"""1"t,,,..I,, separate however.

llWo and there is

Intll'!ril'!mnond
Models
TIMOTHY J. KEHOE
110 introdadion
Hili this we shaU the properties of llWo
oral a model wiRh a finite number of
nnlilDitelv lived consumers, and an overlapping model with an
finfinile number of lived consumers. Both models contain a
cornplete set of markets. 1be Anow-Debreu formulation of the Wal-
Iraman modeJ can. of course, be a UYIIlIillDlIC intpIT.....t"'.
in goods are indexed date. The models that we differ
from the standard model in that we allow an infinite
number of As we shall see, models with infinite numbers of
OOissel$S very different from models with finite of
models are regarded as however: their
...r<,.....rti ..., are in so far as provide into models
but finite. nwnbers of
with a finite of lived consumers sbares
UUIPUI win nrnr...r1ti..", with the standard Arrow-Debreu modei:
aU equilibria are there is no role to be played
outside money, ulllbacked nominal debt; and, third, there are, in gellleliU
.ill finite number of uniqne equilibria. In contrast, the
gelllera,tiOllS model may violate each of these three properties: it may
have that are IK)( it may have in
wbidt outside money plays an and it may a robust
continuum of equilibria. We see that there is a close
between the possibility of of and the role for
outside money. The possibility of is a
one
equilibrium models are be.colirUng iIlICll:asi:nglly
particularly in
The research in abis paper bas been funded by GTant 00. SES-8S00484 from ibe
National Fouildatioll.. M, lhinking 00 ,be i.mIer. discussed in this bas been
l'Iith odIer n:sem::blm same area. In
partiallar, I woukl to thlIDk David Backus, JOIIlIIthIID Buroell. Jotm
GellllSlkop'los, Frank Hahll, Patrick Kehoe, Andreu Mas-Coklil.
Pot.eman:ludi:is, Paul Romer. Micllad Woodford. WiDiam
Levine.
TIMothy J.

ttOOre llIas been to use small general equilibrium models no analyse
macroeconomic issues (see, for example, Lucas 1981, and Kydland and
I?r:escott 1982). Not all macroeconomists have been caught up in this
nrend, however, and the use of explicit general equilibrium models an
macroeconomics has been the subject of much controversy, in which one
side accuses the other of using ad /we and unrealistic models. Many
commentators have interpreted this controversy as an idealogical debate
between monetarists and Keynesians. This interpretation is probably
IlUllfortunate. An explicit general equilibriwn framework impo5CS a
discipline and assures internal consistency. This makes it easy for us to
organize our thinking about economic phenomena and to
dJis thinking to others, mostly because the assumptions of the model have
weU understood implications in this framework. The phrase ad hoc is
much misused in economics. It bas become a synonym for 'yours' and
'bad' and an antonym for 'mine' and 'good' . Most good economic models
are ad hoc in the strict sense that they are designed for a particular
and produce resulls that follow closely from a particular set of
assumptions. The advantage of using explicit general equilibrium models
is that they provide a framework. in which sets of assumptions are easily
nmderstood and compared.
The potential disadvantage is, of course, that the general equilibrium
framework caD become an intellectual straightjacket. Fortunately, how
ever, this framework is ricb enough to allow a wide variety of results. To
BlDustrate this, we enlploy both of our models to answer Barro's (1974)
question of whether government bonds are net wealth. Different models
a m produce very dilJerent answers this question. 'There is a close
relationship between these answers and the sets of assumptions that
dBstinguisb these models.
The models that we study in this paper are both pUJre-excbange models
no production or storage. TlDle is discrete and there is no
u:J1Certainty. Furthermore, both models are stationary in that the struc
mre of preferences and endowments is constant over time. These models
are the simplest to analyse. We indicate, however, how our results extend
more complicated models. We also oompate the structures of the two
models. On one band, the overlapping generations model has similar
properties to a model with a finite number of infinitely lived consumers
1fJho face borrowing and lending constraints. On the other, a model with
21 finite Dumber of infinitely lived consumers has similar .,.-operties to an
overlapping generations model in which parents leave bequests to their
ciBildreD.
:!'o All lalaitel)' u.etI <Co_ Mecie!J
We begio by analysing an economy with a finite number of agents who
consumer over an infinite number of time-periods. There are FU goods,
ffnlertemporal Equilibrium Models :
which cannot be stored, in each period and h consumers. Consumelr ,i ns
characterized by a utility function

2: rj-iUj(c:{" .. . c1",) (1)
,=1
and an endowment vector wi =(MI., ... which he has claim no Dill
each period. Here the discount factor y, satisfies 1> Yj > 0; the mom(m
tary utility function u, is continuously differentiable of the second olTck r
for all positive consumption vectorS, strictly concave, and monotonia Uy
increasing; and the endowment vector wi is strictly positive.
There are two interpretations of this model. 'The first is tbe traditiollaB
Walrasian interpretation in which all trades, including those tbat involve
future delivery of goods, take place in the fust period. In this interpreJa
tion time plays no explicit role and I can be thought of as merely anotti Ie!'
index on commodities. 1be consumer's budget constraint is
GO
2. 2, p;wJ.
':2}
8=1 ,=E
Here p, ,.,. (p'tt ... p",) is the vector of futures prices in period 6 and If:<d.
is the inner product !:7=t p;,e!,.,.
In the second interpretation trades place over time, but there
perfect capital markets and ratiooal expectations. In this simple
the assumption of perfect capital markets means only that consumers (an
borrow and lend as much as they want at a competitively determillliW
interest rate, and the assumption of ntional expectations means tl ,at
consumers have perfect foresight. Let 'I, =(ql" . . . q,.) be the vectorr of
spot prices in period t; let r, be the ioterest rate between I and f + 1; Iud
let be the net lending done by consumer j between t and , + t. We Ulo"ll
iJnterpret m{ as inside money. Consumer j faces a sequence of budl:et
constraints
+ m{ q;wi
+ <5i qiwi + (1 + rl)m{ (13)
urt, + (1 +
Dividing the budget constraint in period g by (1 +,.)(1 + ,7) . . . (li + "'8- Y'
e=2, ... T. and adding up, we obtain
T T
L p;d,+ + Yt)(l -+ r2) ' (U + YT-t) <5i L p;w'
' 0 & .-1
where p, = q,/(D +1"1)(1 + r2) . (1 + r,- i)' in the limit dbis produces lle
same budget constraint as does the first interpretation as long as
lim + V'1)(n+Y2) (ll +8'r_l ) =(]I.
T-
=
.J6tj Timothy i. KehOfi
To that this condition holds, we need to puQ some constraint OOl
ahe real level of debt that we alloW consumer j iO incur. We shalH see thai,
such a constraint, (5) must hold in any equilibrium.
[Let us retum to the tim interpretation of the model. An equilibrium ns
BJ sz:quence of price vectors (p" Pl...) and a sequence of consumption
vedors <&a, &....) for each consumer. j = 1. ... , h. such that each
consumer maximizes utility subject SO his budget constraint (2) and
demand is equal to supply:
/a /a
1=1,2... . {6}

thai any equilibrium must be such that converges;
oaherwise the consumer would have infinite income, and his utility
IDwtimization problem would have no solution. Because Uj is monotoni
cally increasing, he would want to consume infinite amounts of at least
on<e good, which would make equilibrium impossible.
Since every coosumer has finite income, Rhe yalue of the aggregate
endowment must also be finite:
(1)
wi) == 0
This impiies that any equilibrium must be Pareto-efticienL The argument
is due to Debreu (J954). Suppose, 0 Ute rootrary, ahlili ihel!"e BS d!I
lP'Etieto-superior allocation plan (c{, c'z, ...):
.. ..
}: yj-U
14J
{CI), j = . . . , n, (8)
-1 ,=1
shict inequality for some j. ltiIat is feasible:
" II
Li{";;;Lwi, =1.2...
jml
l1ilen ihe consumption sequence (c{, &, ...) must cost at leasa as much
lllS tlhe consumer's income, and strictly more for some consumer;
Ci2> ) would not be utility-maximizing. Consequently,
(f p:e,) > wi).
I-I ,1 1:1 ,ma .=D i=1
SfuItce the Pareto-superior allocation is feasible, however,
(i p;&,) = i p;(f e.),,;;; ifo;( wi). Ill)
I-I .=1 ,-1 I-I
Tillis oontradiction establishes that there can be no allocation that ns
l?i!feto-superior to the competitive allocation and is also feasible.
lniertemporai Equilibrium Models 35J1
That nhe value of the aggregate endowment is finite also implies ttl an
there can be no equilibrium outside money, unbacked deij ,t.
Suppose, to the contrary, that there is an al1ocation tbat is feasible in
which each consumer satisfies the budget constraint
...
L

