a system
of two first order difference equations in the endogenous variables Ct and
kt as well as the forcing variables Zt and Tt. The homogenous part of this
system is of the general form
( all
o
a12) (
a22 kHI
~+I
) = (bll 0) ( ~t
~I b22 kt
) .
" 'V' ., " V .,

:=A :=B
279
Thus the stability properties of the economy depend on the eigenvalues of

the matrix D := A-lB. In standard RBC-models one eigenvalue of D is
larger and the other smaller than unity in absolute value. In this case a
stable solution to the difference equation exists for a linear subspace of the
initial conditions. This means that for one starting value given (L e. the
predetermined value of ko) the other starting value (Le., CO) is uniquely
determined by requiring stability of the solution path. This is the so-called
saddle-path stability of RBC-models.
If the eigenvalues of D were both smaller than unity in absolute value, the
saddle becomes a sink, meaning that for any pair of starting values the econ-
omy converges to the steady state. Since only ko is predetermined, CO can,
in this case, assume any value. It can in particular be driven by sunspot ex-
pectations, which will necessarily be self-fulfilling, since any sunspot shock
in consumption leads the economy on a sustainable convergent path to the
steady state. The economy then displays equilibrium indeterminacy.
The algebra of the above difference equation is quite messy and I will
therefore refrain from computing the eigenvalues of D in terms of the struc-
tural parameters. Rather, I follow Wen (1996), who points out that a neces-
sary condition for D to have both eigenvalues smaller than one in absolute
value is that the determinant of D is smaller than one in absolute value.
(Clearly, this condition is not sufficient.) The determinant of D can be
derived as
1 [ 4>(1+"p)('y-,6(1-6))] 1
det D = 73 1 - a,6 (1 _ 6) -
'Y (1 - 4» (1 +"p) =: 73 [1 - xl·
This expression is smaller than one only if X is positive. Thus one ob-
vious necessary condition for equilibrium indeterminacy is 4> > 0, i.e., the
existence of increasing returns to scale in aggregate private production. A
further necessary condition for sunspot fluctuations is "p > -1, for if 'l/J
were smaller than -1, then the numerator of X would be negative and the
denominator positive, Le., c would be negative. These conditions will be
helpful in the empirical analysis below.
3 Data and preliminary estimates

Quarterly data for the main economic aggregates of the West German econ-
omy are available from 1960 to 1994, since the system of national accounts
had fictitiously separated West and East Germany from 1990 until the
end of 1994. However, data prior to 1975 do not lend themselves easily
to an analysis of the kind intended here, since in this earlier sample the
investment share in GDP is not constant but higher than today and grad-
ually decreasing. Likewise, the consumption share is lower than today and
gradually increasing. These phenomena are probably due to accumulating
productive capital after the severe losses incurred in World War 11. Thus
the data do not allow for a steady state interpretation prior to 1975 and
280
therefore I will use a sample of 80 observations ranging from 1975.1 to

1994.4.
Further, the share of government absorption in GDP is negatively
trended in this sample, while government transfers (as a percentage of
GDP) have a (roughly offsetting) positive trend. The share of net exports
in GDP, incidentally, has roughly the same positive trend. Prior to 1990
the reason for this can at least partially be traced to the rising importance
of transfers to the European Union: First, these transfers somewhat limited
the scope for government absorption in Germany. Second, however, in as
much as the European Union used these transfers for absorption of German
goods (for instance by intervening in agricultural markets), the transfers
artificially increased German exports. Clearly, such exports should actually
be counted as government absorption as well.
In 1990 and the following years, the German government transferred huge
amounts of money to East Germany. These were counted as transfers to
the rest of the world in the system of national accounts that still separated
both parts of Germany. To the extent to which East German regional gov-
ernments used these transfers to buy godds and services in West Germany,
West German exports increased. Again, it would be better to count these
exports as government absorption.
There is no easy correction for these distorting developments in the sys-
tem of national accounts. As an admittedly crude correction I have added
net government transfers to the rest of the world to government absorption
at home and subtracted these transfers from net exports. Better meth-
ods may exist, but the one I have chosen serves its purposes well, since in
their new definitions government absorption and net exports have roughly
constant shares in GDP.
I have used (seasonally adjusted) data from the OECD Quarterly Na-
tional Accounts, since the OECD also provides capital stock data, although
only in a yearly periodicity. I have used the data on quarterly private invest-
ment to construct quarterly capital stock data, assuming that each year's
depreciation is distributed evenly on the four quarters. Since the RBC-
model is of a representative agent type, I divide all series by a population
series provided by the Deutsches Institut fiir Wirtschaftsforschung. 6 The
latter institution also provided (seasonally adjusted) data on hours worked
and on the real wage.
We are now interested in estimation of the parameters qy and 'ljJ. If we
had measures for the percent deviations from the steady state, then the
easiest way would be to estimate the linearized versions of equations (5)
and (9)
(18)
6 The series used is total population ("Wohllbevnlkerung").

281
~ ~ 1 ~ T,y,
Ct-Yt=--zt+--.Lt+ (1 -'f/
.,.) l~t (19)
I-T I-T
But exact measures of the percent deviations are difficult to obtain, since
the random walk growth component is not observable. Proxy variables can
be derived, however, by applying a Hodrick-Prescott (HP) filter to the
logarithms of the level variables and interpreting the deviations from the
Hodrick-Prescott trends as percentage deviations. While estimation of (18)
and (19) with HP-filtered variables is not fully efficient in view of the fact
that the long run information is filtered out, this approach readily provides
a preliminary impression of the magnitudes of the key parameters and
complements the estimation methods implemented below.
I estimate the linearized production function (18) by two-stage nonlin-
ear least squares, assuming the technology shock to be AR(I). The Euler
equation (19) is estimated by GMM in order to exploit the orthogonality
restrictions imposed by the model. In both cases lagged values of depen-
dent and independent variables serve as instruments. For the production
function we obtain 1> = 0.15 (standard error 0.05), and et = 0.78(0.20). For
the Euler equation I fix T = 0.25 which is the long run average share of
government absorption in the sample and estimate 7jJ = -1.10 (0.28). All
these values are significant and of reasonable magnitude. If we identified the
point estimates with the true parameter values, the estimates would imply
that there is no scope for sunspot fluctuations (7jJ < -1) but an excessive
tax burden (T > 1» which would give rise to systematic non-keynesian
effects of fiscal contractions due to unbalanced marginal productivities of
infrastructure services and private capital.
4 Cointegration tests and error correction estimates

Two objections can be raised against the above estimates: First, inducing
stationarity by HP-filtering is completely ad hoc and second, all long-run
information is lost under this filter. It is therefore preferable to work di-
rectly with the instationary level variables and to derive the cointegration
properties of the system. Transforming the hatted variables back into ob-
servables7 , equations (16) and (17) become
'Y -,B (1 - 8)
ClO + -'----'---'----'- (20)
'Y
[1 ~ T In(yt - NXt ) - 1~ T lnGt -lnKt] + Vt
7 N ot.e in part.icular t.he relation Zt = In (Yt - N X t) - In Yt.

282
, - (1- 8) Gt
C20 + ,
In K t (21)
, - ,6 (1 - 8) [ 1 T ]
+ (1 - a),6, 1 _ T In (yt - N Xt} - 1 _ T In G t - In Gt
Here ClO and C20 are constants and Vt is the Euler equation error. Since the
left hand side of the first equation is stationary, so must be the right hand
side, and therefore we see that lnKt cointegrates with In(yt - NXt ) and
In Gt . Using this fact we learn from equation (21) that In Gt cointegrates
with In Kt and that the cointegrating vector is (1, -1).
Proceeding similarly with equation (15) and collecting the constant terms
in C30 results in
1 T
In Gt - 1 _ T In (yt - N X t ) + 1 _ T In Gt (22)
_ 1 + -1/; [
=:C30+- K t +c;f>lnGt +a(l+v)InXt ] ,
(l-a)Inyt
a
an equation that cannot be estimated easily, since In X t is not observed.

Yet we see from (22) that lnyt cointegrates with lnKt,lnGt , and lnXt,
since the right hand side is a cointegrating combination. Solving (5) for
In X t and substituting into (22) we get
1
In Lt = -:---...,.......,.----,- x (23)
(1 + 1/;)(1 + v)
[C30 + 1 ~ T In (yt - N X t ) - 1 ~ Tin Gt - In Gt]
v
+ ( ) [Inyt-(l-a)InKt -c;f>InGtl+uAt
a 1 +v '
C30 + C3K In Kt + C3C lnGt + C3Y In yt + C3NX In (yt - NXt )
+C3G In Gt + UA,t,
which implies that for v :f. 0, In Lt is cointegrated with In yt, In Kt, and
In G t , while for v = 0 In Lt is stationary.
In a system consisting of the six variables In Kt, In Gt , In yt, In(yt-NXt},
In G t , and In Lt and driven by the three independent stochastic trends In X t ,
In Zt, and In Tt we have thus found three independent cointegrating vectors.
283
These form a basis of the cointegration space and are, for convenience,
summarized in Table 1.
Table 1
Coefficients of Cointegrating Vectors
lnKt InCt lnyt In(yt - NXt ) lnGt lnL t

1 -1 0 0 0 0
1 T
0 1 0 -l-T I-T 0
I-a 0 -1 0 cP a~
v
In the empirical analysis, the first step taken is a check of the long-
run implications of the model, i.e., at its cointegration properties. I
will use Johansen's (1988, 1995) general maximum-likelihood-method
to test for multiple cointegration in the system of the six variables
lnKt , In Ct , lnyt,ln(yt - NXt),lnG t , and InL t .
The specification allows for a linear trend in the data by allowing for a
constant in first differences and in the cointegrating relationships. An addi-
tionallinear trend in the cointegrating relationships is not permitted, as the
model supposes that all trend components have been correctly specified8 .
Table 2 gives the results of the Johansen-Tests for the six-dimensional
system, estimated with two lags in first differences. In this and all simi-
lar tables, r denotes the dimension of the cointegration space under HO.
Residuals are approximately white noise with this lag specification, since
the Q-statistic is not significant at the 5%-level. The implied P-value 9% is
rather low, however, indicating that some evidence of autocorrelation may
still exist. The consequences of a longer lag-Iength will be explored below.
8The Johansen-tests have been carried out with Eviews, version 2.0. The program
readily provides Q-statistics for individual residual series, but does not provide proce-
dures to test for residual whiteness in a system perspective. I have therefore tested for
autocorrelation in the six-dimensional error process by using the multivariate version of
the Ljung-Box-Q-test as described, e.g., by Lutkepohl (1991). The procedure was writ-
ten by myself in GAUSS, version 3.2, and uses the correct number of freedoms taking
into account the cointegration restriction on the VAR. The lag-length tested for is four
years (16 quarters).
284
Table 2
Johansen-Test for Cointegration: 2 Lags
Eigen- Trace-. 5% Crit. 1% Crit. r Q(16)

value Stat. Val. Val. (P-value)
0.46 121.6 94.2 103.2 0
0.30 73.6 68.5 76.1 1
0.27 45.9 47.2 54.5 2
0.20 21.2 30.0 35.7 3 576 (0.086)
Normalized Cointegrating Coefficients: 3 Cointegrating Equations

(standard errors in parentheses)
In Kt InCt InYt In(Yt - NXt ) InGt In Lt
1 0 0 -4.08 (2.56) 4.20 (2.38) 6.11 (2.04)
0 1 0 -2.02 (0.79) 1.14 (0.73) 1.08 (0.63)
0 0 1 -1.23 (0.42) 0.68 (0.39) 1.42 (0.34)
We see from Table 2 that the hypothesis of no cointegrating vector is

rejected at the 1%-level and the hypothesis of at most one cointegrating
vector is rejected at the 5% level. The hypothesis of at most two cointe-
grating vectors can be accepted at the 5%-level, but the value of the test
statistic is quite close to the 5%-critical value and hence, since the model
postulates the existence of three cointegrating vectors, I decided to reject
the hypothesis of not more than two vectors as well. The hypothesis of at
most three cointegrating vectors, finally, cannot be rejected at a reasonable
level of significance and is thus accepted.
Turning to the estimated long-run coefficients, recall that under the nor-
malization chosen the coefficients of In(Yt - N X t ), In G t , and In Lt should
be equal in the first and second row of the lower panel of Table 2. This
property is not fulfilled for the point estimates, but the standard errors
are quite large so that a formal test-statistic for the hypothesis of equality
might possibly not be significant.
However, I will not use such a test for in the six-dimensional system. The
reason is that the results given in Table 2 are not particularly stable with
respect to the lag length chosen. For the same system and lag length three
285
(in differences) we obtain, cf. Table 3, evidence for only one cointegrating
vector. Moreover, the point estimates cannot be considered close to those in
Table 2. While the Q-statistic indicates some progress as to the whiteness
of the residuals, it remains unclear, whether the system with two lags is
underparameterized (and coefficient estimates are biased) or whether the
system with three lags is overparameterized (and coefficient estimates are
inefficient). General experience, however, often casts doubt on the reliability
of inference in high-dimensional VARs, so that in the next step I will reduce
the dimension of the VAR by exploiting the zero restrictions given in Table
1.
Table 3

0.41 100.2 94.2 103.2 0
0.33 60.1 68.5 76.1 1
0.20 29.4 47.2 54.5 2 553 (0.247)
Normalized Cointegrating Coefficients: 3 Cointegrating Equations

(standard errors in parantheses)
lnKt lnCt lnYt In(Yt - NXt ) lnG t lnL t
1 0 0 -0.67 (1.07) 1.54 (0.91) 4.80 (1.53)
0 1 0 -1.21 (0.18) 0.46 (0.15) 0.63 (0.26)
0 0 1 -0.77 (0.28) 0.35 (0.23) 1.32 (0.39)
I first check the bivariate system of consumption and the capital stock
for cointegration. It turns out that it is difficult to find an appropriate lag
length specification since, apparently, capital adjustment is rather slow.
Thus, in the absence of other conditioning variables, a lag length of nine
quarters seems necessary to obtain reasonable results. Table 4 shows that
in such a specification, consumption and capital are cointegrated at the
1% level and the estimated coefficient for consumption deviates from the
theoretically expected value of -1 by about one standard deviation. These
results basically confirm the first cointegrating vector in Table 1.
286
Table 4

0.19 22.5 15.4 20.0 0 126 (0.595)
Normalized Cointegrating Coefficients: 1 Cointegrating Equation

lnKt lnCt
1 -1.34 (0.35)
Next I analyze the trivariate system consisting of consumption, output

less net exports, and government absorption, which, according to the model,
should also be characterized by one cointegrating vector. Using three lags
in the Johansen procedure shows that the hypothesis of no cointegration
can indeed be rejected (at the 1% level), while the hypothesis of one coin-
tegrating vector is accepted, cf. Table 5.
Table 5

0.27 37.3 29.7 35.7 0
0.11 13.2 15.4 20.0 1 126 (0.595)

(standard errors in parentheses)
lnCt In(Yt - NXt} lnGt
1 -1.52 (0.17) 0.49 (0.17)
287
The coefficient estimates for In(yt - N X t ) and In Gt independently imply

estimates of T, the steady state share of government absorption in GDP.
These estimates are surprisingly close; they amount to 0.34 computed from
the coefficient of In(yt - NXt ) and 0.33 computed from the coefficient of
lnGt • These values (obtained from second moments) are somewhat too high
if compared with the time series Gt/yt: The mean, i.e. the first moment, is
about 0.25 and the series actually never exceeds 0.28. To see whether the
hypothesis of cointegration can still be maintained with lower values of T, I
set T = 0.25 and subject the series In et -1.33In(yt - N X t ) + 0.33 In Gt to
an Augmented Dickey-Fuller-test. With a constant and three lags included
the test statistic is -3.31, i.e., the series is cl~arly stationary. (The 5% critical
value is -2.90.) Thus I conclude that the three series are cointegrated as
predicted by the model and that the cointegrating vector is compatible with
T = 0.25. Note that the thereby implied coefficients of the cointegrating
vector do hardly differ by more than one standard error from the above
estimates.
To verify the existence of the third cointegrating vector from Table 1,
I now apply the Johansen test to the four-dimensional system consisting
of In Kt, In yt, In Gt , and In Lt. With three lags the hypothesis of no coin-
tegrating vector cannot be rejected at the 5%-level, but, again, the test
statistic is very close to the respective critical value. Thus, in view of the
model implications, I stick to my earlier decision to accept the existence of
one cointegrating vector, cf. Table 6.
Table 6

0.28 46.0 47.2 54.5 0
0.20 21.1 29.7 35.7 1 271 (0.063)

In Kt lnyt lnGt In Lt
0.36 -1 0.59 (0.06) 0.44 (0.28)
The coefficients of the cointegrating vector provide information about the

288
structural parameters Q and cfJ. Taking into account that 1/ is merely a trend
parameter for hours which is easily estimated to be around -0.2 from mean
growth rates, the coefficients of In Kt and In Lt independently imply values
of Q = 0.64 and Q = 0.55. The first of these is quite a reasonable value for
the labor share in total income, while the second is probably a little low.
The difference is not dramatic, however. Both values are lower than the
earlier estimate of 0.78, but the difference is statistically insignificant on
the basis of the standard error obtained in the estimate with HP-filtered
data. The coefficient of In Gt should equal cfJ and should thus give us an
idea of the degree of externality in aggregate production. While a positive
estimate was to be expected from the model, a value of almost 0.6 deems
too high. Quite possibly, this coefficient is subject to some estimation bias
as there may still be some autocorrelation in the residuals, cf. the P-value
of the Q-statistic which is only barely above 5%.
Thus it seems that the long run implications of the model are roughly
fulfilled in the data as far as the cointegration properties are concerned.
However, coefficient estimates are somewhat unsatisfactory. In order to de-
rive more reliable estimates of the structural parameters, I will resort to an
error-correction approach. The equation to be estimated is equation (23),
in which both of these parameters appear. Unfortunately, identification is
not trivial, since there is a nonlinearity in the parameters. Again using the
estimate of T = 0.25, however, we can rewrite equation (23) in a form that
makes it easy to retrieve cfJ and 'Ij; from the regression coefficients
1/
+ Q (1 + 1/) [In Yt - (1 - Q) In Kt - cfJlnGtl
=: C30 + C3ec In ECt + C3Y In Yt + C3K In Kt + C3G In Gt

We therefore have
,J.. _ _ C3G
'I' - ,
C3Y
'Ij; = C3Y + C3K - 1 _ 1.
C3ec
In order to cope with the simultaneous equations bias, I estimate (24)

by instrumental variables (IV). Lagged real wages are quite powerful in
explaining current consumption, so I construct an instrument for In C t by
regressing In C t on log real wages lagged one period and a linear trend. All
other variables I just lag by one period to form instruments. The regression
results are given in Table 7. The implied values of the structural parameters
are Q = 0.67, cfJ = 0.22, and 'Ij; = -1.04.
289
Table 7
Instrumental Variables Results: Dependent Variable is In Lt
Variable Coefficient Std. Error t-Statistic P-Value

CNST -0.57 0.24 -2.40 0.019
Eet -0.40 0.18 -2.23 0.029
lnyt 1.52 0.49 3.09 0.000
In Kt -0.50 0.11 -4.40 0.003
InG t -0.33 0.08 -4.21 0.000
These results coincide quite nicely with the earlier estimates obtained
using HP-filtered data. In both cases the estimates of'l/J and cjJ suggest that
there is no scope for sunspot fluctuations in Germany (since 'l/J is smaller
than -1) and that tax rates may be too high since the point estimates of
cjJ are smaller than the average share of government absorption. However,
these conclusions are still somewhat shaky, since the inferred values of 'l/J
and cjJ are rather close to -1 and 0.25, respectively.
5 Sensitivity analysis
To check the robustness of the results suggested by the point estimates in
the preceding sections, I will compute the maximum eigenvalue of the ma-
trix D for varying configurations of parameters. Matrix D depends on the
six structural parameters a, (3, ,,(, 8, cjJ, and 'l/J. A discount factor (3 = 0.997
implies a real interest rate of roughly 3% and should thus be realistic,
the parameter does not allow for much variation. The parameter "( is de-
termined by the mean growth rates of the main economic aggregates and
does not permit much variation either; with 2% real GDP-growth "( must
equal 1.005. The rate of depreciation, finally, was implicitly determined by
constructing quarterly capital stock data; it equals 0.005 or 2% per year.
Therefore, there is not much scope for variation in these parameters and
hence the stability properties of the model economy essentially depend on
a,cjJ, and 'l/J.
Table 8 displays the maximum eigenvalue (in absolute terms) of D for a
rather comprehensive selection of parameter values, where (3, ,,(, and 8 have
been "calibrated" in the way described above: In the first row of Table 8
I give the maximum eigenvalue of D for the estimated values of a, cjJ, and
'l/J, this eigenvalue is necessarily larger than one since 'l/J is smaller than -1.
Successively increasing 'l/J still results in eigenvalues larger than one, until
'l/J is as high as -0.55. Thus, given the other values are correct, sunspot
290
fluctuations are possible only in the case of a strongly biased estimate of

1/;.
Table 8
Maximum Eigenvalue of D (/3 = 0.997, 'Y = 1.005,8 = 0.005)
a cP 1/; Max. Eigenvalue

0.67 0.22 -1.05 1.0054
0.67 0.22 -0.95 1.0013
0.67 0.22 -0.85 1.0011
0.67 0.22 -0.75 1.0007
0.67 0.22 -0.65 1.0002
0.67 0.22 -0.55 0.9994
0.97 0.22 -0.70 1.0009

0.87 0.22 -0.70 1.0008
0.77 0.22 -0.70 1.0007
0.67 0.22 -0.70 1.0005
0.57 0.22 -0.70 1.0002
0.47 0.22 -0.70 0.9996
0.67 0.02 -0.70 1.0013

0.67 0.12 -0.70 1.0009
0.67 0.22 -0.70 1.0005
0.67 0.32 -0.70 1.0001
0.67 0.42 -0.70 0.9998
In the second and third panel of Table 8 I assume that the our estimate
of 1/; is subject to a strong negative bias and that the true value of 1/; is -0.7.
Varying a, we see that only in the case of a labor share of less than 50%
would the economy display sunspot fluctuations. Such a low value of a is
certainly not realistic. Similarly, setting a = 0.67 again, we would need a
value for cP larger than 0.32 to push both eigenvalues below one. But such
291
a large value of cp is unlikely to hold in reality, as this would imply that the
aggregate production function is homogenous of degree 1~cf> ~ 1.5 or more!
Thus, given our regression results and the above stability analysis, it
seems rather unlikely that the non-keynesian phenomena observed in the
1980s are to be attributed to sunspot fluctuations. Rather, the coefficient
estimates suggest an easy and straightforward systematic explanation for
the effects of reduced government absorption: Note that the estimated value
for cp is 0.22 which is lower than any of the observed values for Tt in the sam-
ple. This suggests that a government policy aiming at a perman reduction
of government absorption, i.e., an attempt to lower the steady-state value
T, increases the steady state value of consumption. (Recall that steady-
state consumption is maximized for T = cp). In this view, the increase in
consumption is due to a government policy that makes resources with low
marginal infrastructure productivity available for private use.
6 Conclusions
At several occasions, the 1980's witnessed non-keynesian effects of fiscal
contractions. Most notably, cuts in government spending in Ireland, Den-
mark and Germany are known to have coincided with increases in private
consumption spending. Self-fulfilling expectations about the effects of stabi-
lization policies may explain these and a diversity of opposing experiences.
This paper formulates a real business cycle model with equilibrium inde-
terminacy, which allows for sunspot fluctuations as well as for systematic
consumption effects of government policy in either direction. Cointegration
tests and Euler equation estimates suggest that the model is approximately
in accord with German data. Parameter estimates imply that the non-
keynesian experiences are not due to self-fulfilling expectations but to a
balancing of the productivity effects of government-provided infrastructure
services and of private capital.
References
Benhabib, J., and Farmer, R.E.A. (1994), Indeterminacy and Increasing
Returns, Journal of Economic Theory 63, 19-4l.
Bertola, G., and Drazen, A. (1993), Trigger Points and Budget Cuts: Ex-
plaining the Effects of Fiscal Austerity, American Economic Review 83,
11-26.
Christiano, L.J., and Eichenbaum, M. (1992), Current Real Business Cycle
Theories and Aggregate Labor-Market Fluctuations, American Economic
Review 82, 430-450.
Correia, 1., Neves, J.C., and Rebelo, S. (1995), Business Cycles in a Small
Open Economy,European Economic Review 39, 1089-1113.
Farmer, R.E.A., and Guo, J.-T. (1994), Real Business Cycles and the An-
imal Spirits Hypothesis, Journal of Economic Theory 63, S. 42-72.
292
Farmer, R.E.A., and Guo, J.-T. (1995), The Econometrics of Indetermi-

nacy: An Applied Study, Carnegie-Rochester Conference Series on Public
Policy 43, 225-27l.
Giavazzi, F., and Pagano, M. (1990), Can Severe Fiscal Contractions be
Expansionary? Tales of Two Small European Countries, NBER Macro-
economics Annual, 75-11l.
Giavazzi, F., and Pagano, M. (1996), Non-Keynesian Effects of Fiscal Policy
Changes: International Evidence and the Swedish Experience, forthcom-
ing in Swedish Economic Policy Review.
Greenwood, .1., Hercovitz, Z. and Huffman, G. W. (1988), Investment, Ca-
pacity Utilization, and the Real Business Cycle, American Economic Re-
view 78, 402-417.
Hansen, G.D. (1985), Indivisible Labor and the Business Cycle, Journal of
Monetary Economics 16, 309-327.
Hellwig, M., and Neumann, M.J.M., (1987), Economic Policy in Germany:
Was There a Turnaround?,Economic Policy 5, 105-140.
Johansen, S. (1988), Statistical Analysis of Cointegration Vectors, Journal
of Economic Dynamics and Control 12, S. 231-254.
Johansen, S. (1995), Likelihood-Based Inference in Cointegrated Vector
Autoregressive Models, Oxford.
Jonsson, G., and Klein, P. (1997), Stochastic Fiscal Policy and the Swedish
Business Cycle, forthcoming in Journal of Monetary Economics.
King, R.G., Plosser, C.I., and Rebelo, S. (1988), Production, Growth, and
Business Cycles: I. The Basic Neoclassical Model, Journal of Monetary
Economics 21, 195-232.
Littmann, D. (1957), Zunehmende Staatstatigkeit und wirtschaftliche En-
twicklung, Koln-Opladen.
Lutkepohl, H. (1991), Introduction to Multiple Time Series Analysis,
Berlin.
Rogerson, R. (1988), Indivisible Labor, Lotteries, and Equilibrium, Journal
of Monetary Economics 21,3-16.
Seater, .1 ..1. (1993), Ricardian Equivalence, Journal of Economic Literature
31, 142-190.
Wen, Y. (1996), Understanding Self-Fulfilling Rational Expectations Equi-
libria in Real Business Cycle Models, manuscript, Hong Kong University
of Science & Technology.
Incentives and risk sharing in a stock market
equilibrium*
Michael MagiU1and Martine Quinzii2
1 Department of Economics, University of Southern California, Los Angeles, CA
90089, U.S.A
2 Department of Economics, University of California, Davis, CA 35616, U.S.A
Abstract. This paper aims to integrate the literature on portfolio choice

and security pricing with the literature on agency costs and capital struc-
ture. It introduces the concept of a stock market equilibrium with rational,
competitive price perceptions (RCPP). Using this equilibrium concept, the
paper finds that incentive considerations induce entrpeneurs (i) to retain
a larger share of their own firm and a smaller share of the equity of other
firms, and (ii) to make much more extensive use of debt than would be
predicted by the standard general equilibrium model of finance. These dif-
ferences translate into higher interest rates and lower risk premia on the
risky securities.
JEL Classification Numbers: D52, D82

Keywords: stock market equilibrium, incentives, risk sharing, capital
structure
1 Introduction
Economists have long been ambivalent on the merits of the stock market.
On the one hand, the capital asset pricing model (CAPM), which is the
basis for the modern theory of finance, emphasizes the merit of the stock
market for diversifying the idiosyncratic risks and sharing the aggregate
risks of productive activity. On the other hand, the traditional view of the
classical economists, revived in modern times by Berle and Means (1932),
*The first. version of t.his paper was present.ed at. the Conference in Honor of Her-
bert. Scarf, Sept.ember 29-30, 1995, Yale Universit.y. We are grateful to Kenneth Arrow,
John Geanakoplos, Peter Hammond, Jean-Jacques Laffont, Peter DeMarzo and Robert
Townsend for helpful discussions, and to participants in seminars at Northwestern Uni-
versity, the Universities of Minnesota and Pennsylvania, the SEDC Conference, ITAM,
Mexico City, the SITE Workshop, Stanford University, the Summer Meeting of the
Econometric Society, California Institute of Technology, and the Economic Theory Con-
ference at Antalya, Turkey, for useful comments.
294
Jensen and Meckling (1976) and the ensuing agency-cost literature, empha-
sized the negative effect on incentives of the separation of ownership and
control implied by the corporate form of ownership. This paper provides
a framework for reconciling these two perspectives and shows the circum-
stances under which the stock market can provide an optimal trade-off
between the beneficial effects of risk sharing and the distortive effects on
incentives.
To study the efficiency properties of the stock market it is natural to
use the framework of general equilibrium. We adopt the simplest model
which permits the simultaneous analysis of production, risk-sharing and
financing decisions-namely the two-period general equilibrium model of
Diamond (1967). In the spirit of Knight (1921) we model the firm as an
entity arising from the organizational ability, foresight and initiative of an
entrepreneur. The activity of a firm consists in combining entrepreneurial
effort and physical input (the value of capital and non-manageriallabor)
at an initial date: this gives rise to a random profit stream at the next
date. In addition to entrepreneurs there is another class of agents which
we call investors: they have initial wealth at date 0 but no productive
opportunities. In the spirit of the principal-agent literature, we assume
that the effort of entrepreneurs is not observable and that the risks to
which firms are exposed are sufficiently complex to make the writing and
enforcement of contracts contingent on states unfeasible (states of nature
are unverifiable). Under these assumptions, markets for channeling capital
from investors to firms and for sharing risks must either be non-contingent
or based on the realized outputs of firms. In this paper we concentrate on
the simplest (linear) contracts: default-free debt and equity. Entrepreneurs
can thus obtain funds for financing their capital investment by drawing
on their own initial wealth, by selling shares of their firms or by issuing
debt; they can diversify their risks by buying shares of other firms. Since
arrangements for financing typically have to be made before production
can take place, we assume that the trades on the debt and equity markets
are made before the entrepreneurs choose the level of effort to invest in
their firms.
Under these circumstances trade on the financial markets will influence
the effort that entrepreneurs invest in their firms. If an entrepreneur fi-
nances his venture by selling most of the shares of his firm, he will not have
much incentive to invest effort in his firm, since most of the payoff from
his effort goes directly to outside shareholders. On the other hand if the
financing is done principally by debt, then typically a high level of effort
will be required to ensure that the firm does not go bankrupt. The effect
on incentives is not however the only consequence of the choice of capital
structure: for the choice of debt and equity also determines the way the
productive risks of the economy are shared. To take an extreme example, if
equity were not traded at all, and if all financing were made by debt, then
no share of the productive risks would be carried by investors - the full
295
burden would fall on the entrepreneurs, who would have undiversified and
leveraged profit streams.
The trade-off between incentives and risk sharing· is the problem that
is studied in the principal-agent literature: the difference is that in the
setting that we consider there is no principal who directly designs a contract
to induce agents (entrepreneurs) to behave in an optimal way. Whatever
incentive schemes there are must somehow be created by the markets. It is
thus natural to ask whether the stock and bond markets can create incentive
schemes which lead to a socially optimal balance between incentives and
risk sharing.
The moral hazard problem posed by the nonobservability of entrepre-
neurial effort only arises when equity is sold, for then the benefit of an
entrepreneur's effort is shared between the entrepreneur and the outside
shareholders, while the cost is born solely by the entrepreneur. If a price
system is to provide appropriate incentives, then it must discourage entre-
preneurs from selling too much equity of their firms. Intuitively this will
only happen if entrepreneurs are aware that the market will ''punish'' them
by a low price for their firms' shares, if they attempt to sell too much of
their equity.
In Section 2 we propose a concept of equilibrium in which markets play
such a disciplining role. It is based on two ideas: first, it assumes that in-
vestors are well informed - they can observe all the financial decisions
of entrepreneurs - and use this information to deduce the effort that en-
trepreneurs will exert. Second, it assumes that entrepreneurs are aware of
this fact: this is formalized by the concept of price perceptions. To decide
whether an investment-financing plan is optimal, an entrepreneur needs to
evaluate what would happen if he were to change this plan: his price per-
ceptions describe how he perceives that the price of his equity would react
to any such change of plan. The price perceptions are assumed to be ratio-
nal (Le., entrepreneurs think that investors will correctly deduce from their
investment-financing decision what their effort and the associated output
of their firm will be) and competitive (an entrepreneur cannot affect the
span of the financial markets and thus the risk premium that investors re-
quire to invest in the risky income stream that he sells). Putting these ideas
together leads to the concept of a stock market equilibrium with rational,
competitive price perceptions (an RCPP equilibrium).
This concept of equilibrium describes markets functioning at their best:
does it suffice to induce a socially optimal outcome? First best optimality
is clearly too demanding a criterion to use in this setting: what is needed is
a extension of the concept of constrained efficiency introduced by Diamond
(1967) which respects both the limited available set of financial securities
and the incentive constraints imposed by the nonobservability of effort. The
associated constrained social optimum problem is in fact equivalent to a
principal-agent problem. In Section 3 we show that an RCPP equilibrium
is constrained efficient: markets can thus be thought of as designing an
296
incentive contract which is the solution of a principal-agent problem. More

precisely, it is the rational price perceptions which provide the incentive
schemes (nonlinear prices) that induce entrepreneurs to choose an optimal
capital structure.
The model that we study makes it possible to integrate two branches of
the literature: the classical literature on portfolio choice and security pricing
(the standard general equilibrium model of finance) and the literature on
agency costs and their relation to capital structure, which following Jensen-
Meckling (1976), have been studied in partial equilibrium models. Having
a model with incentives which contains the classical risk-sharing model as
a special case, permits one to study how the predictions of the standard
model are modified by the presence of incentive effects. In Section 4 we give
examples of RCPP-equilibria and compare the resulting capital structure
and security prices with those of the standard finance model: we find that
in an RCPP equilibrium diversification is less extensive for entrepreneurs,
since incentive considerations induce them to retain a larger share of their
own firm and a smaller share of the equity of other firms than would be
required solely on the basis of risk diversification; furthermore, incentives
induce entrepreneurs to make much more extensive use of debt than would
be predicted by the standard model. These differences translate into higher
interest rates and lower risk premia on the risky securities.
Related Literature. The study of the way ownership structure in busi-

ness enterprise affects incentives has a long tradition in economics. The
classical economists were uncompromisingly in favor of sole proprietorship,
arguing that shared ownership has a negative effect on incentives (Smith l
(1776), Mill2 (1848), Marshall3 (1890)). The idea that share systems can
be explained as a compromise between risk sharing and incentives was in-
troduced in the sharecropping literature by Cheung (1969) and Stiglitz
(1974): for a more recent discussion of shared ownership (and the stock
market) versus sole proprietorship see Hammond (1993). The paper by
Stiglitz was an early contribution to the literature on the principal-agent
problem which subsequently gave rise to an extensive literature (see for
example Sappington (1991) for a survey). Although our paper is not set
up as a principal-agent problem, as we pointed out above the social op-
timum problem defining a constrained Pareto optimum can be expressed
as a principal-agent problem, with the planner acting as a "benevolent"
ISee Book Ill, Chapt.er II of the Wealth of Nations for a criticism of the metayer
system, the share system used in Continental Europe, by which the farmer and the
landowner each obt.ained one half (metarius) of t.he output of t.he farm. See Book V,
Part III for a vehement criticism of joint stock companies.
2 See Book Il, Chapters VI-VIII of Principles of Political Economy for a more bal-
anced assessment of t.he metayer system and Book I, Chapter IX for a discussion of joint
st.ock companies.
3See Book VI, Chapter X and Book IV, Chapter XII of Principles of Economics.
297
principal.
The idea that financial decisions of agents transmit information about
characteristics or actions of agents that are not directly observable or know-
able by the market, has been extensively explored in the finance literature.
Concepts of equilibrium based on this idea and the idea of rational expec-
tations have been used in many partial equilibrium models: for adverse se-
lection in the signaling models of Ross (1977), and Leland and Pyle (1977),
and the subsequent literature [see Harris and Raviv (1992) for a survey];
for problems of moral hazard by Jensen and Meckling (1976), Grossman
and Hart (1982), and Brander and Spencer (1989). This paper differs from
these latter contributions in that it makes explicit in a general equilibrium
setting with moral hazard how the market can resolve (or at least mitigate)
the incentive problems created by asymmetry of information; it also pro-
vides a framework in which the risk-sharing function of financial markets
and their disciplining role in attenuating the agency costs of firms can be
studied simultaneously. This permits the agency costs and benefits of eq-
uity and debt to be balanced against the risk-sharing benefits and costs of
these securities.
A simpler concept of rational expectations is present in all the literature
on general equilibrium with incomplete markets (GEl) which began with
the papers of Arrow (1953) and Diamond (1967), and subsequently gave rise
to an extensive literature [for a survey of results in this area see Magill and
Shafer (1991)]. We have chosen the simplest version of the GEl model with
production, namely Diamond's model, to study how the agency theory of
the firm can be incorporated into a general equilibrium analysis. It is well-
known that a stock market equilibrium in Diamond's model is constrained
efficient but that such a result can not in general be expected to hold in
more complex GEl models. Since the problem of constrained inefficiency
arising in an incomplete markets model with many goods or many periods,
or in a production economy without partial spanning, is not directly related
to the problems posed by incentives, we have chosen to take as a benchmark
the simplest model of a production economy in which financial markets lead
to constrained efficiency in the absence of incentive effects.
An alternative approach to incorporating asymmetric information into
general equilibrium, which is tantamount to extending Arrow-Debreu the-
ory directly to a world with moral hazard and adverse selection, has been
proposed by Prescott and Townsend (1984a, 1984b). The contracts they
consider are lotteries on an abstract consumption space. For the moment
it is not clear to us how the two approaches are related: the contracts they
study seem very different from the standard debt and equity contracts
which are the focus of our analysis.
More recently a number of papers have studied how moral hazard within
the firm affects the pricing of its equity contract [Kahn (1990), Kocher-
lakota (1995), Shorish and Spear (1996)]: these are representative agent
models modified to incorporate the effect of unobservable effort on produc-
298
tion. The findings of Kocherlakota are similar in spirit to those of Section

4 - namely that, when trades are observable, moral hazard does not help
to solve the equity premium puzzle.
The paper is organized as follows. Section 2 presents the basic model
of a stock market economy with moral hazard and introduces the concept
of an RCPP equilibrium. Section 3 analyzes it normative properties, while
Section 4 presents examples of RCPP equilibria, contrasting them with the
equilibria of a standard finance model.
2 Stock market equilibrium

The Model. Consider a two-period model of an economy with production,
in which there is one good (income), and in which an investment of capital
and effort at date 0 gives rise to an uncertain income stream at date 1, the
uncertainty being modelled by states of nature (s = 1, ... , S). There are I
agents: each agent i has an initial wealth wb at date 0 and if agent i is an
entrepreneur, by investing capital (an amount of the good (income)) and
effort ei at date 0 he can obtain the uncertain stream of income at date 1
given by
Fi(i, ei ) = (F{(zi, ei ), ... ,F~(zi, ei )),
where FiU is an increasing function of (zi, e i ) on R~. When agent i
is an investor, we set Fi = O. Each agent has a utility function U i
where Ui(Xi, e i ) is the utility associated with the consumption stream
xi = (xb, xL . .. ,xk) and the effort level ei . U i , which is defined on the
domain R~+I x R+, is increasing in Xi and decreasing in ei . Since the ef-
fort e i of an investor is not productive, it will always be set equal to zero.
Each agent is thus characterized by (U i , wb, Fi) and we let feU, wo, F)
denote the resulting economy with characteristics U = (U I , ... , U I ),
Wo = (W6,"" Wb), F = (FI, ... , F I ).
The characteristics of the economy feU, Wo, F) satisfy the following ad-
ditional assumptions. Agents' utility functions are separable
Ui(xi, ei ) = uh(xh) + ut (xL . .. ,xk) - ci(e i ),
where the functions ub, ut

are strictly concave increasing, and i is convex c
increasing. These functions are differentiable on their domains and satisfy
the boundary conditions4
u~(xh) ~ 00 if xh ~ 0,
4
\7ui. = (8Ui 8u
~, ... , ~
i ) s is the boundary of the
denotes the gradient of ui. and 1JR+
non-negative orthant of 'R s.
299
In short, consumption is essential in all states and effort is essentially cost-

less for small levels of effort.
On the production side, we assume that the production functions have
the multiplicative form
(1)
the function t expressing the specific ability of agent i for transforming

an initial investment of capital and effort (zi, ei ) into a profit stream at
date 1. To permit the same notation to be used for both investors and
entrepreneurs, we adopt the convention that.,.,i == 0 if agent i is an investor.
fi(zi, ei ) is assumed to be a differentiable, increasing function of (zi, ei )
which satisfies fi(O, ei ) = fi(zi, 0) = 0 (both inputs are essential). While
fi is concave in zi reflecting decreasing returns to capital, concavity in ei
is not needed as long as the marginal cost of effort increases faster than its
marginal product (see Assumption MCMP(a) below).
The multiplicative factor structure5 in (1) was first introduced by Dia-
mond (1967). Its principal advantage is that it leads to a competitive pricing
of the firms' risks which is well-defined even if the financial markets are
incomplete. By altering his actions (zi, ei ), entrepreneur i can influence the
expected value of the profit stream of his firm, but he cannot influence the
risk profile "1i of the income stream that he sells, and thus the "risk price"
of this basic income stream. More general risk structures for the produc-
tion functions Fi would require more markets than the basic debt-equity
markets to ensure that an entrepreneur has no influence on the structure
of the financial markets. We leave this case for further analysis and adopt
here the simplest framework in which the assumption of competitive pric-
ing of risks is appropriate, concentrating on the new element introduced by
incentives.
We accept as a fact that the complexity of business risks, when combined
with the unobservability of entrepreneurial effort, makes the writing and
enforcement of contracts contingent on states unfeasible. The opportunities
for sharing the production risks in the economy are those that can be
obtained through shared ownership of the firms. Thus, there is a stock
market on which entrepreneurs, who have the initial property rights to
the profit streams of their firms (since this is the result of their effort and
initiative) can sell a part of their ownership shares to obtain funds for
capital investment, and can buy shares in other firms in order to diversify
their risks. We assume that after selling ownership shares of their firms,
entrepreneurs remain the sole managers of their firms even though they
hold less than 100% of the shares: they are thus "owner-managers" in the
sense of Jensen-Meckling (1976). In addition to obtaining funds by issuing
5 This is ill essence a non linear version of activity analysis, the vector 'l}i constituting
the "activity" (income stream) of firm i (see Tjalling Koopmans (1951)).
300
equity, firms can also issue debt. To simplify the analysis the penalty for
bankruptcy is assumed to be infinite: there is thus a single instrument
traded on the bond market, which is the "default-free" bond.
To make clear how the timing of agents decisions takes place, date 0 is
divided into two subperiods 01, 02. In subperiod 01 entrepreneurs use the
financial markets to obtain the capital required to set up their firms and
to diversify their risks: in the second subperiod 02, after the investment
and financing decisions have been made, firms become "operative" and
entrepreneurs decide on the appropriate effort to invest in the running
of their firms. At date 1 "nature" chooses a state of the world (shock):
production takes place and profit is realized.
In subperiod 0 1 entrepreneur i decides on the amount of capital zi to
invest in his firm, on the amount to borrow bi (if bi > 0, lend if bi < 0),
on the share (1 - O~) of his firm to sell and on the shares 01 of other
firms k -# i to buy: let (i = (01. ... , O~) denote the agent's portfolio of
equity contracts. Since we study the case in which financial markets are
still relatively simple (debt and equity only), we assume that there are no
short sales6 so that (i E n~. Let qo denote the price of the bond and let
Q = (Ql,"" Qf) denote the vector of prices of the firms' shares: thus Qi
is the price offull ownership of firm i, and if agent i is not an entrepreneur
i.e., if Fi(zi, ei ) = 0, then Qi = O. The accountability of agent i requires
that the following budget equations be satisfied
xh wb + qObi - L 01Qk + (1 - ODQi - Zi, (2)

k¥=i
x! _bi + L 01jk(zk, ek )1J: + OUi(zi, ei )1J!, s = 1, ... ,S, (3)
k¥=i
the consumption in each state being non-negative. If a:t

(xL···, xk)
denotes the date 1 consumption stream, and if 1 = (1, ... ,1) denotes the
riskless income stream at date 1, then the S equations in (3) can be written
in the more condensed vector form
a:1 = -bil + L 01jk(zk, ek)rl + OUi(zi, ei)".,i. (4)

k¥=i
The agents' financial transactions (zi, bi , Oi){=1 carried out in subperiod

01 are assumed to be mutually observable. Thus an investor who spends
money buying shares of firm i, knows exactly how this money is used by
entrepreneur i: how much is invested in the firm (zi), how much goes to
private consumption (xb), etc.; he also knows agent i's sources of income
6 This is not essential. If short sales were allowed, to carry out the anlysis we would
need to add the assumption that the income streams 1]i, for 1]i =I 0, are linearly
independent.
301
at date 1, his debt payment -bi , and the dividends he will receive from
the different firms in the economy. What the investor cannot observe when
buying his shares in firm i is the effort entrepreneur i will invest in his
firm: this decision will be made by the entrepreneur in subperiod 02, and
the best the investor can do is to form an expectation about what ei will
be.
Optimal Effort Function. Consider how entrepreneur i chooses his op-

timal effort in subperiod 02' Given that this decision is made after the
financing decision (zi, bi , Oi) has been chosen, the entrepreneur will choose
the effort level ei which maximizes ui (xD - ci (e i ), the date 1 consumption
stream xi being given by (4). If agent i correctly anticipates the effort of
t=
other entrepreneurs (k i), then he will correctly anticipate what his date
1 outside income stream m i will be, where
mi = mi(bi , (Ot)k,ei) = -bil + 2: Otfk(zk, ek)rl (5)

k,ei
The agent's choice of effort is thus the solution of the problem
~~ {ul (mi + O~fi(zi, ei)."i) - ci(e i )} (E)
where the parameters (mi, zi, On E ns x n~ must be such that mi +

OUi(Zi, ei)."i 2: 0 for some ei 2: 0: let 1) denote this domain.
Assumption MCMP (marginal cost-marginal product) .

. (a) For all zi > 0, ci/O/8f~:.i .. ) is increasing and tends to 00 when
e ~ -----> 00.
(b) There is a smooth path ei : [0,1] -----> n+ with ei(O) = 0 and ei/(t) > 0
such that
Assumption MCMP(a) ensures that the problem (E) has a unique solu-
tion, while MCPM(b) ensures that each entrepreneur's technology is suf-
ficiently productive relative to his cost of effort to make it worthwhile
to put his firm into operation: if the entrepreneur were to operate at
(zi, ei ) = (0,0), there would be a way of slightly increasing capital (zi = t)
and effort (e i = ei(t)) so that the increase in marginal utility arising from
the increase in output exceeds the marginal increase in the cost of effort.
Example. If fi(zi, ei ) = (zi)f3(e i )'Y

and ci(ei ) = (ei)O, then MCMP(a) is
satisfied if 8 > 'Y and MCMP(b) is satisfied if 8 > ~. The higher the
power 8, the flatter is the cost curve at zero, and the more readily MCMP
is satisfied.
302
Proposition 1: (i) If Assumption MCMP(a) is satisfied, then for each

(mi, zi, o~) E V the problem (E) has a unique solution
2(mi, i, o~) = arg max {ui(mi + O!Ji(zi, ei )l1i) - ci(ei )} (6)

ei~O
and 2 is differentiable whenever 2(mi, zi, O~) > 0.

(H) If Assumption MCMP(b) holds, then for all xi = (xb,xi) E R!+l
°
with xb > 0, there exist (zi,e i )>> such that
Proof. (i) The first-order condition for the problem (E) is given by
i/( i) S 8 i
C e > o~" u~ (mi
gj:;.( i i) - t L...J 8xt
+ O~fi(zi
t",
ei)'ni) 'T}i
'/S
(8)
ae' z, e s=1 S
P
with equality if ei > 0. Since fi(zi, ,) is increasing, and aau x.
) is decreasing
LHS is increasing. If at ei = 0, LHS exceeds RHS then ei = ° °

by concavity of ui, the RHS of (8) is a decreasing function of ei , while
is the
solution: in the opposite case, since LHS goes to 00 there is a unique ei >
satisfying (8) with equality, and the differentiability of this solution follows
by applying the Implicit FUnction Theorem, noting that the hypothesis of
Proposition 1 implies Cill/Cil > a;t
a e' /IU...
ae'
(ii) Let tl.U denote the difference in utility in (7) between investing
i
(zi,ei(zi)) in activity i and investing (0,0), where ei (-) is the function
defined in MCMP(b) and zi ::::; min {xb/2, 1}, Then
Set k = ug(xb/2), K = \7u1(xl + fi(1, ei (l))l1i) , l1i. Then, since *Er >
0,*!r °
> 0, e'i1 > and ub and ul are concave
By MCMP(b), for zi > ° sufficiently small this expression is positive. 0

303
Note that by (5), the entrepreneur's outside income mi is a function of

his borrowing and of his equity shares in other firms (b i , (Bl)k;ii), so that
his optimal effort is well-defined once he has chosen his financial variables.
We may thus use either the notation ?(mi,zi,Bn as in (6) or ?(zi,bi,(Ji)
to denote an entrepreneur's optimal effort function.
Stock Market Equilibrium. Consider an investor who is thinking of

buying shares of entrepreneur i's firm and can observe his financial deci-
sions (zi, bi , (Ji). It would be "irrational" for the investor not to use this
information to deduce what the most likely effort of entrepreneur i will
be. To be able to deduce ?(zi,bi,(Ji), however, the investor would need to
know in addition to the entrepreneurs financial decisions, his characteristics
(uLci,ji,.,.,i). In the analysis that follows we make the strong assumption
that the agents' characteristics are common knowledge. Thus the investor
can deduce from the financial variables (zi, bi , (Ji) the effort that entre-
preneur i will choose: in short, we suppose that every investor knows the
entrepreneur's effort function ?(zi, bi , (Ji). In practice agents will probably
not have such a precise knowledge of other agents' characteristics - how-
ever they are likely to have a good idea of ''what makes entrepreneurs tick" .
Experienced investors are not readily fooled: they are likely to predict that
an entrepreneur who retains only a small share of his firm and has a lot of
outside income will not exert much effort to make his firm productive.
If investors correctly anticipate, through the price they are prepared to
pay for each firm i, the effect of the financial decisions of entrepreneur i on
the effort that he invests in his firm, then it seems reasonable to suppose
that each entrepreneur will come to understand this. Hence our second
assumption: entrepreneurs know that investors will use their financial deci-
sions as "signals" of the effort that they will exert in their firms. The next
step is to incorporate these two assumptions into a concept of equilibrium.
The description of an equilibrium consists of two parts. The first is the
standard part which enumerates the actions of the I agents, the prices of
the I + 1 securities and the mutual compatibility of their actions under
these prices. The second part describes the entrepreneurs' perceptions of
the way their financial decisions affect the price that the "market" will
pay for the shares of their firms, and ensures that these perceptions are
compatible with the equilibrium prices. Let
denote the price perception of each entrepreneur i and let Q

(Q1, ... ,Q/). Thus Qi(zi, bi , (Ji) denotes the price that entrepreneur i ex-
pects to receive if he sells the share 1 - B~ of his firm, when his other
financial decisions are given by (zi,bi, (B~)k;ii).
304
Definition 1 A stock market equilibrium with price perceptions Qis a triple
«x, e, z, b, 6), (iio, Q)j Q)

consisting of actions, prices and price perceptions such that
(i) for each agent i, (xi, ei) maximizes Ui(;xi, ei ) among consumption-
effort streams such that 7
xh wh + qobi - L Ok01+ Qi(zi, bi , (i)(1 - O~) - zi

k#i
;xi -bil + L 01fk(zk, ek)1Jk + OUi(zi, ei )1{

k#i
for some (zi,bi,fi) E R+ x R x R~j

.. - .- -i -i -i
(11) Qi = Qi(Z ,b ,(J ), i = 1, ... ,Jj
(iii) " ,I
L..,,~=1
li = O',
. ",I -i
(IV) L.."i=10k = 1, k = 1, ... ,1
Thus in an equilibrium with price perceptions Q, each entrepreneur takes

the prices and production plans of the other entrepreneurs as given, and
correctly anticipates the effort they invest in their firmsj he chooses his
own actions, anticipating that those which are observable (his financial
decisions) will influence the price that outside investors are prepared to
pay for their shares in his venture, in the way indicated by the function
Qi(zi, bi , (Ji). By (ii), the price perceptions are consistent with the observed
equilibrium prices Q, and by (Hi) and (iv) , the bond and equity markets
clear.
Without more precise assumptions on the price perceptions Qi(zi, bi , (Ji),
this concept of equilibrium only incorporates the first assumption that we
discussed above - namely that investors have correct expectations - but
it does not yet explicitly incorporate the second - namely that entrepre-
neurs are fully aware of this fact. For example, the equilibrium concept
in Definition 1 would be compatible with myopic expectations of the form
Qi(zi, bi , (Ji) = Oi, i = 1, ... ,1. At first glance this might seem like the nat-
ural candidate for a concept of "competitive" equilibrium. However, this is
not a legitimate use of the assumption of price-taking behavior, since Qi is
not a "per-unit" price, but rather is the price of the whole firm. Compe-
tition means that agents take per-unit prices as given, independent of the
amount that they supply to the market. The "good" sold by entrepreneur i
7Whcncvel' k is not. an entrepreneur, since Fk(zk, e k ) = 0, t.he shares O~ are fictitious:

they are shares of t.he zero vector. In this case we set O~ = 1, O~ = 0, i I=k, so that the
market clearing condition (iv) can be written symmetrically for all agents.
305
to investors is the risk profile ".,i that they can use for taking or diversifying
risks, and we assume that entrepreneur i takes it s price as given. The no-
tion of competition does not however explain how an entrepreneur should
perceive that the "market" will evaluate the personalized part fi(zi, ei ),
namely the "amount" of".,i that we will supply when ei is not observable.
To answer this part, the concept of rational expectations is more appro-
priate than the concept of competition. We are thus led to the following
concept of equilibrium.
Definition 2 A stock market equilibrium with r~tional, c011]JJetitive price

perceptions (RCPP) is an equilibrium ((x, e, b, 0)), (ilo, Q; Q) with price
perceptions in which the perception functions satisfy the following con-
dition: there exist prices (ill, ... , ill) for the firms' basic income streams
".,i, i = 1, ... ,1 such that for i = 1, ... ,1,
(9)
where mi = -bil + L,otfk(zk,ek)".,k. (10)

ki-i
e
Thus to check if his financial decision (zi, fji, i ) at equili~rium is optimal,
entrepreneur i forms expectations about what the price Qi would be if he
were to make an alternative financial decision (zi, bi , Oi). To form these
expectations he takes the price ili of one unit of his income stream ".,i as
given8 , and calculates that the market price of his firm will be tiili' if the
market anticipates his profit will be ti".,i. To evaluate mi in (10) he takes
as given the effort ek that other entrepreneurs (k i= i) make given their
financial choices (e k = e'" (mk, fjk, eZ)). This is the competitive part of his
calculation.
To evaluate what the market anticipates his "output" t i will be, he draws
on his knowledge of investor rationality: he anticipates that the market will
deduce from (mi, zi, o~) what his optimal effort will be, and thus anticipates
that t i will be equal to fi(zi, ?(mi, zi, o~)). This is the rational expectations
part of his calculation.
An RCPP equilibrium describes a situation where entrepreneurial effort
is not observable, but where all participants on the market use all available
information to deduce the likely values of the hidden (moral hazard) vari-
ables - and all agents know this: in short, there is common knowledge of
8Note that the "competitive" price qi can be deduced from the observable market
prices Qi, only if the firm of entrepreneur i is active. For if li(zi, ei ) > 0 then (ii) in
Definition 1 and (9) imply t.hat qi = Qi/li(zi,~). However if li(zi,e i ) = 0, then (ii)
and (9) imply Qi = 0, so t.hat. qi is indeterminate. In this lat.ter case, the concept. of
equilibrium does not. guarantee that the price qi used by ent.repreneur i to reach t.he
decision (Zi, ei ) = 0 is "reasonable", since it. does not. correspond to an objective market
signal. Assumption MCMP(b) avoids the conceptual difficulties t.hat. arise in these cases.
306
rationality.
3 Constrained efficiency
A well-known result of Diamond (1967) asserts that in a model similar to
the one considered in this paper, but in which there are no incentive effects,
the stock market leads to efficient investment and risk sharing, the efficiency
being relative to the existing structure of securities - in short, he proved
that a stock market equilibrium is constrained efficient. When the firms'
profit functions fi(zi, e i ) are independent of e i , so that the effort variables
are omitted, the model we are studying reduces to Diamond's model of the
stock market. Does the constrained efficiency result carry over to the more
general version of the model in which entrepreneurs' incentives are explic-
itly taken into account? Since the stock market cannot achieve risk sharing
without distorting incentives, the question arises whether this trade-off is
achieved in an optimal way at an equilibrium. In their attempt to diversify
their risks, do outside shareholders acquire excessively large holdings in the
firms, leading to undue distortion of the entrepreneurs' incentives to invest
effort in their firms? Or, on the contrary, are the entrepreneurs unduly re-
luctant to sacrifice ownership shares in their profit streams, thus robbing
other agents of potential opportunities for risk sharing? To answer these
questions we need to generalize the concept of constrained efficiency intro-
duced by Diamond to the context of this model. This means introducing a
concept of constrained feasible allocations, which respects the limited trad-
ing opportunities achievable by a system of bond and equity markets, and
in addition respects the incentive constraints imposed by the nonobserv-
ability of effort. Applying the Pareto ranking criterion to this constrained
feasible set leads to the concept of a constrained Pareto optimum.
Definition 3 An allocation (x,e) = (x i ,ei )!=l is constrained feasible if

there exist inputs and portfolios (z, b, 0) = (Zi, bi , Oi){=l E R~ X RI X RV
such that
I I I
Lxh
i=l
Lwh-L i
i=l i=l
(11)
Lb
i=l
i 0 (12)
L
i=l
Ol 1, k = 1, ... ,1 (13)
and for each agent i = 1, ... ,I

xi = _bit + L 01fk(zk, ek)rl
I
(14)
k=l
307
mi = _bi! + L,OUk(zk,e k )17 k (15)

k#i
An allocation (:z:, e) is constrained Pareto optimal (CPO), if it is con-
strained feasible, and if there does not exist any alternative constrained
feasible allocation (x, e) such that Ui(xi, 2) ~ Ui (:z:i , ei ), i = 1, ... ,I with
strict equality for at least one i.
Constrained Efficiency of Stock Market. We can think of a CPO

allocation as being achieved by a "planner" who chooses the variables
which, in equilibrium, are determined by trade on markets, with the objec-
tive of maximizing social welfare. Here the planner chooses the variables9
(xb, zi, bi , (i). The implicit assumption which limits the planners instru-
ments to (xb, Zi, bi ,(i) is that he cannot remove the observational con-
straints of the model, which limit the instruments for risk sharing and make
entrepreneurs' effort impossible to control directly: in particular, the plan-
ner has to respect the fact that entrepreneurs will personally choose their
effort levels based on the incentives created by his choice of investment-
portfolio variables (zi, bi , 8 i ). Proving that an equilibrium is CPO thus
amounts to showing that, given the observational constraints of the model,
there is no way of improving the trade-off between risk sharing and incen-
tives that results from decentralized trade on the markets. In short, even
if a "planner" replaces "markets", he cannot improve on the allocation.
Proposition 2: If £( u, Wo, F) is an economy satisfying the assumptions

of Section 2, then every RCPP equilibrium is constrained Pareto optimal.
Proof. If the equilibrium ((x, e, z, b, 6), (iio, Q); Q) is not CPO, then there
is a constrained feasible allocation (:z:, e, z, b, 8) satisfying (11)-(15) such
that u i (:z:i , ei ) ~ Ui(Xi, ei ), i = 1, ... ,I with strict inequality for at least one
i. By Proposition 1, Assumption MCMP(b) implies that in an equilibrium
all entrepreneurs invest a positive amount of capital and effort in the sector
in which they are productive: as a result all income streams 17 i , with 17i -I 0,
are traded in an equilibrium. Thus the date 1 consumption stream
:z:i = _bi! + L, 01fk(zk, e k )17 k + O~fi(zi, ei )17 i

k#i
would have been available to agent i, (when he in fact chose the equilibrium
consumption xi) had he chosen the investment, debt and ownership in his
. . . -i
own firm (z" b" OD, and the portfolio of shares in other firms (Ok)k#i given
9In order to express the fact that the planner replaces "markets", he must not have
to worry about prices or respecting agents' budget constraints and thus has to be able
to choose the date 0 consumption xb
of agents directly, subject only to the aggregate
feasibility constraint (11).
308
by
~fk(zk, ek) = f)~fk(zk, ek).
Given the outside income mi derived from debt and other firms' securities,
by constrained optimality, his choice of effort ei = ?(mi, zi, f)~) would then
have been optimal. Since (zi, ei ) is preferred or indifferent for all agents
and strictly preferred by at least one agent, the date 0 consumption must
be at least as expensive, and strictly more for some agent: thus
xh ~ wh + iiobi - L Qk~ + Qi(zi, bi , (~)k;ei' f)~)(1 - f)n - zi, i = 1, ... ,I

k;ei
(16)
with strict inequality for some i. Note that by (9), and (ii) in Definition 1,
Qk~ = iikfk(zkek)~ = iikfk(zk,ek)f)~, (17)

Qi(zi, bi, (~)k;ei' O~) iidi(zi, ei ). (18)
Summing (16) over i, using (17) and (18) gives
(19)
By feasibility E{=l bi = 0 and Ek=l f)f = 1, i = 1, ... , I. But then (19)

Imp "I "I "I
. lies L..ti=l Xoi > L..ti=l Woi - L..ti=l Z i , cont r ad·et·
1 mg the const · ed 1ea-
ram &
sibility of (z, e, z, b, 8). 0
The standard framework for studying the optimal trade-off between risk
sharing and incentives is the setting of a principal-agent problem. It is thus
of some interest to note that the planner's problem of finding a CPO can
be expressed as a generalized principal-agent problem. A principal (the
planner), who can be thought of as owning all the resources, looks for a
way of rewarding agents in the economy through the choice of consumption,
investment and portfolio variables, so as to maximize a weighted sum of the
agents' utilities under constraints which limit the risk-sharing possibilities
at date 1 (constraints (14)), the incentive constraints (15), and subject to
a reservation level of utility for himself equal to zero. This latter constraint
can be expressed as the fact that the principal appropriates no resources of
the economy for himself, and is thus equivalent to the resource availability
constraints (11)-(13). If the principal wanted to decentralize the solution
to his social welfare problem by providing agents with incentive contracts,
then he would have to solve the following contract design problem: find
functions <pi : R+ x R x R~ - - R such that a Nash equilibrium of the
game with strategies (zi, bi , 8 i , ei ) for the agents (i = 1, ... ,1) and payoffs
309
Vi(Z, b, (J, e) where
0(%, b, 0, e) ~ u~( ~i(Zi, bi , Oi» + uj t

(bil + e~f'( z', e )'!') -c
i i ( ei )
is a CPO allocation. Proposition 2 asserts that the market provides a so-

lution to this contract design problem given by
cjJi(zi, bi , (Ji) = wh + iiobi - L QkB1 + Qi(zi, bi , (Ji)(1 - B~) - Zi

kfi
where (iio, Q, Q) are the prices and price perceptions of an RCPP equilib-
rium of the economy £(U, wo, F). Note that the contracts cjJi are linear for
investors and nonlinear for entrepreneurs.
To obtain an intuitive understanding for the way in which the market
solves the contract design problem, it is useful to compare the first-order
conditions for constrained optimality with the first-order conditions in an
RCPP equilibrium.
First-Order Conditions for CPO. In view of the boundary assumptions

on the utility functions and assumption MCMP, at a CPO all the variables
Xi are positive and, for entrepreneurs, the variables (zi, ei ) are also positive.
The only non-negativity constraints which need to be taken into account in
deriving the first-order conditions (FOC) are the no-short-sales constraints
B1 2: O. The FOC are more convenient to derive ifthe variables (x, e; z, b, (J)
are replaced by the variables (x, e; z, b, ((ItUkfi, B~)[=l) where the relation
between the two sets of variables is given by
It~ = B1fk(Zk, ek), i 1= k.
The new variables (ItUkfi reflect the fact that the production of firm
k affects agent i only in so far as it affects his outside income Tni. In
these new variables an allocation (x, e) is constrained feasible if there exist
(z, b, ((ItUkfi, B~)[=l) E R~ x RI x RV such that
I I I
Lxh Lwh - L zi , (20)

i=l i=l i=l
I
Lbi 0, (21)
i=l
LIt~ < (22)

kfi
and for each agent i = 1, ... , I
-bit +L It~Tl + B!fi(zi, ei )l1i, (23)
kfi
310
ei = ?(-bil+LJ.£i11k,i,B~). (24)
k#i
A constrained Pareto optimal allocation is a solution of the problem
I
max L vi (u~(x~) + ui(xi) - ci(ei )) ,
i=l
subject to the constraints (20)-(24), where Vi is the relative weight attached
to the utility of agent i. To express the cost of each constraint in units of
date 0 consumption, we divide all the multipliers by the multiplier >'0 in-
duced by the date 0 constraint (20). This gives a set of normalized multipli-
ers (1, qo, (qi, 7ri, €i)I=l) associated respectively with each ofthe constraints
(20)-(24), where 7ri = (7rt, ... ,7r~). The first-order conditions with respect
to the variables (xi, ei , zi, bi , 14, Bn of an entrepreneur i are
8ui1 /8x i8
= 7r~, s = 1, ... ,S; (25)
u il0
Cil
u 0il
= (1 - B~) qi + B~ 1ri . 11i) ~~: _ €i; (26)
1 (1 - B~)~ qi + B~~ 1ri . 11i) 88z~

f~ + €i 8z~'
8~ . (27)
qo 1ri . 1 + €i V'm ? . 1;
i (28)
qk > 1ri . flk + €i "Vmi?' flk, k =/= i; (29)
. . ·1 8?
qi = 1r' '11~ + €~--:-.' (30)
f' 8Bi'
where V' mi? is the vector of partial derivatives (the gradient) of the effort
function ?(mi, zi, B~) with respect to mi = (mL ... , m~) and where (29)
holds with equality if J.£i > o. To these equations should be added the FOe
for the choice of optimal effort by entrepreneur i
ci/(ei ) = B~ V'ui (mi + B~t(i, ei )11i) '11i ~~: (zi, ei )

