Вы находитесь на странице: 1из 26

Continuity Properties of the Core of a Market

Author(s): Yakar Kannai


Source: Econometrica, Vol. 38, No. 6 (Nov., 1970), pp. 791-815
Published by: The Econometric Society
Stable URL: http://www.jstor.org/stable/1909693 .
Accessed: 05/11/2014 10:57

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
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.

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

http://www.jstor.org

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
E C O N O M E T R I C A
38
VOLUME November, 1970 NUMBER 6

CONTINUITY PROPERTIES OF THE CORE OF A MARKET'

BY YAKARKANNAI2

1. INTRODUCTION
ONEOFTHEprincipal difficulties in the theory of equilibrium in free-competition
markets is that the "core" does not coincide with the set of all "competitive
equilibrium allocations" [3, 4]. It has been felt since the time of Edgeworth [6]
that this is due to the fact that the action of a single trader has an effect on the mar-
ket. Therefore, people began to consider "large markets."
Two main solutions were given to the problem of the incoincidence of the core
and the competitive equilibrium. Debreu and Scarf [4] proved that if we let the
number of traders increase, the core will shrink to the set of competitive equilibria.
However, they allowed the number of traders to increase in a very special way.
They dealt with a finite number (say m) of different types of traders and considered
markets with mk traders (k traders of each type); then they let k tend to infinity.
This enabled them to compare the cores of different markets (with different k's),
which would be difficult otherwise, since the cores are subsets of Euclidean spaces
with dimensions varying with the number of traders, and there is no obvious way
to compare such sets.
R. J. Aumann [3] gave quite another solution. He considered a market with a
continuum of traders, and proved that in this case the core coincides with the set
of all equilibrium allocations.
Both the above models are only idealizations of the finite case, and it is not
obvious how well they describe actual markets. It is the aim of the present paper
to show that the continuous model may be regarded as the limit of models with a
finite number of traders. A far-reaching generalization (Theorem A) of the Debreu-
Scarf theorem will be obtained. The question of the continuity of the core (in the
continuous case) with respect to the initial state is also elucidated.
One of the first problems to be solved, before one can even formulate such
theorems, is how to let markets "tend to a limit," where not only the number of
' This research has been sponsored in part by the Logistics and Mathematical Statistics Branch,
Office of Naval Research, Washington, D.C., under Contract F 61052-67-C-0094. Reproduction in
whole or in part is permitted for any purpose of the United States Government. A part of this paper was
written while the author was at the Israel Institute for Biological Research.
2 The author wishes to thank R. J. Aumann for suggesting the problems and for some very helptul

conversations, L. Dubins, B. Peleg, and M. A. Perles for helpful suggestions, and L. S. Shapley for
bringing to the author's attention an inaccuracy in an earlier version of this paper (Research Memor-
andum 10).

791

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
792 YAKAR KANNAI

traders and the commodities held by them varies, but their preferences also vary.
A topology is introduced on the space of preferences in order to overcome this
difficulty. This topology (described in Section 3) may have other applications as
well.
In order to overcome the difficulty of comparing markets with differentnumbers
of traders, the notion of "a representation of a finite market" by means of the
continuous model is introduced. Using these representations, we can attach a
precise meaning to the "tending" of markets to a "limit."
The concepts of s-cores, due to Shapley and Shubik [9], are used (though in a
different form). Whereas in [9] the existence of the s-cores was proved for large
markets using transferable utilities and letting the markets grow in a special way
(fixing the numbers of types, as in [4]), the non-emptiness of the c-core (in our case)
follows easily from the existence of the core in the continuous case [1], for a general
way of letting the markets grow.
Other results deal with continuity properties of cores and 6-cores and with
other properties that are essential to the description of actual markets by a
continuous model (see Section 7).

2. DEFINITIONS, NOTATIONS, AND MAIN RESULTS

Let Q denote the nonnegative orthant of an Euclidean space Rm.A vector in Q


will also be called a commodity bundle (the dimension m of the space indicates
the number of different types of commodities). A partial preference order on Q
will be denoted by >-. We assume that >- is continuous, i.e., the sets {y: y >- x}
and {x: y >- x} are open in Q [3]. There is also an assumption usually assumed
on >- called "desirability of commodities": If x > y (i.e., xi > Yi, i = 1, ..., m)
and x :Ay, then x >- y. (There is also a slightly weaker form of desirability;
see [3]).
If we deal with a market with a finite number (n) of traders, we fix a set {jJi 1
of vectors in Q and call it an initial assignment, assuming that YJ%ji > 0, i.e.,
every commodity is actually present in the market. We let >-j denote the preference
order of the jth trader. A set {X}xj1 of vectors in Q will be called an allocation if
%=ix = ij. An allocation x is dominated by an allocation y via a subset S
of traders if yj >-j xj for j E S and ljesyj = EjTsij (the yj for j 0 S are unimportant).
The set of the allocations that are not dominated via any S is called the core.
An allocation is called an equilibrium allocation if there exists a vector p e Q
such that xj is maximal with respect to >j in the set {x E?Q: p *x < p *ij}, for every
j. The vector p is then called a price vector, and is not to be regarded as a commodity
bundle. The pair (p, x) is called a competitive equilibrium. It is well known [4]
that the set of the equilibrium allocations is a subset of the core, but they do not
coincide even in simple cases. If the preferences are not convex, the core may be
empty [9].
A continuous market, or a market with a continuum of traders, is given by an
initial assignment i(t), which is a measurable function from the unit interval T to Q,
with fT i(t) dt > 0. There is also a preference valued function >, with the following

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 793

property (Aumann-measurability):For all measurable functions x(t), y(t) from T to


Q, the set {t :x(t) >-t y(t)} is measurable. The expression x >-t y is to be read as "the
trader t prefers x to y" or "x is preferredto y by t." An allocation x(t) is an integrable
function, x: T -+ Q, which satisfies ST x dt = ST i dt. We say that x is dominated
by y via S if S is a measurable subset of 1T,(S) > 0, y(t) >-t x(t) for t e S and
fs y(t) dt = fs i(t) dt. (Instead of writing y(t) >-t x(t) for t E S, we may write y >-s x).
The set of all allocations undominated via any S is called the core. A competitive
equilibrium is an allocation x(t) such that there exists a p E?Q with the property
that for almost all t, x(t) is maximal (with respect to >-t) in the set {x: p. x < p i(t)}.
The introduction of the continuous model enables the proof of the following
theorem [3].

THEOREM: Under the above conditions the core coincides with the set of all equili-
briumallocations.

A continuous representation of afinite market is defined as follows. Let {Sj}71


be a partition of T into disjoint measurable sets, u(S) > 0 for every j. Define i(t)
by i(t) = ii for t e Sj, >-(t) = >-j for t e Sj, and represent each allocation {x;}
by x(t) = xj for teSj. Then STidt = X>1ig4S;), STxdt = 1% xj,(S;). If, in
addition, Y(S) = 1/n for every j, then n f j sJ x dt = l2Jcxj. The set of the rep-
resentations of allocations which belong to the core of the finite market will be
called thef-core, and is to be distinguished sharply from the core of the representa-
tion, when the representation is regarded as a continuous market.
The main result of Section 3 is that a separable metric topology can be defined
in a "natural" way for the space of preference orders which are complete,
transitive, reflexive, and continuous, and which satisfy the assumption of desira-
bility. This is the minimal topology that makes the set {(x, y, >-): x >- y} open
in the product topology. When we write >-n -+ >-, we mean convergence in this
topology.
One of our main results is the following theorem.

THEOREMA: If (in, >n) are continuous representationsof finite markets such that
(Si(n))n,oo-+ 0, if Xn is in the f-core of the nth marketfor every n, and if in -+ ,
n+ >-, Xn -+ x almost everywhere, and ST
iT
dt -+ST i dt, ST Xn dt -+ ST x dt, then
x is an equilibriumallocation.