p;Ci, = L p;w' j=l ... Ii (H)

where
m= L
:,
m'*O. {B)
j=n
(U m = 0 but mi =F 0, this is just an equilibrium with transfer paymenb .)
Here m is the stock of outside, or fiat, money, which can be positive l 'lr
negative. Summing these budget constraints over consumers, we obt.wr
t (i p;&,) =(i p;wi) +m. QD- l}
;=i ,=u }=I ,al
Multiplying the feasibility oonditions (6) by prices and summing, ]hO'lll
ever, we obtain
. )
'=0 j 'J-I
Consequently, m =0, which contradicts the assumption mere is a! fU
equilibrium with outside money.
11tis same argument can be used to show that the sequence of budgd
consuaints (3) are equivalent to the single intertemporal budget COil 0
straint (2). For consumer i to have a weD defined maximization problem
and for the concept of equilibrium to make any sense, the limit in (5)
would have to exist. Here, unlike the outside money case that we haVI
bust examined, the variables are chosen by the consumers. Since utilit'
is monotonically increasing, every consumer would want to chaos; !
...) so that the limit in (5) is negative. 1be same argument tha (
predudes an equilibrium with outside money also precludes Qbi ;
possibility.
For an equilibrium to em. however, we must impose a 06 !
ibe real level of debt that we allow COnsumer j to iRalr.
m{/lIp,1I 8 = 1,2.... (16
for some b < O. Otherwise. the consumer would try to run a fonz '
scheme, rolling over an exponentially increasing amount of debt
making the limit in (5) as negative as possible. In such a case, as
argued above, no equilibrium can exist. Any son of bound on tbe Jreal
level of debt, no matter how large in absolute value, precludes nhh
possibility.
f":';"::t
J(,'8 Trmothy J. Kehoe
lIn general, this model has a finite number of IocaISy unique equilibria.
1'01 see abis, we transform the equilibrium conditions using an approach
developed by Negishi (t96Ob) and applied to inlertemporal models
by Bewley (1982). To simplify the exposition, we ignore the possibility of
oomer solutions to the consumer's utility maximization problem. This can
he justified by imposing an additional restriction on Il} (see Kehoe
bvine 1985a). 1be solution to the consumers utility maximizati011l
problem is <:haracterized by the c:onditions
t;-
l
DU;(d,)=).jP; (li1)
ffJf some Lagrange multiplier it; > 0, and the budget oonstraints (2).
(Here DIl/(d,) is the I x II vector of partial derivatives of Ilj.) An
equilibrium is, therefore, characterized by (2) and (17), which are Qhe
utility maximization c:onditions. and (6), wlUch are the market-dearing
cooditioos that demand be equal to supply. 'Ibis ns 2 system with an
infinite number of equations and unknowns.
Consider now the Pareto problem of maximizing a weighted sum of
ilildividual utility functions subject to feasibility wnstlraints:
A ..
max aJ L rl-luAd,)
(18)
/ .. g
Ii Jt
s.t L d,== L wi. i = 1, 2, . ..

lHere aI, ... all are positive utility weighas. fir. solution to iliis problem
os characterized by the conditions
="" j = 1... h, 8 = n, 2, . . . (19}
fur some sequence of vectors of Lagrange multipliers Ra =
(nl" ... , Jr",) > 0, and the feasibility constraints (6). Notice that, jf we
divi<Be (19) by ai' then it bec:omes the same as (17). This is an alternative
','vay of seeing that any c:ompetitive equilibrium is lPareto-efficient, thaa
f:he FlCSt Theorem of Welfare Economics holds.
TIle Second Theorem holds as well: any solution io the Pareto problem
i{ IS) satisfies all of the conditions for a competitive equilibrium except the
3ndividual budget c:onstraints (2). Such a solution can, therefore, be
':fiewed as competitive equilibrium with transfel" payments. The rompeti
live prices are, of course, the Lagrange multipliers 11,. We can compute
iUle transfer payments needed to implemeot as a competitive equilibrium
,!he Pareto-efficient allocation associated with the welfare weights a =
la" . , . , a,.):
...
Ij(a) =2: Jr,(a)'{d,(a} - wi!. j == .. . , h. (20)
, =1
fffflerlemporai Equilibrium Models
Setting these transfers payments equal to 0 produces a characterizatic i'l
equilibria in a finite number of equations and unknowns.
Using the strict concavity of "I' we can demonstrate that itraRlSferr
functions 1/ are c:ontinuous. Also, 'i is homogeneous of degree B JIl\\ CJ1
because 1f, is homogeneous of degree 1 and is homogeneous of degree
0; if we double a, for example, the sequences of consumption veL10n
ahat solve the problem do not <:hange, but the Lagrange multipliers
double. Furthermore, the transfer functions satisfy
Po
L gJ{a} =0 {21]i
/=B
because any soiution Qo !the Pareto /problem satisfies the feasii ;i1ity
oonstraints.
1lte ronditions that charaaerize the equilibria of this are
fonnaJly equivalent to those that charaderize the equilibria of a siali<:.
pure-exchange model with 13 goods. Indeed, the functions "e,t) =
- I
J
( a)1aj have all of the properties of the excess demand functions of fl
model: they are rontiouous; they arc homogeneoltS o[
degree 0; and they obey Walras's Law, 1::...,J"(a) -0. Debreu
has demonstrated that, if the excess demand functions" are rontinu(tUSly
differentiable, then aboost all economies have a finite number of Io'.:ally
unique equilibria. The phrase 'almost an' is, of coune, given a
mathematical meaning. We can use either the transfer functions 'i 01' the
demand functions /; to characterize the equilibria of tbe interteml,;orafi
model that we are considering bere. Kehoe and levioe (l985a) :iJave
shown that Debreu's rt:$Oniog extends to this model. FurthermoR . by
amposing another. fairly weak. condition on uit they are abln ft())
demonstrate that I
J
is indeed c:ontinuously differentiable.
The proof of Debreu's result relies on fairly oompJex mathemEtkafl
macbinery. The intuition behind it is very simple, however. It is, in faa.
the same intuition as Walras had when he counted equations ancjJ
unknowns: There are h equations, li(a) = 0, in h unknowns. ai. Bec:luse
of homogeneity. one of the weights aj is redundant. Because ofr the
adding-up restriction (21), however, one of the equations is alw
redundant. Consequently, the equ.ilibrium c:ondilions can be viewed lM 2l
syetem of II - 1 equations gn Ie - 1 unknowns. Suppose i:mese
equations are independent in the sense that 'j(er) =o. j = 1.. . ,h - n.
and the (13 - i) x (h - 1) matrix of partial derivatives

oal

8a... _
a
1 = (22)
01"'_1 (&) 8t,._1 (ci)
aerl a.. _ 8
370
,,""
Timothy J. Kehoe
is nonsingular. (We have imposed the normalization aA> = i and dropped
filiJe equation 1,.(0') == 0.) Then the inverse function theorem of elementary
calculus says that, in some open neighbourhood of a, it is the only
solution to the equilibrium conditions; that is, ,-8(0) =a. Using
compactness of the set of possible equilibria and the continuity of the
lequ.ilibriwn conditions. we can easily prove that there is a finite number
oj equilibria if J is non-singular at every equilibrium.
If J is singular at some equilibrium, then the intuition says that the
sIlightest perturbation in the functions '/ either make it non-singular or
ei!se make it impossible for there to be a solution near a. Figure 16.1
iRlustrates some possibilities in an economy with two consumers.
To mate some of the concepts that we have discussed in this sectioo
more concrete, let us consider a simple model with one good in each
period and two oonsumers. Suppose that ",(c,) -= uz(c,) =log c, and
WI = Ml2 = 1. lbe only difference between the awo consumers is
Yo < 12 A. solution to the utility maximization 85 characterized by the
<c()ooitions
-1/ ' i
ri C:=A;P, (23}
.. ..
LP,d,= (24)
g
equilibrium satisfies these conditions and the condition ihai demand
t,ta,.i}
/,-
/
I I
I I
1
/
I
-- -
11',

n!!i.ll
lnlertemporai Equilibrium Models
nn
equals supply:
C: +11:: + 2, 8 =1, 2, . . .