This is just the marginal way of expressing the incentive constraint ei =
?(.) in (24). Dividing this equation by u~ to make it comparable with (25)
- (30) gives
cil lI'' .~ ~.8fi
u~ = U i 1r . 11 8e i ' (31)
The first-order conditions with respect to the variables (xi, J.£i) of an in-
vestor are (25) and
qo = 1ri. 1 (28 /)
qk~1ri'11k (=ifJ.£i>O), k=/=i. (29')
311
Economic Interpretation of FOe. Equation (25) defines the present-

value vector 7r i = (7r i, ... , 7r~) of agent i: for any date 1 income stream
v = (Vl' ... ,vs), 7ri . v is the present value to agent i of the income stream
v. The variables (qQ, ql, ... ,qI) are the social values (shadow prices) of the
income streams (securities) (1,1]1, ... , 1]i). Ei , which is the social cost of
the incentive constraint (24), is the social value of (one unit of) effort by
agent i. The equations (28) - (30) and (28') - (29'), i.e., the first-order
conditions with respect to (b i , f-LL O!), express the limited sense in which
there must be equalization of marginal rates of substitution to achieve a
CPO allocation, full equalization being prevented by the fact that income
can only be distributed indirectly using securities, and that the incentive
constraints of the agents must be satisfied.
For each security, 1 or 1]k, the private benefit to agent i of an additional
(marginal) unit of the security is 7r i ·l or 7r i '1]k. If agent i is an investor,
then the private benefit coincides with the social benefit and (28') and (29')
express the equalization of social (marginal) benefit and social (marginal)
cost - these are the standard FOC for an optimal portfolio problem. Sup-
pose now that agent i is an entrepreneur and i =1= k. An additional unit of
security 1 or 1]k creates more than just a direct marginal benefit: since the
agent is an entrepreneur, an increase in his outside income has an indirect
effect - for it changes his effort by ~ei = V'mi? . 1 or V'mi? . 1]k, and
since this effort has a social value Ei , the social value of this indirect effect
is Ei ~ei. If i =1= k, in order for agent i to receive an additional unit of the
security of his own firm, his holding OUi(zi,e i ) must increase by one unit:
this is equivalent to increasing O~ by ],. This increase in the shareholding
of his own firm increases 10 his effort by (11 fi) (a? lao!) , the social value
of which is Ei (lit) (a?laoD. Thus (28)-(30) express equalization at the
margin of the social cost and the social benefit of allocating an additional
unit of 1, "1k or "1i to entrepreneur i where the social benefit is equal to
the private benefit to the entrepreneur minus the indirect social cost of his
changed effort.
The social value Ei of an additional unit of effort by entrepreneur i is
defined by equation (26) which can be written as
. = (.O~ 7rt. . 1.]a t · .afi ) ci ,

Et
atet-. + (1 - OD qatet-. - -."
ub
(26').
Ei is the difference between the social marginal benefit O! 7ri . "1i~ + (1 -

On qi~, namely the benefit to entrepreneur i plus the benefit to "out-
side investors" who receive the share (1 - O!) of his output, and the social
lOIn the text we take the most intuitive case where ae' /a()~ > 0 i.e., increased own-
ership leads to increased effort. It can happen, when bi is sufficiently large, that income
effects make this term negative (see Section 4).
312
marginal cost, which here coincides with the private cost Ci , /u~, since en-
trepreneur i is the only one to bear the cost of his effort. Since effort is
chosen optimally by entrepreneur i, by the "envelope theorem", or more
precisely by the FOC (31), the welfare effect on the entrepreneur of a mar-
ginal change in his effort is zero. Substituting (31) into (26') gives
€i = (1 _ (i) qi 8f~ (32)

, 8e'
The social value of an additional unit of effort by entrepreneur i is the
value to agents other than himself of the additional output that this effort
would create l1: thus €i > 0 (= 0) if and only if e~ < 1 (= 1). When
e~ < 1 the effort of entrepreneur i affects all those agents j who obtain a
share of his profit stream: there is thus an external effect. The incentive
constraint implies that this external effect is not taken into account when
agent i makes his effort decision and this creates a cost €i, which is the cost
of separating ownership and control. This cost is however explicitly taken
into account by the planner when he chooses (zi,bi,li).
The logic underlying the FOC (27) for the socially optimal investment in
firm i should now be clear: the social cost of one unit of investment at date
o must equal the direct social benefit (the first term on RHS of (27)) plus
the indirect social benefit (€i 8? /8z i ) from the increased effort by agent i
induced by this increment to the capital input of his firm.
How the FOC for CPO are Achieved at Equilibrium. Since a stock
market equilibrium is constrained Pareto optimal, entrepreneurs must -
just like the planner in a CPO problem - be induced to take into account
the external effect of their effort on the welfare of others, namely the terms
in €i in equations (26)-(30). In the standard model of competitive equilib-
rium, where prices are assumed to be independent of the quantities chosen,
the price system cannot cope efficiently with externalities. However, in an
RCPP equilibrium, there is a "non-competitive" part, namely the rational-
anticipations component of the perception function Q: while entrepreneurs
take the prices (qi){ of the factors ",i as given, they recognize that the price
that the market will pay for their shares depends on investors' expectations
of the effort that they will make. Since investors can deduce from the en-
trepreneurs' financial decisions what their effort will be, financial decisions
end up playing the role of signals: in the process of choosing their "signals",
entrepreneurs are led to internalize the externality.
The way in which the price perceptions force entrepreneurs to internal-
ize the externality, can be clearly understood by matching the FOC at an
equilibrium with the FOC for a CPO allocation. Consider the maximum
llNote that their benefit is evaluated using qi, and not 1r i 7Ji for j # i, and thus
incorporates the incentive cost of giving them a marginal increment in the income stream
7Ji.
313
problem of an entrepreneur in a stock market equilibrium ((i) in Defini-

tion 1). Let Xi = (A~, Ai, ... , X~) E n~+l denote the vector of multipliers
induced by the 8 + 1 budget constraints: the normalized vector
= ::-r(Al, ... ,AS) = (1fi, ... ,1f s)

. 1 -i -i . .
ftt
Ao
is the present-value vector of agent i at the equilibrium. The first-order
conditions are
8u 1i /8x si -i
7r S' S = 1, ... ,8; (33)
u il
0
cif -i . . 8fi
"'l)'" (34)
et ;
() -t t
i 1r . 1]
u if
0
-i -i i 8fi -i8Qi
1 ()i 1r '1] '""iJ'zt" + (1 - ()i)~;
zt
(35)
-i -i 8Qi
qo 1r . 1 + (1 - ()i) 8bi ; (36)
Qk > fti . 1]k fk + (1 _ 7i) 8Q.i . -i

(= If ()k > 0), k fi; (37)
t 8()1'
Qi fti . 1]i fi + (1 - o~) ~~/ . (38)

t
By paying attention to the way potential shareholders react to his finan-

cial decisions (zi,bi,li), through the partial derivatives (8Qd8z i , etc ... ),
entrepreneur i is led to take their interests into account. With the rational,
competitive price perceptions Qi defined by (9), these partial derivatives
are given by
8Qi _ 8fi _ 8fi 8?

et 7P
zt + qi "'l)'" zt ;
qi "'l)'" (39)
8z i
8Qi 8fi .
iJi "'l)'"
et V' m i e' . 1;
(40)
8bi
8Qi - 8fi V' ..".;, k fk
qi "'l)'"
et mi e . 1] , k fi; (41)
8()~
8Qi _ 8fi 8?
qi 8ei 8()i' (42)
8()it t
Substituting (39) - (42) into (33) - (38), and setting qi = iJi, f.i = (1 -
o!) iJi ~ for i = 1, ... , I, gives the FOe (25)-(31) for a constrained Pareto
optimal allocation.
In letting himself be guided by the price perceptions Qi ( Z i , bi , (}i), an
entrepreneur understands, for example, that if he doubles the share (1- ()n
314
of his firm that he sells, this will not double the income he receives: for
shareholders know that when his ownership share falls, the effort that the
entrepreneur ,!ill invest in his firm will fall, and this is reflected in the
smaller price Qi that shareholders will pay for the shares. He also knows
that if he uses the proceeds of the sale for personal consumption or to buy
shares in other firms, he will get less than if he uses the proceeds to finance
capital expenditure for the firm.
There is an interesting connection between Proposition 2 and the condi-
tions for constrained (second best) optimality in an insurance market with
moral hazard (Hellwig (1983), Henriet-Rochet (1991), Lisboa (1996)). In
the insurance models, nonlinear prices are needed to obtain constrained op-
timality, and in such models the insurance companies are the natural inter-
mediaries for implementing such "second-best optimal" nonlinear pricing.
In the stock market, price perceptions induce nonlinear prices: thus rational
behavior and anticipation on the part of agents can act as an alternative
mechanism for achieving constrained efficiency to having intermediaries
that charge explicit nonlinear prices. 12
4 Qualitative properties of stock market equilibria

In this section we examine how equilibria with incentives differ from the
familiar financial market equilibria based on risk sharing. The results which
are summarized in Tables 1-4 show two types of equilibria for economies
with the following characteristics: there are three (types of) agents, two
entrepreneurs (agents 1 and 2) and one investor (agent 3); there are three
states of nature of equal probability, and agents have additively separable
utility functions
3
Ui(xo, XI, x2, x3, e) = vi(xo) + 8i 2)1/3)v i (x s ) - ci(e)
s=l
vi(x) ";x - ai, a1 = a2 = 0, a3 = 50,

8i = 0.9, ci(e) = {3e''t, {3 = 1.8, "y = 2
Thus the utility functions for date 1 consumption are expected discounted
utility, with vi taken from the LRT (linear risk tolerance) family.1 3 All
12In practice the underwriters who undertake to float an issue of shares on behalf of a
firm help to make clear to the company how the market is going to evaluate their issue of
shares, From the perspective of our model, in addition to matching supply and demand,
their role is to help "entrepreneurs" to form rational, competitive price perceptions.
13For an expected utility function E(v(x)), the risk tolerance is defined by T(x) =
-v'(x)/v"(x). The function v is in the LRT family ifT(x) = A+Bx. A is the intercept
and B is the coefficient of marginal risk tolerance. Here Al = A2 = 0, Aa = -100 and
Bi = 2 for all agents.
315
agents have the same coefficient of marginal risk tolemnce (equal to 2) and
agent 3, with a negative intercept, is less risk tolerant than the others. The
entrepreneurs' production possibilities are given by
Thus activity 1 with mean E(.,,1) = 15 and standard deviation 0"(.,,1) = 5.7
is less productive, but less risky, than activity 2, for which E( .,,2) = 18 and
0"(.,,2) = 8.6. The two activities are positively correlated with correlation
coefficient cor(.,,\ .,,2) = 0.76. The economy has a fixed date 0 wealth:
w6 + w5 + w8 = 400. We consider two distributions of initial wealth between
entrepreneurs and investors given by
(80,80,240) and (20,20,360).
To show how the incentive effects change the predictions of the model
with respect to risk sharing, security prices, and the use of debt versus
equity, when compared with the standard CAPM-like model of finance,
we compute two types of equilibria. First, the RCPP stock market equilib-
rium (Tables 1 and 3); second, the risk sharing equilibrium of the associated
finance economy in which firms have the same physical investment and out-
put (Zi, yi) as in the RCPP equilibrium, but where the production plans
are taken as fixed and independent of the consumption-portfolio choices of
the agents. The consumption-portfolio choices and security prices of this
latter equilibrium are those that would be predicted by an outside observer
knowing the agents' risk-impatience characteristics and the firms' produc-
tion plans, but who is not aware of the feedback between the entrepreneurs'
financial decisions and their choices of effort. Since we have chosen utility
functions in the LRT family and since there are well-known properties for
the equilibria of a finance economy with such preferences, we call this latter
type of equilibrium an LRT equilibrium (Tables 2 and 4).
Comparing RCPP and LRT Equilibria. The main difference between

the two types of equilibria lies in the capital structure of the firms. An LRT
equilibrium is a classical risk sharing equilibrium, and by a well-known re-
sult in the finance literature 1 4 , in such an equilibrium agents have fully di-
versified portfolios, ()i/()~ = 1, i = 1,2,3 (see Tables 2 and 4). By contrast,
in an RCPP equilibrium (Tables 1 and 3) because entrepreneurs know that
retaining an increased ownership share implies an increased equity price,
l1For a summary of the properties of LRT economies, see for example Magill and
Quill7:ii (1996, Section 17).
316
and because increasing debt has the same effect, the incentive effects induce
entrepreneurs to retain a higher proportion of their firm than in an LRT
equilibrium: as a result, entrepreneurs typically make more use of debt to
finance their capital investment in an RCPP equilibrum than in an LRT
equilibirum.
The qualitative difference in capital structure in the two types of equilib-
ria translates into a qualitative difference in the prices of the securities or
equivalently their rates of return (as shown in the last row of Tables 1-4).
If r denotes the rate of interest and if ri - r is the risk premium on the
equity of firm i, where
1 E(yi)
1 +r =-,
qo
1 + ri = Q:-' i = 1,2.
and yi = (y1, ... ,y~) is the date 1 profit stream of firm i, then the rate of
interest is higher and the risk premia on securities are lower in an RCPP
equilibrium than in an LRT equilibrium. Entrepreneurs, by restricting the
supply of their firms' shares that they offer for sale, drive up the prices of
equity contracts, thus lowering their risk premia15. The entrepreneurs who
need outside funds to finance their capital investment resort to increased
borrowing, thereby increasing the rate of interest.
The difference between the incentive effects of equity and debt can be
seen by comparing the RCPP equilibria in Tables 1 and 3. The reduced
initial wealth of entrepreneurs in the latter equilibrium forces them to draw
more extensively on outside sources of funds, their capital investment in
the two equilibria being essentially unchanged: were they to raise funds
exclusively by selling shares in their firms, the negative effect on incentives
would lead to a fall in output and to a fall in the price of their shares.
To avoid this decrease in the price of their equity, entrepreneurs increase
their reliance on debt: incurring debt counterbalances the effect of selling
equity, since increasing debt has a positive effect on incentives 16 , leading
to a higher output and higher equity prices. In the equilibrium of Table 3
the effect of increasing debt dominates the effect of selling equity, so that
effort and output increase (by about 15%).
Qualitative Properties of Effort Function. The way in which owner-
15 This result seems to make the "equity premium puzzle" even more of a puzzle.
However the observed high return on equity comes from capital gains rather than a high
dividend yield, and capital gains are not present in our two-period mode\. A multi period
model would be needed to determine whether the incentive-based restriction of the
supply of equity could be a factor contributing to large capital gains
16 A one unit increase in debt leads to a one unit decrease in consumption in each state
at date 1 and hence to an increase in the marginal utility of consumption in each state.
This increased marginal benefit (payoff) of effort implies that more debt leads to more
effort (see footnote 17). Thus, in this model, the market interprets an increase in debt
as a "favorable signal" .
Table 1: RCPP Stock Market Equilibrium
Wo = (80,80,240)
Xo Xl X2 X3 E(zI) O'(Zl) e z b fh (%) fh (%)
agent 1 64 107 68 28 68 32 0.98 43 17 87 0
agent 2 83 201 29 64 98 74 1.24 85 57 0 80
agent 3 126 158 108 111 126 23 0 0 -74 13 20
aggregate a
273 466 205 203 292 124 7.8 4.7 7.3
returns b (%) L-- ~--
---
.. aggregate consumption at each date in each state which for date 1 is equal to aggregate output
b the last row of the "b" column gives the interest rate r (percent); the last row of the 8; column
gives the risk premium ri - r (percent) for each firm i = 1,2 .
.-------------------------- - - ----- ----
Table 2: LRT Equilibrium

Wo = (80,80,240)
11 Xo 1 Xl 1 X2 1 X3 11 E(zd 1 O'(Zl) 11 b I fh (~) 1 ()2 (%)]
I
agent 1 64 119 44 44 69 35 14 29 29
agent 2 84 157 59 59 92 47 19 38 38
agent 3 125 190 102 102 131 42 -33 33 33
aggregatea
273 466 205 203 292 124 6.4 8.1 11.8
~
returnsb (%) ....
~
a,b same definition as in Table 1

~
Table 3: RCPP Stock Market Equilibrium .....

00
Wo = (20,20,360)
Xo Xl X2 Xa E(zt} O'(Zl) e z b (h (%) (J2 (%)
agent 1 32 82 44 6 44 31 1.16 46 37 71 0
agent 2 51 169 9 41 73 69 1.41 86 72 0 67
agent 3 186 274 180 181 212 44 0 0 -109 29 33
aggregate a
269 525 233 228 329 139 18.5 4.9 7.2
returnsb (%)
-.- -- Table 4: LRT Equilibrium

1
Wo = (20,20,360)
---
I_n ]u~o ·1 Xl X2
1 1 Xa 11 E(zt} I O'(zl)Lb J (Jl (%) 1 (J2 (%) 1
agent 1 33 71 28 27 42 21 7 15 15
agent 2 51 112 43 42 66 33 12 24 24
agent 3 185 342 162 159 221 85 -19 61 61
aggregate a
269 525 233 228 329 139 15.8 8.7 12.5
returnsb (%)
319
Ci!IA PH 0 F e Fro RT FUNCTION
OIWN! !\sHIP :SHA R!
FIGURE 1. Graph of the effort function of entrepreneur 2 for different values of b2 :

starting with the top curve the values are b2 = 102, 87, 72, 57, 37, 22, 7, -8, -93,
-393.
ship and debt jointly influence effort is shown in Figure 1. Entrepreneur 2's
capital investment and his ownership share of firm 1 have been set at the
equilibrium values in Table 1 (z2 = 85, ei
= 0) so that his optimal effort
can be expressed as a function of his ownership share B~ and his debt b2
e2 = h2(B~, b2) = ~(-b21, z2, B~).

Figure 1 shows the graph of the effort function e2 = h2(B~, b2 ) viewed as
a function of B~ for different fixed values of b2 • The graph of the effort
function of entrepreneur 1 has the same general form.
Increasing debt always leads to an increase in effort. 1 7 The effect of
changing the ownership share is more subtle. When an entrepreneur has
positive outside wealth (b2 < 0) then an increase in B~ always increases
effort. When the entrepreneur is endebted (b2 > 0) then for any fixed level
~2 2 ~2
b2 of debt, there is a critical level B2(b2 ) such that for 61 2 < 61 2 effort is a
decreasing function of B~ and for B~ > e~ effort is an increasing function. 18
171t is easy t.o see, by differentiating the first-order condition defining the optimal
effort function ei that the property ae i labi > 0 holds generally when ui is an expected
utility
18This behaviour of ae i la(}~ holds for LRT utility functions with a zero intercept and a
320
The negative slope of the optimal effort function for small values of the
equity share is akin to the income effect dominating the substitution effect
in a standard microeconomic choice problem (interpreting effort as labor
and e~ as a wage, since the reward for effort is proportional to e~fi (zi , ei )).
The equation determining entrepreneur i's optimal effort is
Increasing the agents' ownership share e~ has two effects: the direct (substi-
tution) effect is to increase the marginal benefit from an additional unit of
effort; the indirect (income) effect is to increase date 1 consumption xl and
thus to decrease oui/ox! (assuming additive separability), thus decreasing
the marginal benefit of effort. When ;.cl is small (small e~ and large bi ), the
marginal utility of consumption decreases fast and the negative indirect ef-
fect dominates, leading to the apparently paradoxical result that a reduced
ownership share leads to increased effort. When ;.cl is large (large e~ and
small or negative bi ) marginal utility changes very little with an additional
unit of consumption, and the direct positive effect dominates: hence the
intuitive result that increased ownership leads to increased effort.
When b2 > 0, the effort curves are asymptotic to the vertical axis, im-
plying that effort must increase enormously when e~ --+ 0: this is the
°
no-bankruptcy effect. Since in this model the cost of bankruptcy is infinite,
to be sure that the inequality _bi + e!fi(zi, ei )'1]! ~ is satisfied for all
states, the smaller e~, the greater the effort agent i must expend to stay out
of bankruptey. While shareholders of firm i would be happy to see entre-
preneur i incurring a large debt and owning only a small share of his firm,
the entrepreneur in ehoosing his financial variables (z i , bi , e~) will normally
stay out of this region!
Note that for Z2 fixed and e~ = 0, the pereeption function Q2 is a function
of (b 2 , e~)
so that up to a monotone transformation of the vertical axis, the same

graph illustrates the perception function Q2(b2, e~). Thus the general qual-
itative properties of the way effort responds to debt and ownership share
translate into equivalent properties for the perception function Q2(b2, e~).
In particular selling equity can always be achieved without a drop in the
price, provided debt is incurred at the same time.
coefficient of marginal risk tolerance greater than one (vi(x~) = (x~)Ct with 0 < a < 1).
321
5 Conclusion
With the exception of the well-known papers of Prescott and Townsend
(1984a,b), general equilibrium theory and the economics of asymmetric
information are two branches of economic theory which have remained sur-
prisingly separate. With some exaggeration general equilibrium studies cir-
cumstances under which markets "work", while the theory of asymmetric
information reveals the circumstances which make markets "fail". Prescott
and Townsend argue that in principle markets can resolve problems posed
by asymmetry of information: however, to establish this result, they pos-
tulate the existence of an extensive array of markets for contracts (which
rather like Arrow-Debreu contracts) are difficult to identify in the real
world.
The approach of this paper is somewhat different: it seeks to formalize
in a general equilibrium setting why the markets that we actually observe
for debt and equity may perform rather well even in the presence of moral
hazard. The main requirement, in addition to perfect competition, is that
participants on these markets be rational, and that this rationality be com-
mon knowledge. This is formalized in the concept of rational, competitive
price perceptions: it is the anticipatory aspect of perceptions which pro-
vides the disciplinary forces that induce agents to act in the appropriate
way.
References
Arrow, K.J., "Le Role des Valeurs Boursieres pour la Repartition la
Meilleure des Risques", in Econometrie, Colloques Internationaux du
Centre National de la Recherche Scientifique, 1953, 40, pp. 41-47; Eng-
lish version: "The Role of Securities in the Optimal Allocation of Risk
Bearing", Review of Economic Studies, 1964, 31, pp. 91-96.
Arrow, K.J. and F.H. Hahn, General Competitive Analysis, San Francisco:
Holden Day, 1971.
Berle, A.A. and G.c. Means, The Modern Corporation and Private Prop-
erty, New York: Harcourt, Brace and World, 1932.
Brander, J.A. and B.J. Spencer, "Moral Hazard and Limited Liability: Im-
plications for the Theory of the Firm", International Economic Review,
1989, 30, pp. 833-849.
Cheung, S.N.S., "Transaction Costs, Risk Aversion, and the Choice of Con-
tractual Arrangements", The Journal of Law and Economics, 1969, 12,
pp. 23-42.
Diamond, P.A., "The Role of a Stock Market in a General Equilibrium
Model with Technological Uncertainty", American Economic Review,
1967, 57, pp. 759-773.
Grossman, S. and O. Hart, "Corporate Financial Structure and Manage-
rial Incentives", in: J. McCall, ed., The Economics of Information and
322
Uncertainty. Chicago: University of Chicago Press, 1982, pp. 107-137.

Hammond, P., "Financial Distortions to the Incentives of Managers, Own-
ers and Workers" , Working Paper No. 50, University of Munich, Germany,
November 1993.
Harris, M. and A. Raviv, "Financial Contracting Theory", in Jean-Jacques
Laffont ed., Advances in Economic Theory: Sixth World Congress, Vol-
ume Il, Cambridge: Cambridge University Press, 1992, pp. 64-150.
Hellwig, M., "On Moral Hazard and Non-Price Equilibria in Competitive
Insurance Markets" , Discussion Paper No. 109, University of Bonn, 1983.
Henriet, D. and J.C. Rochet, Microeconomie de l' Assurance, Paris: Eco-
nomica, 1991.
Jensen, M.C. and W.H. Meckling, "Theory of the Firm: Managerial Be-
havior, Agency Costs, and Capital Structure", Journal of Financial Eco-
nomics, 1976, 3, pp. 305-360.
Kahn, J., "Moral Hazard, Imperfect Risk-Sharing and the Behavior of Asset
Returns", Journal of Monetary Economics, 1990, 26, pp. 27-44.
Knight, F.H., Risk, Uncertainty and Profit, Boston: Houghton MifHin, 1921.
Kocherlakota, N.R., "The Effect of Moral Hazard on Asset Prices When
Financial Markets are Complete", Journal of Monetary Economics, 1998,
41, pp. 39-56. Koopmans, T.C., "Analysis of Production as an Efficient
Combination of Activities" , in Activity Analysis of Production and Allo-
cation, T.C. Koopmans ed., Cowles Foundation Monograph, New York:
WHey, 1951.
Leland, H. and D. Pyle, "Information Asymmetries, Financial Structure
and Financial Intermediation", Journal of Finance, 1977, 32, pp. 371-
388.
Lisboa, M.B., "Moral Hazard and Nonlinear Pricing in a General Equilib-
rium Model", Working Paper, University of Pennsylvania, 1995.
Magill, M. and M. Quinzii, Theory of Incomplete Markets, Volume 1, Cam-
bridge: MIT Press, 1996.
Magill, M. and W. Shafer, "Incomplete Markets", in W. Hildenbrand and
H. Sonnenschein, eds., Handbook of Mathematical Economics, Volume
IV, Amsterdam: North Holland, 1991, pp. 1523-1614.
Marshall, A., Principles of Economics, London: MacMillan, 1890.
Mill, J.S., Principles of Political Economy, London: J.W. Parker, 1848.
Prescott, E.C. and R.M. Townsend, "Pareto Optima and Competitive Equi-
libria with Adverse Selection and Moral Hazard", Econometrica, 1984a,
52, pp. 21-46.
Prescott, E.C. and R.M. Townsend, "General Competitive Analysis in an
Economy with Private Information", International Economic Review,
1984b, 25, pp. 1-20.
Ross, S.A., "The Determination of Financial Structure: The Incentive-
Signalling Approach", Bell Journal of Economics, Spring 1977, 8, pp.
23-40.
Sappington, D.E., "Incentives in Principal-Agent Relationships", Journal
323
of Economic Perspectives, Spring 1991, 5, pp. 45-66.

Shorish, J. and S.E. Spear, "Shaking the Tree: An Agency-Theoretic Model
of Asset Pricing", Working Paper, Carnegie Mellon University, 1986.
Stiglitz, J.E., "Incentives and Risk Sharing in Sharecropping", The Review
of Economic Studies, 1974, 41, pp. 219-255.
Smith, A., An Inquiry into the Nature and Causes of the Wealth of Nations,
London: W. Strahan and T. Cadell, 1776.
Mergers and acquisitions, bargaining and
synergy traps*
Maria-Angels Oliva1and Luis A. Rivera-Batiz2
1 Department of Economics, Massachusetts Institute of Technology, 50
Memorial Drive, Cambridge, MA 02142-1347

2 Department of Economics, Universitat Pompeu Fabra, Jaume I Building,
Ramon Trias Fargas, 27, 08005 Barcelona, Spain and McGill University,
Montreal, Canada
Abstract. This paper presents a model of cross-product mergers and ac-

quisitions (M&As) motivated by technology synergies. M&As are modelled
as simultaneous bargaining games with endogenous threat points that de-
pend on the endogeneous structure of conglomeration. We show that simul-
taneous M&As can lead to an equilibrium synergy trap that entails lower
profits than under the status quo in which all firms are independent.
JEL Classification Numbers: C78, F23, G34, 032
Keywords: Mergers, acquisitions, bargaining, integration, synergy
1 Introduction
The causes and social benefits of mergers and acquisitions (M&As) have
been a subject of heated discussions. Concerns have been raised about their
effect on competition and industrial concentration, and about their financial
impact on the firms involved. Empirical evidence shows that M&As activity
leads to large income redistributions that target shareholders tend to reap
gains, and that M&As often result in a "synergy trap" that involves the
destruction of firm value'!
·We wish to acknowledge financial support from Comision Interministerial de Cien-

cias y Tecnologia, DGICYT (Spain), and Direccio General de Recerca of the Generalitat
de Catalunya. We are grateful for the comments of seminar participants at Universitat
Pompeu Fabra, the London Business School, Universidad de Alicante, Oxford University,
and McGill University.
1 For evidence on losses from mergers and acquisitions, see Jensen and Ruback [1983],
Ravenscraft and Scherer [1987], Agrawal, Jafee and Mandelker [1992], and the evidence
surveyed in Sirower [1997]. For evidence showing that the synergy trap is not a general
phenomenon and that both target and acquiring firms can benefit from M&As, see Dodd
and Ruback [1977], Bradley [1980], Bradley, Desai and Kim [1988], Morck, Shleifer and
Vishny [1990], Franks, Harris and Titman [1991]' Healy, Palepu and Ruback [1992]' and
326
This paper presents a model of cross-industry M&As that take place to

exploit technology synergies, but can paradoxically generate an equilibrium
synergy trap in the sense that firms' values and profits are reduced com-
pared with the status quo previous to the M&As. Group-level economies
arising from intra-group transfers of technology provide a reason for M&As
because, ceteris paribus, they afford merging firms a production cost ad-
vantage over independent firms that do not benefit from cross-production
effects. The resulting firm integration is neither horizontal nor vertical,
but rather a form of conglomerate integration based on technological syn-
ergies. The label "congeneric" integration is used to distinguish it from
conglomeration due to diversification, and to stress that integrating firms'
technologies are connected through synergies.
We consider a duopoly partial equilibrium model of two industries. The
possible equilibrium conglomeration structures are (1) two independent
firms in each market, (2) one conglomerate and two independent firms,
one in each market, and (3) two conglomerates. M&As involve simulta-
neous (i.e., parallel) negotiations for the formation of two conglomerates.
The takeover process is realized through bargaining between a raider and a
target firm. The bidding cost of acquiring control is a negotiated takeover
fee that is at least as large as the profits the target firm could obtain by
remaining independent (the bargaining problem's threat point). We show
that the takeover fee is a positive function of market size and the bar-
gaining power of target firms, and a negative function of the variable cost
parameter. The distribution of bargaining power between bidders and tar-
get firms affects the allocation of the benefits from the takeover, but does
not influence whether or not the takeover occurs.
When potential synergies are strong enough and the costs of conglomer-
ation are not too high, simultaneous (i.e., parallel) M&As create two con-
glomerates. If two firms realize an M&A transaction, incentives are created
for other firms to follow suit. The reason is that firms' competitive position
will be eroded if they do not also consolidate to benefit from intra-group
technology transfers. The model also generates an equilibrium in which
there is a single conglomerate that competes with those firms that remain
independent. An equilibrium with a single conglomerate arises when the
costs of conglomeration are at an intermediate level or potential synergies
are not too strong (Figure (4)).
Strategic firm interaction can generate a synergy trap when there are two
conglomerates in equilibrium (but not when there is a single conglomerate).
The combination of synergistic gains and strategic firm interaction encour-
ages technological competition that lowers firms' profits by increasing re-
search costs and inducing a large enough price reduction. This situation
entails an equilibrium synergy trap in which the firms engaging in parallel
Matsusaka [1993J.
327
M&As have lower profits than in the status quo in which all firms are in-
dependent (but higher payoffs than the endogenous disagreement payoff of
competing as an independent firm against a conglomerate).
There is a parameter area with a single conglomerate, even if the costs
of conglomeration are low and potential synergies are strong (Figure (5)).
This paradoxical case arises when the demand elasticity is low enough (or
the research cost parameter is high enough). The reason is that the output
increase that would occur due to the coexistence of two conglomerates
would lead to a price reduction (or an increase in research costs) that is large
enough to preclude the two-conglomerate equilibrium in this parameter
range.
The model is consistent with key features about cross-product M&As.
The model generates parallel takeovers exploiting synergies and arising
as a reaction to rivals' M&As. This result relates to the vast evidence
on takeover waves encompassing contemporaneous cross-industry M&As,
particularly the combination mergers of the late sixties and the eighties
takeover wave (Golbe and White [1988]). In the nineties wave, numerous
M&As constituted strategic investments positioning firms in new markets,
exploiting synergies, and reducing output costs in the face of competitive
pressures. Healy, Palepu and Ruback [1992] report that research expendi-
tures tend to be maintained after the M&A, an implication of our strategic
research model for strong synergies.
Next section describes the stage game and models technology synergies.
Sections 3, 4 and 5 solve the two stage research and output game, the
bargaining problem, and the congeneric integration problem. Section 6 en-
dogenizes the structure of conglomeration and examines synergy traps, and
Section 7 offers conclusions.
2 Stage game and synergies

We consider a three-stage game involving sequential decisions about firm
integration, research spending, and output. The paper focuses on syner-
gistic effects that take place within a firm, that is, between members of a
firm group. Contrary to inter-firm spillovers such as Silicon Valley effects
(that we do not consider), cross-product synergies can only be captured
when firms actually integrate to form a multiproduct firm. For instance,
Japanese conglomerates such as Sharp invest in core technologies that can
be applied to many products (Collis and Noda [1993]).
2.1 Stage game and bargaining process

Following much of the literature on models with firms that both conduct
R&D and produce final output, we solve a duopoly model and consider sep-
arately the R&D and output stages (D'Aspremont and Jacquemin [1988],
Steurs [1995]). In these settings, actions in the research stage are viewed as
328
PRE-STAGE STAGE 1 STAGE 2 STAGE 3
Research Final Output

Independent
Spending Production
Finns Conglomeration
Decision Decision
FIGURE 1. Integration and Research-Output Decisions
a partial commitment device limiting output stage decisions, in the sense

that the level of research determines the cost conditions in the output stage.
This standard setting of firms with R&D is extended by adding a conglom-
eration stage in which the structure of conglomeration is endogenized (the
market structure is an exogenously-given duopoly structure).
The three-stage congeneric M&A game is summarized in Figure (1).
Firm behavior is modeled as a sequential move game and the subgame per-
fect equilibrium is computed using backward induction. We only consider
agents playing pure strategies.
At a pre-stage (t = 0) all firms operate as independent duopoly firms.
Firm integration synergies are not exploited and there are no technology
transfers. One could think of the pre-stage as a situation in which the for-
mation of conglomerates is prevented by the impact of regulation, or factors
that might suddenly change such as weak synergies and high conglomera-
tion and technology adaptation costs.
Stage 1 concerns the congeneric integration decision (M&A decision),
and involves a bargaining game that determines the target firm's price.
The second and third stages entail a Nash-Cournot game determining the
levels of research spending and final output production. Because the option
of creating a new firm is not considered, congeneric integration can only
take place by merger with or acquisition of pre-existing firms.
2.2 Cost functions and synergies

Initially, two firms, labelled "I" and "2," compete in one market, and two
firms labelled "1*" and "2*" compete in the other. Even though symmetry
implies that the identities of raider and target firms are indeterminate,
the exposition is facilitated by treating firm 1 as a potential raider that
aims at taking over target firm "1*" to form multiproduct group "1." Firm
"2" is treated as another raider that aims to take over firm "2*" to form
multiproduct group "2."
Demand functions are linear: P= D- 1 (Q) = a-bQ and P*= D- 1 (Q*) =
a-bQ*, where a, b> 0, P and P* denote prices, Q=ql +q2 and Q*=qi +q2'
The post-merger operating profit functions of groups formed by firms 1 and
329
(2a) Independent FInDs

(2b) Conglomerates
(2e) Independent FInDs_and Conglomerate
FIGURE 20 Technology Transfers Network
1*, and firms 2 and 2*, are given by
IIf (0) = (a - bQ)ql + (a - bQ*)qi - Cl (-) - Ci (0)

(1)
IIf (0) = (a - bQ) q2 + (a - bQ*) q2 - C2 (0) - C2 (0),
where we are adding up the demand and cost functions in both product
lines.
Figure (2) illustrates the profile of intra-group transfers of technology in
the three possible equilibrium structures. Figure (2a) represents the case
of independent firms, which, by definition, cannot benefit from synergies.
Figure (2b) depicts the synergy network that arises when there are two
conglomerates. Each group-member has access to a portion (3 of the tech-
nology developed by the member producing another product. Figure (2c)
presents intra-group transfers of technology when a single group competes
with independent firms.
The previous setup can be illustrated by considering a merger between
a tax consulting company and an accounting consulting company. These
companies have no overlap between their product lines, and their demand
functions are independent, but there are potential synergies between them.
Linear versions of the conglomerate operating cost functions are given
by
(A - Xl - (3xi) ql + 'Yx~ (2)

+ (A - xi - (3Xl) qi + 'Yxi2, Xl, xi 20,
(A - X2 - (3x2) q2 + 'Yx~ (3)

+ (A - X2 - (3x2) q2 + 'YX22, X2, X2 2 0,
330
where we impose the restrictions that levels of output and production costs
must be non-negative. Firms are assumed to have the same operating cost
parameter (A = A*). The synergy coefficient {3 E [0,1] represents cost-
reducing intra-group transfers of technology, and 'Y is the own-research
cost parameter. There is also a fixed cost of conglomeration CM&A. This
cost factor represents the costs of coordinating two firms, and transferring
and adapting technology among group firms.
The variable cost reduction (or process innovation) achieved by research
can be decomposed into a direct and an indirect effect. The first one refers
to the direct cost reduction effect of firms' own research investment, x,
and the second one to the cost-reduction gain {3x due to the R&D activ-
ities of the other group's member. The indirect effect reflects intra-group
transfer of technology among the firms forming the conglomerate, and is
the feature that justifies an M&A motivated by the search for production
cost-reduction. Expressions (Xl + (3xi) and (xi + (3Xl) in equation (2) rep-
resent the sum of the unit cost reduction achieved by own and intra-group
research efforts for firms 1 and 1* of group 1 (the same applies for group
2).
3 Research-output choices
We proceed to solve the research-production problem and determine the
profit functions under the three possible conglomeration structures: two
conglomerates, one conglomerate coexisting with independent firms, and
four independent firms.
3.1 Conglomerates' research-output decisions and intra-group

synergies
Firms have incentives to spend in process innovation that translates into net
advantages against their product market competitors. Each conglomerate
member's research spending Xi and xi is
x'=x~=2(a-A)(1+{3) i=12. (4)

• • 9lry - 2 (1 + (3)2 ' ,
If {3 E [0,1]' and a> A, groups will engage in research when 9lry - 2 (1 + (3)2
> 0. The greater the difference a- A between market size and operating cost
parameters, and the larger the intra-group transfer of technology parameter
{3 E [0,1], the greater the equilibrium research investment.
Substituting for x and x* into the solution for output as a function
of research, qi (x, x*) and qi (x* , x), yields the unique firms' equilibrium
outputs
.= ~= 3(a-Ah i=12 (5)
q. q. 9lry - 2 (1 + (3)2 ' ,.
331
Stronger intra-group synergies stimulate R&D spending, reduce marginal

costs and lead to expanded final output production.
Output prices are obtained by substituting output quantities into the
inverse demand functions
* a [3lry - 2 (1 + ,8)2] + 6bky

(6)
P =P = 9lry - 2 (1 + ,8)2 .
Prices decline with the strength ,8 of intra-group synergistic effects, but

increase with the magnitude of the operating cost parameter A. Larger
synergistic gains reduce operating costs both directly and through increased
research. This leads to higher output and lower prices, driving firms to
accentuate their competitive behavior by a movement toward the more
elastic part of the demand function. The price effect of the level of demand
parameter a is ambiguous and depends on ,8. Output prices increase with
the value of the demand parameter if,8 is small enough (,8 < ~ - 1).
Substituting the solution for production, research, and prices into con-
glomerate operating profits nf =P( q1 +qi}-C1(. )-Ci (-) and nf = P( q2 +
q:i) - C2 (.) - C2 (.) yields
nf (a, A,,8, 'Y, b) nf (a, A,,8, 'Y, b) (7)
2 (a - A)2 'Y [9lry - 4 (1 + ,8)2]

[9lry - 2 (1 + ,8)2f
Operating profits are strictly positive for b > 4(1::)2, augment with market
size a, and decline with the variable cost parameter A.
The presence of a mimicking symmetric rival causes conglomerates' prof-
its to decline with greater synergies' strength. The reason is that excessive
technology creation entails high research costs and leads to disadvanta-
geously low prices. Notice that these synergy effects are due to the joint
effect of both firms' actions. Ceteris paribus, stronger synergies (a higher
,8) do add up to the profits contributed by given firm research facilities.
The point that technology competition entailing high team research costs
can be detrimental to participants, applies to various settings involving the
formation of competing research teams. In particular, research-based indus-
tries such as computers, telecommunications, and biotech are vulnerable to
these effects.
3.2 Independent firms

Independent firms constitute the status quo. The solution for independent
firms can be obtained by making synergies ,8 equal to zero in the two-
conglomerate case. There exists a unique symmetric solution for research
332
spending, final output, and output price
_ _ 2(a-A) _ _ 3,(a-A) p_ a(3lYy-2)+6lYyA

Xl - X2 - 9,b _ 2 ' ql - q2 - 9,b - 2' - 9,b - 2 .
(8)
A firm forming part of a conglomerate invests more in R&D than an inde-
pendent firm (compare equations (4) and (8». How do conglomerate and
independent firms contrast as regards prices and output? In this model,
a conglomerate member firm produces more output and sets lower prices
than an independent firm (compare expressions (8) and (5) for output, and
(8) and (6) for prices).
Substituting the solution for research spending, output and prices into
the profit functions 7r{=Pql-(A - Xl)ql-'X~ and 7r~=Pq2-(A - X2)q2-'x~
yields
(9)
3.3 Partial conglomeration

Partial conglomeration entails one group spread across two industries or
products, competing with an independent firm in each market. The con-
glomerate operating profits ITP are (the P superscript represents a single-
conglomerate structure)
2 (a - A)2, (3lYy - 2)2 [9lYy - 4 (1 + 1'1)2]

IT P (a, A,,8,,, b) = (10)
[27b2,2 - 12lYy (1'12 + 2,8 + 2) + 4 (1 + 1'1)2r'
The operating profits 7r P of independent firms competing with a conglom-
erate are
(a _A)2, (9lYy - 4) [3lYy - 2 (1 + ,8)2r

7rP(a,A,,8,,,b)=----~----------~----------~~n2· (11)
[27b2,2 - 12lYy (1'12 + 21'1 + 2) + 4 (1 + ,8)2]
The conglomerate's profits increase with stronger synergies (a higher 1'1),
while the profits of independent firms competing with a conglomerate
decline with greater synergies. Both conglomerate and independent firms'
profits increase with market size a, and decline with variable cost A.
4 A bargaining approach to M&As

This section examines the bargaining game between potential buyers (or
firm raiders) and sellers (or target firms) leading to the formation of a
multi-product integrated firm. The discussion corresponds to the bid-sell
333
negotiation process that takes place in the first stage of the congeneric
integration game. Given the model's symmetry, the identities of buying
and selling firms are undetermined.
M&As transactions can take place through either negotiations or auc-
tions. Bulow and Klemperer [1996] show that auctions are optimal in a
wide range of cases. For instance, auctions maximize the return of a "bad"
manager that is selling a firm but does not accurately know its potential
value. On the other hand, bargaining is appropriate when synergies are
specific to merging firms, they possess greater information about the merg-
er's value than other firms, or the costs of searching and negotiating with
alternative buyers make auctions expensive.
4.1 The bargaining model

A raider forming a conglomerate must pay a fixed cost CM&A plus the
takeover fee FK required by the target firm. The takeover fee is determined
in a bilateral negotiation involving the potential buyer and seller. It is given
by the fee F K that maximizes the geometrically weighted average of the
surpluses STK and SRK received by the target and the raider (the two
potential members of conglomerate K).
The maximized value F K of the takeover fee is (Hoel [1986])
(12)
where the equilibrium price, FK (-) E {~(o:, a, A, {3, ,,(, b, CM&A), F~(o:,
a, A, {3, ,,(, b, CM&An, is strictly positive and depends on 0: E [0,1], which
represents seller's bargaining power.
In order to solve the bargaining problem and determine the takeover fee,
we need to compute the surpluses of the buyer and seller. The bidder's
surplus, SRK, is defined as the conglomerate's operating profits, minus the
sum of the fixed costs of conglomeration, the takeover fee paid, and the
opportunity cost of buying. The seller's surplus, STK , is the takeover fee
received, FK, minus the opportunity cost of selling the firm. The opportu-
nity cost consist of the profits obtained if the conglomerate is not formed.
Because there are two duopolists in each market, the bargaining problem
between any two firms depends on conjectures formed about the strategies
that other firms follow.
One possible conjecture made by two firms bargaining over the takeover
fee is that other firms have formed a conglomerate. In that case, bidder's
benefits from a conglomerate correspond to the operating profits nO (com-
puted in equation (7)) net of the conglomeration costs (fixed costs CM&A
334
plus takeover fee Ff). The opportunity cost of buying is given by the op-
erating profits the bidder would obtain by remaining independent when
there is a conglomerate in the other market, 7r P , given by (11). Under such
conjecture, the bidder surplus is
(13)
_CM&A - Ff P
-7r (a,A,{3,"(,b).
The other possible conjecture refers to the absence of a coexisting con-

glomerate. In that case, bidder's benefits from a merger or acquisition are
given by a conglomerate's operating profits rrP (10) when it is the only
existing conglomerate. The opportunity cost of buying is given by the in-
dependent firm's profits 7r I defined by (9). The bidder's surplus is thus
SRK (a,A,{3,,,(,b,Fk,C M&A) rrP(a,A,{3,"(,b) (14)
_CM&A - Fk - 7rI (a, A, "(, b),
where Fk is the takeover fee paid under a partially conglomerated struc-
ture.
Similarly, the seller's opportunity cost depends on the specific conjecture
made. If the seller supposes that there is a conglomerate in the market so
that the second seller has accepted its raider's offer, the price of the firm
is given by Ff, and the seller's opportunity cost coincides with the profits
of an independent firm under a partially conglomerated structure, 7r P in
(11). Algebraically,
STK (a,A,{3,,,(,b,Ff) =Ff -7r P (a,A,{3,"(,b). (15)
If the seller conjectures that there is no rival conglomerate, then the ask
price for accepting the bidder's proposal is denoted by Ft;. The opportunity
costs of selling are the profits 7r I obtainable under the independent firms
structure specified in equation (9). The corresponding seller's surplus is
(16)
4.2 The bargaining solution

We proceed to determine the takeover fee Ffi under the conjecture that
there is another conglomerate, and the fee Ft; under the conjecture that
there is no other conglomerate.
Let us conjecture that there is a conglomerate rival, meaning that the
parallel negotiation between a seller and its raider succeeds. Under such
an assumption, bidder and seller disagreement profits 7r P (a, A, {3, "(, b)
are the profits of an independent firm competing with a conglomerate. The
necessary and sufficient first order conditions of the bargaining problem are
. a STK (a, A, {3, "(, b, Ff)
z) 1 _ a = SRK (a , A , (3 ,,,
'V b Fe CM&A)
, K'
335
~ -7r P (a,A,(3",b)
(lIO (a,A, (3, " b) - CM&A) - (~+ 7r P (a, A, (3", b))
ii) STK C,~) = ~ - 7r P (a, A,(3", b) 2: 0
iii) SRK (-.~) = lIo (a,A,(3",b)

_CM&A - ~ - 7r P (a, A, (3, " b) 2: o.
Equation (i) says that the seller's bargaining power relative to the bid-
der's bargaining power is equal to the seller's surplus relative to that of
the bidder. This equation can be solved to determine the takeover fee as
a function of the model's parameters, F~ = ~ (a,a,A,(3",b,CM&A).
Equations (ii) and (iii) express that, in equilibrium, the price received by
the seller should exceed its opportunity cost, and the bidder surplus should
be positive. The total surplus is STK + SRK = lI o - CM&A - 27r P .
Solving the first order condition for ~ (the target firm's price under
the conjecture that there is a rival integrated firm) we obtain
FK a,a,A,(3",b,CM&A)
-0 ( (17)
a (lIO (a, A, (3, " b) - CM&A) + (1 - 2a) 7r P (a, A, (3, " b)
a [(lIO (a, A, (3, " b) - CM&A) - 7r P (a, A, (3, " b)]
+ (1 - a) 7r P (a, A, (3, " b) ,
with F~ (a,a, A, (3", b, CM&A) 2: o. The takeover fee is a weighted average

of the surplus obtained by a conglomerated firm and the opportunity cost
of the target, with the weights corresponding to the bargaining power of
the target and the bidder. It is easy to verify that the fee increases with
the seller's bargaining power a and market size a, and declines with the
cost parameter A. How it changes with (3 depends on the particular point
in the parameter space at which the fee is computed.
Under the conjecture of independent rivals, first order conditions are
obtained by replacing lIo by lIP and 7r P by 7r I in (i)-(iii) above. The
target firms' equilibrium price is given by
F~ (-) = a (lIP (a, A, (3, " b) - CM&A) + (1 - 2a) 7r I (a, A, " b), (18)
FK = -P
where -P FK ( a,a,A,(3",b,CM&A) > 0, II P ( a,A,(3",b)IS. gIven.
by
(10), and 7r I (a, A, " b) by (9). The total surplus with a single conglomerate
is STK + SRK = lI P - CM&A - 27r I •
336
Coalition
DO NOT
CD~liti~ CONGLOMERATE CONGLOMERATE
FULL PARTIAL
CONGLOMERATE
CONGLOMERATION CONGLOMERATION
PARTIAL INDEPENDENT
DO NOT
CONGLOMERATE CONGLOMERATION RRMS
FIGURE 3. Congeneric Integration Game
5 Mergers and acquisitions equilibrium

In this section we formally specify the game equilibrium concept employed
and examine the equilibrium conditions. We develop a simple separation
property showing that agents' relative bargaining power influences the dis-
tribution of the benefits among sellers and bidders, but does not affect the
merger and acquisition decision.
5.1 Merger and acquisitions game and payoffs

Figure (3) illustrates the congeneric integration decision. The merger and
acquisition game involves four players. The duopoly structure means that
there are at most two potential coalitions each formed by two players. When
two negotiating parties agree on the takeover fee, for a given conjecture
about the result of the other simultaneous (i.e., parallel) bargaining game,
they form a conglomerate. Equilibrium requires all agents' conjectures to
be sustained. If negotiating parties fail to reach an agreement, both players
remain independent firms and their payoffs coincide with the disagreement
payoff's.
The target and raider in a potential coalition K are denoted by T K and
RK, respectively. Agents' strategies are: (1) {conglomerate, conglomerate},
in which case the negotiators reach an agreement to buy-sell; and, (2) {con-
glomerate, do not conglomerate}, (3) {do not conglomerate, conglomerate},
or (4) {do not conglomerate, do not conglomerate}, in which case there is
noM&A.
Payoff's contingent on all firms' strategies are defined by
(19)
337
[(F~, ne (-) - F~ - CM&A) , (F~" ne (.) - F~, - CM&A)]

two conglomerates;
[(FC, nP (-) - FC - CM&A) , (7r P (.), 7r P (.))]
partial conglomeration;
partial conglomeration;
independent firms,
where the pairs (V TK , V RK ) and (v T K' , V RK') denote the payoffs of the po-
tential seller and buyer of coalitions K and Kt. Expressions ne, n P, 7r P
and 7r1 are given by (7), (11), (10), and (9), respectively. Notice that sub-
tracting the opportunity costs from the Nash equilibrium payoffs yields
agents' surpluses.
5.2 Conglomeration equilibrium conditions

In this subsection we discuss the M&As equilibrium conditions and show
that the M&A decision is not affected by seller's bargaining power, a E
[0,1]. The bargaining process consists of the redistribution of firms' sur-
pluses from merging, but the takeover decision is not affected by a. The
F
takeover fee K E {~, F~} is a function of agents' conjectures about
the structure of conglomeration, which must be sustained in equilibrium.
Full Conglomeration (two conglomerates) arise in equilibrium if merging
firms' payoffs are no less than their payoffs as independent firms. The fol-
lowing conditions have to be satisfied simultaneously (given the conjecture
that there is a rival conglomerate):
(1) Raiders' operating profits ne (equation 7) are greater or equal than
the sum of the takeover costs F~ (17), the M&A costs CM&A, and the
raiders' opportunity cost of merging 7r P (11)
ne (a, A, {3, ,,(, b) ~ ~ + CM&A + 7rP (a, A, {3, ,,(, b), where
(20)
F~ = a (ne (a, A, {3, ,,(, b) - CM&A) + (1 - 2a)7rP (a, A, {3, "(, b).
338
(2) Target firms' price Ilk

is greater or equal to their opportunity costs
1['P (11), given the conjecture that there is a rival conglomerate
-=:C
F K ;: =: 1['
P( a,A,/3,,),,b). (21)
Substituting Ilk,
1['P, and n c into (20) and (21), and simplifying terms,
we get the following full conglomeration condition:
nC (a, A, /3, ,)" b) - CM&A ;: =: 21['P (a, A, /3, ,)" b). (22)
In a Partial Conglomeration equilibrium (a single conglomerate and two

independent firms), the following conditions should be satisfied:
(1') Raider's profits nP(equation 10) when conjecturing that there is no
other conglomerate minus the fixed costs CM&A, are greater or equal than
the sum of the takeover cost F~ (18), and raider's opportunity cost 1['1 (9),
np ( a,A,/3,,),,b ) -C M&A ;::=:F

-P K ( a,a, A ,/3,,),,b) +1[' 1 ( a,A,,),,b)
where (23)
F~ = a (np (a, A, /3, ,)" b) - CM&A) + (1 - 2a)1['1 (a, A, ,)" b).
(2') The price F~ obtained by the target should not be less than its
opportunity cost 1['1
-P 1
F K ;::=:1[' (a,A,,),,b). (24)
(3') The rivals remain independent. The relevant conditions are any of
(3'a) the second potential raider does not find it profitable to conglom-
erate
1['
)
P (a,A,/3,,),,b;::=:n C ( a,A,/3,,),,b ) -C M&A-C
-FK; (25)
(3'b) the firms that were conjectured to remain independent have to find
the sale to a potential raider unprofitable,
1['P (a,A,/3,,),,b) ;::=:~, (26)
where 1['P and ~ are defined in (11) and (17), respectively.

Partial conglomeration arises as equilibrium if
(i) the raider's operating profits nP minus the fixed M&A cost exceeds
two times the opportunity cost 1['1 of the target and the raider
n P (a, A, /3, ,)" b) - CM&A ;::=: 21['1 (a, A, ,)" b) (27)
(obtained by substituting for F~ into conditions (1') and (2')), and

339
(ii) a second conglomerate would be unprofitable
2Jr P (a,A,,8,),,b) ~ IT e (a, A,,8,)', b) - CM&A (28)
(obtained by substituting the takeover fee in (20) into (25)).

Independent Firms arise in equilibrium if, when conjecturing that there
is no other conglomerate, firms prefer not to conglomerate,
Jr I (a,A,)"b ) ~ ITP (a,A,,8,),,b ) - C M&A - -P

FK (a,a,A,,8,),,b ) , (29)
where the takeover fee is F~ (a,a,A,,8,),,b,C M&A)= a(ITP(a, A, ,8, )',

b)_CM&A)+ (1-2a)Jr I (a, A, )', b), with IT P and JrI given by (10) and (9).
Substituting F~ into (29) yields that the profits firms could obtain from
merging, conjecturing that there is no rival conglomerate, are less than the
combined payoffs they could obtain in case of disagreement,
(30)
The full M&As equilibrium condition (22), the single conglomerate equi-
librium conditions (27) and (28), and the independent firms condition (30)
do not depend on the value of sellers' bargaining power a E [0,1]. There-
fore, the decision to merge is not affected by the distribution of bargaining
power. However, sellers' bargaining power affects the allocation of the sur-
pluses.
6 Conglomeration and synergy traps

This section endogenizes the industrial conglomeration structure and ex-
amines synergy traps. Even if the duopoly structure is maintained, compe-
tition between conglomerates can displace all independent firms, and there
is the possibility that equilibrium entails competition between a conglom-
erate and independent firms. The gains and losses deriving from M &As de-
pend on the industrial conglomeration structure generated. Equilibria with
two conglomerates can result in a synergy trap, but a single conglomerate
always gains while independent firms competing against a conglomerate
always lose.
6.1 Equilibrium conglomeration structure and multiple

equilibria
The single-conglomerate inequality condition (28) is the reverse of the two-
conglomerate equilibrium inequality condition (22), except at the boundary.
This means that full conglomeration and partial conglomeration structures
are both equilibria only when the inequalities above hold as equalities;
otherwise they are mutually exclusive in equilibrium. The same type of in-
compatibility arises between partial conglomeration and independent firms
340
11
INDEPENDENT FIRMS
1 8
FIGURE 4. Nash-Equilibrium Solution (a - A= 3, b =1, , = 2)
equilibria, because expression (27) is the reverse of condition (30), except

at the boundary. This argument characterizes the areas in which there is
multiple equilibrium with partial conglomeration and either full conglom-
eration or independent firms.
The conglomeration model studied entails four familiar parameters re-
lating to demand size and slope (a and b) and variable and research costs
(A and ,). There are two additional parameters measuring the cost of con-
glomeration (CM&A) and the benefits from conglomeration ({3). Because
the structure of conglomeration does not depend on bargaining power (a),
we can plot the key results on the ({3, CM&A) plane.
6.2 Simulations
The results of the simulations performed are summarized in Figures (4)
and (5). From the equations characterizing boundaries II and CC (defined
in the appendix), our simulations show that changes in the parameters are
mostly limited to shifting the curves' position up or down, keeping the
general shapes of the boundary between the two-conglomerate and single-
conglomerate equilibria illustrated in the Figures.
One typical pattern of conglomeration is depicted in Figure (4) (con-
structed for a=lO, A=7, b=l, "y = 2); the appendix contains the equations
for each boundary. Independent firms constitute the unique equilibrium
structure in the area above the curve I 1. In this area, conglomeration costs
are "too large" relative to intra-firm synergies and do not justify total M&A
costs.
Partial conglomeration arises in equilibrium when the intra-firm synergy
parameter is in the range between curves labelled II and CC. M&A costs
341
clA&A
INDEPENDENT ARMS
FULL CONGLOIAERAllON
0.84 {J
FIGURE 5. Nash-Equilibrium Solution (a - A = 3, b = 'Y = 2)
are "too large" to permit two firms to merge, but not for a firm to con-
glomerate while the other firms remain as independent firms. The duopoly
structure is maintained after the M&A, but independent firms competing
with a conglomerate always lose in comparison with the status quo (com-
pare equations (9) and (11)).
Full. conglomeration equilibria arise when the gains obtained from imple-
menting group R&D are large relative to the M&A fixed costs associated
with implementing the synergies (the area below curve CC). In this area,
the profits 'lfP of an independent firm competing with a conglomerate are
lower than the net profits obtained under conglomeration. Therefore, par-
allel M&As take place in both markets.
Figure (5) depicts the case of a low demand elasticity (high b) or a
high research cost parameter ,. It is constructed for a=lO, A=7, b=, = 2
(compare with Figure (4)). The figure illustrates the possibility that there
is a single conglomerate even if the costs of conglomeration are low and
potential synergies high (Figure 5). This paradoxical case arises when the
demand elasticity is low enough (or the research parameter high enough).
In that case, the market can only sustain one conglomerate. The reason is
that the output increase due to the coexistence of two conglomerates would
lead to a price reduction (or research cost increase) that is large enough to
preclude the two-conglomerate equilibrium.
6.3 Equilibrium synergy traps

By equilibrium synergy trap we refer to the prisoner's dilemma-like situ-
ation that can arise in this model. When the two potential teams play a
non-cooperative game, abstaining from joint maximization, conglomeration
342
results in lower profits than what firms could obtain in their status-quo as
independent firms. Because status quo profits do not depend on synergies
or M&A costs, comparisons between pre- and post-conglomerate gains do
not depend on the initial point on the ({3, CM&A) plane. The analysis of
gains is equally valid when conglomeration results from deregulation (with
no change in {3 or CM&A), a large enough increase of {3, or an appropriate
reduction of CM&A.
A synergy trap can allude to losses by the raider (raider's synergy trap),
by the target (target's synergy trap), both of them (joint synergy trap), as
well as to losses for the conglomerate as a whole (that is, the aggregate of
sellers and buyers lose, although not necessarily all of them). In this model,
integration losses (called the synergy paradox) result from the prisoners'
dilemma problem involved in the formation of two rival conglomerates.
The synergy trap phenomenon is not due to entrepreneurs' mistakes or
myopic behavior, and is not the result of managers acting under imperfect
information, as frequently stressed (see Sirower [1997]). Even though there
are synergy traps, firms engage in M&As in equilibrium because if they did
not integrate with another firm, they would be even worse off.
Algebraically, raider-firm synergy traps are defined by
nG (a, A, {3, ,)" b) - pC - CM&A :s 71"1 (a, A, ,)" b). (31)
Substituting the takeoverfee, pC = a(nG_cM&A)+ (1-2a)7I"P given by (17),

and operating profits (7), (11) and (9), into condition (31) yields
nG (a,A,{3, ,),, b) - (CM&A +7I"P (a,A,{3, ,)" b) +71"1 (a,A,,),,b))

(32)
The threshold level of sellers' bargaining power, athreshold,R, that gener-

ates a synergy trap for the raider is obtained when (32) holds as an equality.
We get that a raider synergy trap arises if seller's bargaining power exceeds
the raider's threshold level
threshold R nG - (CM&A + 7I"P + 71"1)
a 2 a '= nG _ (CM&A 271"P) + , (33)
where athreshold,R 2 0 if nG 2 CM&A + 7I"P + 71"1, and athreshold,R < 1 if

7I"P < 71"1.
Target-firm synergy traps are defined by
(34)
The threshold level of sellers's bargaining power, athreshold,T, that delimits
the target's synergy trap region is obtained by substituting the takeover
343
fee given by (17), and operating profits (7), (11) and (9), into condition
(34). Solving for the target's threshold, we obtain that a seller synergy
trap occurs when
7r I - 7r P
a < athreshold,T = (35)
- lI c - (CM&A + 27r P ) ,
where athreshold,T <1 if lI C _CM&A_7r P ;:::: 7rI so that lI C _C M &A_27r P ;::::
7r I -7r P .
The model is able to generate a synergy trap when there are simulta-
neous M&As (but not when there is a single conglomerate). When both
inequalities (31) and (34) hold simultaneously, we have joint synergy traps,
that is, both targets and raiders lose. For instance, by directly calculat-
ing firms' profits with a=lO, A=7, b= "( = 2 (as in Figure (5)), f3 = 0.5
and CM&A = 0.3, it can be checked that conglomerates' profits are less
than independent firms' profits. Strategic firm interaction can generate a
synergy trap when there are two conglomerates in equilibrium because the
combination of synergistic gains and strategic firm interaction encourages
technological competition that lowers firm profits by increasing research
costs and inducing a large enough price reduction. This situation entails an
equilibrium synergy trap in which firms engaging in parallel M&As have
lower profits than in the status quo in which all firms are independent (but
higher payoffs than the endogenous disagreement payoff of competing as
an independent firm against a conglomerate).
7 Conclusions
We have presented a model of M&As leading to the formation of multiprod-
uct firms motivated by technology synergies. The model endogenizes the
industrial conglomeration structure, and examines the effects of competi-
tion between conglomerates, and competition between a conglomerate and
independent firms. There is an equilibrium synergy trap in which the prof-
its of all the firms engaging in simultaneous (i.e., parallel) M&As are lower
than in the status quo (as independent firms). Conglomeration arises de-
spite of losses, because the endogenous disagreement payoffs of the bargain-
ing problem are the profits from remaining independent while competing in
disadvantage against a rival conglomerate (rather than with independent
firms).
The analysis suggests extensions to incorporate asymmetric information,
match the dynamics of M&As waves, and consider research that creates new
products or inputs. We could thus generate alternating processes of merger
waves, demergers, and divestitures, depending on the nature of techno-
logical change (Markides [1995]). For instance, if the costs of integrating
research among new products is initially very high, incentives for mergers
could take some time to develop. Also, the extent of synergies and the evo-
lution of technology implementation costs can be endogenized and related
344
to human capital and the quality of available inputs. On the financial side,
one can consider the role of auctions, poison pills, taxes, and other mecha-
nisms and restrictions influencing the gains' allocation between sellers and
buyers, and introduce a realistic financial structure comprising shares and
debt.
Mergers and acquisitions representing technology enhancing processes
can have a significant impact on industrial concentration, firms' competi-
tive position and the vitality of industries. In the nineties, the restructuring
wave affecting computers, telecommunications, pharmaceuticals and other
technology-based industries, altered market conditions at a global scale.
The examination of the microeconomic foundations of such phenomena
might provide useful insights about the "real" basis of macroeconomic be-
havior and financial markets.
Appendix
The boundaries of the areas in Figures (4) and (5) can be derived as follows.
The equilibrium condition that characterizes a two-conglomerate structure
is given by condition (22). Substituting the firm's operating profits defined
in expressions (7) and (11) into (22), we obtain that the parameter con-
dition for an equilibrium two-conglomerate structure (the area below the
curve CC in Figures (4) and (5)) is
2 (a - A)2 ')' [9lYy - 4 (1 + ,8)2]

_ _ _ _---"_ _ _---:~---"- _ CM&A
[9lYy - 2 (1 + ,8)2f
(a - A)2 ')' (9lry - 4) [3lry - 2 (1 + ,8)2f

> 2 2.
[27b 2')'2 - 12lYy (,82 + 2,8 + 2) + 4 (1 + ,8)2]
Curve CC is defined by
CM&A = 2 (a _ A)2 ')'[ 9lYy - 4 (1 + ,8)2 2 _

[9lYy - 2 (1 + ,8)2]
(9lry - 4) [3lYy - 2 (1 + ,8)2r

-----~-----"-------n2J·
[27b2')'2 - 12lry (,82 + 2,8 + 2) + 4 (1 + ,8)2]
Independent firms arise as an equilibrium when two times the profits from
remaining an independent firm exceed the profits from a single conglom-
erate that competes with two independent firms (condition (30)). Substi-
tuting the profit equations (10) and (9) on inequality (30), yields the inde-
pendent firms parameter condition (the area above the curve II in Figures
345
(4) and (5))

2 (9,b - 4)-y (a - A)2
(9,b - 2)2
[9b! - 4 (1 + ,6)2]
2 (a - A)2, (3b! - 2)2 -=-_____
_________ ~ _ CM&A
>
[27b2,2 - 12b! (,62 + 2,6 + 2) + 4 (2 + ,6)2] 2
Curve I I is defined by
CM&A = 2 (a-A)2[ 9,b-4

, (9,b - 2)2
(3b! - 2)2 [9b! - 4 (1 + ,6)2]

2J·
References
Agrawal, Anup, Jeffrey F. Jafee and Gershon N. Mandelker (1992), "The
Post-merger Performance of Acquiring Firms: A Reexamination of an
Anomaly." Journal of Finance 47, pp 1605-1621.
Bradley, Michael (1980), "Interfirm Tender Offers and the Market for Cor-
porate Control." Journal of Business 53, pp 345-376.
Bradley, Michael, Anand Desai and E. Han Kim (1988), "Synergistic Gains
from Corporate Acquisitions and their Division between the Stockholders
of Target and Acquiring Firms." Journal of Financial Economics 21, pp
3-40.
Bulow, Jeremy and Paul Klemperer (1996), "Auctions versus Negotia-
tions." American Economic Review 86(1), pp 180-193.
Collis, David J. and Tomo Noda (1993), Sharp Corporation: Corporate
Strategy. Boston: Harvard Business School Publishing.
D'Aspremont, Claude and Alexis Jacquemin (1988), "Cooperative and
Non-Cooperative R&D in Duopoly with Spillovers." American Economic
Review, 78(5), pp 1133-1137.
Dodd, Peter, and Richard S. Ruback (1977), "Tender Offers and Stock-
holder Returns: An Empirical Analysis." Journal of Financial Economics
5, pp 351-373.
Franks, Julian, Robert Harris and Sheridan Titman (1991), "The Post-
merger Share-price Performance of Acquiring Firms." Journal of Finan-
cial Economics 29, pp 81-96.
Golbe, Devra L. and Lawrence J. White (1988), "A Time-series Analysis
of Mergers and Acquisitions in the U.S. Economy," in Alan J. Auerbach
Corporate Takeovers: Causes and Consequences. Chicago: University of
Chicago Press. pp 265-307.
346
Healy, Paul M., Krishna G. Palepu, and Richard S. Ruback (1992), "Does
Corporate Performance Improve After Mergers?" Journal of Financial
Economics 31, pp 135-175.
Hoel, Michael (1986), "Perfect Equilibria in Sequential Bargaining Games
with Nonlinear Utility Functions." Scandinavian Journal of Economics
88(2), pp 383-400.
Jensen, Michael C. and Richard S. Ruback (1983), "The Market for Corpo-
rate Control: The Scientific Evidence." Journal of Financial Economics
11, pp 5-50.
Markides, Constantinos C.(1995), Diversification, Refocusing, and Eco-
nomic Performance. Cambridge, Mass., and London, England: MIT
Press.
Matsusaka, John G. (1993), "Takeover Motives during the Conglomerate
Merger Wave." Rand Journal of Economics 24(3), pp 357-378.
Morck, Randall, Andrei Shleifer and Robert W. Vishny (1990), "Do Man-
agerial Objectives Drive Bad Acquisitions?" Journal of Finance 45, pp
31-48.
Ravenscraft, David J. and Frederic M. Scherer (1987), Mergers, Sell-Offs
and Economic Efficiency. Washington, D.C .. Brookings Institution. .
Sirower, Mark L. (1997), The Synergy Trap: How Companies Lose the Ac-
quisition Game. New York. The Free Press.
Steurs, Geert (1995), "Inter-industry R&D Spillovers: What Difference Do
They Make?" International Journal of Industrial Organization 13, pp.
249-276.
Protectionism versus non-protectionism
under cost uncertainty in Cournot and
Stackelberg markets*
Benan Zeki Orbay
istanbul Technical University, Faculty of Management, Ma<;ka 80680, Istanbul,

Turkey
Abstract. This paper compares optimal protectionist trade policies with

optimal non-protectionist policies in an environment where governments
have private information about the marginal costs of the domestic firms. A
country with an efficient firm is always better off not protecting its firm and
subsidizing both domestic and foreign firms compared to being protectionist
by taxing imports. However, under some circumstances, a country with an
inefficient firm does better being protectionist and taxing imports.
JEL Classification Numbers: F12, F13
Keywords: Trade policy, protectionism
1 Introduction
There has been a considerable amount of research on strategic trade in the
last decade. In their seminal paper, Brander and Spencer (1985) demon-
strate that export subsidies can be welfare improving when export mar-
kets are imperfectly competitive. In most studies following Brander and
Spencer optimal strategic trade policies are shown to be sensitive to the
type of market competition, product nature and number of firms. Most re-
cently, Long and Souberyran (1997) and Bandyopadhyay (1997) show that
cost heterogeneity and demand elasticities are important determinants of
strategic trade policies. Introducing incomplete information to a Brander
and Spencer type model, Cooper and Rlezman (1989) show that optimal
policy choices depend on the amount of the noise in the demand intercept.
In particular, when the amount of noise is small, direct quantity control is
found to be optimal. However, when the amount of noise is large, export
*1 am grateful to David Collie, Bjorn Jorgensen, Semih Koray, Peter Neary, Murat
Sertel, Hakan Drbay, !;,ule Ozler and the participants of the 1996 Economic Theory
Fest in Istanbul, 1996 European Meeting of Econometric Society in Istanbul, and the
Ill. International Meeting of the Society for the Advancement. of Economic Theory in
Antalya, for helpful comments and fruitful discussions.
348
subsidies are optimal for countries with a small number of firms and export
taxes are optimal for the countries with a large number of firms. Following
Cooper and Riezman (1989), Arvan (1991) considers a tax-subsidy game
between governments under demand uncertainty. In Cooper and Riezman's
model, governments choose the policy type at the first stage and the policy
level in the second stage. In Arvan's model, choices can be made before
or after observing the demand intercept. Hence, the outcome is gener-
ally asymmetric (except the case where industry concentration is identi-
cal between the countries). One government acts first, like a Stackelberg
leader, and makes its choice before observing the demand intercept. The
other government acts as a follower choosing its policy after making the
observation. Shivakumar (1993) considers both subsidies and quotas in a
model similar to Arvan's. He shows that quotas are preferable for both
countries in a duopolistic international market. The timing of the policy
depends on the noise in the intercept. Hwang and Schulman (1993) intro-
duce non-intervention, as a distinct policy choice, into Cooper-Riezman's
model. They show that although non-intervention does not arise as a Nash
equilibrium, including it as an explicit policy option changes (some of) the
results of Cooper and Riezman. Qui (1995) investigates the effects of non-
linearity of policies by endogenizing the firms' choices of strategic variables
(quantities or prices) under market uncertainty.
Most of the studies in strategic trade policy literature deal with de-
mand uncertainty. However, only very few studies consider cost uncertainty.
Maggi (1992) and Qui (1994) are among these few studies. Maggi (1992)
considers non-linear policies and shows that when governments do not know
the costs of their domestic firms, the strategic trade policy game results in
a higher output and lower profits in comparison to a complete informa-
tion game. Qui (1994) considers screening and signalling problems in the
policy game. He shows that the government chooses an information reveal-
ing menu policy under the Cournot competition, but an information con-
cealing uniform policy under the Bertrand competition. In a recent study,
Brainard and Martimort (1997) also investigate the screening effect when
governments propose menus of linear contracts to the firms depending on
their cost declarations.
This paper also focuses on cost uncertainty. Our model differs from the
above models in several dimensions. First, we introduce home consumption
by letting two countries trade between themselves. Second, the governments
have full information about domestic firms' costs but have incomplete in-
formation on foreign firms' costs. In contrast, in both Maggi (1992) and
Qui (1994), governments have incomplete information on domestic firms'
costs, and foreign firms' costs are either common knowledge or not im-
portant for countries' trade policy choices. We argue that, in reality, it is
relatively easier to obtain information about domestic firms compared to
foreign firms. In our model, it is publicly known that homogeneous prod-
ucts can be produced with either low constant marginal cost technology
349
or high constant marginal cost technology. Governments have two policy

choices: they can impose a tariff on imports or subsidize exports. Similar to
the model of Cooper and Riezman (1989), they commit to one of these two
policies at the first stage and determine the policy level at the second stage.
The costs of the foreign firms are revealed between the first and the second
stages. Hence, governments choose levels of the policies under complete
information but they choose policy types under incomplete information.
Firms decide how much to produce and where to sell after observing the
governments' trade policy choices, at the third stage. We analyze firms'
decisions under two different scenarios. In the first scenario, firms behave a
la Cournot in both domestic and foreign markets. In the second, they act
as Stackelberg leaders at home.
In this paper we show that, in the protectionist environment, taxing im-
ports is preferred by both governments irrespective of the types of their
firms and their behavior. The intuition is that when the rival country im-
poses import taxes, taxing imports is a better policy choice for each coun-
try. Because of the disadvantageous position due to the import taxes in the
foreign market, domestic firms' gains in profit are higher in the domestic
market. In addition, there is a positive gain due to import taxes. Hence,
it is better for the countries to support their firms in the domestic market
instead of in the foreign market.
We also consider a different environment where a free trade agreement al-
lows governments to impose non-protectionist policies, such as excise taxes
or subsidies, without distinguishing the country of origin of the product.
In such an environment, a country is better off subsidizing both domestic
and foreign firms when the domestic firm has high efficiency and taxing
both firms when the domestic firm has low efficiency (except for the case
where domestic firms are Stackelberg leaders and the probability of facing
an efficient foreign firm is very low).
Welfare comparisons, under both type of market structure, indicate that
a country with an efficient firm is always better off signing a free trade
agreement and subsidizing both firms. A country with an inefficient firm
also does better with a free trade agreement if the relative level of the
foreign firm's efficiency is low under both types of market structures. After
the information is revealed, countries do not regret their decisions for most
cases.
This paper is organized as follows. Section 2 describes the model. Pro-
tectionist and non-protectionist policies under alternatives are analyzed in
Section 3 and Section 4. In Section 5 we compare the findings of Section 3
and 4. Concluding remarks are in Section 6.
2 The Model
In this paper, we assume that there is one firm located in each of two
countries. Firms produce a homogeneous good and sell it in the domestic
350
Governments Cost Governments Firms

select of foreign choose choose
policy --t firms --t policy --t production
type is revealed levels quantities
FIGURE 1.
and foreign markets. The constant marginal costs can be either high or
low. Formally, firm i's marginal cost is represented by Ci which is a random
variable that can take values Cl and Ch where Cl < Ch. We assume that
Ci and Cj are independent and identically distributed and that each has a
probability distribution represented by the vector IL = (ILL, ILh) where ILL and
ILh are the respective probabilities of Ci = Cl and Ci = Ch. For simplicity,
fixed costs and transportation costs are assumed to be zero. Country i's
publicly known inverse demand function for the homogeneous good is
(1)
where Xi = x~ + X; is the total output and x~ and x; represent the amount

of outputs produced by firm i and firm j, respectively, in country i's market,
{i, j} = {I, 2}. The efficiency of each firm is defined by Oi = a - Ci. Let
a = a - Cl denote high efficiency and {3 = a - Ch denote low efficiency; thus,
we have Oi E {a,{3}.1 2
3 Protectionist Policies
In this section, we consider an environment where countries are not en-
gaged in a free trade agreement to prevent intervention. We assume that
each country has two policy choices: taxing imports and subsidizing ex-
ports. 3 Furthermore, each government knows only the marginal cost of the
domestic firm. At the first stage governments select the policy types and
at the second stage they choose its level. In between these two stages, the
costs of the foreign firms are revealed. Then, at the third stage firms make
their production decisions. Figure 1 represents the timing of these events.
Note that firms decide under full information. Thus, the profit of firm i
1 The tilde is dropped to represent the realized value of the random variables.
2We will only consider the region where et/ (3 :'S 2 throughout the paper, because when
et/(3 > 2, an inefficient firm can not survive in the simple Cournot market. (Note that
subsidies or the leadership advantages may relax this condition)
3 Of course, governments could use both policies together. However, we keep the model
as simple as possible ill order to see the effects of the cost uncertainty on each policy
choice more clearly. Obviously, the next step should include a wider range of policies
including mixed policy choices.
351
can be written as
where si is the amount of export subsidy imposed by country i, and t j is

the amount of import tax imposed by country j.
3.1 Cournot market

First we assume that firms behave a la Cournot. Maximizing (2), firm i
decides how much to sell in the domestic market and how much to export.
This maximization results in the following Cournot equilibrium outputs:
2B· - B· - sj +
. = _'_--::..J-::--_ ti
__
x~
, 3 (3)
-J. _ 2B·' - B·J + 2s i - 2t j

xi - 3 . (4)
The social welfare function of Country i is defined as follows:
where r\Bi' Bj ) = (9 i +9r it-

ti
)2 is the consumer surplus of country i,
ifi (Bi' Bj) the equilibrium profit of the firm i is expressed as
(6)
where
i-i(B. B.) _ t i (2B j -Bi +2si _2ti)
t Xj J' , - 3
is the import tax revenue and
is the export subsidy loss. Governments choose one of the two policies si
and t i and compute the welfare W (B i, Bj) setting the other policy variable
to zero. The Nash equilibrium policy levels are described in the following
three equations: 4
4These policy levels are computed by maximizing Eqn.(5) with respect to si or t i

(for both realized values of the random variable OJ)' taking the other country fs policy
variable as given.
352
. -' ()j ()i

(r,P) = ("3'3")'
( ....i -j) = (4()i - 3()j 6()i - ()j)
s ,t 14' 14 .
The first equation is for export subsidies in both countries. The second is
for import taxes in both countries. Finally, the third is for export subsidy
in country i and import tax in country j. Neary (1994) shows that in the
standard Brander and Spencer export subsidy game, governments impose
high subsidies to low cost firms. Our findings strengthen this paradoxical
result. The optimal subsidy levels are high for the low cost domestic firms
(high efficiency firms) even when domestic consumption is included in the
model and the foreign country is allowed to impose import taxes. As stated
in Neary (1994), the source of higher subsidies for the more efficient firms is
simply the incentive of the governments to increase domestic firms' profits.
Since the profit-shifting effect is higher for relatively efficient firms, it is
optimal to give higher subsidies to these firms.
It is important to note here that such a paradoxical result is not found
for import taxes. For the asymmetric case, where a foreign country imposes
an export subsidy, the optimal import tax level is higher when the domestic
firm's efficiency level is higher. When import taxes are imposed bilaterally,
the optimal tax level is only a function of the foreign firm's efficiency level.
The tax level is low for the low efficiency firm.
In Tables 1 through 4 (located at the end of the paper) we present the
payoffs for the trade policy game at optimal policy levels given above. Table
1 is for the case where both firms, domestic and foreign, are efficient. Table
2 is for the case when the domestic firm is efficient and the foreign firm is
inefficient, etc.
3.2 Stackelberg market

In this section, we assume that firms are Stackelberg leaders in their do-
mestic markets. When domestic firms have leadership advantage the equi-
librium outputs are described as
~ _ 2Bi - Bj - sj + ti (7)
Xi - 2 '
~j _ 3()i - 2()j + 3s i - 3t j (8)

Xi - 4 '
{i,j} = {I, 2}. The profits and consumer surplus at the Stackelberg equi-
librium, respectively, are
353
The 92timal policy levels are obtained by maximizing social welfare func-
tion W i (Bi' Bj), at the above Stackelberg equilibrium. The optimal tax lev-
els presented below are respectively for the following three scenarios: (i)
both countries impose export subsidies, (ii) both countries impose import
taxes, and (iii) country i imposes export subsidy and country j imposes
import tax.
( ~ -;:;j) = (3Bi - 2B j 3B j - 2Bi)
8,8 3' 3 '
( -,i -j) = (7B j - 2Bi 7Bi - 2B j )
t ,t 19' 19 '
( ~ +:i) = (18Bi -16B j 21Bi -lOBj )
8 ,t 39' 39 .
Some of the results of the Cournot case are robust when the domestic
firms are Stackelberg leaders. Countries continue to give higher subsidies
to the more efficient domestic firms independent of the policy choice of the
foreign government (import tax or export subsidy). When both countries
impose import taxes, however, the result is slightly different. Specifically,
the efficiency level of domestic firm plays a role in determining the level
of import tax. The higher the domestic firm's efficiency the lower is the
import tax level. In Tables 5 through 8 (located at the end of the paper)
we present the social welfare levels of the trade policy game at the above
optimal policy levels.
3.3 Policy game

In this section we first present the solution of the policy game under com-
plete information, then present the results for the incomplete information
game. The solution of the simple games given in Tables 1 through Table
8 indicates that regardless of firm behavior and type, taxing imports is al-
ways the unique equilibrium in the protectionist environment. This result
differs from the standard result in the existing literature. The standard
result is that under quantity competition, export subsidy is the outcome of
the strategic trade policy game. This difference is due to the discrepancies
in the structures of the models. In our set-up domestic consumption is in-
cluded and the strategy spaces of governments differ because they contain
import taxes.
We introduce incomplete information to the policy game assuming that
governments know only their domestic firms' efficiency. Thus, they know
at least which of the two games are going to be played. For instance, for
the Cournot case, a government with an efficient domestic firm knows that
it is playing either the game given in Table 1 or the game given in Table 2.
Expected social welfare for country i can easily be computed by using the
publicly known probability distribution of marginal costs, as described in
the next equation.
EWi(Bi) = 1L1Wi(Bi,a) + (1-1L1)W i (B i ,J3), Bi E {a,J3}. (10)
354
In order to describe the Bayesian Nash Equilibrium of the policy game,

we define a type contingent strategy space5 , {TT, TS, ST, SS}, where TT
represents taxing imports regardless of the type of the firms, TS repre-
sents taxing imports when the domestic firm is efficient and subsidizing
exports when it is inefficient, etc. (note that the first letter represents the
strategy of the governments with efficient firms and the second letter rep-
resents the strategy of the governments with inefficient firms). Equilibrium
analysis shows that the strategy pair (TT,TT) is a Bayesian Nash Equi-
librium (BNE) of the policy game (see the appendix). In other words, in
the protectionist environment, taxing imports is the best policy choice for
the governments regardless of the efficiency of their firms under both type
of market structures. Notice here that introducing incomplete information
does not change the equilibrium.
The intuition of the results is clear. Given that the foreign country is
taxing imports, the best policy for the domestic government is to respond
with the same policy. This is because the domestic firm looses an advantage
in the foreign market due to the import taxes. Hence, instead of supporting
this firm in the foreign market, it is better to support it in the domestic
market where it can have substantial profit gains. 6 Thus, in the advanta-
geous position, the profit increase of the domestic firm due to taxation of
imports is more pronounced for the Stackelberg case. In summary, under all
circumstances analyzed above, net welfare gains, when imports are taxed
(given that imports are taxed in the foreign market) 7 , are higher than the
net gains when exports are subsidized.
4 Non-protectionist policies
In this section, we consider an environment where governments can not pro-
tect their firms due to a free trade agreement. They can only impose excise
taxes (or subsidies) without distinguishing the country of origin. In this
environment, the policy choice and level choice stages of the policy game
shrink to one stage. The decisions about the amount of the excise taxes
to be imposed are made before the cost information of the foreign firms is
revealed. The intuition behind this assumption is that before engaging in
a free trade agreement, countries need to know all of the industrial policies
imposed in the partner country. Therefore, countries have to declare their
policy levels before signing the agreement. Under these circumstances, the
social welfare function of the country i at the Cournot market and Stack-
5 A type contingent strategy is defined as a mapping from the type space to the
strategy space.
6This result resembles Neary (1994)'s result, where governments give higher subsidies
to the lower cost firms.
7Net welfare gains are profit gains of the domestic firms plus the tax revenue minus
the consumer surplus losses.
355
elberg market equilibria can be written, respectively, as follows:
where subscript n corresponds to non-intervention and Vi and vi are the

unit excise taxes imposed by country i and country j, respectively. The
expected welfare of country i is
(13)
(14)
respectively, in the Cournot and the Stackelberg markets. Maximizing
Eqns. (13) and (14) with respect to vi of country i and country j, under
each different market structure, we obtain the following dominant strategy
equilibrium excise taxes for the countries with efficient and inefficient firms
for Cournot and Stackelberg markets, respectively:
-i( ) _ (a - ,8)(JLI - 1) -i(,8) _ (a - ,8)JLI

v a - 2 ,v - 2 (15)
.,.,.;, 5JLI(a -,8) - (6a - 5,8).,.,.;, 5JLI(a -,8) -,8

v (a) = 11 ' v (,8) = 11 . (16)
It is easy to see from (15) and (16) that if the domestic firm is efficient
the government chooses to subsidize both domestic and foreign firms in-
stead of taxing them. However, if the domestic firm is inefficient taxing
both firms is always a better policy choice with one exception. This excep-
tion is the case where the domestic firm is a Stackelberg leader and the
cost disadvantage or the probability of facing an efficient foreign firm is
sufficiently low. The intuition behind this result is as follows. Subsidizing
both firms creates welfare gains by increasing consumer surplus due to the
increase in output, and by increasing domestic firm's profit. Obviously, the
total subsidy given to the foreign firm is a loss. When the domestic firm
is efficient, the foreign firm's expected market share is low and thus, the
subsidy loss is not as high as the gains. On the other hand, taxing both
firms creates a gain through the tax revenue, but the effects of taxes on
consumer surplus and the domestic firm's profit are both negative. Since
356
these negative effects of taxation are pronounced for the countries with effi-
cient firms, subsidization is a better policy choice. When the domestic firm
is inefficient, the foreign firm has a larger market share, and therefore, tax
revenues obtained from the foreign firm becomes higher. However, in the
case of subsidization, the loss through the subsidy given to the foreign firm
is much more pronounced. As a result, taxation turns out to be a superior
policy to subsidization. The reason that subsidization can still be a better
policy in the Stackelberg market is because of the domestic firm's lead-
ership advantage. This advantage is bigger when the cost disadvantage of
the domestic firm is not substantial or the probability of facing an efficient
foreign firm is not high (i.e., a - (3 or J..L1 is sufficiently low). In sum, when
the domestic firm is efficient, subsidizing firms is a superior policy choice
to imposing taxes. However, when the domestic firm is inefficient, imposing
excise taxes becomes a better policy choice under most circumstances.
5 Comparison of protectionist and non-protectionist

policies
Suppose that the governments make a choice between joining a (bilateral)
free trade agreement and not joining the agreement but protecting their
firms. We assume that due to a strong enforcement mechanism (punish-
ment), it is impossible to cheat after signing the agreement. Thus, we only
consider the bilateral cases of intervention and non-intervention. The fol-
lowing propositions are the results of the comparison of the protectionist
and non-protectionist policies:
Proposition 1 If the domestic firm is efficient, a country is better off with
the free trade agreement (subSidizing both domestic and foreign firms).
Proof. See appendix.
Proposition 2 If the domestic firm is inefficient, a country is better off
with the free trade agreement (taxing both firms) under both type of market
structures if ~ is sufficiently low.
These propositions tell us that free trade is more likely to be preferred
to intervention. Governments choose not to sign the agreement and protect
their firms only when the domestic firm is inefficient and the cost disad-
vantage is substantial.
Proposition 3 When domestic firms have leadership advantage the con-
dition for countries to prefer free trade is more restrictive.
Proof. The proof is very simple. k 2(J..LI) and k2(/-LI) are the upper limits
of ~ for countries to prefer free trade for Cournot and Stackelberg cases,
respectively, and k 2(/-LI) > k2(J..LI), VJ..LI'
357
As in Neary (1994), the result that protectionism is more likely when

domestic firms have leadership advantages looks paradoxical. This paradox
can be explained by considering the profit shifting effect. In contrast to the
Cournot case, profit gains from the taxation of imports are higher due to
the advantageous position of the domestic firm. As a consequence countries
prefer protection for relatively lower values of ~.
Note that both k2(JL1) and k2(JL1) are decreasing functions of JLl' When JLl
is sufficiently low both of these conditions are not binding, i.e., free trade is
always preferred. When JLl ---t 1, k 2 (JL1) approaches to 17-pn ~ 1.17 and
k2 (JLI) approaches to 154243-2~~g~4996217 ~ 1.09. Therefore, the restriction
on ~ is quite strict for the high values of JLl for both cases, i.e., it is more
likely to observe intervention when JLl is high.
The above propositions describe the environment in which countries are
better off signing the free trade agreement. It is obvious that a free trade
agreement is possible when both countries agree to join. Thus, if ~ < k2 (JL1)
holds for the Cournot case and ~ < k2(JL1) holds for the Stackelberg case,
we should observe free trade in this two-country world. But, it must be
noted that when both countries have efficient domestic firms, we should
always observe free trade.
As we know from Section 3, governments have complete information
at the ex post stage of the trade policy game. We can assume that the
cost information of the foreign firms is also revealed after some interaction
between the countries in the free trade environment. Thus, countries will
have a chance to reexamine their decision using this new information at the
ex post stage. The following proposition gives us an insight into whether
there is a possibility for the countries to regret their decision to sign the
free trade agreement.
Proposition 4 Countries with efficient domestic firms and countries with
inefficient domestic firms facing inefficient foreign firms always prefer free
trade at the ex post stage.
Proposition 5 Countries with inefficient domestic firms facing efficient
foreign firms prefer free trade at the ex post stage if ~ is lower than the
related threshold values for the Cournot market and the Stackelberg market.
The ex post stage comparisons indicate that under most circumstances
countries will not experience regret at the ex post stage when they sign
the agreement in the interim stage. In fact, when they prefer protectionism
in the interim stage it is possible that they may regret it at the ex post
stage. The only situation where they would prefer to protect their firms at
the ex post stage is when a country with an inefficient domestic firm faces
358
an efficient foreign firm, and ~ is higher than the threshold values for the
relevant markets.
Most of the previous strategic trade studies do not consider non-interven-
tion in their models other than the zero equilibrium subsidy/tax rate. Dif-
fering from the others, Hwang and Schulman (1993) treat non-intervention
as a distinct policy choice and show that bilateral non-intervention never
arises as a Nash equilibrium outcome of the policy game. However, under
some equilibrium conditions, one country subsidizes exports while the other
does not protect its firms. Contrary to their findings, our results indicate
that free trade is preferred to intervention in most cases. The difference in
results is not surprising if one considers that our model includes domes-
tic consumption and non-intervention, as a result of a bilateral free trade
agreement. Furthermore, in our free market environment, governments are
able to impose a tax or a subsidy in their domestic markets without dis-
tinguishing the country of origin of the firms.
Strategic trade literature is criticized for being extremely dependent on
the assumptions of the models. For instance, the number of firms and the
market structure are very effective determinants of policy choices. This
dependency on assumptions prevents the results from being of use for pol-
icy makers. Obviously, it is not easy to get information about the market
structure or even about the number of firms, especially if the policy makers
need to know the exact number of foreign firms. Some economists suggest
that it is very likely that a wrong policy will be chosen due to lack of in-
formation. This paper shows that, in fact, being engaged in a free trade
agreement might be the best thing to do for the governments unless the
domestic firms are very inefficient.
As with most of the models in economics, our model has its own limita-
tions. Restricting ourselves to the two type case is one limitation. However,
our intuition suggests that extending the model from the two type case to
a case with continuum of types should not affect the general results. On
the other hand, if we stick to the two type case but increase the number
of firms in each country, the only likely effect will be on the threshold for
intervention. Preferring intervention for a country might require an addi-
tional restriction on the number of firms besides the restriction on the ratio
of efficiencies. The robustness of our results to changing the market struc-
ture from quantity competition to price competition is not clear. Previous
studies show that market structure is important in choosing the optimal
trade policy. For this reason, it might be worthwhile to extend the model
to Bertrand competition to check the sensitivity of our results. Increasing
the policy options of our governments is another possible extension for a
further research agenda.
359
Appendix
Equilibrium analysis of the trade policy game
-==i .. ...-......
We denote W {p~, p3 j ()i,()j), W~{P~, p3 j ()i,()j) as the ex post stage welfare
levels of Country i, in Cournot and Stackelberg m~kets respectively. pi
and pj represent the policies chosen by Country i and Country j and ()i
and ()j are the efficiencies of the domestic and foreign firms.{Note that
these expost stage welfare levels are given in Table 1 through Table 8).
The following conditions are necessary and sufficient for (TT, TT) to be
a Bayesian Nash Equilibrium (BNE) of the policy game in the Cournot
market and Stackelberg market.
=i i=i =i =i
J1.1 W (T, Tj a, a) + J1.h W (T, Tj a, (3) ~ J1.1 W (8, Tj a, a) + J1.h W (8, Tj a, (3)
(17)
W i (T, Tj f3, a)
J1.1 = + J1.h=1.
W (T, Tj f3, (3) ~ J1.1 = 1.)
W (8, Tj f3, a + J1.h =1.
W {8, Tj f3, f3 )
(18)
J1.1 Wi{T, Tj a, a) + J1.1 Wi{T, Tj a, (3) ~ J1.1 Wi{8, Tj a, a) + J1.h Wi{8, Tj a, (3)
(19)
Conditions (17) and (19) are for the country with an efficient firm in
the Cournot and Stackelberg markets respectively. After substituting the
relevant welfare levels from Table 1 through Table 8, these conditions can
be reduced to
r2{136 + 93J1.1) + 297{1 - J1.1) - 204r{1 - J1.1) > 0 (21)

3969 -
r2{ 416964 - 171467J1.1) + 771025{1 - J1.1) - 942492r{1 - J1.1) > 0 (22)
5856864 -
where r = a/ f3.
It is easy to see that (21) and (22) are always satisfied.
Similarly, conditions (18) and (20), which are for the country with an
inefficient firm in Cournot and Stackelberg markets, respectively, reduce to
297r 2J1.1 + 229 - 93J1.1 - 204r 1£1 > 0

(23)
3969 - .
360
771025r 2J.l1 + 245497 + 171467J.l1 - 942492rJ.l1 > 0

(24)
5856864 - .
Since (23) and (24) are also always satisfied, (TT, TT) is a BNE in both
the Cournot and Stackelberg markets. Moreover, one can see that there is
no other pure strategy BNEj therefore (TT, TT) is the unique equilibrium
in both markets.
Proof of Proposition 1. Substituting (15) and (16) into (11) and (12),
for the country with an efficient domestic firm, respectively, we obtain the
non-intervention equilibrium welfare functions in the Cournot and Stack-
elberg markets as follows:
W!.(a) = (a 2(37 - 23J.l1 + 2J.l~) + a{3(34J.l1 - 34 - 4J.l~)

+{32(13 - 15J.l1 + 2J.lf))/36 (25)
(a 2(4994 - 3239J.l1 + 117J.l~) + a{3(5118J.l1 - 4884 - 234J.l~)

+(32(1762 -1879J.l1 + 117J.l~))/3872 (26)
W!.(a) > J.lIW\T, Tj a, a) + J.lhW\T, Tj a, (3) needs to be satisfied for

free trade to be a better choice for country i with an efficient firm in the
Cournot Market. This condition can be simplified to
(r2(107 - 111J.l1 + 18J.l~) + r( -102 + 138J.l1 - 36J.l~)

+18J.ll - 27J.ll +9 > O.
The condition for the Stackelberg market,
W~(a) > J.ll Wi(T, Tj a, a) + J.lh Wi(T, Tj a, (3),

can be simplified to
(r2(606386 - 602031J.l1 + 42237J.l~) - r(617012 - 701486J.l1 + 84474J.l~)

+(57218 - 99455J.l1 + 42237J.l~) > O.
Both of the conditions hold for all J.ll E [0,1] and r E (1,2) •
Proof of Proposition 2. Substituting (15) and (16) into (11) and (12),
for the country with an inefficient domestic firm, respectively, we obtain the
non-intervention equilibrium welfare functions in the Cournot and Stack-
elberg markets as follows:
361
(28)
Free trade is better for country i with an inefficient firm in the Cournot
market if
=-=i -=-=i =-=i
W n(6) > 111 W (T, T; (3, ex) + I1h W (T, T; (3, (3).
This condition reduces to
r2( -9111 + 1811f) - r(66111 + 3611~) + 1811~ + 75111 + 14 0
324 > .
Let us denote the left hand side of the above condition as f(r). The roots
of f(r) = 0 are
11yPl + 6 R + J14(1 + 12111)

yPl(6ILI - 3)
11 yPl + 6 R - -/'-14"-;-(1""-+----:12:--IL-'-I)
yPl(6ILI - 3)
If ILL < ~ then f(r) is a concave function of r, and k1(ILI) < 0 and k 2(ILI) >
1. Thus, f(r) > 0, iff r < k2(ILI)' If ILL > ~ then f(r) becomes a convex
function of r , k2(ILI) < k 1(111) and k1(ILI) > 2. Therefore, again f(r) > 0,
iff r < k 2(ILI)'
Free trade is better for the country i with an inefficient firm in the Stack-
elberg market if
which reduces to
g(r) = (42237ILI + 14981)r2ILI - (532538 + 84474ILI)rlLl
+46592 + 517557ILl + 42237IL~ > O.

The roots of g(r) = 0 are
266269Ji1l + 42237 Ht
+ 8J955902385111 - 10906168)
Ji1l(42237ILI + 14981)
266269yPl + 42237 R - 8y'=95=5"""'90=2"""'38=5'--IL-l----,-,10::-: "9..". ,06::-:-1"""68:-:'")
yPl( 42237ILL + 14981)
g(r) is always a convex function of r. If ILL > 91~;90~2136:5 ~ 0.11, k1(ILI) >
k2(ILI)' Thus, g(r) > 0 iff r < k2(ILI) or r > k1(ILI)' Since k1(ILI) > 2 ,
the constraint r > kl (ILL) is not binding. If ILL < 0.11, then 2 < kl (ILL) <
k 2(ILI), thus, g(r) > 0 always holds. Therefore, we can say that g(r) > 0 if
r < k2(ILI) or ILL < 0.11 . •
362
Proof of Proposition 4. When the domestic firm is efficient and when

the domestic firm is inefficient, but, the foreign firm is efficient, the following
conditions are necessary and sufficient for free trade to be better than
intervention at the ex post stage.
=:i 4r2 65r 2 =:i

Wn(a,a) = 9 > 162 = W (T,T;a,a),
Wn (a, f3) = 15r2 - 12r + 5

18 >
113r2 -102r + 54
162
= Wi(T T· f3)
' , a, ,
- . 4r2 65r 2 -i
W~(/3,f3) = 9 > 162 = W (T,T;/3,/3),
--. 117r 2 325r 2 --.

W~(a, a) = 242 > 722 = W 2
(T, T; a, a),
Wi ( /3) = 274r 2 - 244r + 87 618r 2 - 592r + 299 = Wi(T T· f3)

n a, 242 > 722 ' , a, ,
--. 117r 2 325r 2 --.
W~(/3,/3)= 242 > 722 =W2 (T,T;/3,/3).
It is very easy to see that all of the above conditions are always satisfied. •
Proof of Proposition 5.
Wn (/3 , a ) = 5r2 - 12r + 15

18 >
54r2 - 102r + 113 = Wi(T T· /3 )
162 ' , ,a
is satisfied if, and only if, r < -113.,/23 ~ 1.26.
--Wi (/3 ) = 87r 2 - 244r + 274 299r 2 - 592r + 618 = --Wi(T T· /3 )

n , a 242 > 722 ' , ,a
is satisfied if ~ < -4113J;1~~ ~ 1.11 for the Stackelberg case. There-

fore, free trade is better than intervening in the market when r < 1.26 for
the Cournot case and when r < 1.11 for the Stackelberg case. •.
363
Strategies Payoffs
(8,8) (15a 2
32 '
15( 2 )
32
(8,T) (101a 2
294 '
449( 2 )
882
(T,8) (449a 2
882 '
101( 2 )
294
(T,T) '( ,65a 2 65a 2,)

162 , 162
Strategies Payoffs
(35a 2 -36a§+16§2 16a 2 -36a§+35§2)
(8,8) 32 , 32
(195a 2-170a§+76§2 422a 2 -878a§+905§2)
(8,T) 294 ' 882
(113a 2-102a§+54§2 54a 2 -102a§+113§2)
(T,8) 162 ' 162
(905a 2 -878a§+422§2 76a 2 -170a§+195§2:)
(T,T) 882 ' 294
Strategies Payoffs
(16a 2-36a§+35§2 35a 2 -36a§+16§2)
(8,8) 32 , 32
( 76a 2-170af3+ 195f32 905a 2 -878af3+422f32 )
(8,T) 294 , 882
( 54a 2 -102a§+ 113132 113a 2 -102a§+54§2 )
(T,8) 162 , 162
(T,T) (422a 2-878a§+905§2 195a 2 -170a§+ 76132~)

882 ' 294
Strategies Payoffs
(8,8) ( 15f32 15f32)
32 ' 32
(8,T) (101f3 2 44913 2 )

294 ' 882
(T,8) (449f32 101132)

882 ' 294
(T,T) ( 65 13 2 6513 2. )'

162 ' 162
364
Table 5: ((h, ( 2 ) = (a, a)

Strategies Payoff
(S,S) (350<2 350<2)
72 ' 72
(S,T) (66230<2 41630<2)

16224' 8112
(T,S) (41630<2 66230<2)

8112 ' 16224
(T,T) (3250<2 3250<2.)'

722 , 722
Strategies Payoff
(S,S) ( 1220<2 -1560<IH69{32 690<2 -1560<{3+ 122{32 )
72 , 72
(S,T) ( 127320<2 -106920<{3+4583{32 61320<2 -133800<{3+ 11411{32 )

16224 ' 8112
(T,S) ( 6180<2 -5920<{3+299{32 2990<2 -5920<{3+618{32 )

722 ' 722
(T,T) (114110<2 -13380<{3+6132{32 45830<2 -106920<{3+12732{32.)

8112 ' 16224
Strategies Payoff
(S,S) (690<2_1560<{3+122{32 1220<2_1560<{3+69(32)
72 , 72
(S,T) ( 760<2 -1700<{3+ 195{32 9050<2 -8780<{3+422{32 )

294 ' 882
(T,S) ( 2990<2 -5920<{3+618{32 1130<2 -5920<{3+299{32 )

722 ' 722
(T,T) (4220<2 -8780<{3+905{32 1950<2 -1700<{3+76{32.)

882 ' 294
Strategies Payoff
(S,S) ( 35{32 35(32)
72 ' 72
(S,T) (6623{32 4163(32)

16224' 8112
(T,S) (4163{32 6623{32 )

8112 , 16224
(T,T) (325{32 325{32. )

722 ' 722
365
References
Arvan, L., "Flexibility versus Commitment in Strategic Trade Policy Un-
der Uncertainty: A Model of Endogenous Policy Leadership," Journal of
International Economics, 1991, 31,341-355.
Bandyopadhyay, Subhayu, "Demand Elasticities, Asymmetry and Strategic
Trade Policy," Journal of International Economics, 1997, 42, 167-177.
Brainard, S. Lael and David :Martimort, "Strategic Trade Policy with In-
completely Informed Policymakers," Journal of International Economics,
1997, 42, 33--{)5.
Brander, J. A. and B. J. Spencer, "Export Subsidies and International
Market Share Rivalry," Journal of International Economics, 1985, 18,
83-100.
Cooper, R. and R. Riezman, "lincertainty and Choice of Trade Policy in
Oligopolistic Industries," Review of Economic Studies, 1989, 56, 129-140.
Eaton, J. and G. Grossman, "Optimal Trade and Industrial Policy Under
Oligopoly," Quarterly Journal of Economics, 1986, 101, 383-406.
Hwang, H.S. and C.T. Schulman, "Strategic Non-intervention and the
Choice of Trade Policy for International Oligopoly," Journal of Inter-
national Economics, 1993, 34, 73-93.
Long, Ngo Van and Antoine Soubeyran, "Cost Heterogeneity, Industry
Concentration and Strategic Trade Policies," Journal of International
Economics, 1997, 43, 207-220.
Maggi, G., "Strategic Trade Policy under Asymmetric Information," 1992.
Stanford University mimeo.
Neary, P. J., "Cost Asymmetries in International Subsidy Games: Should
Governments Help Winners or Losers?," Journal of International Eco-
nomics, 1994, 37, 197-218.
Qiu, Larry D., "Optimal Strategic Trade Policy Under Asymmetric Infor-
mation," Journal of International Economics, 1994, 36, 333-354.
Qiu, Larry D., "Strategic Trade Policy under Uncertainty," Review of In-
ternational Economics, 1995, 3 (1), 75-85.
Shivakumar, R., "Strategic Trade Policy: Choosing Between Export Sub-
sidies and Export Quotas Under Uncertainty," Journal of International
Economics, 1993, 35, 169-183.
The canonical extensive form of a game form:
symmetries
Bezalel Peleg1,2, Joachim Rosenmiiller 1 and Peter Sudholter2
1 Institut fUr Mathematische Wirtschaftsforschung, Universit1l.t Bielefeld
D-33615 Bielefeld Germany

2 Center for Rationality and Interactive Decision Theory, The Hebrew
University of Jerusalem, Israel
Abstract. This paper is the first in a series of papers in which we plan to

exhibit to any noncooperative game in strategic or normal form a "canon-
ical" representation in extensive form that preserves all symmetries of the
game. The operation defined this way will respect the restriction of games
to subgames and yield a minimal total rank of the tree involved. Moreover,
by the above requirements the "canonical extensive game form" will be
uniquely defined.
Part I is dealing with isomorphisms of game forms and games. An au-
tomorphism of the game is called motion. A symmetry of a game is a
permutation which can be augmented to a motion. Some results on the
existence of symmetry groups are presented. The context to the notion of
symmetry for coalitional games is exhibited.
JEL Classification Numbers: C70, C72.

Keywords: Games, extensive form, normal form, strategic form
1 Introduction
This paper is the first part of an investigation which is devoted to the
study of the relationship between a game in strategic form and its possible
representations by extensive games. (A representation of a strategic game
G is an extensive game r whose normal form is G.) We should emphasize
that our starting point is a game in strategic form. The transition from the
extensive form to the strategic form as defined by von Neuman and Morgen-
stern (1944) has already been investigated extensively (see Kohlberg and
Mertens (1986) for a recent treatment of this topic). The transition in the
opposite direction is considered "trivial" and conceptually straightforward.
It is the purpose of our work to show that this is not true: The choice of a
method of representation of strategic games by extensive games which re-
spects symmetries of strategic games leads to difficult conceptual problems
368
FIGURE 1. Two representations
and deep mathematical results. Some of the problems will be illustrated by

the well known example of the Battle of the Sexes.
Example 1.1 Consider the following version of the Battle of the Sexes
C 8 C S
(See, e.g., Myerson (1991, p.98) for a verbal description of the game.)
Here C is going to a concert and S is going to a soccer game. Player 1 is the
wife and player 2 is the husband. There are two potential focal equilibria
in this game: (C,C) and (S,S) (see the beautiful discussion in Myerson
(1991, Section 3.5)). Now, according to the present standard convention of
game theory one can represent G by an extensive game in the following two
different ways (see Figure 1).
The foregoing convention which leads to multiple representations has the
following two problematic aspects:
(1) The transition to the extensive form might influence focality. Consider
the extensive game r 1. It is common knowledge in this game that
player 1 moves first. Therefore she has the option to choose C before
player two makes his choice. Thus, it seems to us that in r 1 the pair
(C, C) of strategies is more likely to be played than (8,8). Our feeling
is supported by the experimental work of Rapoport (1994). Clearly,
in r2 the pair (8, S) may be the dominant focal equilibrium rather
than (C, C).
(2) The transition to the extensive form may destroy symmetry. The
game G is "symmetric" in the following sense: It has an automor-
phism, which permutes players 1 and 2 (for a definition of automor-
phism, i.e., an isomorphism of G to itself, see Harsanyi and Selten
(1988, Section 3.4)). This automorphism is given explicitly in our
369
Example 3.6 (1). However, r 1 and r2 are totally asymmetric; more

precisely, if r = r 1or r = r2, then there is no non-trivial automor-
phism of r that respects the temporal ordering of moves in r.
The discussion in the last paragraph leads naturally to the following basic
question:
Let G be a game in strategic form. When is G "symmetric"? (In particular,
is the Battle of the Sexes a symmetric game?)
Quite surprisingly there is no satisfactory answer available to this ques-
tion. If we follow our mathematical intuition and define a strategic game
G to be symmetric if all possible joint renamings of players and strate-
gies are automorphisms of G (see Harsanyi and Selten (1988, P.71) for the
precise definition of renaming), then the class of symmetric games reduces
to the trivial class of all games whose payoff functions are constant and
equal. Also, this definition is incompatible with the definition of symmetric
bimatrix games (see van Damme (1987, p.211)).
Our answer to the basic question is indirect. A symmetry of G, according
to our definition, is a permutation 1[' of the players for which there exists
an automorphism a = (1[','11) of G (here 'P,which is compatible with 1[', is
a renaming of strategies ). Thus, our definition of symmetries (of strategic
games) is different from that of Harsanyi and Selten (1988, p.73). The
game G is symmetric if every permutation of the players is a symmetry of
G. Thus, in particular, the Battle of the Sexes is symmetric according to
our definition. Our definition has the following desirable properties:
(1) The class of symmetric games is a nontrivial interesting class.
(2) It is possible to use similar ideas to define symmetries of extensive
games (see Definition 4.16).
(3) It is possible to compare the symmetry groups of a strategic game
and its coalitional form (see Theorem 4.11).
As far as we could check, symmetries of games in extensive form, which
preserve the partial ordering on the nodes that is induced by the game tree,
were not considered previously. Thus, our treatment of symmetry groups
of extensive games is entirely new.
We now present our solution to the problem of representing the Battle
of the Sexes by an extensive game.
The above representation is faithful because, according to our definitions,
it has the same symmetry group, namely 2:( {1, 2}), as the Battle of the
Sexes. Also, it is quite obvious that it has "minimum" graph in the class
of all faithful representations (of the Battle of the Sexes).
Our goal in this work is to generalize the foregoing construction to all
finite strategic games. This task turned out to be very difficult. We just
mention here two of the highest hurdles.
370
FIGURE 2. The 'canonical Battle of Sexes'
(1) Consider a 2 x 3 two-person game. Such a game has no symmetries.

Therefore, there is no hope to find a "canonical" representation just
for this game. Thus, we have to add the requirement that our repre-
sentations of 2 x 3 games are consistent with our representations of
2 x 2 games. Formally, the only way to get canonical representation
is to consider mappings from game forms to extensive game forms,
that are defined on rich enough domains.
(2) Given a "square" n-person strategic game form, that is a game form
with the property that all the players have the same number of strate-
gies, it is not clear how to find for it a minimal and faithful represen-
tation. (Observe that a square game form allows for complete sym-
metry between the players.) When the number of players is greater
than two, then there is no obvious solution to problem of representing
square game forms. Indeed, we started with the simplest ("atomic")
representations of square game forms and built "symmetrizations" of
such "atoms" in order to obtain faithful (Le., symmetry-preserving)
representations.
We now review the contents of the paper. Game trees and their isomor-
phisms are presented in Section 1. Isomorphisms respect all partitions as
well as the partial ordering of the tree. Section 2 introduces strategic and
extensive preforms. A strategic preform specifies only the set of players
and the strategy sets. Extensive preforms, similarly, specify only the set of
players and the game tree. The definitions of strategic and extensive games
and game forms are also reviewed in Section 2.
Isomorphisms of strategic and extensive preforms, game forms, and
games are introduced in Section 3. Our definition of isomorphisms of ex-
tensive games seems to be new. An automorphism of a game G is called
motion. A symmetry of G is a permutation of the players that is applied
by some motion. After defining formally the symmetry group of a game we
present some results on existence of symmetry groups.
371
The existence of canonical representations of game forms will be investi-

gated in Part H.
2 Prerequisites
The structure of strategic games and game forms will be discussed in Sec-
tion 2. As for extensive games and forms some prerequisites are necessary
which we will deal with presently. Most readers familiar with the topic
could just browse or entirely skip this section.
We start out with a pair (E, -<) where E is a finite set (the nodes) and
-< is a binary relation on E such that (E, -<) is a tree. The root of the tree
is denoted by xo, the generic element is ~ and the set of endpoints is BE.
A play is a sequence
x = (xo, Xl, ... , XT),

satisfying Xi -< Xi+l (i = 0, ... , T -1), such that Xo is the root and XT E BE
holds true. A path is a sequence of consecutive nodes.
The distance of nodes is measured by the rank function defined via
r(xo) = 0 and r(() = r(~) + 1 if ~ -< ~/. Then the nodes
{~I r(~) =t} =: C(E,-<,t)

constitute the level t; thus C is the level function. In addition we shall
employ the notion of maximal rank
rmax(E, -<) := max {r(~) I ~ E E}

as well as that of total rank
R(E, -<) = E r(~).
~E8E
An extensive game (form) is also formulated with the aid of partitions;

we introduce three partitions as follows:
P is the player partition (a partition of E - BE). The names of the players
will be assigned later (Section 2); however, we assume that there
is a distinguished element Po E P (which may be the empty set)
representing the chance moves. All other player sets are assumed to
be nonvoid. Let p = (P~)~EPo be a family of probability distributions
(of chance moves), i.e., let p~ be a probability on the successors of ~
for every ~ E Po. We assume that p~ (~/) is positive for every successor
(of ~.
Q represents the information partition. Q is a refinement of P; thus an
element Q E Q , Q ~ P , is an information set of the player who com-
mands the elements of P. In particular it is required that Q refines
Po to singletons, that is,
372
Q E Q, Q ~ Po =} Q = {O for some ~ E E.
Also, any two elements ~,e' E Q E Q belonging to the same element

Q of Q should have the same number of successors. Moreover, every
information set contains at most one node of any play.
C is a family of partitions representing choices. This item refers to the

above defined structures (E, -<), P, Q. Denote by
C(~) := {e' I ~ -< e'} (1)
the successors of ~ E E and write for Q E Q
C(Q) := UC(~). (2)

€EQ
Now we assume that for any Q E Q we are given a partition C(Q) of

C(Q) such that
8 E C(Q) =} 18 n C(~) 1= 1 (~E Q).
(We use (~ E Q) to indicate the quantification "for all ~ E Q".)

Now C := (C(Q))QEQ denotes the system of choices.
We will refer to the structure (E, -<, P, Q, C,p) as a game tree. In addi-
tion we shall omit Q or p, if P = Q or Po = 0, respectively. That is, we
simplify the notation accordingly, if players have but one decision or there
are no chance moves.
Now we continue by defining various operations to be performed on game
trees. To this end let (E, -<) and (E', -<') be trees and consider a bijective
mapping 4> : E ---- E'.
We shall say that a bijective mapping 4> respects (-<, -<') if
(3)
holds true.
Similarly, if P and pI are partitions of subsets B of E and B' of E',
respectively, then we shall say that a bijective mapping 4> respects (P, PI)
if, for all pI E pI it follows that 4>-l(p l ) E P and also 4>-1 (B') = B hold
true.
Thus, if we want to consider game trees, it is clear in which way a map-
ping respects the player partition and the information structure. To respect
the decision structure is a property defined in a straightforward way.
373
Definition 2.1 A game tree (E, -<, P, Q, C,p) is isomorphic to a game tree
(E' , -<' ,P', Q', c' ,p') if there is a bijective mapping c/J : E ~ E' (an
isomorphism between the game trees) which satisfies the following proper-
ties:
(1) The mapping c/J respects (-<,-<'), (P,P'), and (Q,Q').
(2) c/J(Po ) = Po and p't/>(t;)(c/J(O) = pt;(~') holds true for ~ E Po and

~' E C(~).
(3) For Q E Q the mapping c/J respects (C(Q), C'(Q')), where Q' E Q' is
the unique information set which satisfies c/J( Q) = Q'.
Note that (2) makes sense in view of (1), the bijectivity of c/J, and the
underlying tree structure.
3 Forms and games

There is a definite hierarchy within which we want to approach both "ver-
sions" of a game, the strategic one and the extensive one. This hierarchy
is characterized by the concepts of preform, game form, and game.
The preform is the pure underlying structure listing the environment of
the players' actions.
Definition 3.1 A strategic preform is a pair e = (N, S), where N is a

finite set (the set of players, I N I~ 2) and S = f1iEN Si is the product of
finite sets Si i= (/) (i E N) (the strategy sets).
Thus, the strategic preform determines just the players and the domain
of the payoff functions to be completed later.
Analogously, the extensive preform is described by an environment as
follows:
Definition 3.2 An extensive preform is a tuple
E = (N, E, -<, P, Q, C,p; [), (4)
where N is as in Definition 3.1 and the next six data have been described in
Section 1. Moreover, [ : P - {Po} --t N is a bijective mapping; [ assings
the nodes of PEP - {Po} to [(P) E N, naturally we use [-1 (i) = Pi to
refer to these nodes.
In order to discuss the concept of a game form, we assume that an ab-

stract set of outcomes A is defined, together with a mapping h which spec-
ifies the outcome corresponding to the choice of every strategy profile by
the players.
374
FIGURE 3. An extensive preform
Definition 3.3 (1) A strategic game form is a quadruple g = (ej A, h) =

(N, Sj A, h). Here e is a preform, A a finite set (the outcomes) and
h : S - t A is a surjective mapping called outcome function. g is called
general if h is a bijection.
(2) An extensive game form is a tuple

"(= (€jA,1]) = (N,E,-<,P,Q,C,pjLjA,1]). (5)
Here again, € is an extensive preform and A is a finite set while the

surjective 1] : 8E - t A assigns outcomes to endpoints of the graph
(E, -<).
Thus, we have augmented preforms with "outcomes", "results" of strate-
gic activities, or "alternatives". All of these do not result in "utilities" for
players but constitute a way to describe the consequences of strategic be-
havior.
Turning now to games, we want to convey the notion that payoffs are
assigned in "utils" - and that expected payoffs may occur as the result of
strategic behavior.
Definition 3.4 1. A strategic game is a tuple
G = (eju) = (N,Sju) (6)
such that
(7)
where Ui is player i's utility function or payoff function.
2. Analogously, an extensive game is a tuple
r = (€j v) = (N, E, -<, P, Q, C,pj Lj v) (8)
such that
(9)
375
a C 00
FIGURE 4. An extensive game form
Vi again is player i's payoff depending on endpoints of the graph

(E, -<) .
Remark 3.5 Of course there is a close relation, between games and game
forms. Formally, if g and'Y are (strategic and extensive) game forms and
U: A ---+ /RN
is a ("utility") function defined on outcomes, then
u := U 0 h, v:= U 0 'fJ (10)

induce games G and r which we will conveniently denote by U *9 and U *'Y,
thus we have
(11)
as well as
(12)
There are marked differences between games and game forms with re-
spect to symmetries - even observed at this early stage.
To this end, consider Figure 5. Here the first sketch represents a game
form with A = {a,b,c}. If we put U: A ---+ /R{1,2}, U(a) = (1,1) (a EA),
then the resulting game r = U * 'Y is represented in the second sketch.
It would appear that, whatever our definition of symmetries will be, this
game is completely symmetric.
Now turn to the third sketch. This game in extensive form apparently
has lost some of the symmetries. One can imagine (we haven't defined
the normalized strategic form mapping as yet) that the strategic version
looks exactly like the one of the second sketch - this fact being due to
the possibility of forming expectations with respect to the probabilities
involved. We may perceive that the game of the third sketch stems from
the game form of the fourth sketch. Note however that a possible result of
strategic behavior could be a pair (a, b) reSUlting from a "lottery" choosing
a and b with probability (~, ~) - this procedure would greatly blur the
picture of symmetries of a game.
376
FIGURE 5. Extensive game form and games
4 Isomorphisms, motions, and symmetries

Isomorphisms are first of all defined for strategic preforms: we want to
rename the players and simultaneously reshuffle the ordering according to
which the strategies are listed.
Thus, given strategic preforms e = (N, S) and e' = (N, S'), we consider
bijective mappings 7r: N ~ N and 'Pi : Si ~ S7r(i) (i EN). 7r renames the
players and 'Pi maps strategies of player i into strategies of his double.
The pair (7r, 'P) of course induces a simultaneous reshuffiing of strategies,
i.e., a mapping
Definition 4.1 An isomorphism between strategic preforms e and e' is a

family (7r, 'P) of bijective mappings
7r : N ~ N, 'Pi : Si ~ S~(i) (i E N)
such that
(7r,'P)e = (7r,'P)(N,S):= (7rN,'P7r(S)) = (N,S') (14)
holds true.
Now turn to extensive preforms. Here, we want to rename not only the
players but also the nodes of the graph. Clearly, the decision structure as
well as the information structure should be preserved. But, in addition,
as we have made it clear in Section 2, the "order of play" should not be
disturbed. Thus we come up with the following:
377
Definition 4.2 Let
15 = (N, E, -<, P, Q, C,p;~)

and
E' = (N , E' , -<' , pI , Q' , C' , p'., ~/)
be extensive preforms. A isomorphism between E and E' is a pair of mappings
(-rr, cp) such that -rr is a permutation of N, cp : E - t E' with the following
properties.
(1) cp is an isomorphism between the underlying game trees (cf. Definition

2.1).
(2) -rr(~(P)) = t'(cp(P)) (P E P)
Thus, we write
(-rr, cp)(N, E, -<, P, Q, C,p;~) (15)

._ (-rr(N), cpE, -<, cpP,cpQ,cpC,Pcp-l; -rr 0 ~ 0 cp-l)
(N , E' , -<' ,pi , Q' , C' , pi., ~I)
Now, let us turn to isomorphisms of forms. Next to the structure of the

preform we also want to preserve the nature of the pair (A, h) or (A, 1]),
respectively. As we consider outcomes as basically different, we will regard
a bijective mapping p : A - t A not as particularly relevant. Thus, for
example, we would like to consider game forms represented by
(;!) and G~)

as to be essentially equal, i.e., they are isomorphic; the appropriate mapping
will involve a bijection p: A - t A , a - t a , b - t C - t b. This consideration
motivates the following definition:
Definition 4.3 Let 9 = (e; A, h) and g' = (e ' ; A', hi) be strategic game
forms. An isomorphism of 9 and g' is a triple (-rr, <p; p) such that (-rr, <p) is
an isomorphism between e and e' while p : A - t A' is a bijection satisfying
Here, <p7r is the mapping defined in (13). That is, we have
(-rr, <p; p)(e; A, h) := ((-rr, <p)e; p 0 h 0 (<p7r)-1) = (e' ; A', h').
An isomorphism is outcome preserving if A = A' and p is the identity.

378
Another way of looking at it is suggested by the observation that the

following diagram is commutative.
<p"
S --t S'
h 1 1 h'
A ~ A'
Analogously a reshuffling of outcomes should not disturb the nature of an

extensive form. Since an isomorphism between extensive preforms carries
endpoints into endpoints, the outcome function r/ is defined on elements
<p(~) of the image graph, the result should yield outcomes as previously up
to a bijective p, i.e., we should have p('TJ(~)) = 'TJ'(<p(~)) for ~ E BE.
Thus we have
Definition 4.4 An isomorphism of two extensive game forms 'Y = (E; A, 'TJ)
and 'Y' = (E'; A' ,'TJ') is a triple (Jr, <p; p) such that (Jr, <p) is an isomorphism
between E and E' and p : A ---* A' is a bijective mapping such that 'TJ' o<p = pO'TJ
holds true. That is, we have
An isomorphism is outcome preserving if A = A' and p is the identity.

The reader may benefit from the following diagram:
BE BE'
'TJ 1 1 'TJ'
A ~ A'
Finally let us turn to games. Of course the procedure is now familiar, all
we have to do is to explain the kind of action a permutation (renaming) of
the players formally performs on n-tuples of utility functions, i.e., on u or
v, respectively.
Clearly, if (Jr, cp) is an isomorphism between the strategic preforms e and
e', and u and u' are tuples of utilities defined on Sand S', respectively,
then the utility of i's image 7r( i) E N should be given by
(16)
thus indicating that we rename players and strategies simultaneously.
This defines the action of the pair (Jr, cp) on tuples of utility functions via
«Jr, cp)U)7r(i) (cp7r(s)) .- Ui(S). (17)

We have thus explained
379
(18)
Analogously, within the extensive set-up, if we have an isomorphism
err, r/J) of preforms E and E' , cf. Definition 4.2, and if v : aE
- t JRN is
a utility N-tuple defined on the endpoints of (E,-<), the action of (7r,r/J),
i.e.,
(19)
is given by
(20)
Thus we have
Definition 4.5 (1) Let G = (e; u) and G' = (e'; u') be strategic games.
An isomorphism between G and G' is a pair (7r, cp) such that (7r, cp) is
an isomorphism between e and e' (see Definition 4.1 and (14)) and
u~(i)(cp11"(s)) = Ui(S) (i E N, sE S). That is, we have
(7r, cp)G = (7r, cp)(e; u) = ((7r, cp)e; (7r, cp)u) = (e', u') (21)
(see (17) and (18)).
(2) Let r = (E,V) and r' = (E', v') be extensive games. An isomorphism
between rand r' is a pair (7r, r/J) such that (7r, r/J) is an isomorphism
between E and E' (see Definition 4.2) and v~(i)(r/J(~)) = Vi(~) (~ E
aE, i EN). That is, we write
(7r, r/J)r = (7r, r/J)(E; v) = ((7r, r/J)E, (7r, r/J)v) = (E', v') (22)
(see (19) and (20)).
So far we have finished the presentation describing isomorphisms between

preforms, game forms, and games.
We are now in the position to discuss symmetric games. At first, this
notion might seem to be obvious: one is tempted to argue that a symmetry
of a game G is just an automorphism of G , i.e., an isomorphism (7r, cp), 7r :
N - t N, CPi : Si - t S11"(i) , such that
(7r,cp)G=G
holds true. However, this approach will not work appropriately. Keep in
mind that we want to speak about symmetries with respect to players, for
good reasons a rearrangement of the order of the of strategies is irrelevant.
The reader is referred to the introduction for a discussion of this viewpoint.
380
Example 4.6 (1) The "Battle of the Sexes" represented by the following
pair of matrices
(23)
allows for two "symmetries '. Consider the automorphism (7r, 'P) given
by 7r = id, 'Pi = id (i = 1,2) as well as the one given by 7r : 1 ~
2 ~ 1 and 'Pi(Si) = 3 - Si (Si E Si = {1, 2}).
(2) If G is indicated by
(24)
then we feel that there are absolutely no symmetries with respect to the
players. Nevertheless, there are nontrivial automorphisms of G, e.g.,
7r = id combined with 'Pi : Si ~ Si, i.e., 'Pi : {1, 2, 3} ~ {1, 2, 3} ,'PI:
1 ~ 2 ~ 3 ~ 1 and 'P2 = id, i.e., if player 1 renames his strategies
it does not change the game but this exhibits no "symmetry" of the
game.
Thus, automorphisms cannot be "symmetries" and "symmetries" should
disregard the pure rearrangement of strategies which does not involve inter-
changing players. Or more to the point, if we rename players and strategies,
the "symmetry" involved should only reflect the "essential" similarities of
players.
Definition 4.7 A motion of a strategic game G is an automorphism (7r, 'P)
of G. A motion (7r, 'P) is impersonal if 7r is the identity (and 'Pi : Si ~
Si(i EN)).
Remark 4.8 It is not hard to see that motions enjoy the structure of a
group. Indeed, if (7r, 'P) and (0", 1/J) are motions of a strategic game G, then
define 1/J ® 'P via
(25)
such that
('P ® 1/J)i : Si ~ SU(7r(i» = Suo7r(i) (i E I). (26)
The product of (7r, 'P) and (0", 1/J) is then
(0", 1/J)(7r, 'P) = (0"7r,1/J ® 'P) (27)

Clearly, the unit motion is given by
381
(id, id) = (idN, (id s ;)iEN )

This way the motions of G constitute a group. Clearly, the impersonal
motions constitute a subgroup.
Remark 4.9 The subgroup of impersonal motions of the group of motions
of a game is normal.
The proof of this statement will be left to the reader. However, it now
becomes conceivable that "disregarding the impersonal part of a motion"
as suggested by Example 4.6 mathematically amounts to introducing the
quotient group of motions with respect to impersonal motions.
Definition 4.10 Let G be a stmtegic game. M = M(G) denotes the group
of motions. I = I( G) ~ M denotes the subgroup of G constituted by the
impersonal motions. The group of symmetries of G is the quotient group
M(G)
S=S(G) = M/I = I(G)' (28)
A game G is symmetric if its symmetry group is isomorphic to the full

group of permutations of N called E(N), i.e., if
S(G) ~ E(N) (29)

holds true.
We would like to further motivate the definition of symmetries by point-
ing out that there is a natural relation between the concept of a symmetry
we have developed so far for noncooperative games and the "natural" con-
cept "coalitional" or "characteristic" function.
To this end, let G = (N, S; u) be a strategic game, and let (-rr, ip) be a
motion of G. We consider the coalitional game (N, v) which is derived from
G in the standard way (see, e.g., McKinsey (1952, Chapter 17).
Theorem 4.11 Let (-rr, ip) be a motion of a stmtegic game G. Then 7r is a
symmetry of (N, v), that is, v(7r(T)) = v(T) for all T ~ N.
Proof. Let T ~ N, T -I 0, N. For Q ~ N, Q -I 0, N, denote SQ = I1iEQ Si.
Then b.(ST)(b.(SN\T)) is the set of all correlated strategies of T(N\T).
By definition
v(T) = max min

l1EA(S.r) TEA(SN\T)
Now let i E T, a E b.(ST) and T E b.(SN\T)' Then
Ui(lT, T) = L L a(s)T(t)Ui(S, t).