This theorem is a generalization of the main theorem of Debreu and Scarf in [4],
which deals with markets with a fixed number (say k) of types of traders, having n
traders of each type, and strictly convex preferences. In that case traders of the
same type get the same commodity bundles in allocations which belong to the
core. Hence the cores of the markets with nk traders may be regarded as subsets
of the core of the market with k traders.
The main theorem of [4] states that the intersection (over n) of these subsets
coincides with the set of all equilibrium allocations. This follows from Theorem A
in the following way. Let us represent the market with k traders by the partition

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
794 YAKAR KANNAI

JSj}j=l, and represent the initial allocation by i1(t) and the preferences by >-1(t).
We represent the market with nk traders by subdividing each Si into n disjoint
sets having the same measure. Then in(t) _ i1(t), >-n(t) -1(t). Let {xj}x =
belong to the intersection, and let x1 be its representation. Then the functions
xn(t) x1(t) belong to thef-cores of the nth markets, and all the assumptions of
Theorem A are satisfied trivially. Hence the limit x(t)- xn(t) is an equilibrium
allocation, so that {xj}xj= is an equilibrium allocation too. Thus the main
theorem of [4] is a special case of Theorem A.
Another result is the next theorem.

THEOREM
B: If (in, > ,,)and (i, >-) are continuousmarkets,xn are allocations in the
cores of the markets(in, >-n), and if in -+ i, xn -+ x, >-,n -+ >- almost everywhere,
ST in dt -+ST i dt, ST Xn dt ST x dt, thenx is in the core of (i, >).
-+

This theorem may be called "upper semicontinuity." Lower semicontinuity


is incorrect (see a counter-example at the end of Section 4). If we allow s-cores,
however, then we have a kind of lower semicontinuity.
An allocation {xj}1% is in the strong s-core of the finite market {ij, >-sj}1% if
Yi >- jes xj implies, not (YjeS Yj)k < max {(y ieS J)k - , O} for 1 < k s< m (here (Y)k
denotes the kth coordinate of y). We shall write a 9 b for the vector whose kth
coordinate is max {ak- bk, O} for 1 < k < m. Also, we shall use the symbol E
to denote both the number E and the vector (e, . . . , e), since no confusion will
arise. With these notations, the condition for the s-core is that yj >-jes xj implies
not TsYj ' Ej2esije e. Since the order x < y is incomplete, we must be careful
not to confuse the negation of < with >. If yj >-jes xj implies not EjesYj < EjeS
ij 9 se where s is the number of elements in S, then {x} % 1 is said to be in the weak
s-core of the finite market {ij, >}>j11. These definitions do not use transferable
utilities and do not coincide exactly with those of Shapley and Shubik [9]. How-
ever, they have almost the same intuitive meaning.
There are no exact analogues of the weak and strong s-cores in the case of the
continuous market (i, >). Therefore we use a different terminology in the con-
tinuous case. An allocation x(t) is in the ordinary s-core of the continuous market
(i, >-) if y(t) >-s x(t) implies not fs y dt < fs i dt 9 e. If y(t) >-s x(t) implies not
f sY dt < f;Si dt 9 ,i(S)e, then x is said to be in the restricted s-core of (i, >).
THEOREMC: Let (i, >-), (in, >-,,) be continuousmarkets,in(t) + i(t), >n(t) -+>4
a.e., where in and i belong to a weakly sequentially compact set in L1. Let x be in the
core of (i, >). Then for every E > 0 and for every sequence Xn of functions with
Xn -+ x a.e. and f Txndt =ST in dt, Xn belongs to the ordinary s-core of (in, >n) for
all large n.

For restricted s-cores we have the following theorem.

THEOREMC': Let (i, >), (in, >n) be continuousmarkets,let i be essentially bounded,


let the range of >-(t) be conditionallycompact,and let in(t) + i(t), >n(t) _+ (,

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 795

the convergencebeing uniforma.e. in T. Let x be in the core of (i, >). Thenfor every
E > 0 and for every sequence Xn of functions such that Xn X+ X uniformly a.e. and
STxn dt = STin dt, Xn belongs to the restricted s-core of (in >n) for all large n.

REMARK: If x belongs to the ordinary e'-core of (i, >), then (under the conditions
of Theorem C) the xn belong to the ordinary E + e'-core of (in a ,). If x is essen-
tially bounded and belongs to the restricted e'-core of (i, >-), then (under the condi-
tions of Theorem C') the xn belong to the restricted E + e'-core of (in, >n).

Theorem C', together with the existence theorem for the core of a continuous
market [1], may be used to prove existence theorems for s-cores. For example,
if we deal with markets with nk traders (n traders of each type) and represent them
by the same representations which were used to show that Theorem A implies the
Debreu-Scarf theorem (subject only to the condition lt(Sj) = 1/k, 1 < j < k),
we may obtain allocations in the s-core for large n in the following way..Let x be an
allocation in the core of (il, >-1), regarded as a continuous market. Then x is
essentially bounded (see p. 815).
Let {{SJ)}Jnk l} l be a sequence of partitions that are refinements of the
partition {S1,..., Sk} and such that ,y(S(P))= Jlnk, and let xn(t) be a sequence of
allocations, such that xn(t) is constant on S(n) (j = 1, . .. , nk),and x -+ x uniformly
a.e. (such a sequence may be constructed without difficulty). By Theorem C' the
xn belong to the restricted s-core of (in, >n) for large n, and by their construction
they represent allocations of the markets with nk traders, which belong to the
weak s-core. In [9] existence of the weak s-core (in the sense of [9]) is proved using
the same way of letting the markets grow. Similarly, Theorem C' shows that if
we have a sequence of finite markets which may be represented in such a way that
the partitions form a dense sequence of partitions in T, the in converge uniformly
a.e. to an essentially bounded limit i, and >-n -+ >- uniformly a.e., then for every
positive e, the weak s-cores for all but a finite number of the markets are not empty.
This, clearly, is a more general proposition than the Shapley-Shubik theorem on
existence of weak s-cores. (On the other hand, Shapley and Shubik do not assume
desirability and continuity of preferences).
Theorem C could be applied similarly, but the resulting s-cores of the finite
markets are weaker than the weak s-cores. An existence theorem for strong
s-cores demands, presumably, a much more delicate argument as well as stronger
assumptions on the preference orders.

THEOREM D: Let (i, >), (in, >-n) be continuous markets,in, i belonging to a weakly
sequentially compact set in L1. Let the function Xn belong to the ordinary g'-core
ofthe nth market(n = 1, 2, . ..) and letxn +x, i+ni, >n >a.e.,T xdt = fTi dt.
Thenfor every E > 0, x belongs to the ordinaryE + e'-core of (i, >).

Theorems C and D may be considered as "continuity" theorems of the ordinary


s-core. Other results follow.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
796 YAKARKANNAI

THEOREM E: Let (i, >), (in, n) be continuous markets, let xn be in the ordinary
?
-core of (in, >-n), let in i, -- >- a.e., let STxn dt = STin dt, and let in(t), i(t),
Xn(t)< z(t) a.e. for an integrable z. Thenfor every c > 0, Xn belongs to the ordinary

THEOREMF: Let (i, >-) be a continuous market, x an allocation in its core. Then
there exist simplefunctions in, Xnn >- n with in -+ i, Xn -+ x, >-n + >- a.e., fT indt =
ST xn dt = ST i dt, an integrablez with in(t), i(t), Xn(t),x(t) < z(t) a.e., such thatfor every
c > 0 xn is in the ordinary ?-core of (in, >-n) for all sufficiently large n.
The corresponding theorems for the restricted c-cores follow.

THEOREM D': Let (i, >-), (in, >n) be continuousmarkets,let i be essentially bounded,
let the range of >-(t) be conditionally compact, and let in -i j, >-n,-- >- uniformly
a.e. Let thefunction Xn belong to the restricted c'-core of the nth market(n = 1, 2,. . .)
and let Xn ? x uniformly a.e. (so that T x dt = STi dt). Then for every c > 0, x
belongs to the restricted c + c'-core of (i, >-).

THEOREM E': Let (i, >-) be a continuousmarket, let i be essentially bounded,and


let the range of >-(t) be a conditionally compact set in E.. For every c > 0, there exists
a neighborhood c1 of (i, >-) in Loo x such that if (i1, >-1) e 1 is a continuous
H

market with ST i dt = ST il dt and if x is an essentially bounded allocation in the


restricted c'-core of (ij, > 1) then x is in the restricted c + c'-core of (i, >-).

THEOREMF': Let (i, >-) be a continuous market, let i be essentially boundedand


let the range of >-(t) be conditionally compact. Let x be an allocation in the core of
(i, >-). Then there exist simple functions in, Xn >-n with in + i, Xn X
x, >n >
uniformly a.e., STin dt = ST Xn dt = STi dt, such that for every c > 0, Xn is in the
restricted c-core of (in, >n)for all sufficientlylarge n.

The importance of Theorem F' lies in the fact that it enables us to regard a
continuous market with an allocation in its core as a "limit," in a suitable sense,
of finite markets and allocations in their weak c-cores (if the initial distribution
is essentially bounded). Similarly, from Theorem F one concludes that any
(integrable) continuous market with an allocation in its core may be regarded
as a limit of finite markets and allocations in their "weaker" c-cores.

3. CONSTRUCTIONOF A TOPOLOGYON SPACES OF PREFERENCES

A relation > on Q will be called a preference order if it is complete, transitive,


and reflexive. If x > y we shall say that x is preferred or indifferent to y. If x > y
and y > x, we shall say that x is indifferent to y and write x y. If x > y but not
x y, we shall say that x is preferredto y and write x >- y. We shall always assume
that there exist vectors x, y such that x >- y, i.e., we exclude the "trivial" preference
order.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITYPROPERTIES 797

A preference order is called continuous if the set {(x, y): x >- y} is open in the
product space Q x Q; in other words, if Xn >+ x, Yn-+ y, and x > y imply Xn >_ Yn
for all sufficiently large n. It is well known [1, p. 3] that a continuous preference
order may be represented by a continuous utility, i.e., there exists a continuous
function u: Q -+ R such that x >- y > u(x) > u(y).
Denote by " a set of preference orders. We want to introduce a "natural"
F
topology on E. A plausible requirement on a topology on is that if x >- y and
if xn -+ x, Yn-+ y, and >-n -+ >- (in the topology of E), then Xn >-n Yn for all
sufficiently large n. In terms of open sets it says that the set A = {(x, y, >-): x >- y}
is open in the product space Q x Q x E. Therefore, if there is a minimal topology
on which makes the sets A open, then this may be considered a "natural"
topology on E. (This notion of natural topology is closely related to the compact-
open topology of function spaces [8]).
We are able to show that there is a natural topology on ~, to characterize it,
and to show that it has a countable basis. Assuming more on >-, we shall be able
to introduce a metric on E which will induce that topology.
Arrange the rational balls (i.e., the balls with rational center and rational radius)
in Qin a sequence {Si}. Set ,ij = {>- E : x >- y for all x c Si and y E Sj} (a superior
bar denotes closure). We shall denote the fact that x >- y for all x E A and for all
y E B by A >- B. (For example, Ei1 = { >-e : Si >- Sj}.) Of course, E 1 can be
non-empty only if Si n S= 0.

THEOREM3.1: Let E be a set of continuous preference orders. Then the minimal


topology on Efor which the set A = {(x, y, ): x >- y} is open in Q x Q x E exists
and is equal to that topology which has the class { di,j}as a sub-basis.

PROOF.The set A is open in that topology, for if (xo, yo, >- 0) E A, then (by the
continuity of >-0) there exist open sets U, V, with yo E V and xo E U, such that
U >-0 V We can find rational balls Si, Sj with xO E Si, yOE Sj, Si c U, Si c V
Then >-0e i,j and (xo,yo >-0)eSi x Si x Ei,j which is an open set in the
product topology.
F
Let Y be a topology on for which A is open. Let Si, Si be rational balls with
disjoint closures. We shall-show that E i,jis open in E. Let >- Edi,j (if E i,jis empty,
then it is trivially open). Then Si x Si x J>01c A. According to a theorem of
Wallace [8, Theorem 5.12], it follows from the compactness of Si, Sj, and {>-0} that
there exist open sets Ui, Uj E Q, Ve A Sk c Uk (k = i, j), and >-0 E V, such that
Ui x Ui x V c A. Hence Ui >- Uj for each >- E V and therefore >-o E V c ij. "

Hence i,j1is open in F.

REMARKS:(1) The choice of Q as the space on which the preferences are defined
is not essential for this theorem (any locally compact second countable space will
do).
(2) The class of sets {Ei,j} is countable. Hence the class of all finite intersections
of members of {J"Ei } is countable. Therefore the topology defined in Theorem 3.1
has a countable basis.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
798 YAKAR KANNAI

Let us assume now that the preferences satisfy the "desirability" condition.
Let 3 denote from now on the set of all preferences which are derived from com-
plete, reflexive, transitive, and continuous orders and which satisfy the desirability
condition. Every >- E 3 may be represented by a continuous utility u in the follow-
ing way. For each x e Q there exists a vector y on the principal diagonal of Q such
that x y (y exists because of continuity, completeness, and desirability, and is
unique because of desirability). Let u(x) = IYI(IYI is the Euclidean norm of y).
It is easy to see that u is a continuous utility and satisfies the condition u(x) = lxl
for all x on the principal diagonal of Q, and there exists a constant C depending
only on m (the dimension of Q) with 0 < u(x) < Clxl for all x E?Q. Denote by C*
the set of the utilities obtained in this way.

THEOREM3.2: The minimaltopologyon 3 whichmakesA = {(x, y, >-): x >- y}


open in the product space Q x Q x 3 is inducedby the metric

(3.1) P(> 1, >2) = max >l(X)-u>44


1 +
xe?2 ix12

(U>-i is the utility in C* representing >-i).

PROOF:The construction of u insures the existence and finiteness of the max


in (3.1). Let >-0 E3 i,j. Then Si >-0 Si, so that for every x c i, yE
Y Si, uo(x) > uo(y)
(where u0 is the utility in C* representing >-0). The continuity of u0 and the com-
pactness of Si, Si imply the existence of a positive 3 with u0(x) - u0(y) > 3. Set
R = maxxesi.uj(l + iX12). Then

JueC*. mxIu(x)1+ - 1X12


u cCrnax uo(x)i <2 __

c {ueC*: max lu(x)-uo(x)l <


xc-sjusj 2

Hence for every x E Si, y E Si, u(x) > uo(x) - 6/2 ) uo(y) + 6/2 > u(y), so that
-0{>:p(>-, >0) < 3/2R} c Eij, and therefore ij is open in the metric
topology.
Let >-0 E, ? > 0. It will suffice to prove that the ball {>:p(>-, >-0) < E} is
open in the minimal topology. Obviously,

u(x) - uo(x) lu(x)l + luo(x)l 2Cixl


1
1 + IX12 l + IX12 1+ jx1-2-'

Hence there exists an R such that lxI> R implies

U(X) -
Uo(X)<
+ Ix12

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 799

for every u E C*. Therefore

(3.2) { p(>-: >-?) < el = max Iu>_(X)


+ |X) < ?}-
We shall show that for every compact subset K c Q and for all reals a, ,
the set {>-:VxeK, a < u>(x) <1} is open in the minimal topology. The sets
Ax = {(y, >-): x >- y} and A, = {(x, >-): x >- y} are open as cuts of open sets, and
therefore Az), = Az n Ay = {(x, >-): z >- x >- y} is also open in Q x 3. Let us
assume that >- is such that for every x E K, z >-1 x >-1 y. Then K x I>-c Az,y
and therefore there exists an open U containing >- 1 (by the theorem of Wallace)
such that K x U c Azy. Hence { >-: xEx K, z >- x >- y} is open in the minimal
topology. Let y, z be on the principal diagonal such that IYI= a, IzI= /3(the cases
a < 0 or ,B = cause no trouble). Then u(y) = a, u(z) = ,Bfor every u E C* and
therefore {> :Vx E K, a < u>(x) <13} = {> :Vx E K, z >- x >- y} and hence our
assertion is proved.
The set {x e Q: lxl < R is compact. By uniform continuity there exist open
neighborhoods of every point in that set, in which the oscillation of uo is less than
e/2. Hencethereexist rationalballs Ski with maxxesk. uo(x) - minxesk uo(x) < e/2.
We may choose a finite number (N) of them which covers all {x: lxl < R}. The set
Hi={>.-'vfxek
>_:VX C_ Sk, minyesk.uo(Y)
minyc-Sk iyeSk,
- e/2 <
u>(x) < max (y) +
u0Oy +e2e/2} is open
-

in the minimal topology, as we have just seen. We have that x E Ski and >. E R'i
-e < u>(x) - uo(x) < e < lu>,(x)- uo(x)l < e. Hence if >- E nQ= Hi', then
xl R implies lu>-(x)- uo(x)l <e and therefore also max1x1<R(IU>(x) - uO(x)/
A
(1 + IxI 2)) < e. Hence by (3.2), >-o Eni {>- :p(>-, >- 0) < -} and therefore
the ball is open in the minimal topology.

REMARKS: (1) The second part of the proof is analogous to the proof of some
properties of the compact-open topologies.
(2) The metric in (3.1) is in no way "natural." We could have defined C* in
other ways and could have chosen other expressions of the metric which would
have induced the same topology on 3. However we shall use the special form (3.1)
in various subsequent calculations, not merely the fact that we have a "natural"
separable metric topology.
(3) Now that we have a topology on the space of preferenceorders, we may define
' be measurable if the inverse
a function >- -(t): T -+ to image of every open
set in 3 is measurable in T. In fact, this condition is.equivalent to "Aumann-
measurability" ([3]; see also Section 2). It is obvious that the present condition
implies A-measurability. For, if x, y are any measurable Q valued functions, then
{t: x(t) >.- y(t)} {t: 3i,jx(t)e Si A y(t) e Si A >-t e Hij}

U {t:x(t)eSi A y(t)e3j A Xt i,jl


i,j
-
U [x '(Si) n y '(S3)n > -
i,j
and here every term is measurable.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
800 YAKAR KANNAI

On the other hand, A-measurability implies measurability of the inverse image


of open sets in ~E.It suffices to show that the inverse image of every open ball is
measurable. For every fixed a, f,, x the set
(3.3) { t: '< u,(x) '< #

is measurable if ut is the utility in C* corresponding to >-t. This is seen if one


chooses y such that ut(y) = # for every t and sets y(t) y, etc. Let us be given the
ball {>-: p(>-, >-0) -< e. Then (as we have seen above) there exists a compact
set K such that our ball corresponds to

{u: bVxE K, lu(x)- u0(x) < }

Hence we have to prove that

{ t: max lut(x)
-
1 u0(x)
xeK +1X12 4
is measurable. Let {xj} be a dense sequence in K. Then

{t:maxIut(x) - u0(x)I <


<
=+x:ft0X) a
tx=
yn x I-K
{ xeK 1+ IX12 1+ IX2 n}
-
=un {t
1u'(xj) uo(xj)I
1 + jXj12 il<?

=un n i
tt: Iut(xi)- uo(xj)I

e-n (1 + IXjl2)}

= u n {t: uo(xj) - (8- 1)(1 + IXjl2)

< Ut(xJ) < uO(xJ) + -!)(1 + Ix1l2)}

and every set in the last expression is measurable, since it is of the type (3.3).

4. PROOFS OF THEOREMSA AND B

The main ideas of the proof of Theorem A (in particular the construction of
J(u) and p) are similar to those of Debreu-Scarf [4] and Aumann [3]. However,
the details are much more complicated, since in spite of the fact that z + i(t) >.t x(t)
implies z + in(t) >-t,n Xn(t) for n > N(t) for almost every t, N(t) depends on t. We
have to find subsets of T whose measure is bounded away from zero, on which N
can be found independent of t. We do this using compactness and uniform con-
vergence on those subsets. First we prove a simple lemma.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 801

LEMMA: Let fn(t),f (t) be real measurablefunctions on T, let f (t) be finite a.e. and
let fn(t) -+ f(t) a.e. Thenfor every ( > 0, there exists an M; > 0 and a set A,6c T
with p(A,6)< ( such that Ifj(t)I,If (t)l < Mafor all t 0 Aj andfor all sufficientlylarge n.

PROOF: There exists an M such that p({t:1I f(t)l > M}) <(/2. According to
Egoroff's theorem there exists a set B,6with p(B6,) < 5/2 andfn(t) -+ f(t) uniformly
in T- B,6.Define A6 = B,u {t: If(t)l > M} and M = M + 1, and the lemma is
proved. (The lemma may also be proved without using Egoroff's theorem.)

A:
PROOF OF THEOREM For every z E EM,let us define R(z) = {t: z + i(t) >-t x(t)},
Rn(z) = {t :z + in(t)>-n,t Xn(t)}.Set Z = {z: z is rational and y(R(z)) = 0}. Then
uzeZR(z) is a null set. Define U = T - u zR(z), //(t) = {z: z + i(t) >t x(t)} and
define zl(U) to be the convex hull of u tcuf(t).

The origin is not an interior point of a(U).


PROPOSITION:

Assume, on the contrary, that 0 is an interior point of X(U). Then a whole neigh-
borhood of 0 is contained in J (U). The sets 0(t) are relatively open. Hence there
exist m + 1 rational vectors zi and non-negative rational numbers jti such that
m m
2f gizi = r, r > ?, 2i pti= 1, zi c i(ti), ti c U.
i=O i=O
Then r is also rational and because of desirability, zi E f(ti) implies (zi + r) E i/(ti).
But IT o0i(zi + r) = 0. Therefore we may assume that SITluizi = 0. According to
the definition of U, zi E if(ti) implies 1(R(z1))> 0. There is only a finite number of
i's. Hence there exists a positive K with pu(R(zi)), c for every i. Also, zi + i(t) >-t x(t)
for all t E R(zi). According to the lemma, there exists a set AK/8, a constant No,
and a compact set K in Emsuch that lt(AK/8) < K/8 and x(t), xn(t),zi + in(t)belong
to K for t 0 AK/8 and n > No. According to Egoroff's theorem there exists a set
B with ,u(B) < K/8 such that xj(t) -+ x(t), in(t) + i(t), >-n(t) >-(t) uniformly for
t 0 B. But
-
p(>(t) >-(t)) = max Iun,(X) utx)

Hence un,(x) -+ ut(x) uniformly for x E K and t 0 B, since 1 + 1x12, C for x E K.


Define 9#(e) suPxeK,IyIle lut(x + y) - ut(x)l.Then pt(e) > 0 for e > 0, pt(e) -+ 0
for e -+ 0, and for every fixed e, pt(e) is a measurable function of t. For if {yj} is a
dense sequence in {y; IYI< 4j, and {xj} is dense in K, then

(pt() = sup lut(xi + yJ)- ut(xj)I


,J

and each of the functions lut(xi + yi) - ut(xj)lis measurable.


The >-n are simple functions from T to 3, and therefore their limit >- is also
measurable and, by Remark 3 after Theorem 3.2, >- is Aumann-measurable.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
802 YAKAR KANNAI

Hence it is easily proved that the functions u,(x(t))and u,(zi + i(t)) are measurable.
By the definition of R(zi), u,(zi + i(t)) - u,(x(t)) > 0 for t e R(zi). Therefore there
exists a measurable set Bi c R(z,) and a ( > 0 such that i4(Bi) < K/8 and
(4.1) u,(zi + i(t)) - u,(x(t)) > ( for teR(zi) - Bi.
Set T1= {t: (p(8) < (/4}. Then ? < i' implies TED T and u,T4 = T. Hence
there exists an go with 4u(T- T) < K/8 and

(4.2) sup ju,(x + y) - u,(x)l < -4 for t e T,,.


xeK, Iy| <o

Let R'(zi) = R(zi) n Teo- A/8 - B - Bi. Then 1i(R'(zi))> K/2. xj(t) -+x(t),
ij(t) -+ i(t) uniformly in R'(zi). Hence there exists an N1 such that
(4.3) Ixn(t)- x(t), lin(t)- i(t) < go for n > N1, t e R'(zi).
By (4.1), (4.2), (4.3), and the definition of B, we get for n > max (NO,N1) that

(4.4) ut(zi + i(t)) - ut(xn(t)) > -2'" for t e R'(zi).

By the uniform convergence of ut,n(x)for x e K and t e R'(zi), there exists an N2


such that

(4.5) lut,n(x)- ut(x)I < -for xeK, teR'(zi), n > N2.

Hence, by (4.3), (4.4), and (4.5), for n > max (NO, N1, N2) and t e R'(zi),
(4.6) Ut,n(Zi + UO(t) - Ut,n(Xn(0)) > ?

Therefore, zi + i(t) >-t,n Xn(t) for t e R'(zi)and for all sufficiently large n, so that
Rn(Zi)D R'(zi). Since there are only a finite number of i's, we have that for large j,
R1(z) n R'(zi) for each i, and especially Iu(Rj(zi)) > K/2.
The ,ui are rational non-negative numbers, and we may write i = si/s where si,
s are non-negative integers. Since (Ij, >-j) is a representation of a finite market,
every set Rj(zi)is a union of sets of the type S(k), each set S(') representing a "trader."
Since y4(S(j))j -+ 0 but u(Rj(zj))> K/2 > 0, we may choose (for j sufficiently
large) s different "traders" for each "coalition" Rj(zi).Let us choose si traders from
each such coalition, and give them the commodity bundle zi + iJ{tk,X)(tk,jeSkj)
RO(zi),1 < k < si). Then
m Si m m Si
Y, E (Zi + ij(tk,i)) = E SiZi + E E ij(tk,i)
i=O k=1 i=0 i=0 k= 1

m m Si m St
= S E JiZi + E E iJ(tk,i) = E Z ij(tk,i).
i=O i=0 k=1 i=0 k=1

Hence we may complete this to an allocation, and zi + ij(ti) >-n xj(ti), contrary
to the assumption that xi is in thef-core of the jth market, and the proposition is
proved.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 803

The proof of Theorem A may now be completed in much the same way as in
[3]: Using f i = f x, one shows that there exists a non-zero vector p such that
x >- x(t) p x > p i(t) (p. x = 0 is the equation of the supporting plane to aI(U)
.

through the origin), and then shows that actually p x > p. i(t) if x >- x(t).

PROOF OF THEOREM B: According to the fundamental theorem of [3j, there exist


price vectors pn,which belong to the allocations x,, in the sense that (p,,,x) is a
competitive equilibrium. W.l.o.g., I ai = 1 (Pn= (pn,..., pm)).Then the pnare
contained in a compact set, so that we may choose a convergent subsequence out
of {P}, and w.l.o.g.P- p (and then also ITm_ip = 1) and p > 0. We shall show
that (p, x) is a competitive equilibrium for (i, >-). For, if y >-, x(t), then y >-t,n xn(t)
for sufficiently large n (since >-n,t + >-S, and xn(t) -* x(t)), so that Pn *Y > Pn i-(t)
p i(t), and therefore Pn Y y p *i(t). (This is true for almost all t.)
The rest of the proof the same as in [3], as at the end of the proof of Theorem A.
is

> ~~~~~
n

FIGURE 1

The converse of Theorem B (i.e. lower semi-continuity) is false. Let i(t) _i(t)
(1, 1), let >-t be the preference order whose indifference curves are the lines
xi + x2 = constant, and let >-n(t) be strictly convex preferences converging to >-t.
Then the core of the markets (xn >n) consists of the single allocation xn(t) (1, 1),
whereas every allocation which satisfies xl(t) + x2(t) _ 2 is in the core of the limit-
ing market. (This is another reason why Theorems C and C' are important.)

5. SOME LEMMAS ON 8-CORES AND RELATED TOPICS

The definitions of 8-cores in the case of a finite market and the continuity of
preferences imply almost immediately that the core is the intersection of all the

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
804 YAKAR KANNAI

.-cores (8 > 0), since if y >-s x and Yj,syj = Tjhsij, it is possible to find y' near y
which satisfy y >- jes xj and Xjcsy9< lj2sij E e for e sufficiently small.
It is not obvious at all that if (in the continuous case) y(t) >-s x(t), we may find
measurable y'(t)with the properties y'(t)>-s x(t) and fSi dt e e > y' dt. We need, S
in fact, the "principle of borel choice" [2, proposition 2.1] which asserts that if F
is a borel subset (actually less is required) of T x Enwith the property that for every
t E T, {x E En: (t, x) E F} 7 0, then there exists a Lebesgue measurable function
f (t) such that (t, f (t)) E F for every t. (After redefinition on a null set we may choose
f (t) borel measurable.) We shall make constant use of this fact in the discussion
which follows.

THEOREM 5.1: The core is the intersection of all the ordinary 8-cores, and is the
intersectionof all the restricted 8-cores.

PROOF: It is obvious that if x is in the core then x(t) is also in the 8-core, for every
e > 0. Assume that x is in the ordinary 8-core for every 8 > 0, but x is not in the
core. Then there exists a coalition S, u(S)> 0, and there exists a function y: S -+ Q
such that fs y dt = fs i dt and y >- x. For every t E S there exists (by continuity of
preferences)an 8 > 0 such that y(t) e 8 >-, x(t).
We shall show that the set {(t, 8): y(t) e 8 >-, x(t)} is borel measurable in the
product space T x Q. Let {rn} be the sequence of all rational numbers. Define
Tn= {t: y(t) e) rn >- x(t)}. By Aumann measurability Tnis measurable, so that
Tnx (0, rn) and therefore UnTn x (0, rn) are also measurable. If y(t) e9 >-, x(t),
then there exists (by continuity) an rn > 8 with y(t) E rn >- x(t), so that (t, 8) e Tnx
(0, rn). Hence {(t, 8): y(t) E) 8 >-, x(t)} = UnTn x (0, rn)and therefore it is measur-
able.
Using the principle of borel choice we may choose (and redefine suitably)
measurable 8(t) > 0 with y(t) 8(t) >te sx(t). fs 8(t) dt = 8' and 8' is positive, and
eJ

fs [y(t) E9?(t)]dt = fs i dt 9 8'. Hence x does not belong to the ordinary (?'/2)
core, a contradiction. The assertion of the theorem about restricted 8-cores follows
from the fact that the restricted 8-core is contained in the ordinary 8-core.

We shall now prove three lemmas which are needed for the proofs of Theorems
C, D, E, F, and C', D', E', F'. The principal difficulty is that y(t) >- tX(t) [y(t) >- ,n Xn(t)]
implies y(t) >-n,Xn(t) [y(t) >- X(t)] for n > N, but N depends on t, and even the
addition of a constant to the left sides does not improve the matter, unless, as
Lemmas 5.1, 5.2, 5.3 show, the y(t) (x(t)) are restricted to compact sets in Q and the
convergence is uniform. Therefore we have to choose subsets of T, where the desired
properties hold.

LEMMA5.1: Let K be afixed compact set in Q. For every x EcQ, 8 > 0, and >-, there
exist 81 > 0 and 82 > 0 such that

(ii) IX'
-
XI < ?61, P(>_, )< ?2, K -3Y >- X >Y + ?>- X,

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 805

PROOF: (i) Define B = {y: y + e x}. Then B is a closed set and therefore
B r- K is compact, and B r- K -< x. Hence {x} x B nrK x {>} c A which is an
open set in the product space (see Theorem 3.1). By a theorem of Wallace [8,
Theorem 5.12] there exist neighborhoods U c 2, Vc 3 with x E U, >- E V4
such that U x B r'K x Vc A. Hence if x'eU, >-'eV, then BrK-<'x' and
therefore K 3 y' >- x' implies y s B, i.e., y + e >- x.
(ii) Define C = {z: z > x,z - e/2e K}. Then C + e/2 is compact and (C +
e/2) x {x} x { >-} c A. Hence there exist neighborhoods U', V', such that
x E U', >- E V', and (C + e/2) x U' x V' c A. Therefore x' E U' and >-'E V'
imply z >-' x' for z E C + e/2. But if K 3 y >- x, then (by desirability) y + e/2 > x
and therefore (y + e) E C + e/2.
Since Q and 3 are metric spaces we may choose el > 0 and 82 > 0 with the
stated properties (i) and (ii).

LEMMA5.2: Let x(t), >-(t) be borel measurable. We may choose e1(t), e2(t) borel
measurable such that (i) and (ii) are satisfied almost everywhere (for x = x(t) and
= >-(0).

PROOF: We shall use the principle of borel choice. Denote by P,(y, x', >') the
proposition
[lx' -x(t)l <el] A [P(>-(t), )< 2] A [K DyS -x']
= [y + e >-(t)x(t)]
and by Pv(y,x', >-') the proposition
[Ilx'-x(t)l < c1] A [P(> (t), )< 82] A [K 3y >- x']
= [y + e/2 >-(t) x(t)].
We have to show that we may choose e1(t),e2(t)from the set
P = {(t,e1,e2):e1 > O, 2 > 0,(VyeK)(Vx'ec2)(V>-'E ,)P(y,x', >')}.
Let {yiJ, {>-'}, {4X} be dense sequences in K, 3, and Q respectively. Then if
(Vyi)(Vx)(V>-'k)Pt(yi, xj, >-') is true, then by desirability
(Vy e K) (Vx' )() >-' eE)Pt(y, x', >-')
is true. Hence the set P contains the set
Q = {(t,1, 82) 1 > 0, 82 > 0, (yi)(VX)(VXj )Pt(Yi, Xj -'})
It suffices to show that Q is measurable. It is well known that p D q -_ pVq
and {z: Vb,Aa(z)}= nQ {z: A,(z)}. Hence

Q= n n {(t, e1, 82): [IX- x(t)I1 81 > 01 V [p(>(t), >'k)


j k V

82 > ?] V [Yi ---,k XJ]1 V(YE + 2 F(t) X(t)l

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
806 YAKARKANNAI

The measurability assumptions of the lemma imply that every set on the right is
borel measurable in the product space. The proof for (ii) is similar.

LEMMA 5.3: Let (i, >), (in, >_n) be continuous markets, in i+ ', >n X+ >, Xn X -+

a.e.; in, i are contained in a weakly sequentially compact set in L1. Then for every
e > 0, 9 > 0, c > 0, there exist a set A c X a positive (, and a compact set K c Q
such that
(1) for every n, fA idt < e, fA indt < e;
(2) xn(t) x+ X(t), in(t) + i(t), >-n(t) -+>-(t) uniformlyfor t s A;
(3) i(t), in(t) E Kfor sufficiently large n and t s A; and
(4) e1(t),e2(t)(which are defined according to Lemma 5.2 for x(t), >- (t), cK, and e)
satisfy e1(t),e2(t) . ( > Ofor t 0 A.,

PROOF: It is well known [5, p. 294] that limg(E)o SfEindt = 0 uniformly in n,


since the inare contained in a weakly sequentially compact set in L1. Hence, using
Egoroff's theorem, we may find a set A1 such that fA, i dt, fA1indt < e/3, and
in(t) -+ i(t), > n(t) -+ >- (t), Xn(t) -+ x(t) uniformly in T- A1. Since i(t) is finite a.e.,
>
4u({t: IIi(t)IIK k})k> -+0. Hence there exist M and A2 such that IIi(t)IIKo< M
for t 0 A2 and |A2i dt, fA2 in dt < s/3. For n sufficiently large and t s A1 u A2,
IIn(t)II, < M + 1, since in(t) + i(t) uniformly in T - A1. Set K = {x eQ?:
lixiK.0 < M + 1}. Choose el(t), e2(t) that satisfy (i), (ii) for Z, >-(t), x(t), and cK.
Then e1(t), e2(t) > 0 a.e. and are measurable. Hence we may find A3 and ( > 0
with SA3i dt,IA3 in dt < e/3 and e1(t), 82(t) > (5> 0 for t OA3 . Defining A = A1 u
A2u A3, we have all the conclusions of the lemma.

Lemmas 5.1 through 5.3 are not sufficient to deal with blocking by coalitions
S if fs i dt is not well in the interior of Q, since in that case we have to consider
fs i dt E e and not fs i dt - e. We shall sketch the necessary modifications in the
lemmas, leaving the details to the reader.
Let ek be the kth unit vector in Rm.In modifying Lemma 5.1 we are interested in
the conclusions (i)' y + eek > x for all 1 < k < m, and (ii)' y + eek >-' x' for all
1 < k < m. We define the set B' by B' = {y'; y + eek >_ x for all 1 < k < m} and
obtain the appropriate numbers el and 82in the same way as in Lemma 5.1. It is
easy to choose e1(t) and e2(t) (as in Lemma 5.2), in a measurable way, and then to
have e1(t) and e2(t) bounded away from zero outside a small exceptional set.

6. PROOFSOF THEOREMS
C, D, E, F, AND C', D', E', F'

PROOF OF THEOREMC: Assume that e* > 0 satisfies ,(E) < e* => in dt, SE
fE i dt < e/4, and set C = m/e* (m = dim Q). Let A, K, e1(t),(, M be as in the state-
ment and proof of Lemma 5.3, the constants e and e there being equal to e/4. There
exists a number N such that if n > N, then Ilin(t) - i(t)li, P(>n(t), >(t)),
lIxn(t) - x(t)oo < (, i(t), in(t) E K for t 0 A and IfF in dt - F i dtII < e/4 for every
measurable F c T. We shall show that xn is in the e-core for n > N. Assume that
S is a coalition with 4u(S)> 0; y: S -+ Q is integrable and Y SnS Xn for some

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 807

n > N. W.l.o.g. we may assume that there exists a set B c S with 4u(B)< e* and
te S - B > IIy(t)II. C(M + 1).Otherwise there exists a set B c S with 1(B) > 8*
and IIy(t)II> C(M + 1) for t e B. Then at least one coordinate of y is greater than
00

C(M + 1) in a set B' c B with 1(B') > e*/m. Hence

(6.1) ydt -> ydt C(M + 1)-= M + 1, so that

J'y dt$ M + 1.

On the other hand,

i dt <
~~ TS
r~~A
Ai dt 0 + I
S-A
i dt < (M + 1)4(S - A) +
40

<M + 1 +-.
4'
Hence

(6.2) {indt-8 < M + 1.

Therefore it is impossible that fS Ydt < fS in dt - 8, and we need not worry about y.
Define B u A = D. Since n > N we have by Lemma 5.3 and the definition of N
that y(t) >-,n Xn(t) for teS - D implies y(t) + e/4 >-,x(t). Hence y(t) + 8/4 >-s-Dx(t).
Since x is in the core of the market (i, >), we conclude that
'
(6.3) { + 4 dt {
(y
S-D -D dt.
Obviously,

(6.4) J'ydt + J-D y d+ S-D y dt.

Moreover, SDi dt =
fA i dt + SB i dt < (8/4) + (8/4) = (2/4)e (since 4(B) < 8*), and
therefore,

(6.5) { idt + 42J idt.


S-D 4

From (6.3), (6.4) and (6.5) it follows that fS y dt + (3/4)e $f i dt, i.e.,

(6.6) JY dt i
J'idt - 38.

O < e/4, so thatfS Ydt 4


in dt - fs i dt 11
We have IfSJ fS in dt - 8. Using the remark
made at the end of Section 5 we can also show in a similar way that sSY dt A
fS in dt e 8. Thereforexn is in the ordinary8-core.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
808 YAKAR KANNAI

PROOF OF THEoREmC': W.l.o.g. assume that there exists a constant C such that
IIx(t)I, IIij(t)
IIxj(t)11K, II, < C. Let an 8 > 0 be given. Set K = {ye2Q,
II, IIi(t)
< (2C2m/g) + C}. Then K is compact. Set inffeT,zeKIuU(z
IIYIIOO + e/4) - u,(z)l = 6.
There exists a sequence {ti4, {zi4 with ut,(zi + g/4) - u,,(zi)-+ 6. Since K is compact
and since { >-: 3(t e T) >- (t) = > } is a conditionally compact set, we may assume
that z e K, >- e exist with >.- , zi -+ z. Hence u5(z + e/4) - u-(f) = 6,
-

so that by desirability 6 > 0. Using a similar compactness argument we find


that there exists a p > 0 such that x, y e K, Ix - yl < p - lut(x) - ut(y)l < 6/3
for t e T Now let N be such that p( >-(t), >- (t)) < (6/3(1 + C2 + (4C4m2/82))) and
IIin(t) - i(t)II,o< 9/4, IIx,(t)- x(t)II, < p for n > N, and assume that there exist
S c T and y(t) such that Y >- nsXn
II, < (2C2m/j) for t e B and p(S - B) <
W.l.o.g. there exists a set B c S with IIy(t)
(8/2C)u(S). Otherwise, there exists a set D c S with p(D) > (8/2C)y(S) and
t e D => jy(t)jjO > (2C2m/e). Hence at least one coordinate of y is greater than
(2C2m/.) in a set D' c D with j(D') > (em(S)/2Cm).Then

ydt
y0 |.t y dt > /u(D') C p
C(S) )t indt |

and we need not worry about y.


To sum up, we know that for t in B,

(6.7) UtY(t)+ 4) - Ut(Y(t))


> ;
.6~~

(6.8) 1Ut(Y(t)- 4~~~~ < _;


Ut,n(Y(0))l

6
(6.9) IUt,.(X.(t)) - Ut(Xn(t))l< 3;

(6.10) |Ut(Xn(t))-Ut(X(0)) < 3.

But

uty((t) + U- u(x(t)) = ut Y(t) + - Ut(y(t)) + ut(y(t))

+ - + -
-Ut,n(Y(t)) Utm,(y(t)) Ut,n(Xn(t)) Ut,n(xn(t)) Ut(Xn(t))

+ ut(xn(t)) - Ut(X(t))

It follows from (6.7H6.10) and y >-nB xn that ut(y(t) + e/4) - ut(x(t)) > 0, or that
.
(6.1 1 ) x t. +

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 809

Since x is in the core of (i, >-), we conclude that


.6
(6.12) |y dt $|i dt - (B).

But IS-B i dtIloo < C4(S - B) < C (8/2C)1.(S)= (?/2)8(S) and IlIfs
indt-
fs i dt I < (e/4)/(S), so that

(6.13) j ydt < { indt - r,(S).

Using a similar argument we can show that for n sufficiently large fs y dt $


f indt e 88(S).
If we assume only that x is in the ordinary (restricted) v'-core of the market
(i, >-), then, adding - e'[ - 8',u(B)] to the right hand sides of (6.3), (6.5), and (6.6)
[(6.12) and (6.13)], we see that xMis in the ordinary (restricted)s + s'-core.

PROOF OF THEOREM D: Define 8*, C, A, K, M, gi(t), and 6 as in the proof of Theo-


rem C. Assume that there exists a measurable S c T with pu(S)> 0 and an integ-
rable y: S -Q ? with y >-s x. An analogous argument to that given in the proof of
Theorem C shows that we may assume w.l.o.g. that there exists a set B c S with
p(B) < 8* and y(t) E CK for t 0 B. Let N be such that n > N implies 11 in(t) -i(t)II .,
11Xn(t)- (t11 00, xP(>(t, >40t) < ) in(t),i(t) E-K for t 0 A, lIfF indt-IfF i dtll%< E/4
for every measurable F c T. Set A u B = D. Then (according to the construction)

(6.14) y(t) + 4SXns-DfXn

Since xn is in the ordinary g'-core, (6.14) implies that

(6.15) X y +_ 4dtt X in dt -8'.


S-D 4 S-D

Therefore we may conclude as in the proof of Theorem C that

(6.16) ydt A c{idt-(8 + 8').

The reader may supply the details which show that

J ydt J idt E( + ').

REMARK: If the xn belong to the ordinary en-coreand n-+ 0 then by Theorem D,


x belongs to the ordinary 28'-core, and by Theorem 5.1 x belongs to the core.
In particular, if the xn belong to the core, then x belongs to the core of the limiting
market (i, >-). This gives us a second proof of Theorem B, which does not depend
on the existence of a price vector. However, this proof is valid with some restric-
tions on in and i (weak sequential compactness) which are unnecessary, in view
of our earlier proof of Theorem B.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
810 YAKAR KANNAI

PROOF OF THEOREM E: Since z(t) is integrable, there exists an e* > 0 with


,u(E) <E* SE indt,SEi dt < E/6, and let us set c = m/E*.Slightly modifying the
proof of Lemma 5.3, we see that there exists a set A T and a compact set K c Q,
such that SAi dt, SAindt < E/6 for every n, in(t)- i(t), >-n -+ >-(t) uniformly for
t 0 A, i(t), in(t),xn(t)e K for all n and t 0 A.
By desirability, Ut(y + E/6) - u,(y) > 0 for almost every t, and by continuity of u,
there exists a 3(t) > 0 with u,(y + E/6) - u,(y) > 6(t) for y E cK (since K is compact).
By the principle of borel choice [1], we may have chosen a. measurable b(t) > 0
with ((t) < 3(t), i.e. u,(y + e/6) - u,(y) > b(t) for almosf every t and for every
y E cK. There exists a set A1 c Twith p(A1) < E*and b(t) > ( > 0 for t 0 A1.
We have constructed A so that >-n -+ >- uniformly for t 0 A. Hence Ut n(Y)-+ UX(Y)
uniformly for y E (cK + (E/6))and t 0 A. Let N be such that n > N implies that
IUtn(Y)- ut(y)l < (/4 for y E (cK + (e/6)) and t 0 A, IISF indt - SF i dtIl x < 8/6
for every F c T. If y >-S xn we may assume w.l.o.g. (in the same way as in the
proofs of Theorems C and D) that there exists a set B c S with I(B) < 6* and
y(t) e cK for t 0 B. Set D = A u A1 u B; U,(Y(t))> U,(Xn(t))and ut,(y(t) + 8/6) -
ut,n(y(t)) > (/2 for t E S - D. Hence Ut,n(y(t) + e/6) > U(,ny(t)) + (/2 > ut(y(t)) +
(/4 > ut(xn(t)) + (/4. Since Xn(t)e K for t 0 A, Iut(xn(t))- ut,n(xn(t))I < (/4, so
that ut,n(x(t)+ s/6) > Ut,n(Xn(t)). Hence y(t) + 8/6 >-nS D Xn-
Since xn is in the ordinary e'-core of the market (in, >- n), we may proceed as in the
proofs of Theorems C and D and find that fs y dt $; fs i dt e (s + e').

F: By
PROOF OF THEOREM Theorem C it suffices to prove the existence of in, Xn,
>n zwith in -?i, xn x, >- n,-+ >-, and JT indt = STXndt = JT i dt, in(t),xn(t),i(),
-+
x(t) < z(t) a.e., (and in, Xn, >.-n are simple), since then in ~xn, i, x are contained in a
weakly sequentially compact set in L1. Since E is a separable metric space, we may
show as in [7, p. 36] that there exists a sequence of simple functions { >n(t)} with
>40 (t+>-(t) a.e. (In [7] this theorem is stated for B-space valued functions, but the
same proof works for functions with values in separable metric spaces.) It is easy
to construct in(t),xn(t),having the required properties.

PROOFOF THEOREMD': Let C, K, (, p, and N have the same meaning as in the


proof of Theorem C'. Let y >-s x. Considering the identity
8 8
Utn (At) + - =
ut,n(xn(t)) utn(Y(t) + - ut(YtY +

+ ut y(t) + - - ut(y(t)) + ut(y(t)) - ut(x(t)) + ut(x(t)) - ut(xn(t))

+ ut(xn(t)) - Ut,n(Xn(t))
for t in a set B chosen as in the proof of Theorem C', we find (for n > N) from (6.7)-
(6.10) that y(t) + e/4 >- nB xn(t). Since xn(t) is in the restricted a'-core of (in, >- n),
it follows that
8
y(t) + dt $;j'i dt - a'y(B).

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 811

An argument similar to the argument at the end of the proof of Theorem C' yields
the conclusion of Theorem D'.

PROOF OF THEOREM E': Let C, K, and ( be the same as in the preceding


proofs. Let U = e
{i2(t) Loo, >- 2(t)e T; 112(t) -i(t)I L, < e/4, P( >2(t), >- (t)) <
(6/3(1 + C2)). Let y >-s x where x is in the restricted g'-core of (ij, >- 1) where
(i1, >- 1)E U, and ST il dt = fT i dt, and let B be chosen as in the proof of Theorem
C'. Using the identity

u1.,(y(t) + - u1,,(x(t))= u1.,(Y(t)


?W - u,(t) +

+ ut(y(t) + 4 - ut(y(t))+ ut(y(t))- u(x(t)) + u,(x(t)) - uj,(x(t))

for t e B (here u1,t is the utility function of >- 1(t)), and observing that analogous
inequalities to (6.7)-(6.9) hold for (i, >) e U, we find that y(t) + 8(4 >-1B x(t) and
the theorem follows.

The proof of Theorem F' is similar to the proof of Theorem F (but simpler) and
will be omitted.

7. DIFFERENT MARKETS WITH THE SAME INITIAL DISTRIBUTION

G. Debreu has remarked that there exists a serious difficulty in the application
of the continuous model to the (finite) economic situation. It may occur that (i, >-),
(1, S) describe the same economic situation, but their cores are not equivalent.
Suppose, for example, that in the economy there are two types of commodities
and that all the traders have the same preference order >-, whose indifference
curves are shown in Figure 2.

x,

FIGURE 2

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
812 YAKAR KANNAI

x2

1 2 XI
FIGURE 3

Let ij(t) = 1 + 2t (mod 1),j = 1, 2. Then the allocation

i( ) ={ 1+ 2t if t > ,

x2(t) = 2 + 2t
is in the core.
The heavy segment in Figure 3 shows the initial assignment i(t), each point on the
segment representing the commodity bundle of two traders, and the dashed seg-
ments represent x(t) (each point on these segments is the commodity bundle of
one trader).The vector (1, 1) is a price vector.

FIGURE 4

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 813

Now let !,(t) -1 + t, j = 1, 2. Every point on the heavy segment is the com-
modity bundle of one (and only one) trader. No equilibrium allocation of this
market is such that both the dashed segments represent commodity bundles held
by traders. For, if (p, X(t))is a competitive equilibrium, the X(t) is on the line {y:
p *y = p l(t)} and this line intersects the heavy line only once. Hence if one of the
dashed lines is in the image of X(t), then every point on the heavy line was used,
so that none of the points of the second line is obtained. (See Figure 3.)
Thus we have two markets which represent the same economic situation, for
we have no reason to prefer i(t) to l(t) when we want to describe a uniform distribu-
tion of goods in the initial state, even if we may regard the market as the union of
two markets. We see that there exists an allocation in the core of one of the models
with no allocation in the core of the other model having the same distribution of
commodity bundles. It is not difficult, however, to see that there is a sequence X'(t)
of equilibrium allocations of (i, S) such that the distributions of the xF(t)tend to the
uniform distribution (see Figure 4).
This fact led R. J. Aumann to the following conjecture.

Conjecture: If (i, >-), (i, VII)represent the same economic situation and if x is an
allocation in the core of (i, >-), then there is a sequence X, of allocations in the core
of (i, ;) whose distributions converge to the distribution of x.
Before attempting to prove such a conjecture, we have to define precisely when
(i, >-) and (l, :) represent the same economic situation (or have the same distribu-
tion). This is not obvious if >-(t) is not a constant (as it was in the preceding
example). However, we shall give a definition here, and we will prove the existence
of a sequence xn(t) in the restricted c-core of (i, S) converging in distribution to
x(t) (assuming that i is essentially bounded and >-(t) is contained in a compact set).
DEFINITION:Let S be a topological space and f, g: T -+ S be measurable. We
say thatf and g have the same distribution if for every open U c S, ,u(f - '(U)) =
1(g- '(U)).
Let f: T -+ S be measurable. Denote by uf-' the measure defined on the borel
subsets of S by ,uf- '(V) = p(t E T: f (t) E V) for a borel subset V c S.
LEMMA 7.1: Let S be a metric space and let f, g: T -+ S be measurable. Then f
and g have the same distribution if and only iffor every real valued, continuous,and
boundedfunction p defined on S, fS p dluf-' = fs p dug- '.
PROOF: If f and g have the same distribution then for every open U c S,
(f-'(U)) f= 1u(-'(U)) and therefore 1,uf-(U) = g-'(U) for every borel U, so
that jif- ' g,g'- and fs5pduf -' = fs(pdMg '. Assume that fsJ dif'- =
'
psPd,ug- for every bounded, continuous, and real valued p defined on S, and let U
be an open subset of S. Define x(x) = max [1 - p(x, S - U), 0]. Then X is real
valued, continuous, and bounded, X(x) = 1 if and only if p(x, S - U) = 0, i.e.,
if and only if x ? U (since S - U is closed). If x E U, then 0 < X(x) < 1. Hence
1 - [X(x)]nn-, Xu(x),where Xu(x)is the characteristic function of U.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
814 YAKAR KANNAI

But

(1 - [x(x)]n) dmlf- 1 = f (1 - [X(f(t))])dt,

(1 - [X(X)1n)
dug 1 = f (1 - [X(g(t))],)dt

so that by Lebesgue's dominated convergence theorem,

(1 - [X(x)]n dj{f - 1 Xu(f(t))dt = p(f '(U)),


and

f(1 - [X(x)]n) 1 -j(g1(U)).


dlg-
But JS(1 - d,g- 1 = JS(1 - [X(x)]n)
[X(x)]n) dmf-; hence 1u(f '(U)) = 1u(g '(U)).

DEFINITION: Let fn,ff: T -- S be measurable (S a topological space). We say


that fn converge in distribution to f if for every real valued, continuous, and
bounded function defined on S,

fp dpuf- 1 - J'p duf'-

LEMMA 7.2: Let S be a metric space, let fn, f, gn: T S be measurable,let fn, gn
have the same distribution,and letfn f a.e. Then gn f in distribution.

PROOF: By Lemma 7.1, fs p dmf,7' = fsp dgg- for every continuous and
bounded p: S -+ R. Hence it suffices to show that

J pdjif -n
X dff.
p
But

fp dltf7' = T
p(fn(t)) dt, djf -1 = p(f(t))dt

and since p is continuous, p(fn(t)) -(+p(f (t)) a.e. in T, and since p is bounded we
may use Lebesgue's dominated convergence theorem to see that ST(p(fn(t))dt
f TP(f (t)) dt.

The two markets (i, >-), (i, 5) have the same distribution if the functions (i, >):
T -Q ? x 3, (i, S): T -* Q x 3 have the same distribution. (It is clear that 2 x 3
is a metric space since Q and 3 are metric spaces). Let x be an allocation in the core
of (i, >-), so that there exists a price vector p such that (p, x(t)) is a competitive equili-
brium. We assume that i is essentially bounded, i.e., there exists a constant M

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions
CONTINUITY PROPERTIES 815

with Ii(t)l< M a.e. (This is no serious restriction if we deal with models of actual
markets.) Then the sets {y eQ2:p y < p i(t)} are bounded a.e. since p > 0,
[3, p. 46] and therefore x(t) is also essentially bounded. By Theorem F' there exist
,,xn n >-fnwithfT indt = STi dt, in i, >-n
in >-a.e. and for every E> O, xn isin the
restricted s-core of (in, >-n), and n, >-n are simple functions. By Theorem F' in, Xn
are uniformly essentially bounded. According to the construction of (in >n) (see
the proof of Theorem F and [7]), it is clear that there exist simple (in, n) having
the same distribution as (in, >-n) and in -? 1,7n a.e. In fact, for every n there
exists an automorphism Un of T onto itself such that ln(t) = in(Un(t)), Sn(t) =
>n(Un(t)). Define xn(t) by Xn(t) = xn(Un(t)). Then STXndt = STXndt = T in dt =
ST idt = ST i dt, and for every E > 0, xn is in the restricted s-core of the market
(in, i;n) for large n. Hence, by Theorem E', Xn is in the restricted s-core of (1, S)
for large n, and by Lemma 7.2 xn -+ x in distribution.
Note that if we do not assume that ;(t) is contained in a compact set, we get a
sequence Xn(t)of allocations in the ordinary s-core of (i, >-) (assuming still that i is
essentially bounded).

The Hebrew University of Jerusalem

Manuscript received March, 1968.

REFERENCES

[1] AUMANN, R. J.: "Existence of Competitive Equilibria in Markets with a Continuum of Traders,"
Econometrica, 34 (1966), 1-17.
[2] "Integrals of Set-valued Functions," Journal of Mathematical Analysis and Application,
12 (1965), 1-12.
[3] "Markets with a Continuum of Traders," Econometrica, 32 (1964), 39-50.
[4] DEBREU, G., AND H. SCARF: "A Limit Theorem on the Core of an Economy," International
Economic Review, 4, (1963), 235-246.
[5] DUNFORD, N., AND J. T. SCHWARTZ:Linear Operators, Part I. Interscience Publishers, New York,
1958.
[61 EDGEWORTH,F. Y.: Mathematical Psychics. Kegan Paul, London, 1881.
[7] HILLE, E., AND R. J. PHILLIPS: Functional Analysis and Semi-Groups, American Mathematical
Society Coll. Publications 31, 1957.
[8] KELLEY,J.: General Topology. Van Nostrand, 1955.
[9] SHAPLEY,L. S., AND M. SHUBIK: "Quasi-Cores in a Monetary Economy with Non-convex Prefer-
ences," Econometrica, 34 (1966), 805-827.

This content downloaded from 152.81.7.38 on Wed, 5 Nov 2014 10:57:07 AM


All use subject to JSTOR Terms and Conditions

Вам также может понравиться