1rhe Pareto pli'Oblem is
GO ..
max at LA
"" ...-1110
get
0+ ""' '-II
ogc, TO 2 La 1'2
2
(:!6)
,I:: t r=- R
s.L 11:: + -= 2, 0= n, 2, . ..
A. solution to this problem is characterized by the oonditions
O'jyj-!lei, = JIl" j = n, 2,
and (25). These equations can easily be solved to yiekB
-1 '- n 2
(;!8)
" t P j-"
0'11'1 + C'U2
X, = (m.t.-
I
-+ a2rl-
2
)12.
The transfer payments needed to implement as a competitive equilibril,m
ahe allocation associated with the weights a. and a2 are therefore
.. I
Ig(1lr1> Ilrz) =2: Jf,(c, - 1) _ 'l'y. n _ Y I'=.
.=1 I 2
2 =t a2 a.
(30)
'z(a., (2) = .1r,(c, -1) -.--)
u=1 - Jl - Yt .. "2
Notice that these functions are continuously differentiable, are bomog.e
neous of degree 1, and sum to O.
The llllique equilibrium of this model is found by setting these transi\!u'
payments equal to O. It is at =(1- Yt)/(1 - yJ. 0'1 = 1. Notice thai tVle
value of the aggregate endowment is finite since
i n,(l + i) = 25: (1 - Y. ia-
I
+
, e 8 ,cD l-yz
2
(3 1)
= - 12 .
There is, of course, no outside money in this model. There is. boweve r,
inside money: consumer 1, who is more impatient than consumer
spends more than his endowment early in his life. later he
less, paying back his debt. In the limit, his mnsumption in each peri()rll
approaches 0 and consumer 2'5 consumption approaches 2.
313
'""
;1)12 TlIYIOlhy J. Kehoe
A:IJ On....... Geueratioas Mood
1n this section we consider an overlapping generations model in which
'lhere is a single good in each period and a single consumer, who lives for
iJWO periods, in each This is tbe model originaUy developed
!by Samuelson (1958) and analyzed extensively by Gale .(1973). B.n the
'DCxt section we discuss more general models.
The consumer born in period 8, t = li. 2,.. . . solves the utility
pcoblem
unax u(G.
(32}
s.t. p,c! + =P,li'l +PHI Wz
We make the same sort of assumptions on u and (WI> W2) as in the
!)JI'evious secliOil. As in the previous model. we can also think of this
r;:onsumer as facing two budget constraints:
q.c: +nr =q,w, } 33)
q,+lc!+I = q,+J W:z + (R +r,)m'.
l1f we normalize the spot prices so that q,...1 :::: q, = 1, divide tlJe second
oonstraint by (1 + r,). and add both together. we l!:aIl produce a single
!!Judget constraint in which P,/P'+I = (1 +r,).
'The solution to this pcoblem is cbara<:terized by (be conditions
aUlD 8 }
"C.. CJojIP =
(34)
au f I
"'.... (C,. C,+l} = A,P,"'1
.
and tbe budget constraint in (32). Given the strict concavity of ii, this
oonsumer has continuous excess demand functions y(p" P.+i) = c! - w,
wheD young and z(p.. PHI) = - Wz wben old. TIle form of the budgea
ooustraint impUes that these functions are homogeneous of degree 0 in
(p" P,+!) and obey Walras's Law:
P,y(p,. P,+I) +p,z(P .. 1',+,) o. (35)
Consider. for example, the case where fI(e!. e!+l) = loge; + y log C;+I
The excess demand functions can easily be computed using (32) and (34).
They zre
p,w, +P,+!"'2 - YP,WB +P,...1 W2 }
y(P"P,.l)
(1 + y)p, w, = (1 + y)p,
(36}
y(p,w, +p,+ , "'2) yp,w.-
2(p" PHI) -Wz = .
(1 + y)PHO (1 + y)P, ... i
InleTtemporal Equilibrium Models
Notice that these functions do indeed satisfy continuity, bomogclI eity
and Walras's Law.
in addition to the consumers born in periods 1, 2,.... there us Ml
ronsumer who is alive only in period 1 and who solves the problem

(31)
s.u. = PI wg + m.
Here m, which can be positive. negative, or zero, is the stock of oolside
money held by generation O. Xf m is non-negative, then it is tasil,
interpreted as fiat money. Even if it is negative, however, tben are
institutional stories to go with it. Think of an institution that makes Cuans
to consumers when they are young. The institution oollects the repay
ments of these loans when the coosumers are old and uses them to nate
loans to the young oonsumers in tile next generation. TItere U f '. off
course, many other interpretations.
this consumer bas preferences for, and endowment of, th(C
IIirst good, we need oot be careful about specifying Uo or The el ;cess
demand function for this consumer is
m
Zo(Ph m}=Pi '
An equilibrium of this model is a stock of outside money mand a
sequence (flh P2, ...) that satisfies the conditions that excess demanj be
equal to 0 in every period:
zJ,.PI, m) +yWu. P2) =0
[n period! 1 and
Z(PH. p,) + y(p" P,... ,) =0 .(40)
UR period 8, 8 = 2. 3, . . ..
One way to compute the equilibria of this model, developed by iaie
(1973) and Cass. Okuno, and Zilcha (1919), is to use the offer curve, abe
wage of (y(p" P,... '), z(p" PH')' This curve passes tbrough the oi'lgin.
stays always m the second and fourth quadrants. and intersects rrays
tbrough the origin only once (except at the origin itself). In fact, Wld,:as's
!Law (35) tells us that
z(p,. P,+8)/Y(P" PHI = - P,/P,+I; (4ll}
that is, the point where it intersects the ray with slope - p,!p,+. its
coordinates excess demands at (Pit P,+1) In addition, the oller c.uve
always y> -WI and z> -Wz.
For example, in our simple log-linear example, we can use the fOf1l!tui8)
11'or yep,. 1"+1) in (36) to solve for p,lp,+O in terms of)l and substitut<r 1lhe
D
3?.IJ
Timothy J. Kehoe
rresoRt in&o the fonnula for z(p,. P"'i) obtain the oilIer CURVe:
YK'ltIl'2 W2
z= -- (412)
(I + y)2y + (1 + y)yw! 11 + y.
TilDe result is pictured in FIgUre 16.2.
Kn general, there are two steady Siaaes, inOation fJ > 0, such
tltat the price sequence p, =fJ' satisfies
z(fJl-I, 11) +y(fJ', 11'+1) =: z(ll. fj) +y(li. P) =@.
These are given by the two intersections of aile offer curve with the tiiDe
\lRurougb the origin with slope - n. z = - y. There is only one steady stale
film the degenerate case where the slope of the offer curve is -! a t!he
origin.
The steady state where fJ =1 Pareto-dominates the steady state the
origin. One way to see this is to show that the consumption plan found by
solving the representative consumer's maximization HKoblem when P, =
PHR also sotves the problem of maximiziog the utiliay of steady-state
oonsumption plan:
max 14(C1,

lS.t. CR YCa= Wi -+ MIa
2
,
i
II

Y
A. --

lnleJ1emportal Equilibrium Models 3" 5
'
Alternatively, notice ihai, since DO trade ns always feasible, tbe consum!f
can only be better off if be chooses to tnde. Indeed, a simple reveall':d
preference argument implies that the consumer prefers the net trade
[y(l. 1), z(l. 1)] to any point that lies on or to the left of the line wi ,1b
slope -1. (Look at Figure 16.2 again.)
To compute equilibria besides the two steady stales, we start m i lu
= mIPI and read horizontally to me line with slope -1 to find the Vail Ie
of y for which y(fi .. pJ = - ZG. We then read vertically to the offer culve
to find the point [Y(PI> Pz). z(fi" 1'2). We now continue by readir lg
horizontally to the ray with slope -! to find the value of y for willi( h
y(P2. fi3) = - z(P .. P2). This process is illustrated in Figure 16.3. ne
offer curve in the figure corresponds to the case where )'W, > "2. Noti( e
that, for auy value of sucb that =,;,Ip, < z(l, 1), there is an
equilibrium that converges to tbe autarkic steady state in which tbere s
no trade. (There is a natural lower bound on ';'/PI provided by bt l\!
ahis is indepeodent of the offer curve of (y, z).) The price sequence 5
computed by nonnalizingpl = I, dieD using the slope of tbe line througb
abe origin passing through the offer curve at [y(fiJ, P2), z(fh, P2>J to finj
P2, using the slope of the line through the origin passing througb the offer
curve at ["(P2, P3). z(pz. ;;3) to find p,. and so on. Notice that e'ler l
equilibrium of this model, except for the one that starts at ';'{PI '""
z(1. !), involves nnJIatiolli. At the autarkic steady slate p, which is
\
z


31f
Tunothy 1. Kehoe
negative of the reciprocal of the slope of the offer culrVe at the origin, is
rw.lwz> U.
!Not only is there a oootinuum of equilibria in this example, but outside
money plays a crucial role and equilibria are DOt necessarily Pareto
efficient. Observe that any equilibria that starts with 0< mlfh < z(l, 1) is
l?areto-dominated by the equilibrium with ,;,Ip, = z(l. 1): the first
geilleration prefers the highest Zo possible, and subseqllent generations are
worse off the further they are from [y(l, 1), z(l, 1)1 and the closer they
ar@ Uo autarky. In fact, equilibria with higher m/p,
those with lower starting-points. In the next section we shall see that the
equilibria with iii IPI < 0, altbough not necessarily Pareto-dominated by
equilibria with "'lpI:= z(1, 1), are not Pareto-eflicient. As Shell (1911)
haJS indicated, this failure of the FlTSt Welfare Theorem depends on the
double infinity of consumers and goods. Althougb it is possible to mimic
llB'Iils failure of the Farst Welfare Theorem in a model with aDcomplete
markets, as done, for example, by Cass and Yaari (1966), it should be
stressed tbat it occurs even if all markets are complete.
!Figure 16.4 depicts the offer curve for the log-lioear model where
ytY, < "2. Notice that, for any values of m/P1 such that ,nIpI <0, there is
m equilibrium that to the steady state where {J =1. There is
aBoo an equilibrium that starts with ,nlpi = 0 and stays at the autarkic
si,!8dy state. Here jJ = < 1. This equilibriUDll is Pareto-efficient
Jns. ll(!).(l
intertemporal Equilibrium Models :),17
since the value of the aggregate endowment is finite:
+ + i: jJ'- Wa + M.Iz) == + 191, +L. (w. + Wz). (A 5)
,=2 A - fJ
As we shall see in the nexl section, aU of the equilibria of modleB! we
I?areto-eflicient.
These two examples suggest ihree hypotheses. FlCSt. any indetennimlCY
of equilibrium is connected to inflation if there is positive outside
Second, all equilibrium price paths converge to some steady statc. Thi ."d,
any indeterminacy of equilibrium is associated with a non-zero stad, of
outside money. We now study counter.examples to the tirsi
propositions. In the next section we shan see that the third, although iliJle
in any model with one good in each period and consumers who live f1JI1l'
two periods, fails in more general models.
The log-linerar examples that we have analysed have the property
as the price ratio P,/P,.-I increases, ,(P.. PHI) decreases and z(p,. P' d )
increases. This means that the demand functions , and z exhibit gf( ,35
substitutability. Consider the offer cllrve depicted in Figure 16.5. HErre
gross substitutability fails in ihe backwards-bending scdion of the
curve. Notice that, for any value of ';'/P1 sufficiently close to I).
there is an equilibrium that converges to the steady state where fJ = R.
The crucial feature of this offer curve is that the slope of the offer cut' "e
z
y
1F!tz. M.5
318
Timothy J. Kehoe
aa (yill, 1), z(l, 1)1 is positive and !ess than one. There are also equilibria
that start with m/PI near, or even equaJ to. z(1, 1) and converge ao the
autarkic steady state: whenever there are two vaiues of z that correspond
to a single y, we have a choice of two ways to read from the line with
siope -1 to the offer curve.
'Jl'"hls example also has equilibria that do not converge to any steady
Consider the offer curves in Figure 16.6. Here there is a two-period
qcle Zo, ZIP Zo, Z" The second offer curve is the reflection of the
fiBt across the line with slope -1. Cyc8es are points where these two
z. -y
)'. -z
fig. RUi
cwves intersect, where
(y(p" p,), z(p" p,)) = - Z(P2, PI)' - Y(P2, PI)}' (46)
This implies that mand (PI' P2. Pa, h, ...) ate an equilibrium of this
\!D00Ie1. The possibility cycles in this sort of model was first pointed out by
Gale (1973). Benhabib and Day (1982) have shown that there are . also
examples with equilibria that do not converge to any steady state or 10 en
cycle of any length. The possibility of such strange behaviour, often
rrefcmd to as chaotic dynamics. has been extensively by
f,.
(1985).
intertemporal Equilibrium Models : 19
4. GeaenU O"erbppiag Geaendioas Models
We now lum our attention to overlapping generations models with mllll1lY
goods in each period and many consumen; in each generation. If we aU' iW
many goods and many consumers, the assumption of two periods of life is
completely general: BaJasko, Cass, and SheD (1980) present a
procedure for redefining periods and generations that converts a model m
which consumers live for any finite Dumber of periods into one in wbk:h
they live for only two. Suppose that consumelS live for k periods. ThmlJ
redefine generations so that generations -k +2, -k + 3, ... ,0 becon :e
generation 0, generations 1,2, ... " -) become generation 1. and : 0
on. Redefine periods in the same way. Figure 16.7 illustrates this
procedure for the case k =4. Notice that each generation lives for ju'lt
iWO redefined periods. If there are n ,0005 in each origioal period, theue
are (k -1)n goods. indexed by date. in each redefined period. If
are h consumers in each original generation, there are (k - U)'1
consumers in each redefined generation.
The model with many consumers and many goods has the sam;
potential for equilibria that are Pareto-inefficient and equilibria witl Q
unbacked nominal debt as does the simple model of the previous section ,
234 567 8 S
It 0 0 0 0 0'0 0 0 I
It C 0 co: 0 000
X It I 0 Q 0: 0 ij 0
--------r-------r------
x x It I X 00 1 0 DO)
I I
o X X I X X 0 I 0 0 0
G x'
I
A X X I
I
ij u 0 ;
_________L _______
o DO:X x XIX CO l
I ,
000(0)( ltlX X C 2
, ,
0 1)( o 0 I
L
G
_______
0 It
, ________
X )(
_ ___ .______
o 0 0 I 0 c 0 l x x )( I
o 0 0:0 0 0:0 X)( 3
I I
o 0 0:0 0 0 10 0 )(
o I
2 3
Period
1fiI. nlu
-2
-1
o
2
c:
..g 3

4
5
6
7
8

300 Trmothy J. KehGtE
Hi hali e'Ven more potentiaJ for indeterminacy of equilibria. Consumer j nD
generntioo , solves the problem
max ",(yj +IN" + wz}
u. ,,:1. +,,;..z! =O.
(47)
lHere )Ii> Z{, "'I. Pt, and PHI are aU n-dimensionaB vectors. if bis
0(p J
exce5 demaDd functions are Y IJ Pt+;) and z (p" P,+I)' then the
aggregate excess demand functions for generation 8 are y(p" P,+I) andl
z( DIl" 81Hol) where, for
r - h
y(p" P.+.} .. Ly/(p" PHI} {48)
i-I
We that y and z are conlinuousiy difJerenoabJe fOfr ail strictly
positive vectors (P,. P.+I). are homogeneous of degree 0 iii
.. f !'+i)' and obey Walras'slaw,
P;y(p" P,+n) +P:+IZ(P" P,+l} !!!!ij. (49)
JlKn arllditioo, there is an old generation. alive only in the tina period. abaa
!bas tl11e aggregate excess demand function .. m). We assume that Zo is
oontifJlUOusly differentiable for aJl strictly positive'price vectoq PD and an
open interval of moaey stocks m that includes 0, is homogeneous of
ciiegree 0 in (Pit m). and obeys Walras'$IaVi.
p:Zo(p., m) liE m. (SO)
Am equilibrium of this model again is a stock of outside money mand a
sequence of price vectors (PI' P2...) that satisfies (39) and (40) where
Ilhe 'Yariables are reinterpreted as vectors. To see the possibility of
indeterminacy. let us count equations and unknowns in the equilibrium
ronditioDS. The equilibrium amdition in the first period,
z.{J .. efr)+ y(.Pi.Pa) =0, {Sn
rontains $'I equations in 2n + 1 unknowns. Since the equations are an
homogeneous, we can impose a normalization to reduce this to 1n
IlUlkIilOWDS. The equiJibriam conditions in subsequent periods,
Z(p,-i, p,) +y(Pl> P.+I) ::: 0, g = 2, 3, ... , (52)
each add n equations and n unknowns. The entire system therefore has n
degrees of freedom. If we set ", =0 II priori, there is one fewer unknown.
and this reduces the degrees of freedom io n - 1. The idea is tbat we
choose m, Ph and h to satisfy (51) and then use (52) as II nonlinear
difference equation to determine Pl, P . ..
'Jrilte problem with simply counting equations and unknowns is tbat we
do 00i aJways know wbether we can use {52) to continue an equilibrium
prire sequence for arbitrary (Pit P2). lin /Figure 16.3, fOIr example, fif we
i nlertemporai Equilibrium MOlkls
2':H
start with any value of m/PI above z(l. I), we can continue li lle
equilibrium for a few periods but eventually we reach a situation whf re
we cannot continue because z exceeds "'I and there is no offer ClU"Ve
read to vertically! In general. we want to situations where \!Ie
cannot use (52) &0 compute a positive value of P'+I as a function of P, _I
and P, . One way to do this is w require that the equilibrium
sequence converge to a steady state at which the matrix of part ial
derivatives of P,+I) with respect to P'+I is non-singular. This implies
that in some open neighbourhood of the steady state, for fixed (P,- I, pll}'
tbe function Z(P'_I. Pt) +Y(P., .) is invertible. 1be implicit functiullI
theorem tells us that in this neighbourhood PHI can be compuir"!d
uniquely as a function of P,). Restricting our attention to tHs
neighbourhood of the steady slate, we can avoid the problem iUustTatuCi
in figure 16.5. where tOOre may be more titan one P.+. that satisfies Rile
equilibrium conditions. 1bis restriction may force us. however, to
some equilibria.
A steady state of this model QS lR vector of relative prices p and! <lBi
inflation factor fJ such that P' == (52). There are two types :)JI
steady states: nominal steady states, in whidt there is a DOo-uro arnolD III
of nominal debt transferred from generation to generation, and reaH
steady states, in which there is no such transfer. Notice that in iili l y
equilibrium the amount of nominal debt transferred from generatiulIl
to generation stays constant over time: (SO) and (51) imply tltaQ
-P;Y(Pe. pz} == .. Ifr); Walras's Law implies that piz(Plt P2) =
-p;y(p., pz); (51) impJies that - P21(P2, Pl) = P2); and 50 OiD.
"The steady-state condition is
zur-ip, fJ'p) +y(fJ'p, =z(p, fJp) + yep, flp) =o.
This implies that p'z(p, (Jp) +p'y(p, (Jp) ... O. Walras's Law implies tlHa
p'y(p,(Jp)+!Jp'z(p,!Jp)=O. Subtracting one from another, we obtailll
(1-,8)p'z(p, fJp) ... o. This says lbat fJ =1 at any nominal steady stat.! .
Kehoe and Levine (1984b) prove that almost aU economies are such dun
fJ :1= 1 at every real steady state.
Balasko and Shell (1980) and Burke (1987) have shown that ZI
necessary and sufficient condition for Pareto efficiency of ao equilibril!'l!i
is abat ..
L I/p,,,-I =00.
(5l j)
.=1
Here IIp,11 =(p;p,)lf2. the standard Euclidean norm. They impose a!
nniform curvature condition 00 indifference surfaces that is natural in
stationary environment. Notice that any equilibrium that converges !to
steady state where fJ> K, an inflationary steady state, is
since the sum in (54) converges. Any eqqjJjbrium that converges Ro
steady where fJ l!ii: n. howcve!l. is Pareto-efficient since ihe SUlllll i Illl

382
Timothy .11. Kehoe
(54) cJIiverges. In fact it is easy to show that if fJ = 1 ahe equilibrium
allocaoon maximizes a weighted sum of utilities of the consumers in III
representative generation subject to steady-state consumption
constraints.
When there are many goods in each period and many consumers in
each generation, there is no need for there to be a unique nominal sleady
state and a unique real steady state as there are in the example of the
previous section. Even with one good in each period, but more than one
consumer in each generation, there can be multiple real steady states,
a1thoogh there is a unique nominal steady state. Consider, for example. a
static exchange model witb mUltiple equilibria. Such a model
is ea5l' Uo construct in an Edgeworth box; see Shapley and Shubik (1m)
aIlI example. Now convert this into an overlapping generations model
in which there are two consumers in each generation with the same
preferences for and endowments of the two goods in the two periods of
heir lives. The multiple equilibria of the static model are real steady
states of the overlapping model in which each consumer
trades only with the other consumer in the same generation. This
iilustllltes the point that real steady states are not, in general. autarkic, as
ilieyare in the simple model. With many goods in each period. not even
nominal steady states need be unique. Kehoe and Levine (l984b) prove,
Iliowe"lfcr. that in general eNery economy bas an odd-in particular, a
non-zero-number of oominaJ steady states an odd number of real
Sieady states. Furthennore, the matrix: of partial derivatives of y with
respect to its second vector of arguments is almost always non-singular at
every steady state.
To analyse the behaviour of equilibrium price sequences that converge
eo steady state, Kehoe and Levine (l985a) linearize the equilibrium
oonditjons (51) and (52). The local stable manifold theorem of dynamical
systems theory says that the behaviour of the nonlinear system near the
steady slate is qualitatively the same as that of tbe linear system (see
[Win 1980). They consider the set of price pairs (PI. P2} that satisfy the
equmbrium condition in the first period and lead to convergence to the
steady state when employed as starting conditions for the nonlinear
difference equation (52). This set is a manifold. a set of points that is
Bocaliy equivalent to an open subset of a Euclidean space of dimension
than 211. (The prototypical manifold is a linear subspace.) Kehoe
and !Levine demonstrate that this manifold can bave dimension as large as
Fe if is outside money and as large as n - i if there is no money.
This manifold can also have dimension as small as 0, un which case il
consists of isolated points. (The best linear approximation to this
manifold near the steady state is the intersection of the stable subspace of
the uinearized version of (52) with the sel of vectors that satisfy the
linearized version of (51) .) Almost all economies are such that any small
l mertemporm Equilibrium Models

perturbation produces an economy with the same qualitative propertil$.
Kehoe and Levine also prove that there are robust examples of stea 'ly
stales with no equilibria at all that aH\verge to them. This cannot
with only one good in each period because Walras's Law implies that m
and p can be chosen so that the sleady-state price vector (P. /Jp) satisfies
Zo(p,m)+y(p,/Jp)=O. Consequently, the steady slate itself i ; l91l
equilibrium.
Notice Ihat we can use a similar irick to that used to convert economi es
with consumers who live for k periods inro economies in which they Ii'/e
for two to convert the study of equilibria that converge to cydes of aLlY
finite length into the study of equilibria that converge to steady states .
Suppose that an economy has a k-period cycle in the sense tin t
(PH .. . P,+Ic) = (fJ'PIo . . , {J'Plc) satisfies (52). Redefine generatiolils
so that, for example, generations i , 2, ... k become generation
Similarly, redefine goods. A k-period cycle is now a steady state of ni e
redefined model.
De RbnIiu EqlliYaIeKe 1IbeOl'elll .
In 18!7 Ricardo asked the question. Does it make any difference whetht r
a government finances an increase in expenditure by raising taxes or h,\,
selling bonds? (See Ricardo 1951: 244-9.) The simple answer that
came up Wilh, although he realized that there were complications, W" $
t hat it makes DO difference. because consumers anticipate that they hav
to pay more taxes in the future if there is a bond sale SO that
government can malce interest payments. This is at odds to Keynes';
answer to the &arne question; that a bond-financed increase in govern
ment expenditure has the full multiplier effect, but that a tax-financecl
increase has a much smaller balanced multiplier effect. 1be crucia l
distinction between the two analyses is that in one consumers' saving. ;
behaviour is altered by the bond issue and in the other it is not. HI
reduces to, as Barro (1974) puts it, Are government bonds net wealth?
!Let us first answer this question using our model with infinitely live"
consumers. We introduce into that model a government that
goods g, = (gl" ... ,gN) in period I, t = 1, 2, . ... We require that thit
expenditure pattern be feasible in tbat
III
0". g, ".}: wi.
6=I , 2. _ "
(55)
; =1
Suppose first thaQ these purchases are financed by lump-sum
-g: ... r,o so that the government budget balances in every period:
h
L -r{=p;g" g = 2. ...

J=!
185
J8.e.
Timothy J. KehO
Then the budget consuaini faced by coosumer j as
.. ..
L p;d, =L (p;wi -
(57)
1'*1
uppose, on the other hand, that the governmenq issue bonds b"
8 = U. 2, . that pay interest at the competitively detennined interesi
rate. it finances these interest payments by lump-sum taxes w,. The
government must balance its budget in tbe sense tbat the presen!
dis.oounted value of its expenditures is equal to the present discounlecl)
value of its revenues:
.. "(" )
h,=
tID _
(58)
LP;8, L L Wr
,=1 '=1 j=1
1:;'=1 h, shows up on both sides of the budget constraint since tbe presenft
disoounted value of Ii bond is equal to sum of the interest payments on iu.
The consumer's budget constraint becomes
- GO ..
L (P:c, +bO = L (p;wI w,) + }: II.
11-1 ,-I 1-1
(59)
.. ..
L P;c, =L(P; wi - 11,).
,_1 ,=1
Here is the net purthase of bonds by consumer j in IJ)eriod g and!
" (OO)


Notice that, if
(61)
L -G= 2 fY"
,=1 ,.:::1
two models are identical in their essentials. in particular, tbe agenis
face the same budget constraints. This is the Ricardian Equivalence
Theorem. There are a number of important maintained bypotheses.
First, there are perfect capital markets. .This implies that each consumelT
faces a single budget constraint. Second, all taxes are lump-sum;
1'\,\,<,.1\ otherwise, relative prices would be distorted in different ways by different
. . _" taxation scbemes. taxes are not redistributional. In other words,
consumers face the same total tax bill under the two taxation scheme;
oaherwise, relative prices would change because of income effects.
We are not claiming that the equilibrium is the same as ir g, = 0, 1=1,
2, ... Since the government is consuming some of the goods tbat would
oiliet'wise have gone to consumers, this cannot be the case. Governmeni
fiscal policy always bas real effects. II is the way it ns financed that Ds
arreievant.
lmenemporm Equilibrium Models
Hi is difficult to give the Ricardian Equivalence Theorem an ioterpn tao
tion in an overlapping generations model: alternative tax schemes t\.'la t
!ime tax collections differently necessarily have redistributional effe,cts
because consumers are alive at different times. There are very :ial
situations in wbich differeut tax schemes do not affect the equilibricn HQ
does not matter, for example, in which period of life ronsumers pay ta'Ies
as long as each consumer faces a single budget constraint, all taxes ;u-e
lump-sum, and each consumer faces the same total tax biD under '.he
different schemes. Rather than say that tbe Ricardian EquivalcliICe
1beorem does not hold in an overlapping generations model. we shollM
say that the range of tax schemes that do not affect the equilibria is much
more limited in an overlapping generations model than it is B!ll 2lJ1
infinitely lived consumer model.
Another problem with interpreting the Ricardian EquivaleMCtl
Theorem in a model with infinite numbers of consumers and goods is, as
we have seen, that there is no reason for !;;'_I p;g, to converge. The
government can therefore issue bonds that it need never pay back. Th!Se
bonds act like injections of outside money and are, tberefore, net.W8fl&: 'l9leA.
figure 16.8 depicts an example with a steady state in which 8, = g :> 0
every period. Here inflation erodes the value of the initial stock
outside money al the same rate as that at which the value of the
lFig,lIM
386 Tunothy J. Kehoe
stode of government bonds increases. The total real stock of nominal
debt. outside money and bonds, remains constant at z +g. 1l1e steady
interest rate is r =liP -1 <0. Notice that, even though the
government is consuming g > 0 of the single good in every period, this
equilibrium Pareto dominates the autarkic equilibrium where it consumes
notiling. Examples of this sort are discussed by Sargent (1987: 0. 7).
Barro (1974) has argued that the Ricardian Equivalence Theorem
holds for overlapping generations models in which consumers include
their otrspring's utility in their own utility functions. Since their offspring
similarly value the utility of their own offspring, this can make 1Il
consumer's utility maximization problem into the problem of maximizing
utility of an infinitely lived family. 1be problem with this story is that,
in general. we have to allow some consumers io pass on debts. as well as
bequests, to their offspring. In tbis case we would want consumers io
include their progenitors' utility in their own utility functions. Think of
our example of the two infinitely lived consumers with log-linear utility
fUlllctions as a model of such families. One family of consumers
asymptotically consumes notbing. They use almost aD of their income to
service their family debt. which they inherit (com their progenitors and
pass on to their offspring. This sort of problem always occurs if different
families have different discount factors in reduced-form utility
fuiilctions. Institutional arrangements in modem societies make this
feature of the bequest story very unrealistic. As Barro himself points out,
nf a is at a comer solution because of a non-negativity oonstraint
00. bequests. the family faces a sequence oR" budget coostrainls that
cannot be aggregated into one.
a model with infinitely lived consumers who face liquidity
OO!JlsQraints can have similar chnacteristics to an overlapping generations
model (see Woodford 19863. ior example.) If we cannot reduce the
consumer's utility maximization pt"oblem to one with a finite number or
budget constraints. then we cannot prove that the value of the aggregate
emoowment is finite. Consequently, equilibria Deed not be Pareto
efficient, and there may be equilibria in which outside money plays 2
rroIle. Even our argument that there is a finile nwnbu of equilibria faUs
apart. The essential feature of that argument is that each consumer
characterized by a single Lagrange multiplier A
J
=1/aj . Rf the consumer
caHUlOt equate his marginal utility of income in diJlerent periods, then he
aci>:S, to some extent, like a sequence of dillerent consumers. There may
be iJ robust continuum of eqUilibria, and the Ricardian Equivalence
11'!eorem need not hold.
Ii. rOl" FWIe Models
What does our analysis of the overlapping generations model tell us
about the properties of models with large. but iiinite. number or
//nlerlemporlll /Equilibrium Models
381'
consumers and! goods? Suppose that we truncate the modei aa 5 )lIII1e
period T using a terminal young generation Yr(PT, m) analogous to the
initial old generation Zo(P.. m). Outside money now corresponds ', 0 III
transfer from the terminal young generation to the initial old. TItef e is
now a finite number of equilibrium conditions:
zJ..Ph if!) +Y(Pt. P2) = 0
z(p" P2) +Y(h, Pl) = 0 62}
z(Pr-I' PT) +'!T(ftn ria) =O.
All of the equilibria of this model are Pareto-efficient. In general. tiller: is
a one-dimensional continuum of equilibria indexed by the real tran,.fer
payment tn/\lftdl.
This method of truncating this model is often equivalent to specil1il!llg
expectations of prices in periods after the model ends. For example, 'We
could specify YT(Pr, m) as YT(Pr, IIPTII (Jp) where (P, P) is a steAdy
state. Here, of course. m=: IIPrll {JP'Z(PT7 IIPTII (Jp). (See Auelrbac!1l.
KotJikoff, and Skinner 1983 for an application of this approach.)
Consider a situation where one equilibrium Pareto-dominates anol ber
in the infinite horizon model. Each of these can be made an equilibr ium
of tbe truncated model with a suitable choice of Yr. Since both of !the
equilibria of the truncated model are Pareto-efficieot, the equilibr:ulIIIll
that dominates in the infinite borizon model must assign some memhers
of the terminal generation lower utility tban does the inferior eq ui
librium. Notice tbat the functions Yr do not necessarily bear l illy
relationship to utility maximization by generation T illl the infinite
borizon model. If T is large enougb. the model is clear: by sacrificing the
welfare of one generation, all otbers are made beUer off, and socie1ly . \5 a
whole is made better off from a utilitarian viewpoint.
In an infinite horizon model there can be n dimensions of indeler
minacy if there is outside money and n - 1 dimensions if there is not. '(he
single dimension of indeterminacy that shows up because of fiat mOl ley
conesponds to the indeterminacy parunetrized by the real tnuder
payment. What about the otber dimensions? To answer this question, let
US suppose that we have two equilibrium price sequences. (Pt.P2, . ' .)
and (Ph P2, .. ), which both converge to tbe same steady state. Supp>se
too thaI both involve tbe same real stock of outside money, .
WI m
--1:1-
lIP111 lIPIII"
If we truncate using a tennioai young genention
1TWr. m) = YT(PT, fin-I)' {1i3)
m =iii, (Pn. P2 . , [fly} is an eql!ilibrium. If we truncate wiRb l he
388
-
Tunothy J.
Pl fp,
pi I
T
lFII. n631
analogous choice of YT. then (PI, PZ.. fir) is an equilibrium. Figure
6.9 depicts this sort of situation. For large enough T. ({Jr. PT+') is going
to be arbitrarily close to (PT. (IT+I) no matter bow far apart are PI and
Pn. indeterminacy of equilibrium therefore corresponds to sensitivity to
terminal conditions. sensitivity of initial prices that becomes more acute
as the time horizon T becomes larger. See Kehoe and Levine (1986) for
nwneuical simulations of an eumple with this propery.
We should point out one other way of reducing an overlapping
generations model to a model with a finite time horizon. Suppose that in
eve!)' period the probability that the world ends before the next period is
P. (]I < P < 1. It is then natund to assume that consumer j in generation g
solves the expected utility maximization problem
WI! + (1 - p)vlr.+ Wa. WZ)} (64)
S.t. p,1, +P,+lz, = O.
Here is his utility function if the world ends before the second period
of his tife and Vj is his utility function if it does not. Even though the
world ends in finite time with probability 1. this model is identical to an
overlapping with an infinite horizon. It may have equilibria that are not
rareto-efficient, equilibria in which outside money plays a role. and
equmlJ)ria with one or more dimeosions of indeterminacy.
lntertemporllK /Equilibrium Models
1. EueasioDS aaci CoIIdasioIls
The results presented in this paper can be extended ao more
models. Kehoe et aI. (1988) have exteoded this analysis of tbe model m( ill
2 finite number of infinitely liyed agents to similar models that
production and capital accumulation. The only difficulty is in ensuwlg
tbat the transfer functions used in the equilibrium conditions ale
continuously differentiable. Muller and Woodford (1985) have extended!
ihis analysis of the overlapping generations model to models that indue! e
infinitely lived assets. and production. They find that
presence of infinitely lived consumers or infinitely lived assets can forc e
the value of the aggregate endowment to be finite. This rules out Paren!)
inefficiency of equilibria and outside money_ It does not rule ot ,fl
indeterminacy of equilibria. however.
Do Pareto inefficiency of equilibria and outside money depend! ()i'1l
there being an infinite number of consumers or on some
baving finite life-spans? Kehoe (1986) considers a simple
model in which there is an infinite number of consumers woo all live fClT
ever. This model has equilibria that are Pareto-inefficient and equiJibni'l
with outside money. It also has equilibria with several dimensions () f
indetenninacy .
As we have seen. indeterminacy is a U'elatively separate issue f!rolia
Pareto inefficiency and outside money. Kehoe et Ill. (1986b) consider
abstract model with infinite numbers of and goods. ]be onU,r
prices that are allowed assign finite value to the aggregate endowmen,
This rules out Pareto inefficiency and outside money. Even so, there ar(;
Il'obust exarnple& with any dimension of indeterminacy. The reason
this indetermioacy is tbat we cannot reduce the equilibrium conditions
a finite number of equations and unknowns. These authors also find tha l
&here is a finite number of locally unique equilibria if consumers am
similar enough. This generalizes our results on economies with a
number of consumers.
Santos and Bona (1986) and Geanakoplos and Brown (1985)
extended the results of Kehoe and Levine (l985a) for stationary ,
pure-exchange, overlapping generations models to models with 00IiJ '
stationary structures. Like Kehoe and Levine, these authors need
restrict their attention to equilibria that remain close to each other I
some sense. They find that, even in a noo-statioaary environment, there
are n dimensions of potential indeterminacy if there is outside money aM.
1'1 - 1 dimensions if there is
One disturbing aspect to the potential indetennioacy of equilibria ir,
ahat it occurs for some values of the parameters of a model but not fOl
others. We would like to somehow classify the parameter values rot
390
Timothy./J. !Kehoe
whiclhl indetenninacy does not occur. A first step in this direction has
!been tak.en by Balasto and Shell (1981), who consider a model with
many goods in each period but a single two-period-lived consumer with a
Cobb-lDouglas utility function in each generation. TIley prove that there
Us HUG indeterminacy without outside money and only one dimension of
ondererminacy with it. Geanakoplos and Polemarchakis (1984) have
shoWlII that the essential feature of this analysis is that tbe single
uwo-period-lived consumer bas intertemporally separable preferences.
[{eooe and !Levine (l984a) have shown further that any small perturbation
m model with a single two-period-lived consumer with intertemporally
sepauable preferences, even if it introduces small heterogeneities among
consumers or small interdependencies in consumption ove/!' time, results
ma IlVIOdeI with these same features.
A more significant finding is that of Kehoe el m. (l986a), who cousideJr
geneiJ'ai pure-exchange, overlapping generations economies with many
goods in each period and many consumers in each generation, ill which
all demand functioos edlibit gross substitutability. TItey prove that there
n5 a i1lnique equilibrium if there is no money; although there may be 3l
one-dimensional indeterminacy with outside money, there is at most one
equiiibrium for each level of real outside money in the first period.
!Furthermore, their analysis is global rather than local. If the economy is
stationary, then there is a unique nominal steady state and there is il
wniqK real steady state, and every equilibrium converges to one of them.
Unfortunately, there are plausible examples that violate gross sub
stitullability. Kehoe and Levine (1986) consider a model with a single
,oociI in each period and a single three-period-lived consumer in each
generation. 1bey give tbis consumer a constant elasticity of substitutiollJ
function and show that, for plausible parameter values, this model
indeterminacy without outside money and more than one
dimension of indeterminacy with it. Moreover, they choose the crucial
parameter. the elasticity of substitution in consumption over time. to
agree with empirical evidence.
nne present analysis has on the differences between models
witlu a finite number of infinitely lived consumers and overlapping
generations models. Yet these two types of models have important
properties in COmmon. In both, for example, equilibria always exist.
Sinoe the equilibriwn conditions for a model with a finite number of
nnfiniteJy lived consumers can be transformed using Negishi's (196Ob)
approach into those of a model with a finite nwnber of goods, it is
stf'anghtforward to prove the existence of equilibria in such models. This
is done, for example, by Kehoe a al. (1988). Proving the existence of
equilibria in overlapping generations models involves more subtle issues.
Considering the limit of a sequence of truncated economies. Halasko et
!lit. :J.1980) prove the existence of an equilibrium with riB = 0 in :m
Kmertempoyai EqlUiJibrium Models 39)'
pure-exchange, overlapping generatiQns model. For general models witlh
countably many consumers and goods, Burke (1986,1988) and Wilsoll
(1981) have proven the existence of equilibria. The presence Qf outsfide
money may be necessary, however, for an equilibrium tQ exist.
Another property that these two iypeS of models have in commolll
tbat the Second Welfare Theorem holds: any Pareto-efficient allocation
can be supported as a competitive equilibrium witb transfers. This
proven fQr tbe overlapping generations model by Balasko and
(1980). The role that outside money plays in supporting a
allocation can be interpreted as such a transfer. Unfortunately, Cass et (61.
(1979) and Millan (1981) have examples in which no
allocation can be supported as a competitive equilibrium by giving aJ
transfer ooly to the first generation. Burke (1987). however, shows thaQ a
transfer to the first generation does support efficiency if it is foUowed by :at
sequence of taxes on subsequent generations. Furthennore, the sum
real tax payments can be made arbitrarily smaU.
How much of this analysis extends to intertemporal models with
uncertainty? If all markets are complete, titen the analysis of the modlefi
with a finite number of infinitely lived consumers remains the same. Hill
particular. all equilibria are Pareto-efficient, there is no role for outside
mouey, and there is generically a finite number of equilibria. Goods are
indexed by histories of states of nature as weD by date. (See Kehoe anell
Levine 1985b for an analysis of a model of this sort.) In a stoch.asftic
overlapping generations model the assumption of complete markets iis
unnatural. however. In a deterministic setting we have argued thai on
makes no difference whether aU trade takes place in the first period or
takes place sequentially; in a stochastic setting setting this is no longer tbe
case. Consumers would want to make trades in periods before they are
born tQ insure themselves againsa being born into UDfavourable clilfo
cumslances. Dutla and Polemarchakis (1985) present an analysis of aJ
simple stochastic overlapping generations model and show the differerK:e
between equilibria with complete markets and equilibria where COlll
sumers are allowed to trade only during their own lifetimes.
As we have seen, models with infinitely lived consumers who fue
incomplete markets have similar properties to overlapping generations
models. Bewley (1980, 1983), Scheinkman and Weiss (1986), and Levine
(1986) analyse simple stochastic models in which there are infinitely livecll
consumers who are constrained in their borrowing and lending decisions.
Not surprisingly. they find that such models have equilibria that arre
Pareto-inefficient and equilibria in wbicb outside money plays an impor
tant rolc. Presumably, these models also have indeterminate equilibrim.
but this property has not received much attention.
The most worrying property of the overlapping generations modeD 8s
probably its potential for indeterminate equilibria even if there ns 1l1\i!J)
39.1

" i
3 9 L ~ Timothy J . . Kehoe
ou;lside money; There are two reasons for this. First, indeterminacy
nwJces the model unsuitable for comparative statics analysis. Second, it
makes the concept of perfect foresight problematical. Multiplicity of
equihoria of any sort presents difficulties for an economist interested in
usnng a model to do a comparative-statics analysis of the impact of s
change in parameters. Suppose, however, that a model has a finite
lIIumber of locally unique equilibria that vary continuously with its
parameter values. (Almost all static general equilibrium models possess
these properties.) Then the economist could hope that, by appealing to
history to justify focusing on one particular equilibrium. and to a (usually
unspecified) dynamic adjustment process to justify focusing on .the
displacement of that equilibrium after a change in parameter values,
comparative statics Still makes some sense. Even these hopes vanish if
there is a continuum of equilibria.
The idea underlying perfect foresight in a model with no uncertainty is
the same as . that underlying the rational expectations hypothesis in a
mooel with uncenainty: the agents know the structure of the model and
use it to predict the relevant values of future variables. If the model does
not make determinate prediction, then hypothesis of perfect foresight
becomes less attractive. If there is a continuum of perfect foresight paths,
the theory is incomplete. Geanakoplos and Polemarchakis (1986) argue
that indeterminacy leaves room for factors like fixed nominal wages and
animal spirits of investors. AF. we have seen in our discussion of the
Ricardian Equivalence Theorem, if there is a continuum of equilibria,
some may bave Keynesian features and some may not.
One way to try to make the theory complete would be to fix the values
0): some variableS in the first period, for example the real money stock or
a relative price. Even this approach tails if y(p, , PHI) is not always 'an
mvertible function of p,+ 1. With the backwards-bending offer curve in
Figure 16.5, for example, there is an infinite number of equilibria even if
we fix the value of mlpl: at every point where there are two values of
P!+l such that z(p,-t. p,) + y(p" Pt+I) = 0, we have a choice of a different
price path to foUow.
Modelling expectations has long been a difficulty in economic theory.
Keynes (1973/1936), for example, realized the importance of expecta
tions formation, but claimed to work with a model in which the time
period was short enough so that expectations could be taken as
exogenous. The simplest way of making expectations endogenous is to
nl8ke them adaptive as done by, for example, Friedman (1968) and
Phelps (1967). The equilibria of the overlapping generations . models
would be generically determinate if we specified expectations as either
exogenous or adaptive: since values of past variables can be taken as
exogenous in any period, the equilibrium conditions reduce to a system
of a finite number of equations in the same finite number of unknowns.
lnlertemporal Equilibrium Models
Computing the equilibria of such a model would reduce to computing the
equilibria of a sequence of models that look like static models.
The indetennmacy of equilibria in the overlapping generations models
is aU the more wonying because it can be aSsociated with the existence of
self-fulfilling prophecies. Even though the prefererices, endowments, and
technology of an economy are detenninistic. a random variable can affect
the equilibria merely because agents expect it to. This phenomenon is
referred to as a 'SUDSpot', although actual sunspots may actuaUy affect the
technology of an economy (see, Mirowski 1984), and may not be
themselves stochastic (see, Weiss 1985). There is a large and growing
literature on sunspots. A very incomplete list of references includes:
Azariadis (1981). Azariadis and Guesnerie (986), Cass and Shell (1983),
and Farmer and Woodford (1984). Woodford (1986b) presents an
example in which agents employ a simple leaming rule and the economy
converges to a perfect-foresight sunspot equilibrium.
Just as worrying as iodetenninacy of equilibria is the possibility that aJJ)
economy may have no equilibrium that converges to a steady state. If the
path followed by equilibrium prices is chaotic or periodic of a very long
length, the perfect-foresight hypothesis is unattractive for a different
reason: it requires too much computational power of the agents of the
model. Any theory of expectations fonnation that is designed to cope
with the problem of indetenninacy of equilibrium must also be able to
relax the requirement of perfect foresight when equilibrium price
dynamics are chaotic or periodic of very long length. Unfonunately, as
Benhabib and Nishimura (1985) and Boldrin and Montrucchio (1986)
have shown, even the model with a finite number of infinitely lived agenks
can have equilibria that exhibit periodic or chaotic dynamics.
The above analysis of intertemporal general equilibrium models has
provided us with a clear understanding of why Pareto inefficiency and
outisde money occur in the overlapping generations model but not in the
model with a finite number of infinitely lived consumers. It is also clear
how these properties manifest themselves in a truncated version of the
model. Although we have attained some understanding of the possibility
of indeterminacy, we are stili faced with the dilemma that indetenninacy
is symptomatic of an incompleteness of the modeL What is needed is a
serious theory of expectations formation . .