SEST tESN\T
382
Furthermore let us tentatively use the notation
(Ct'i)iET =: CP'T and (CPi)iEN\T =: CPN\T'
Then CP'T : ST --t S1T(T) and CPN\T : SN\T --t S1T(N\T) are bijections and for
each i ET,s EST, and t E SN\T
holds true.
Hence, for a E ~(ST) and r E ~(SN\T) we obtain
L a(s)r(t)ui(S, t) =
The mapping CPT : ~(ST) --t ~(S1T(T») given by cpT(a)(s)

a((cp'T)-l(s)) is an affine isomorphism between ~(ST) and ~(S1T(T»)' Sim-
ilarly we define the affine isomorphism CPN\T : ~(SN\T) --t ~(S1T(N\T»)'
Using these notations we obtain
L L L a(s)r(t)ui(S, t) = L ui(a, r)
iET SEST tESN\T iET
L Ui(cpT(a) , CPN\T(r)).
iE1T(T)
Therefore it follows that
v(T) = max min

<7E~(ST) TE~(SN\T)
max
<7E~(ST)
mm
TE~(SN\T)
L Ui(cpT(a), CPN\T(r)) =
iE1T(T)
max
aE~(S"'(T)
min
TE~(S"'(N\T)
L Ui(o-, f) = v(1l'(T))
iE1T(T)
holds true in view of the fact that CPT and CPN\T are bijections. •
383
Corollary 4.12 The symmetry group of G is contained in the symmetry

group of (N, v).
Example 4.13 Let G be given by (0,0)(0,0). Then the symmetry group of

G is trivial (ISll =1= IS2J), and (N,v) is symmetric. Thus, the Corollary is
sharp.
Remark 4.14 One obtains similar results by considering the a or f3 NTU

coalitional games which are derived from G.
Example 4.15 Next, let G be described as follows. There are 3 players,

each of them having 2 strategies. Player 1 chooses strategies from SI =
{small, BIG}. His payoff is always -1, and does not depend on the choice
of strategy triples. The payoffs of players 2 and 3 in the two situations
player 1 may choose are represented as follows; player 2 chooses t(op) or
b(ottom) and player 3 chooses l(eft) or r(ight).
r 1 r
(~~) (~n
t
small
b
(30)
1 r r
(:~) (;~)
t
BIG
b
The symmetries of the Battle of the Sexes are to some extend revived
within this example; player 2 may play his role in the "small" version of a
modified Battle of the Sexes as well as player 1 in the "BIG" version.
More precisely, there is a (personal) motion (7r, 'P) given by 7r : 1-t
1 , 2 -t 3 -t 2 , 'PI : BIG -t small -t BIG; 'P2: t -t r , b -t 1 ; 'P3 :
r -t t , 1 -t b, which leaves G untouched. This motion involves player l's
exchange of strategies. It can be seen that the only further motion is (id, id)
(one cannot exchange player 1 with any of the others and exchanging the
others, while player 1 remains fixed requires the above 'P)'
Thus, the quotient group is {id, 7r} or, loosely speaking, "players 2 and
3 are symmetric". Thus, motions have to be incorporated in the definition
of symmetries but, as the next example shows, this is not a sufficiently
comprehensive definition depending on the context.
On the other hand consider the following example where a strategic game
G for 2 players is indicated by
384
1 2 1 2
1 1
2 2
G admits of exactly four motions (1r, cp). Indeed, if we denote the

nontrivial permutation of two indices by T : 1 ~ 2 ~ 1 and
the trivial permutation by id, then we can list these motions to be
(id, (id, id)), (id, (T, T)), (T, (id, T)), and (T, (T, id) ).
Thus, we have four motions, but G enjoys only two symmetries (the first
and second pair of motions coincide on the player set). It should be pointed
out that, contrary to the "Battle of the Sexes" example (cf. Example 4.6),
both the nontrivial motions (T, (id, T)) and (T, (T, id)) are noncyclic, i.e.,
Tk = id for some k E IN does not imply that (cp7y(S) = S holds true for
all S E 8.
Within the framework of extensive structures, the definition of symme-
tries is now rather analogous to those given in the context of strategic
structures.
Definition 4.16 Let f = (E, v) be an extensive game. A motion (1r, cP) of

f is an automorphism of f and M = M(f) denotes the group of motions.
I = I(f) is the normal subgroup of impersonal motions, i. e., automor-
phisms of shape (id, cP). The group of symmetries is the quotient group
S = S(f) = M/I = M(f) (31)

I(r)
and f is symmetric if S(r) ~ 'L.(N) holds true.

In the remainder of this section we will shortly indicate a characterization
of all possible groups of motions. Let G = (N, 8; u) be a strategic game
and let M be the set of motions of G. We define an equivalence relation
"-' M on B = N x 8 in the following way:
(i,S)"-'M(j,t) ifthere exists (1r,cp) M such that (1r(i),cp11"(s)) = (j,t)

E
(32)
"-'M is an equivalence relation because M is a group.
Theorem 4.17 M is covering, that is, if (1r, cp) is an (n + 1)-tuple of

bijections, 1r : N - t N, CPi : 8 i - t 811"( i) (i E N), such that
(1r(i), cp11"(s))"-'M(i, s), for all (i,s) E B (33)
then (1r, cp) EM. (Here and in the sequel we use n =1 NI.)
385
Proof. By (32) and (33) for each (i,s) E B there exists (T,p) E M such
that (T(i), pT(S)) = (1I"(i) , <p7r(s)).
Hence,
U7r(i) (<p7r(s)) = UT(i)(pT(S)) = Ui(S),
that is, (11", <p) is a motion of G. •
We now prove the converse result. Let e = (N,8) be a strategic preform.
A transformation of e is an (n + 1)-tuple (11", <p) of bijections, 11" : N --+
N, <Pi : 8 i --t 87r( i) (i E N), that is, it is an automorphism of e. Denote by
Aut( e) the group of all transformations. For H C Aut( e) we define'" H by
32 (with M replaced by H). Similarly, H is covering if each (11", <p) E Aut(e)
that satisfies
(1I"(i) , <p7r(s))"'H(i, s), for all (i, s) E B, (34)
is a member of H.
Theorem 4.18 If a subgroup H C Aut(e) is covering, then there ex-
ists a vector u H of payoff functions for e, u H : 8 --t RN, such that
M(N,8;u H) = H.
Proof. Let It, ... , h be the equivalence classes of "'H. Choose real numbers
Cl < ... < Ck and define u H : 8 --t RN by Ui(S) = Cl if (i,s) Eh
Claim 1. If (11", <p) E H then (11", <p) E M(N, 8; u H ). Indeed, let i E N

and s E 8. Then (1I"(i) , <p7r(s)) "'H (i, s). Hence U7r(i) (<p7r(s)) = Ui(S) by the
definition of u H .
Claim 2. If (T,'Ij;) E M(N,8;u H ) then (T,'Ij;) E H. Indeed, for
every (i,s) E B we have UT (i) ('Ij;T(S)) = Ui(S). By the choice of
u H , (T(i), 'IV (s))'" H(i, s). Because H is covering, (T, 'Ij;) E H. •
Given a strategic preform e = (N, 8) some subgroups of Aut( e) may not
be the group of motions of (N, S; u H ) for any u H : S --t RN.
Example 4.19 Let N = {I, ... , n}, and 8 1 = ... = 8 n = {I, ... , l} where
2 S l S n - 2. Denote by An the group of even permutations of N. Then
H = {(11", (id, ... , id)) 111" E An}

is a subgroup of Aut( e) which is not covering. Indeed, if 11" is an odd per-
mutation, then (1I",(id, ... ,id)) is not in H. However, for each (i,s) E B
there exists T E An such that (11" (i), id7r (s)) = (T( i), idT(s)), that zs,
(1I"(i) , id7r(s))"'H(i, s).
The following result is a corollary of Theorems 4.17 and 4.18.
Corollary 4.20 Let e = (N,8) be a strategic preform and let He 2::(N)
be a group of permutations. Then there exists a vector of payoff functions
u H : 8 --t RN such that H = S(N, 8; u H ) ~f and only if there is a covering
group H* of automorphisms of e such that H = H* /I(H*), where I(H*)
is the (normal) subgroup of impersonal automorphisms in H*.
386
Proof: Necessity. If H = S(N, S; u H ) then

H = M(N,S;u H )
I(N,S;u H )
and M(N,S;u H ) is covering by Theorem 4.17.

Sufficiency. If H = H* jI(H*) and H* is covering then, by Theorem 4.18,
there exists a vector of payoff functions u H : S - t RN such that H* =
M(N, S; u H ). Because H = H* jI(H*), H = S(N, S; u H ) by definition. •
Example 4.21 Let, again, N = {I, ... , n} and SI = ... = Sn = {I, ... , l}
with l ~ n - 1. We claim that for every subgroup H of 2:(N) there exists
a vector u H : S - t RN such that H = S(N, S; u H ). Indeed, let H c 2:(N)
be a group. Consider the following group H* of automorphisms of e
H* = {(-rr, (id, ... , id)) Irr E H}
We shall prove that H* is covering. Indeed, let (7r,(ipl, ... ,ipn)) satisfy
(7r,ip7r(s))""'"'H.(i,s) for all (i,s) E B. Consider a pair (i,sm) where i E N
and Sm = (m, ... , m). By assumption
(7r(i), ip7r(sm)) = (7r(i) , (ip7r-1(1) (m), ... , ip7r-1(n) (m)) ""'"'H. (i, sm)
for all i EN. Therefore ip7r-1(i)(m) = m for all i E Nand 1 ~ m ~ l.
Thus ipl = ... = ipn = id. Now choose (i, s) = (n, (1,2, .... , n - 1, n - 1)).
By our assumption there is fr E H such that
(7r(n) , i~ (s)) = (fr(n) , id* (s)).

Therefore, 7r = fr. Thus H* is the group of motions of some game (N, S; u H )
whose group of symmetries is clearly H.
References
van Damme, E. (1987). Stability and Perfection of Nash Equilibria.
Springer-Verlag, Berlin.
Harsanyi, J.C. and Selten, R. (1988). A General Theory of Equilibrium
Selection in Games. MIT Press, Cambridge, Mass.
Kohlberg, E. and Mertens, J.-F. (1986). "On the Strategic Stability of
Equilibria". Econometrica 54, 1003-1037.
McKinsey, J. (1952). Introduction to the Theory of Games. McGraw-Hill,
New York.
Myerson, R.B. (1991). Game Theory. Harvard University Press, Cambridge,
Mass.
von Neumann, J. and Morgenstern, O. (1944). Theory of Games and Eco-
nomic Behavior. Princeton University Press, Princeton.
387
Rapoport, A. (1994). "Order of Play in Strategically Equivalent Games in

Extensive Form". University of Arizona, Thcson, AZ 8572l.
RosenmUller, J. (1981). The Theory of Games and Markets. North-Holland
Publishing Company, Amsterdam.
Bargaining in a changing environment*
Clara Ponsati1 and J6zsef Sakovics2
1 CODE i Departament d'Economia i Historia Economica, Universitat
Autonoma de Barcelona 08193 Bellaterra, Spain

2 Institut d'AnaJisi Economica (CSIC),Campus UAB, 08193 Bellaterra, Spain
Abstract. We study bilateral negotiations where both parties have outside

options of uncertain value. First, we argue that the relevant attribute of
an outside option is the time by which its value is revealed rather than the
time up to which it is available. Second, we show that a random outside
option is potentially worth much more - and never less - than its straight
expected value. Third, we prove that, rather than by threats, delays might
be caused by the bargainers' waiting for their outside option to become
available and/or to reveal its value. Finally, we point out that as additional
outside options are added to the opportunities of a bargainer, the strategic
advantage of the original outside options may decrease.
JEL Classification Numbers: C78
Keywords: Random outside option, bargaining
1 Introduction
Bargaining does not take place in a vacuum, but rather in a changing
environment. The opportunities available to players if they decide to quit
bargaining change, and the options' known or uncertain characteristics also
may vary over time. In this paper we provide insight on how bargaining
develops in changing environments by analyzing a bargaining game where
both players enjoy outside options of uncertain value that arise and cease
to be available over time.
We will first argue that the relevant attribute of an outside option is
the time by which its value is revealed rather than the time up to which
it is available. Second, we will show that a random outside option is po-
tentially worth much more - and never less - than its straight expected
·We have benefited from the comments and suggestions of two anonymous referees on
an earlier version of this work. Support to Ponsati through research grants PB-96-1192
from nGCYT and SGR-96-75 from CIRIT and to Sakovics through HCM program of
the EC contract CHRX-CT94-0489 "Games and Markets" and from nGCYT through
grant PB93-0679 is gratefully acknowledged.
390
value. Third, we will prove that, rather than being caused by threats, de-
lays might be caused by the bargainers' waiting for their outside option
to become available and/or to reveal its value. Finally, we will point out
that the prospect of several future outside opportunities is not necessary
better than the prospect of a single one, since as additional outside options
appear the strategic advantage of the original ones may decrease.
Here we propose a sample of four stylized bargaining stories consistent
with the distinctive elements of our model, i) that the value of outside
opportunities is uncertain, ii) that both players enjoy the possibility of
these opportunities, and iii) that strategic capabilities depend crucially on
the dates the options appear and mature and on the dates uncertainties
are removed.
1) Arbitration: The Teachers Union bargains with the School Board of

City A. Each party can unilaterally call for an arbitrated decision. The
arbitrator has been nominated recently and has no record on school cases.
However she has been called to arbitrate in a similar case in City B and
will be announcing a decision in 10 days.
2) Sports: Soccer Star A bargains over his contract with team B while
he is also waiting to hear an offer from another team next week. Medical
reports on another player are also due next week. Depending on these
medical reports, the alternative player will or will not be fit to play as
a perfect substitute of A. Agreements must take place before the season
starts in two weeks.
3) Home-ownership: Buyer B and seller S are negotiating over the sale
of S's home. There is another house that B likes but its owner is away on
a trip until next week. S has an appointment with another potential buyer
also next week.
4) Exclusive vs. standard designs: Two firms, F and f, bargain over a con-
tract that could make f the provider of an exclusively designed component
in F's production. There is a competitive market for standard components
where F and f can buy and sell respectively. Both firms can adapt their pro-
duction to standard or exclusively designed components, but not to both.
Both firms have complete information on the costs and benefits of the ex-
clusive design alternative. In contrast, there is uncertainty on how well the
standard design adapts to F's needs, and there is uncertainty on f's costs
to produce the 'standard' component. The technical departments at both
firms are carrying out research to remove this uncertainty and their reports
are due on day D.
With uncertainty about the sizes of the outside options, as in all our
bargaining stories, the most relevant question is: When does the uncertainty
resolve? A moment of reflection suffices to see that if it is never resolved
391
then -in the absence of risk aversion 1_ the uncertain options would be
strategically equivalent to fixed options with a value equal to the expected
value of the former ones. Therefore, a crucial element of our model is the
revelation date, a parameter which gives the time period at the beginning of
which the uncertainty is resolved, via a random draw according to the prior
distribution. In addition, we introduce two more dates that are significant
in an option's life: the first date it is available and the last period in which it
can be taken, its maturity date. Moreover, in a bargaining situation which
is symmetric - in the sense that both parties can make proposals - we find
it more realistic that both players be offered outside opportunities. Moving
from one-sided to two-sided outside options has dramatic effects2 and the
combination of uncertainty and two-sided options yields results that are in
sharp contrast with the earlier results of Shaked and Sutton (1984) (where
the "outside option principle" 3 was formulated) and Shaked (1994)4, where
one-sided, certain outside options are added to a Rubinstein game.
The maturity date makes our model practically finite, since the subgame
following it is the original Rubinstein game, which has a unique subgame-
perfect equilibrium. This makes possible the use of backward induction to
resolve the game, yielding a unique solution despite the two-sided options
environment. We show that, apart from this important conceptual effect
- as long as it is posterior to the revelation date - the magnitude of the
maturity date is irrelevant.
The outside options affect the nature of equilibrium in several ways.
Players can take their outside options before their values are revealed if
their distributions are sufficiently attractive, for the outside option's value
is not discounted after it is selected. Further the players may wait to learn
their outside option before negotiating. Finally, even a mid level outside
option value may lead to an extreme division of surplus. Once it is credible
for the proposer to accept her outside option after the current period, then
the responding player accepts any offer equal to her reservation value or
1 Although the effect risk attitudes on the play of the game is certainly of interest, in
this paper we assume risk neutrality, and concentrate on the strategic relevance of an
"ill-specified" threat without considering the risks involved.
2In Ponsati and Sakovics (1998a) we make this point in a model without uncertainty.
3 This principle says that unless the solution without the possibility of opting out
gives a lower payoff to one of the players than her outside option, the outside options
do not affect the outcome of the negotiation. Note that as Sutton (1986) and Dalmazzo
(1992) show, the OOP does not hold in such a neat way if the model is augmented by
an exogeous risk of breakdown or a shrinking cake. However, it is generically true that
the outside options cannot be identified with the disagreement point.
4He shows that if it is the proposer who can threaten to take his outside option,
the strategic consequences are markedly different because if the outside option exceeds
his continuation value in case he does not leave the game he can appropriate the entire
surplus, by making a take-it-or-leave-it offer. Thus, there exist a range of outside options
(strictly between zero and one) for which there exist multiple equilibria.
392
better.
The main insight of our model is that an uncertain outside option has
many of the properties of a stock option. The strategic value of the outside
option is a function of the right-tail of its cumulative distribution. The
distinction between outside options and stock options is that an outside
option is never exercised (if the values of the outside options leave room for
surplus), so that the exact value of the outside option is not important so
long as it is above a threshold to make its acceptance a credible threat. The
potential for exercising an option at a later date if its value is sufficiently
large increases the payoff for a given player in any agreement.
As an extension, we examine the case when a player can choose among
several outside options. Our main findings here are that whether an option
has an effect on the outcome depends exclusively on the upper end of the
support of its distribution and that the presence of an alternative option
decreases the strategic value of a given outside opportunity.
A complementary, more general, motivation for our work is the obser-
vation that in the real world many "games" are played at the same time
(possibly involving the same player in multiple games) and therefore the
evolution of game A has direct effects on a player's preferences (and there-
fore her strategy) in game B. In our Soccer tale, A may expect a negotiation
with the alternative team to start after hearing a first offer. In our home-
ownership tale, the owner of the alternative house and B (or S and the
potential buyer that will visit her next week) may e._ter a negotiation. In
both cases these negotiations will influence and be influenced by the events
in our main story. Of course, the straightforward way to deal with such situ-
ations would be to model the set of interacting games as a unique game thus
internalizing all the externalities. However, we believe that understanding
simple, two-person games can provide important insight that will help in
understanding of the more complicated games. 5 Therefore, our approach
is to concentrate on bilateral bargaining while taking into account at least
the most important cross-effects. 6 In this paper, these boil down to the
consideration of random outside options.
Finally, a few words on related literature. Models of strategic bargaining
with symmetric but imperfect information are very scarce. Vislie (1988)
5We do not wish to play down the importance of the research carried out on decen-
tralized markets (see Osborne and Rubinstein (1990) for a survey of the pre-llilleties
literature; see also Hendon and Tranaes (1991), Moldovanu (1993) and Hendon et al.
(1994)). However, the primary concern of this literature is to see whether these markets
implement Walrasian equilibria and if not how efficient they are, while they treat the
fine details of the externalities in a fleeting manner.
6 One of the obvious effects of this genre is the consequence of bounded rationality:
players with limited memory, computing power, etc. will have to "ration" their resources
among the games they are simultaneously playing. Without discounting the importance
and interest of this type of considerations, in this paper we would like to concentrate on
another aspect. of t.he externalities.
393
adds uncertainty about the presence of a second potential worker in the

model of Shaked and Sutton (1984) and derives the corresponding unique
equilibrium. Avery and Zemsky (1994a) consider a model in which ex-
ogenous shocks affect the gains from trade. The outcome of each shock is
revealed after the proposer made her offer but before the responder's move.
Consequently, the responder can use this "private" information to reduce
the proposer's "first mover" advantage. In equilibrium, the first proposal
is not accepted for all possible realizations of the shock and therefore there
will be delay with positive probability before agreement occurs. This in-
efficiency, in turn, may cause multiple equilibria, and hence deterministic
delay, according to the Money Burning Principle (see, Averyand Zemsky
(1994b)). Finally, Merlo and Wilson (1995) also analyze a stochastic model
of sequential bargaining, although they only consider the variability of the
available surplus and do not allow for (random) outside options, they also
obtain equilibria where the players disagree in some states of the world
because they both expect to improve their shares in the future.
The paper is organized as follows. Section 2 presents the set up, defini-
tions and instrumental results that are used for characterizing equilibria in
Section 3. Section 4 addresses the model with multiple successive outside
options. Section 5 concludes.
2 The strategic value of a random outside option

As mentioned above, we use the Rubinstein procedure as the base model of
our bargaining process. In addition, we assume that there is an outside op-
tion 7 available to each player that they can take at the end of every period.
The values of the options, a = (aI, (2), are random variables, distributed
according to the joint cdf F : [0, 1] x [0, 1] -+ [0, 1], which is common knowl-
edge between the players. The unconditional expected value of the option
of Player i is denoted by E i . In addition, the options are characterized by a
three-tuple - which, for simplicity, we assume is the same for both of them:
{Ta, (:::;), Tr, (:::;), Tm (00 ) }. Ta is the first period in which the options are
available. If a player takes her option at any time before it is available, her
payoff is discounted up to this time. At the beginning of period T r the un-
certainty is revealed, and the realization of the random variables becomes
common knowledge. Finally, T m is the maturity date of the options: if they
have not been taken at T m or before, they cease to be available.
Let us start the analysis by defining the most crucial attribute of a ran-
dom option:
Definition 1 The strategic value of the random option of Player i, Ai, is

the utility he expects to obtain in the unique subgame-perfect Nash equilib-
7For brevity's sake, from here on we will refer to it simply as an option.

394
rium of the subgame starting at TT.8

As Lemma 1 demonstrates, the strategic value of an option is always
well defined, and it is never less, and is often significantly more than its
expected value. In order to clarify the intuition behind the lemma, we first
present the result for the case where only one player has an option.
Lemma 1 If only Player 1 has an option, its strategic value to he~ is
i) If TT = T m ,
if 1 proposes at TT,
otherwise.
ii) If TT < Tm,
if 1 proposes at TT,
otherwise.
Proof. It follows from Lemma 2 below. •
Note that when the maturity date coincides with the revelation of the
realization of the option,10 if there is an arbitrarily small positive proba-
bility that the option realizes a value above that period's Rubinstein share
(or, if TT is odd, an even lower value) then, irrespective of the rest of the
distribution of her option, Player l's continuation value will exceed what
she would obtain in the Rubinstein game starting in period TT. Therefore,
a random outside option is potentially worth much more - and never less
- than its straight expected value; hence it is to be evaluated at its "cer-
tainty equivalent": the option's strategic (or conditional) expected value.
8 Note that this definition refers to the gross effect of the option, taking into account
the continuation value that would result in its absence.
9Note that when only one player has an option Al + A2 == l.
IOIf the revelation date preceeds the maturity date the situation is less clear cut, but
it is qualitatively similar.
395
It is interesting to observe that while this is effectively an "option value"

effect, the parallel with options in financial markets applies with the roles
reversed: it is the continuation value not the outside opportunity that plays
the role of an option (understood as a derivative security). That is, when
a player buys the opportunity to get to an agreement with his bargaining
partner after he learns about the value of his outside option, he is effectively
buying an option not to have to take his outside option.
Lemma 2 If both players have options, their strategic values to the players
are (i being the proposer in the period ):11
i)IfTr = T m,
Aj=1!8F(1~8'1!8)+ ff
Xj>-rt&,xi>fFo,Xl +x2::S1
xjdF(x)
A j =1!8 F (1:8'1!8)+ II
Xj,xi>fFo,Xl +X2::S 1
xjdF(x)
Proof. If the sum of the realizations of the options, al + a2, exceeds one,
then at least one player will prefer taking her option right away to any
other feasible equilibrium payoff, and thus the result obviously holds true.
Otherwise, we have two cases to consider, Tm even and Tm odd. If is even,
the option matures at a date when Player 2 makes the offer. Now, observe
that, since after this date the game reverts to a standard Rubinstein game,
in the last period of the options' life Player 1's unique subgame-perfect
equilibrium payoff is al if a2 2: l~/j' and max{ aI, l!/j} otherwise. Similarly,
if is odd, Player l's unique subgame-perfect equilibrium offer (which is
always accepted) is 1 - a2 if al 2: l~/j and min{l - a2'1!/j} otherwise.
To see this, note that if a player can threaten to opt out in case her offer
llWe adopt the convention that the first argument of F(.,.) is Player i's option.
396
is not accepted, she can claim the whole surplus - less the other players'
option value -, whenever such a threat is credible. 12 Moreover, note that the
continuation value of the proposer at Tm in case she stays is her Rubinstein
share in the next period, which has a present value of 1~{j for her. Thus
opting out is credible if and only if her outside option is larger than this
value. If taking the option by the proposer is not credible then, by the
Outside Option Principle, that threat simply has no effect and either the
responder earns the value of his option or the Rubinstein outcome prevails.
This completes the proof of i).
If Tr < Tm, using the Shaked-Sutton backward induction argument, we
can derive the players' continuation values at Tr. We have two cases to
consider, depending on whether the realized options are above or below
the threshold value for credibility for the proposer's option, 1~{j.
If neither option is credible, then, by the Outside Option Principle, they
do not affect the negotiation, since 1~{j ~ 1!{j. Therefore, the players'
continuation values at T r are simply their appropriate (depending on who
makes the offer at Tr) Rubinstein shares.
If at least one player's option exceeds 1~{j' then in all periods between
(and including) Trand T m - 1, the unique subgame payoff for Player i is
1 - aj if she is the proposer and ai otherwise. To see this, first note that,
if the claim is true a period then it is also true for the previous period,
since the options will always be credible for the proposer (he expects aj in
the next period, which has a discounted value of 8aj ~ aj). As we saw in
the first paragraph of this proof, if the proposer's option is credible, the
unique subgame equilibrium is the one posited above. Thus, all we have
left to prove is that for period T m - 1 the result holds also. To see this,
let j be the proposer in that period. This implies that in period T m i's
continuation values are ai or max{ai' 1!{j}' whenj's option is credible and
when it is not, respectively. Then i's option is credible in period Tm - 1,
since 8ai ~ ai, and since 1~{j ~ ai, whenever j's option is not credible, by
assumption. •
Lemma 2 is an interesting result on its own right, since in its proof the
unique subgame perfect equilibrium of a complete information alternating-
offer game with a finite lived outside option for each player is characterized.
Note that the maturity date is only relevant insofar as it is equal or not
to the revelation date, independently of the identity of the last proposer
before the option expires. Moreover, when the revelation date is different
from the maturity date, the only difference is that the responder needs a
lower realization of her option to influence the outcome. Otherwise, the
maturity date is irrelevant. In fact, the exact date of maturity need not
be known to the players to obtain our results, which can be extended for
12This argument was first put forward in Shaked (1987/1994).

397
any (common knowledge) distribution over maturity dates that has a finite
support.
Observe also that, even though we allow for opting out in every period,
we still obtain a unique equilibrium, unlike Shaked (1987/1994). There, one
player has an option, and that player's choice to exercise the option can be
credible in some but not all periods. Here, since both players have outside
options, if the option is a credible choice by one player in a given period,
then it is also a credible choice by the other player in the previous period
(since he will be offered only the option next period, the option today is
a better payoff and thus taking it is a credible threat). Thus if an option
is to be credible in the future, then both options are also credible at all
earlier stages of the game. This, in principle, does not rule out multiplicity:
in Ponsati and Sakovics (1998a) both players can opt out in every period,
their options are credible and still there is multiplicity. The difference,
however, is that our present model is not stationary (it is effectively finite
horizon) and therefore the continuation values are fixed (that is, unique)
in period TT, which, by backward induction, yields a unique outcome in all
previous periods too.
3 The SPE with a single outside option for each

player
Knowing the value of the subgame starting at TT , we can derive the solu-
tion of the entire game. Let us start with a useful definition:
Definition 2 Take a Rubinstein game without the possibility of opting out

and with exogenously imposed continuation values, AI and AIl (AI+ AIl :s
i j13 , for the players in period T. The backward inducted strategic value
(BISVj of the period T proposer at t < T is that player's continuation value
in period t.
The BISV is calculated via simple backward induction. Suppose it is

Player i who proposes in period T. Then, in equilibrium, in period T-1
Player j has to offer at least 8Ai to i, while she is willing to offer no more
than a share of 1 - 8A j to him. Note that, by assumption, 8Ai :s 1 - 8A j ,
so she will offer 1 - 8A j to i, giving her a payoff of 1 - 8Ai (since i will
accept). In consequence, in period T-2 i has to offer 8(1- 8A i ) to j, and so
on. Overall, Player 1's continuation value in period t < T is given by the
following table - Player 2's payoff is one minus this value. (Row gives the
proposer in period t, Column the proposer in period T):
13Note that if at some point in time the discounted sum of the players' strategic values
exceeds 1 then at least one of them will always prefer waiting until period T r to agreeing
at any value.
398
1 2
1 A
I
+-m-
8T - t l_cS T - t 1- A II 8 T - t -'1+6
cS_cS T - t
A 8T-t+cS-cST-t 1- A II 8 T - t -'1+6
l_cS T - t
2 I -m-
Table 1
Let us now interpret A as the strategic value of an option in period T.
Observe that the option is effective (that is, it modifies the partition that
would result in its absence) if and only if A is larger than its owner's Ru-
binstein share in period T, independent of the size of T (c.f. the remark
following Lemma 1). This observation should be interpreted as a generaliza-
tion (and strengthening) of the Outside Option Principle. 14 The important
fact here is that even though the option is to be discounted in case it is
taken, it is its undiscounted value what matters when it is used as a threat.
In a parallel way, the value of the outside option at the beginning of the
bargaining game is significantly more than simply its discounted strategic
value. Whenever the option is effective, its (strategic) value at time zero
is the sum of two terms: its discounted strategic value and an increasing
(!) proportion (1 - 8T ) of the Rubinstein share. 15 By the proof of Lemma
2 it is quite clear why is it A what matters instead of the straight ex-
pected value of the option. The other term is a direct consequence of the
Rubinstein formulation.
How does the strategic advantage derived from having the option trade
off against the "first mover advantage"? The direct comparison of the
four different situations with different bargaining power (1 or 2 makes the
first/last offer) is not feasible because the different delays that necessarily
arise between the first offer and the revelation date. Instead, we propose
the comparison of two hypothetical situations, where both the strategic
value and the revelation date, T-t, are assumed to stay constant while we
change the identity of the first proposer. In this case, the difference between
being a first or a second mover is ~:;:~, just as in the game without outside
options.
Let us now return to the characterization of the equilibria of our game
with stochastic outside options. The equilibrium behavior from period T r
on we have already settled. To fully characterize the equilibrium we use
backwards induction, simply observing that the equilibrium moves in pre-
vious periods are more restricted depending on which is the unique16 sub-
14Rubio (1994) has shown a similar result, for a time-varying outside option with ex
ante known value.
15Note, however, that the equilibrium share still has a decreasing trend in T.
16We assume that when players are indifferent between two actions they choose the
one that ends the game earlier.
399
game equilibrium in some period. As the revelation date increases there

are four possible phases in the equilibrium play. These must follow each
other in a monotone order. However, depending on the parameters, some
of the phases may not come about. The phases as the revelation date moves
into the future are i) "dealing out" the opponent at the discounted expected
value of his option; ii) taking the options before their value is known; iii)
"dealing out" at the discounted expected value of the other's continua-
tion (which includes an "option value"); iv) waiting for next period. We
summarize this exercise in the following straightforward proposition.
Proposition 3 The game with an outside option for each player has a
unique SPE. It can be computed by backwards induction as follows:
i) Compute the unique equilibrium behavior from period TT on (see

Lemma 1). Using these strategic values, the equilibrium action pro-
files at TT - 1 can be derived easily. Depending on the parameters,
these must be one of the following:
a) waiting for next period's payoff;

b) agreeing at a value holding the responder to the discounted ex-
pected value of her continuation payoff next period;
c) taking their respective outside options (without knowing their
realization) ;
d) agreeing at a value holding the responder to the (possibly) dis-
counted expected value of her option.
ii) For all t ::; TT - 1, if the equilibrium action profiles at tare
a) then the equilibrium action profiles at t -1 are (a), (b), (c) or

(d);
b) then the equilibrium action profiles at t -1 are (b), (c) or (d);
c) then the equilibrium action profiles at t -1 are either (c) or (d);
d) then the equilibrium action profiles at t -1 must be (d).
Let us now interpret what the proposition tells us (this will also provide
a sketch of the proof). First of all, it formalizes the intuition that if an
outside option has high option value (that is, its prior distribution has large
variance) then the players may find it in their interest to delay agreement
until their information about their outside opportunities improves. If the
revelation date is "too far" in the future then the players will prefer to
end the game immediately. If neither player prefers her option to it they
will agree at the corresponding BISV (b). Note that if in some period
t < TTboth players prefer agreeing at the BISV of one of them to waiting,
400
then they do so in all previous periods as well. 1 7 If at least one does then
they either both take the options (c) or they agree at the value which gives
the responder exactly as much as he would have obtained (in expectation)
opting out (d). Now, if the options become available sufficiently before
their revelation date then it may be the case that in periods previous to
ones where they would agree, they take their options (since the options do
not get discounted). Finally, if in a period before Ta they both take their
options then in (sufficiently) earlier periods agreement will be the unique
equilibrium, since the discounted sum of the expected values of the options
decreases below one.
Having described the nature of the bargaining process when the nego-
tiators have a single outside option, we now turn our attention to the case
where the same player may choose among several outside opportunities.
4 Multiple options
In this section we analyze the case when Player 1 (the first proposer) has
several options available to her if she leaves the bargaining game. Specif-
ically, assume that there are N options, and option i matures at Tk and
its value is distributed according to Fk. Without loss of generality, we as-
sume that Tl ::; T2 ::; ... ::; T N. In order to be able to concentrate on the
issues arising from the multiplicity of options, in this section we assume
that Player 2 has no outside option and that Tk = Tk = Tt:" for all k.
Let us start by a generalization of the concept of backward inducted
strategic value for the multi-option setting.
Definition 3 The backward inducted strategic value of the set of options

(j,j+l, ... ,Nj at time Tk, denoted by Bk(j), is the equilibrium continuation
value of Player 1 in that period, only taking into account the existence of
options (j,j+l, ... ,Nj.
The way to calculate the BISVs is presented in the following lemma:
Lemma 4 The backward inducted strategic values of the options are inter-
related according to the following system of equations;18
if Tk is even
if Tk is odd
17 Simply observe that all four elements in Table I satisfy the relationship that de-
creasing t by one gives a value that is strictly greater than multiplying that element by
o.
18Note that BN(N) is the strategic value of the option maturing last.
401
Bk + 1) ot(k) + 1 6,*(k)
(k if T.k is odd
Bdk+1)= { +1 1+6' :J
B (k + 1) ot(k) + 6_6'·(k) if T.k is even
k+1 1+6 ' :J
where t(k) = n+1 - Tk and t * (k) = t(k) if t(k) is odd and t(k) +1
otherwise.
Proof. The first equation follows from the same argument as the proof
of Lemma 1 (2). The second equation is a straightforward application of
Table 1..
Now we are ready to state the main result of this section:
Proposition 5 The unique outcome of the multi-option game, which is
supportable by a subgame-perfect equilibrium, is immediate agreement, with
Player 1 obtaining B l (1)OT1 -1 + 11 !;·
(where T* = T1 ifT1 is even and
T 1-1 otherwise), unless Tl = 1, in which case she obtains the whole surplus
if B1 (2) :S al (the realized value of the first option) and Bl (2) otherwise.
Proof. Note that B l (1) can be considered as the strategic value of a single
outside option whose value realizes at time T l . The rest follows from Table
1. The results for T1 = 1 follow from the same argument as the proof of
Lemma 1 (2) . •
An immediate observation to make is that the upper end of the support
of an option's distribution function is a sufficient statistic for the entire
distribution. 19 The following corollary is then straightforward:
Corollary 6 If and only if the upper end of the support of all the options
maturing in odd (even) periods is no greater than 1~6 (1!6)
the outside
options have no influence on the bargaining outcome.
As a consequence, the availability of several options over time gives rise to
a Coasian result: whenever one of the later options is effective, in periods
where she is the proposer Player 1's bargaining power is reduced with
respect to the case where she only has option 1, just as a monopolist seller
of a durable good drives down the price competing against herself. This
reduction takes the form that some outside options that on their own would
give a strategic value of one, will cease to be effective because of the later
option's presence.
To illustrate how the potential effectiveness of earlier options is affected
by a later option being effective, we present the following example:
Example 1 Assume that the last option that is locally effective (that is,
it has a strategic value that is different from the Rubinstein continuation),
19By the first equation in Lemma 1 it follows that this is so for "local" effectiveness.
But "global" effectiveness is just the local one checked sequentially.
402
matures in period 6 and it yields a strategic value of A(6) = !.

We are
now going to calculate what the maximal upper ends of the supports of the
distributions of the (potential) options in all the previous periods are so
that the last option remains the one determining the terms of agreement.
Since period 5 is odd, by Lemma 1 that period's option is effective iff

it is greater than 8A(6)=~. If the option is not effective then Player l's
subgame-perfect share in that period is 1- 8(1-!) = 2;8. Then in period
4 the option has to give Player 1 8 2;8 in order to be credible (as well as
effective, since we are in an even period). This will also be her subgame-
perfect share if the option is not effective. In period 3 then the effectiveness
limit is 822 ;8, while Player l's subgame-perfect share isl-8 (1 - 8 2 ;8) =
1 - 8 + 82 2;8. In period 2 then the effectiveness limit, as well as Player
l's share, is 8 - 82 + 83 2;8. Finally, in the first period the effectiveness
limit is 82 - 83 + 842 ;8, while Player l's share is 1-8 (8 - 82 + 832;8) =
1 - 8 + 82 - 83 + 84 2;8. In Table 2 we can appreciate that while the effect
of the option is diminishing the further it is, it never vanishes: in every
preceding period it increases both the effectiveness limit of the current
option and Player l's share, with respect to the case with no effective
options.
Period 1 2 3 4 5
Rub. share 128 64 128 64 128
Eq. share 129 66 132 72 144
Indiv. eff. lim 32 64 32 64 32
Eq. eff. lim. 33 66 36 72 48
Table 2. (8 = ~ and the unit of measure is 1~2)
5 Final remarks
In many bargaining situations outside options are uncertain and they arise
and cease over time. Proposals that may seem attractive (unattractive) at
the start of the game may cease to be so as the uncertainty disappears
or as time approaches the maturity date of the outside opportunities. We
have analyzed the effects of this kind of uncertainty in the outcome of a
negotiation.
The crucial influence of the outside opportunities available to the play-
ers on the outcome of a bargaining game has long been recognized (Nash
(1950)). However, once the confusion among the definitions of "status quo
point," "disagreement point" and "threat point" was first pointed out and
then satisfactorily settled (Sutton (1986), Binmore, Shaked and Sutton
403
(1989)), outside options have seemingly exited from the research agenda.
The implicit argument behind this neglect was that, since we know exactly
their effects, why should we unnecessarily complicate our models with those
extra details?
We have argued here and in related work (see Ponsatf and Sakovics
(1998a,1998b)) that there are still interesting questions that can be ad-
dressed and new insight to be gained in the context of bargaining games
a la Rubinstein-Stahl with outside options. In this paper we have claimed
that considering uncertainty about the outside options is a natural exten-
sion to the literature that deserves attention. Moreover, since allowing both
players to take an outside option has important effects (compared to models
with only one-sided outside options), the combination of these two elements
yields novel insights on the role of outside opportunities in bargaining.
There are a number of ways in which our models could be extended.
The most obvious generalization would be to add asymmetric information,
players could learn privately the value of their options, about the proba-
bilities, about the revelation date etc. Another interesting scenario is when
the revelation of information is not instantaneous but the players learn over
time about the prospects of their outside opportunities. In these cases, we
would also have the additional consideration of how much a player's actions
(offers) reveal about what she knows at the time.
References
Avery, C., and Zemsky, P. (1994a) "Option Values and Bargaining Delays,"
Games and Economic Behavior 7, 139-153.
Avery, C., and Zemsky, P. (1994b) "Money Burning and Multiple Equilibria
in Bargaining," Games and Economic Behavior 7, 154-168.
Binmore, K., Shaked, A. and Sutton, J., (1989) "An Outside Option Ex-
periment," Quarterly Journal of Economics 104, 753-770.
Dalmazzo, A. (1992) "Outside Options in a Bargaining Model with Decay
in the Size of the Cake," Economics Letters, December, 417-42l.
Hendon, E., Sloth, B. and Tranaes, T., (1994) "Decentralized Trade with
Bargaining and Voluntary Matching," Economic Design 1, 55-78.
Hendon, E. and Tranaes, T., (1991) "Sequential Bargaining in a Market
with One Seller and Two Different Buyers," Games and Economic Be-
havior 3, 453-466.
Merlo, A. and Wilson, C., (1995) "A Stochastic Model of Sequential Bar-
gaining with Complete Information," Econometrica 63, 371-399.
Moldovanu, B. (1993) "Price Indeterminacy and Bargaining in a Market
with Indivisibilities," Journal of Mathematical Economics 22, 581-597.
Nash, J., (1950) "The Bargaining Problem," Econometrica 18, 155-162.
Osborne, M., and Rubinstein, A. (1990) Bargaining and Markets, Academic
Press, San Diego.
Ponsatf, C., and Sakovics, J. (1998a) "Rubinstein Bargaining with Two-
404
Sided Outside Options," Economic Theory, forthcoming.

Ponsatf, C., and Sakovics, J. (1998b) "Two-Sided Options and Stationary
Uncertainty in Bargaining," mimeo, Universitat Autonoma de Barcelona,
March.
Rubio, J., (1994) "Time-dependent outside option in an alternating offers
bargaining model," mimeo, Euskal Herriko Unibertsitatea, October.
Shaked, A., (1994) "Opting Out: Bazaars versus 'Hi Tech' Markets," In-
vestigaciones Econ6micas 18(3), 421-432. (previously circulated as LSE
discussion paper 1987/159)
Shaked, A. and Sutton, J., (1984) "Involuntary Unemployment as a Perfect
Equilibrium in a Bargaining Model," Econometrica 52, 1351-1364.
Sutton, J., (1986) "Non-Cooperative Bargaining Theory: An Introduction,"
Review of Economic Studies 53, 709-724.
Vislie, J., (1988) "Equilibrium in A Market with Sequential Bargaining and
Random Outside Options," Economics Letters 27, 325-328.
An extension of the nonatomic assignment
model*
Doraiswamy Ramachandran I and L. Ruschendorf'2
1 Department Of Mathematics and Statistics, California State University, 6000
J Street, Sacramento, CA 95819-6051, U.S.A

2 Institut Fur Mathematische Stochastik, Albert-Ludwigs-Universit!Lt,
Eckerstrasse 1, D-79104 Freiburg, Germany.
Abstract. We investigate the nonatomic assignment model introduced by

Gretsky, Ostroy and Zame (1992) under general conditions on the economic
agents and utility functions. This extension allows considering various func-
tion spaces (price processes and risk curves) as models for the market.
We show that the assignment problem based on "inequality" constraints
is equivalent to a corresponding transportation problem under "equality"
constraints. This equivalence leads to a general duality theorem in this set-
ting. In order to model some external regulations on the market requiring
at least a certain minimum level of activity on the part of the agents, we
then introduce a modified assignment model with constraints on the agents.
We establish duality theorems for this modified assignment problem and
existence results for optimal solutions.
JEL Classification Numbers: cn, C78

Keywords: Assignment problem, duality theorem
1 Introduction
The assignment model of Shapley and Shubik (1972) has been extended to
the following version with a continuum of buyers and sellers by Gretsky,
Ostroy and Zame (1992) called the nonatomic assignment model:
Consider the set X I of buyers and the set X 2 of sellers each having a dis-
tinct house to sell, and associate two probability spaces (Xi, Ai, Pi), i = 1,2
with the buyers and sellers respectively, where Pi represent the popula-
tion distributions. If buyer Xl and seller X2 were to transfer ownership
of the house X2, then the monetary value of this transfer between the
pair (Xl, X2) can be represented by a measurable function h(Xb X2) on
"The work is supported by the V.S. - Germany Cooperative Science Program, NSF
grant. INT-9513375 and DAAD grant 315/PRD/fo-ab.
406
(Xl X X 2 ,A I ®A2 ). Alternatively, h(XI,X2) is the profit available to the

pair (Xl, X2). The goal of the assignment problem then is to match buyers
and sellers so as to maximize the total available profit. Define an assign-
ment of buyers to sellers as a measure J.L on (Xl X X2, Al ® A2) and inter-
pret it as a statistical summary of the activities in the housing market; i.e.,
J.L( E X F) is the distribution of buyers in E purchasing from sellers of houses
in F. Call an assignment J.L feasible for (Pt, P2) if J.L( E x X 2) ~ PI (E) and
J.L(XI x F) ~ P2(F) for all E E Al and F E A 2 • The assignment problem
then is to maximize the expected profit; that is, to find a feasible assignment
J.L to attain the optimal value
This problem leads to the dual problem of finding Ai-measurable func-

tions It, i = 1, 2 to attain the dual functional
Assuming Xl and X2 to be the closed unit interval [0,1] and the objective
function h to be upper semicontinuous, Gretsky et al. (1992) show that
(i) there exist optimal solutions to the resulting assignment

problem and its dual, and
(ii) there is no gap between the primal and dual values, i.e.,
S(h) = J(h).
They also present a game theoretic version via the core and a mar-
ket economy version via Walrasian equilibrium and establish the effective
equivalence of the solution concepts of the three versions.
In this paper, we treat the nonatomic assignment model under very gen-
eral conditions on the economic agents and cost functions; we also discuss
some variants of the problem. We first show that the assignment problem
based on "inequality" constraints as considered by Gretsky et al. (1992)
is equivalent to a corresponding transportation problem under "equality"
constraints. Based on a general duality theorem for transportation prob-
lems, this leads to a broad solution which encompasses the case of compact
metric spaces with upper semicontinuous cost functions considered by Gret-
sky et al. (1992). This level of generality allows considering various function
spaces as models for the market; for example, consider the situation where
risks of stocks or goods represented by the development of the price over
a time period (rather than the price at a fixed time point) are sold to a
group of buyers such as insurance companies by the associated owners.
Next, we introduce the nonatomic assignment model with additional con-
straints on the economic agents in the form of finite measures which bound
407
the population distributions of the buyers and sellers; this arises in markets
where certain minimal level activities are imposed by regulatory conditions.
By a modification of the finitely additive approach to duality theorems as
developed in Riischendorf (1981), we establish a general duality theorem
for this case along with some useful consequences.
2 The nonatomic assignment model

Let (Xi, A, Pi), i = 1,2, be two probability spaces. A probability /1 on
(Xl x X 2, Al ® A2) is said to have marginals PI and P2 if
and
/1(Xl x A2) =: /1 0 7l"2(A 2) = P2(A 2) for all A2 E A2.
Let M = M(Pl , P2) = {/1 on Al ® A2 : /1 0 7l"i = Pi, i = 1, 2}. For a
given measurable function h on (Xl x X2, Al ® A2)' the marginal problem
is concerned with the solution of
(1)
Gretsky et al. (1992) introduced the following nonatomic assignment

model: Consider two finite measure spaces (not necessarily probabilities
to begin with) (Xi, A, Ti), i = 1,2 ,where without loss of generality
Ihll ~ IIT211· Let
< <
M- = M-(Tl,T2)={/1 on Al®A2:/107l"i~Ti,i=1,2}.
For an Al C59A2-measurable function h on Xl x X 2 , the assignment prob-

lem seeks to find a solution to
Our first aim in this section is to show that this nonatomic assignment
problem of Gretsky et al. (1992) can be reduced to a corresponding marginal
problem (1) described in the preceding paragraph. To this end, define Xl =
Xl U {8} and let
_h- {h on Xl x X2
a on {8} x X 2 .
By taking T1 = Tl +(IIT211-lh 11)106 we have IIT111 = IIT211 and without loss

of generality we can take IIT111 = IIT211 = 1 so that they are probabilities.
- <
Let Al = a({A1,{8}}). For 7I E M-(T1,T2), it is easy to check that
408
f.L = 'jIIX t XX 2 E M:S(71,72) and JX 1 XX2 h d:ji = JXI XX2 h df.L. Conversely,
if f.L E M~ (Tb T2), then extend f.L /' 'jI by 'jI(C) = f.L(Cn(X1 x X 2)); again,
it can be checked that 'jI E M:S (71, 72) with Jh d:ji = J h df.L. Thus
sup{j h d:ji: 'jI E M:S (71, 72)}
sup{j h df.L: f.L E M:S (71, 72)}
S:l
<
XA 2 (h) (3)
and so in the Gretsky et al. (1992) model we can take 71,72 to be proba-
bilities (by replacing them with 71,72, h if needed).
<
Next, if f.L E M- (71, 72) then letting H = {h 2: O} and f.LH = f.LIH ,we
get
j h df.L:::; j h df.LH = j h+ df.LH
where h+ = max(h,O) and f.LH 01fi :::; 7i,i = 1,2; so f.LH E M:S(71,72)
as well. Hence in (2) (respectively (3)), one can restrict attention just
< <
to those f.L E M- (71, 72) (respectively'jI E M- (71, 72)) with the prop-
erty f.L({h < O}) = 0 (respectively'jI({h < O}) = 0). For each f.L E
M:S (71, 72), J h+ df.L = J h+ df.LH = J h df.LH :::; S:S(h) and J h df.L :::;
J h+ df.LH :::; s:S (h+) whereby
(4)
Hence in (2) we can further assume that h 2: O.
To reduce the problem finally to (1), we denote the probabilities 7i by
Pi, i = 1,2. Given E > 0, let f.L€ E M:S(P1,P2) be such that Jh+df.L€ >
s:S (h+) - E. Since f.L€ E M:S (Pb P2), we have Qi = f.L€ 01fi :::; Pi. We
now choose a 1/ E M(P1 - Ql, P2 - Q2) and take f.Lo = f.L€ + 1/. Then
f.Lo E M(P1, P2) and
j h+ df.Lo j h+ df.L€ + j h+ dl/
> j h+ df.L€
> s:S (h+) - E.
Hence
S(h+) 2: s:S (h+).
The other inequality being trivial, we get
(5)
409
From (4) and (5) it follows that
(6)
We have thus shown that in defining feasible assignments in the

nonatomic assignment model of Gretsky, Ostroy and Zame (1992) it is
sufficient to restrict attention to those assignments whose marginals are
the prescribed probabilities.
In order to derive the solution of the assignment problem in this general
set-up, for an Al ® A2-measurable function h on Xl x X2, consider the
dual functional
J(h) = JAl~M2(h) = inf{t;

2
Jhi dPi : hi E £l(Pi ) and h::; hI EEl h2}.
Let (AI ® A2)m = {h : h is (Xi, A, Pi), i = 1,2, measurable and

h ~ It EEl h for suitable Ii E £1 (Pi)} which contains all bounded, Al ® A2-
measurable functions. A probability measure P on (n, A) is called per-
fect (equivalently, the probability space (n, A, P) is called perfect) if for
every A-measurable, real-valued function f on n we can find a Borel set
BI c fen) such that p(j-l BI) = 1; all probabilities on the Borel subsets
of a complete, separable metric space are perfect and perfect probabilities
have many desirable features for applications in probability theory (see Ra-
machandran (1979)). Based on the considerations in this section and the
duality theorem established in Ramachandran and Rllschendorf (1995) for
the transportation problem with fixed marginals, we have the following gen-
eral solution for the nonatomic assignment model without any topological
assumptions on the underlying spaces.
Theorem 1 Let (Xi, Ai, Pi), i = 1,2, be arbitrary probability spaces where
at least one of PI, P2 is perfect. Then [§': (h) = S(h+) = I(h+)] for h E
(Al®A2)m.
Comments (i) For h E (AI ® A2)m with J(h) < 00 there exist solutions
to the dual problem
:3 hi E £l(Pi ) with h::; EEli hi and J(h) = ~

~
Jhi dPi .
(ii) When Xi are topological spaces with Pi as tight measures then for
upper semicontinuous functions h there exists a maximal measure Jlh E
M(Pl , P2) such that
410
3 Assignments with constraints on the agents

In this section we introduce a modification of the assignment model where
the marginals of admissible assignments are bounded above by population
measures Ti. Suppose that by some external regulations on the market at
least a certain minimal level of activity is required. This assumption leads
to considering also lower bounds on the agents given by finite measures
Ai ~ Ti· Without loss of generality we may assume that IIT111 = IIT211 = 1
and IIA111 = IIA211. It turns out that this model is technically more involved
than the pure assignment model and needs some new tools to be developed.
Let (Xi, Ai), i = 1,2 be two measurable spaces and let Ai ~ Ti be finite
measures on (Xi,Ai) with IIA111 = IIA211 ~ Ihll = IIT211 = 1 and define
MA,'T = {fL on Al ® A2 : (7)

fL is a measure with Ai ~ fLi = fL 0 7ri ~ Ti, i = 1, 2}, (8)
which is the class of feasible assignments with restrictions on the activity

of the agents. The optimal assignment problem in this set-up is to solve
A suitable dual functional is given by

2
h,'T(h) = inf(~~~)J fi+dTi - J fi-dAi) : fi E .c1 (Ai), EBi fi 2 h} .
• =1
We obtain the following duality theorem.

Theorem 2
A1) Let at least one of the measures Tb T2 be perfect and let U denote
the class of all functions h which can be approximated uniformly
by functions of the form L~=l Qi1AixBi,Ai E A 1 ,Bi E A 2. Then
[SA,'T(h) = h,'T(h) for all h E U.]
A2) If (Xi, A) are Hausdorff spaces and Ti, i = 1,2 are tight (Radon)
measures then for all bounded upper semicontinuous functions
h [SA,'T(h) = h,'T(h).]
A3) If h is bounded then there exist fi E .c1 (Pi) such that [h,'T(h) =
J
L~=l (J f i+dTi - f i- dAi).]
Outline of the Proof The proof follows the approach in Ruschendorf
(1981,1991). In the first step, the set of measures in MA,'T are replaced
by finitely additive measures. Application of the Hahn-Banach theorem
and the Riesz representation theorem establish duality in the finitely addi-
tive setting. Based on the integration theory for finitely additive measures
411
one obtains from this the duality theorem in (AI) for the class U. Since
perfectness of one of the marginals implies a-additivity of such finitely
additive measures on the algebra of measurable rectangles in the product
space (see Ramachandran (1979), chapter 3), in the topological case the
tightness property of Ti allows us to obtain duality for bounded continuous
functions. Then one can establish compactness of the class of dual admis-
sible functions as in Ruschendorf (1981). This implies (A3) and, based on
this compactness, we obtain that the dual functional is a-continuous up-
wards (see Kellerer (1984), Prop. (1.28)). This implies that duality can be
extended to the upper semicontinuous functions.
Remarks Similar to the above model with M)..,'T specifying upper and
lower bounds on the population measures, we can also consider markets
with external regulations on the matching of buyers and sellers. Consider
(Xi, Ai, Ti), i = 1,2, representing the buyers and sellers where the popula-
tion measures Ti are assumed to be normalized. Let Cc Xl X X 2 denote
the set of admissible interactions among the economic agents imposed on
the market by regulations or technical conditions and let
be the set of all admissible assignments concentrated on C. The following

assignment problem for regulated market arises: Find an optimal solution
of
<
S~(h) J
= sup{ h dp,: p, E M~< }.
In the case when Xi are Hausdorff spaces and the measures Ti are tight,
the following solution is obtained for any profit h(X1,X2) which is either
lower or upper semicontinuous or uniformly approximable by functions of
the form Li Qi 1Ai X Bi :
Letting h+ = max(h, 0) and taking the dual functional I defined by
1(g) = inf{ ~
•
J gi dTi : gi E .c1 (Ti)' gl EB g2 2: g},
we have the duality

S: (h) = 1(h+ le).
Details can be found in Ramachandran and Ruschendorf (1997).
References
Gretsky, N.E., Ostroy, J.M., and Zame, W.R.: The nonatomic assignment
model. Economic Theory, 2, 103-127 (1992).
Kellerer, H.G.: Duality theorems for marginal problems. Z. Wahrschein-
lichkeitstheor. Verw. Geb. 67, 399-432 (1984).
412
Ramachandran, D.: Perfect measures I and H. I SI-Macmillan Lecture

Notes Series, 5,7, Macmillan, New Delhi, (1979).
Ramachandran, D. and Rtischendorf, L.: A general duality theorem for
marginal problems. Probability Theory and Related Fields, 101, 311-319
(1995).
Ramachandran, D. and Rtischendorf, L.: Duality theorems for assignments
with upper bounds. Proceedings of the third international conference on
distributions with fixed marginals and moment problems held at Prague.
Amsterdam: Kluwer (1997).
Rtischendorf, L.: Sharpness of frechet bounds. Z. Wahrscheinlichkeitstheor.
Verw. Geb. 57, 293-302 (1981).
Rtischendorf, L.:Frechet bounds and their applications. In Dall' Aglio, G
et at. (eds.) Advances in probability distributions with given marginals,
151-188. Amsterdam: Kluwer (1991).
Testing decision rules for multiattribute
decision making*
Christian Seidl1 and Stefan Traub 1
1 Institut fur Finanzwissenschaft unci Sozialpolitik,
Christian-Albrechts-Universit1it zu Kiel Olshausenstr. 40 D-24098 Kiel,

Germany
Abstract. This paper investigates the existence of an editing phase and

studies the compliance of subjects' behaviour with the most popular mul-
tiattribute decision rules. We observed that our data comply well with the
existence of an editing phase, at least if we allow for a natural error rate of
some 25%. We also found a satisfactory performance of certain groups of
subjects for the conjunctive rule, for the elimination-by-aspects rule, for
the majority rule, and for the maximin rule. Our data suggest, however,
rejection of the prominence hypothesis and of the maximax rule. Thus, our
experiment sheds light on the existence of an editing phase and on the use
of various multiattribute decision rules.
JEL Classification Numbers: C91, D46, D80, A14

Keywords: Sequential decision making, editing phase, prominence hy-
pothesis, elimination-by-aspects hypothesis, conjunctive decision rule, ma-
jority rule, multiattribute decision making
1 Introduction
When we designed an experiment of sequential decision making to em-
pirically investigate whether subjects satisfy the axiom of independence
of irrelevant alternatives, we became increasingly suspicious of the useful-
ness of distinguishing between static and dynamic decision models. For a
long time, psychologists have stressed that a si,ngle decision is the result
of perceptual, emotional and cognitive processes, which all contribute to a
dynamic process in which the decision maker seeks and evaluates informa-
"This paper was init.iat.ed when t.he first. aut.hor spent. three weeks at t.he Center of
Economic Research at Tilburg University. He wants t.o express his indebtedness to the
hospit.alit.y of the Cent ER. Financial support of the Landeszent.ralbank fUr Hamburg,
Mecklcnhurg--Vorpommern und Schleswig-Holstein and the European Community under
Contract CHRX-CT94-0647 is gratefully acknowledged. For useful comments we are
indebted to Eric van Damme and to Professors Albers and Brockhoff.
414
tion sequentiallyl. This insight led us soon to the question which decision
rules are observed by subjects, or, to put it more precisely, which decision
rules are not at variance with subjects' decisions. Fortunately, our exper-
iment equipped us generously with data, which allowed us to embark on
this research. In particular, we investigate the existence of an editing phase
in which dominated alternatives are eliminated. Furthermore, we test sev-
eral multiattribute decision rules for their explanatory power of consistency
with subjects' choices.
Section 2 develops the theoretical underpinning of our paper. It starts by
explaining the dynamics of decision making through decision makers' effort
to minimize cognitive effort. Then we proceed to review the most impor-
tant multiattribute decision rules. Section 2.3 examines whether decision
processes are multi-phased.
Section 3 gives a detailed description of our experiment, and Section
4 contains the results of our study. Section 4 parallels Section 2 in its
structure to facilitate the comparison of more theoretical considerations
with the empirical evidence of our experiment. Section 5 concludes.
2 How to simplify decision problems

2.1 Minimizing cognitive effort
A decision maker's task of seeking, gathering and evaluating information
involves a considerable cognitive effort or cost of thinkingf. As different
decision rules may require different amounts of cognitive effort, decision
makers who try to minimize the amount of cognitive effort will, during the
decision process, tend to apply simpler rules before they try rules which
require more cognitive effort3 .
Now, which decision rules can be assumed to be simpler rules in mul-
tiattribute decision making? Multiattribute decision rules can be divided
into noncompensatory or noncommensurable and compensatory or com-
mensurable decision rules. The distinction between these types of decision
rules is straightforward. Under a noncompensatory rule, the abundance in
some attribute cannot compensate for the deficiency in another. Under a
compensatory rule, trade offs among all attributes must be defined. A com-
1 Cf., e.g., Montgomery and Svenson (1976), 283; Svenson (1979), 86.
2For an interesting theory of the cost of thinking cf. Shugan (1980). The cost of
optimization was analysed by Conlisk (1988). Von Winterfeldt and Edwards (1982) and
Harrison (1994) have blamed experimental economists for insufficient rewards used in
their experiments, which were not attractive enough to compensate subjects for their
decision cost. They argue that insufficient rewards could have been the cause for much
observed falsification of correct theories. For a fully-fledged model of decision costs and
subjects' performance in experiments cf. Smith and Walker (1993).
3Montgomery and Svenson (1976), 288f.; Svenson (1979), 107; Shugan (1980), 100;
Russo and Dosher (1983).
415
pensatory rule must state exactly which improvement in attribute k of a

choice alternative just compensates for a given deterioration in attribute e
to keep the respective choice alternative equally attractive. Therefore, the
attractiveness of a choice alternative under a compensatory decision rule is
usually addressed as its utility4.
Empirical work supports the "general conclusion that subjects find it
rather difficult to weight and 'trade off' values in a compensatory man-
ner .... information that does not require 'in the head' transformation is
preferred in the interest of cognitive economy.5" This empirical evidence
has induced Slovic and MacPhillamy to suggest a common-dimension ef-
fect stating that attributes are weighted more heavily in the comparison of
choice alternatives when they are common. On grounds of minimizing cog-
nitive effort, subjects eschew making transformations of information prior
to its use 6 •
When trying to minimize cognitive effort, the decision makers may, of
course, use a plurality of decision rules 7 rather than following the pre-
script of utility maximization as has been assumed by mainstream micro-
economics. Empirical evidence has shown that some of the decision rules ap-
plied are used subconsciously and cannot easily be communicated8 , which,
however, does not invalidate the use of a plurality of decision rules.
Empirical research has shown that decision makers apply decision rules
sequentially in order of increasing cognitive effort. This means that they
use first noncompensatory rules to whittle down the number of choice al-
ternatives by eliminating alternatives, thereby simplifying the decision task
considerably. Compensatory decision rules are subsequently used to analyse
the simplified decision problem. Indeed, noncompensatory rules are applied
to decision problems with many alternatives whereas compensatory rules
dominate in decision problems with only few choice alternatives 9 • Let us il-
lustrate this with a quotation of Payne. He argues that "the less cognitively
demanding decision procedures, conjunctive and elimination-by-aspects,
might be called early in the decision process as a way of simplifying the
decision task by quickly eliminating alternatives until only a few alterna-
tives remained as choice possibilities. The subject might then employ one
of the more cognitively demanding choice procedures ... 10" Many other
authors have followed him in perceiving individual choice procedures as
4Montgomery and Svenson (1976), 285.

5 Slovic and MacPhillamy (1974), 193.
6Cf., e.g., Slovic and MacPhillamy (1974); Slovic and Lichtenstein (1968); Payne and
Braunstein (1971). Notice that Slovic and MacPhillamy's common-dimension effect is
similar to Slovic and Lichtenstein's compatibility effect.
7Keen (1977), 33.
sCf. Bem (1972); Nisbett and Wilson (1977); Fischhoff, Slovic, and Lichtenstein
(1980). Contrary to these views cf. Smith and Miller (1978) and White (1980).
9Cf. Payne (1976), 382; Russo and Dosher (1983); Johnson and Meyer (1984).
lOPayne (1976), 384f.; cf. also Payne (1982), 398f.
416
multi-staged processes l l .
These considerations describe nothing more than a general pattern of se-
quential decision making. Subjects may, however, follow different decision
rules. Some subjects may be more attracted by particular decision rules,
others may be more attracted by others. In this study, we investigate first
whether decision processes are multi-phased where a rough screening ap-
plies at the outset of a decision problem, and second which decision rules
are capable of explaining the actual decisions of palpable groups of subjects.
In the rest of this section, we will first consider a menu of rules for mul-
tiattribute decision making. Then we shall investigate whether individual
decision processes indeed consist of several phases which can be clearly
discerned.
2.2 A menu of decision rules

In this section we shall concisely characterize the most important multiat-
tribute decision rules 12 . As to the notation, let A := {ai I i = 1,2, ... ,n}
denote the set of decision alternatives ai, let D := {d k I k = 1,2, ... ,K}
denote the set of attributes, and let aik denote the level of attribute k in
the decision alternative i. A decision alternative is thus defined by a vector
of attribute levels, i.e. ai = (ail, ai2, ... , aiK). Moreover, let ai > aj denote
that all components of ai are greater than the respective components of aj ,
ai ~ aj that no component of aj is greater than the respective component
of ai, and let ai 2': aj denote that (ai ~ aj)&-,(ai = aj). The weak prefer-
ence relation between choice alternatives is denoted by t , >- and ~ being
its asymmetric and symmetric components. For A' ~ A, E(A') denotes the
set of alternatives which have not been eliminated by some decision rule.
2.2.1 Noncompensatory decision rules

The most important noncompensatory decision rules are the dominance
rule, the conjunctive rule, the disjunctive rule, and the lexicographic rule
with its variations.
(1) Dominance rule:
llCf., e.g., Simon (1955); MacCrimmon (1968b); Mintzberg, Raisinghani and TMoret
(1976); Montgomery and Svenson (1976); Nisbett and Wilson (1977); Wright and Bar-
bour (1977); Kahneman and Tversky (1979), 274ff.; Svenson (1979); Tversky and Sat-
tah (1979), 542f.; Shugan (1980); Montgomery (1983), 350ff.; Russo and Dosher (1983);
Dawes (1988), chapter 4; Tversky, Sattah and Slovic (1988), 372.
12Cf., e.g., Coombs and Kao (1955); Coombs (1964); Dawes (1964); Montgomery and
Svenson (1976), 285ff.; Fishburn (1978); Svenson (1979), 89ff.; Shugan (1980), 100.
417
The dominance rule declares ai as superior to aj, if aj has no better

attribute level than ai, but ai outperforms aj with respect to at least one
attribute. The dominance rule provides for pruning the set of feasible choice
alternatives by eliminating those choice alternatives which happen to be
dominated by some other choice alternatives. It seems to be a generally-
agreed-upon decision rule. Notice, however, that it can (save for rather
special causes) only shrink the set of viable choice alternatives without
being able to single out a best one.
(2) Conjunctive rule:
E(A) = {ai E A I ai ~ cl,

where c = (Cl, C2,"" CK) is a vector of minimum attribute levels to be
satisfied by eligible choice alternatives. A conjunctive rule is thus a multi-
dimensional generalization of the satisficing principle suggested by Simon
(1955). It requires any eligible choice alternative to satisfy certain minimum
standards for all attributes.
In other words, this means that a conjunctive rule evaluates choice al-
ternatives on their least relevant attributes. Dawes has provided a good
instance for a conjunctive rule: "For example, in order to stay alive an in-
dividual must have every vital organ functioning above a certain level (by
definition). The fact that a given individual has an excellent liver, heart,
and lungs will not compensate for the fact that a doctor has just removed
his single kidney; an individual may have radically inferior organs, but he
will not die unless one ceases to function above a minimum level. Life or
death depends upon one's worst vital organ." 13
Notice that a conjunctive rule can, like a dominance rule, serve merely
as a selection procedure which narrows down the set of eligible choice al-
ternatives. It can, however, also be shaped as a choice rule which directly
determines the optimum choice alternative. This is accomplished by vary-
ing the vector of minimum attribute levels until only one choice alternative
passes all requirements.
(3) Disjunctive rule:
E(A) = {ai E A I 3aik 2 Ck for some k = {1, 2, ... , K}} ,

where C = (Cl, C2, .•• ,CK) is a vector of attribute levels of which at least
one has to be satisfied. A disjunctive rule thus requires any eligible choice
13Dawes (1964), 105. Cf. also Einhorn (1970), 223; Einhorn (1971), 14f.; Montgomery
and Svenson (1976), 285; Payne (1976),367; Wright and Barbour (1977), 94f.; Svenson
(1979), 89.
418
alternative to satisfy certain minimum standards for at least one attribute.

This means that a disjunctive rule evaluates choice alternatives on their
greatest merits. If the choice alternatives happen to be candidates to fill
some very special position, then they are evaluated according to their great-
est talents regardless of their other attributes14 • A disjunctive rule can, like
a dominance rule, serve only as a screening device which whittles down the
set of eligible choice alternatives. Raising the c's sufficiently it can also be
sharpened to isolate a single best choice alternative. Then it comes close
to a lexicographic rule.
(4) Lexicographic rule:
Let d' denote a vector of attributes dk, k = 1, 2, ... ,K, arranged in
decreasing importance of the attributes, i.e., d~ is considered to be more
important than d~ if m < £.
aim ajm, ail > ajl. for all m, 1 ~ m < £,

and for some £ = 1,2, ... , K' ~ ai )-- aj;
E(A) = {ai E A I ~ aj : ajm = aim, ajl. > ail for all m,

1 ~ m < £ and for some £ = 1,2, ... ,K'}.
A lexicographic rule declares ai as superior to aj, if ai has a better at-

tribute level than aj in the most important attribute for which the attribute
levels differ. The lexicographic rule has spawned many offshoots, such as the
lexicographic semiorder rule15 , the elimination-by-aspects (EBA) rule 16 ,
the constant-ratio (eR) rule 17 , the elimination-by-tree (EBT) and the hi-
erarchical elimination (HE) rules 18 (which happen to be equivalent), and
the prominence hypothesis 19 (PH).
The elimination-by-aspects hypothesis was pioneered by Restle and
Shepard2o • Building upon their work, Tversky developed a theory describ-
ing choice as an elimination process governed by successive selection of
aspects or attributes proceeding in the order of their importance21 • He il-
lustrates this with the following example: "In contemplating the purchase
of a new car, ... the first aspect selected may be automatic transmission:
this will eliminate all cars that do not have this feature. Given the remain-
ing alternatives, another aspect, say a $3000 price limit, is selected and
14Cf. Dawes (1964), 105; Einhorn (1970), 223; Montgomery and Svenson (1976), 285;
Svenson (1979), 89.
15Tversky (1969).
16Tversky (1972a,b).
17Luce (1959; 1977).
18Tversky and Sattah (1979).
19 Tversky, Sattah, and Slovic (1988).
20Restle (1961); Shepard (1964a,b).
21Tversky (1972a,b).
419
all cars whose price exceeds this limit are excluded. The process continues
until all cars but one are excluded. 22 " Adhering to the (then dominating)
conception of choice as a stochastic phenomenon23 , Tversky conceived of
evaluation functions of the various aspects (or attributes) of choice alter-
natives and modelled the choice probabilities of the various alternatives as
increasing functions of the values of the relevant aspects. As the values of
the various aspects are, however, not known a priori, Tversky chose an ex-
perimental design presenting the same stimuli repeatedly to his subjects.24
This provided him with choice frequencies of the decision alternatives which
he could use for the estimation of the values of the aspectS. 25
Because of considerable data requirements of the elimination-by-aspects
rule, Tversky and Sattah later developed a hierarchical elimination process
which is more parsimonious with respect to data than the former one26 .
This elimination-by-tree rule or hierarchical elimination rule is applicable
whenever the decision problem is presentable in the form of a decision
tree in which the decision maker eliminates various subsets of alternatives
sequentially according to some hierarchical structure. The choice behaviour
is assumed to be context dependent, i.e., driven by the preferences among
the various attributes. The main significance of this model seems to be its
greater parsimony with respect to data requirements. Its basic message is
not different from the elimination-by-aspects rule.
The prominence hypothesis suggested by Tversky, Sattah and Slovic27
is some kind of truncated lexicographic rule. It draws on Slovic's more-
important-dimension hypothesis 28 . Carrying out four experiments, he ob-
served "that people resolve choices between equally valued, multiattribute
alternatives by selecting the alternative that is superior on the more im-
portant attribute or dimension. 29 "
Slovic's early work later induced the development of the prominence hy-
pothesis. It says essentially that the more prominent attribute will weigh
more heavily in choice. This suggests that, in a way, the prominence hypoth-
esis endeavours to materially establish the preference order of attributes
which would eventually define a lexicographic rule. It seems, however, that
the formulation of the prominence hypothesis has not progressed far beyond
the two-attribute world30 .
22Tversky (1972a), 285.

23Cf. Luce (1958; 1959); Block and Marschak (1960); Chipman (1960); Marschak
(1960); Becker, DeGroot, and Marschak (1963a,b). It was mainly Debreu's (1960) cri-
tique of Luce (1959), which prompted Tversky's (1972a,b) work.
24Tversky (1972a), 292.
25Tversky (1972a), 293f.
26Tversky and Sattah (1979).
27Tversky, Sattah, and Slovic (1988), 372.
28Slovic (1975),281.
29Slovic (1975),286.
30 Notice that Tversky, Sattah, and Slovic (1988), 376 and 380f!., use the prominence
420
When discussing their results, Tversky, Sattah, and Slovic stress the ne-
cessity of further investigations into the prominence hypothesis: "Although
the prominence effect was observed in a variety of settings using both in-
trapersonal and interpersonal comparisons, its boundaries are left to be
explored. How does it extend to options that vary on a large number of
attributes? ... With three or more attributes ... additional considerations
may come into play. For example, people may select the option that is su-
perior on most attributes ... In this case, the prominence hypothesis does
not always result in a lexicographic bias.,,31
2.2.2 Compensatory decision rules

I) Without interattribute comparability
(5) Majority rule 32 :
#{aik I aik > ajk, k = 1,2, ... ,K}

> #{aiklaik<ajk, k=1,2, ... ,K}~ai>-aj;
E(A) {aiEAI~ aj:#{ajklajk>aik, k=1,2, ... ,K}

> #{ajk I ajk < aik, k = 1,2, ... ,K}}.
The majority rule mimics the old wisdom that more arguments in favour
of an alternative are better than less arguments. It assumes that subjects
just count attributes in favour of the respective alternatives and opt for that
one which has more attributes on its side. This decision rule may violate
transitivity and is thus not immune to preference cycles. This means that
there may well exist A' ~ A such that E(A') = 0.
11) With interattribute level comparability
(6) M aximin rule:
hypothesis in conjunction with the compatibility principle to explain the preference

reversal phenomenon [for details cf. Lichtenstein and Slovic (1971)]. According to the
compatibility principle, "the weight of any input component is enhanced by its compat-
ibility with the output ... , For example, the pricing of gambles is likely to emphasize
payoffs more than probability because both the response and the payoffs are expressed
in dollars." [376]
32Russo and Dosher (1983), 683, call this rule majority of confirming dimensions.
421
(7) Maximax rule:
E(A) = {ai E A I ~ aj: max*{ajk} >- max*{aik}}'

k k
The asterisks at min and max indicate that the minimization or max-
imization occurs with respect to preferences among attributes, because
attributes may have different dimensions which cannot be directly com-
pared. Alternatively, we can use (partial) utilities defined on the various
attributes. Notice that the maximin rule is an extension of the conjunc-
tive rule and the maximax rule is an extension of the disjunctive rule if
interattribute level comparability holds33 .
IIl) With cardinal interattribute comparability
(8) Linear multiattribute utility rule:
K K
L Wk vk(aik) > L Wk vk(ajk) => ai >- aj;
k=l k=l
K
E(A) = {ai E A I ai E argmax{L Wk vk(ajk)}},
J k=l
where the Wk'S denote the weights of the attributes and the Vk(') 's denote
the conditional value functions of the attributes. The linear multiattribute
utility rule plays a prominent role in models of multiattribute decision
making34 • It has given rise to other developments such as the use of the
Minkowski metric and the generalized mean.
(9) Multiplicative utility rule:
K K
Il[l + AWk vk(aik)] > Il[l + AWk vk(ajk)] => ai >- aj;
k=l k=l
K
E(A) = {ai E A I ai E argm~x{II[l +AWk vk(ajk)]}},
J k=l
33Shugan (1980), 100, neglects this requirement. There is also confusion about this
condition in Dawes' (1964) article.
34Cf., e.g., Yntema and Torgerson (1961); for a good survey up to the seventies cf.
Slovic and Lichtenstein (1971), in particular 677ff. For later surveys cf. Weber and
Borcherding (1993) and Borcherding, Schmeer, and Weber (1995). For the rigorous es-
tablishment of the linear multiattribute utility rule cf. Farquhar (1977) and Dyer and
Sarin (1979).
422
where A is a general scaling factor35. Einhorn has used offshoots of the

multiplicative utility rule to mimic a conjunctive rule by
and a disjunctive rule by
pretending that these functional forms are noncompensatory rules 36 . How-

ever, inspection shows that these are in fact compensatory rules. Rather
than approximating conjunctive or disjunctive rules, the former can ap-
proximate a maximin rule, and the latter a maximax rule.
2.3 Are decision processes multi-phased?

When testing the hypothesis that decision processes are multi-phased, we
should preferably focus on the simplest possible and largely undisputed
case. This seems to be the hypothesis that the first part of decision processes
consists of an editing phase in which dominated alternatives are eliminated
from further consideration. Among noncompensatory decision rules, the
dominance rule is certainly the least demanding and the most convincing
one. It is not sensible to choose an alternative which is in no respect better,
but for some attributes strictly worse than some other choice alternatives.
This hypothesis was made particularly lucid by Kahneman and Tversky
(1979). In their prospect theory, they distinguished two phases in a choice
process: "an early phase of editing and a subsequent phase of evaluation.
The editing phase consists of a preliminary analysis of the offered prospects,
which often yields a simpler representation of these prospects. In the sec-
ond phase, the edited prospects are evaluated and the prospect of highest
value is chosen. ... The function of the editing phase is to organize and
reformulate the options so as to simplify subsequent evaluation and choice.
Editing consists of the application of several operations that transform the
outcomes and probabilities associated with the offered prospects. 37 " One
of the major operations of the editing phase is the detection and elimina-
tion of dominated alternatives 38 • Kahneman and Tversky maintain that
35Cf. Keeney (1974).

36Einhorn (1970), 227; Einhorn (1971), 3.
37Kahneman and Tversky (1979), 274.
38This assumption is also necessary to immunize prospect theory against violations
of stochastic dominance. Cf. Kahneman and Tversky (1979), 284: "Direct violations
of [stochastic] dominance are prevented, in the present theory, by the assumption that
423
"dominated alternatives ... are rejected without further evaluation. 39 "

These considerations, taken from decision making under risk, readily
carry over to multiattribute decision making under certainty. In this field,
too, many scholars have argued in favour of multi-staged decision processes
whose early stages consist of a screening phase (preceeding the evaluation
phase proper), in which alternatives are eliminated from the option set
following some noncompensatory decision rule4o • The most fundamental
elimination procedures in the screening or editing phase should obviously
encompass the dominance rule. As this rule seems to be the only undis-
puted elimination rule41 in the editing phase, we restrict ourselves to testing
whether an editing phase exists in which the dominance rule applies.
In the course of our experiment, we, therefore, investigate whether domi-
nated alternatives are really eliminated from the choice set. As the possibil-
ity cannot be ruled out that also undominated alternatives are eliminated
in the screening phase, we will focus on searching the subjects' short lists
of alternatives for dominated alternatives.
Taking up a suggestion of Farquhar and Pratkanis, we will consider the
structure of our subjects' short lists with respect to k-dominated alterna-
tives, which means that exactly k options dominate the respective alterna-
tive42 in a subject's short list. In this terminology, undominated alternatives
are referred to as O--dominated alternatives.
3 The experiment
3.1 Stimulus material
The stimulus was an evaluation sheet of the data of 25 applicants for the
position of a chief secretary to be hired. The subjects were told that they
should imagine themselves to be successful entrepreneurs and, since they
were short of time, they entrusted the screening of the applicants for this
position to a professional recruitment agency.
dominated alternatives are detected and. eliminated prior to the evaluation of prospects."
Notice that cumulative prospect theory avoids this problem; cf. Tversky and Kahneman
(1992).
39 Kahneman and Tversky (1979), 275. Notice that the rejection of dominated alterna-
tives has been stated by MacCrimmon (1968a), 15-17, as his third postulate of rational
choice. In this empirical investigations, MacCrimmon reports some violation of this pos-
tulate, but adds that, in the interview after the experiment, all subjects repealed their
violations of MacCrimmon's third postulate, attributing the violations to "carelessness
in their reading or thinking about the problem".
40 Cf. the references in footnote 11.
41Montgomery (1983) has ventured to model decision processes in which decision
makers apply first transformations of the alternatives' attributes in order to render the
dominance rule applicable even in such cases in which there was no dominance in the
primary data.
42Parquhar and Pratkanis (1993), 1223.
424
The recruitment agency assigns a code number to each applicant and

evaluates the applicants with respect to six attributes, viz. 43 :
(i) IQ ... quotient of intelligence defined with a mean value
100, a standard deviation of 15, and the assump-
tion that intelligence is normally distributed;
(ii) ST ... proficiency in shorthand and typewriting to be
measured along a scale ranging from 1 to 100
points;
(iii) L ... proficiency in foreign languages measured as a
weighted index (the weights reflecting the needs
of the firm) along a scale ranging from 1 to 100
points;
(iv) AM ... appearance and good manners of an applicant

points;
(v) EXP /PROF experience and proficiency in office work measured
along a scale ranging from 1 to 10 points;
(vi) COMP ... proficiency in working with personal computers,
points.
The basic evaluation sheet is depicted as Table 1 (at the end of the pa-
per). It exhibits a simple structure. The applicants numbered land 2 excel
with respect to the first attribute, where alternative al dominates alterna-
tive a2, although alternative a2 is, in general, rather similar to alternative
al. Moreover, alternative al is D-dominated (undominated), whereas al-
ternative a2 is l-dominated (by alternative ad, so that, when alternative
al drops out, alternative a2 becomes a O-dominated alternative and can
replace alternative al as some kind of a similarly structured second best
alternative. This pattern is repeated for the six pairs of alternatives al to
a12·
Alternative a13, too, is O-dominated. It is constructed such that the
values of all its attributes are third best. For instance, only alternatives al
and a2 have better values for the first attribute, only alternatives a3 and
a4 have better values for the second attribute, and so on for all attributes.
The remaining alternatives al4 to a25 are at least 2-dominated. All are
dominated by al3 and by at least one alternative of the first twelve al-
43There is evidence that there is some effect of attribute ranges on attributes' weights
in multiattribute decision making; cf. von Nitzsch and Weber (1993). However, we see no
possibility to control for these effects. We could hardly do more than keeping attribute
ranges constant for the two parts of the experiment.
425
ternatives. For example, a14 is dominated by a13 and a1; a15 and a20 are
dominated by 10 alternatives each; etc. The dominance structure is shown
in Table 2 (at the end).
This structure of choice alternatives as exhibited in the basic evaluation
sheet is, of course, too revealing in this form to be presented to the sub-
jects. Therefore, we employed a randomization of the lines of Table 1 and
presented an evaluation sheet with randomly permutated lines to our sub-
jects. (Notice that the chosen randomization was the same for all subjects.)
As we wanted to eliminate biases from data presentation44 , we chose the
matrix form of data presentation, as this mode had proved to engender the
least distortions45 •
For the second part of the experiment, we told subjects that, after several
years, the chief secretary has been transferred to support the establishment
of a new branch of the firm, and a new secretary was to be hired. As the
recruitment agency had performed well, it is again entrusted with the eval-
uation of the applicants. For the second part, we used essentially the same
set of alternatives. In order to camouflage this fact, we employed a different
randomization of the lines, and re-arranged the columns, using AM as the
first attribute, IQ as the second, EXP /PROF as the third, L as the fourth,
COMP as the fifth, and ST as the sixth. We explained the re-arrangement
of columns by telling the subjects that the recruitment agency had changed
its evaluation reports so as to enlist the features which concern an appli-
cant's personality in the first places, and the more technical properties
only thereafter. Moreover, one or two of the previous alternatives (to be
explained in the next section) were deleted, so that a subject was presented
an evaluation sheet containing 23 or 24 of the original 25 alternatives of
the first part of the experiment, arranged, however, in a different order.
3.2 Response method

Several days before the start of the first part of the experiment, the subjects
were introduced to the problem (Le., recruitment of a chief secretary) and
received the agency's evaluation sheet (Le., a randomized version of Table
1). They were told that they should carefully analyze it and think about
a short list of candidates, about the candidate who should be chosen to
be employed, and about the relative importance of the various attributes.
Furthermore, subjects were asked to enter their names in a time-table and
to show up at the agreed time for the first part of the experiment.
The experiment was administered on a computer. The subjects first en-
tered their personal data to enable us to join the individual responses of
44Cf., e.g., Montgomery and Svenson (1976), 287; Bettman and Kakkar (1977), 234;
Svenson (1979), 99; Payne (1982), 391; Russo and Dosher (1983), 677; Johnson and
Meyer (1984), 531ff. and 538.
45Cf. the third experiment of Bettman and Kakkar (1977), as well as Russo (1977).
426
the first and the second part of the experiment. Then the subjects were
asked to order the six attributes according to their importance. They could
state indifference as well as strict preference. We then asked the subjects
for the short lists of their most preferred candidates. They could nominate
up to ten candidates. Then we asked the subjects to state which candidate
they wanted to hire. After the respective response, the subjects were told
that the chosen candidate had just recently withdrawn her or his applica-
tion. The subjects should kindly make another choice. After having done
that, the subjects were informed that this very candidate had meanwhile
accepted another offer and was, therefore, no longer available. They were
told that the recruitment agency had assured that all of the remaining can-
didates were still available. The subjects should kindly accept the agency's
apologies and make one more choice. This concluded the first part of the
experiment. It provided us with data on subjects' short lists and with data
on three actual decisions made for each subject.
We then prepared a second evaluation sheet as described above. The first
best alternative of the first choice was the one to be deleted. As we expected
a distinct preference for alternative a13, we deleted also this alternative, if
a13 happened to be the first or second best alternative in the first part of
the experiment. If the alternative a13 emerged as first best, then the second
best alternative, too, was eliminated. Then the subjects were invited to
indicate the short lists of their most preferred candidates (up to ten) and
the candidates to be hired with first and second priorities.
As the second part of the experiment started only some two weeks later46,
and, as the second evaluation sheet was sufficiently camouflaged, we ex-
pected our subjects to feel as if they had to deal with a separate choice
problem. The second part of the experiment provided us again with data on
subjects' short lists and with data on two decisions made for each subject.
3.3 Procedure
The subjects were 45 students of the University of Kiel, mostly students of
Economics, in their third or fourth year.
The subjects were introduced to the experiment on December 15 and
16, 1994, respectively, and received the first evaluation sheet. At the same
time, they entered their names into a time-table, which allowed them 15
minutes at the computer. The first part of the experiment took place in
the time period between December 19 and December 22, 1994, which left
them more than a weekend to thoroughly analyze the choice problem before
answering our questions. We then processed the second evaluation sheet,
which started from a common re-arrangement of columns and another
46We chose this short spell to exclude major changes of preferences, which would have
invalidated the results gained from our experiment.
427
common randomization of lines 47 • Furthermore, the evaluation sheets were

individualized by deleting the subjects first best alternative of the first part
of the experiment and also the alternative al3 if it had been chosen as the
second best alternative. If the alternative al3 emerged as first best, then
the second best alternative, too, was eliminated. Then these individualized
evaluation sheets were re-numbered (carrying now 23 or 24 applicants)
and were sent by mail to the subjects' private addresses on December 29,
1994. This left them ample time to analyze the second evaluation sheet.
The second part of the experiment took place in the time period between
January 4 and 6, 1995.
All subjects were promised at least 10 Deutsch Marks as an honorarium
for their efforts of participation in the experiment, provided that they had
not given obviously absurd answers, which would indicate their carelessness
in treating this experiment. This proviso was made with the intention to
induce the subjects to undergo a thorough and earnest analysis of the
choice problem before making their choices. Indeed, it had not proven to
be necessary to deny a subject his or her honorarium. We finally paid the
subjects 12 Deutsch Marks each for their participation.
There is a widespread conviction 48 that experiments should conform with
Smith's precepts, in particular with saliency and dominance49 • Saliency re-
quires that subjects are guaranteed the right to claim a reward which is
increasing (decreasing) in good (bad) outcomes of an experiment. Domi-
nance requires that the reward structure dominates any subjective cost as-
sociated with participation in the experiment. Whereas these precepts are
largely undisputed for experiments with outcomes with a natural priority
order (i.e., more money is better than less), it is dubious for experiments in
which respecting undiluted individual preferences is vital. Otherwise, the
reward scheme would distort subjects' behaviour in favour of the values
imposed by the experimenter's reward scheme. This led us to pay our sub-
jects a fixed honorarium. We could have linked it with subjects' effort, e.g.,
design it as a fixed payoff per time unit, but we could not exclude other
distortions of such a reward schedule. As we were fortunate enough to have
interested our subjects in the experiment, and, as we felt that we owe our
subjects some reward in return for their effort, we relied on the combina-
tion of subjects' interest and a moderate lump sum reward. In spite of the
modesty of the financial reward, all subjects participated in both parts of
the experiment, which demonstrates their vivid interest in our experiment.
There is, finally, the problem of the reliability of our data. Subjects are
notoriously susceptible to mistakes and errors in their responses. For in-
stance, "they could misunderstand the nature of the experiment; they could
47This common basic structure of the evaluation sheets of the second part of the
experiment was used to keep possible framing effects to their minimum.
48C£., e.g., Harrison's (1994) recent paper.
49Cf. Smith (1982), 931 and 934.
428
press the wrong key by accident; they could be in a hurry to finish the ex-
periment; they could be motivated by something other than maximizing the
welfare from the experiment per se. 50 " There have been several attempts
to measure subjects' natural error rates 51 • They suggest a natural error
rate of 15-25%, Camerer's 31.6% and Battalio, Kagel and Jiranyakul's less
than 5% being, as it seems, outlyers. As to the error rate of our data we feel
that Table 6 below would provide some good clues. If we take the failure
to choose undominated alternatives in the ultimate decisions as our nat-
ural error rate, this gives us i~ = 26,67%. If we take the failure to choose
alternatives which have been nominated as members of the short lists, we
get a natural error rate of Is
= 15,56%. These two figures delineate pretty
well the interval of commonly recognized error rates.
4 Results
4.1 Testing the editing phase and the elimination of dominated
alternatives
Kahneman and Tversky (1979) and many other authors 52 have claimed
that decision makers approach a decision problem in two phases. In the
first, the so-called editing or screening phase, they simplify the decision
problem, and in the second phase, the evaluation phase proper, they finally
select the choice to be made. One of the more prominent features of the
editing or screening phase is the elimination of all dominated alternatives.
We shall, therefore, test whether the short lists did not exceed the set
of undominated alternatives and whether dominated alternatives were in
fact eliminated by our subjects in a preliminary phase of solving the posed
decision problem. In the first part of our experiment, there are exactly seven
undominated alternatives (Le., the odd-numbered alternatives from al to
a13), six 1-dominated alternatives (Le., the even-numbered alternatives
from a2 to a12), and twelve k-dominated alternatives with k 2:: 2 (Le.,
the alternatives with numbers from a14 to a25). In the second part of the
experiment, there are at least six undominated alternatives, at least four 1-
dominated alternatives, and at least eleven k-dominated alternatives with
k 2:: 2.
We have two groups of data to test the hypothesis of the elimination
of dominated alternatives, viz. the subjects' short lists and their choices
made.
We consider first the subjects' short lists. The results are displayed in
50Hey and di Cagno (1990), 292.

5iCf., e.g., Camerer (1989), 81; Starmer and Sugden (1989), 170; Battalio, Kagel, and
Jiranyakul (1990), 47, note 13; Harless and Camerer (1994), 1263; Hey and Orme (1994),
1296, 1318, and 1320f.
52 See the references in footnote 11.
429
Table 3 for the first part of the experiment, and in Table 4 for the second.
(Both tables are at the end.) In the columns of these tables we list the
number of alternatives in the respective short lists, in the rows the number
of dominated alternatives in the short lists (irrespective of the degree of
domination). Of course, there cannot be more dominated alternatives than
the respective numbers of alternatives in the short lists, which means that
there are only blanks in Tables 3 and 4 below their main diagonals.
Tables 3 and 4 seem to document little evidence of the hypotheses of the
existence of an editing phase and of the elimination of dominated alter-
natives in the editing phase. If all subjects would have complied with the
conditions of the editing phase53 , then we should observe in Tables 3 and
4 nonzero entries only in the first seven (six for Table 4) columns of line
1, which would then, of course, be replicated in the sum line. This means
that the hypotheses of the existence of an editing phase and the elimina-
tion of dominated alternatives are for the first part of the experiment only
consistent for 15.56% CZs) and for the second part only for 4.44% (425) of
all subjects.
We can also have a look at the percentages of subjects whose short lists
do not exceed the number of undominated alternatives. This gives us rates
of 64.44% (~~) for the first part and 55.56% (~~) for the second part of the
experiment. However, this does not explain the occurrence of dominated
alternatives even within this restricted set of short lists. We can perhaps
interpret it as subjects' inability to correctly identify any undominated
alternative.
On average, we observe close to 2 dominated alternatives in the average
short list in part one, and close to 2.5 dominated alternatives in the average
short list of part two of the experiment. No wonder that applying a Mann-
Whitney U-test 54 to test the null hypothesis of the marginal distribution
(1,0,0, ... ,0) against the actual distributions of the right margins of Tables
3 and 4 rejects the null hypothesis at the 1% significance level.
This provokes, of course, the question of the intensity of rejecting the
editing hypothesis. Table 5 informs us about this.
Table 5 shows us that the intensity of the rejection of the hypothesis of
the existence of an editing phase and the elimination of dominated alterna-
tives is modest. 68.44% [61.59% in part two] of all alternatives in the short
lists are undominated alternatives and more than 94% of all alternatives in
the short lists are made up of undominated or 1-dominated alternatives.
These percentages may as well be interpreted as conditional probabilities
that a choice alternative is undominated given that it is a member of a
short list. The conditional probabilities are 0.6844 for the first and 0.6159
53 Remember that this means that the short lists should not exceed the set of undomi-
nated alternatives and all undominated alternatives should be eliminated from the short
lists.
54Cf., e.g., Mood, Graybill, and Boes (1974), 522ff.
430
Table 5: k-dominated Alternatives in the Short Lists

Total O-dominated 1-dominated k-dominated
number of alternatives alternatives alternatives
alternatives (k = 0) (k = 1) (k ~ 2)
in the
short lists number % number % number %
Part I 282 193 68.44 76 26.95 13 4.61
Part 11 289 178 61.59 95 32.87 16 5.54
for the second part of the experiment. The respective pure chance proba-
bilities that an alternative happens to be undominated is 0.28 (;,,) for the
first part of the experiment and 0.26 (263) for the bulk of the second part
of the experiment. These figures lie well below the conditional probabili-
ties, which demonstrates that undominated alternatives loom larger in the
subjects' considerations than dominated alternatives.
The figures shown in Table 5 could indicate that, although subjects aim
at eliminating dominated alternatives from their short lists, their discrimi-
nation between dominated and undominated alternatives is somewhat im-
perfect, which could then explain that 1-dominated alternatives count for
about one fourth to one third of all alternatives in the short lists 55 . Jux-
taposing undominated alternatives in the short lists to k-dominated alter-
natives (k 2 1) gives us 31.56% misperformances for the first part of the
experiment. Although this figure is rather high, it is surprisingly close to
Camerer's 31.6% error rate56 . If we allow for an error rate of 25%, this
means that more than 90% of our subjects would truly comply with the
editing hypothesis, but were prevented by their natural error rate from
accomplishing this goal.
Concerning the second part of the experiment, our subjects' performance
deteriorated. Now 38.41% of k-dominated alternatives were in the short
lists. Allowing an error rate of 25%, this gives still a moderate inconsistency
with the editing hypothesis of 13.41%.
Let us now look at the choices made and at the same time check whether
the first, second (and third) best alternatives were contained in the short
lists. The data is displayed in Table 6.
It shows that some 85% of all choices, taking all five choices in a line,
are contained in the short lists. Considering 45 X 5 = 225 actual choice
acts, the hit rate increases to some 97% of choices being in the short lists.
55 Notice that subjects' performance deteriorates in the second part of the experiment.
5 6 Camerer (1989), 81.
431
Table 6: Structure of the Choices Made

Total First best Second best Third best
number of alternatives alternatives alternatives
cases
undom in SL undom in SL undom in SL
Part I 45 45 45 42 43 38 40
Part 11 45 43 45 43 45 - -
This perfomance shows an improving pattern: In the second part of the

experiment, 100% of all actual choices were contained in the short lists.
We find, however, decreasing fractions of undominated alternatives as the
best choices when previously chosen alternatives became irrelevant S7 . For
instance, in the first part of the experiment after two consecutive choices the
best choices of the remaining alternatives shrink to some 84% undominated
alternatives onlyS8.
4.2 Testing various decision rules

4.2.1 Testing the conjunctive rule with endogenous cutoff
scores
As a good secretary should satisfy good standards of all relevant attributes,
and as the disjunctive rule has performed rather unsatisfactorily in other
empirical research s9 , we shall restrict our interest to the conjunctive rule
and to various lexicographic rules among the non compensatory decision
rules. This section considers the conjunctive rule. Rather than pretending
that some minimum attribute levels have been forced upon the decision
maker by some outside authority60, we are taking up Dawes' suggestion
of deriving the cutoff scores of the attributes endogenously61, following
from the decision maker's goal to whittle down the choice set to a desired
number of alternatives which should remain in the short list for further
evaluation. This method can be used to determine a subject's short list of
57Notice that our calculations considered, of course, the changing patterns of undom-
inated alternatives as previously chosen alternatives become unavailable.
58This influence of chosen alternatives which become irrelevant is studied in another
paper. An alternative hypothesis of an editing phase is the compatibility test of image
theory. The test of compatibility is rather tedious and requires sophisticated numerical
methods. Therefore, we investigate it in a separate paper [cf. Seidl and Traub (1998)].
59Cf. Einhorn (1970; 1971). We shall see below (Section 4.2.6) that the maximax rule,
which is closely related to the disjunctive rule, performs equally poor.
60This was the way Wright and Barbour (1977) modelled the screening phase of their
experiment.
61 Dawes (1964), 105ff.
432
choice alternatives (and thus model the subject's editing phase) as well as
to determine his ultimate choice if he intends to employ the conjunctive
rule right to the end of his decision process.
Let us illustrate the latter case for 25 choice alternatives and 6 attributes.
If the subject rates all attributes equally and wants to select exactly one in
25 alternatives according to a conjuctive rule, then he will determine the
cutoff scores of the attributes endogenously from the formula
p6 = 215 => p = 25-t = 0.5848 .

This means that the cutoff scores begin for each attribute after the 58.48%
best attribute values and eliminates the 41.52% of the alternatives with
the worst attribute values. Proceeding sequentially, this leaves the decision
maker with exactly
25 X (0.5848)6 ~ 1
choice alternative left. Proceeding along these lines, we find the cutoff scores
for six equally rated attributes and 22 to 24 alternatives (i.e., for all option
sets arising in our experiments) from 62
p6 2~ => P(24) = 0.5888;

I
23 => P(23) 0.5930;
I
22 => P(22) 0.5974.
The cutoff scores are then gained from the acceptance of the best ipCi)
(i = 22,23,24,25) values of the respective attribute values.
Suppose now that the subject has a strict order of the attributes, say
d l )- d 2 )- d3 )- d4 )- ds )- d 6 • Everything else remaining the same, we can
model this as
p2I =
25
2:..
=> P = 25-* = 0.858.
This gives for the various attributes:
PI = p6 0.399;
P2 pS 0.465;
P3 = p4 0.542;
P4 p3 0.631;
Ps p2 0.736;
P6 = pI 0.858.
62If necessary, we indicate the number of alternatives, to which the cutoff probability
refers, in br ackets.
433
The cutoff score for the first attribute begins after the best 39.9% attribute
values and ends for the sixth attribute close after the best 85.8% attribute
values. This reflects that the subject is more demanding with respect to
more important attributes requiring higher cutoff scores. Suppose the cho-
sen alternative becomes invalid. Then, for the next step of the decision,
this procedure has to be repeated for 24 choice alternatives, etc.
H the subject wants to whittle down his choice set to 10 alternatives
(forming then his short list resulting from the editing phase) by means of
a conjunctive rule, then he derives his cutoff scores from the formula
6 10
p = 25 ::;. p = 0.858,
if the subject rates all attributes equally.

Alas, tempting as this model of a conjunctive rule looks, this procedure
is not immune to path-dependency. The short list or the ultimately chosen
alternative may well depend on the order in which the various attributes
are screened. Therefore, sensitivity analyses (applying different orders of
screening the attributes) are expedient in cases of equally rated attributes.
In cases of a preference order of attributes, this provides a natural sequence
for the screening of attributes.
The path-dependency caused excessive computational effort. For a single
decision of a subject with equally rated attributes we had to calculate
6! = 720 combinations. This is easily seen: We can start with each of the
six attributes, continue with each of the five remaining attributes etc., until
we have checked all 720 combinations. As this procedure has to be applied
to all of a subject's five decision, we had to calculate 3, 600 combinations of a
single subject with equally rated alternatives. This immense computational
effort induced us to confine ourselves to investigate the conjunctive rule only
for the actual choice taken, rather than investigating also the short lists.
Whenever a subject chooses an alternative which is at the same time also
an element of an optimum combination, this subject scores a hit. Table 7 (at
the end) summarizes the data gained from investigating Dawes' conjunctive
rule.
Table 7 reports the structure of hits at all five decisions (denoted by Di to
Ds) made. 1 denotes a hit according to the conjunctive rule, 0 signals that
the subject failed to make his or her decision according to the conjunctive
rule. All hit combinations which did not actually occur were, of course,
deleted from Table 7. Di to D3 concern the first part of the experiment,
D4 and Ds the second. Allowing for at most one error in five decisions, we
see that the choice behaviour of some 20% of our subjects can be modelled
according to Dawes' conjunctive rule.
Table 7 also reveals some interesting relationships about conditional
probabilities. For instance, out of the 15 subjects which conformed at least
twice with the conjunctive rule in the first part of the experiment, 12 con-
formed with the conjunctive rule for their first choice in part two, and
434
7 conformed for both choices in part two with the conjunctive rule. This
gives conditional probabilities63 of 0.8 and 0.47 respectively. This shows
again that a group of subjects look as if they acted in conformity with the
conjunctive rule.
Table 8 (at the end)informs on the hit rates at different decision or-
ders. We see that the hit rates are markedly higher whenever the decision
problem is analyzed from scratch (first and fourth decisions) and is thus
unadulterated from the frustrating experience of past choices which had
become invalid.
4.2.2 Testing the eUmination-by-aspects rule

Recall that Tversky (1972a,b) modelled the elimination-by-aspects rule in
the framework of stochastic choice theory. Our experimental design, which
presents a certain choice situation only once to a subject (and does not
generate data on the full preference orderings of all alternatives) does not
allow calculations as required by these decision rules. However, their more
basic message seems to be attractive enough to try our best to test them
in some rudimentary forms applicable to our data. We can, for instance,
apply a between--subjects analysis and look for some positive correlation
between the attractiveness of alternatives and the number of other alter-
natives which they dominate 64 , or we can apply some other pattern elimi-
nation processes.
Starting with the second approach, we test the elimination-by-aspects
rule using an algorithm which gradually lowers the attribute values and
checks at each step whether some of the actual alternatives dominates these
sham vectors. To illustrate, the first sham alternative is composed by the
best attribute values in each row of Table 1, viz.
(120,90,95,10,10,10) .
Of course, no alternative dominates this vector. Then we compose a sham

vector using the second best values of all attributes. This gives
(118,88,92,9,9,9) .
Again, no alternative dominates this vector. The next vector constructed

in this way is the alternative a13, which places this alternative at the top.
63The condition being that the subject conformed at least twice with the conjunctive
rule in the first part of the experiment.
64Cf. Tversky (1972a), 295, to see that this is a valid interpretation of the elimination-
by-aspects hypothesis. Many other scholars, too, have found empirical evidence that the
addition of dominated alternatives increases the attractiveness of the now dominating
alternatives. Cf., e.g., Huber, Payne, and Puto (1982); Huber and Puto (1983); Tyszka
(1983); Ratneshwar, Shocker, and Steward (1987); Wedell (1991).
435
Proceeding further in this way, we arrive after some other sham vectors at
the sham vector
(104,73,72,5,4,4) ,
which is dominated by alternative ag. Applying this algorithm further gives
us for the top five items the preference order
al3 >- ag >- alO >- al >- a22 .
Table 9 examines the performance of the top three alternatives in part

one of the experiment.
Table 9: The Performance of a13, ag, and aID
al3 >- ag >- alO 9 times chosen
al3 >- ag at top 15 times chosen
al3 37 times first; 5 times second; 1 time third
This result is very appealing. Table 9 shows us that the ordering of

choices al3 >- ag >- alO emerged in 20% of all responses. If all 25 alternatives
are considered, this preference ordering would evolve by chance only with
a probability of 0.000072464 = 7.2464 X 10- 5 • If only the first thirteen
alternatives are taken as eligible, it still can arise only with a probability
of 0.00058275 = 5.8275 x 10- 4 . The ordering al3 >- ag at top was chosen in
one third of all responses. Considering all 25 alternatives, it would evolve
by chance only with a probability of 0.0016667 = 1.6667 x 10- 3 . If only the
first thirteen alternatives are considered as eligible, the ordering al3 >- ag
at top arises only with a probability of 0.0064102 = 6.4102 x 10- 3 . This
shows that the elimination-by-aspects rule provides a rather good forecast
of subjects' actual choices.
The second part of the experiment was tested separately, because, as a
consequence of error, subjects may have failed to realize a13 >- ag >- alO in
the first part of the experiment. Starting their decision process from scratch
in the second part of the experiment may turn out to be more successful.
Yet the choice situations may well be different. Therefore, we used the
above algorithm to individually compute the best alternatives under an
elimination-by-aspects rule and ask whether the subjects indeed followed
the respective orders, say ai >- aj. This is summarized in Table 10.
Table 10: The Performance of the Best Alternatives in Part Two

a;, >- aj 8 times chosen
a;, at top 22 times chosen
a;, second 11 times chosen
This shows again results which are far above the realizations which would
evolve from pure chance.
436
Another way to test the elimination-by-aspects rule is to explore the

correlation between the attractiveness of alternatives (in terms of actual
choices) and the number of alternatives which they dominate. For the first
choice in the first part of the experiment, we have no problems, because
all subjects are faced with the same option set. Correlating thus the actual
choices of the various alternatives against the number of the alternatives
which they dominate gives a correlation coefficient of 0.6037319.
One might surmise that this high correlation coefficient is influenced by
the fact that a13 was chosen 37 times as the first best choice, and a13
dominates 12 other choice alternatives. Therefore, we calculated another
correlation coefficient, taking the 37 subjects who had chosen a13 as their
first best alternative as a basis. These 37 subjects all faced the same op-
tion set for their second best choice. Correlating their second best choices
against the number of alternatives which they dominate yields a correla-
tion coefficient of 0.6035214, which is nearly equal to the former correlation
coefficient.
Things become more complicated for the second part of the experiment,
because subjects face different decision problems. We tried to allow for this
by weighting alternatives for their availability. For instance, a13 was only
available for three subjects in part two and was chosen two times. This gives
a ratio of i. Weighting this for availability gives ~ X 45 = 30. That is, we
pretend that a13 would have been chosen by 30 subjects if it were available
for all subjects. This procedure is not entirely satisfactory, as it feigns that
there was a fictitious choice involving (as this procedure works out) some
80 subjects instead of 45, but, as a correlation coefficient normalizes the
figures anyway, we thought that we should try this approach. We find a
correlation coefficient of 0.8215491.
These results show that we observe a distinctly positive correlation be-
tween the attractiveness of an alternative (in terms of actual choice) and
its number of dominated other alternatives.
With respect to our results we are entitled to say that the choice behav-
iour of at least 20% of our subjects is consistent with the elimination-by-
aspects rule.
4.2.3 Testing the prominence hypothesis

There seems to be no clearcut recommendation as to how to test the promi-
nence hypothesis in a multiattribute decision problem with K attributes
for K > 2.65 We are trying a rank correlation between the alternatives in
the short lists and indicated ranks of the attributes.
As logit and probit do not yield significant estimates, we employ a linear
probability model of the alternatives in the short lists and the order of
the attributes as indicated by the subjects. The linear probability model is

437
derived from
6
7ri = f30 + L f3 k Q,ik +Si ,

k=l
where 7ri is accorded the value 1 if alternative i was a member of the short
list, and 0 otherwise. The f3k's represent the weights of the standardized
attribute values
A aik-minj{ajk}
aik := maxj
{ } - mInj
ajk . { ajk } '
f3 0 denotes a constant, and Si denotes the error term. Because of het-

eroscedascity of the error term we employed a two-stage Aitken (weighted
least squares) estimation for the f3 k 's (deleting alternatives for which
7rT st rt. [O,lJ after the first stage of the estimation)66. Then all f3k's which
did not satisfy the 5% significance level were eliminated (Le., set equal to
zero). The remaining f3 k 's were then transformed into ranks, treating f3 k 's
which did not differ by more than 5% as identical. For identical coefficients,
a mid-ranking procedure was applied. Then we computed Spearman's rank
correlation coefficients for the ranks of the attributes derived in this way
and the ranks of the attributes as indicated by the subjects. This gave us
two rank correlation coefficients for each subject, one for the first part and
one for the second part of the experiment.
The result is not particularly encouraging. The rank correlation coeffi-
cients do not point strongly in anyone direction. We observe positive, nega-
tive, and near-zero rank correlation coefficients. The average rank correla-
tion coefficient is 0.3003 for the first part and 0.3625 for the second part of
the experiment. Using these two sets of 45 individual rank correlation coef-
ficients to compute the ordinary (Bravais-Pearson) correlation coefficient,
we get a value of 0.0118, which means that there is hardly any correlation
between the rank correlation coefficient of the first and the second part of
the experiment. Thus, the rank correlation analysis rejects the prominence
hypothesis.
We also tried out two other methods to test the prominence hypothe-
sis with similarly poor results. Therefore, we refrain from reporting them
here 67 .
4.2.4 Testing the majority rule

The majority rule asserts that subjects form their preferences on choice
alternatives by pairwise comparisons, simply counting the attributes in
favour of one or the other. If we have an equal number of attributes in
66Cf. Goldberger (1964), 249f.j Gujarati (1988), 468-473.

67Einhorn, too, failed to discover a marked relationship between the ratings of at-
tributes and subjects' decisions. Cf. Einhorn (1971), 17: "The ratings of importance for
each subject for each attribute were collected and analyzed. Unfortunately, there was no
consistent pattern of use for any particular variable".
438
favour of either of two alternatives, the majority rule indicates indifference.

Let us code ai )- aj, i indicating the line and j the column of an adjacency
matrix, by 1, ai aj by 0, and ai ~ aj by a hyphen, we can form the
I'V
priority structure of the majority rule in Table 11 (at the end), which
corresponds to Table 2 for the dominance structure.
From Table 11, we immediately see that, neglecting ties, three alter-
natives form the unrivalled winners in the order al3 )- ag )- aw, which
compares favourably with our test of the elimination-by-aspects rule. Fur-
thermore, we see that the majority rule is plagued by so many intransitiv-
ities, that any subject would simply be lost without having the powerful
instrument of an adjacency matrix at hand.
Therefore, we confined ourselves to testing our subjects' internal inconsis-
tencies of their actual choices. Suppose, for instance, that a subject's pref-
erences for alternatives, as revealed by his actual choices, are al )- a2 )- a3
for the first part of the experiment. Denoting weak priority under a major-
ity rule (preference or indifference) by R and strict priority by P, we should
expect alRa2Ra3 if the majority rule is indeed followed. Comparing these
two orders, the first one being given, we could find up to three priority
violations for part one of the experiment. For instance,
means exactly one priority violation. Table 12 lists the priority violations
observed in our experiment.
Table 12
Priority Violations of the Majority Rule
Priority violations
0 1 2 3
Part I 22 22 1 0
Part II 32 13 - -
It is interesting to find some structure in the priority violations of the

majority rule. A probit analysis shows that the subjects with more priority
violations in the first part of the experiment exhibit a greater tendency
for priority violations in the second part of the experiment 68 • Moreover,
this tendency increases with a rising number of priority violations in the
first part of the experiment. As to our notation, X denotes the number of
priority violations in the first part of the experiment, and <.p( o+,BX) gives us
the probability of a priority violation in the second part of the experiment
68Notice, however, that the coefficient f3 in our probit estimate is significant only at
the 7% level.
439
given X priority violations in the first part of the experiment, where <P
denotes the standard normal distribution function. Table 14 shows that
the probability of a priority violation in part two is an increasing function
of the number of priority violations in part one of the experiment.
Table 13
Probit Estimates for the Majority Rule
Label a f3
Coeff. -0.9622 0.6946
Std.Error 0.3084 0.3825
t-Stat. -3.1199 1.8159
Level of Sign. 0.0018 0.0694
Table 14
Cond. Probabilities of Priority Violations in Part Two
x 0 1 2 3
CI>(a + f3X) 0.1685 0.3936 0.6664 (0.8686)
This obviously suggests that we have two rather stable cohorts among
our subjects. One cohort seems to make consistent use of the majority rule,
whereas the other cohort does not seem to pay particular attention to it.
This conjecture is confirmed by Table 15 which shows the violations of the
majority rule.
Table 15: Violations of the Majority Rule

Number of Violations Conditional Probabilities, given
of the Majority rule Frequency n Violations in Part One
in part one in part two no. of % n=O n=l n=2
subjects
0 0 18 30.0 0.8182 - -
0 1 4 8.9 0.1818 - -
1 0 14 31.1 - 0.6364 -
1 1 8 17.8 - 0.3636 -
2 1 1 2.2 - - 1.0
Table 15 corresponds pretty well to the results of Table 14. The condi-
tional probabilities of a violation of the majority rule in part two, given
440
no violation in part one, are 0.1685 (Table 14), and 0.1818 (Table 15). For
one violation in part one the respective probabilities are 0.3936 (Table 14),
and 0.3636 (Table 15). For two violations in part one we have 0.6664 from
Table 14, and 1.0 from Table 15. Moreover, Table 15 shows us that 30% of
our subjects exhibit a behaviour which is fully consistent with the majority
rule.
4.2.5 Insufficient data base to test the linear and the

multiplicative multiattribute utility rules
Any experimenter must accept compromises when conducting an experi-
ment. Testing the linear multiattribute utility rule
K
L wkvk(aik)
k=l
and the multiplicative multiattribute utility rule

K
Il[1 + AWkvk(aik)]
k=l
requires a special experimental design which is attuned to this task.

First, we would have to test for mutual preference independence [for
the linear rule] or for weak difference independence [for the multiplicative
rule] of the attributes 69 • These conditions warrant that the Vk (. )-functions
can be evaluated independently of the values of all other attributes. If an
experiment shows serious violations of these requirements, then the linear
and/or the multiplicative multiattribute utility rule do not apply.
If the independence requirement is met, the vk(-)-functions and the
weights Wk have to be elicited. The vk(-)-functions can be derived either
by methods of numerical estimation (direct rating, category estimation,
ratio estimation, curve drawing), or by methods employing indifference re-
lations (difference standard sequence, bisection, dual standard sequence,
sequential trade-off). The weights, too, can be derived either by meth-
ods of numerical estimation (ranking, direct rating, ratio estimation, swing
weights) or by using indifference relations (cross-attribute indifference,
cross-attribute strength of preference). Selected combinations of deriving
the vk(·)-functions and the weights Wk have been given special names,
such as SMART (simple multiattribute rating technique), difference value
measurement, conjoint measurement, weak-order model, etc. 70
Our data base contains just our subjects' rankings of attributes, which
can be put to good use to derive the weights Wk, but we have no data
6 9 For details cf. Dyer and Sarin (1979),812-815; Keeney (1974).

7oFor details cf. von Winterfeldt and Edwards (1986), chapters 7 and 8; Keeney and
Raiffa (1976), chapter 3.
441
whatsoever to derive the vk(·)-functions. The data requirements to derive

vkO-functions for six attributes are indeed formidable and require exper-
iments which focus entirely on this task. Although our data base allows
testing several multiattribute decision rules, it is decisively insufficient to
test the linear or the multiplicative multiattribute utility rules71.
We also considered making use of the independent multinomial logit
model, which has been used in the analysis of brand choice behaviour. This
approach estimates the weights Wk from
where P(ai I A) denotes the probability that the choice alternative ai is

chosen from the set of alternatives A.72 However, this approach assumes
that the subjects are homogeneous and behave in accordance with the
random utility model. As our subjects exhibit rather diverse rankings of
the attributes, common attribute weights for all subjects would come up to
a gross misspecification of the choice model. Therefore, we did not consider
the independent multinomial logit model as a serious candidate to mimick
the linear multiattribute utility rule. It may be appropriate as a description
of brand choice, but not for multiattribute choices in general.
4.2.6 Testing the maximin and maximax rules

Following Einhorn (1970, 1971), we approximate the maximin rule as
and the maximax rule as
The (}:ik'S are the values of the attributes of the various choice alternatives,
normalized on the unit interval. The exponents f3 k were derived from the
subjects' attribute rankings in the following way. For indifference between
all attributes we used f3k = i 'if k = 1,2, ... ,6. For strict preference or-
derings, we used 261 as an exponent for the highest ranked attribute, i1
as an exponent for the second highest ranked attribute, etc., ending with
2\ for the last ranked attribute. If we have, for instance, one attribute at
71 Meyer and Johnson (1995), GI83f., report a poor performance of the linear multi-
att.ribute ut.ility rule.
72Cf., e.g., Manrai (1995), 5.
442
the highest rank, three attributes second ranked, and two attributes at the
third rank, we applied the exponents 131 for the highest ranked attribute,
121 for each of the three second ranked attributes, and A for each of the
two third ranked attributes. These examples should suffice to clarify the
rule of construction of the exponents.
Table 16 informs on the compliance of subjects' responses with the max-
imin rule. It reports the structure of hits at all five decisions (denoted by
D1 to D 5 ) made. 1 denotes a hit according to the maximin rule, 0 a failure
to comply with the maximin rule. All hit combinations which did not actu-
ally occur were deleted. D1 to D3 concern the first part of the experiment,
D4 and D5 the second.
Table 16 shows that, while 20% of our subject acted in conformity with
the maximin rule in the first part of the experiment, not a single subject
acted for both choices in the second part of the experiment in conformity
with the maximin rule. Thus, it seems that the conformity of agents' choices
with the maximin rule decreases as the experiment continues.
How can subjects' compliance with the maximin rule be evaluated when
we allow for some error, e.g., missing the maximin rule by just one alter-
native. Table 17 informs about this.
Table 17 shows that, allowing for an error of missing the best alternative
according to the maximin rule but by one alternative, we have a hit rate
(except the 5th choice) of more than 64%. Thus, although the maximin
rule is not followed for all of a subject's choices, our data show that this
rule generally enjoys great attention among our subjects.
In contrast, things are not very encouraging for the maximax rule. The
maximax rule involves the choice of the parameter c, where c > 1. Our
calculations show us that the choice alternatives endorsed by the maximax
rule are extremely sensitive to the value of c. For c = 1.05, only 9 subjects
are in conformity with the maximax rule for the first choice of part one of
the experiment. This figure rises to 15 for c = 1.1, and to 26 for c = 1.2. The
hit rate increases sharply with rising c, which reflects that greater values
of c reduce the differences between the choice alternatives by assigning the
same maximum value to more alternatives. Therefore, we conclude that the
maximax rule is not a good explanation of individual behaviour and we do
not bother the reader with numerical results of our tests of this rule.
5 Conclusion
We utilized a choice experiment presented in the context of recruiting a
secretary to investigate whether decision processes are multi-phased and
which decision rules are consistent with subjects' choice behaviour. In the
most elementary form, mulit-phased decision processes should manifest
at least in the form of an editing phase, in which dominated alternatives
are eliminated from further consideration, and a decision phase proper.
Furthermore, we tested whether individual choice behaviour is consistent
443
with various decision rules which were surveyed in Section 2.2.

We have found that, applying a strict yardstick, the existence of an edit-
ing phase and the elimination of dominated alternatives is not confirmed.
We observed an average of more than two dominated alternatives in the
short lists and rejection rates of more than 60%. However, the intensity of
rejecting the existence of an editing phase and the elimination of dominated
alternatives is modest. 60% and more of all alternatives in the short lists
are undominated alternatives and more than 94% of all alternatives in the
short lists are either undominated or I-dominated alternatives. Allowing
for an error rate of some 25% implies that some 90% of our subjects would
truly comply with the editing hypothesis, a picture which is only obscured
by their natural error rates.
As to the testing of decision rules, we tested the conjunctive rule with
endogenous cutoff scores, the elimination-by-aspects rule, the prominence
hypothesis, the majority rule, and the maximin and maximax rules.
The conjunctive rule was tested using endogeneously determined cutoff
scores to determine the subjects' ultimate choices. We found that some
20% of our subjects seem to consistently apply the conjunctive rule for
their ultimate choices.
We tested the elimination-by-aspects rule using an elimination algo-
rithm which provided us with preference orderings of alternatives. These
fitted surprisingly well. Their good performance ranks far above the random
choice probabilities, which means that the behaviour of a substantial part
of our subjects is consistent with the elimination-by-aspects rule. Trying
correlations between chosen alternatives and the numbers of the alterna-
tives which they dominate (another proxy for the elimination-by-aspects
rule) provided encouraging results, too.
In order to test the prominence hypothesis, we employed a linear prob-
ability model applied to the alternatives in the short lists. Then, for each
subject, we computed Spearman's rank correlation coefficient between the
ranks of the attributes derived in this way and the ranks of the attributes as
indicated by the subjects. Our results suggest a rejection of the prominence
hypothesis.
The majority rule, on the other hand, has a remarkably good perfor-
mance. For the first part of the experiment, it yields the same top ranked
alternatives in the order a13 >- ag >- alO as the elimination-by-aspects rule
and, thus, shares its good performance. Moreover, we found evidence that
our subjects split into two cohorts, one complying with the majority rule
and the other one disregarding it.
The maximin rule did remarkably well. Allowing for a natural error to
miss the best alternative but by one alternative gave us a hit rate of more
than 64%. The maximax rule, in contrast, provided rather discouraging
results and should, therefore, be discarded. This shows that our experiment
has shed light on the existence of an editing phase and on the spread of
using various multiattribute decision rules.
444
Table 1: Basic Evaluation Sheet

Appl.
COMP
No.
1 120 70 75 6 8 6
2 118 65 73 5 7 5
3 95 90 67 8 7 8
4 94 88 66 8 7 7
5 97 68 95 8 6 8
6 96 66 92 7 5 8
7 101 72 59 10 8 6
8 100 69 57 9 7 5
9 104 75 72 8 10 7
10 103 73 69 8 9 7
11 108 81 62 6 7 10
12 107 79 60 6 7 9
13 109 85 82 8 8 8
14 105 62 70 5 7 6
15 91 59 62 5 7 5
16 88 81 55 8 7 5
17 79 66 64 7 6 6
18 92 57 80 6 5 7
19 88 63 50 7 4 5
20 96 60 55 6 6 4
21 99 48 52 5 6 5
22 100 71 65 7 7 7
23 102 66 48 6 6 7
24 96 75 51 5 4 8
25 104 62 46 6 7 5
Table 2: The Dominance Structure of the Choice Alternatives
2 4 5 678 10 11 12 13 14 15 16 17 18 19 20 21 222324251E
6
2 2
3
4 4
5 5
2
7 4
3
9 9
10 7
11 7
12 5
13 12
14 2
15 o
16
17 o
18 o
19 o
20 o
21 o
22 5
23 o
24 o
25 o
E o o o o o o o 2 10 3 7 3 11 10 11 3 5 3 5
1 means that alternative i (line) dominates alternative j (column). The line sums indicate the number of alternatives which are
dominated by the alternative of the respective line. The column sums indicate the number of alternatives which dominate the
alternative of the respective column (k-dominance).
~
446
Table 3: Dominance Structure of the Short Lists ( Part One)

N umber of
dominated Number of alternatives
alteruatives in the short lists
in the
short lists 1 2 3 4 5 6 7 8 9 10 sum
0 0 1 3 2 0 0 1 -~
7
1 0 0 4 3 3 2 0 0 12
2 0 1 0 4 0 1 1 2 -~
9
3 - - 0 0 0 0 3 2 1 4 1()
4 () () () 1 2 0 3 6
5 ~-
- _. () 0 () () () 1 1
6 - - -
0 () 0 () () ()
7 () 0 () 0 0
8 - () 0 0 0
9 0 () ()
1() - - -- - - () ()
SUlll () 1 8 5 7 2 6 5 3 8 45
Table 4: Dominance Structure of the Short Lists (Part Two)
() () () 1 () 1 () -* -
2
1 0 () 1 3 4 () 1 -* 9
2 0 2 4 4 3 1 1 1* 16
3 - () 0 0 1 1 2 2 -* 6
4 () 0 1 1 1 2 5 10
5 - - - () 0 () () 1 1 2
6 - - -
0 () () () () ()
7 -
0 () () 0 ()
8 ~- - () () () ()
9 - - () 0 0
1() - ~- - - () 0
SUlll () () 4 7 9 5 4 4 6 6 45
* N ou-blank possibility for one subject ouly
447
Table 7: Conformity with the Conjunctive R.ule
# Dl D2 D3 D4 Ds Frequency %
of hit.s
0 0 0 0 0 0 5 11.11 11.11
1 0 0 0 0 8 17.78
0 1 0 0 0 4 8.89
1 0 0 0 1 0 1 2.22
0 0 0 0 1 1 2.22 31.11
1 0 1 0 0 1 2.22
1 0 0 1 0 3 6.67
1 0 0 0 1 2 4.44
2 0 1 1 0 0 1 2.22
0 0 1 1 0 2 4.44
0 0 1 0 1 2 4.44 24.43
1 1 0 1 0 2 4.44
1 0 1 1 0 1 2.22
3 1 0 1 0 1 1 2.22
1 0 0 1 1 2 4.44 13.32
1 1 1 1 0 2 4.44
4 1 1 0 1 1 2 4.44
1 0 1 1 1 1 2.22 1l.l0
5 1 1 1 1 1 4 8.89 8.89
Sum 45 lOO 100
Table 8: Hit R.ates at Different Decision Orders

Decision order # of hits Hit rate
First decision 29 64.44%
Second decision 15 33.33%
Third decision 15 33.33%
Fourt.h decision 20 44.44%
Fifth decision 15 33.33%
t
00
Table 11: Priority Structure of the Majority Rule
2 3 4 5 6 7 8 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
0 0 0 0 0 1 0
2 0 0 0 0 0 0
3 0 0 0 1 0
4 0 0 0 0
5 0 0 0 0
6 0 0 0 0
7 0 1 0 0
8 0 0
0 1 0
10 0 0
11 0 0
12 0 0 0
13 0
14 0 0 1
15 0 0 0
16 0 0 0 0 0
17 0 0 0
18 0 0 0 0 0
19 0 0
20 0 0 0 1
21 0
22 0 0
23 0 0 0 0
24 0 0 0
25 0 0
449
Tahle 16: Suhjects' Choices Following the Maximin Rule

# of hits Dl D2 Da D4 Ds Frequency % Group %
1 0 0 0 0 5 11.11
1 0 1 0 0 0 2 4.44
0 0 1 0 0 1 2.22
0 0 0 1 0 2 4.44 22.22
1 1 0 0 0 2 4.44
1 0 1 0 0 2 4.44
2 1 0 0 1 0 3 6.67
1 0 0 0 1 4 8.89
0 1 0 1 0 1 2.22
0 1 0 0 1 1 2.22 28.89
1 1 0 1 0 4 8.89
1 1 0 0 1 2 4.44
3 1 0 1 1 0 5 11.11
1 0 1 0 1 1 2.22
0 1 1 1 0 1 2.22 28.89
1 1 1 1 0 5 11.11
4 1 1 1 0 1 4 8.89 20.00
Sum 45 100.00 100.00
Table 17: Compliance with the Maximin Rule

Actual Chosen alternative Chosen alternative Chosen alternative
Choices complies with is second best ranks lower than
maximin rule according to second best
maximin rule according to
maximin rule
1st 37 3 5
2nd 22 10 13
3rd 19 10 16
4th 21 11 13
5th 12 12 21
450
References
Battalio, RC., Kagel, J.H., and Jiranyakul, K (1990) "Testing Betwen
Alternative Models of Choice Under Uncertainty: Some Initial Results",
Journal of Risk and Uncertainty 3, 25-50.
Becker, G.M., DeGroot, M.H., and Marschak, J. (1963a) "Stochastic Mod-
els of Choice Behavior", Behavioral Science 8,41-55.
Becker, G.M., DeGroot, M.H., and Marschak, J. (1963b) "An Experimental
Study of Some Stochastic Models for Wagers" , Behavioral Science 8, 199-
202.
Bern, D.J. (1972) "Self-Perception Theory", in: L. Berkowitz (ed.), Ad-
vances in Experimental Social Psychology, Academic Press: New York,
1-62.
Bettman, J.R, and Kakkar, P. (1977) "Effects of Information Presenta-
tion Format on Consumer Information Acquisition Strategies", Journal
of Consumer Research 3,233-240.
Block, H.D., and Marschak, J. (1960) "Random Orderings and Stochastic
Theories of Response", in: I. Olkin et al. (eds.), Contributions to Prob-
ability and Statistics: Essays in Honor of Harold Hotelling., Stanford
University Press: Stanford, 97-132.
Borcherding, K, Schmeer, S., and Weber, M. (1995) "Biases in Multiat-
tribute Weight Elicitation", in: J.-P. Caverni, M. Bar-Hillel, H. Barron,
and H. Jungermann (eds.), Contributions to Decision Research, North-
Holland: Amsterdam, 3-28.
Camerer, C.F. (1989) "An Experimental Test of Seven Generalized Utility
Theories", Journal of Risk and Uncertainty 2, 61-104.
Chipman, J.S. (1960) "Stochastic Choice and Subjective Probability", in:
D. Willner (ed.), Decisions, Values and Groups, Pergamon Press: Oxford-
London-New York, 70-95.
Conlisk, J. (1988) "Optimization Cost", Journal of Economic Behavior and
Organization 9, 213-228.
Coombs, C.H. (1964) A Theory of Data, Wiley: New York.
Coombs, C.H., and Kao, RC. (1955) "Nonmetric Factor Analysis", Re-
search Bulletin No. 38, Engineering Research Institute, University of
Michigan: Ann Arbor.
Dawes, RM. (1964) "Social Selection Based on Multidimensional Criteria",
Journal of Abnormal and Social Psychology 68, 104-109.
Dawes, RM. (1988) Rational Choice in an Uncertain World, Harcourt
Brace Jovanovich: San Diego.
Debreu, G. (1960) "Review of RD. Luce, Individual Choice Behavior: A
Theoretical Analysis", American Economic Review 50, 186-188.
Dyer, J.S., and Sarin, RK (1979) "Measurable Multiattribute Value Func-
tions", Operations Research 27, 810-822.
Einhorn, H.J. (1970) "The Use of Nonlinear, Noncompensatory Models in
Decision Making", Psychological Bulletin 73, 221-230.
451
Einhorn, H.J. (1971) "Use of Nonlinear, Nonc6mpensatory Models as a

Function of Task and Amount of Information", Organizational Behavior
and Human Performance 6, 1-27.
Farquhar, P.H. (1977) "A Survey of Multiattribute Utility Theory and Ap-
plications", TIMS Studies in the Management Sciences 6, 59-89.
Farquhar, P.H., and Pratkanis, A.R. (1993) "Decision Structuring with
Phantom Alternatives", Management Science 39, 1214-1226.
Fishburn, P.C. (1978) "A Survey of Multiattribute/Multicriterion Evalua-
tion Theories", in: S. Zionts (ed.), Multiple Criterion Problem Solving,
Springer: Berlin, 181-224.
Fischhoff, B., Slovic, P., Lichtenstein, S. (1980) "Knowing What You Want:
Measuring Labile Values", in: T. Wallsten (ed.), Cognitive Processes in
Choice and Decision Behavior, Erlbaum: Hillsdale (N.Y.), 117-14l.
Goldberger, A.S. (1964) Econometric Theory, John Wiley & Sons: New
York.
Gujarati, D.N. (1988) Basic Econometrics, 2nd ed., McGraw-Hill: Singa-
pore.
Harless, D.W., and Camerer, C.F. (1994) "The Predicitve Utility of Gen-
eralized Expected Utility Theories", Econometrica 62, 1251-1289.
Harrison, G.W. (1994) "Expected Utility Theory and the Experimentalist-
s", Empirical Economics 19, 223-253.
Hey, J.D., and Di Cagno, D. (1990) "Circles and Triangles: An Experimen-
tal Estimation of Indifference Lines in the Marschak-Machina Triangle",
Journal of Behavioral Decision Making 3, 279-306.
Hey, J.D., and Orme, C. (1994) "Investigating Generalizations of Expected
Utility Theory Using Experimental Data", Econometrica 62, 1291-1326.
Huber, J., Payne, J.W., and Puto, C. (1982) "Adding Asymmetrically Dom-
inated Alternatives: Violations of Regularity and the Similarity Hypoth-
esis" , Journal of Consumer Research 9, 90-98.
Huber, J., and Puto, C. (1983) "Market Boundaries and Product Choice:
Illustration Attraction and Substitution Effects", Journal of Consumer
Research 10, 31-44.
Johnson, E.J., and Meyer, R.J. (1984) "Compensatory Choice Models of
Noncompensatory Processes: The Effect of Varying Context" , Journal of
Consumer Research 11, 528-54l.
Kahneman, D., and Tversky, A. (1979) "Prospect Theory: An Analysis of
Decision Under Risk", Econometrica 47, 263-29l.
Keen, P.G.W. (1977) "The Evaluing Concept of Optimality", TIMS Studies
in the Management Sciences 6,31-57.
Keeney, R.L. (1974) "Multiplicative Utility Functions", Operations Re-
search 22, 22-34.
Keeney, R.L. and Raiffa, H. (1976) Decisions with Multiple Objectives:
Preferences and Value Tradeoffs, Wiley: New York.
Lichtenstein, S., and Slovic, P. (1971) "Reversals of Preferences Between
Bids and Choices in Gambling Decisions", Journal of Experimental Psy-
452
chology 89, 46-55.

Luce, R.D. (1958) "A Probabilistic Theory of Utility", Econometrica 26,
193-224.
Luce, R.D. (1959) Individual Choice Behavior: A Theoretical Analysis,
John Wiley & Sons: New York.
Luce, R.D. (1977) "The Choice Axiom After Twenty Years", Journal of
Mathematical Psychology 15,215-233.
MacCrimmon, KR. (1968a) "Descriptive and Normative Implications of
the Decision-Tree Postulates", in: K Borch and J. Mossin (eds.), Risk
and Uncertainty, Macmillan: London and New York, 3-23.
MacCrimmon, KR. (1968b) "Decision Making Among Multiple-Attitude
Alternatives: A Survey and Consolidated Approach" , Rand Corporation,
Santa Monica, RM-4823-ARPA.
Manrai, A.K (1995) "Mathematical Models of Brand Choice Behavior",
European Journal of Operations Research 82, 1-17.
Marschak, J. (1960) "Binary-Choice Constraints and Random Utility Indi-
cators" , in: KJ. Arrow et al. (eds.), Mathematical Methods in the Social
Sciences 1959, Stanford University Press: Stanford, 312-329.
Meyer, R., and Johnson, E.J. (1995) "Empirical Generalizations in the
Modeling of Consumer Choice", Marketing Science 14, G180-G189.
Mintzberg, H., Raisinghani, D., and Theoret, A. (1976) "The Structure of
Unstructured Decision Processes", Administrative Science Quarterly 21,
246-275.
Montgomery, H. (1983) "Decision Rules and the Search for a Domi-
nance Structure: Towards a Process Model of Decision Making", in: P.
Humphreys, O. Svenson and A. Vari (eds.), Analyzing and Aiding Deci-
sion Processes, North-Holland: Amsterdam, 343-369.
Montgomery, H., and Svenson, O. (1976) "On Decision Rules and Infor-
mation Processing Strategies for Choices among Multiattribute Alterna-
tives" , Scandinavian Journal of Psychology 17, 283-29l.
Mood, A., Graybill, F.A., and Boes, D.C. (1974) Introduction to the Theory
of Statistics, 3rd. ed., McGraw-Hill: Singapore.
Nisbett, R.E., and Wilson T.D. (1977) "Telling More Than We Can Know:
Verbal Reports on Mental Processes" , Psychological Review 84, 231-259.
Payne, J.W. (1976) "Task Complexity and Contingent Processing in Deci-
sion Making: An Information Search and Protocol Analysis", Organiza-
tional Behavior and Human Performance 16, 366-387.
Payne, J.W. (1982) "Contingent Decision Behavior", Psychological Bulletin
92,382-402.
Payne, J.W., and Braunstein, M.L. (1971) "Preferences Among Gambles
with Equal Underlying Distributions", Journal of Experimental Psychol-
ogy 87, 13-18.
Ratneshwar, S., Shocker, A.D., and Steward, D.W. (1987) "Toward Un-
derstanding the Attraction Effect: The Implications of Product Stimulus
Meaningfulness and Familiarity" , Journal of Consumer Research 13, 520-
453
533.
Restle, F. (1961) Psychology of Judgment and Choice, John Wiley & Sons:
New York.
Russo, J.E. (1977) "The Value of Unit Price Information", Journal of Mar-
keting Research 14, 193-20l.
Russo, J.E., and Dosher, B.A. (1983) "Strategies for Multiattribute Binary
Choice", Journal of Experimental Psychology: Learning, Memory, and
Cognition 9, 676-696.
Seidl, C., and Traub, S. (1998) "A New Test of Image Theory", Organiza-
tional Behavior and Human Decision Processes, (forthcoming).
Shepard, R.N. (1964a) "Attention and the Metric Structure of the Stimulus
Space", Journal of Mathematical Psychology 1,54-87.
Shepard, R.N. (1964b) "On the Subjectively Optimum Selection Among
Multiattribute Alternatives", in: M.W. Shelly and G.L. Bryan (eds.),
Human Judgments and Optimality, John Wiley & Sons: New York.
Shugan, S.M. (1980) "The Cost of Thinking", Journal of Consumer Re-
search 7, 99-111.
Simon, H.A. (1955) "A Behavioral Model of Rational Choice", The Quar-
terly Journal of Economics 69, 99-118.
Slovic, P. (1975) "Choice Between Equally Valued Alternatives", Journal of
Experimental Psychology: Human Perception and Performance 1, 280-
287.
Slovic, P., and Lichtenstein, S. (1968) "The Importance of Variance Pref-
erences in Gambling Decisions" , Journal of Experimental Psychology 78,
646-654.
Slovic, P., and MacPhillamy, D. (1974) "Dimensional Commensurability
and Cue Utilization in Comparative Judgment" , Organizational Behavior
and Human Performance 11,172-194.
Smith, E.R., and Miller, F.S. (1978) "Limits on Perception of Cognitive
Processes: A Reply to Nisbet & Wilson" , Psychological Review 85, 355-
362.
Smith, V.L. (1982) "Microeconomic Systems as an Experimental Science",
The American Economic Review 72, 923-955.
Smith, V.L., and Walker, J.M. (1993) "Monetary Rewards and Decision
Cost in Experimental Economics", Economic Inquiry 31, 245-26l.
Starmer, C., and Sugden, R. (1989) "Probability and Juxtaposition Effects:
An Experimental Investigation of the Common Ratio Effext", Journal of
Risk and Uncertainty 2, 159-178.
Svenson, O. (1979) "Process Descriptions of Decision Making", Organiza-
tional Behavior and Human Performance 23, 86-112.
Tversky, A. (1969) "Intransitivity of Preferences", Psychological Review
76,31-48.
Tversky, A. (1972a) "Elimination by Aspects: A Theory of Choice", Psy-
chological Review 79, 281-299.
Tversky, A. (1972b) "Choice by Elimination", Journal of Mathematical
454
Psychology 9,341-367.
Tversky, A., and Sattah, S. (1979) "Preference Trees", Psychological Re-
view 86,542-573.
Tversky, A., Sattah, S., and Slovic, P. (1988) "Contingent Weighting in
Judgment and Choice", Psychological Review 95, 371-384.
Tyszka, T. (1983) "Contextual Multiattribute Decision Rules", in: L.
Sj5berg, T. Tyszka, and J.A. Wise (eds.), Human Decision Making, Bok-
f5rlaget Doxa: Lund, 243-256.
Von Nitzsch, R., and Weber, M. (1993) "The Effect of Attribute Ranges on
Weights in Multiattribute Utility Measurements", Management Science
39, 937-943.
Von Winterfeldt, D., and Edwards, W. (1982) "Costs and Payoffs in Per-
ceptual Research", Psychological Bulletin 91, 609-:-622.
Von Winterfeldt, D., and Edwards, W. (1986) Decision Analysis and Be-
havioral Research, Cambridge University Press: Cambridge.
Weber, M., and Borcherding, K. (1993) "Behavioral Influences on Weight
Judgments in Multiattribute Decision Making" , European Journal of Op-
erational Research 67, 1-12.
Wedell, D.H. (1991) "Distinguishing Among Models of Contextually In-
duced Preference Reversals" , Journal of Experimental Psychology: Learn-
ing, Memory, and Cognition 17, 767-778.
White, P. (1980) "Limitations on Verbal Reports of Internal Events: A
Refutation of Nisbett & Wilson and of Bern", Psychological Review 87,
105-112.
Wright, P., and Barbour, F. (1977) "Phased Decision Strategies: Sequels
to an Initial Screening", TIMS Studies in the Management Sciences 6,
91-109.
Yntema, D.B., and Torgerson, W.S. (1961) "Man-Computer Cooperation
in Decisions Requiring Common Sense" , IRE Transactions of the Profes-
sional Group on Human Factors in Electronics, HFE 2,20-26.
Collusion and distribution of profits under
differential information*
Konstantinos Serfes 1 and Nicholas C. Yannelis2
1 Department of Economics, SUNY at Stony Brook, Stony Brook, NY
11794-4384, U.S.A
2 Department of Economics, University of Illinois at Urbana-Champaign, 330
Commerce West Building, 1206 South Sixth Street,Champaign, 1L 61820, U.S.A
Abstract. We examine a Cournot game with differential private informa-

tion. We study collusion under different information rules, i.e., when firms
pool their private information, use their common knowledge information,
or decide not to share their private information at all. We put the industry
profits under the three different information schemes in a hierarchy. In addi-
tion, we look at the incentive compatibility problem and we show that only
collusion under common knowledge information is incentive compatible.
Finally, we deal with the issue of how the industry profits are distributed
among the firms, in a way that asymmetries are captured. We propose the
Shapley value as a proper way to distribute the industry profits among the
firms. We also point out that the a-core associated with the Cournot game
with differential information is non-empty.
JEL Classification Number: cn, C72, D82

Keywords: Collusion, differential information, Shapley value, alpha-core
1 Introduction
We study the collusion of firms with differential information. A game with
differential information consists of a finite number of firms, where each firm
is characterized by its strategy set, its payoff function, its private informa-
tion (which is a partition of an exogeneously given probability measure
space) and a prior. When firms collude, they choose an output level that
maximizes joint expected profits. The information firms can use in the col-
lusive agreement varies. Firms may pool their information, may use their
private information, or they may choose to use their common knowledge
information. Each type of information sharing yields different profits and
most importantly creates different incentives to the individual firms for
'The paper benefit.ed from discussions wit.h .Jingang Zhao.

456
misreporting their true information.

The main focus of our paper is to address the following questions: i)
How should firms share their private information in collusive agreement
such that industry profits are the highest? ii) How are the collusive prof-
its under different types of information sharing compared to profits from
non-cooperative production? iii) How should colluding firms share their pri-
vate information in a coalitional incentive compatible way? iv) How should
industry profits be distributed among firms in a way which captures the
contribution or the "worth" of each firm to total profits?
Other work on the subject [e.g., Donsimoni et al. (1986) and Crampton
and Palfrey (1990)] follow an approach similar to ours by assuming that
firms abide by the cartel agreement. The problem of explicit collusion in an
industry with heterogeneous firms and private information was first con-
sidered formally by Roberts (1983). He derives properties of the incentive
compatibility constraints associated with a revelation game. He found that
without side payments, if firms are sufficiently similar, then monopoly col-
lusion cannot be achieved, but if side payments are allowed such collusion is
possible with a dominant strategy mechanism essentially equivalent to the
Vickrey (second-price) auction. Rotemberg and Saloner (1990) investigate
a price leadership scheme in a differentiated products duopoly in which the
firms are asymmetrically informed. Crampton and Palfrey (1990) study the
issue of cartel enforcement when the cost of each firm is private informa-
tion. An enforceable cartel is one which is feasible, incentive compatible and
individually rational. They show that if defection results in either Cournot
or Bertrand competition, the incentive problem in large cartels is severe
enough to prevent the cartel from achieving the monopoly outcome. Laf-
font and Martimort (1997) study collusion of agents whose objectives are
not aligned with that of their organization under asymmetric information.
Our model is different from the above ones. In particular, we have a gen-
eral model and we address the issue of collusion in a differential information
game for the first time. We show that collusion under the pooled informa-
tion yields the highest industry profits. However, this type of information
sharing is not coalitional incentive compatible. 1 We present examples with
two firms where one firm can distinguish between two states of nature and
the other cannot and the firm with the "superior" information finds it prof-
itable to misreport the true state of nature to the other firm. Only collusion
under the common knowledge information is coalitional incentive compati-
ble. It is important to emphasize here that we look at the coalitional incen-
tive compatibility and not at the individual incentive compatibility as, for
example, Crampton and Palfrey (1990). An individual incentive compatible
1 A collusive agreement is coalitional incent.ive compatible when there does not exist
a coalition of firms that can misreport the true state of nature and benefit its members.
For a precise definit.ion see definit.ion 7.1.
457
outcome may not be coalitional incentive compatible which in turn means

that coalitions of firms, rather than individual firms, may have an incentive
not to report truthfully the realized state of nature. We also propose that
a sensible rule for allocating production and distributing the profits among
the firms is according to the Shapley value of each firm. The Shapley value
rule yields individually rational and Pareto optimal outcomes, captures the
informational asymmetries between the firms as well as the contribution of
each firm to the total profits. We also point out that the a-core of the
differential information game is non-empty. We provide several examples
that illustrate and clarify our results.
The rest of the paper is organized as follows. In Section 2, we provide the
notation and definitions. In Section 3, we present the model and Section 4
contains existence results. In Section 5, we outline three information rules
and in Section 6 we rank the industry profits under the different information
rules. Section 7 addresses the incentive compatibility issue. In Section 8,
the issue of profit distribution is addressed. Finally, Section 9 contains two
illustrative examples.
2 Notation and definitions

2.1 Notation
JR.l denotes the I-fold Cartesian product of the set of real numbers.
JR.~ denotes the positive cone of JR.l •
JR.~+ denotes the strictly positive elements.
2A denotes the set of all non-empty subsets of the set A.
o denotes the empty set.
\ denotes set theoretic subtraction.
2.2 Definitions
If X and Y are sets, the graph of the set-valued function (or correspon-
dence), <jJ : X - t 2Y is denoted by
G", = {(x,y) E X x Y: yE <jJ(x)}.
Let (0" F, JL) be a complete, finite measure space, and X be a separable

Banach space. The set-valued function <jJ : 0, - t 2x is said to have a mea-
surable graph if G", ® (3(X), where (3(X) denotes the Borel (J-algebra on X
and ® denotes the product (J-algebra. The set-valued function <jJ : 0, - t 2x
is said to be lower measurable or just measurable if for every open subset
V of X, the set
{w En: <jJ(w) n V 1= 0}
is an element of :F. It is well known that if <jJ : 0, - t 2X has a measurable
graph, then <jJ is lower measurable. Furthermore, if <jJ(. ) is closed valued and
lower measurable then <jJ : 0, - t 2x has a measurable graph. A theorem
458
of Aumann tells us that if en, F, J.L) is a complete finite measure space,

X is a separable metric space and cl> : n -4 2x is a non-empty valued
correspondence having a measurable graph, then cI>{. ) admits a measurable
selection, i.e., there exists a measurable function f : n -4 X such that
few) E cI>(W) , J.L - a.e.
Let (n,F,J.L) be a finite measure space and X be a Banach space. Fol-
lowing Diestel-Uhl (1977), the function f : n -4 X is called simple if there
exist Xl> X2, ... , Xn in X and aI, a2, ... , an in F such that 2:~=1 XiX ai where
Xa{w)
" = 1 if w E ai and Xa(w) , = 0 if w tJ- ai. A function f : n -4 X
is said to be J.L-measurable if there exists a sequence of simple function
fn : n -4 X such that limn-->oo Ilfn(w) - f{w)/I = 0 for almost all wEn. A
J.L-measurable function f : n -4 X is said to be Bochner integrable if there
exists a sequence of simple functions {In : n = 1,2, ... } such that
lim r
n-->oo ln
/lfn(w) - f(w)/ldJ.L(w) = O.
In this case we define for each E E F the integral to be
lEr f(w)dJ.L(w) = n-->oo

lim r fn{w)dJ.L{w).
lE
It can be shown [see Diestel-Uhl (1977), Theorem 2, p.45] that if f : n -4 X
is a J.L- measurable function then f is Bochner integrable if and only if
In /If(w)/ldJ.L(w) < 00.
For 1 ~ p < 00, we denote by Lp(J.L, X) the space of equivalence classes
of X-valued Bochner integrable functions x : n -4 X normed by
It is a standard result that normed by the functional /I. /lp above, Lp(J.L, X)
becomes a Banach space [see Diestel-Uhl (1977), p.50j.
3 The Cournot game with differential information

We assume that there are n firms, {i = 1, ... , n}, that produce an output
q = {ql, ... , qn}. The subscript -i will be used to denote all firms other
than firm i. Let (n, F, J.L) be a complete probability measure space. We
interpret n as the states of nature of the world and assume that it is large
enough to include all events that we consider to be interesting. As usual
F denotes the a-algebra of events and J.L is a common probability measure.
Let Y be a separable Banach space denoting the production space.
Definition 3.1: A Cournot game with differential information is a set
C = {( Qi, 7ri,.1i, J.L) : i = 1, ... ,n}, where
i) Qi : n -4 2Y is the random production set of firm i;
459
ii) 1ri: Q(W) --. IR is the random profit function2 of firm i, (where Q(w) =
Ql(W) x ... X Qn(w));
ill) Fi is a sub iT-algebra of F, which denotes the private information of
firm i;
iv) J.L is a probability measure on n denoting the common prior.
Let LQi denote the set of all Bochner integrable and Fi-measurable se-
lections from the production set Qi of firm i, i.e.,
LQi = {qi E Ll (J.L, Y) : qi : n --. Y is Fi - measurable and

qi(W) E Qi(W) and J.L - a.e.}.
Let LQ = LQl X ... x LQn. Given a Cournot game, a production plan for
firm i is an' element qi E LQ;.
The ex-ante expected profit function 3 of firm i, ITi : LQ -+ IR is defined
as 4
ITi(qi, q-i) = 1 wEn
1ri(qi(W), q-i(w))dJ.L(W).
A Cournot-Nash equilibrium for C = {(Qi, 1ri, Fi,J.L) : i = 1, ... ,n} is an

element q* E LQ such that for all i,
We can now state the assumptions needed to prove the existence of a

Cournot-Nash equilibrium.
(A.l)
2 If pew) : Q(w) ..... R is the inverse demand function and C i : Qi(W) ..... R is the cost
function of firm i, then 1T;(q(W) = p(q(W»qi(W) - Ci(q;(W». We could have allowed
the payoff function 1T to depend also on the state of nature w. The results of the paper
remainualid.
3The entire analysis would go through if instead of the ex-ante profit function we
used the interim one. That is, the conditional (interim) expected profit function of firm
i Ili(',' ) : LQi x Q-i(W) ..... R is defined as
Il; (qi, q-i) = 1

w'EE;(w)
1Ti(W', qi, q-i (W'»ki(W'IEi(W»dJ.l(w'),
where
ki(w'IEi(W» = {o qi(W')
---
f';'EEi(W) q;(w)dJ.«w)
ifw ' E E;(w)
J
is the prior of agent i, (where ki is a Hadon-Nikodym derivative such that ki(W)dp(w) =
1 and Ei(W) denot.es the event in firm i ' s partition which contains the realized state of
nature).
4For simplicity we assume that the profit function does not. depend on n. As we
mentioned above all t.he results of the paper remain valid.
460
Qi : n ~ 2Y , is a non-empty, convex, weakly compact-valued and

integrably bounded correspondence having an Fi-measurable graph, i.e.,
GQi E Fi ® B(Y).
(A.2)
i) For each i, 7riC" ) : Q(w) ~ JR, is weakly continuous.

ii) The function 7ri is concave in the i-th coordinate for all i.
iii) 7ri is integrably bounded.
4 Existence of a Cournot-Nash equilibrium

We can now state the first existence result. We assume that there exists a
finite or countable partition Ai, (i = 1, ... , n) of n, and the a-algebra Fi is
generated by Ai.
Theorem 4.1: Let C = {( Qi, 7ri, Fi, ft) : i = 1, ... ,n} be a Cournot game
satisfying (A.1) - (A. 2). Then, there exists a Cournot-N ash equilibrium.
Proof: For each i, define the correspondence 'Pi : LQ_i ~ 2LQi by
Also define the correspondence F : LQ ~ 2LQ by
As in Yannelis and Rustichini (1991), we will show that the correspondence

F satisfies all the hypotheses of the Fan-Glicksberg Fixed Point Theorem.
It can then be easily checked that a fixed point of the correspondence F
is by construction a Cournot-Nash equilibrium for C. We will complete the
proof in three steps.
I. LQ is non-empty, convex, weakly compact and metrizable.
The non-emptiness of LQ follows from the Aumann measurable selection
theorem. Also, since each Qi is non-empty, convex and weakly compact, it
follows from Diestel's Theorem that each LQi is a weakly compact subset
of Ll (ft, Y). Obviously, each LQi is convex. Furthermore, since each LQi is
a weakly compact subset of a separable Banach space L 1 (ft, Y), it is also
metrizable [for more details see Yannelis and Rustichini (1991), Theorem
5.1].
Il. The function TIi is weakly continuous for each i.
Since, by assumption, 7ri is concave, weakly continuous and 7r i is inte-
grably bounded, the result follows by an application of Theorem 2.8 in
Balder and Yannelis (1993).
Ill. Each correspondence 'Pi : LQ_i ~ 2 LQi , is non-empty, convex valued
and weakly U.S.c.
461
Since II(q) is a concave function of qi on L Qi , it follows that 'Pi is convex

valued. By virtue of Berge's Maximum Theorem, it follows that 'Pi is weakly
u.s.c. Finally, an appeal to Weierstrass' Theorem it is guaranteed that 'Pi
is also a non-empty valued correspondence.
Now since each 'Pi is non-empty, closed, convex valued and weakly u.s.c.,
it follows that likewise is F : LQ ~ 2LQ • Thus, the correspondence F
satisfies all the conditions of the Fan-Glicksberg Fixed Point Theorem.
Consequently, there exists some q* E LQ such that q* E F(q*). D
5 Collusion under different information rules

It is a well known result that a Coumot-Nash equilibrium may not be
Pareto optimal. In other words, there is a surplus that has not been ex-
tracted by the firms. If the firms collude and play a cooperative game,
then a Pareto optimal outcome will be reached. This problem has been
examined when firms have symmetric information. However, in the pres-
ence of differential information there may be different ways for the firms
to collude, depending on how they want to share their private information.
Before we proceed, let's define the three different information rules that we
will consider in the sequel.
Definition 5.1: A Pooled information rule is the one where firms share
their information, i.e., the information they use is, Fj = Vi=l Fi, j =
1, ... , n, where V denotes the join. 5
Definition 5.2: A Private information rule is the one where firms use
their own private information, i.e., Fi, i = 1, ... , n.
Definition 5.3: A Common knowledge information rule is the one where
firms use only the information that is common to them, i.e., Fj =
l\i=l Fi, j = 1, ... , n, where 1\ denotes the meet. 6
Let L~i denote the set of all Bochner integrable and Vi=l Fi- measurable
selections from the production set Qi of firm i, i.e.,
L~i {qi E L1 (IL, Y) : qi : n ~ Y is Vi=l Fi - measurable and

qi(W) E Qi(W), IL - a.e.}.
Also let L~ = L~I X ... x L~n'

Let LQi denote the set of all Bochner integrable and l\i=l Fi - measurable
selections from the production set Qi of firm i, i.e.,
{qi E L1 (IL, Y) : qi : n ~ Y is l\i=l Fi - measurable and

E Qi(W),IL - a.e.}.
5That is the smallest IT-algebra containing all of the sub IT-algebras Fi, i = 1, ... , n.
6That is the largest IT-algebra contained in all of the sub IT-algebras Fi, i = 1, ... , n.
462
Also let LQ = LQ\ x··· x LQn.

A collusion equilibrium under the pooled information rule for C
{( Qi, 7ri, F i , JL) : 1, ... ,n} is an element q* E L~ such that
n
IJP(q*) = max I)Ii(q).
qEL~ i=l
A collusion equilibrium under the private information rule for C

{( Qi, 7ri, Fi, JL) : i = 1, ... ,n} is an element q* E LQ such that
A collusion equilibrium under the common knowledge information rule

for C = {(Qi, 7ri, Fi, JL) : i = 1, ... , n} is an element q* E LQ such that
Next we present the second existence result of this paper.

Theorem 5.1: Under assumptions (A.l}-(A.2), a collusion equilibrium
exists for all the information rules.
Proof: Notice that the objective function is weakly continuous and L~, LQ
and LQ are non-empty and weakly compact. Therefore the maximum is
attained and the argmax is the set of all equilibrium points. 0
6 Comparison of profits under the three information

rules
In this section, we will put the industry profits under the three different
information rules in a hierarchy. It is known that under symmetric informa-
tion the industry profits when firms collude are greater than or equal to the
industry profits derived from the Cournot-Nash game. But what happens
under differential information?
Let IJP(q*) , fi(q*) and IIC(q*) be the value functions under the three
information rules, as defined in the previous section.
Proposition 6.1: IIP(q*) 2 fi(q*) 2 IIC(q*).
Proof: First observe that 1\":=1 h S; Fi, i = 1, ... , n, S; V":=l Fi. This implies
that LQ S; LQ S; L~. Since the objective functions are the same, the desired
result follows. 0
Let IIN(q*) denote the industry profits derived from the Nash game.
Proposition 6.2: IIP(q*) 2 fi(q*) 2 fIN(q*).
463
Proof: Obvious. 0
However, as the following proposition indicates, we cannot compare the
industry profits derived under the common knowledge rule with those de-
rived from the Cournot-Nash game. The reason is that when firms collude
using the common information, essentially they throwaway some valuable
information, which means lower profit. On the other hand, the joint maxi-
mization alone, gives them higher profits. In this general setting we cannot
tell which effect outweighs the other.
Proposition 6.3: The industry profits derived from the collusion under
the common knowledge information rule and the ones derived from the
Cournot-Nash game are not comparable.
Proof: Consider a Cournot game with two firms {1,2}, three states of
nature, i.e., n = {a, b, c} and one homogeneous output q. Each state occurs
with the same probability. Each firm's private information is given by the
following partition of the state space,
F1 = {a,b,c},F2 = {{a}, {b}, {c}}.

The inverse demand function is, p(w) = 5 - 1.5(q1(W) + q2(W)). The cost
function which is measurable with respect to each firm's private information
is: For firm 1, C 1(W,q1(W)) = .4qr for all wEn and for firm 2,
q~ ifw=a
C2(w, Q2(W)) = { .4Q~ if w = b
.8Q~ if w = c.
Notice that firm 1 has trivial information, while firm 2 has complete infor-
mation. The following production plan is a Cournot-Nash equilibrium,
Q1(W) = 1.00275, for all wE nj

.699 if w = a
Q2(W) = { .919 ~f w = b
.759 1fw=c.
Observe that the production plan is also measurable with respect to each
firm's private information. The expected industry profits from the Cournot-
Nash game are 3.296.
Now assume that firms collude using the common knowledge information
rule. Since the information that is common to both of them is the trivial
information, the production plan must be Gonstant in all states. This is,
Q1(W) = .9194, Q2(W) = .502, for all wEn.

464
The expected industry profits are 3.5537.

Thus, in this example the profits from collusion with common knowledge
information are higher than the profits from the Cournot-Nash game. Next
we present an example where the profits from the Cournot - Nash game are
higher than the profits from collusion with common knowledge information.
Consider now a Cournot game with two firms {I, 2} that produce a ho-
mogeneous output q. Uncertainty is generated by the marginal cost func-
tions Ci : n ~ lR+, i = 1, 2. Firm 1 has trivial information and its cost
function is: Cl = Cl qr,
where Cl = 5 for all wEn. Firm 2 has complete
information and its cost function is: C2 = c2q~. We assume that C2 is dis-
tributed uniformly on [0, 10]. The above information about the Cournot
game is common knowledge. Moreover, the inverse demand function is
p(w) = 50 - 1.5(q1(W) + q2(W)), Since firm 1 has trivial information, its
production plan will be constant across all states, while firm's 2 produc-
tion plan will be contingent on each realization of the random variable C2.
Therefore, the Cournot-Nash equilibrium is
22.5122
q1(W) = 3.317, for all wEn and q2(W) = ( ).
1.5 + C2 w
The expected industry profits from the Cournot-Nash game are 174.747.
Now assume that firms collude using the common knowledge information
rule. This now implies that production must be constant across all states.
The production plan is
Q1(W) = Q2(W) = 3.125, for all wE O.
The expected industry profits now are 156.25. 0
7 Coalitional incentive compatibility

One of the basic questions that one may ask is whether the different equi-
librium notions we defined previously are coalitional incentive compatible.
That is, whether a coalition of firms has an incentive to misreport the
true state of nature and benefit its members. This is an important ques-
tion especially for the collusion equilibrium. If a collusion equilibrium is
not coalitional incentive compatible, then it is not sustainable. We define
rigorously below the notion of coalitional incentive compatibility which is
related to the one in Krasa-Yannelis (1994).
Definition 7.1: An output function q E LQ is said to be coalitional
incentive compatible if and only if the following does not hold: There exist
a coalition of firms 7 S c I and two states a, b that members of 1\ S cannot
distinguish (i.e., a and b are in the same partition for the firms in 1\ S)
7I is the set. of all firms.

465
and such that members of S are better off by announcing b whenever a

has actually occurred. Formally, q E LQ is said to be coalitional incentive
compatible for C if it is not true that there exist coalition S, and states a, b
with8 a E niftsEi(b), such that 9 7ri(qS(b),qI\S(b)) > 7ri(qS(a),qI\S(b», for
all i E S, that is, each firm in coalition S is strictly better off announcing
that state b occurred rather than the true state a and firms not in S are
unable to distinguish between state a and b.
It turns out that a Cournot-Nash equilibrium is incentive compatible.
Also a collusion equilibrium under the common knowledge information rule
is coalitional incentive compatible. However, a collusion equilibrium under
the pooled information rule and under the private information rule may
not be coalitional incentive compatible.
Proposition 7.1: A Cournot-Nash equilibrium for C = {(Qi, 7ri, .ri, J.l) :
i = 1, ... ,n} is incentive compatible.
Proof: Since we are dealing with a non-cooperative concept it is appropri-
ate to reduce the coalition S to the singleton coalition, i.e., S == {i}. Then,
-i denotes all the firms but i. Suppose that q* E LQ is a Cournot-Nash
equilibrium and there exist a, b, where a E E_i(b), such that
First, since q~i is .r_i-measurable, it is implied that q~i(a) = q~i(b). Thus,

for all WE Ei(a) n E_i(a) and t E Ei(b) n E_i(a),
7ri(q;(t) , q~i(t» > 7ri(q;(W) , q~i(W)).
Now consider the following production plan for firm i,
if wE Ei(a) n E_i(a)
otherwise.
It follows that
This contradicts the fact that q* is a Nash equilibrium. 0

Proposition 7.2: A collusion equilibrium under the private information
rule for C =. { (Q i, 7ri , .ri, J.l) : i = 1, ... , n} may not be coalitional incentive
compatible.
8 Ei(b), is the event in firms' information partition that contains the realized state b.
9 qS and qI\S are vectors of outputs for firms in coalition S and I \ S respectively.
466
Proof: Consider two firms {1,2} that produce a homogeneous output.

The state space is n = {a, b, c} where each state occurs with probabil-
i
ity and the private information of each firm is: :Fi = {{a, b}, { c}} and
F2 = {{a, cl, {b}}. We denote by ql(W), q2(W) the production in state W
of firm 1 and firm 2 respectively. The inverse demand that firms face is
p = (5 - 1.5(ql(W) + q2(W))). The marginal cost function of each firm,
which is measurable with respect to each firm's private information is
if W = a,b
ifw = c
.25 if w = a, c
C2(W) ={
1 if w = b.
A collusion equilibrium under the private information rule is
qi(a) = qi(b) = .916,qi(c) = .416,
q2(a) = q2(c) = .916,Q2(b) = .416,

and the profit of each firm in each state is
7r1(a) = 1.833,7rI(b) = 2.52, 7r1(C) = .832,

7r2(a) = 1.833,7r2(b) = .832, 7r2(C) = 2.52.
The ex-ante expected profit for the industry is III + Il2 = 1.725 + 1. 725 =
3.45.
Suppose that state b occurs. Firm 2's profit is .832. However, if firm 2
reports that state a occurred and produce as if a had actually occurred, its
profit is 1.146. Since 1.146 > .832 the collusion equilibrium with differential
information is not incentive compatible. 0
Remark: It follows from the above proposition that a collusion equilibrium
under the pooled information rule may not be incentive compatible as well.
The reason is that although information is now symmetric, still firms cannot
distinguish between the states that could not distinguish before the pooling
took place. Hence, the above proposition is applicable here as well.
Proposition 7.3: A collusion equilibrium under the common knowledge
information rule is incentive compatible.
Proof: It is rather obvious, since now there do not exist states a and b
such that one firm can distinguish between the two and the others cannot.
o
467
8 Distribution of profits
So far, we showed that when firms collude (under the pooled and the private
information rules), industry profits are higher than the profits obtained
by the Cournot-Nash game. Moreover, profits may be higher when firms
collude under the common knowledge information rule. The question that
remains to be answered is how are these extra profits being distributed
among the firms in a way that captures the contribution of each firm to
the total profits.
8.1 The private value production plan

We propose that each firm should be rewarded according to its contribution
to the total profits. One way to do this is to reward each firm according
to its Shapley value. Then a production plan will be determined, so as
each firm will get its Shapley value. This production plan will be the value
production plan.
As in the definition of the standard value allocation concept, we must
first derive a transferable profit game (TP) in which each firm's profits are
weighted by a factor Ai, (i = 1, ... , n), which allows for profits comparisons.
In the value allocation itself no side payments are necessary.ID A game with
side payments is then defined as follows:
Definition 8.1.1: A game with side payments r = (I, V) consists of a

finite set of agents (firms) I = {I, ... ,n} and a superadditive, real valued
function V defined on 21 such that V(0) = O. Each S C I is called a
coalition and V(S) is the "worth" of the coalition S.
The Shapley value of the game r (Shapley 1953) is a rule that assigns
to each firm i a "payoff", Shi, given by the formula,ll
Shi(V) = ~ (ISI-1)1!;:[I-ISI)! [V(S) - V(S \ {i} )].

S:J {i)
The Shapley value has the property that L:iEI Shi(V) = V(I), i.e., the
Shapley value is Pareto efficient. Moreover, it is individually rational, i.e.,
Shi 2 V({i}),Vi.
We now define for each Cournot game with differential information, C,
and for each set of weights, {Ai : i = 1, ... , n}, the associated game with
side payments (I, VI) (we also refer to this as a "transferable profits" (TP)
game) as follows:
IOSee Emmons and Scafuri (1985, p.60) for further discussion.

11 The
Sltapley value measure is the sum of the expected marginal contributions a firm
can make to all the coalitions of which it is a member.
468
For each coalition S C I, let
We are now ready to define the private value production plan.

Definition 8.1.2: An output plan q E LQ is said to be a private value
production plan of the Cournot game with differential information, C, if
there exist .Ai ~ 0 (i = 1, ... , n, which are not all equal to zero), with
where Shi(VI) is the Shapley value of firm i derived from the game (I, VI),
defined in (8.1).
The above definition says that the expected profits of each firm multiplied
by its weight .Ai must be equal to its Shapley value derived from the (TP)
game (1, VI).
An immediate consequence of Definition 8.1.2 is that the private value
production plan is individually rational (profits for firm i are greater than
of equal from the ones derived from the Cournot-Nash game). This follows
immediately from the fact that the game (I, VI) is superadditive for all
weights. In addition, it is Pareto efficient. 12
We are now ready to state the first existence result of this section.
TheoreIn 8.1.1: LetC = {(Qi,1l"i,Fi ,/L): i = 1, ... ,n} be a Cournot game
as defined in Section 3, satisfying assumptions (A.l}-(A.2).
Then, a private value production plan exists in C.
Proof: This result can be proved along the lines of Krasa and Yannelis
(1996), Theorem 1. 0
ReInark: One can easily show that a pooled information (where the in-
formation that is being used is the pooled information) value production
plan 13 exists as well.
8.2 The COInInon knowledge value production plan

We now introduce another notion of a value production plan for the
Cournot game with differential information. The difference stems from the
measurability restriction on the type of production plans. It is an ana-
log of the coarse core of Yannelis (1991). We call it a common knowledge
value production plan, since the information that is being used is the com-
mon knowledge information. As we saw in the previous section a common
12 For more details see Krasa and Yannelis (1996), p.169.

13 Since it is not incentive compatible we will not examine it thoroughly.
469
knowledge production plan is coalitional incentive compatible. Therefore,

it is of great importance if we know that there is also a way to distribute
the surplus (if there is any) in a manner that the contribution of each firm
is rewarded.
We now define for each Cournot game with differential information, C,
and for each set of weights, {Ai: i = 1, ... ,n}, the associated game with
side payments (1, Vf) (we also refer to this as a "transferrable profits" (TP)
game) as follows:
For each coalition S cl, let
subject to
i) for each i, qi is Ai=l Frmeasurable.
The common knowledge value production plan can now be defined as in
definition (8.1.2), except that we replace (8.1) by (8.2) and also replace VP
by Vc.
Thus, in contrast to the private value production plan, we now require the
production plan within a coalition to be based on the common knowledge
information. Notice that the common knowledge value production plan is
coalitional incentive compatible. However, we cannot prove a general exis-
tence theorem. In fact, if, for example, one firm has "trivial" information
and the other has "full" information, the common knowledge information
implies that the trivial information must be used and therefore the super-
additivity condition of the function Vf(· ) may be violated, i.e., there can
exist coalitions S, T with S n T = 0 and Vf(S) + Vf(T) > Vf(S U T).14 In
the proof of proposition 6.3, we present an example with two firms where
the Cournot-Nash equilibrium yields higher profits than collusion under
common knowledge information, which destroys the superadditivity condi-
tion. This causes problems with the existence of a common knowledge value
production plan. Therefore, we cannot prove a general existence theorem
of a common knowledge value production plan.
9 Examples
Below we give examples with two firms, with differential information, that
collude using the common knowledge information rule and the distribution
of profits is determined by a common knowledge value production plan.
These examples illustrate how the common knowledge production plan is
determined and also show that firms with superior information, while keep-
ing the other characteristics of the firms (Le., marginal cost) fixed, have
higher Shapley value and higher share of the industry profits. The profits
14See Krasa and Yannelis (1996), p.177, for more details.

470
from collusion are higher than the Cournot-Nash profits and the Shapley
value of each firm captures its contribution to the total industry profits. It
is important to note here that in these examples the value allocation is in
the core and therefore the cartel can be viewed as stable, in the sense that
no coalition of firms can deviate from the cartel agreement and become
strictly better off.
Example 9.1
Consider two firms {I, 2} that produce a homogeneous product. The
state space is n = {a, b} where each state occurs with probability and !
the private information of each firm is: :F1 = {{ a}, {b} } and :F2 = {{a, b} }.
We denote by q1 (w), q2 (w) the production in state w of firm 1 and firm 2
respectively. The inverse demand that firms face is: p = (5 - ,B(W)(q1(W) +
q2 (w) ) ). The marginal cost is zero for both firms. The slope ,B takes on the
following values:
.8 ifw =a
,B(w) ={
1.2 if w = b
A Cournot-Nash equilibrium is
q~(a) = 2.2916, q~(b) = 1.25,

q2(a) = q2(b) = 1.666.
Notice that the production is measurable with respect to each firm's private
information. The ex-ante expected profit for the industry is III + Il2 =
3.03819 + 2.77778 = 5.81597.
Now suppose that the two firms collude under the common knowledge
information rule. The information they use now is the trivial information
and the optimum total production is 2.5. The expected industry profits are:
6.25. The problem that arises is how this surplus will be distributed among
the two firms. Or put it in different words, what is the production that will
be assigned to each firm? Without taking the information superiority of the
first firm into account, both firms are identical. Hence, one solution would
be just to split the profits. However, this is not a "fair solution" since firm
1 contributes more to the coalition than firm 2 does. The value production
plan allocation we discussed above provides a more sensible outcome.
The Shapley value of the two firms (with Al = A2 = 1) is
1 1
Sh 1 = "2[6.25 - 2.77778] + "2[3.03819] = 3.25521
1 1
Sh 2 = "2[6.25 - 3.03819] + "2[2.77778] = 2.99479.
Then, a value production will be a solution to the following problem:
(5 - (q1 + q2))q1 = 3.25521

471
(5 - (ql + q2))q2 = 2.99479.

Hence, a value production plan is ql = 1.30208 and q2 = 1.19792. To
conclude, the firm with the superior information gets rewarded in the value
production plan, by being assigned a higher level of production and thus
higher profits (Ill = 3.25521 and Il2 = 2.99479).
In the next example, the asymmetry of information comes from the cost
side.
Example 9.2
Consider two firms {I, 2} that produce a homogeneous product. The
state space is n = {a, b} where each state occurs with probability ~ and
the private information of each firm is: F2 = {{a}, {b} } and FI = {{a, b} }.
We denote by ql (w), q2 (w) the production in state w of firm 1 and firm 2,
respectively. The inverse demand that firms face is p = (5 - 1.5(ql(W) +
q2(W))). The marginal cost of each firm, which is measurable with respect
to each firm's private information, is
.8 if w = a
C2(W) = { ,
1.2 if w = b
I ifw=a
{
CI(W) = 1 ifw=b.
A Cournot-Nash equilibrium is
q~(a) = q~(b) = .888889,

q2(a) = .955556, q2(b) = .822222.
The ex-ante expected profit for the industry is III + Il2 = 1.18519 +
1.19185 = 2.37704.
Now suppose that the two firms collude under the common knowledge
information rule. The information they use now is the trivial information
and the optimum total production is 1.33333. The expected industry profits
are 2.66667.
The Shapley value of the two firms (with Al = A2 = 1) is
1 1
Sh l = 2[2.66667 - 1.19185] + 2[1.18519] = 1.33,
1 1
Sh 2 = 2[2.66667 - 1.18519] + 2[1.19185] = 1.33667.
Then, a value production will be a solution to the following problem:
472
(5 - 1.5(ql + q2))q2 - q2 = 1.33667.

Hence, a value production plan is ql = .665 and q2 = .668333. To conclude,
as in the above example, the firm with the superior information gets re-
warded in the value production plan, by being assigned a higher level of
production and thus higher profits.
Remark 1: Alternatively, one could have used the notion of the a-core
which is defined as follows: We say that q E LQ is an a-core of the game C
if
it is not true that there exist S c I and (Yi)iES E IIiEsLQi such that
for any zI\S E IIi'isLQilIIi(yS,zI\S) > IIi(q) for all i E S.
It follows that under our assumptions in Section 3 the a-core is non-

empty [see Yannelis (1991)]. A collusive agreement that is an element of
the a-core is individually rational, Pareto optimal and coalitional stable.
Although these are clearly desirable properties, we do not have a straight-
forward way of selecting an element from the core that would capture the
"worth" of each firm. To this end, the Shapley value provides a relatively
easy way of figuring out the contribution of each firm to the total profits
and how to distribute them among the firms.
Remark 2: In the two firms case, the Shapley value is in the core and there-
fore in this case the duopoly with differential information can be viewed
as stable. This is not the case for more than two firms unless the corre-
sponding TU game is convex. Zhao (1998) provides necessary and sufficient
conditions for the deterministic TU game to be convex. In a subsequent pa-
per we intend to examine the conditions which guarantee the convexity of
the side-payments game defined in (8.1) or (8.2).
References
Balder, E.J., and N.C. Yannelis (1993): "On the Continuity of the Expected
Utility," Economic Theory," 3, 625-643.
Crampton, P.C., and T.R. Palfrey (1990): "Cartel Enforcement with Un-
certainty About Costs," International Economic Review, 31, 17-47.
Diestel, J., and J. Uhl (1977): "Vector Measures, Mathematical Surveys,"
American Mathematical Society, 15. Providence.
Dondimoni, M.P, N.S. Economides, and H.M. Polemarchakis (1986): "Sta-
ble Cartels," International Economic Review, 27,317-327.
Emmons, D. and A.J. Scafuri (1985): "Value Allocations-An Exposition,"
in C.D. Aliprantis et al., eds., Advances in Equilibrium Theory. Springer,
Berlin, Heidelberg, New York.
473
Krasa, S., and N.C. Yannelis (1994): "The Value Allocation of an Economy
with Differential Information," Econometrica, 62, 881-900.
Krasa, S., and N.C. Yannelis (1996): "Existence and Properties of a Value
Allocation for an Economy with Differential Information," Journal of
Mathematical Economics, 25, 165-179.
Laffont, J.J., and D. Martimort (1997): "Collusion under Asymmetric In-
formation," Econometrica, 65, 875-91l.
Roberts, K. (1983): "Self-Agreed Cartel Rules," working paper, IMSSS,
Stanford University.
Rotemberg, J.J., and G. Saloner (1990): "Collusive Price Leadership," The
Journal of Industrial Economics, XXXIX, 93-11l.
Shapley, L.A. (1953): "A Value of n-person Games," in H.W. Kuhn and
A.W. Thcker, eds., Contributions to the Theory of Games, Vol. H. Prince-
ton University Press, Princeton, NJ, 307-317.
Yannelis, N.C., (1991): "The Core of an Economy with Differential Infor-
mation," Economic Theory, 1, 183-198.
Yannelis, N.C. and A. Rustichini (1991): "Equilibrium Points of Non-
Cooperative Random and Bayesian Games," in Positive Operators, Riesz
Spaces, and Economics, C.D. Aliprantis, K.C. Border and W.A.J. Lux-
emburg, eds., Springer-Verlag.
Zhao, J., (1998): "A Necessary and Sufficient Condition for the Convexity
in Oligopoly Games," Mathematical Social Sciences, forthcoming.
Predicting proposal configurations in
cooperative games and exchange economies
Plnton Stefanescu
University of Bucharest, Faculty of Mathematics, 14 Academiei St., Bucharest

70109, Romania
Abstract. One considers stable configurations of proposals predicting

coalition formation and effective payoffs, which are called uniform com-
petitive solutions. Such "solutions" exist for almost all properly defined
cooperative games and, therefore, can be proposed as substitute of the
core. The new existence results obtained in the present paper concern also
the case when the coalitional function of a game has empty values. Plll con-
cepts and results are implemented in the competitive analysis of exchange
economies.
JEL Classification Number: cn

Keywords: Coalition formation, core
1 Introduction
During the last three decades, many authors have proposed different ex-
tensions of classical solution-concepts of cooperative games, in order to
cover situations when such solutions don't exist or when they seem to be
inadequate.
The competitive solutions have been introduced by McKelvey, Ordeshook
and Winer (1978) as any stable configuration of proposals. By a proposal
the authors mean an offer that a coalition can make, which in turn is for-
mally defined as a pair consisting of a coalition and a feasible utility-vector
which is effective with respect to it. Unlike in earlier solution-concepts, the
coalitions in a competitive solution are not necessarily mutually disjoint.
The un~form competitive solutions are basically modified versions of the
above concept. They appear first in Stefanescu (1993). In an extensive
comparative study of competitive and uniform competitive solutions of TU
games (Stefanescu, 1994), it is shown that, although the two concepts differ,
any competitive solution for non-trivial games is a uniform competitive
solution too. However, in the general framework of NTU games, significant
existence results of competitive solutions are missing. The main advantage
of the new concept of uniform competitive solution is that one can prove
476
its existence for almost all properly defined cooperative games.

To explain intuitively the basic ideas underlying the uniform competitive
solution, let us turn to the definition of the core. Roughly speaking, the core
is the totality of the utility vectors which are feasible and non-objectable;
i.e., any utility vector of the core is supported by the grand coalition and
no other coalition can improve upon it.
Suppose now that the players have committed to form mutually disjoint
coalitions and each coalition makes an offer to its members. If no other
coalition can make its members better off, one arrives at a first extension
of the core-concept; the set of proposal configurations defined above. Each
of them is stable, in the sense that no other coalitions are likely to form
because they cannot corrupt members making them better off. Moreover,
for superadditive games, the offers of the coalitions in a stable configuration
define a feasible utility vector which belongs to the core. (Therefore, the
core is nonempty.) In summary, one can predict non-objectable feasible
pay-off together with the coalitions to be formed.
Next, extending again the above ideas, consider that the coalitions in a
stable proposal configuration are not disjoint. If any two coalitions of the
configuration make the same offer to each common member, then all these
coalitions are likely to be formed. Obviously, one arrived to a qualitatively
different situation, because two nondisjoint coalitions cannot be formed si-
multaneously, and then, one cannot predict the formation of all coalitions
of the configuration. What one can predict is that these coalitions have
the best chance to be formed because no other coalition can make its offer
more attractive. For a game which does not admit other solutions, (par-
ticularly, when the core is empty), it seems to be the best substitute. In
a wide sense, likewise earlier solution-concepts, a uniform competitive so-
lution recommends rational outputs to be negotiated by the players. Note
that, as it results from the above discussion, when the core is nonempty,
then to each core-utility one can associate a uniform competitive solution.
Arguing somehow differently, the concept of aspiration leads to simi-
lar conclusions. The term "aspiration" has been introduced by Bennett
(1983) for TU games, but its earlier appearances under other denomina-
tions to several authors were also known. The formal definition, in the
general framework of NTU games, is given in Bennett and Zame (1988).
Likewise the classical solution-concepts, this notion focuses on the utility
received by the players. An aspiration is a utility-vector (generally, un-
feasible) with the properties that for each player there exists at least one
coalition supporting it, and that no coalition can make its members obtain
more than this vector says. As shown in Stefanescu (1994), to each uniform
competitive solution one uniquely associates a utility-vector which is an
aspiration.
The next four sections of the present paper are organized as follows.
Section 2 introduces two variants of the uniform competitive solution
concept, emphasizing their properties. One argues why a uniform compet-
477
itive solution is stable and one proves its rationality.

The results of Section 3 concern the existence of uniform competitive s0-
lutions. For NTU games one considers both the classical case of non-empty
valued coalitional functions and the case when the emptiness is allowed. In
this way one extends earlier results of Stefanescu (1993). Due to the rela-
tionships between uniform competitive solutions and aspirations, one can
extend also the existence results concerning aspirations. For TU games, the
existence results are obtained as consequences of the main theorems.
Section 4 discusses the validity of the above results in the framework of
games defined by bounded-valued coalitional functions.
The basic ideas and results of Sections 2-4 can be implemented in the
analysis of exchange economies. The last section of the paper concerns this
subject.
2 Stability of proposal configurations

The formal framework is usual in the NTU cooperative game theory; a
game is represented as a pair (N, V), where N = {I, 2, ... , n} is the set
of players and V is the coalitional function defined on the family of all
coalitions, 2N. The elements of the n-dimensional Euclidean space, denoted
by ~N, are called utility-vectors. For V we adopt one of the commonly used
representations; if C E 2N ,then V(C) is a subset of the ICI-dimensional
Euclidean space, denoted by ~c. Any element of the set V(C) is referred
to as a C-effective payoff. As we have already noted, an N -effective payoff
will be also referred to as a feasible payoff.
Some special notations will be used in what follows. For any vector x =
(Xl,X2, ... ,xn ) E ~N, Xc = (Xi)iEC stands for its projection onto ~c, and
for any subset A ~ ~N, one denotes prcA = {xc I x E A} (when C = {i},
then we will write, simply, priA.) If x and y are two vectors of the same
dimension, then x 2 y means Xi 2 Yi for all i; x > y means x 2 y but
x =I- y, and x ~ y means Xi > Yi for all i.
Turning back to the formal definition of a game, note that almost every-
where in the literature, V (C) is described as the set of all vectors represent-
ing utilities that the coalition C can guarantee for its members. Particularly,
this means that any such vector can be achieved by a suitable play. Since it
is hard to believe that a particular coalition can determine a game output
and the grand coalition cannot do it, then we are forced to consider that if
x E V(C) then x must have an extension in V(N).,i.e., x E prcV(N). In
view of this fact, a quite natural property for cooperative game is expressed
by the following definition:
Definition 1 The coalitional function V (the game (N, V) ) is monotonic

if C, DE 2N and C cD, always imply V(C) ~ prcV(D).
Monotonicity says that utilities which are effective for a coalition are
478
still available for all members of this coalition when they commit to form
a larger coalition.
The most general situation described by the pair (N, V) is of the non-
transferable utility (NTU) game. A transferable utility (TU) game may also
be represented as above. Traditionally, a TU game is defined by its real-
valued characteristic function v : 2N ---t ]R, where v( C) is the maximum
total payoff that the members of C can make independent of the actions
of players outside of C. A commonly used representation of a TU game in
the coalitional function form is then (N, V), where
V(C) = {x E ]RC ILXi ~ v(C)}. (1)

iEC
The concept of the uniform competitive solution, defined below, captures
the idea of the stability of a set of proposals made by some coalitions which
prevent the formation of rival coalitions.
Definition 2 A proposal of the game (N, V) is any pair (x, C) where C E
2N and x E V(C) nprcV(N).
Intuitively, a proposal is an offer that a coalition can make for its mem-
bers. The associated payoff is effective for the coalition but must be extend-
able up to a feasible payoff, that is, actually be achievable by the entire set
of players.
Remark 1 If the game is monotonic then the meaning of this notion ba-
sically reduces to that of C-effectiveness.
For the next definitions we will consider a finite collection of proposals,
K = {(xC,C)IC E Cl, where C S;;; 2N \{0} covers N, i.e., UCECC = N.
Definition 3 K is a uniform competitive solution (u.c.s.) if it satisfies the
following two conditions:
1. xf = xf for i E CnD and any coalitions C, DEC. (2)

2. :J no proposal (u, S) such that (3)
(i) Ui :2: xf whenever i E S n C and C E C,
(ii) Uj >xf for at least one j E Sand C E C such that j E C.
Weakening the second condition one obtains a new variant:

Definition 4 K is a weakly uniform competitive solution (w.u.c.s.) if it
satisfies (2) and
:J no proposal (u, S) such that Usnc »x~nc whenever C E C. (4)
The basic ideas in the above definitions are to ensure that any (weakly)
u.c.s. is a stable proposal-configuration.
479
This stability has two components: internal stability and external stabil-
ity. Internal stability requires by (2) that each player has identical prefer-
ences for the offers coming from different coalitions when he is a potential
member of them. Therefore, there are no objections from inside.
External stability, expressed in the two variants by (3) and (correspond-
ing to the weaker version of the Pareto principle) by (4), says that no out-
side coalition can threaten the existing configuration, because there does
not exist a coalition which can make its members better off.
As it immediately follows from the definitions that any solution is
coalitionally-rational.
Proposition 1 If K ={(xC,C)IC E C} is a u.c.s. (w.u.c.s.), then xC is a
Pare to optimum (weakly Pareto optimum) ofV(C) nprcV(N), for every
CEC.
Remark 2 For monotonic games, coalitional mtionality says that every
coalition involved in a solution offers its members an optimal effective pay-
off·
The usual meaning of individual rationality is that each player should
receive at least the best utility he can obtain himself. In the present frame-
work, we define the individually-rational payoff level of the player i E N
by
Vi = {SUPV({i})npriV(N), ifV({i})-/=0
-00, if V({i}) = 0.
Now, let us associate to each u.c.s. (w.u.c.s.) a utility vector w with the
components
Wi = xfwhenever i E C E C.
By (2), W is well-defined. Call it the ideal payoff vector associated with
K. Obviously, the i-th component of w is exactly the utility promised to
player i by all coalitions of C where he is a potential member.
Proposition 2 If w is the ideal payoff vector associated with the u. C.s.
(w.u.c.s.) K, thenwi 2: Vi, for all i E N.
In summary, each u.c.s. (w.u.c.s.) is individually-rational because it pre-
dicts to each player a utility not less than the one he can guarantee for
himself, and coalitionally-rational because each coalition directly involved
can offer its members an optimal effective payoff.
In the remainder of this section, we exemplify the previous definitions.
The games in all three examples have empty core but admit a u.c.s. or
w.u.c.s.
Example 1 Consider the three person TU game with the chamcteris-
tic function v where v({i}) = 1,i = 1,2,3; v({1,2}) = v({1,3}) = 4;
v({2,3}) = 3; v({1,2,3}) = 5.2.
480
Define V by (1).
The proposal-configuration
K = {( (2.5,1.5), {I, 2}), ((2.5,1.5), {I, 3}), ((1.5,1.5), {2, 3})}

is a u.c.s. and the associated ideal payoff vector is w = (2.5,1.5,1.5).
Example 2 n = 3. Define the coalitional junction V by
V({i}) = [O,l],i = 1,2,3; V({1,2}) = [0,1.5] X [0,2];
V( {I, 3}) = [0,2] x [0,1.5]; V( {2, 3}) = [0,1.5] x [0,2];

V( {I, 2, 3}) = {x E ~3lxl + X2 + X3 ~ 3.5}.
The proposal configuration
K ={((1.5, 1.5), {I, 2}), ((1.5, 1.5), {I, 3}), ((1.5, 1.5), {2, 3})}
is a w.u.c.s. and its associated ideal payoff vector is w = (1.5,1.5,1.5). One
can easily verify that the game does not admit any u.c.s ..
Example 3 n = 4. The coalitional junction V is given by
V({1,2}) = {x E ~2Ixl+X2 ~ 4}; V({1,2,3}) = {x E ~3Ixl+X2+X3 ~ 3};
V({1,2,4}) = {x E ~3lxl + X2 + X4 ~ 3});

V({l, 2, 3, 4} = {x E .wIXl + X2 + X3 + X4 ~ 2}); V(C) = 0, otherwise.
The proposal configuration
K = {( (1,1,1), {I, 2, 3}), ((1,1,1), {I, 2, 4})}

is a u.c.s .. The associated ideal payoff is w = (1,1,1,1). Note that the game
is ordinally convex.
3 Existence of uniform competitive solutions

As already mentioned, the present formalism allows emptiness of the val-
ues of the coalitional function. Unlike the classical approach, V(C) may
be empty for some nonempty coalitions. Such situations actually appear,
"simple games" being one of the most representative.
To avoid any confusion, everywhere in this section, a cooperative game
will be identified by a triple (N, V, W), where W is the family of all winning
coalitions, i.e.,
W = {C E 2N I V(C) # 0}.
As usual, 0 ~ W. If V satisfies (AA), then W is a monotonic class, i.e.,
C E W, C c D, for some D ~ N => DEW. (5)

481
The emptiness of the values of the coalitional function could be a serious

drawback for the existence of classical solutions. For instance, convexity
fails to be sufficient for the nonemptiness of the core. The game in Example
1 is (ordinal) convex, but the core is empty. However, the existence of a
U.C.s. can be proved whatever assumptions on W we make.
Everywhere in this section, the cooperative game (N, V, W) is assumed
to satisfy the following four axioms:
(A.I) V(C) is closed in jRc, for every C E W.
(A.2) For every C E Wand y E jRc the set {x E jRclx E V(C),x 2: y} is
bounded.
(A.3) For every C E W, if x E V(C) and y :S x for some y E jRc, then
yE V(C).
(AA) V is monotonic.
Axiom (A.3), referred to as comprehensiveness, allows for the free dis-
posal of utility for any coalition, while (A.I) and (A.2) are rather technical
conditions.
Note that for any TU game its NTU representation defined by (1) satisfies
all (A.I)-(AA).
The existence results of this section, concern both u.c.s. and w.u.c.s. and
apply to NTU or TU games represented as above. For the proof of the
main theorem, it is useful to represent the solutions in a compact form.
The starting point is the remark that if K = {(xC, C)IC E C} is a (weakly)
uniform competitive solution, then it basically consists of the ideal payoff
vector wand the collection of coalitions C. Indeed, from the pair (w, C)
one easily obtains K taking xC = Wc for every C E C. In fact one can
easily prove the following proposition as an immediate consequence of the
previous definitions.
Proposition 3 Let C ~2N\{0} be any collection of coalitions for which
UCECC = N. Assume that the utility vector wE JRN is such that
Wc E V(C) nprcV(N) whenever C E C (6)
and
:3 no proposal (u, S) such that u > ws(u » ws). (7)
Then, K ={(x C , C)IC E C} is a u.c.s. (w.u.c.s.).
Proof. Note that a utility vector can be associated to several u.c.s.'s as
their ideal payoff vector, but a pair (w, C) uniquely defines a solution.
For any utility vector w, denote C(w) = {C E 2Nlwc E V(C) n
prcV(N)}. Obviously, if (w,C) represents a u.c.s. (w.u.c.s.) for some cov-
ering C of N, then (w,C(w)) also defines a U.C.s. (respectively, a w.u.c.s.).
Conversely, if (w, C(w)) represents a u.c.s. (w.u.c.s.), then (w, C) is so, when-
ever C ~ C(w) and UcECC = N. One can say that (w,C(w)) is a maximal
482
u.c.s. (w.u.c.s.) with respect to the ordering "~" on the set of proposal-
configurations.
A first existence result concerning w.u.c.s. shows that this solution-
concept is actually an extension of the core concept.
Proposition 4 If the core C(N, V) is nonempty, then for every w E
C(N, V) the proposal-configuration J( = {(w,N)} is a w.u.c.s.
Proof. Immediately follows from the previous proposition. •
Remark 3 A converse implication for games satisfying (A.4) also holds.
In this case, if J( ={w,C} is any w.u.c.s. such that NE C(w), then w E
C(N, V). (Therefore, the core is not empty.)
Now, the main existence result for w.u.c.s. follows.
Theorem 5 Any cooperative game (N, V, W) satisfying (A.l}-(A.4) ad-
mits a w.u.c.s, provided that W -=I 0.
The proof is based on the following construction, similar to the "reduced
game technique" (see Peleg, 1986).
Pick a coalition T E W, minimal with respect to "~" (i.e., there is no
SEW such that SeT ). Let z be a Pareto-optimum of VeT) (Le.,
x E VeT), x ~ z imply x = z). Clearly, if T is a singleton set, say T = {k},
then z = Vk = max V( {k}) = max V( {k}) n PTk V(N) (by (A.l), (A.2) and
(AA». If I TI ~ 2, then for any y ::; VeT) the set {x E V(T)lx ~ y} is a
nonvoid compact in jRT (by (A.l) and (A.2) and admits a Pareto-optimum.
Obviously this one is a Pareto-optimum of VeT). Note also that by (A.4)
z E PTTV(N).
Set M = N \ T and define the reduced game (M, VM, WM) by
VM(C) = UP~T{X E jRcl(x,zp) E V(CU PH
for all 0 -=I C ~ M.
The next two lemmas will be proved under the assumptions of the the-
orem.
Lemma 6 WM = 2M \ {0} and VM satisfies all conditions (A.l}-(A.4).
Proof. To prove the first assertion, it is sufficient to observe that CuT E W
(from monotonicity ofW) and z E VeT) ~ PTTV(CUT), and so VM(C) -=I 0
for every C ~ M, C -=I 0.
Show now that VM verifies (A.l) to (AA):
(A.l) is obviously satisfied, since each term of the union representing
VM(C) is the section of a closed set.
To check (A.2), pick a vector y E jRc, for some C ~ M. Then
{x E VM(C)lx ~ y} = up~dx E jRcl(x, zp) E V(C U P),x ~ y} ~

UP~TPTC{X E V(CuP)lxc ~ y,xp ~ zp}
483
and each set in the last union is the projection of a bounded set.
Since the comprehensiveness of VM is obvious, let us verify (A.4). Assume
that C c D ~ M. Then, for every P ~ T, V(C U P) ~ prcupV(D U P)
and, consequently, prcV(C U P) ~ prcV(D UP). Hence,
VM(C) = UPS;;TprC(V(CUP) n {x E lRcUPlxp = zp}) ~

UPS;;Tprc(V(D U P) n {x E lRDUPlxp = zp}) = prcVM(D) . •
Lemma 7 Assume that the reduced game has a w.u.c.s. and let w be its
associatedidealpay-ofJ. Then the pair (u,C(u)), whereu = (w,z) represents
a w.u.c.s. of the game (N, V, W).
Proof. Since (w,C(w)) is a w.u.c.s. of the reduced game, one has
UCEC(w)C = M. Moreover, for each C E C(w), Wc E VM(C), so that
there exists P ~ T such that (wc, zp) E V(C U P) and thus CUP E C(u).
Since T E C(u) it follows that UC'EC(u)C' = M U T = N.
Now, to prove the lemma, it suffices to verify (7) of Proposition 3.
Let (x, S) be a proposal of the game (N, V, W). Clearly, S = T, or,
S n M "# 0. Suppose that x » us. Obviously, S = T is impossible, since
UT = z and z is Pareto-optimum of V(T). Then we must have S = CUP,
with 0 "# C ~ M and P ~ T. By (A.3), (xc, zp) E V(S) and hence,
Xc E VM(C). But Xc »uc = Wc, contradicting the definition of w . •
Now we will prove the theorem in two steps.
1. Consider first the case when W =2 N \ {0}. The existence of a w.u.c.s.
will be proved by induction on n = 1Nl. For n = 1, V({l}) is a non-
empty closed upper-bounded subset of lR so that, {(Wl' {I})}, where Wl =
maxV({l}), is a w.u.c.s. Assume the existence of w.u.c.s. for any game
with at most n -1 players, and consider the n-person game (N, V, W). Set
M = N \ {n} and form the reduced game. Since WM = 2M \ {0}, it follows
by induction that the reduced game has a w.u.c.s. Hence, by Lemma 6, the
game (N, V, W) has a w.u.c.s.
2. The general case when W "#0. Apply Lemma 6 to the reduced game
properly defined. The reduced game has a w.u.c.s (by the first step of the
proof), so the original game also has a w.u.c.s ..•
By similar arguments, the existence of a u.c.s. can also be proved if the
following condition is fulfilled:
(A.5) For any c > 0, S ~ N, x E V(S) and i E S, there exists y E V(S)
such that Yi = Xi - c, and Yj > Xj for all j E S \ {i}
Basically, this axiom states that, within the set of effective payoffs of a
given coalition, it is always possible to improve the utility of other players
if one player accepts to diminish his own utility.
The game in Example 1 doesn't satisfy (A.5) and doesn't admit a u.c.s.
Therefore, in the next existence theorem this condition cannot be removed.
484
Theorem 8 Any coopemtive game (N, V, W) satisfying (A.l}-(A.5) and

W =10 admits a u.c.s..
Proof. In fact, we can prove that for a game satisfying (A.5), any w.u.c.s.
is a u.c.s ..
Let K be a w.u.c.s. represented in the compact form by (w,C) and a
proposal (x,8) such that x > Ws. Pick an i E 8 such that Xi > Wi and put
£ = Xi - Wi. By (A.5) there exists a proposal (y,8) such that Yi = Xi - ~
and Yj > Wj for allj E 8\ {i}. But this means Y ~ ws, which is impossible.
Thus K is a u.C.S ..•
Assuming (A.5), Proposition 4 can also be strengthened:
Proposition 9 Let (N, V) be any game satisfying (A.5). If wE C(N, V),

then K = {(w,N)} is a u.c.s ..
Proof. Follows from Proposition 4 using the same arguments as in the

proof of Theorem 8. •
Finally let us discuss the structure of the coalition configurations of the
u.c.s. At first glance, the opportunity of a special structure (balanced or
separating collection) to motivate the concept of competitive solution is not
obvious. Moreover, in the general framework of the present paper (where
the emptiness of the values of the coalitional function is allowed), the set
of winning coalitions may not contain a balanced or separating collection.
However, respecting the intuitive and economic interpretation of such
structures, we can inquire whether the existence of a u.c.s. with a structured
configuration of coalitions can be guaranteed in some particular cases. A
positive answer can be given in the classical case W = 2N \ {0}.
Denote by "1 the set of all coverings of N, "1 = {C c;;.2 N I UCECC = N}.
For every subset "1' of "1 let (N, V-y')' consider the game with the coalitional
function V defined by
V-y' (8) = { V(8),

=I N
if 8
UCE-Y' {x E RN I Xc E V(C), VC E C}, if 8 = N.
One can easily verify that for every "1' c;;. "1 with {N} E "1', the games
(N, V) and (N, V-y') have the same uniform competitive solutions. Particu-
larly, if 'Yb is the family of all balanced collections, then the game (N, V-Yb)
is balanced. If (N, V) satisfies all assumptions of the existence theorems,
then (N, V'Yb) has nonempty core. Let U E C(N, V'Yb ). Then there exists
C E 'Yb such that Uc E V(C) for all C E C. By the monotonicity of (N, V),
for each C E C there exists u C E V(N) such that ug = Uc. Clearly, K
= {(u C , C) ICE C } is a u.c.s. of (N, V-Yb)' thus it is a u.c.s. of (N, V) with
a balanced configuration of coalitions.
485
4 Bounded valued coalitional functions

An alternative representation of the cooperative games includes bounded-
ness conditions in the spirit of the "individual rationality" of the imputa-
tions in classical TU theory. A standard definition of V in this case restricts
its values in the positive orthant of the utility spaces. Accordingly, we can
modify the axioms assumed in the previous section replacing (A.1), (A.2),
(A.3) by the following:
(A.l') V(C) is closed in IR~, for every C E W.
(A.2') For every C E W , V(C) is bounded in IR~.
(A.3') For every C E W, if x E V(C) and y :::; x for some y E IR~, then
yE V(C).
For a TU game with characteristic function v, the NTU representation
should now be
V(C) = {x E R~I LXi:::; v(C)}. (8)
iEC
Obviously, V satisfies (A.l')-(A.3').

The method of proof used in Theorem 5 continues to work in the present
formalization to obtain the following result.
Theorem 10 If the cooperative game (N, V, W), where W i- 0, satisfies

(A.l'), (A.2'), (A.3') and (A.4), then it admits a w.u.c.s.
The next result is the counterpart of Theorem 8 and can be obtained
under an additional condition which is weaker than (A.5).
(A.5') For any S ~ N, x E V(S), i E S and for Yi = Xi - € and Yj > Xj for
alljES\{i}.
Theorem 11 If the cooperative game (N, V, W), where W i- 0, satisfies

(A.l'), (A.2'), (A.3'), (A.4) and (A.5') then it admits a u.c.s ..
Proof. Following the proof of Theorem 8, it suffices to show that if w is
the associated ideal payoff of a w.u.c.s K and Y > Ws for some proposal
(y, S), then there exists y' E V(S) nprs(N) such that Y' ~ Ws. Obviously,
w E R~, so that, if Yi > Wi for some i E S, then Yi - ~ 2: 0, where
€ = Yi - Wi· By (A.5') it gives the existence of Y' E V(S) nprsV(N) such
that y~ = Yi - ~ and yj > Xj for all j E S\ {i}, a contradiction. •
The case of TU games requires a special examination.
If the NTU representation is defined as in (8) then (AA) may not be satis-
fied. A special situation is when the characteristic function v is monotonic,
i.e., Cc D =} v(C) :::; v(D). Then the associated NTU game is monotonic
and, therefore, it has a u.c.s. Other conditions for existence are discussed
in Stefanescu (1994).
486
5 Competition in exchange economies

Uniform competitive solutions have natural correspondents in exchange
economies. When the core is nonempty, each core-allocation can be recom-
mended to all agents. When this is not the case, it is expected that several
coalitions will bid for members based on different allocations available to
them. Thus, one arrives at situations when any logical prediction is to be
based on a configuration of proposals.
Let us begin by explaining the formal framework of this section.
In a compact form a (neoclassical) exchange economy will be represented
as
where N = {1,2, ... ,n} is the set of agents, (i is the i-th agent's initial
endowment, and Ui is real valued function representing his preferences over
the commodity space.
One considers the finite-dimensional case, with m commodities, and iden-
tifies the set of all consumption plans of each agent with the positive orthant
JR+ of the commodity space. Then Ui : JR+ - JR, and if ui(a) > ui(b) we
will say that the agent i strictly prefers the consumption plan a to b. Each
agent has a non-zero non-negative endowment (i.e., (i > 0, for all i E N)
and the total endowment ( = EiEN (i is assumed to be strictly positive
« »0.)
Any redistribution of the total endowment to agents is called an alloca-
tion. Formally, an allocation is an n-tuple a = (aI, a2, ... , an) of consump-
tion plans, ai E JR+, i E N. Therefore, the set of all allocations is defined
as A = {x = (aI, a2, ... , an) I ai E JR+, i E N, EiEN ai = O.
For any set C of agents (coalition), denote by A(C) the set of all alloca-
tions where the members of C share their own endowment, i.e.,
A(C) = {a E AI Lai = L(i}'

iEC iEC
Now the counterparts of U.c.S. (w.u.c.s.) can be defined. Let
N = {(aC,C)IC E C}
be a finite configuration where C ~ 2N \ {0}, UcECC =N and, for each

C E C, a C E A(C).
Definition 5 N is a competitive negotiation set (c. n.s.) if
ui(af) = ui(af) whenever i E CnD for any two coalitions C, DE C (9)

487
and
:3 no S E 2N and b E A(S) such that

(i) ui(bi ) 2:: ui(af) whenever i E S n C and C E C,
(ii) uj(bj) > uj(ar) for at least one j E Sand C E C such that j E C.
(10)
Definition 6 N is a weakly competitive negotiation set (w.c.n.s.) if it sat-
isfies (9) and
:3 no S E 2N and b E A(S) such that ui(bi ) > ui(af)

whenever i E S n C for some C E C. (11)
Now it is easy to show that an exchange economy has competitive nego-

tiation sets if and only if an associated cooperative game admits uniform
competitive solutions. The two standard NTU games associated with the
economy [ correspond respectively, to the unbounded and the bounded-
valued versions of the coalitional function. Thus, V is defined by either
V(C) = {x E ~CI Xi:::; ui(ai) for some a E A(C) and for all i E C} (12)
or
V(C) = {x E ~~I Xi:::; ui(ai) for some a E A(C) and for all i E C}. (13)
The next two propositions hold no matter which variant of V is consid-

ered.
Proposition 12 The associated NTU game of an exchange economy is
monotonic (V satisfies (A·4))·
Proof. Indeed, if C c D and x E V (C), pick an a E A( C) such that
Xi :::; ui(ad for all i E C and define b E A(D) by bi = ai, if i E C,
and bi = (i if i E D\C. Further, define y E ~D by Yi = Xi for i E C and
Yi = ui(bi ) for i E D\C. Then yE V(D) and Yc = x, thus, x E prcV(D) . •
The next proposition justifies the formal similitude expressed by the de-
finitions of the uniform competitive solutions and competitive negotiation
sets.
Proposition 13 N = {(aC, C)IC E C} is a c.n.s. (w.c.n.s.) of the econ-

omy [ if and only if K = {(xC, C)IC E C}, where xf = ui(af) for all
i E C E C, is a u.c.s. (w.u.c.s.) of the associated NTU game.
Proof. Let us prove the c.n.s-u.c.S version.

First assume that N is a c.n.s of the economy [. Clearly, by the definition
of V, xC E V (C) nprc V (N) for C E C. Since property (2) of K immediately
488
follows from (9) let us verify (3). To the contrary, assume that (y, S) is a
proposal of the game (N, V) which verifies (i) and (ii) of (3). Then there
exists b E A(S) such that ui(bi ) ~ Yi for all i E S. Hence, band S satisfy
(i) and (ii) of (10), contradicting the definition of N.
Conversely, assume that lC is a u.c.s of the associated game of E. To
each proposal (xC, C) E lC corresponds an allocation aC E A(C) such that
xf ~ ui(af) for all i E C. One has xf = ui(af) for all i E C, because xC
is Pareto-optimum of V (C) n prcV (N). Choose one a C for each xC and
form N = {(aC,C)IC E Cl. Obviously, N satisfies (9). Show that it also
satisfies (10). If it is not that the case, then there are Sand bE A(S) which
verify (i) and (ii) of (10) with respect to N. But then, the proposal (y, S),
where Yi = ui(bi ), i E S, verifies (3) with respect to lC, a contradiction. •
The next existence results use this proposition and the results of the
previous section.
Theorem 14 If the preferences in the exchange economy E are expressed

by continuous junctions, then E admits a w.c.n.s ..
Proof. It suffices to show that the associated NTU game (12) satisfies the
assumptions of Theorem 8. By Proposition 12 it satisfies (A.4) , and the
comprehensiveness trivially follows from the definition. For each i E N,
the continuous function Ui is bounded on the compact [0, (j. Hence, V
has nonempty upper bounded values and, therefore, condition (A.2) is also
satisfied. To verify (A.I), assume that the sequence (xt) C V(C) converges
in ]Rc to some x. Then there exists a sequence (at) in A(C), with the
property that x~ ~ ui(aD for every i E C and for all t. Since A(C) is a
compact, there exists a subsequence of (at) that converges to an allocation
a E A(C). The continuity of Ui implies that Xi ~ ui(ai) for every i E C.
Hence, x E V (C), which proves the closedness of V (C) . •
Definition 7 The function U : ]Rm --t ]R is said to be mono tonic if u( a) >

u(b), whenever a > b, for a, b E ]Rm.
Theorem 15 If the preferences in the exchange economy E are expressed

by continuous and mono tonic functions, then E admits a c. n.s ..
Proof. One applies Theorem 11 to the associated games (13). For that,
one should prove that axiom (A.5') is also satisfied.
Without loss of generality, assume that Ui(O) = 0 for all i E N. Pick a
coalition C with 101 ~ 2 and an x E V(C). Then there exists a E A(C)
such that Xi ~ ui(ai) for all i E C. The non trivial case is when x> O. Then
ui(ai) > 0 for at least one i E C (by the monotonicity of Ui ). Pick then c
such that 0 < c ~ Xi. By the continuity of Ui one has that Xi - c = Ui(Aai)
for some A E (0,1). Take the allocation b, where bi = Aai, bj = aj + Ib[~l'
if j E C\{i} and bj = (j if j E N\C. Obviously, b E A(C) and the
489
monotonicity of utilities implies that uj(bj ) > uj(aj) for all j E C,j f:. i.
Then, Y verifies (A.5 / ), where Yi = Xi - € and Yj = uj(bj ) for j E C\{i} .•
References
Bennett, E. (1983) "The Aspirations Approach to Predicting Coalition For-
mation and Payoff Distribution in Sidepayments Games". International
Journal of Game Theory, 12: 1-28
Bennett, E. and Zame, W.R. (1988) "Bargaining in Cooperative Games".
International Journal of Game Theory, 17: 279-300
McKelvey, R.D, Ordeshook, P.C and Winer, M.D (1978) "The Competitive
Solution for N-Person Games Without Transferable Utility, With an Ap-
plication to Committee Games". The American Political Science Review,
72: 599-615
Peleg, B.(1986) "A Proof that the Core of Ordinal Convex Game is a von
Neumann-Morgenstern Solution". Mathematical Social Sciences, 11: 83-
87
Stefanescu, A. (1993) Competitive Solutions and Uniform Competitive So-
lutions for Cooperative Games". Social Science Working Paper 868, Cal-
ifornia Institute of Technology, Pasadena
Stefanescu, A. (1994) "Solutions for Transferable Utility Cooperative
Games". R.A.I.R.O. Rech.Oper. 28: 369-387.
International financial equilibrium with risk
sharing and private information*
Bart Taub
Department of Economics, University of Illinois at Champaign,1206 South Sixth

Street, Champaign, IL 61820, U.S.A.
Abstract. There are two economies. Within each economy individuals un-
dergo idiosyncratic and common shocks to their endowments. The idiosyn-
cratic shocks are independent across individuals and as such can be pooled,
providing perfect mutual insurance. The common shocks are not internally
insurable. The two countries can trade risk in the common shocks however,
and standard ideas about trade in goods apply: there is a price of foreign
risk in terms of domestic risk, and beneficial trade occurs at this price. The
standard idea that small countries benefit from trade goes through in the
risk domain as well.
The mutual benefits of trade in risk can collapse if information about
the idiosyncratic endowment processes is private. In that instance an
information-eliciting contract can provide insurance of the idiosyncratic en-
dowment processes even under autarky. However, if the two countries open
to trade in the common risk processes the domestic information contracts
are affected; the degree of insurance can be reduced because of spillover
from the effects of the international risk sharing. The desire to insure both
idiosyncratic and common shocks can conflict to the point that one country
prefers autarky. In contrast to the standard result, the benefits of trade can
shrink and become negative as country size shrinks. These findings would
be somewhat academic were it not for the fact that within the model,
the information-eliciting contract generates a dynamic allocation identical
to that of an asset equilibrium. Thus, a theoretical foundation exists for
countries rationally restricting trade in assets.
JEL Classification Number: G15
Keywords: International financial equilibrium, risk sharing
*Earlier versions of this paper were presented at the 1997 meetings of the Society for
the Advancement of Economic Theory, the 1997 Summer Meetings of the Econometric
Society, Arizona State University, and the Federal Reserve Bank of Atlanta. I thank
Ayhan Khose, Dan Bernhardt, Hector Chade, Marco Espinosa, Alejandro Manelli and
a referee for helpful comments.
492
1 Introduction
That international trade is beneficial is a cornerstone of economics. The
idea rests on standard competitive and general equilibrium: if there is con-
vexity and competition, there is gain from exchange. The idea dictates
that countries remain as open as possible to international trade, with any
deviation from full openness viewed as pathological. It is typically contem-
plated in the context of static equilibrium with multiple goods; in dynamic
settings it goes through if goods indexed by time are treated as different
goods. It is this latter setting that will be examined here. As is standard
in macroeconomic theorizing, a composite static good will be assumed,
collapsing all exchange to intertemporal exchange. The standard theorem
dictates that countries should trade intertemporally; the exchange will take
the visible form of assets exchanged for goods, and subsequent repayment
of the goods. By this logic countries should be borrowing and lending in
the international market, and enjoying benefits from doing so.
Such borrowing and lending, aside from smoothing the mundane frictions
of goods exchanges, exists essentially to smooth fluctuations of production
and consumption. To the extent that such smoothing is successful, bor-
rowing and lending can be viewed as insurance against those fluctuations.
Unsurprisingly, an international trade equilibrium in such insurance has
the standard efficiency property: regardless of country size, internal char-
acteristics, or volatility of endowment, trade is beneficial.
But this result depends on one assumption: that information about out-
comes is publicly observable. If that assumption is dropped and information
is private, the possibility develops that opening to international trade is
damaging. Because of this, countries can rationally restrict trade in assets,
making them appear inefficiently closed. The reason the standard result is
overturned is that if information is private, institutions constructed to ap-
propriately elicit private information can be radically affected by opening
to trade, thence forcing small countries to bear excessive risk.
I investigate this idea in a model that is similar to the two-country model
of Lucas [20], embellished with idiosyncratic shocks to individual endow-
ments within each country; it is the information about those shocks that is
private. International trade in the absence of private information would not
affect the treatment of those local shocks; it only affects risks that cannot
be diversified autarkically. But under private information, the properties
of international exchange are strongly affected by the need to elicit local
idiosyncratic information. In welfare terms, this effect can be thought of as
amplification of the direct effects of exchange.
One could argue for the relative insignificance of private information in
real settings. It is a tenet of this paper that private information is in fact
significant because money and bonds, which are ubiquitous in real settings,
reflect its presence. This idea is developed in [32], [31], and [30], and I will
refer to it as asset equivalence here. When settings are contemplated in
493
which there is no private information, the intertemporal exchange institu-

tions that arise are in fact equity markets, not asset markets.
Under autarky (in the international-trade sense of the word) each coun-
try operates its own risk sharing contract, generating asset equilibria of the
above kind. The motive for trade is as with public information: endowment
fluctuations have aggregate components that are uninsurable within each
country, but which are statistically independent and hence potentially in-
surable across countries. This risk is inefficiently borne under the autarkic
contract.
Opening to international risk sharing, which I will refer to as trade even
though there is only one static good, allows for this risk to be shared. Un-
der trade, risk sharing entails pooling the aggregate endowment processes.
Unlike situations in which many individuals pool independent endowment
fluctuations and so effectively eliminate them, the aggregate shocks must
be explicitly shared: foreign aggregate shocks are then partially borne by
the domestic country and vice versa. Under full information, there is an
explicit compensation for this risk via the price of that risk. A small coun-
try holds a proportionally smaller quantity of foreign endowment than the
domestic endowment it is able to insure and consequently holds less risk as
well. This is no more than an expression of the desirability of trade in risk
instead of goods.
Under private information, the price that develops is complicated by
the additional requirement that exchange must be founded on incentive-
compatible information transmission, and this can prevent the price from
appropriately compensating for the extra burden of foreign risk. This is
because in order to be incentive compatible, the intertemporal pattern of
signals must be used in exchange, and small countries are indistinguishable
from large ones in such patterns. In concrete terms, one can think of a
small country, such as Belarus, joining the international financial market. In
borrowing and saving, Belarus must face international risks not necessarily
directly correlated with its inherent domestic risks by holding foreign assets;
one could think of callable foreign loans, which will be demanded in a
foreign banking crisis, or the holding of foreign currency and consequent
exposure to its price level fluctuations, as examples of this risk. But the
need to simultaneously insure private idiosyncratic shocks constrains this
facet of risk sharing. In contrast to standard results in trade theory, a small
trading partner might therefore wish to limit the extent of trade and risk
sharing because of this problem.
This result will be developed by first examining the full information
setting as a benchmark, followed by an analysis of the private information
setting. Before proceeding with the model, I set out the mechanics of the
frequency domain approach to contract theory that will be used.
494
2 LQ contract mechanics
The linear-quadratic technique lends itself to a "building-block" approach.
Stochastic processes and their realized histories can be contemplated and
manipulated as objects. Contracts can be represented as functions and their
interactions with the individuals and elements of the economy can be rep-
resented by multiplication operations. Information flows are captured by
simple internal properties of the contracts which have direct analogues in
ordinary matrix algebra.
The elementary building blocks of a simple scarce-information model,
then how they fit together, will now be set out to illustrate these points.
Begin by supposing that there is a single individual acting autarkically.
The individual has a quadratic, additively time separable objective
2: {jt(Ct -
00
-Eo c)2,
t=o
where Ct is consumption and {j the discount factor. There is a bliss point at
c. Some standard technical assumptions are that the endowment process,
at = A(L)et, is covariance-stationary, with Gaussian, iid, mean-zero inno-
vations et, and with serial correlation described by a function of the lag
operator A(L), and with appropriate restrictions on the form of A.
One can view the individual as trying to hit a stochastically moving
target, A(L)et. In an amalgamation of many such individuals, there is a
potential to offset the target via mutual insurance. If the endowment re-
alizations are common to all individuals, no insurance is possible. But if
each individual's innovations et are independent of all others, those with
high positive at can trade with individuals with negative at. If there were
an insurance contract, and if it operated to perfectly offset residual endow-
ment realizations, the stochastic component of consumption in each period
would be A(L)et - h(L)A(L)et = 0, where h is some contractually deter-
mined function that operates on each individual's endowment process. All
risk would be eliminated.
If information about each individual's endowment process is private, such
insurance depends on successfully inducing individuals to truthfully reveal
their endowment realizations to the contract. This revelation is accom-
plished via incentive constraints. There are two kinds of incentive constraint
that need to be considered in the infinite horizon, linear-quadratic setting.
The first can be thought of as an adjustment cost constraint-it has the
natural interpretation of the cost of sending signals. I treated this explic-
itly in [32], but as I showed in [31], this constraint can be put into the
backround by noticing that its main effect is to ensure that the policies of
individuals are well-behaved.
The second incentive constraint is central to the analysis. The histories
of individuals' signals of their states are in essence infinite-dimensional vec-
tors. The contract weights these vectors in a way that unravels the vector,
495
revealing the whole history of states--particularly the current realization-

completely. This weighting corresponds to matrix multiplication, with the
matrix expressible as a filter-a function of the lag operator. 1
The private-information contract operates in the following way. First,
individuals observe their endowment realizations and each period send a
signal, sA(L)et, to the contract. Second, the contract takes the entire his-
tory of that signal as an input and applies a filter h-a set of weights for
each lag of the history-to it. This forms the insurance payment paid to the
individual (or, if the signal is of a good endowment realization, premium
charged).
Individuals are assumed to understand the nature of the contract and
the internal structure of h, and this structure directly influences the for-
mula for SA. They would, if possible, subvert the contract by inverting the
filter-reporting its mirror image-in order to attain complete insurance
not only of the stochastic process at but to attain their bliss point as well.
Just as when all insurees attempt to attain complete insurance in a sta-
tic adverse selection setting of the sort definitively analyzed by Rothschild
and Stiglitz [26], insurance would be infeasible. The solution in the linear-
quadratic setting is to make use of the matrix-like behavior of the contrac-
t's weighting filter, designing it to be noninvertible. This noninvertibility is
easily generated by putting a so-called zero into the filter. (It is a solution
of the equation h(z) = 0, hence the name.) The zero makes the contract
incentive-compatible, inducing individuals to report truthfully, but as in
static models, it is at the cost of providing incomplete insurance.
It remains to choose the zero, and to give it an economic interpretation.
In [32], I showed that if the endowment process is a nonstochastic constant,
the zero must be related to the discount factor. In [31], I generalize this
finding, showing that if there is an aggregate component of the endowment
process that is stochastic, then for the contract to be feasible the zero
is determined by the aggregate process. The aggregate component of the
endowment process is uninsurable, and the correct zero in h removes the
temptation for individuals to attempt to insure it.
The zero has a second interpretation. Its inverse, when multiplied by the
inverse of the discount factor, is exactly the gross interest rate that clears
the market in a corresponding lending equilibrium. Here is the concrete ex-
pression of asset equivalence. An unencumbered lending equilibrium, then,
reflects the information strategy of an insurance contract.
1 Green [11] originated the idea of a contract operating on histories of signals as a

way to elicit information in a dynamic insurance setting. There is a burgeoning literature
that uses t,his contract approach: [2], [12], [19], [21], [23], [24], [25], [27], [37], [38]' and
[18], as well as [30], [31], and [32].
496
3 The economic and technical setting

There are two countries, initially in autarky. Each country consists of a
continuum of individuals who attempt to maximize their utility in the face
of ongoing stochastic shocks to their endowments. Those shocks have a de-
gree of statistical independence so that individuals can gain from sharing
the risk through some exchange mechanism; when one individual's endow-
ment is high, he can share it with another whose endowment is low in
exchange for a later quid pro quo. The sharing mechanism must overcome
two obstacles to succeed. First, an equilibrium must ensure commitment
to the ongoing execution of the mechanism, because there is an incentive
to renege when it comes time to deliver the quid pro quo. Second, it must
be informationally consistent: if the shocks are only privately observable,
the mechanism must be able to elicit appropriate information to provide
a foundation for redistribution. 2 For that reason I continue here with the
strategy of modelling an information-constrained risk-sharing contract and
then use the equivalence to draw conclusions about asset equilibria.
The two countries will be designated 1 and 2; economy 1 will be thought
of as the domestic economy where appropriate, but the two economies
will for the most part be assumed to be roughly symmetric. Each has a
stochastic domestic endowment process, Yit, where
YIt = aIt + bit;

where ait = Ait(L)eit, and bit = Bi(L)uit. The innovation process eit is
idiosyncratic white noise: for an individual x in the continuum of individuals
Xi,
E[ei,t+k(X)I( ... , ai,t-l(X), ait(x))] = 0, k > 0;
cov[ei,t+j(X), ei,t+k(x)I(.·., ai,t-l(X), ait(x))] = { ~ j=l:k

. k' J',k
J=
> O.
and for distinct individuals x and Y,
The bi-processes are similar and independent of the ai processes, but the
innovations are common across individuals
E[Ui,t+k(X) 1(· .. , bi,t-l (x), bit (x ))] = 0, k > 0;
2Under full information, a contract in which commitment must be internally gener-

ated is not always feasible, and even when it is feasible, the properties of the contract
are such that contractual consumption lacks a realistic degree of variability; see [33] and
[18]. In this paper I consider only the second issue, private information, and take the
existence of adequate commitment as given (but see [31]). In previous work I have ar-
gued that within the technical confines of the type of model I use here, the appropriate
private information mechanism is indistinguishable from an asset equilibrium.
497
COV[Ui,t+i (X), Ui,t+k(X) I(... , bi,t-l (X), bit (X))J = {~~ ~ ~ ~ , j, k > O.
and for distinct individuals x and y,
cov[Ui,t+i (x), Ui,t+k(Y) I(... , bi,t-l (x), bit (x)), (... , bi,t-l (y), bit(y))J = a~.
Because the effects of interest in this paper have to do with serial correla-
tion, the variances will be normalized:
a e2 -_ a2 -1 .
u -
I make the assumption that the law of large numbers holds, in the sense
that the realized average idiosyncratic innovation and idiosyncratic com-
ponent of endowment is zero, replicating the mean of the distribution, in
both economies [17, lOJ. The idiosyncratic part of endowment is therefore
insurable in the sense that pooling it yields zero consumption.
The two countries can differ in size (or have different innovation vari-
ances, but I am suppressing this as redundant); I normalize them so that
country 1 has size 8 and country 2 has size 1- 8. In the risk sharing analysis
below, these relative sizes will translate into variances.
Each individual x has a discounted, infinite horizon, additively time-
separable, quadratic objective:
00
- L,6t Ct (x)2.
t=O
Clearly, bliss is obtained at a consumption value of zero. (There is no free

disposal.) Henceforward the index x will be suppressed.
4 Converting to the frequency domain

Whiteman [39] showed that linear-quadratic optimization problems with
stationary driving processes of the kind stated above can be reformulated
as optimization problems in the frequency domain. The central observation
is that discounted utility can be translated as follows. Suppose Ct = C(L )et.
Then the following equivalence holds:
E, [~fl'c1l = E, [~fl'(C(L)e,)2l
~f,6tc;~~ 1 C(z)C(,6z-1) dz
t=O 27rZ lIzl=l z
The equivalence in (i) is driven by the fact that the expectation of the
complicated object (C(L)et}2 is radically simplified because
o k-::j:.·
E[et-ket-iJ ={1 k =~ .
498
The equivalence in (ii) is driven by a similar fact:
This treatment underscores the fact that the discounted utility and profit
objectives are simply norms of the infinite-dimensional vectors that char-
acterize the functions A(L), B(L), and so on. Therefore the following ad-
ditional equality holds:
~1 C(z)C(,Bz-I) dz (~) IIClI~,

27n lIzl=1 z
where the subscript "2" accounts for the fact that C can be thought of as an
element of the space L 2 • A demonstration of these equivalences is presented
in [39]. In any case, this norm and frequency-domain representation results
in notational and analytic convenience, because a variational approach to
solving the model can be used. Welfare calculations can then be solved by
relatively straightforward contour integration. This approach is used in the
remainder of the paper.
5 Trade and the price of risk under full information

There are three regimes to consider when there is full information. First, if
there is no exchange, welfare consists of the expected utility of consuming
endowment. In frequency-domain notation, this is3
Second, with full information, the idiosyncratic risk can be perfectly di-
versified since the endowments are mutually independent. Under full infor-
mation and no risk sharing across countries, but with internal risk sharing
that exploits this mutual independence, the idiosyncratic risk is perfectly
shared across individuals, but no insurance of the domestic aggregate risk
is possible. Welfare is then simply
reflecting the nondiversifiable component of endowment.

Finally, under cross-country risk sharing, a fraction Al of country 1 's ag-
gregate process is transferred to country 2 and a fraction A2 symmetrically
3 Here is the first instance showing how the linear-quadratic formulation allows the
additive separability of payoffs generated by idiosyncratic and aggregate stochastic en-
dowment processes. For a more general treatment of the separability of aggregate and
idiosyncratic components of endowment processes, see [4J and [5J.
499
is transferred from country 2 to country 1. The country-1 consumption

process is 4
(5.1)
and welfare is now
-(1- A1)211Bl112 _ (1 ~28)2 A~IIB2112

with a corresponding expression for country 2. The second term is the risk
sharing term; note how it is weighted by country size.
There must be some relationship between the elimination of risk from
shedding the domestic endowment process through A1, and the consequent
acceptance of risk from the foreign process via A2. It will be assumed that
this tradeoff is linear, and the assumption will be validated as an equilib-
rium below. The two countries choose their respective Ai under this linear
trade-off. Thus,
A2 = pAl. (5.2)
The objective for country 1 is then
2 2 (1 - 8)2 2 2)
max ( -(1 - A1) IIB111 - 2 PA111B211 .
Ai 8
The objective thus represents a portfolio problem in that the endowment
process can be divided into holdings of domestic and foreign assets. This
a straightforward portfolio problem in which means have been suppressed.
The price p can be thought of as the price of insuring the B1 process, with
the cost being increased holdings of the risky process B 2 •
The first-order condition is
2(1- A1)IIB l I1 2 _ 2(1 ~/)2 '\lp211B2112 = 0

for country 1, and symmetrically
2 82 2
2p(1 - pAl) IIB211 - 2 (1 _ 8)2 A111B111 =0
for country 2. These are two equations in the two unknowns p and Al. Note
that the second order condition is negative; if a price-of-risk equilibrium is
found, it will be stable. Solving yields
82 IIBl112
(5.3)
p = (1 - 8)2 IIB211 2'
4 Tesar [36] used this type of portfolio scheme in a time-domain formulation. The
portfolio weights, Ai, are static. The evolution of endowment is continuous, in essence
changing the wealth of each country; it is an attribute of this linear-quadratic formulation
that this changing wealth would not change the portfolio weights. As an example of how
this property works explicitly, see [29], appendix B.
500
Thus, the price of insuring risk is reduced by decreasing country size and
by decreasing domestic risk relative to foreign risk.
1
(5.4)
l+p
Also,
(5.5)
In equilibrium the following condition holds, showing that an internal so-

lution exists:
(5.6)
Intuitively, increasing A1 is good for country 1 since it transfers risk; this

formula for A1 is increasing in domestic risk IIBl112 as it should be, and
decreasing in foreign domestic risk. Thus, as domestic variance or size in-
creases relative to foreign variance and size, the amount of foreign risk that
must be borne per unit of domestic risk transferred rises. Looking at it
another way, the price of foreign endowment p can be thought of as the
price of insurance of domestic endowment, and is increasing in country size
and in domestic risk as well. (Note that both 0 < A1 < 1 and 0 < pAl < 1
in equilibrium.)
Welfare. The welfare of both countries in equilibrium can now be cal-
culated simply by substituting the equilibrium values of A1 and p into the
objectives. Thus, equilibrium welfare for country 1 is
-(1 - A1)211Bl112 _ (1 -26)2 A~IIB2112.

6
Using (5.3)-(5.6), this simplifies further to
--P-IIB 2
l+p l I1 ,
so welfare is a fraction of the (negative) pure autarky welfare. A small price

of insurance p translates into improved welfare, as it should intuitively. Still
more algebraic simplification yields
(5.7)
This shows that welfare is a generalization of the variance ratio of the sort
that appears in signal extraction problems. Observe that it is beneficial to
be a small country, or to be a low-variance country; this will be reversed
under private information.
The world portfolio. It is straightforward to see that in the full-
information llicardian equilibrium residents of both countries hold the same
501
portfolio composition, differing only in the size of the portfolio. Substituting

(5.6) into (5.1), the country-1 consumption process is
Correspondingly, the country-2 consumption process is
The consequence is that the aggregate consumption process in each country

is except for magnitude the same process. Thus at this theoretical level risk
sharing is maximized. But as will be noted in the discussion of the literature,
this is somewhat at odds with the facts.
6 Insurance under autarky and private information

Under autarky, there is a central organization in each country that real-
locates endowment across individuals. The reallocations are based on the
histories 5 of signals ( ... ,Sai,t_l' sait) and ( ... ,Sbi,t_l' Sbit ) that individuals
choose to send. The reallocations are assumed to be a stationary linear
mapping hi (L) (sait + Sb it ); the function hi (L) will be referred to as a filter.
Individuals choose their signal process with the knowledge of the structure
of the contract hi. In order to be feasible in the aggregate, the contract
must be such that the signal of the aggregate process is zero; hi can be
designed so that this is the case. The design is in fact simple: the contract
is such that each individual's signal is equivalent to their managing a stock
of credit in order to smooth consumption, with an interest rate structure
that is determined by the parameters of the stochastic process bt . 6
5If information about each individual's endowment process is private, such insurance
depends on successfully inducing individuals to truthfully reveal their endowment real-
izations to the contract. This revelation is accomplished via incentive constraints. One
approach would be for all individuals to announce the state of their endowment processes
each period. Individuals with endowments higher than the bliss point-the mirror image
of the current aggregate endowment-would average their endowments with individuals
below the bliss point. Truthful reports and some risk sharing could occur; indeed many
individuals could at.tain t.he bliss point.. However in that. instance there would always
be individuals wit.h a residual demand for insurance, and it. would be feasible to share
the risk int.ert.emporally at the cost of reducing t.he intraperiod exchanges. Intertem-
poral exchange requires that signals sent by individuals about t.heir states be tracked
intertemporally--indeed, such histories of signals offer richer possibilities for efficient
exchange-and this is t.he framework that is used here.
60ne can set. up such a contract. more formally as a constrained optimizat.ion problem
in the frequency domain, as demonst.rated in the appendix of [28]. The solution still rests
on finding the appropriate zeroes in filters, the method used here.
502
For the purpose of this paper, I will focus on the examples in which the
component processes are AR(l):
As demonstrated in the appendix of [28], discounting may be safely sup-

pressed in the primary analysis. In that case it is easy to show that the
contract must be the filter
The value of the insurance contract can then be calculated.

The autarkic equilibrium in each economy would be for each economy to
have an internal interest rate so that no attempt would be made to bor-
row or save the aggregate endowment processes, bit. In the corresponding
insurance contract representation, the insurance filters would have zeroes
preventing the insurance of the aggregate b-processes.
The operation of the contract. The objective of an individual in country
1 under autarky and a domestic contract is
00
- L:.Bt(Ylt - h1(L)st)2,
t=O
where St = salt + Sbw As I argued in [32], the cost of signalling can be

structured as an adjustment cost that leaves the intertemporal structure of
the contract and consumption processes unaffected; I use this as license to
suppress the cost of signalling from consideration.
In frequency-domain and operator notation, the objective is to solve
The separation of the idiosyncratic and aggregate terms occurs because of

their mutual independence of the innovations elt and Ult. The first element
is the difference between the idiosyncratic endowment process and the fil-
tered signal of the process. The second element is the difference between
endowment and filtered signals of the aggregate process.
Using standard results from [39], [32], the first order conditions are
SAl = h}l[(hi)-lhiAl]+
SB I = h}l[(hi)-lhiBl]+.
These can be solved explicitly in special cases.
An AR(l) example. Under autarky, the insurance filter will prevent
insurance of the common aggregate component, bt . Let the processes be
503
AR(l): Al(Z) = (1 - alz)-t, and Bl(z) = (1 - blz)-l. Utility under au-

tarky is then
As demonstrated in [31], the autarkic filter will therefore take on the non-
invertible form hl(z) = bl (l - b1l z), with the invertible factorization
hl{z) = (1 - blz). The utility value is then
(6.1)
and from Corollary AA, the value of insurance is
This must be compared to the utility under open international asset flows.
7 International risk sharing under private

information
Now let there be a common filter h(L) operating between the two countries.
The contract must meet feasibility requirements. There is a single h-filter
and contract because if there were two filters, arbitrage would be possible
in the sense that individuals could insure against one filter, then take the
residual and insure it against the other, and so on. The essential feature
of the common filter h is that unlike autarky, it need not suppress the
aggregate processes Bl and B2; instead, they can now be shared by a filter
that is not determined by their structure. In the ideal case as we shall
see, that structure is more like the structure in a model with no aggregate
process, as developed in [32J. That model behaves like one in which there is
an aggregate process that is a random walk. This shows that the essential
feature of the risk sharing contract is the mutual independence of the locally
uninsurable processes.
It should be emphasized that the aggregate processes remain observable
as in the autarkic model of the previous section. However, the type of an
individual in the sense of country is assumed unobservable. For this reason
no individual can promise a fraction of an aggregate process in exchange for
that of the other country. This in turn means that inappropriate signals of
aggregate processes must be suppressed just as in the autarkic model. (The
appendix treats the situation in which types are observable. In that situ-
ation, Ricardian trade of aggregate risk coexists with a private insurance
contract.)
504
Let h(z) = <p(I- <p-lZ), and h(z) = (1- <pz), 0 < 4i < 1. This can now
be used to calculate the signal. Under trade, the objective is
Note that the ffiter h is not subscripted by country, but is common to

both; the actual composition of the filter cannot be determined simply by
the Bi processes, and is left ambiguous at this stage. The resulting first
order conditions are
SAl = h-l[(h*)-lh* All+
SB l = h-l[(h*)-lh* Bll+·
Equilibrium resource conditions. Recalling that the relative sizes of the
domestic and foreign economies are 8 and 1 - 8 respectively, consump-
tion in the two countries must match the combined weighted endowment
processes, but the foreign risk sharing must be weighted by country size.
The equilibrium condition is then
h h* 1 - 8 h h* h h*
8[(Bl - X[ h* Bll+) + -8-X[ h* B2l+l + (1 - 8)[(B2 - X[ h* B2l+)
8 h h*
+-1J;.".[-:::-Bll+l
- uh h*
= 8Bl + (1- 8)B2'
and observe that it will automatically be satisfied. Thus, if the domestic
country is small, the term 16"8 ~[f.B2l+ will generate a large variance,
representing the excessive risk sharing undertaken by the small country.
An AR(J) example. Now assume that all endowment processes in both
countries are AR(I) as in the autarky analysis. From Corollary AA, the
autarkic value of insurance is
Under trade, we have
( <P-al)2 1
l-al<P l-a~+
( <p - b l)2 1
1 - bl<P 1 - b~
minus the risk sharing cost in the third term, and where <p is defined as
above. The third term in utility can be calculated more directly as follows:
505
q; - b2 1 112 ( q; - ~ ) 2 1
= - 11 1- b2q; 1- b2z = - 1- b2q; 1 - ~ ,
which has the same form as the insured terms. This can now be used to
calculate the welfare gain for country 1 relative to no contract: continuing
with the AR(l) example, it is
( 1q;-- alalq; )211 Al 112 + ( 1q;-- bb1q; )211 Bl112 _ (1-/)2 ( q; - b2 )21IB2112,
8 1- b2q;
1
(7.1)
with a symmetric value for country 2. The result should be compared with
the welfare value in (6.1): there is a potential change in the first term, the
value of the insured idiosyncratic component of endowment, since q; need
not equal b1 . The attempt to share risk internationally thus spills over
into the domestic risk-sharing arrangement! There is now insurance of the
formerly uninsurable aggregate process bIt as reflected in the second term,
but there is an additional burden of risk from sharing the foreign country's
aggregate process as reflected in the third term.
An essential feature of this expression and its symmetric foreign coun-
terpart is that for some parameters it may be impossible to simultaneously
improve the welfare of both countries by opening to trade! This is because
the third term, the additional utility burden of sharing the foreign risk, in-
creases without limit as the relative size of the domestic country, 8, shrinks:
a small country can be forced by the international contract to bear too
much of the foreign risk. 7 Thus, exactly contrary to standard trade theory,
small countries might rationally wish to avoid trade. This is not a standard
rent-seeking explanation for the distortion of trade; all the individuals in
a small country could agree that trade is undesirable. This conclusion is
special for certain parametric values and can be reversed for others, but
even in those cases it can be shown that trade is diminished and distorted,
if not closed off. 8
The contract thus fails to adjust a price of risk in response to country size,
unlike the full-information case. The reason is that the contract, in order to
be incentive compatible, must use the intertemporaZ pattern of signals from
individuals about their states. This forces the insurance of idiosyncratic and
aggregate shocks to be coupled under trade. The contract cannot rely on
individuals to report the relative weight of their endowment processes in
7 One might reasonably ask how individuals are forced to bear foreign risk in reality.
With asset equivalence including equivalence to currency equilibria, the interpretation is
that countries sharing a common currency automatically share risk through fluctuations
in the price level.
BBoyd and Smith (6), using an embellished version of the Diamond [8] overlapping
generations model, also find that standard predictions of trade models can be thrown
off by informational asymmetries. Their result, that investment can flow perversely from
poor to rich countries, similarly hinges on the interaction between domestic differential
information and international exchange.
506
an incentive-compatible fashion. If this were permitted, individuals would

report an inverse of their true size, and a false report of their true state so
as to eliminate all risk, which is not feasible in the aggregate.
8 Identical but independent aggregate and

idiosyncratic processes
While the results above demonstrate the extreme case in which trade is
blocked by one country, it is also possible for welfare-improving trade to
occur, but with its structure and welfare gains constrained by the private-
information contract. In order to investigate this situation the asymmetry
in aggregate processes will be removed in this section. Each country's aggre-
gate process is identical to the other's in intertemporal structure, but with
independent innovations. If the international contract does not insure the
aggregate processes, then trade is irrelevant. Their mutual independence
allows risk sharing of the aggregate processes, however, but this requires
deviating from the autarkic contract, and again the excessive aggregate for-
eign risk can spoil the gains for the small country. Nevertheless the small
country will prefer trade that is a slight modification of its autarkic con-
tract, as some diversification can be achieved this way.
Some intuition stems from considering how an ideal contract would work
if there were many small countries so that all aggregate risk could be mu-
tually diversified. This is no different from the situation in which many
massless individuals mutually diversify risk. It is known from [28] that in
that instance the most efficient contract induces the maximum amount
of persistence of consumption. 9 The two country model allows some ten-
dency of the contract to increase persistence, constrained by the excessive
aggregate risk bearing effect of the small country. This effect cannot be
neglected as it can in the case of a continuum of small agents or countries,
and therefore the contract "chokes" before going all the way to full persis-
tence. However, some sharing of the aggregate risk can take place if there
is enough symmetry.
In order to illustrate this point, let the two countries have identical AR(l)
at and bt processes: Al = A2 = A, and BI = B2 = B, and correspondingly
al = a2 = a, bl = b2 = b, but maintain the independence assumption.
Insurance will be motivated solely by independence in that case, rather than
by differences in the aggregate processes. Then the objective in country 1
is
_(1- (~(al))2) IIAI12 _ (1- (~(bt})2) IIBI12

h(at) h(b l )
9 As discussed in that paper, this extracts the maximum amount of usable information
about endowments.
507
The second and third terms have the same form, so they can be combined
If 8 = 1/2, so that the countries are equally sized, the third term, which
is the gain from insurance of the aggregate processes, becomes zero. But
if 8 -# 1/2, then the gain from the third term must be negative for the
smaller country, as it is bearing excess risk. The only way to shut down
this negative gain is to set cjJ = b, eliminating all gain from international
risk sharing. Because gains potentially exist in the first term, arising from
improvements in the trade contract's treatment of idiosyncratic processes,
the smaller country might still preferred a limited degree of trade.
Welfare. Is there an efficient value of cjJ? It would exist ifthe payoff in (8.1)
were simultaneously maximized for some value of cjJ. The first derivative of
(8.1) with respect to cjJ is the expression
2 cjJ - a 28 - 1 cjJ - b
(8.2)
(1 - acjJ)3 + 2---;52 (1 - bcjJ)3·
As long as the country is large (8 > 1/2), the second derivative of this
expression is positive; therefore there is no internal stationary point. The
optimal cjJ is then a corner solution at cjJ = 1 or cjJ = -1. The interpretation
is that it is optimal to have as much persistence as possible in the contract.
A highly persistent consumption contract can develop even with just two
individuals as long as their innovations are independent.
However, for the small country (8 < 1/2), it might not be optimal to
bear this much risk; it will prefer an interior contract with less persistence.
There is thus an asymmetry in the stability properties of the contract. But
it is not the case that both countries will prefer an interior solution.lO Even
for a small country, some risk sharing can be valuable; the marginal value
lOIf there are multiple countries the impetus is toward the persistent contract, because
the risk sharing cost shrinks. Proof: Let there be N identical countries with independent
aggregate processes. The excess risk shared is ~ x ~. This shrinks with N. Thus each
small country sees the cluster of other countries as "small" because the risk is divided
evenly.
508
of changing f/; can is nonzero at the autarkic value of f/; = b if b i- a. In that

instance (8.2) becomes
f/;-a
2 (1 _ af/;)3 i- O.
The gains from deviating from autarky can therefore be positive even for
a small country.
As in the standard theory of trade, the largeness of the countries leaves
the equilibrium concept that selects a specific value of f/; unresolved aside
from the issue of Pareto efficiency. Regardless of the equilibrium concept,
it is clear that in this special case, trade will have positive value for both
countries. It is also clear that the serial correlation of consumption will be
inefficiently nonpersistent if there is a large country effect. Finally, the pat-
tern of idiosyncratic consumption will be strongly affected by risk sharing.
Under autarky, idiosyncratic consumption is ruled by the uninsurable ag-
gregate; under trade, there is no aggregate and this can make consumption
highly persistent. Thus, while trade did not affect the complete insurance
available under autarky, here trade amplifies the direct effects on the ag-
gregates by affecting the insurability of the idiosyncratic processes.
9 Welfare comparisons
The formulas for welfare when there is international risk sharing under full
information (5.7) and under private information (8.1) can now be com-
pared. Restating them,
(5.7)
(8.1)
One difference is obvious: under full information, there is perfect risk shar-
ing of the idiosyncratic shocks, which is not the case under private informa-
tion. (This leads to amplification of the aggregate effects.) The difference
in the aggregate shock terms is harder to state. But it is easier to see the
potentially large difference by considering the extreme case of equally sized
countries; in that case the term (28 - 1)/8 vanishes, leaving the uninsured
residualllBll12; the private information risk-sharing contract then provides
no insurance at all for the aggregate component, in contrast to the full
information case.
Suppose now that 8 < 1/2. Then the small country won't want to maxi-
mize f/; because 28 - 1 < 0 and there is a risk sharing burden, so that the
welfare gain for the smaller country is negative. Nevertheless it is possible
that the beneficial effects on the idiosyncratic insurance component of wel-
fare is such that this effect is small relative to the benefits and the efficient,
509
high-4> contract will be chosen. But as 8 shrinks, the risk sharing burden
eventually becomes too high and the high-4> contract will be refused; the
benefits of trade under private information shrink as country size shrinks,
a reversal of the full information case. In the real world, this refusal will
look like deliberate isolation and abstinence from international trade, but
it is instead an avoidance of risk. This isolation would not occur under full
information, recalling from the analysis of that case above that the price
of risk would appropriately adjust in that instance. Thus, the presence of
private information, even if it is only about idiosyncratic processes, can
seriously impinge on standard notions of efficient behavior. But given the
apparent necessity of private information in realistic models of assets, this
seems unavoidable.
10 Discussion
All these private information results do depend on trade mattering in the
sense of differences in endowment processes; if the aggregate endowment
processes are not independent, there would be no gains from trade possible
in the private information setting. Note also that it is the privacy of the
idiosyncratic at process, not the aggregate bt process, that constrains the
wide characteristics of the private information contract, even though that
risk is not shared internationally. Thus "local conditions" restrict aggregate
interactions.
Despite the negative normative implications of the model, it is attractive
as a positive model. Countries, especially small ones, do in fact close them-
selves off from each other. An empirical expression of this closure would be
unexpectedly weak inter-country correlation of consumption, precisely as
observed by Backus, Kehoe and Kydland [3J. More recently, Athanasoulis
and van Wincoop [1] present an empirical case that there are substantial
unrealized gains from international risk sharing. More than that, countries
aggressively wall off their currencies and their credit systems from inter-
national interaction. Because of the asset equivalence of the model, this
behavior is rationalized as a positive matter.
It is well known that risk can alter the standard implications of standard
models of international trade. Recent cases in point include Hoff [15J, who
finds that uncertainty can reverse the Heckscher-Ohlin model's implication
that trade in goods alone equalizes factor prices. Tesar [35] shows that
nontraded goods diverts domestic portfolio holdings, damping international
risk sharing. Feeney and Jones [9] also study integration of state-contingent
claims as risk-sharing into a real model with traded and nontraded goods.
They show that the type and magnitude of risk aversion can cause less
rather than more smoothing of consumption across states when trade is
opened but there are nontradeable goods. A rough interpretation of the
results here is that private endowment processes are nontradeables and so
dampen exchange in the spirit of these prior models.
510
Cole and Obstfeld [7] model international risk sharing and show that it
might have few gains once goods arbitrage occurs. As in Lucas's model,
there is a representative agent in each country: as Imrohoroglu [16] demon-
strates, omitting effects from risk-sharing arising from the heterogeneity of
individuals can significantly distort welfare calculations; the spillover effect
in the scarce-information model of trade developed here suggests that this
point is relevant here.
That private information interferes with international exchange has been
noticed by Obstfeld and Rogoff [22, pp. 401-407] in a two-period model.
Aside from the fact that the model here entails repetition, Obstfeld and
Rogoff model aggregates as being private, an assumption they themselves
question because of the two-period restriction [22, p.402, footnote 54],
whereas here it is the privacy of information at the individual level that
drives the results.
11 Conclusion
The standard idea of comparative advantage extends to the abstract do-
main of risk, even in dynamic situations. With the addition of private in-
formation, the idea can break down. This would be an intellectual curiosity
were it not for the fact that assets-money and bonds-seem inescapably to
be generated by private information. Such asset structures abound in real
economies, suggesting that private information permeates them. While the
breakdown of the idea of comparative advantage is lamentable, the seeds of
a theory that explains the empirical incompleteness of trade in risk emerge
from the model. Moreover, the theory suggests the reason why markets
tend to involve hierarchical agglomerations of traders and economies: there
are critical masses of traders whose endowment processes possess enough
statistical independence to generate diversification gains that outweigh the
losses incurred from spillover effects on domestic idiosyncratic insurance.
Appendix A
Lemma A.I: Suppose f is such that f f* = 1. Then
Proof.
IIA - f[f* A]+112 = ~

2~z
f (A - f[f* A]+)(A* - f*[f* A]*) dz
+ z
= IIAI12 - ~
2~z
ffA*[f*A] dz
+ z
- ~
2~z
ff*[f*A]* (A -
+
J[f*A]+) dz
Z
511
= IIAI12 - 2~i f f A*[/* Al+ d: - 2~i f [/* Al~(f* A - [/* Al+) d:.
where the last step follows from the hypothesis about f. The factor in the
last integrand, (f* A - [/* Al+), has only strictly negative powers of z; the
other factor, [f* Al+, has only nonpositive powers. The integrand therefore
has only strictly negative powers of z and the integral is zero. _
Observe that f == hh- 1 satisfies the hypothesis. The insurance compo-
nent of the idiosyncratic part of utility is therefore
< hh- 1A *, [h*(h*)-l Al+ > .
and the higher this quantity, the better is the value of insurance. This
quantity can be calculated for each element of the objective. Another lemma
is useful at this point.
Lemma A.2: Let A(z) = (1- az)-t, B(z) = (1- bzt1. Then
< f*A, [g*Bl+ >= f(a)g(b)(A, B) =< f,A >< g,B >< A,B >.
Proof. This is a straightforward application of [34], Lemma C.3. _

Corollary A.3: Let A(z) = (1 - az)-l. Then
< f* A, [/* A]+ >= f(a)211AII2.
Proof. Immediate. _
Corollary A.4: Let A( z) = (1 - az) -1. Then the value of insurance is
< f* A, [/* A]+ >= f(a)211AII2.
Proof. The value of insurance is

-IIA - J[/*A]+II 2 + IIAII2 = -IIAII2+ < fAo, [/*A]+ > +IIAII2
=< fAo, [/*A]+ >= f(a)211AII2._
The following lemma is useful in the derivations of section 7.
Lemma A.5: Let A(z) = (1 - az)-l, h(z) = 4>(1 - 4>- l z), and h(z) =
1- 4>z. Then
[ h(Z-l) A z] _ 4>-a _1_ i (1)

h( Z-l ) () + - 1 - a4> 1 - az ( )
and
A(z) _ ~(z) [~(Z-l) A(Z)] = 1 - 4>2 _l_(ii) (2)

h(z) h(Z-l) + 1-a4>1-4>z
Proof. Result (ii) follows from algebra and result (i). Result (i) follows
from direct evaluation of the annihilate. _
512
Appendix B
Observable types in a private-information setting. The purpose of this ap-
pendix is to explore the idea of full-information exchange of aggregate
processes while retaining the private-information structure of the idiosyn-
cratic processes. This can be interpreted as individuals' national types being
observable, even if their endowment processes remain private. The aggre-
gate process from each country is then observable and it is feasible for an
individual to trade aggregate endowment in the full information sense, and
yet still need to use the filter to signal and insure idiosyncratic endowment.
As was shown in the full-information case, both countries will hold port-
folios resulting in a common endowment process, and the insurance filter
will therefore be based on that common process.
The common consumption process is defined by the factorization of
An important feature of the solution is that if B 1 (L)ult and B2(L)U2t are

both AR(l) processes with different AR parameters, then the consump-
tion process will not be AR(l). Rather, it will be AR(2), MA (1). This
complicates any analysis in which the aggregate consumption process must
interact with an incentive compatible filter. In particular, the MA compo-
nent induces a lag in the world insurance filter, and the filter must have two
zeroes in order to suppress signalling of the two AR components. Thus the
world ffiter will be significantly more restrictive than the domestic filters,
and the no-trade result is still a possibility.
Analytically, this approach means that the the welfare analysis of the
full-information model can be recycled into the private-information model.
Let us assume that each individual's country of residence is observable; in
that case, each resident can credibly promise an endowment stream equiv-
alent to the domestic aggregate endowment process to an international
market. Equivalently, the central insurance mechanism in each country can
average the endowments reported to it and act as an agent in international
exchange of that process. Each individual in each country then has the
world endowment process as determined by the full-information portfolio
model.
The world filter suppresses signalling and insurance of this aggregate
world consumption process, and therefore the gains from trade of the ag-
gregate processes duplicate those of the full information model. Ricardian
trade in aggregate risk then coexists with insurance of domestic idiosyn-
cratic processes via a filter; this is roughly like equity (in the foreign aggre-
gate process) and debt (arising from signalling and insurance of idiosyn-
cratic endowment processes) coexisting in more conventional settings. In
this instance, the portfolio diversification gain from trading the aggregate
process is unambiguously positive as was established above, and there is
513
therefore no loss from bearing excess foreign aggregate risk. But spillover
effects on domestic idiosyncratic insurance are still possible and have the
potential to dominate the portfolio gains; it is this possibility that will now
be explored.
The analytical burden shifts to (i) solving for the world consumption
process, (ii) writing down the consequent world h-filter, and (iii) calculating
the idiosyncratic welfare component in each country. Since the world filter
is more restrictive, we know that the idiosyncratic welfare component will
fall; we can compare it with the gain in the aggregate component. (Note
also that since the world aggregate consumption process in the standard
analytical example has both AR components, the filter will also suppress
reporting of each domestic aggregate endowment process taken separately.)
We can use Corollary AA to check the effect on the insurance gain under
the more complicated world filter. The world filter is
where the z term is present to suppress the MA component of the world

consumption process. It is then easy to see that
Thus, we can substitute
into the formula of Corollary AA. This yields the idiosyncratic insurance
gain of
2 ( b2 - a ) 2 ( b1 - a ) 2 1
a I-b 2a I-b 1a l-a 2 .
The no-trade insurance gain, in which the domestic insurance filter is 1 -
bI 1 z yields a gain of
( bl - a)2 1
"f = 1 - b1a 1 - a2 ·
The gain under trade is thus a multiple of this,
2(b2 -a)2
a 1 _ b2 a "f.
It is immediately apparent that this gain can be a fraction of the no-trade

gain, and even that the gain can be eliminated entirely-this happens if
~ = a. Thus, the Ricardian gains in the risk domain that arise from opening
to trade can be offset or even overbalanced by the losses incurred from
reducing the insurance value of the filter.
514
The gain from shedding risk on the aggregate process in llicardian fash-
ion can be obtained from the difference between no-trade utility of the
aggregate and expression (5.7)
For AR(l) processes this reduces to
(1 - 8)2(1 - b1)-1(1 - b~)-l (1- 8)2

2 2
8 (1 - bi)-l + (1 - 8)2(1 - b~)-l - 8 (1 - b~) + (1 - 8)2(1 - b~)'
Using AR(l) processes this can be calculated and added to the (poten-
tially negative) gain for the idiosyncratic insurance term. The net gain from
the idiosyncratic and llicardian components is
We can immediately and easily examine the case in which all insurance is
lost by opening to trade, namely if ~ = a. In that instance opening to trade
adds a zero a to the insurance filter, eliminating idiosyncratic insurance.
Setting b2 = a, the gain is then positive if
This inequality can be violated for 8 sufficiently close to unity if b1 =I a (and

therefore b1 =I b2 )-the opposite of the previous finding. Thus large coun-
tries that enter trade can be hurt if the trading partner has an aggregate
process with different persistence. Since it is large countries that are hurt
due to the presence of small countries whose signals must be suppressed,
this result resembles standard adverse selection.
Under alternative parameterizations small countries can be hurt by trade
as well. In the extreme case 8 = 0, and the gain from initiating trade is
(a (1b- ~aa )2)

2 2 -
-
(1 - b1a
1
a
)2 1 - l
b1 l
-
a + 1 - 11 ' 2
which can be negative if b2 = a; this is most easily seen if a ;:::::: 1 and b1 ;:::::: O.
In this instance it is the spillover effect that dominates the (negative) gains:
while insurance of the aggregate process is improved by opening to trade,
515
insurance of the idiosyncratic process is choked off by the incompatibility of

the foreign aggregate process with insurance of the domestic idiosyncratic
process. In general then there is no guarantee that trade will occur.
A third characterization of the results is that trade will occur if the
spillover effects are guaranteed to be small due to the lack or insignificance
of domestic insurance. In the extreme case there is no domestic insurance
possible when b1 = a; even if l>2, = b1 = b there is no spillover effect and
there is automatically a gain of
(1 - 8)2 1
82 + (1 - 8)2 (1 - bY)
from risk-sharing of the aggregate processes, which is positive.
Observe that the common portfolio attainable through the ful1-
information Ricardian exchange is not attainable through the insurance
filter. The insurance filter suppresses any insurance of the domestic aggre-
gate process, and therefore does not itself generate risk-sharing with the
trading partner. In this sense the Ricardian and insurance exchange might
be viewed as separable, in the sense of allowing equity and debt assets to
coexist. But again this interpretation depends crucially on the notion that
the aggregates can be treated as observable.
References
[1] Athanasoulis. Stefano, and Eric van Wincoop (1996), "Growth Uncer-
tainty and Risksharing," working paper, Iowa State University.
[2] Atkeson, Andrew and Robert E. Lucas Jr., (1992), "On Efficient Dis-
tribution with Private Information," Review of Economic Studies 59,
427-453.
[3] Backus, David, Patrick Kehoe, and Finn Kydland (1992), "International
Real Business Cycles," Journal of Political Economy 100, 745-75.
[4] Bergin, James, and Dan Bernhardt (1992), "Anonymous Sequential
Games with Aggregate Uncertainty," Journal of Mathematical Eco-
nomics 2, 543-562.
[5] Bergin, James, and Dan Bernhardt (1995), "Anonymous Sequential
Games: Existence and Characterization of Equilibria," Economic Theory
2, 543-562.
[6] Boyd, John, and Bruce Smith (1995), "Capital Market Imperfections,
International Credit Markets, and Nonconvergence," working paper, Fed-
eral Reserve Bank of Minneapolis.
[7] Cole, Harold, and Maurice Obstfeld (1991), "Commodity 1fade and
International Risk Sharing," Journal of Monetary Economics 28,3-24.
516
[8J Diamond, Peter (1965), "National Debt in a Neoclassical Growth

Model," American Economic Review 55, 1126-1150.
[9J Feeney, JoAnne, and Ronald Jones (1994), "Risk Aversion and Inter-
national Markets: Does Asset Trade Smooth Real Income?," Review of
International Economics 2(1), 13-26.
[1OJ Feldman, Mark, and Christian Gilles (1985), "An Expository Note on
Individual Risk without Aggregate Uncertainty," Journal of Economic
Theory 35, 26-32.
[l1J Green, Edward J. (1987), "Lending and the Smoothing of Uninsur-

able Income," in Contractual Arrangements for Intertemporal Trade,
(E. Prescott and N. Wallace, eds.), Mineapolis: University of Minnesota
Press.
[12J Green, Edward and Soo Nam Oh (1991), "Contracts, Constraints and
Consumption," Review of Economic Studies 58, 883-899.
[13J Hansen, L. and T. Sargent (1980), "Formulating and Estimating Dy-

namic Linear Rational Expectations Models," Journal of Economic Dy-
namics and Control, Vol. 2.
[14J Hansen, L. and T. Sargent (1981), "Linear Rational Expectations

Models for Dynamically Interrelated Variables," in Rational Expecta-
tions and Econometric Practice, Vol. 1, Robert E. Lucas, Jr. and Thomas
Sargent, eds. Minneapolis: University of Minnesota Press.
[15J Hoff, Karla (1994), "A Reexamination of the Neoclassical Trade Model
under Uncertainty," Journal of International Economics 36(1), 1-28.
[16J Imrohoroglu, Ay§e (1992), "The Welfare Cost of Inflation under Im-
perfect Insurance," Journal of Economic Dynamics and Control 16, no.
1 (January), pp. 79-91.
[17J Judd, Kenneth, "The Law of Large Numbers with a Continuum of IID
Random Variables." Journal of Economic Theory 35 (1985), 19-25.
[18J Kocherlakota, Narayana (1994), "Implications of Efficient Risk Shar-
ing without Commitment," Review of Economic Studies 63(4), 595-610.
[19J Lucas, Robert E., Jr. (1992), "On Efficiency and Distribution," The
Economic Journal 102, 233-47.
[20J Lucas, Robert E., Jr. (1982), "Interest Rates and Currency Prices in a
Two-Country World," Journal of Monetary Economics 10 pp. 335-360.
[21J Marcet, Albert and Ramon Marimon (1992), "Communication, Com-
mitment and Growth," Journal of Economic Theory 58, 219-49.
517
[22] Obstfeld, Maurice, and Kenneth Rogoff (1996), Foundations of Inter-

national Macroeconomics. Cambridge: MIT Press.
[23] Phelan, Christopher (1995), "Repeated Moral Hazard and One-sided

Commitment," Journal of Economic Theory 66, 488-506.
[24] Phelan, Christopher (1994), "Incentives and Aggregate Shocks," The

Review of Economic Studies 61, 681-700.
[25] Phelan, Christopher and Robert M. Townsend (1991), "Computing

the Optimal Insurance-Incentive Tradeoff for Multiperiod, Information-
Constrained Economies," Review of Economic Studies 58, 853-881.
[26] Rothschild, Michael, and Joseph Stiglitz (1976), "Equilibrium in Com-

petitive Insurance Markets: An Essay on the Economics of Imperfect
Information," Quarterly Journal of Economics XC, no. 4 (November),
629-650.
[27] Spear, Stephen, and Sanjay Srivastava (1987), "On Repeated Moral
Hazard with Discounting," The Review of Economic Studies 54, 599-617.
[28] Taub, B. (1997), "Optimal Policy in a Model of Endogenous Fluc-

tuations and Assets," Journal of Economic Dynamics and Control 21,
1669-1697.
[29] Taub, B. (1997), "Dynamic Agency with Feedback," Rand Journal of

Economics 28(3), 515-543.
[30] Taub, B. (1994), "Currency and Credit Are Equivalent Mechanisms,"

International Economic Review 35, 921-56.
[31] Taub, B. (1994), "Aggregate Fluctuations, Interest Rates and Re-

peated Insurance under Private Information," Journal of Economic Dy-
namics and Control 18, 433-465.
[32] Taub, B. (1990), "The Equivalence of Lending Equilibria and

Signalling-Based Insurance under Asymmetric Information," Rand Jour-
nal of Economics 21(3), 388-408.
[33] Taub, B. (1989), "Dynamic Consistency of Insurance Contracts Under

Enforcement by Exclusion," Journal of Economic Dynamics and Control
13,93-112.
[34] Taub, B. (1986), "The Tradeoff between Social Insurance and Aggre-
gate Fluctuations," Information Economics and Policy 2(4), 259-276.
[35] Tesar, Linda (1993), "International Risksharing and Nontraded

Goods," Journal of International Economics 35, 69-89.
518
[36] Tesar, Linda (1995), "Evaluating the Gains from International Risk-
sharing," Carnegie-Rochester Conference Series on Public Policy 42, 95-
143.
[37] Thomas, Jonathan and Tim Worrall (1990), "Income Fluctuation and
Asymmetric Information: An Example of a Repeated Principal-Agent
Problem," Journal of Economic Theory 51, 367-390.
[38] Wang, Cheng (1995), "Dynamic Insurance with Private Information
and Balanced Budgets," The Review of Economic Studies 62, 577-95.
[39] Whiteman, Charles (1985), "Spectral Utility, Wiener-Hopf Tech-
niques, and Rational Expectations," Journal of Economic Dynamics and
Control 9, 225-240.
Remark on extended price equilibria
Rabee Tourky
Department of Economics, University of Melbourne, Parkville, Victoria 3052,

Australia and School of Business, La Trobe University, Bundoora, Victoria
3083, Australia
Abstract. Economies with infinitely many commodities fail to pass the

standard equilibrium existence test with linear price systems. On the other
hand, Peleg and Yaari [25], Aliprantis, Brown, and Burkinshaw [2], and
Araujo and Monteiro [4] show that the existence of equilibrium in economies
with infinitely many commodities is obtainable, under standard assump-
tions, with extended price systems (functions to the extended real line).
All, however, assume that preferences are complete preorderings and satisfy
some strict monotonicity assumption. We show that the same conclusion
holds when preferences are neither monotone nor transitive. Our commod-
ity spaces are Lp, 1 :::; p :::; 00, and price systems are measurable functions.
We also consider the space ca(8,~) (finite measures on (8, CS)) and extend
the results to economies with commodity differentiation. Finally, we show
that under the assumptions of Aliprantis, Brown, and Burkinshaw [2] their
equilibrium extended price systems coincide with ours.
JEL Classification Number: D51

Keywords: Infinite commodities, extended price system
1 Introduction
The usual approach to the theory of value with infinitely many commodi-
ties is that suggested by Debreu [9] and Hurwicz [13] in which the linear
function concept is a formalization of price system (see also Radner [26]).
Generally, however, economies with infinitely many commodities fail to
pass the standard existence test if we require that price systems be of the
Debreu-Hurwicz type (see [15, 27, 32]).
The conventional solution to non-existence in the Debreu-Hurwicz
framework is to assume that preferences satisfy a cone condition, which
Mas-Colell [20] called uniform properness (see also [1, 3, 21, 27, 30]). De-
spite the attractiveness of these results and related generalizations, the
inability to prove existence under the standard finite dimensional assump-
tions seems to suggest an inherent inadequacy in the Debreu-Hurwicz
framework.
520
An alternative framework is provided in the work of Malinvaud [18, 19],

Peleg and Yaari [24, 25], and Peleg [23]. A discernible theme of these writers
is that the one to one correspondence between price systems and linear
functions is not primitive to economic analysis. What is essential is that
price systems retain the interpretation of being lists of prices and value be
calculated by a generalized method of summing the product of prices and
quantities. A strong argument in favor of this framework can be found in
Peleg [24].
Work pertaining to the Malinvaud-Peleg-Yaari framework has been com-
paratively scarce. This is surprising since meaningful existence theorems
have been obtained, under standard finite dimensional assumptions, within
this framework. The aim of this paper is to contribute to the research by
extending available results and, perhaps, further motivating the study of
economies within this alternative framework.
The classic equilibrium result within this framework is that of Peleg
and Yaari [25]. Peleg and Yaari establish the existence of equilibrium for
exchange economies embedded in the space of all real sequences. Their price
systems are positive real sequences and value is calculated as the sum of
the product of prices and quantities. Thus, some commodity bundles may
have infinite value at equilibrium.
An important extension of Peleg and Yaari's theorem was established by
Aliprantis, Brown, and Burkinshaw [2, Theorem 4.13]. They prove a related
result in topological vector lattices with quasi-interior points. Such spaces
include the Lebesgue spaces Lp, 1 :S p :S 00. Their extended price systems
are additive functions from the positive cone of the commodity space to the
extended real line and give finite value to the total endowment. It has been
suggested by Araujo and Monteiro [4, p. 409] that these extended price sys-
tems are not entirely satisfactory and need not be economically meaningful.
Presumably, Araujo and Monteiro, who deal with the commodity spaces
Lp, mean that these equilibrium extended price systems do not coincide
with an analytical form of a price system that retains the desired economic
interpretation. This issue deserves investigation since it is known that, at
equilibrium, Aliprantis, Brown, and Burkinshaw's price systems coincide
with Peleg and Yaari's price systems, if we assume the assumptions made
by Peleg and Yaari (see [2, p. 1135]).
In a more recent generalization of Peleg and Yaari's result, Araujo and
Monteiro [4] establish the existence of equilibrium in the spaces Lp, 1 :S
p:S 00. Their price systems are positive measurable real functions (see also
Bewley [6, p. 516]). They show that we can weaken the assumption that the
consumption set is the positive cone of the commodity space by assuming
that it is a 'proper set', as in Back [5], and that preferences are strictly
monotone in the direction of the properness constant.
The distinguishing feature of Aliprantis, Brown, and Burkinshaw [2],
Araujo and Monteiro [4], and Peleg and Yaari [25] is that preferences are
not assumed to be uniformly proper, or even pointwise proper. All these
521
results, however, assume that preferences are complete preorderings that

satisfy some strict monotonicity assumption. These assumptions are crucial
for the existence proofs of Aliprantis, Brown, and Burkinshaw [2] and Peleg
and Yaari [25] which are core equivalence arguments. Though transitivity
is apparently not crucial to the representation argument in Araujo and
Monteiro [4, Lemma 1], their special monotonicity assumption is used to
show that equilibrium price systems are countably additive set functions.
In this paper, we show that when the commodity space is Lp, 1 :s: p :s: 00,
the properties of local nonsatiation, continuity, convexity, and irreflexivity
of preferences are sufficient to guarantee the existence of equilibrium. These
conditions are among the weakest assumptions that guarantee the existence
of equilibrium in the finite dimensional exchange model. The space Loo
has been used in models of economies with uncertainty and economies in
which activity takes place over infinitely many intervals or continuously
over time (e.g., [6, 17, 26]). The space L2 has been used as a model in
finance economics (e.g., [7,8, 10]) and Ll was considered by many authors
(e.g., Ostroy [22]).
The usual commodity space for economies with commodity differentia-
tion is the space of finite Borel measures on a compact set, which does not
have a quasi-interior point. This space is considered by Aliprantis, Brown,
and Burkinshaw [2] and Araujo and Monteiro [4]. The results, however, are
not satisfactory since equilibrium is only shown on a truncated economy
and not the whole. We consider the space 00(8, ~), of count ably additive
measures on a a-field, ~, of subsets of 8. We show how with some logi-
cal restrictions on the consumption sets our theorems imply a meaningful
equilibrium existence result for economies with commodity differentiation.
Informally we place the following restriction on consumer demand:
Consumers do not demand a strictly positive amount of a com-
modity if the total initial endowment of the commodity is zero.
Finally, we show that, under the assumptions made in [2], equilibrium
extended price systems of Aliprantis, Brown, and Burkinshaw [2] coin-
cide with positive measurable functions. The core equivalence theorems of
Aliprantis, Brown, and Burkinshaw are very general, and all that is needed
to give them economic interpretation is to show that the price system na-
tion is economically meaningful.
Section 2 sets out the model and statements of results. Section 3 provides
an application of the results to economies with commodity differentiation.
The proofs are in Section 4. Our argument is different from that of Araujo
and Monteiro. Like Araujo and Monteiro we prove equilibrium existence
in a dense subspace, Aw (so do Aliprantis, Brown, and Burkinshaw [2]).
Araujo and Monteiro then construct an isomorphism between Aw and Loo,
and carry the analysis to that space. We, however, construct a sequence of
Loo subspaces in Aw and proceed with a limiting operation similar to the
standard proof of the theorem on the representation of Lp conjugates.
522
2 Models and results

The commodity set is a positive, u-finite, and complete measure space
(S, ~, j.L). A commodity bundle is a function, f, in Lp(S,~, j.L), 1 ~ p ~ 00,
such that f(s) ~ 0 j.L-a.e. There is a finite number m> 0 of consumers and
e~h consumer is indexed by i E {I, ... , m}.
Definition 1 An economy is a set (Lp, (Ki)~1' (Wi)~1' (Pi)~1)' where
(1) Vi, Ki is a non-empty subset of D; = U E Lp: f(s) ~ 0, j.L-a.e.}

and is called the ith consumer's consumption set;
(2) Vi, Wi is a point in Ki and is called the ith consumer's initial endow-
ment;
(3) Vi, Pi is a mapping from Ki into 2Ki, and is called the ith consumer's
strict preference map.
Let W = Z::1 Wi, this point is called the total endowment. A point in
n:1 Ki is called an allocation. A price system is a j.L-measurable function,
g, such that g(s) ~ 0 j.L-a.e. The v.alue of a commodity bundle, f, relative
to a price system, g, is Isg(s)f(s)j.L(ds).
Definition 2 An equilibrium is a pair {g, Ud}, where 9 is a price system
and Ud is an allocation, such that
(1) E:'l(!i(s)-Wi(S)) :::; 0 J.L-a.e. {fi} is called an attainable allocation;

(2) Vi, Isg(s)h(s)j.L(ds) > Isg(s)wi(s)j.L(ds) if hE PiUi);
(3) Vi, Isg(s)!i(s) ~ Isg(s)wi(s)j.L(ds) < +00.
The first theorem deals with economies in Lp, 1 ~ P < 00.
Theorem 1 Let (Lp, (Lt), (Wi), (Pi)) be an economy such that 1 ~ P < 00.
Assume that
(1) Vi, Wi(S) >0 j.L-a.e.;
(2) Vi, Vf ELt, f rt. PiU);

(3) Vi, VfELt, the sets PiU) and {f} U PiU) are convex;
(4) Vi, Vf E Lt, PiU) is norm open in Lt;
(5) Vi, Vf E Lt, Pi- 1U) = {g: f E Pi(g)} is u(Lp, L~) open in Lt;
(6) Vi, Vf E Lt, PiU) is not empty.
Then there is an equilibrium.

523
Remark 1 We can replace condition 1(3) by the following weaker condi-

tion:
(*) Vi, Vf E Lt, PiU) is convex and there exists h E PiU) and E > 0
such that h(s) ~ EW(S) jL-a.e., and a:h+(l-a:)f E PiU) fora: E (0,1].
Condition (*) is implied by the conditions in almost all the referenced

equilibrium existence results. It is, however, not implied by the conditions
in Toussaint [29j, where it is assumed that f E PiU). We can also weaken
the requirement that the consumption set be Lt by allowing for more general
consumption sets on the lines of the results in Back [5}.
Only minor changes to the statement of Theorem 1 are required when

the commodity space is Loo. We do not assume the strong assumption that
W is in the norm interior of Lt, (Bewley [6] assumes that w(s) 2 E > 0,
jL-a.e., see also [12, 16]). Denote by 7(Loo, L 1) the Mackey topology of the
duality (Loo, L1)'
Theorem 2 Let (Loo, (Lt,) , (Wi), (Pi)) be an economy. Assume that
(1) Vi, Wi(S) > 0 jL-a.e.;
(2) Vi, Vf ELt" f ~ PiU);
(3) Vi, Vf E Lt, the sets PiU) and {f} U PiU) are convex;
(4) Vi, Vf E Lt" PiU) is 7(Loo, L 1) open in Lt,;
(5) Vi, Vf ELt" Pi- 1U) = {g ELt,: f E Pi(g)} is a(Loo, L 1) open in

Lt,;
(6) Vi, Vf E Lt" PiU) is not empty.
Then there is an equilibrium.
A corollary of Theorem 1 can be obtained for the space ca(S, <;S), which
consists of all scalar valued set functions that are defined and are count ably
additive on <;S. The norm IIAII is the total variation of A. We extend, in the
obvious manner, Definition 1 to economies in ca(S, <;s).
Theorem 3 Let (ca, (ca+), (Pi)' (Wi)) be an economy. Assume that
(1) Vi, Wi i- 0, and VE E <;s, wi(E) > 0 if weE) > 0;

(2) Vi, VA E ca+, A ~ PiCA);
(3) Vi, VA E ca+, the sets PiCA) and {A} U PiCA) are convex;
(4) Vi, VA E ca+, PiCA) is norm open in ca+;

524
(5) Vi, V)" E ca+, Pi- l ()..) = p': ).. E Pi ()..)} is norm open in ca+ and the
complement of Pi- l ()..), relative to ca+, is convex;
(6) Vi, V)" E ca+, Pi ()..) is not empty.
Then there is an attainable allocation Pd and a positive measurable

function 9 such that fsg(s) ..i(ds) :::; fsg(s)wi(ds) < +00 for all i, and
fsg(s)"(ds) > f g(s)wi(ds) for all w-absolutely continuous).. E Pi()..i).
Finally, we investigate the price system notion of Aliprantis, Brown, and

Burkinshaw [2]. Let Aw = U~=ln[-w,w], where [.,.] denotes an order in-
terval. Denote the positive cone of Aw by At and the positive cone of Aw
by;e (when p = 00, the closure of Aw is in 7(Loo, Ll)). The gauge of
the set [-w, w] is a norm on Aw. Under this norm w is an interior point of
At. Since order intervals are bounded in Lp, 1 :::; p :::; 00, then this norm
topology, denoted p, on Aw is finer than the relative topology on Aw as
a subspace of Lp" It was observed by Aliprantis, Brown, and Burkinshaw
that when consumption sets are the positive cone of the commodity space
then issues concerning the economic optimality of an attainable allocation
depend only on the truncated economy in At.
Definition 3 Let (Lp, (Lt), (Pi), (Wi)), 1 :::; p :::; 00, be an economy. An
extended price system is a function, 7r, from;e into [0, +00], which sat-
isfies 7r(w) < +00, and V(f,g) E ::;e x::;e,7r(f + g) = 7r(f) + 7r(g).
An extended equilibrium is a pair {7r, {Id} such that (a) 7r is an extended
price system, (b) {li} is an attainable allocation, (c) Vi, Vh E Pi(!i) n;e
we have 7r(h) > 7r(Wi), and (d) Vi, 7r(fi) :::; 7r(Wi).
Remark 2 This is a modified version of [2, Definition 4.12], which is con-

cerned with topological vector lattices, complete preorderings, and a quasi-
is monotone, in the sense that °: :;

equilibrium concept. It was observed in [2] that an extended price system 7r
y :::; x implies 7r(y) :::; 7r(x), and .finite
on At. Furthermore, there is a linear functional, T, on A w, which is an
extension of 7r1 +.
Aw
Let t be a complete preordering on Lt. t is convex if O'.f + (1- O'.)h ~ f

for all f, h ~ f, and 0'. E (0,1). t is continuous if t is a closed set in
Lt x Lt· t is strictly monotone if h t= 0, h E Lt implies that f + h ~ f
for all f. The notation (., (.), (ti), (.)) shall mean (., (-), (Pi =~i)' (.)).
Theorem 4 Let (Lp, (Lt), (ti), (Wi)), 1 :::; p <

Assume that
°(p = 00), be an economy.
(1) w(s) > 0 j.t-a.e.;

(2) Vi, ti is norm continuous (7(Loo, Ld continuous);
525
(3) Vi, ti is convex;

(4) Vi, ti is strictly monotone.
If {-7r, {Id} is an extended equilibrium then there is a price system 9

such that {g, {Id} is an equilibrium and Isg(s)f(s)J-L(ds) = 7r(f) for all
f E A-:;.
Remark 3 If we use the definition of the extended price equilibrium func-
tion given in the second part of [2, Theorem 4.13J then 7r and 9 coincide
on all of Lt.
3 Commodity differentiation models

Let S be the set of commodities. A point s E S is a complete list of the
characteristics of the commodity in question. Let ~ be a a-field of subsets
of S. Consumers choose positive finite measures on (S, ~). Let K be the
collection of all such measures. Given a consumption bundle A E K, the
number A(E) is the total amount of all commodities with characteristics
in E that are available for consumption. Let the total endowment, w, be
an element of K. We restrict the consumption sets to elements, A, of K for
which
A(E) = 0 if w(E) = 0, E E ~.
This restriction has a natural economic interpretation. It says that no indi-
vidual can demand a positive amount of commodities with characteristics
in E if commodities with characteristics in E do not exist in any amount.
Thus, the consumption set is the positive cone of ca( S, ~, w), the space of
all finite positive w-absolutely continuous measures on (S, ~), and Theorem
3 implies the existence of equilibrium.
In contrast, Jones [14] assumes that S is a compact metric space and ~
is the Borel field of S. A price system is a real function that is positive,
bounded and ~-measurable. The consumption set is all Borel measures on
(S,~) and the value of a choice, A, with respect to a price system, g, is
defined by the formula Is g( s) A( ds).
4 Proofs
Most of the work is done in the proof of Theorem 1. Before proving the
theorems we need the following lemma.
Lemma 5 Under the hypotheses of Theorems 1 and 2 there is a positive
linear functional, T, on Aw and an allocation, {Ii}, such that
(1) 2::'1 (fi(S) - Wi(S)) :::; 0 J-L-a.e.;

(2) T(w) > 0;
526
Proof: The proof of the lemma follows easily from the methods in the
literature concerning such spaces (e.g., [6, 16, 32]). We prove it here for
completeness.
First note that At has W as an interior point in the topology p defined
above. Denote by A~ the topological dual of (Aw, p). The set
Ll = {q E A~: q(w) = 1, q 2:: O}
is convex and a(A w, A~) compact.

Let :F be the set of all finite dimensional subspaces of Aw which contain
each Wi. Direct :F by inclusion. Denote an element of:F by Ea. We know
from the many finite dimensional results (e.g., [28, p. 20]) that for each Ea
there is a positive linear form qa on Ea, and an allocation {It} for which
- 2::1 Ut(s) - Wi(S)) ~ 0 IL-a.e.;

- qa(W) = 1;
- Vi, Vh E PiUt) n Ea, qa(h) 2:: qa(Wi).
For each a let iia E A~ be the positive linear form on Aw that is an

extension of qa'
We know that the order interval [0, w] is a(Lp, L~) compact for 1 ~
p < 00 and a(LOCJ' L 1 ) compact for p = 00. Thus, for 1 ::; p < 00 there
is {T, Ud} E (Ll, a(Aw, A~)) X (Lp, a(Lp, L~))m that is a cluster point
of the net {iia, Ut}}. While for p = 00, the cluster point is {T, {Ji}} E
(Ll, a(Aw, A~)) x (LOCJ' a(LOCJ' Ll))m. Without loss of generality, we take
the net as converging.
Clearly
- 2::1 (Ji(s) - Wi(S)) ~ 0 IL-a.e.;

- T(w) = 1.
Suppose that 9 E Pi(Ji) and 9 E Aw. Then eventually 9 E Piun and
9 E Ea. Thus eventually iia(g) 2:: iia(wd and it must be the case that
T(g) 2:: T(Wi)' 0
Proof of 1: Let T and Ud be the positive linear function on Aw and

the allocation given in Lemma 5, respectively.
We may assume without loss of generality that W is strictly positive
everywhere. Define a sequence of sets {Dn} by Dn = {s: w(s) > 2~}' We
have Dn C D n+1, for each n and U~=lDn = S. Since the measure space is
a-finite there is an increasing sequence, {G n }, of elements of ~ such that
527
J1.(G n ) < +00 for each n and U~=lGn = B. Define the sequence, {En},
by En = Dn n Gn . Each En is an element of ~. Also, it is clear that
U~=lEn = B, that En C En+1, and that J1.(En) < +00 for all n. For each
n let XRn be the characteristic function of En. It is not difficult to see that
XEn E Aw for all n.
Let Loo(En) denote the set of all J1.-€ssentially bounded functions in
Lp(B,~, J1.) vanishing outside En. Since Loo(E) = U~lr[-XEn' XEJ, then
Loo(En) C Aw.
Let ~(En) be the restriction of ~ to En and J1. En be the restriction of
J1. to ~(En). Loo(En) is equivalent to Loo(En, ~(En), J1. EJ and the latter
space can be regarded as a subspace of Lp(B,~, J1.).
The restriction of T to each Loo (En' ~(En)' J1. En) is positive and thus con-
tinuous in the J1.-essential supremum norm topology of that space. For each
n define a set function An;::: 0 on ~(En) by An(F) = T(X F) for F E ~(En)'
This set function is in ba(En,~(En),f.LEJ, and T(J) = fEn f(S)An(ds) for
f E Loo(En).
Clearly, An = An+1 on elements of ~(En). For each n decompose An into
its unique purely finitely additive and count ably additive parts. That is,
An = A~ + A~ where the notation is the obvious one. From uniqueness we
get A~ = A~+l on elements of ~(En)'
For each n let gn be the Radon-Nikodym derivative of A~. Each gn is
a unique element of L1 (En' ~(En), f.LEJ and thus can be regarded as an
element of L1 (En), the space of all functions in L1 (B, ~,f.L) that vanish
outside En. Since each gn is unique then we must have gn(s) = gn+1(s) for
J1.-almost every sEEn. Thus, the limit g( s) = limn gn (s) exists f.L-a.e. Also,
g( s) = gn (s) for f.L-almost all sEEn. From the countable additivity of f.L we
see that g is J1.-measurable. We will show that {g, {Id} is an equilibrium.
We have the following statements
T(J) = le" f(S)An(ds) = len f(S)A~(ds)+ len f(S)A~(ds), f E Loo(En),
and fEn f(S)A~(ds) = fs f(s)g(s)f.L(ds) for all f E Loo(En).

Take an arbitrary i. Let hE Pi(Ji). Define the sequence {h n } by
n if h( s) > n and sEEn,

hn(s) ={ h(s) if h(s) :::; n and sEEn, (1)
o if s tt En
for each n.
Evidently, hn(s) ;::: 0 f.L-a.e., and hn E Loo(En). Also the sequence {h n }
is increasing and converges f.L-a.e. to h. For each n we may choose a subset
Fn of En such that A~(En \ Fn) = 0 and f.L(Fn) :::; 2~" Define the sequence
{h~} by
hin -- h nXEn\Fn' n= 1, ... (2)
528
The sequence h~ converges to h in 11 . 11. Thus, h~ is eventually in Pi(fi).

iE
We have n h~(s),x.:;(ds) = 0 for n = 1, .... Furthermore, we have
and T(h~) = Is h~(s)g(s)j.L(ds) 2: T(Wi) for all h~ E Pi(!i).

Because g(s) 2: 0 j.L-a.e., and hn(s) 2: h~(s) j.L-a.e., then
for all h~ E Pi(!i). Clearly, the sequence of positive j.L-measurable functions

{hng} is increasing and converges j.L-a.e. to hg. Thus, Is h(s)g(s)j.L(ds) 2:
T(Wi) for all hE Pi(!i), and all i.
Since T(w) > 0 then there is some j such that T(wj) > O. From this and
the fact that Pj (fj) is not empty we see that g is not a j.L-null function.
Since g is not a j.L-null function, the functions gWi do not vanish j.L-a.e.,
and Iswi(s)g(s)j.L(ds) > 0 for all i. Define the increasing positive sequence
{w~} converging j.L-a.e. to Wi as in (1). It is easy to see that, from T 2: 0,
,x.:; 2: 0 and ,x.~ 2: 0, the following equation holds:
Thus,
f w~(s)g(s) ~ T(Wi) < +00,

0<
isf wi(s)g(s)j.L(ds) = lim
n is
i = 1, ... , m.
Hence, Ish(s)g(s)j.L(ds) 2: Iswi(s)g(s)j.L(ds) < +00, for all h E Pi(!i), and

all i.
We show that the inequality in the statement above is strict. Consider
a function h E Pi(fi). Then for some a E (0,1) close enough to 1 we have
ah E Pi(!i). Because the j.L-integral of each Wig is strictly greater than 0
and finite, the inequality must be strict to avoid contradiction. We can now
restate the result
Is h(s)g(s)j.L(ds) > Is wi(s)g(s)j.L(ds) < +00, hE Pi(!i), i = 1, ... , m.

All that remains is to show that each !i lies within the budget con-
straint. For each i it is clear that !i(s)g(s) ~ w(s)g(s) j.L-a.e. Thus, fig is
in Ll (S, CS, j.L) for all i. Take an arbitrary i. Take the point h E Pi(!i). Let
{h n } be the net given in (1). For some n large enough hn E Pi(!i) and
hng E L 1 (S, CS, j.L). For each a E (0,1), ka; = ahn + (1 - a)fi E Pi(fi) and
ka;g E Ll (S, CS, j.L). If we let a approach 0 then ka;g converges to fig in the
529
norm of Ll(S,CS,P,) and fsfi(S)g(s)p,(ds) ~ fSWi(S)9(s)p,(ds). From the

positivity of 9 and the Lemma above it must be the case that
is fi(S)g(s)p,(ds) = is wi(s)g(s)p,(ds), i = 1, ... , m.

This proves the theorem. D
Proof of 2: The proof of Theorem 2 follows with little adjustment to the

proof of Theorem 1. We will only list the adjustments needed.
Instead of (1) define a sequence {h n } by
hn = hX En , n = 1, .... (3)
The sequence {h n } converges to h in T(Loo,Ll). Let {h n } be only that
part of the sequence eventually in Pi(fi). Whenever we refer to (1) in the
previous proof refer to (3) in this proof.
Instead of (2) for each n choose an E small enough and a corresponding
Fn c En such that P,(Fn) :S E, A~(En \ Fn) = 0 and define
(4)
Such a choice is possible by [11, Theorem 8, p. 292; Theorem 11, p. 294;
Theorem 1, p. 430] and [31, Theorem 1.19, p. 50].
The rest of the details follow immediately from the proof of Theorem 2.
D
Proof of 3: Observe that ca( S, CS, w) is equivalent to Ll (S, CS, w), by the
Radon-Nikodym isomorphism. We need to show that the image of the econ-
omy under this isomorphism satisfies the conditions in Theorem 1. The
consumption set is the positive cone of L 1 . The image of w is Xs ' and the
image of each Wi is strictly positive w-a.e. Also, the continuity of prefer-
ences is immediate. Thus, there exists an equilibrium for the economy in
Ll with a price system that is a positive measurable function.
The value of a bundle x E Lt under 9 is fsg(s)x(s)w(ds). Let x be the
image of some A E ca(S, CS,w)+ then fsg(S)A(ds) = fsg(s)x(s)w(ds). D
Proof of 4: Let T be the extension of 71"1 + to Aw as in Remark 2. Then

Aw
T is a positive linear function on Aw, and {T, {fi}} satisfies the conclusion
of Lemma 5 with the inequality in (3) being strict. Also, from [2, Theorem
4.10] T is order continuous on Aw. That is if Un} is an increasing sequence
in Aw converging to f E Aw then TUn) converges to TU).
We follow the procedure in the proof of 1. Let {En} be the sequence of
sets given in that proof. Since the restriction of T to each Loo(En) is order
continuous then each An is count ably additive. Consequently, each A~ = o.
Thus, for all f E Loo(En) we have fsg(s)f(s)p,(ds) = TU). Since T is
order continuous we also have TU) = fsg(s)f(s)p,(ds) for all f E Aw. D
530
References
[1] C. D. Aliprantis, "On the Mas-Colell-Richard equilibrium theorem", J.
Econ. Theory 74 (1997), no. 2,414-424.
[2] C. D. Aliprantis, D. J. Brown, and O. Burkinshaw, "Edgeworth equi-
libria", Econometrica 55 (1987), 1109-1137.
[3] A. P. Araujo and P. K. Monteiro, "Equilibrium without uniform condi-

tions", J. Econ. Theory 48 (1989), 416-427.
[4] A. P. Araujo and P. K. Monteiro, "The general existence of extended

price equilibria with infinitely many commodities", J. Econ. Theory 61
(1994), 408-416.
[5] K. Back, "Structure of consumption sets and existence of equilibrium

in infinite-dimensional spaces", J. Math. Econ. 17 (1988), 89-99.
[6] T. F. Bewley, "Existence of equilibria in economies with infinitely many

commodities", J. Econ. Theory 4 (1972), 514-540.
[7] G. Chamberlain and M. Rothschild, "Arbitrage, factors structure, and

mean variance analysis on large asset markets", Econometrica 51 (1983),
1281-1304.
[8] G. Chichilnisky and G. NI. Heal, "Competitive equilibria in Sobolev

spaces without bounds on short sales", J. Econ. Theory 59 (1993), 364-
384.
[9] G. Debreu, "Valuation equilibrium and Pareto optimum", Proc. Nat.

Acad. Sci. 40 (1954), 588-592.
[10] D. Duffie and C. Huang, "Implementing Arrow-Debreu equilibria by
continuous trading of a few long-lived securities", Econometrica 53
(1985), 1337-1356.
[11] N. Dunford and J. T. Schwartz, "Linear operators. part I: General

theory", Pure and Applied Mathematics, Interscience Publishers, INC.,
New York, 1958.
[12] M. Florenzano, "On the existence of equilibria in economies with an
infinite dimensional space", J. Math. Econ. 12 (1983), 207-220.
[13J L. Hurwicz, "Programming in linear spaces", Studies in Linear and
Non-Linear Programming (K. Arrow, L.Hurwicz, and H. Uzawa, eds.) ,
Stanford University Press, 1958.
[14] L. E. Jones, "A competitive model of commodity differentiation",
Econometrica 52 (1984), 507-530.
531
[15J L. E. Jones, "Special problems ansmg in the study of economies

with infinitely many commodities", in: Models of Economic Dynamics:
Springer-Verlag Lecture Notes in Economics and Mathematical Systems
264 (H. Sonnenschein, ed.), Springer-Verlag, Berlin-New York, 1986.
[16J M. A. Khan, "A remark on the existence of equilibria in markets with-
out ordered preferences and a Riesz space of commodities", J. 1\1ath.
Econ. 13 (1984), 165-169.
[17] M. Majumdar and R. Radner, Shadow prices for infinite growth pro-
grams, Techniques of Optimisation (A. V. Balakrishnan, ed.), American
Press, New York, 1972.
[18] E. Malinvaud, "Capital accumulation and efficient allocation of re-

sources", Econometrica 21 (1953), 233--268.
[19J E. Malinvaud, "Efficient capital accumulation: A corrigendum",

Econometrica 30 (1962), 570--573.
[20] A. Mas-Colell, "The price equilibrium existence problem in topological

vector lattices", Econometrica 54 (1986), 1039-1053.
[21] A. Mas-Colell and S. F. Richard, "A new approach to the existence of

equilibrium in vector lattices", J. Econ. Theory 53 (1991), I-I!.
[22J J. M. Ostroy, "On the existence of Walrasian equilibrium in large-

square economies", J. Math. Econ. 13 (1984), 143-164.
[23J B. Peleg, "Efficiency prices for optimal consumption plans: I1", Israel
Journal of Mathematics 9 (1971), 222-234.
[24] B. Peleg and M. E. Yaari, "Efficiency prices in infinite-dimensional
spaces", Int. Econ. Rev. 11 (1970), 369-377.
[25] B. Peleg and M. E. Yaari, "Markets with a countably many commodi-
ties", Int. Econ. Rev. 11 (1970), 369-377.
[26] R. Radner, "Efficiency prices for infinite horizon production pro-
grammes", Rev. Econ. Stud. 34 (1967), 51--66.
[27] S. Richard and W. Zame, "Proper preferences and quasi-concave util-

ity functions", J. Math. Econ. 15 (1986), 231-247.
[28] W. J. Shafer and H. F. Sonnenschein, "Some theorems on the existence

of competitive equilibrium", J. Econ. Theory 11 (1975), 83-93.
[29] S. Toussaint, "On the existence of equilibrium in economies with in-

finitely many commodities and without ordered preferences", J. Econ.
Theory 33 (1984), 98-115.
532
[30] N. C. Yannelis and W. R. Zame, "Equilibria in Banach lattices without

ordered preferences", J. Math. Econ. 15 (1986), 85-110.
[31] K. Yosida and E. Hewitt, "Finitely additive measures", Trans. Amer.
Math. Soc. 72 (1952), 46-66.
[32] W. Zame, "Competitive equilibria in production economies with an
infinite-dimensional commodity space", Econometrica 55 (1987), 1075-
1108.
Program of the Ill. International Conference of the Society for
the Advancement of Economic Theory
ECONOMIC THEORY AND APPLICATIONS
June 16-21, 1997
Antalya, TUrkey
Program Committee
A. Alkan (Chair) ........................... Bogazi<;i University, Istanbul

C. D. Aliprantis ...................... University of Indiana, Indianapolis
S. Altug ........................................ Ko<; University, Istanbul
R. M. Anderson ....................... University of California, Berkeley
Y. Balasko ........................................ University of Geneva
D. Cass ......................... European University Institute, Florence
E. Dierker ......................................... University of Vienna
F. Forges .................................. Universite de Cergy-Pontoise
B. Grodal ....................... University of Copenhagen, Copenhagen
T. Ichiishi ............................. Ohio State University, Columbus
A. Khan ........................... Johns Hopkins University, Baltimore
A. Kirman ....................................... University of Marseille
M. Magill ................ University of Southern California, Los Angeles
P. McAfee ................................... University of Texas, Austin
H. Moulin .................................... Duke University, Durham
M. Quinzii ............................... University of California, Davis
M. R. Sertel ............................... Bogazi<;i University, Istanbul
B. Smith .................................... University of Texas, Austin
W. Trockel ....................................... University of Bielefeld
N.C. Yannelis ................. University of Illinois, Urbana-Champaign
W.R. Zame ........................ University of California, Los Angeles
Organization
Ahmet Alkan
Mehmet Barlo
Kemal Badur, Ipek Ozkal, Sinan Sarp<;a, Pinar TUtu§
Sponsors
The Scientific and Technical Research Council of TUrkey; TUrkish Academy

of Sciences; Yapl Kredi Bankasl; Istanbul Stock Exchange; Karma Inter-
national; Bogazi<;i University, Center for Economic Design; TUrk Ekonomi
Bankasl
534
Sessions
Monday
Opening Assembly
Welcoming by A. Alkan, C.D. Aliprantis, N.C. Yannelis
Asymmetric Information and Core

Chair: Bezalel Peleg, Hebrew University, Jerusalem
• Interim Design of Core Mechanisms, Gabrielle Demange (EHESS and

DELTA, Paris), R. Guesnerie
• Cooperative Interim Contract and Re-Contract: Chandler's M-Form

Firm, Tatsuro Ichiishi (Ohio State University, Columbus), M. R. Ser-
tel
• The Observable Information Core in Club Economies with Asymmet-

ric Information, Myrna Wooders (University of Toronto), C. Li
Public Good Provision

Chair: FUad Aleskerov, Bogazi<;i University, Istanbul
• Polling Mechanisms and the Demand Revelation Problem, Robert

Gary-Bobo (Universite de Cergy-Pontoise), T. Jaaidane
• Majority versus Average Rule, Alain Trannoy (Universite de Cergy-

Pontoise), R. Renault
• Concern for Status, Rank-Order Contests and Contributions to a

Public Good, Unal Zenginobuz (Bogazi<;i University, Istanbul)
Matching
Chair: R. Preston McAfee, University of Texas, Austin
• Marriage and Admissions via Graphs, Michel Balinski (Labarotoire

d'Econometrie, Ecole Polytechnique, Paris)
• Implementation of Matching Rules on a Restricted Domain, Tarik

Kara (Bilkent University, Ankara)
• Implementation of the Core of a Marriage Problem, Bezalel Peleg

(Hebrew University, Jerusalem), S.C. Suh
Axiomatic Analysis: Allocation Schemes, Processes, Equilibria
Chair: Yves Sprumont, University of Montreal
• A Note on Population Monotonic Allocation Schemes for Transferable

Utility Games, Theo Driessen (University of Twente)
535
• Axiomatic Analysis of Economic Transformation Processes, Andrzej

Malawski (Cracow Academy of Economics)
• Public Competitive and Evaluation Equilibria: Extending the Scan-

dinavian Consensus Approaches, Antonio Villar (University of Ali-
cante), P. Hammond
General Equilibrium I
Chair: Nicholas Yannelis, University of illinois, Urbana-Champaign
• Genericity with Infinitely Many Parameters, Robert Anderson (Uni-

versity of California, Berkeley), W. Zame
• Determinacy of Competitive Equilibria in Economies with Many

Commodities, Chris Shannon (University of California, Berkeley)
Clubs
Chair: Birgit Grodal, University of Copenhagen
• From Each According to Their Capacity to Pay: Differentiated Pric-

ing in Clubs, Laura Razzolini (University of Mississippi)
• Clubs and the Market, Suzanne Scotchmer (University of California,

Berkeley), B. Ellickson, B. Grodal, W. Zame
Dynamic Oligopoly
Chair: Suzanne Scotchmer, University of California, Berkeley
• Renegotiation in Repeated Oligopoly Interaction, Joseph Farrell (Uni-

versity of California, Berkeley)
• Consumption Externalities, Coordination and Advertising Wars, Tu-

vana Pastine (Bilkent University, Ankara), I. Pastine
• Dynamic Price Competition with Costly Production Adjustment,

Xavier Vives (Institut d'Analisi Economica, CSIC, Barcelona), B.
Jun
Dynamic Equilibrium and Growth

Chair: William Brock, University of Wisconsin, Madison
• Does Money Matter? A Deterministic Model with Cash in Advance

Constraints in the Labor Market, Erdem BB.§c,;i (Bilkent University,
Ankara), I. Saglam
• On Consumption Growth and Asset Prices in Stochastic Growth Mod-

els, Ralph Chami (University of Notre Dame), T. Cosimano, C. Ful-
lenkamp
536
• On Flows, External Effects and Multiple Long-Run Paths in a Two-

Sector Model of Equilibrium Growth, Jean-Pierre Drugeon (University
of Paris 1)
Mathematical Methods I
Chair: Semih Koray, Bilkent University, Ankara
• Pseudo-Supermodular Games: Theory and an Application, Elettra

Agliardi (University of Bologna)
• On the Indices of Zeros of Nash Fields, Fabrizio Germano (University

of Lausanne), S. DeMicheli
Mathematical Methods 11
Chair: Tatsuro Ichiishi, Ohio State University, Columbus
• The Existence of the Satisfactory Point, Adam Idzik (Polish Academy

of Sciences, Warsaw)
• Nonatomic Assignment Model with Upper Bound Constraints, Do-

raiswamy Ramachandran (California State University, Sacramento),
L. Ruschendorf
Thesday
Cooperative Games I
Chair: Herve Moulin, Duke University, Durham
• Orderings, Excess Functions and the Nucleolus, Jean Derks (Univer-

sity of Maastricht), H. Peters
• Agenda Independence in Allocation Problems with Single Peaked Pref-

erences, Carmen Herrero (University of Alicante), A. Villar
• The Select ope for Cooperative Games, Hans Peters (University of

Maastricht), J. Derks and H. Haller
Cooperative Games 11
Chair: Carmen Herrero, University of Alicante
• Sharing Externalities in Joint Production: Rationing, Additive Cost

Sharing and Beyond, Herve Moulin (Duke University, Durham)
• Randomized Uniform Allocation Mechanisms and Single-Peaked Pref-

erences of an Indivisible Good, Hiroo Sasaki (Waseda University)
• Ordinal Cost Sharing, Yves Sprumont (University of Montreal)
Dynamic General Equilibrium I

Chair: Mark Huggett, Centro de Investigacion Economica, ITAM, Mexico
537
• Pricing Rules When Agents Have Non-Additive Expected Utility and

Homogeneous Expectations, Rose-Anne Dana (University of Paris,
Dauphine)
• The We~fare Effects of Transfers in Overlapping Generations Models:
An Application of the Theory of Duality, Christian Ghiglino (Univer-
sity of Geneva), M. Tvede
• Endogenous Strategic Business Cycles, Steve Spear (Carnegie-Mellon
University, Pittsburgh), A. Goenka and D. Kell
Dynamic General Equilibrium Il
Chair: Clara Ponsati, Universitat Autonoma de Barcelona
• Collusive Temporary General Equilibrium, Milan Horniacek (Charles
University, Prague), C. Bejan and F. Bidian
• Aggregate Precautionary Savings: Infinitely-Lived Agents, Mark
Huggett (Centro de Investigacion Economica, ITAM, Mexico), S. Os-
pina
• Valuation Bubbles and Sequential Bubbles, Jan Werner (University of
Minnesota), X.D. Huang
Mathematical Methods III
Chair: Charalambos D. Aliprantis, Indiana University & Purdue University,
Indianapolis
• Operators Preserving Disjointness, Yuri Abramovich (Indiana Uni-
versity & Purdue University, Indianapolis)
• Some Aspects of Noncommutativity with Applications in Economics,
Vesna Pasetta (Cornell University, Ithaca, N.Y.)
• Revealed Preference Revealed, Sydney Afriat (Bilkent University,
Ankara)
Social Choice and Allocation
Chair: Alain Trannoy, Universite de Cergy-Pontoise
• Degree of M anipulability of Social Choice Procedures, Fuad Aleskerov
(Bogazic;i University, Istanbul), E. Kurbanov
• On Allocative Stability, Herve Cres (University of Pennsylvania), M.
Tvede
• Acyclicity of Fuzzy Preferences, Manabendra Dasgupta (University
of Alabama, Birmingham), R. Deb
• Aggregation of Preferences with a Variable Set of Alternatives, Jean-
Francois Laslier (Universite de Cergy-Pontoise)
538
Wednesday
General Equilibrium and Information I
Chair: Robert Anderson, University of California, Berkeley
• Equilibrium and Optimality in Incomplete Market Model with Default
and Credit History, Subir Bose (Iowa State University)
• Do Liquidity Constraints Matter for Ricardian Equivalence'?, Marcos
de Barros Lisboa (Stanford University)
• International Financial Equilibrium with Risk Sharing and Private
Information, Bart Taub (University of Illinois, Champaign)
General Equilibrium and Information 11
Chair: Martine Quinzii, University of California, Davis
• Endogenous Probabilities and the Information Revealed By Prices,
Tom Krebs (University of illinois, Urbana-Champaign)
• Information at a Competitive Equilibrium, Enrico Minelli (University
of Brescia), H. Polemarchakis
• Financial Innovations with Endogenous Risk, Jesus Santos (Univer-
sity of Chicago)
Dynamic Processes and Finance
Chair: Jan Werner, University of Minnesota
• Bifurcation Routes to Complex Dynamics in Evolutive Models in Eco-
nomics and Finance, William Brock (University of Wisconsin, Madi-
son)
• Risk and Return in a Dynamic General Equilibrium Model, Levent
Akdeniz (Bilkent University, Ankara), D. Dechert
• An Evolutionary Approach To Financial Innovation, Thorsten Hens
(University of Bielefeld), M.O. Bettzuege
Stock Markets
Chair: Michael Magill, University of Southern California
• Portfolio Insurance and Incomplete Derivative Markets, Charalambos
D. Aliprantis (Indiana University & Purdue University, Indianapolis),
D. Brown and J. Werner
• The Significance of the Market Portfolio, Stefano Athanasoulis (Iowa
State University), R. Shiller
• Incentives and Risk Sharing in a Stock Market Equilibrium, Martine
Quinzii (University of California, Davis), M. Magill
539
Oligopoly I
Chair: Egbert Dierker, University of Vienna
• On the Foundations of Cournot Oligopoly, Rabah Amir (Wis-

senschaftszentrum, Berlin)
• Protectionism Versus Non-Protectionism with Stackelberg Leader Do-

mestic Firms, Benan Zeki Orbay (Istanbul Technical University)
• Skewness of the Wage Distribution in a Firm and the Substitutability

of Labor Inputs, Rene van den Brink (Tilburg University
Sequential Games I
Chair: Joachim Rosenmuller, University of Bielefeld
• The Strategic Sequential Bargaining Model is Generically Incompati-

ble with the Cooperative Bargaining Theories, Yaacov Bergman (He-
brew University, Jerusalem)
• Backward Induction is not Robust: The Parity Problem and the Un-
certainty Problem, Steven J. Brams (New York University), D.M.
Kilgour
• On Subjective Uncertainty and Myopic Behavior, Doron Sonsino (Fac-

ulty of Industrial Engineering and Management, Technion, Israel)
Interacting Agents
Chair: Shlomo Weber, Southern Methodist University
• A Formal Approach to Market Organisation: Choice Functions, Mean

Field Approximations and Maximum Entropy, Alan Kirman (EHESS
and University of Marseille), J. P. Nadal, G. Weisbuch, and O. Ch-
enevez
• Rules of Thumb and Local Interaction, Akos Valentinyi (University

of Southampton)
• Technological Diversity with Localized Learning and Network Exter-

nalities, Murat Yildizoglu (Universite Louis Pasteur, Strasbourg)
Bargaining and Negotiation

Chair: Walter Trockel, University of Bielefeld
• Endogeneous Reference Points and the Adjusted Proportional Solu-

tion for Bargaining Problems with Claims, Carmen Herrero (Univer-
sity of Alicante)
• Bargaining in a Changing Environment, Clara Ponsati (Universitat

Autunoma de Barcelona), J. Sakovics
540
• Optimizing Multi-Stage Negotiations, Matthias Raith (University of

Bielefeld), R. John
• Loss Aversion and Bargaining, Jonathan Shalev (CORE, Universite
Catholique de Louvain)
General Equilibrium and Moral Hazard
Chair: Antonio Villanacci, University of Florence
• Moral Hazard in Economies with Multiple Commodities: Existence of
Competitive Equilibrium, Alessandro Citanna (Carnegie-Mellon Uni-
versity, Pittsburgh), A. Villanacci
• Overlapping Generations and Moral Hazard, Massimo Morelli (Iowa
State University)
Growth and Policy
Chair: Bruce Smith, University of Texas, Austin
• Political Risk and Irreversible Investment, Sumru Altug (KoC; Uni-
versity, Istanbul), F. Demers and M. Demers
• The Evolution of Modern Educational Systems. Technical vs. General
Education, Distributional Conflict and Growth, Graziella Bertocchi
(University of Modena), M. Spagat
• The Role of Credit Market Concentration in Promoting Technologi-
cal Progress, Nicola Cetorelli (Virginia Polytechnic Institute & State
University)
Thursday
Mechanism Design and Implementation I
Chair: Gabrielle Demange, EHESS and DELTA, Paris
• Ex Post Individually Rational Trading Mechanisms, Francoise Forges
(Universite de Cergy-Pontoise)
• The Optimal Design of Markets, Matthew Jackson (Kellogg Graduate
School, Northwestern University), S. Brusco
• Learning in Distributed Systems: Implications for Implementation,
Scott Shenker (Xerox Research Center, Palo Alto), E. Friedman
Mechanism Design and Implementation II
Chair: Francoise Forges, Universite de Cergy-Pontoise
• The Limit-Price Mechanism, Jean-Francois Mertens (CORE, Univer-
site Catholique de Louvain)
• The Anonymous Core: Characterization and Implementation for
Weighted Majority Games, Massimo Morelli (Iowa State University)
541
• Just~fiability of Bayesian Implementation in Oligopolistic Markets,

Ismail Saglam (Bilkent University and University of Michigan), S.
Koray
Voting
Chair: Salvador Barbera, Universitat Autunoma de Barcelona
• Multi-Member District Congressional Elections, Stephen Calabrese
(KoC; University, Istanbul)
• Majority Voting in Representative Democracy, Jean Laine (ENSAE,
Paris), G. Laffond
• Limited Lambda-Majority Voting Paradoxes, Jozsef Mala (Budapest
University of Economics)
Electoral Systems Design and Voting
Chair: Murat Sertel, Bogazic;i University, Istanbul
• Electoral Evolution, Salvador Barbera (Universitat Autunoma de
Barcelona), M. B. Maschler and J. Shalev
• Consistency and Electoral Systems Design, Semih Koray (Bilkent
University, Ankara)
• Strategyproof Division of a Private Good when Preferences are Single-
Dipped, Hans Peters (University of Maastricht), B. Klaus and T. Stor-
cken
• Strong Equilibria of Voting Games, Remzi Sanver (BogaziC;i Univer-
sity, Istanbul), M. R. Sertel
Matching 11
Chair: Michel Balinski, Laboratoire d'Econometrie, Ecole Poly technique ,
Paris
• Multi-Object English Auction with Object Dependent Preferences for
Money, Ahmet Alkan (Bogazic;i University, Istanbul)
• Buying Several Indivisible Goods, Carmen Bevia (Universitat Au-
tonoma de Barcelona), M. Quinzii and J. Silva
• Equitable Allocation of Divisible Goods and Market Allocation of In-
divisible Goods, Tatsuro Ichiishi (Ohio State University, Columbus),
A. Idzik
Experimental Studies
Chair: Alan Kirman, EHESS and University of Marseille
• Assessment Biases in Utility Functions, Christian Seidl (Christian
Albrechts University, Kiel)
542
• The Diamond Paradox: Experimental Evidence, Abdullah Yava§

(Penn State University), M. Sefton
• Network Model of Markets of Homogenous Products, V. Kalashnikov

(CEMI, Moscow)
Spatial and Network Competition

Chair: Scott Shenker, Xerox Research Center, Palo Alto
• Network Competition with Reciprocal Proportional Access Charge

Rules, Toker Doganoglu (Christian Albrechts University, Kiel) , Y.
Tauman
• Balance of Power in a Model of Spatial Competition with Rank Ob-

jectives, Marci Rossell (Federal Reserve Bank of Dallas), S. Weber
• Group Formation in Games with Externalities, Shlomo Weber

(Southern Methodist University, Dallas)
Learning
Chair: Yaw Nyarko, New York University
• Learning in Bayesian Games by Bounded Rational Players, Nicholas

Yannelis (University of Illinois, Urbana-Champaign)
• The Truth Is in the Eyes of the Beholder or Equilibrium in Beliefs

and Rational Learning, Yaw Nyarko (New York University)
• Real- Time Hierarchical Resource Allocation, Timothy Van Zandt
(Princeton University, New Jersey)
• Fast Learning in Organizations, Andreas Blume (Center for Economic

Research, Tilburg University and University of Iowa)
Dynamic Processes
Chair: Davis Dechert, Wisconsin University, Madison
• The Mechanisms of the Regional Programs' Financing, Vladimir

Bourkov (Institute of Control Sciences, Russian Academy of Sciences,
Moscow)
• Location in a von Thunen Structure: Urban Evolution and Urban

Planning, Beatrice Dadoun (CREST-LEI Bureau, ENSAE, Paris)
Topics in Macroeconomics I
Chair: Valerie Bencivenga, University of Texas, Austin
• The Economic Effects of Restrictions on the Government Budget,

Christian Ghiglino (University of Geneva), K. Shell
543
• Aggregate Fluctuations, Financial Constraints and Risk Sharing,

Pamela Labadie (George Washington University)
• Limited Enforcement and Time Consistent Foreign Investment, Anne
Villamil (University of Illinois, Urbana-Champaign)
Special Session on Mathematics
Chair: ~afak Alpay, Middle East Technical University, Ankara
• On Weakly Compact Operators, A. Wickstead (Queen's University of
Belfast)
• On the Invariant Subspace Problem, Charalambos D. Aliprantis (In-
diana University & Purdue University, Indianapolis)
• Operators Preserving Disjointness, Yuri Abramovich (Indiana Uni-
versity & Purdue University, Indianapolis)
Friday
Matching III
Chair: Tarik Kara, Bilkent University, Ankara
• Specialization During Education and Labor Market, Hakan Orbay
(Kot; University, Istanbul)
• Bargaining and Investment in a Two-Sided Matching Model, An-
drew Postlewaite (University of Pennsylvania), H. L. Cole and G.
J. Mailath
• Fairness and Efficiency in Student Placement Problems, Tayfun Son-
mez (University of Michigan, Ann Arbor
Auctions
Chair: Matthew Jackson, Kellogg Graduate School, Northwestern Univer-
sity
• Auctioning Entry into Tournaments, R. Preston McAfee (University
of Texas, Austin)
• Welcoming the Middlemen: Limiting Competition in Auctions by Ex-
cluding the Final Consumers, Subir Bose (Iowa State University), G.
Deltas
• The Multi-Unit Auction with One Seller and Few Bidders, Yvan
Lengwiler (Swiss National Bank)
Cooperative Games III
Chair: Hans Peters, University of Maastricht
• Optimal Entry and Contribution of the Players, Kunio Kawamata
(Keio University)
544
• Relation Between the Largeness of the Core and Exactness, Jos

Potters (University of Nijmegen), A. Biswas, S. Sinha and T.
Parthasarathy
• Predicting Proposal Configurations in Cooperative Games and Ex-
change Economies, Anton Stefanescu (Bucharest University)
Sequential Games 11
Chair: Jean-Francois Mertens, CORE, Universite Catholique de Louvain
• Learning and Incentive Compatible Mechanisms for Public Good Pro-
vision: An Experimental Study, Fang Fang Tang (Hebrew University,
Jerusalem), Y. Chen
• The Canonical Extensive Form of a Game, Joachim RosenmuIler
(University of Bielefeld), B. Peleg and P. Sudhoelter
• An Ordinal Selection of Stable Sets in the Sense of Hillas, Dries Ver-
meulen (University of Maastricht), M.J.M. Jansen
General Equilibrium 11
Chair: Hildegard Dierker, University of Bonn
• General Equilibrium with Multi-Person Households, Hans HaIler (Vir-
ginia Polytechnic Institute & State University)
• Allocation of Commodities and Consumers, Hans Gersbach (Univer-
sity of Heidelberg), H. Haller
• Hierarchial Allocation Schemes, Hans Gersbach (University of Hei-
delberg), H. HaIler
General Equilibrium III
Chair: Hans HaIler, Virginia Polytechnic Institute & State University
• Extension of Sonnenschein's Demand Theory, Farhad Husseinov
(Bilkent University, Ankara)
• Characterization of Utilities, Values, Distributions in a Model of a
Lindahl Type Economy, Alexander Sotskov (Bogazic;i University, Is-
tanbul)
• Equilibria in Mixed Economies of Arrow-Debreu Type: Existence and
Pareto-Efficiency, Hans Wiesmeth (Technische Universitdt Dresden),
B. Vasilie
Fiscal Policy I
Chair: Ay§e Imrohoroglu, KoC; University, Istanbul
• Non-Keynesian Effects of Fiscal Contractions: Theory and Applica-
tions for Germany, Bernd Lucke (Free University of Berlin)
545
• Unemployment vs. Mismatch of Talents: Reconsidering Unemploy-

ment Benefits, Ramon Marimon (European University Institute, Flo-
rence), F. Zilibotti
Fiscal Policy 11
Chair: Pamela Labadie, George Washington University
• Social Security with Heteregeneous Populations Subject to Demo-

graphic Shocks, Gabrielle Demange (EHESS and DELTA, Paris), G.
Laroque
• Is Altruism Important for Understanding the Long-Run Effects of So-

cial Security?, Luisa Fuster (Universitat Pompeu Fabra and Univer-
sity of Western Ontario)
• Two Computations to Fund Social Security, Selahattin Imrohoroglu

(KoC; University, Istanbul), H. Huang and T. J. Sargent
• Saving and Work Disincentives: Means of Testing Social Security,

Elizabeth Powers (University of Illinois, Urbana-Champaign)
Topics in Macroeconomics 11
Chair: Hasan Ersel, BogaziC;i University, Istanbul and Yapi Kredi Bank
• Technological Change, Public Finance and Income Distribution in De-

veloping Countries, Valerie Bencivenga (University of Texas, Austin),
E. J. Green
• An EMU with Different Transmission Mechanisms?, Giorgia Giovan-

netti (University of Florence), R. Marimon
• Inflation and Financial Market Performance, Bruce Smith (Univer-

sity of Texas, Austin), J. Boyd and R. Levine
Empirical Choice
Chair: Sumru Altug, KoC; University, Istanbul
• The Security Issue Decision: Evidence Prom Small Business Invest-

ment Companies, Hesna Genay (Federal Reserve Bank of Chicago)
• A Discrete Choice Dynamic Model of Human Capital Decisions, Cem

Mete (State University of New York, Stony Brook)
Saturday
General Equilibrium IV
Chair: Erdem Ba§c;i, Bilkent University, Ankara
• The Stability of Inflation Target Policies, Ramon Marimon (European

University Institute, Florence), E. Barucci and G. 1. Bischi
546
• Genernl Competitive Analysis with Asymmetric Information, Alberto

Bisin (Massachusetts Institute of Technology and DELTA), P. Got-
tardi
• On the Connection Between Correlated Equilibria and Sunspot Equi-
libria, Julio Davila (Universidad Autonoma de Barcelona)
Oligopoly 11
Chair: Hans Wiesmeth, Technische Universitdt, Dresden
• Shareholders'Surplus, Profits, and the Maximization of Shareholders'
Real Wealth, Egbert Dierker (University of Vienna), B. Grodal
• Product Differentiation and Market Power, Hildegard Dierker (Uni-
versity of Bonn) , E. Dierker
• Poor Consumers on the Product Line, Martin Peitz (University of
Alicante)

978 3 662 03750 8

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

978 3 662 03750 8

Uploaded by

Copyright:

Available Formats

Studies in Economic Theory

Springer-Verlag Berlin Heidelberg GmbH

M. Ali Khan and Nicholas C. Yannelis (Eds.)

Charalambos D. Aliprantis, Kim C. Border

Charalambos D. Aliprantis and Kim C. Border

Mordecai Kurz (Ed.)

Proceedings of the Third International Meeting

Prof. Charalambos D. Aliprantis

Prof. Nicholas C. Yannelis

Library of Congress Cataloging-in-Publication Data

In 1990, the Society for the Advancement of Economic Theory

A. Alkan, C. D. Aliprantis and N. C. Yannelis

Pseudo-supermodular games: theory and an application ................ 1

A degree of manipulability of known social choice procedures ......... 13

On the properties of stable many-to-many matchings under responsive

Cost uncertainty, taxation, and irreversible investment ................ 41

The significance of the market portfolio: theory and evidence ......... 73

Does money matter? A deterministic model with cash-in-advance

Welcoming the middlemen: restricting competition in auctions by

On the connection between correlated equilibria and sunspot

Ex post individually rational trading mechanisms .................... 157

Allocation among multi-member households: issues, cores, and

Valuation equilibrium revisited ...................................... 201

The equal-distance rule in allocation problems with single-peaked

Optimal entry and the marginal contribution of a player ............. 233

Non-keynesian effects of fiscal contractions: theory and applications

An extension of the nonatomic assignment model .................... 405

Testing decision rules for multiattrihute decision making ............. 413

Collusion and distribution of profits under differential information .,. 455

Predicting proposal configurations in cooperative games and

Department of Economics, University of Bologna, 40126 Bologna, Italy

Abstract. In this paper we introduce the notion of pseudo-supermodular

JEL Classification Numbers: C60, C70, C73

2 we introduce the relevant mathematical preliminaries and present the

2 Monotonicity of optimal solutions and

Proof. Suppose that f is not pseudo-supermodular, that is, there exist

By the (SC) and (A) properties we get

hence x V x' E <p(t', S').

hence, x 1\ x' E <p(t, S).

of ]Rm where m = L:::~=l mi. Elements of L are called strategy profiles,

STEP 2. The proof proceeds as in Milgrom and Roberts (1990), using

Remark 3. The set of Nash equilibria of a pseudo-supermodular game is

3 Pseudo-supermodular games and selection

is a regular selection dynamics if it satisfies for any w

Monotonicity requires that if pure strategy Xi receives a higher payoff

iih yo (t) _ WiYl (t) < 8 for all t sufficiently large,

Review, 80, 511-527

Fuad Aleskerov 1 and Eldeniz Kurbanov2

1 Department of Economics, Bo~azi~i University, Bebek, Istanbul 80815 Turkey

and Institute of Control Sciences, Moscow, 117806 Russia

JEL Classification Number: D71

Keywords: Social choice rules, degree of manipulation, efficiency of ma-

be the profile of agents' true preferences, while

where do is the number of profiles in which manipulation takes place! .

1 We also studied an extended version of Kelly's index. Denote by ~k the number of

J.2 - L):in L~=l Zi

The indices It and 12 were also used in [13].

Ziax = max(Zl, ... , ZKJ.

4 Social choice rules

XILY {:} card{i E N I XPiY} > card{i E N I yPix}.

Definition 4. Upper contour set of an alternative x in the relation P is

4.1 Scoring rules

where n+(a, P) = card{i E N I Vy E A aPiy}.