Journal of Mathematical Economics Equilibrium theory with

Journal of Mathematical Economics Equilibrium theory with

Journal of Mathematical Economics 44 (2008) 1266–1283 Contents lists available at ScienceDirect Journal of Mathematical Economics journal homepage: ...

461KB Sizes 0 Downloads 11 Views

Recommend Documents

General equilibrium theory - NYU Economics
Nov 4, 2014 - Walras Law: pz (p, ω)=0, ∀p >> 0; lower boundedness: ∃s such that zl (p, ω) > −s, ∀l ∈ L; boun

journal of mathematical economics - Elsevier
In the Editors view, the formal mathematical expression of economic ideas is of vital importance ... to any field of eco

Journal of Mathematical Economics Inductive game theory - CiteSeerX
M. Kaneko, J.J. Kline / Journal of Mathematical Economics 44 (2008) 1332–1363. 1333 situation. Nevertheless, experienc

ECONOMIC THEORY AND MATHEMATICAL ECONOMICS
Franklin M. Fisher and Karl Shell. The Economic Theory of Price Indices: Two Essays on the Effects of Taste, Quality, an

Four Aspects of the Mathematical Theory of Economic Equilibrium
Vancouver, 1974. Four Aspects of the Mathematical Theory of. Economic Equilibrium. Gerard Debreu*. The observed state of

Journal of Mathematical Economics Unbounded - Ceremade
Arbitrage and equilibrium in unbounded exchange economies with satiation. Journal of Mathematical Economics 42 (6), 661â

Positive Theory of Equilibrium
and they are equal to 0 (because of Walras law p.Z(p) = 0):. Z(p)=0. 3. f satisfies the assumptions of Kakutani Theorem.

Equilibrium theory
Abstract: We consider a standard Walrasian pure exchange economy. From an economic point of view, equilibrium in such a

General Equilibrium with Endogenously Incomplete - NYU Economics
are upper-hemicontinuous convex valued correspondences. Proof It trivially suffices to prove upper-hemicontinuity for d

Mathematical Methods of Economics
mathematical problems in engineering, in crystallography, in geophysics, and in cell biology. At Caltech he has twice re

Journal of Mathematical Economics 44 (2008) 1266–1283

Contents lists available at ScienceDirect

Journal of Mathematical Economics journal homepage: www.elsevier.com/locate/jmateco

Equilibrium theory with unbounded consumption sets and non-ordered preferences Part I. Non-satiation夽 Dong Chul Won a,1 , Nicholas C. Yannelis b,∗ a b

College of Business Administration, Ajou University, Woncheon-dong, Yeongtong-Gu, Suwon, Kyunggi 443-749, South Korea Department of Economics, University of Illinois at Urbana Champaign, 330 Wohlers Hall 1206 S. Sixth Street, Champaign, IL 61820, United States

a r t i c l e

i n f o

Article history: Received 17 August 2005 Received in revised form 25 April 2008 Accepted 25 April 2008 Available online 15 May 2008 JEL classification: D51 D52 C62 G12

a b s t r a c t A new condition is introduced for the existence of equilibrium for an economy where preferences need not be transitive or complete and the consumption set of each agent need not be bounded from below. The new condition allows us to extend the literature in two ways. First, the result of the paper can cover the case where the utility set for individually rational allocations may not be compact. As illustrated in Page et al. [Page Jr., F.H., Wooders, M.H., Monteiro, P.K., 2000. Inconsequential arbitrage. Journal of Mathematical Economics 34, 439–469], the no arbitrage conditions do not apply to an economy with a non-compact utility set. Second, we generalize the arbitrage-based equilibrium theory to the case of non-transitive preferences. © 2008 Elsevier B.V. All rights reserved.

Keywords: Non-ordered preferences Unbounded consumption sets Arbitrage

1. Introduction The consumption set need not be bounded from below in an asset market economy where unlimited short sales are allowed. The existence of a Walrasian equilibrium with unbounded-from-below choice sets was initially addressed in Hart (1974) who introduced a condition on preferences eventually known as a no arbitrage condition. To generalize the condition of Hart (1974) on preferences, different arbitrage notions have been introduced in the literature (see for example, Hammond, 1983; Page, 1987; Werner, 1987; Chichilnisky, 1995; Page et al., 2000; Dana et al., 1999; Allouch, 2002, among others). The arbitrage conditions are not only sufficient but also necessary for the existence of a Walrasian equilibrium in certain cases. The notion of arbitrage, however, has some limitations as a conceptual framework for explaining equilibrium beyond either the transitivity of preferences or the compactness of the utility set for individually rational allocations. First of all, equilibrium theory with unbounded consumption sets makes use of the assumption of transitivity of preferences. The no

夽 We wish to thank Nizar Allouch, Cuong Le Van, Guangsug Hahn, and a competent referee for their comments and suggestions. ∗ Corresponding author. 1

E-mail addresses: [email protected] (D.C. Won), [email protected] (N.C. Yannelis). The author gratefully acknowledges the research grants of the Guwon Scholarship Foundation.

0304-4068/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jmateco.2008.04.005

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1267

arbitrage conditions of the literature are based upon a utility representation of preferences.1 One may be tempted to consider a naive extension of the notions of arbitrage to the case with non-transitive preferences simply by dropping the transitivity of preferences in their definition. We provide an example where the naive extension of the no arbitrage conditions fails to explain the existence of equilibrium with non-transitive preferences. Specifically, the set of Pareto optimal allocations is compact but the naively extended conditions fail in the example. This example illustrates the difficulty of finding sufficient conditions for the existence of equilibrium with non-transitive preferences as an extension of the no arbitrage conditions. To the best of our knowledge, there is no literature which attempts to extend the notion of arbitrage to the case of non-transitive preferences in the framework of Hart (1974). Another limitation of the arbitrage-based approach lies in its capability to explain the existence of equilibrium even in the case with transitive preferences. As illustrated in Page et al. (2000), the no arbitrage conditions do not apply to the case where the utility set for individually rational allocations is not compact. Specifically, the no arbitrage conditions are no longer necessary for the existence of equilibrium in economies with a noncompact utility set. The purpose of this paper is to prove the existence of equilibrium with unbounded consumption sets and non-transitive preferences. To this end we introduce a new condition which subsumes as a special case all the arbitrage conditions found in the literature in two respects. First, it covers the case where the utility set for individually rational allocations is compact and moreover, it explains the existence of equilibrium in the counterexample of Page et al. (2000) to the no arbitrage conditions. Second, we generalize the arbitrage-based equilibrium theory to the case of non-transitive preferences. In particular, the result of the paper applies to the aforementioned economy with non-transitive preferences in which the set of Pareto optimal allocations is compact, an equilibrium exists, but any known conditions are violated. In other words, the economy of the example has all the desired properties except for the transitivity of preferences, however, the existence of equilibrium cannot be explained by any conditions found in the existing literature. Furthermore, our analysis covers interdependent preferences. Thus, the equilibrium existence result of the paper includes as a special case not only all the equilibrium existence results with unbounded consumption sets but also gives as a corollary the standard equilibrium existence results without transitivity or completeness of preferences. One example of the economies with non-ordered preferences on unbounded choice sets is a recent development of the capital asset pricing model (CAPM). Traditional CAPMs assume that agents’ preferences are represented by a mean-variance utility function. Recently, Boyle and Ma (2006) show that the return–risk relationship of the CAPM can hold in the case where agents are risk averse with respect to mean-preserving spread (MPS).2 The MPS-risk-averse preferences need not be transitive and thus, admit no utility representation in general. It is also widely recognized in the literature that transitive preferences fail to explain important anomalous phenomena such as preference reversal.3 For example, experimental methods, Grether and Plott (1979) and Loomes and Sugden (1991) among others document the violation of transitivity of preferences by detecting the preference reversal phenomenon. There have been also controversies over the presence of money pump which is frequently mentioned as a refutable evidence against non-transitive preferences. These controversies, however, are quite misleading from the viewpoint of equilibrium theory. Non-transitive preferences may allow a triad of objects {A, B, C} which contributes to money pumps. Suppose that there exists an agent who strictly prefers A to B, B to C, and C to A, and he is currently endowed with A. Then an arbitrageur can pump money out of his pocket by offering repeatedly to exchange A for C, then C for B, and then B for A at a fee small enough to induce him to accept each offer. The literature of decision theory argues that money pumps are hard to find in the real world and therefore, the non-transitivity of preferences is not convincing. It will be illustrated later (Example 3.1.2) that the naive arguments are irrelevant to confirming the non-transitivity of preferences because the money pump never works and therefore, is not observable in equilibrium. The paper is organized as follows: In Section 2 we present an auxiliary theorem which generalizes the standard equilibrium existence results. In particular, the auxiliary theorem is proved via an extension of the abstract equilibrium theorem of Shafer and Sonnenschein (1975), and Borglin and Keiding (1976), and it is the main mathematical tool used to prove our main theorem which is the focus of Section 3. Appendix contains all the proofs of the results of the paper. 2. A first extension of the classical equilibrium existence theorem 2.1. Notation For a set A in a finite-dimensional Euclidean space R , the following notation will be used. • 2A denotes the set of all subsets of the set A.

1

Allouch (2002) is an exception because preferences are allowed to be incomplete. However, Allouch (2002) assumes that preferences are transitive. An agent is said to exhibit MPS-risk-aversion if for any random payoffs X and Y = X + , he prefers X to Y whenever E() = 0 and Cov (X, ) = 0. For more rigorous treatment of MPS-risk-aversion, see Boyle and Ma (2006). 3 Experimental work of Berg et al. (1985) documents that arbitrage profits can be extracted from the optimal choices which reveal the preference reversal phenomenon. Preference reversals arise when an agent prefers lottery A to lottery B but sets a higher selling price on B than on A. Such behavior violates transitivity of preferences. 2

1268

• • • • • • • •

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

conA denotes the cone generated by the set A, i.e., con A = ∪≥0 A. clconA denotes the the closure of the cone generated by the set A. coA denotes the convex hull of the set A. clA denotes the closure of the set A. intA denotes the interior of the set A. ∂A denotes the boundary of the set A. rintA denotes the relative interior of the convex set A.4 ∂r A denotes the relative boundary of the convex set A, i.e., ∂r A = (clA) \ (rintA). We adopt the following additional notation.

• x =



 x2 j=1 j

denotes the Euclidean norm of the vector x = (x1 , . . . , x ) ∈ R .

• xn → x denotes the convergence of the sequence {xn : n = 1, 2, . . .} to the point x. • C ◦ (x, r) denotes the open ball in R centered at the point x with radius r > 0. • C(x, r) denotes the closed ball in R centered at the point x with radius r > 0. 2.2. Definitions For two nonempty subsets Z and Y in R , consider a correspondence ϕ : Z → 2Y . Let clϕ, intϕ and coϕ denote the correspondence from Z to 2Y which has the value clϕ(z), intϕ(z) and coϕ(z) for all z ∈ Z, respectively. The correspondence ϕ is said to have an open graph if Gϕ ≡ {(z, y) ∈ Z × Y : y ∈ ϕ(z)} is open in Z × Y . The correspondence ϕ is said to have open lower sections if the set ϕ−1 (y) = {z ∈ Z : y ∈ ϕ(z)} is open in Z for every y ∈ Y and ϕ is said to have open upper sections (or open-valued) if ϕ(z) is open in Y for every z ∈ Z. The correspondence ϕ is said to be lower semi-continuous if for every open set V of Y, {z ∈ Z : ϕ(z) ∩ V = / ∅} is open in Z and ϕ is said to be upper semi-continuous if for every open set V of Y, {x ∈ Z : ϕ(x) ⊂ V } is open in Z. 2.3. Auxiliary theorem: a generalization of the classical Walrasian equilibrium existence theorems We consider the exchange economy which is populated by finitely many agents in I. We let I denote both the number and the set of agents and  the finite number of commodities. For each i ∈ I, let ei ∈ R denote the initial endowment and Xi ⊂ R X denote the choice set of agent i ∈ I. We denote the exchange economy by E = {(Xi , ei , Pi ) : i ∈ I} where Pi : X → 2 i is a preference correspondence where X = i ∈ I Xi . For points x ∈ X and yi ∈ Xi , we read yi ∈ Pi (x) as “agent i strictly prefers yi to xi provided / i.” For example, given a binary relation i ⊂ X × X, we can define Pi as follows: that the other agents choose xj for all j =

∀x = (x1 , . . . , xI ) ∈ X,

Pi (x) = {yi ∈ Xi : (x1 , . . . , xi−1 , yi , xi+1 . . . , xI ) i x}.

The preference ordering i is so general that it allows interdependence among agents and need not be either transitive or complete. For each p ∈ R \ {0} and each i ∈ I, we define the sets ˇi (p) = {xi ∈ Xi : pxi < pei } and Bi (p) = {xi ∈ Xi : pxi ≤ pei }. / ∅ and Xi is convex. Notice that clˇi (p) = Bi (p) whenever ˇi (p) = Definition 2.3.1.

An equilibrium for the exchange economy E is a pair (p, x) ∈ (R \ {0}) × X such that

(i) xi ∈ Bi (p) for all i ∈ I, (ii)  Pi (x) ∩ Bi (p) = ∅ for all i ∈ I, and (iii) (x − ei ) = 0. i∈I i The pair (x, p) is a quasi-equilibrium for the economy E if it satisfies (i), (iii), and the following condition. 

(ii ) Pi (x) ∩ ˇi (p) = ∅ for all i ∈ I.



(x − ei ) = 0}. For each x ∈ X and each i ∈ I, let Ri (xi ) = {z ∈ X : xi ∈ / Let F denote the set of feasible allocations {x ∈ X : i∈I i coPi (z)}. We set R(e) = {x ∈ X : ei ∈ / coPi (x) for all i ∈ I}. An allocation x ∈ R(e) is called individually rational. Such individual / coPi (x) for all i ∈ I.5 We set H = F ∩ R(e). rationality is appropriate in the sense that if x is an equilibrium allocation, then ei ∈ Let clH denote the closure of H in X. A point x ∈ H is an allocation which is feasible and individually rational. In the special case that the preference ordering of agent i is defined on Xi and representable by a quasiconcave utility function ui for all i ∈ I, R(e) is equal to the set {x ∈ X : ui (ei ) ≤ ui (xi ) for all i ∈ I} and therefore, it is convex. We assume that E satisfies the following conditions for all i ∈ I.

4 5

The relative interior of A is the interior of A in the smallest affine subspace of R which contains A. For all i, let i denote the strict ordering on Xi induced by a reflexive ordering  ∼ i on Xi , i.e., for all zi , xi in Xi , zi i xi if zi  ∼ i xi but not xi  ∼ i zi . In this

case, one might be tempted to define the set of individually rational allocations as



i∈I

{xi ∈ Xi : xi  ∼ i ei }. This is fine as far as i is transitive for all i ∈ I. It

is not the case, however, with non-transitive preferences because xi i ei does not imply Pi (xi ) ⊂ Pi (ei ), as illustrated in Example 3.1.2.

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

B1. B2. B3. B4. B5. B6.

1269

Xi is a closed, nonempty and convex set in R . ei is in the interior of Xi . Pi is lower semi-continuous. For all x ∈ X, xi ∈ / coPi (x). For all x ∈ H, Pi (x) = / ∅. Let x be a point in H. Then for each zi ∈ coPi (x) and vi ∈ Xi , there exists  ∈ (0, 1) such that zi + (1 − )vi ∈ coPi (x).

Assumption B5 states that no satiation occurs on the set of feasible and individually rational allocations. Assumption B6 holds true when Pi (x) is relatively open in Xi for all x ∈ H.6 As shown later, this condition is not required for the existence of quasi-equilibrium. It is used only in verifying that a quasi-equilibrium is an equilibrium. Define i : X → 2Xi by i (x) = coPi (x). For ei ∈ Xi , notice that i−1 (ei ) = {x ∈ X : ei ∈ i (x)} = {x ∈ X : ei ∈ coPi (x)}. Observe that X \ i−1 (ei ) = {x ∈ X : ei ∈ / i (x)} = {x ∈ X : ei ∈ / coPi (x)} = Ri (ei ). Since Pi is lower semi-continuous under the condition B3, i−1 (ei ) need not be open and therefore, Ri (ei ) need not be closed. Consequently, H need not be closed.7 By B4, ei ∈ / coPi (e) for all i ∈ I and therefore, e ∈ R(e). Since e ∈ F, it follows that e ∈ clH. In particular, clH is not empty. Below we prove a very general Walrasian equilibrium existence theorem, which will be used in the proof of our main result. Auxiliary Theorem 2.3.1. if H is bounded.

Suppose that E satisfies the assumptions B1–B6. Then there exists an equilibrium for the economy E

Proof.



See Appendix B.

The auxiliary theorem is based on the assumption B5 and the condition that H is bounded. Thus, this result subsumes as a special case the classical existence theorems of Shafer (1976) and Gale and Mas-Colell (1975), among others, which are built upon the condition that F is bounded. As shown in Appendix, the proof of Auxiliary Theorem is distinct from the standard proof of the classical theorems because it involves delicate arguments in verifying that candidate equilibrium allocations lie in H. 3. Main results We briefly review the notion of arbitrage used in the literature and provide generalizations of the no arbitrage conditions to the case with non-transitive preferences. Two examples are given to motivate the need for the new conditions for the existence of equilibrium. The first example borrowed from Page et al. (2000) illustrates that the no arbitrage conditions are not useful to explain equilibrium when the utility set for allocations in H is not compact. Specifically, the no arbitrage conditions are no longer necessary for the existence of equilibrium in economies with the noncompact utility set. Another limitation of the arbitrage conditions is that they explicitly or implicity assume the transitivity of preferences. One may be tempted to consider a naive extension of the notions of arbitrage to the case with non-transitive preferences simply by replacing ‘transitive preferences’ by ‘non-transitive preferences’ in their definition. The second example shows that such a naive extension of the no arbitrage conditions fails to explain the existence of equilibrium with non-transitive preferences. Moreover, the set of Pareto optimal allocations in H is compact but the naively extended conditions fail in the example.8 The second example illustrates the difficulty of finding sufficient conditions for the existence of equilibrium with non-transitive preferences as an extension of the no arbitrage conditions. This example is also used to demonstrate the irrelevance of the money pump arguments against non-transitive preferences. In the second subsection, we introduce a new condition which can cover the cases not only of non-transitive preferences, but also of noncompact utility sets. In particular, this condition subsumes all the previous no arbitrage conditions as a special case. Moreover, it covers the counterexample of Page et al. (2000). Finally, the main theorem of the paper is provided under the condition which further generalizes the condition of the second subsection.

6





Suppose that Pi (x) is relatively open in Xi . Let zi be a point in coPi (x). Then there exist z  ∈ Pi (x), z ∈ Pi (x) and ˛ ∈ [0, 1] such that zi = ˛z  + (1 − ˛)z . i

i



i

i

Let vi be a point in Xi . Since Pi (x) is relatively open in Xi , there exists  ∈ (0, 1) such that z  + (1 − )vi ∈ Pi (x) and z + (1 − )vi ∈ Pi (x). It follows that 

i

i

zi + (1 − )vi = ˛[z  + (1 − )vi ] + (1 − ˛)[z + (1 − )vi ] ∈ coPi (x). i i 7 If each P has open lower sections and is convex-valued, then R(e) is closed. But this is not warranted by B3. For example, the budget correspondence is i lower semi-continuous but does not have open lower sections. See Yannelis and Prabhakar (1983) for an example of a lower semi-continuous correspondence which does not have open lower sections. 8 A feasible allocation x ∈ X is said to be Pareto optimal if there is no feasible allocation y ∈ X such that y ∈ P (x ) for every i ∈ I. i i i

1270

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

3.1. Arbitrage conditions and counterexamples For illustrative purposes, we consider the case where preferences for agent i are represented by a reflexive ordering  ∼ i on Xi for all i ∈ I. Let i denote the strict ordering on Xi induced by  ∼ i ; for all zi , xi in Xi , zi i xi if zi  ∼ i xi and not xi  ∼ i zi . Define the (x) = P i (xi ), correspondence P i : Xi → 2Xi by P i (xi ) = {zi ∈ Xi : zi i xi }. In this section, we assume that for all x ∈ X and all i ∈ I, Pi and Pi is convex-valued.9 For each xi ∈ Xi , we set Ri (xi ) = {zi ∈ Xi : not(xi i zi )}.10 Then it is easy to see that R(e) = i ∈ I Ri (ei ). Dana et al. (1999) and Page et al. (2000) extend the notions of arbitrage used in Werner (1987) and Page and Wooders (1996) to the case where H need not be bounded, and they compare various notions of arbitrage. Remarkably, Dana et al. (1999) provide a condition which is equivalent to the compactness of the utility set { ∈ RI+ : ∃x ∈ H s.t. 0 ≤ i ≤ ui (xi ) for all i ∈ I} when preferences of agent i are represented by a quasiconcave function ui : Xi → R.11 Allouch (2002) further generalizes the notion of arbitrage to the case where preferences are transitive but need not be complete. CPP. (i) The preference ordering  {xn } in H, there ∼ i on Xni is transitive and reflexive for each i ∈ I. (ii) For each sequence n n n k k k exist a subsequence {x } and a sequence {y } in X convergent to some point y ∈ clH which satisfies yi i xi k for all nk and for all i ∈ I. Allouch (2002) shows that the CPP condition is equivalent to the compactness of the utility set of allocations in H when preferences are numerically representable. The aforementioned literature, however, does not explain the existence of equilibrium in the following two examples. Example 3.1.1. Page et al. (2000), and Monteiro et al. (2000) provide an example where the economy has an equilibrium but the utility set is not compact. They show that the example does not satisfy any known conditions based on arbitrage. It is assumed in the economy that I = {1, 2}, X1 = X2 = R2 , e1 = e2 = (0, 0). Agent 1’s utility function is given by



u1 (v) =

v1 , v1 − , v2

if v1 ≤ 0, or v2 ≥ −1 if v1 ≥ 0, and v2 ≤ −1

while preferences of agent 2 is given by u2 (w) = w1 + 2w2 . The utility set is not compact.12 It is easy to see that the economy has an equilibrium (p, v, w) where p = (1, 2), and v = (2, −1) and w = (−2, 1) are an optimal choice for agent 1 and 2, respectively. Example 3.1.2. We consider a two-agent, two-good economy where agents have non-transitive preferences in R2 with e1 = e2 = (0, 0) ∈ R2 . For all v = (v1 , v2 ) ∈ R2 , we introduce the sets



W 1 (v)

=



W 2 (v) =

{v = (v1 , v2 ) ∈ R2 : v1 > 0 & v2 ≥ 0} ∪ {(0, 0)} {v = (v1 , v2 ) ∈ R2 : v2 ≥ v2 }

if v = (0, 0) if v = / (0, 0)

{v = (v1 , v2 ) ∈ R2 : v1 ≥ 0 & v2 > 0} ∪ {(0, 0)} {v = (v1 , v2 ) ∈ R2 : v1 ≥ v1 }

if v = (0, 0) if v = / (0, 0).

For each i = 1, 2, we define



P 1 (v) = P 2 (v) =



2 R++ {v =

2   i 2  ∼ i on R such that v  ∼ i v if v ∈ W (v). Then it follows that for all v ∈ R ,

(v1 , v2 ) ∈ R2 : v2 > v2 }

2 R++ {v = (v1 , v2 ) ∈ R2 : v1 > v1 }

if v = (0, 0) if v = / (0, 0) if v = (0, 0) if v = / (0, 0).

Since (−1, 2) ∈ P 1 (1, 1) and (1, 1) ∈ P 1 (0, 0) but (−1, 2) ∈ / P 1 (0, 0), P 1 is not transitive. Preferences of agent 1 are incomplete because (0, 0) and (−1, 1) are not comparable. Similarly, we can show that P 2 is non-transitive and incomplete. It is easy to see that the sets R1 (e1 ) and R2 (e2 ) is written as R1 (e1 ) = {v = (v1 , v2 ) ∈ R2 : v2 ≥ 0} R2 (e2 ) = {v = (v1 , v2 ) ∈ R2 : v1 ≥ 0}. We show that the economy satisfies all the conditions imposed by B1–B6. Clearly, P 1 and P 2 are convex and open valued. We claim that they are lower semi-continuous. For a point v ∈ R2 , let z = (z1 , z2 ) be a point in P 1 (v). Let vn → v in R2 . Suppose that v = (0, 0). Then we see that z1 > 0 and z2 > 0. Since vn → (0, 0), z2 > v2n for sufficiently large n and therefore, z ∈ R2++ ∩ {v = (v1 , v2 ) ∈ R2 : v2 > v2n }. It implies that z ∈ P 1 (vn ).

9 10 11 12

By Proposition A.1 of Appendix, the convexity assumption can replace B4 without loss of generality. Notice that “not (xi i zi )” means either “not (xi  ∼ i zi )” or “zi  ∼ i xi ”. Dana et al. (1999) and Page et al. (2000) also provide an excellent review of various notions of arbitrage and their economic implications. It is easy to see that u1 (n, −n) → 1 and u2 (−n, n) → ∞ as n → ∞.

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1271

Suppose that v = / (0, 0). Since R2 \ {(0, 0)} is open, vn is in R2 \ {(0, 0)} for sufficiently large n. On the other hand, z2 > v2n so that we have z ∈ P 1 (vn ) for sufficiently large n. Therefore, we conclude that P 1 is lower semi-continuous. By the same argument we can show that P 2 is lower semi-continuous. Since P 1 and P 2 are convex and open valued, B6 trivially holds for the economy. Thus, we see that the economy satisfies B1–B6. In Section 1, we argue that money pumps are not observable in equilibrium. Clearly, the triad {(0, 0), (1, 1), (−1, 3)} constitutes a money pump against agent 1. For any equilibrium price p ∈ R2+ , (1, 1) is not budget-feasible in equilibrium. Agent 1 cannot accept the offer to exchange (0, 0) for (1, 1) at any positive fee because of the budget constraint. Thus, the money pump does not work in equilibrium. We consider an extension of the CPP condition to economies with non-transitive preferences simply by dropping (i) and keeping (ii) of the CPP condition. We show that the naive extension of the CPP condition is not satisfied with the example. For each n, set x1n = (−an , bn ) and x2n = (an , −bn ) for some an and bn in R. We assume that xn = (x1n , x2n ) is in H for all n. Then we have an ≥ 0 and bn ≥ 0. Suppose that an and bn are strictly positive and increasing for all n, and an → ∞ and bn → ∞. Then the distance of P i (xin ) from the origin goes to infinity as n → ∞. Thus, there is no bounded sequence {yn } which satisfies yin ∈ P i (xin ) for all n and all i = 1, 2, and therefore, (ii) of the CPP condition is violated.

Remark 3.1.1. The economy of Example 3.1.2 has the desired properties in terms of preferences except for the nontransitivity. Moreover, the initial allocation {(e1 , e2 )} is the unique Pareto optimal allocation in H and thus, the set of Pareto optimal allocations in H is trivially compact. Nonetheless, there is no literature which covers Example 3.1.2. The classical works do not apply simply because consumption sets have no lower bound. The arbitrage-based literature following the seminal work of Hart (1974) is not applicable because preferences are not transitive. Moreover, the existence of equilibrium in Example 3.1.2 cannot be explained by the version of the CPP condition of Allouch (2002) where the transitivity of preferences is dropped. In the subsequent sections, we will discuss new conditions which can cover Examples 3.1.1 and 3.1.2. 3.2. New conditions without transitivity We provide a new sufficient condition for the existence of equilibrium without transitivity of preferences. This condition encompasses all the arbitrage-related conditions as a special case. Remarkably, it is illustrated that the utility set need not be compact under this condition. In particular, our new condition is satisfied in the economy of Example 3.1.1 which is given by Page et al. (2000). For a point x ∈ X and i ∈ I, we set ri (x) = max{xj , j = / i}. / i. We make the following assumption. Note that the closed ball C(0, ri (x)) contains xj for all j = B7a. There exists h ∈ I such that for any sequence {xn } in H, there exist a subsequence {xnk }, and a sequence {ynk } convergent to a point y ∈ clH such that for all nk , Ph (ynk ) ⊂ con[Ph (xnk ) − {eh }] + {eh }

(1)

and for all i = / h, Pi (ynk ) ∩ C(0, rh (xnk )) ⊂ con[Pi (xnk ) ∩ C(0, rh (xnk )) − {ei }] + {ei }.

(2)

It will be illustrated that B7a may hold in economies with a noncompact utility set and therefore, it is strictly more general than the no arbitrage conditions of the literature. The asymmetric treatment of agents in the condition B7a deserves a special remark. Remark 3.2.1. The asymmetric treatment of agents in B7a plays a crucial role in extending the no arbitrage conditions beyond the class of economies which have the compact utility set for individually rational allocations. There are two conceivable conditions which give symmetric treatment to agents. The first one is the following alternative to B7a: for any sequence {xn } in H, there exist a subsequence {xnk }, and a sequence {ynk } convergent to a point y ∈ clH such that for all nk and all i ∈ I, Pi (ynk ) ∩ C(0, r(xnk )) ⊂ con[Pi (xnk ) ∩ C(0, r(xnk )) − {ei }] + {ei }.

(3)

where for a point x ∈ X, r(x) = max{xi , i ∈ I}. This alternative condition is quite restrictive because, as illustrated below, it may fail to explain the existence of equilibrium in economies with a noncompact utility set. The other conceivable symmetric treatment of agents is to apply the restriction on agent h represented by (1) to the other agents: for any sequence {xn } in H, there exist a subsequence {xnk }, and a sequence {ynk } convergent to a point y ∈ clH such that for all nk and all i ∈ I, Pi (ynk ) ⊂ con[Pi (xnk ) − {ei }] + {ei }

(4)

This condition is weaker than B7a but is not required for the existence of equilibrium. For example, the condition is satisfied in the economy with two agents and two goods where preferences of the agents are represented by a utility function u1 (v) = v2 and u2 (v) = v1 on R2 , but the economy has no equilibrium.

1272

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

Fig. 1. A non-compact utility set.

The following condition is a simple extension of the CPP condition of Allouch (2002) to the case where preferences need not be transitive. B7b. For any sequence {xn } in H, there exist a subsequence {xnk }, and a sequence {ynk } convergent to a point y ∈ clH such that for all nk and all i ∈ I, Pi (ynk ) ⊂ Pi (xnk ).

(5)

Condition B7b can be considered as a simple extension of the CPP condition of Allouch (2002) to the case where preferences need not be transitive. Clearly, B7a subsumes B7b and therefore, the CPP condition of Allouch (2002) becomes a special case. As mentioned earlier, the CPP condition is equivalent to the compactness of the utility set of allocations in H when preferences are numerically representable. Moreover, the no arbitrage conditions used by Werner (1987); Page (1987) or Dana et al. (1999) imply the compactness of the utility set and therefore, they also imply B7a.13 The following example shows that B7a may hold with the noncompact utility set for allocations in H. Example 3.2.1. We continue with Example 3.1.1. The utility set for allocations in H is not compact. As illustrated below, however, the economy satisfies B7a. We recall that there exists an equilibrium (p, v, w) where p = (1, 2), and v = (2, −1) and w = (−2, 1). Note that w + v = 0. Let {(vn , −vn )} be a sequence of allocations in H. By setting x1n = vn and x2n = −vn for all n, we can match the notation of B7a. If {(vn , −vn )} is bounded, we set y1n = vn and y2n = −vn for each n. Then it is trivial to see that {xn } and {yn } satisfy (1) and (2). From now on, we assume that vn  → ∞. We set y1n = v and y2n = w for all n, and y1 = v and y2 = w. Note that for each n, P 2 (y2n ) = P 2 (w) is the open half space above the line through b and B. Since (vn , −vn ) ∈ H and vn  → ∞, without loss of √ generality, we can assume that vn lies in the fourth quadrant of R2 and vn  > 5 for all n (see Fig. 1). Then −vn lies in the arc AB while vn in the arc ab. Noting that e2 = (0, 0), we see that for all n, con[P 2 (−vn ) − {e2 }] − {e2 } = conP 2 (−vn ), and conP 2 (−vn ) is the open half space above the line through b and B. Recalling that w = −v, for all n we have P 2 (w) = con[P 2 (−vn ) − {e2 }] + {e2 }. Since e1 = (0, 0) and b is the best point for agent 1 in the arc ab, it follows that P 1 (b) = P 1 (v) ⊂ P 1 (vn ) for all n and therefore, P 1 (v) ∩ C(0, vn )

⊂ P 1 (vn ) ∩ C(0, vn ) = [P 1 (vn ) ∩ C(0, vn ) − {e1 }] + {e1 } ⊂ con[P 1 (vn ) ∩ C(0, vn ) − {e1 }] + {e1 }.

Thus, we conclude that the economy satisfies B7a.

13

A complete comparison of the no arbitrage conditions can be found in Dana et al. (1999), and Page et al. (2000).

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1273

In Remark 3.2.1, we have mentioned that the alternative condition to B7a where every agent is subject to the restriction represented by (3) may fail to explain the existence of equilibrium. This is true in this example. It is easy to see that for each n with −vn ∈ P 2 (w), P 2 (w) ∩ C(0, vn ) ⊂ / P 2 (−vn ) ∩ C(0, vn ), and therefore, P 2 (w) ∩ C(0, vn ) ⊂ / con[P 2 (−vn ) ∩ C(0, vn ) − {e2 }] + {e2 }. Thus, the restriction on each agent imposed by (3) does not hold for the economy. 3.3. Main existence theorems In this section, we provide the main existence theorem of the paper based on a condition which generalizes B7a. As illustrated below, B7a calls for further generalization because it does not cover the economy of Example 3.1.2. The exemplary economy is standard except for the non-transitivity of preferences. Nevertheless, the existence of equilibrium of the economy cannot be explained by any known conditions including B7a. The main existence result of the paper is based on a generalization of B7a which covers the economy of Example 3.1.2. The new condition to be discussed is motivated by the following distinct characterization of the equilibrium conditions. For a point x ∈ X, we define the set G(x) =



clcon[Pi (x) − {ei }].

i∈I

It is shown below that the set G(·) summarizes the conditions for x to be a quasi-equilibrium for the economy E. Lemma 3.3.1. Suppose that for all x ∈ H and all i ∈ I, Pi (x) is convex and xi ∈ clPi (x). Then for a point x ∈ H, G(x) = / R if and only if there exists a nonzero vector p ∈ R such that pz ≥ 0 for all z ∈ G(x), i.e., pei ≤ pzi for all zi ∈ clPi (x) and all i ∈ I. Proof.

See Appendix B.



This lemma shows that for a point x ∈ H, G(x) = / R is a necessary and sufficient condition for the existence of p ∈ R such that (p, x) is a quasi-equilibrium for the economy E. The condition that G(x) = / R will be guaranteed in the proof of the main theorem of the paper by a fixed point theorem. In addition to the conditions of Lemma 3.3.1, suppose that B2 and B6 hold for the economy. Then by the same arguments made in Step 5 of the proof of Theorem 2.3.1, we can show that (p, x) is an equilibrium for the economy E. The following result is immediate from the assumptions B2 and B6 and the results of Lemma 3.3.1. Proposition 3.3.1. Suppose that B2 and B6 hold, and for all x ∈ H and all i ∈ I, Pi (x) is convex and xi ∈ clPi (x). Then for any x ∈ X, x ∈ H and G(x) = / R if and only if there exists a nonzero p ∈ R such that (p, x) is an equilibrium for the economy E. Proposition 3.3.1 shows that for an allocation x ∈ H, G(x) = / R is a necessary and sufficient condition for the existence of equilibrium. Thus for an allocation x ∈ H, the set G(x) fully characterizes the conditions for x to be an equilibrium allocation. For a point x ∈ X and i ∈ I, we define the set Gi (x) = clcon[Pi (x) − {ei }] +



clcon[Pj (x) ∩ C(0, ri (x)) − {ej }].

j= / i

Let Ei (x) denote the economy which is the same as E except for that agent i has the consumption set Xi and each j = / i has the consumption set Xj ∩ C(0, ri (x)) which is equal to the consumption set Xj truncated by the closed ball C(0, ri (x)). The economy Ei (x) will be used in the proof of the main theorem of the paper. The following corollary shows that the set Gi (x) summarizes the conditions for x to be equilibrium of the economy Ei (x). Corollary 3.3.1. Suppose that for all x ∈ H and all i ∈ I, xi ∈ clPi (x). Let x ∈ H and i ∈ I such that Pj (x) ∩ C(0, ri (x)) = / ∅ for every j= / i.14 Then Gi (x) = / R if and only if there exists a nonzero p ∈ R such that (p, x) is an equilibrium for the economy Ei (x). Example 3.1.2 illustrates that arbitrage-related conditions may not be useful for the existence of equilibrium with nontransitive preferences. By taking advantage of the properties of the set Gi (·) for each i, we provide new conditions which subsume the no arbitrage conditions as a special case and are relevant to the case of non-transitive preferences. B7. There exists h ∈ I such that for any sequence {xn } in H with {rh (xn )} increasing to infinity, there exist a subsequence {xnk }, and a sequence {ynk } convergent to a point y ∈ clH such that for all nk , Ph (ynk ) − {xhnk } ⊂ Gh (xnk )

(6)

14 By the same arguments made in Step 2 of the proof of Theorem 3.1.1, we can show that x ∈ clP (x) and P (x) ∩ C(0, r (x)) = / ∅ for every j = / i imply j j j i xj ∈ cl[Pj (x) ∩ C(0, ri (x))].

1274

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

Fig. 2. A simple example without satisfying B7a.

and for all j = / h, [Pj (ynk ) ∩ C(0, rh (xnk ))] − {xjnk } ⊂ Gh (xnk ).

(7)

Remark 3.3.1. We have emphasized the importance of the asymmetric treatment of agents in Remark 3.2.1. Since B7 is a generalization of B7a as shown below, Remark 3.2.1 applies to B7 as well. Lemma 3.3.2. Proof.

Suppose that xi ∈ clPi (x) for all x ∈ X and for all i ∈ I.15 Then B7a implies B7.

See Appendix B.



The following illustrates an economy which satisfies B7 but does not B7a. Thus, B7 is strictly more general than B7a. Example 3.3.1. We show that the economy of Example 3.1.2 satisfies B7 but does not satisfy B7a. For each n, set x1n = (−an , bn ) and x2n = (an , −bn ) for some an and bn in R. We assume that xn = (x1n , x2n ) is in H for all n and ank + bnk → ∞. The condition xn ∈ H implies that an ≥ 0 and bn ≥ 0. Since x1n  = x2n , we have ri (xn ) = xin  for all i = 1, 2 and all n. For each n, we take xn such that x2n is on the line a. Let {y2n } be a bounded sequence in R2 (e2 ). Since x2n  → ∞, without loss of generality, we can assume that y2n  < an . Then y2n ∈ Q 2 (e2 ) ∩ C(0, x2n ). As shown in Fig. 2, the cone con[P 2 (x2n ) − {e2 }] + {e2 } is an open set which has the lines 0a and 0b as its boundaries. Thus, we see that P 2 (y2n ) ∩ C(0, x2n ) ⊂ / con[P 2 (x2n ) − {e2 }] + {e2 }. / con[P 1 (x1n ) − {e1 }] + Let {y1n } be a bounded sequence in R1 (e1 ). By the same argument, we can show that P 1 (y1n ) ∩ C(0, x1n ) ⊂ {e1 }. Thus, B7a does not hold for the economy. It is trivial to see that (e1 , e2 ) is a unique equilibrium allocation for each E1 (xn ). That is, any nonzero allocation in F cannot be an equilibrium allocation of E1 (xn ). Then by Corollary 3.3.1, G1 (xn ) = R2 and therefore, B7 holds trivially in this example. With all these preliminary results out of the way, we can now turn to the main existence theorem of this paper. Theorem 3.3.1. Suppose that E satisfies the assumptions B1–B6. Then there exists an equilibrium for the economy E if the condition B7 is satisfied. Sketch of Proof. The rigorous proof of the theorem will be given in Appendix B. The following sketch will be useful to understand the idea of the proof. When the consumption sets are not bounded, traditional approaches to the existence proof rely on the truncation method introduced by Debreu (1959). Moreover, truncations of the consumption sets need to be taken sequentially so that the whole consumption sets can be covered in the limit when F is bounded. In the case where F is unbounded, however, the truncation method alone does not work because no sequential truncations can contain F as a whole. Assumption B7 allows us to circumvent this problem. The idea is to take advantage of information elicited from each truncated economy. (It is worth noting that the truncation method used here differs from the conventional truncation methods in that one agent is allowed to make choices in the whole consumption set in all the truncated economies.) By applying Auxiliary Theorem 2.3.1, we show that each economy with the truncated consumption sets has an equilibrium

15

As mentioned earlier in this section, this condition does no harm to the existence result as far as B4 is satisfied.

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1275

under Assumption B7. Again by B7, there exists a bounded sequence of allocations which are asymptotically supported by the sequence of equilibrium prices for the truncated economies. The limit of those prices and bounded allocations turns out to be an equilibrium of the economy E.  This theorem is a generalization of all the existence results in the literature in several respects. The condition B7 along with the other conditions B1–B6 are sufficient for the existence of equilibrium. As illustrated in Example 3.2.1, the utility set for allocations in H need not be compact under the condition B7. Preferences need not be transitive and moreover, they are interdependent and satisfy the condition B3, weak continuity assumptions on preferences. Most importantly, B7 is much weaker than the no arbitrage conditions used in the literature because all other arbitrage conditions cannot apply to the case where either the utility set is not compact or agents have non-transitive preferences. Remark 3.3.2. One of the referees of the paper suggested an interesting alternative condition for B7 where agents are treated symmetrically. Its formulation is based upon the following definition. Definition 3.3.1. A sequence {Tn } of sets in R is called a compactification procedure of the economy E if it satisfies the following conditions (i) (ii) (iii) (iv)

for each n, Tn is non-empty, closed and satisfies Tn ⊂ Tn+1 , for each x ∈ R and each  > 0, there exists n such that C(x, ) ⊂ Tn , for each n, the set H ∩ TnI is bounded, and for each i ∈ I and each x ∈ R(e), it holds that Pi (x) ∩ T1 = / ∅.

If each Xi is bounded from below, then there exists z ∈ R such that Xi ⊂ {x ∈ R : x ≥ z} for all i ∈ I. In this case, a trivial compactification procedure {Tn } is available where Tn = {x ∈ R : x ≥ z} for all n. The following is the condition suggested by the referee as a possible substitute for B7. B7s. There exists a compactification procedure {Tn } such that for each {xn } with xn ∈ H ∩ Tn , there exist  > 0 and y ∈ clH which satisfy16 Pi (y) ∩ C(yi , ) − {ei } ⊂ Ls{con[Pi (xn ) ∩ Tn − {ei }]}. Clearly, agents are symmetrically treated in B7s. As shown below, an advantage of B7s over B7 is it makes the existence proof much simpler. It is easy to check that B7s is weaker than B7b. We also see that the compactification procedure {Tn } defined by Tn =



(v1 , v2 ) ∈ R2 : v2 ≥ −

1 2

+

1 n



(v1 + n) − n, v2 ≥ −

1 2



1 n





(v1 + n) − n

satisfies B7s in the economy of Example 3.1.1. A disadvantage of B7s, however, is that no systematic theory is yet available which answers the existence of a compactification procedure {Tn } which satisfies the conditions of B7s. Indeed, there would be no definite way of checking B7s in the simple economy mentioned in the end of Remark 3.2.1.17 We now provide an existence theorem based on assumption B7s. Theorem 3.3.1s. Suppose that E satisfies the assumptions B1–B6. Then there exists an equilibrium for the economy E if the condition B7s is satisfied.18 Proof.

See Appendix B.



4. Concluding remarks We have shown the existence of equilibrium in an economy with non-ordered preferences and unbounded-from-below consumption sets. The consequences of the paper not only subsume the arbitrage-based equilibrium theory as a special case but also can cover the case with non-compact utility sets. The main results of the paper stated in Theorem 3.3.1 depend critically on the assumption B5 which excludes the satiability of preferences. One possible extension is to examine the effect of satiation on equilibrium. Such an extension is attempted in Won and Yannelis (2002a,b) as a sequel to the current paper. Allouch and Le Van (2006); Allouch et al. (2006), and Martinsda-Rocha and Monteiro (2007) address the equilibrium existence problem with satiable preferences under the condition that satiation occurs both inside and outside the set of feasible and individually rational allocations. Won and Yannelis (2002a,b) differ from them in that it also covers the case where preferences are possibly satiated only inside the set of feasible and

16 If {A } is a sequence of subsets of R , then Ls{A } is the set of limit points of {A }, i.e., b ∈ Ls{A } if and only if there exists a subsequence {a } such that n n n n nk ank ∈ Ank for each nk and ank → b. 17 Let us call the aforementioned economy ‘economy Z’. Any compactification procedure could be picked up to be tested against the conditions of B7s in the economy Z. It would fail. But this does not say anything about whether B7s holds in the economy Z because there is still an inexhaustible set of compactification procedures to be tested. That is, there is no way of seeing whether B7s holds in the economy Z. 18 The proof of Theorem 3.3.1s to be shown in Appendix B heavily rely on the suggestions of the referee.

1276

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

individually rational allocations. As mentioned in Won and Yannelis (2002a,b), such a distinction is particularly important in the context of the capital asset pricing models. By specializing the framework of Won and Yannelis (2002a,b) in asset pricing models, Won et al. (2008) show the existence of equilibrium in the capital asset pricing model with heterogeneous expectations where mean-variance utility functions reach satiation due to the absence of risk-free assets. One challenging problem is to extend the outcomes of the paper to cover the case discussed in Won (2001) where contract curves are unbounded. Appendix A. Equilibrium in abstract economies As a preliminary step  for the existence of equilibrium, we introduce an abstract economy. For each i ∈ I, let Xi be a nonempty set in R . We set X = i ∈ I Xi . An abstract economy = {(Xi , Ai , Pi ) : i ∈ I} is a set of ordered triples (Xi , Ai , Pi ) where Ai : X → 2Xi and Pi : X → 2Xi are correspondences. The abstract economy provides a simple but powerful conceptual framework for studying an exchange economy in a general setting. Definition A.1.

A quasi-equilibrium for is a point x ∈ X such that for all i ∈ I,

(i) xi ∈ clAi (x) (ii) Pi (x) ∩ Ai (x) = ∅. The point x ∈ X is an equilibrium for if it satisfies (i) and the following condition 

(ii ) Pi (x) ∩ clAi (x) = ∅. We are now ready to provide the following preliminary theorem which will be useful in proving the existence of quasiequilibrium of an exchange economy. Theorem A.1. A1. A2. A3. A4. A5.

Let = {(Xi , Ai , Pi ) : i ∈ I} be an abstract economy satisfying the following conditions for each i ∈ I

Each Xi is convex, compact and nonempty in R . Each Pi is lower semi-continuous. Ai is convex-valued, nonempty-valued and has an open graph. clAi is upper semi-continuous. xi ∈ / coPi (x) for all x ∈ X.

Then has a quasi-equilibrium, i.e., there exists x∗ ∈ X such that for all i ∈ I, (i) xi∗ ∈ clAi (x∗ ), and (ii) Pi (x∗ ) ∩ Ai (x∗ ) = ∅. Proof. For each i ∈ I, define i : X → 2Xi by i (x) = [coPi (x)] ∩ Ai (x). Clearly, i is convex-valued. For each i ∈ I, let Ui = {x ∈ X : / ∅}. Since Pi is lower semi-continuous, by Proposition 2.6 in Michael (1956) coPi is lower semi-continuous. Hence, by i (x) = Lemma 4.2 of Yannelis (1987), i is lower semi-continuous.19 It follows from the lower semi-continuity of i that for each i ∈ I, Ui is open in X (recall that Ui = {x ∈ X : i (x) ∩ X = / ∅}). / ∅ for some i ∈ I. It is easily seen in case (a) There are two cases to be examined: either (a) Ui = ∅ for all i ∈ I or (b) Ui = that for all i and for all x ