You are on page 1of 13

Signalling with Many Signals

Author(s): Maxim Engers


Source: Econometrica , May, 1987, Vol. 55, No. 3 (May, 1987), pp. 663-674
Published by: The Econometric Society

Stable URL: https://www.jstor.org/stable/1913605

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/1913605?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

The Econometric Society is collaborating with JSTOR to digitize, preserve and extend access to
Econometrica

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
Econometrica, Vol. 55, No. 3 (May, 1987), 663-674

SIGNALLING WITH MANY SIGNALS

BY MAXIM ENGERS'

This paper examines a market with asymmetric information where there are many signals
available and where both the costs of signalling and product value may depend on many
privately known characteristics. Under a weak condition on the relationship between the
marginal cost of increasing the signals and the product value, a separating set exists whereby
the value of every seller's product is inferred from the seller's optimal choice of signals.
The separating set constructed is Pareto-dominant and corresponds to recently proposed
equilibrium notions in signalling and screening models.

KEYWORDS: Screening, asymmetric information, self-selection, multidimensional sig-


nalling, informational equilibrium.

1. INTRODUCTION

PROSPECTIVE BUYERS may find it hard to determine the value of goods whose
quality varies from seller to seller. Although sellers know the value of their wares,
their incentive to make inflated claims can undermine their credibility. Suppose,
however, that each seller chooses the level of an activity called a signal, and
buyers observe this level. The signal can then communicate the value of the
product to buyers. Examples include education signalling the productivity of
workers (Spence, 1973, 1976) and deductible amounts signalling the risk class of
insurees (Rothschild and Stiglitz, 1976). If low-quality sellers mimic the signal
choices of higher-quality sellers, the signal is uninformative. Thus, for signalling
to work, a necessary assumption is that sellers of higher-valued products have
lower costs of increasing the signal level. Under weak regularity conditions this
assumption is also sufficient for signalling to convey information about product
quality (Riley, 1979; Engers and Fernandez, 1987).
This paper generalizes this sufficiency result to apply for any number of signals
and quality attributes. More specifically, this paper proves under general condi-
tions the existence of a Pareto-dominant separating set. A separating set consists
of signal-price offers by buyers such that if sellers each choose their best offer
from the set, then buyers earn zero profits on each transaction. This is separating
in the sense that products of different values are sold at different prices and thus
are associated with different signal levels.
A separating set corresponds to what Quinzii and Rochet (1985) call a separat-
ing equilibrium. The notion of equilibrium is still unsettled: there are various
ways of modelling the interaction between the informed sellers and the uninfor-
med buyers (Stiglitz and Weiss, 1983). In screening models, buyers act first, each
announcing signal-price offers to sellers who then each choose an optimal offer
from all those announced; in signalling models the sellers act first, choosing
signal levels, and then buyers make price offers. It is worthwhile outlining the
relationship between the Pareto-dominant separating set which is the focus of
this paper, and the equilibrium outcome in some of these models.

' The author thanks John Riley, Marius Schwartz, and Jonathan Skinner for helpful suggestions
and Peggy Claytor and Beverly Lecuyer for typing.
663

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
664 MAXIM ENGERS

In a screening model where buyers simultaneously announce all their signal-


price offers, any Nash equilibrium set of offers must be a Pareto-dominant
separating set. If it is not a separating set, then some buyer must be making
positive profits on some transaction which are typically offset by losses incurred
on other transactions at the same signal level. By offering a slightly higher price
at an appropriately higher signal level, a rival buyer can draw away the profitable
part of the first buyer's business. So the nonseparating situation is not a Nash
equilibrium. A similar argument establishes the Pareto-dominance. Riley (1979)
showed that for a wide class of screening models, no Nash equilibrium exists in
pure strategies and instead proposed the reactive equilibrium: a set of offers by
buyers which does not incur losses overall and which ensures that any buyer
introducing a new offer will incur losses after other buyers react to the new offer
by introducing additional offers. Any reactive equilibrium must be a separating
set; if not, some buyer could make a new offer which earned the buyer nonnegative
profits on each seller who accepted it, as long as the original offers stood. It turns
out that the reactive equilibrium exists under very general conditions and is
precisely the Pareto-dominant separating set, S, shown to exist in this paper,
both in the case of one signal (Engers and Fernandez, 1987), or many signals
(Engers, 1984).
Cho and Kreps (1986) have recently analyzed a signalling model where infor-
med workers choose education levels first and then uninformed employers make
competing wage offers. From the infinitely many Nash equilibria, they select one
on the basis of an intuitive criterion which corresponds to the stability requirement
of Kohlberg and Mertens (1986). This outcome, too, turns out to be the Pareto-
dominant separating set derived in this paper.
Only a few studies have examined the case of more than one signal (Kohlleppel,
1983; Quinzii and Rochet, 1985; Wilson, 1985). When many signals are available,
it is harder to show that a separating set exists since sellers are no longer completely
ordered by their cost of increasing the signal; typically one seller will have a
lower cost of increasing some signals than another seller, but have a higher cost
of increasing other signals. Both the value of the seller's product and the seller's
signalling cost depend on the seller's underlying quality vector or type, but in
both relationships the different quality parameters can interact in a complicated
way. Nevertheless, exactly as in the one-signal case, a condition may be found
which is sufficient to ensure that a separating set exists. The condition says roughl
that the relationship between the marginal cost of signalling vector and the value
of the seller's product is decreasing and quasiconvex. Since any monotonic
function of a single variable is automatically quasiconvex, the result encompasses
the standard one-signal result as a special case. The proof proceeds in two stages.
First, Section 3 outlines a very general environment and introduces two con-
cepts. A sure thing is a set of offers such that buyers earn nonnegative profits on
each transaction. Thus any buyer offering a sure thing will never incur losses
regardless of what other buyers offer. The union of all sure things is called the
sure offer set, denoted S. Section 4 uses an additional assumption to prove that
S is a separating set. Since every separating set is a sure thing, S contains all

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
SIGNALLING 665

separating sets and hence Pareto-dominates them. The additional assumption


says that if any offer set is such that some buyer is earning positive profits on
some transaction, then some other buyer can introduce an offer which earns
nonnegative profits on all sellers who accept it, as long as the original offers
stand. The new offer is called a sure thing given the original offers.
The assumption of Section 4 is not expressed in terms of the exogenous data.
This is remedied in the second stage where Section 5 shows that in an environment
with many signals the assumption may be replaced by the quasiconvexity condi-
tion referred to earlier. If a buyer is earning positive profits on some seller, the
separating hyperplane theorem is used to construct a new offer which lures away
that seller and attracts only sellers whose products are worth more than is being
offered for them. The new offer is thus a sure thing given the current set of offers
and the result follows. Section 6 concludes by providing an example to show
that without the quasiconvexity condition the result does not hold.

2. RELATED WORK

Most work on signalling has concentrated on the one-signal case, but other
authors have recently independently investigated multidimensional signalling.
Wilson (1985) has characterized a separating set in a monopoly model where
price and advertising signal product quality. Kohlleppel (1983) provided the
following labor-market example to argue that with more than one signal yi and
more than one underlying characteristic qi, complete separation is generally not
possible.

Productivity V(q, y) = q,

Signalling cost C(q, y) = Y2+ Y1+Y2


q2 q1 q2

This is illustrated in Figure 1. Through any point q* draw the hyperbola q1q2=
q*q2*. Pick any point q' above this curve and with q' < q* as shown. It is evident
that all signalling costs are lower for q' than for q*; in particular, the marginal
cost of increasing the signals:

aC(q',y) oC(q*, y)
y < for i = 1, 2.
ayi ayi

So, given any set of signal-wage offers, q' will choose levels of the signals
which are at least as high as those chosen by q*. But since q* has higher
productivity than q' it is clear that no separating set exists. This is the case because
the marginal cost of signalling varies directly with productivity. Under such
conditions, no separating set exists even in the standard one-signal case.
Quinzii and Rochet (1985) examine the case where there is an equal number
of characteristics and signals and where the cost of signalling takes the specific
form

C(q y) = Yi +Y2+ * *+Yk


q1 q2 qk

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
666 MAXIM ENGERS

q2
q

q q*
qq

FIGURE 1-Kohlleppel's example.

They prove that a separating equilibrium exists if the function mapping marginal
cost to productivity is convex. In this paper we show that only quasiconvexity
is needed and, more importantly, that the restrictions on the form of the cost
function and the number of signals and unobservables are unnecessary.

3. THE ENVIRONMENT

The market consists of informed sellers each selling one unit of the commodity
and (initially) uninformed buyers. Each seller's commodity is characterized by
a quality attribute q E Q, possibly a vector, which the seller knows but which
buyers cannot observe. Each seller chooses a signal y from a compact set Y
Buyers each obtain a value of V(q, y) from each unit of the commodity they buy
from a seller of type q signalling at level y. The function V is bounded and
continuous. Here Q is an arbitrary set. In Section 5, Q is taken to be a compact

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
SIGNALLING 667

metric space with V continuous on Q x Y. This ensures that V is bounded which


allows the straightforward compactness argument in Lemma 1 below.
The market operates as follows. First buyers tender sets of offers, each offer
being a signal-price pair (y, p) representing a commitment by the buyer to pay
p for a unit of the commodity to any and every seller who chooses the signal y.
Then sellers choose their best signal-price pairs amongst all those offered by
buyers. The preferences of each seller are characterized by a continuous utility
function U(q, y, p) which is strictly increasing in p.

3.1. The Basic Assumptions

Let Q denote the set of all types q.

ASSUMPTION 1: Y, the set offeasible signals, is a compact metric space.

ASSUMPTION 2: The value function of buyers V: Q x Y-> R is bounded, and for


all q E Q, V(q, y) is continuous on Y.

Let p=inf{V(q,y): qeQ,yG Y}, p=sup{V(q,y): qGQ,ye Y} and let P


denote [p,fi].

ASSUMPTION 3: The preferences of sellers are given by the utility function

U:Qx YxP-*R

where for all q c Q, U(q, y, p) is continuous on Y x P, and strictly increasing in p.

3.2. Market Choices

Since buyers are profit-maximizing competitors we assume, without loss of


generality, that they announce offers in Y x P. Sellers choose from the set of all
offers made by buyers. To ensure each seller's choice problem has a solution we
shall require this set to be closed (or equivalently, compact). We reserve the term
offer set to denote a closed subset of Y x P. Given a nonempty offer set A, then,
a utility-maximizing choice exists for each type q E Q, since U is continuous. The
choice need not be unique: we thus use the following tie-breaking rule: If a seller
of type q has more than one utility-maximizing choice from offer set A we assume
the seller breaks the tie by (i) selecting an optimal choice which maximizes buyer
profit, V(q, y) -p; (ii) if ties still remain, selecting the lowest remaining choice
according to some given ordering on Y x P.
Since the set of offers maximizing a continuous function over a compact set
is compact and not empty, this rule defines a unique choice by the seller from
any nonempty offer set A for each q E Q. We denote this choice by c(q, A) or by
tA(q)s(yA(q),pA(q)). The profit buyers make on a seller of type q is then
V(q,yA(q))-pA(q) which we denote by ITA(q). If A is 0, the empty set, we
adopt the convention that wA(q) =0 for all q. Finally if A is a nonempty offer

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
668 MAXIM ENGERS

set with 7-TA(q) =0 for every q e Q we say A is a separating set; in this case
V(q, yA(q)) = pA(q) for each q, so sellers of products of different values choose
different offers and, in particular, different signal levels yA(q) (since U is increas-
ing in p). B is a sure thing if B is an offer set such that VB(q) ? O for all q e Q.
The sure offer set (denoted by S) is the union of all sure things. We now show
that S is closed, itself a sure thing, and contains all separating sets.

4. RESULTS

LEMMA 1: The sure offer set S is a sure thing.

PROOF: We show that the closure of S which we denote by B, is a sure thing,


and so is contained in S by definition. Since any set is contained in its closure,
we have S = B and the result follows.
To check that B is a sure thing, note that by construction B is closed.
Let q c Q. We show that rB(q) - 0, which completes the proof. Since the choice
t = (yB(q), pB(q)) E B, the closure of S, we can find a sequence {Bi} of sure things
and of offers ti E Bi such that {ti} converges to t. Let t' denote (YBI (q), PB, (q)). Th

U(q, ti) - U(q, t') since t' is optimal and


V(q,yB,(q))-PB,(q) ; O since Bi is a sure thing.
Since B is compact {t'} has a subsequence that converges to some t' = (y', p') c B.
Taking limits above, since U and V are continuous,

U(q, t) -,- U(q, t'),

V(q, y') - p'?_ O.

Since t is chosen over t', the first inequality is an equality and the tie-breaking
rule then ensures that

7rs(q)-= V(q, yB (q)) - PB(q) >- V(q, y' )- p ? O


(by the second inequality above) which completes the proof.

LEMMA 2: The sure offer set, S, is not empty and contains every separating set.

PROOF: Any separating set is a sure thing and is thus contained in S, the union
of these. if y c Y then {(y, p)} is a sure thing; hence (y, p) E S.

DEFINITION: B is a sure thing given A if A is an offer set, B is a nonempty


subset of Y x P, disjoint from A, with A u B an offer set and ITA,B(q) ? 0 for al
qEQ such that c(q,AuB)EB.

AssuMPTION 4: For any offer set A satisfying iTA(q) > 0 for some q E Q, there
is a sure thing given A.

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
SIGNALLING 669

THEOREM 1: Under Assumptions 1 to 4 the sure offer set S is a separating set


and contains all separating sets.

PROOF: By Lemma 2, we need show only that S is a separating set, i.e. that
lrs(q) =0 for all q c Q. By Lemma 1, 'rrs(q) - 0 for all q c Q. We suppose there
is some q c Q with vs(q) > 0 and show this leads to a contradiction. By Assump-
tion 4 there is a sure thing given S; call it B. Since S is a sure thing it follows
easily that S u B is a sure thing. So S, the union of all sure things, contains S u B
and hence contains B. This contradicts the fact that B, a sure thing given S, is
disjoint from S. Q.E.D.

5. SIGNALLING WITH MANY SIGNALS

To generalize the standard signalling results to environments where y and q


may be vectors rather than scalars, we make a simplification: instead of sellers'
preferences being given by the general utility function U(q, y, p) we assume that
this utility function takes the form p - C(q, y). The function C(q, y) may be
thought of as the total cost of signalling at level y for a seller of type q.
If q and y are scalars the standard assumptions are that product value to
buyers V(q, y) is increasing in q and that the marginal cost of signalling, MC(q, y)
(the partial derivative of C(q, y) with respect to its second argument), is
continuous and strictly decreasing in q. We reformulate these assumptions
in a way which at first may seem awkward yet permits us to generalize to the
many-signal, many-unobservable case. Given y and p consider the set B =
{MC(q, y): q E Q, V(q, y) < p}. Our assumptions imply that B consists of all
marginal costs MC(q, y) lying on or above some threshold level; that is, if x E B
and q* is such that MC(q*, y) - x then MC(q*, y) c B, i.e. V(q*, y) - p. In fact
if x is any convex combination of elements of B, then x lies on or above the
threshold and so if q* is such that MC(q*,y) - x, then V(q*,y) > p as before.
This last condition, M5 below, is sufficient to ensure the conclusion of Theorem
1 holds in the many signal case.
Condition M5 says that a weighted average of marginal cost vectors correspond-
ing to sellers of products worth no more than p, never is uniformly lower than
the marginal cost vector of a seller q* of a product worth more than p. To
understand why such a condition is needed, recall that to separate a high-value
seller q* from a low-value group, we need an increase Ay in the signal vector
which is less costly for q* than for the others. For any type q, the extra cost of
increasing the signal by Ay is approximately MC(q, y) - Ay. If M5 were violated,
then for any Ay in R+ the extra cost to q* would be (to a first-order approximation)
at least as high as the weighted average for the group. Thus there would be no
signal increment Jy which is less costly for q* than for every member of the
low-value group.
Assumption M6 (below) says that an offer of price p, the maximum produc
value, contingent on any signal vector on the upper boundary of Y belongs to
the sure offer set S. Since S is a sure thing, when S is offered, all sellers are paid

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
670 MAXIM ENGERS

a price no greater than the value of their products and so all sellers signal at
levels inside the upper boundary except possibly sellers of the highest-valued
products. M6 in effect ensures that the upper bounds on signal levels are high
enough to allow sufficient room for separation. It is easy to construct examples
to show that M6 is needed for Theorem 2 (see Engers and Fernandez (1986)
Example 2; alternatively with two sellers of different-valued products, take Y to
be a one-point set).

Assumptions

ASSUMPTION Ml: The set Q of types is a compact metric space.

ASSUMPTION M2: Y is a compact convex subset of R .

ASSUMPTION M3: V: Q x Y-> R is continuous, and nondecreasing in each yi.

Since Q x Y is compact, V is bounded. Let p, fp, and P be defined as before.

ASSUMPTION M4: The preferences of sellers are represented by the utilityfunction


U(q, y, p) p - C(q, y) where C, the cost-of-signalling function, is differentiable
with respect to y with derivative MC(q, y). Both C(q, y) and MC(q, y) are con-
tinuous on Q x Y.

ASSUMPTION M5: If p E P, y E Y then if x is a convex combination of elements


of the set {MC(q, y): q E Q and V(q, y) - p} and MC(q*, y) - x, then V(q*, y) - p.

ASSUMPTION M6: S contains (y, ft) for all y on the upper boundary of Y.

THEOREM 2: Under Assumptions Mi to M6 the sure offer set S is a separating


set and contains all separating sets.

PROOF: By Theorem 1 we need establish only that Assumptions M1-M6 imply


Assumption 4.
Given offer set A and q* E Q such that irA(q*) > 0, we construct a sure thing
given A. Denoting c(q*, A) by t* = (y*, p*), suppose V(q*, y*) > p*. This implies
p* < p and hence y* does not lie on the upper boundary (by Assumption M6)
since we may assume A contains S (if it did not, S\A would be a sure thing
given A).
Choose p^ such that V(q*, y*)>p>p*.
Let H denote the convex hull of the set

M = {MC(q, y*): q E Q and V(q, y*)_ p


and let X denote the set {x E R n: x - a for some a E H}. X is easily shown to
be closed and convex. By Assumption M5 the vector MC(q*, y*) does not lie

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
SIGNALLING 671

in X. By the Minkowski separating hyperplane theorem, there exists a vector


hER' such that

(1) h-MC(q*,y*)<l<h-x forall xeX.

Since X is not bounded above, h - 0. Since M is contained in X, the right-hand


inequality implies:

(2) h - MC(q, y*) > 1 for all q such that V(q, y*) pt.
Consider the offer tA = (YA, PA) where YA = y* + Ah and pA = p* + A. Given any
neighborhood N of t* we can choose A > 0 and sufficiently small (say A < AO)
that tA E N, since y* is not on the upper boundary of Y
U(q, tA) -U(q, t*) = A -[C(q,y*+Ah)-C(q,y*)]

= A[1 -f(q, A)],

where f(q, A) = (1/A)[C(q, y* + Ah) - C(q, y*)].


As A tends to zero, f(q, A) converges to h - MC(q, y*). Since this is continuous
on Q by M4 we may extend f to be continuous on Q x [0, AO] by defining
f(q, 0) = h MC(q, y*).
By (1) f(q*,0)<1, hence for A sufficiently small, say A <Ai,f(q*,A)<1 so

U(q*, tk) -U(q*, t*)-- A[ 1-f(q*, A)] >O


and thus tA - A.
By (2) f(q, 0)> 1 for all q in the compact set {q: V(q, y*)/ '}. Let b denote
the minimum value of f(q, 0) on this set. Since f is continuous on a compact set,
it is uniformly continuous, so we can choose a 8>0 such that if the distance
d((q, A), (q', A')) < 8, thenf(q', A') -f(q, A) < b-1. Choosing A < 8, if V(q, y*) S
p we have
f(q, A) >f(q, 0) + 1-b , b + 1-b = 1.

Now U(q, tA ) - U(q, t*) implies f(q, A) - 1 which implies that V(q, y*)>
Choose A < min {Ao, A1, 5, p-p*}. Then U(q, tA) - U(q, t*) implies V(q, YA)
V(q, Y*)> A> PA and so {tA} is a sure thing given A and the theorem follows.
Q.E.D.

6. AN EXAMPLE

Under Assumptions Ml to M6 the sure offer set S is the Pareto-dominant


separating set. We now provide a simple example with two signal variables, two
unobservable quality parameters and three seller types, which does not satisfy
M5 and for which the two sets are not the same.

Q = {a, b, c} where a = (1, 2), b = (2, 1), and c = (54 4),


y=[0,y]2 for Y92,
V(a,y)= V(b,y)=1, V(c,y)=2 forall ye Y,

U(q, y, p) = p - C(q, y) = p ----,


ql q2

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
672 MAXIM ENGERS

so

U(a, y, p) =p - Y1-2
2'2
U(b, y, P) = P - Y2, and

U(c, y, p) =P -(YI +Y2).

Note that the marginal cost vector MC(q, y) associated with type q is independent
of y, and MC,a = (1, ), MCb = (4, 1), and MCc = (4, 4).
To illustrate Assumption M5 we plot the marginal cost vector of each type
and label it with the value of the product of a seller of that type. This is shown
in Figure 2.
We show that the sure offer set S has types a, b, and c each choosing the offer
t = (0, 0, 1); hence S is not separating. Note first that {t} is a sure thing and that
it provides a utility of 1 for types a and b. No other choice from a sure thing
can give a or b as high a utility. Now suppose c were to choose some other offer
t' = (y, p). Since a and b choose t, we must have U(a, t) - U(a, t') and U(b, t) ?
U(b, t'). Since the cost of signalling is linear in y these can be expressed:

lp-MCa y and 1:p-MCb y

where * denotes the scalar product of vectors in R'.

MCb (V=1)

\ MC (V=2)
4/5 ,

aC\
ac
a Y2\
1/2 Mo0 (V=1)

1/2 4/5 1
a
aYl

FIGURE 2-Marginal costs of signalling such that S is not separating.

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
SIGNALLING 673

MCa, y-p-1 and MCb- y p-1;


hence

'(MCa + MCb) * y - p - 1 and so


MC, * Y > I(MCa + MCb) * y- p-1.
Thus type c will not choose (y, p) either. As long as there is some convex
combination of the marginal cost vectors of the low-valued sellers a and b, which
is in each component below the marginal cost vector of the high-valued seller c,
this problem will arise, preventing the conclusion of the theorem from holding.
Thus in this example MS is necessary as well as sufficient for Theorem 2, given
the remaining assumptions.
Note that separating sets do exist for this example. One consists of the three
offers (0, -, 1), (2, 0, 1), and (-, , 2) which has types a, b, and c choosing the
first, second, and third offer respectively. Note that the choice (0, 0, 1) from S is
strictly preferred by all sellers to any separating outcome.
We haven't yet characterized S in this example but the reader should have no
difficulty verifying that it is the subset of Y x P bounded above by the indifference
surfaces for types a and b passing through (0, 0, 1); that is,

S = {(yI, Y2, p) E Yx P: P -Y-1 and p -jY2;1}.

If we altered the example by reducing V(q, y) by .9 everywhere and we allowed


sellers to achieve zero utility if they neither signal nor sell their product, no
separating set exists. Quinzii and Rochet (1985) have provided an elegant alterna-
tive example of nonexistence when Q = Ri for k > 2:

C(q, y) = Yi Yk
ql qk

and V(q, y) = ||MC(q) 1 = (ql 2+

Department of Economics, University of Virginia, Charlottesville, VA 22901,


U.S.A.

Manuscript received September, 1985; final revision received July, 1986.

REFERENCES

CHO, I., AND D. KREPS (1986): "Signalling Games and Stable Equilibria," Quarterly Journal of
Economics, 102, forthcoming.
ENGERS, M. P. (1984): "The Existence and Uniqueness of Signalling Equilibria," unpublished Ph.D.
dissertation, UCLA.
ENGERS, M. P., AND L. FERNANDEZ (1987): "Market Equilibrium with Hidden Knowledge and
Self-Selection," Econometrica, 55, 425-440.
KOHLBERG, E., AND J-F. MERTENS (1986): "On the Strategic Stability of Equilibria," Econometrica,
54, 1003-1037.
KOHLLEPPEL, L. (1983): "Multidimensional 'Market Signalling'," Discussion Paper, Universitit
Bonn.

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms
674 MAXIM ENGERS

QUINZII, M., AND J-C. ROCHET (1985): "Multidimensional Signalling," Journal of Mathematical
Economics, 14, 261-284.
RILEY, J. G. (1979): "Informational Equilibrium," Econometrica, 47, 331-360.
ROTHSCHILD, M., AND J. E. STIGLITZ (1976): "Equilibrium in Competitive Insurance Markets:
An Essay on the Economics of Imperfect Information," Quarterly Journal of Economics, 90, 629-649.
SPENCE A. M. (1976): "Competition in Salaries, Credentials and Signaling Prerequisites for Jobs,"
Quarterly Journal of Economics, 90, 51-74.
(1973): Market Signaling: Information Transfer in Hiring and Related Processes. Cambridge,
MA: Harvard University Press.
STIGLITZ, J., AND A. WEISS (1983): "Sorting Out the Differences Between Screening and Signaling
Models," mimeo, Columbia University.
WILSON, R. (1985): "Multi-dimensional Signalling," Economics Letters, 19, 17-21.

This content downloaded from


200.16.86.81 on Thu, 01 Jun 2023 19:42:35 +00:00
All use subject to https://about.jstor.org/terms

You might also like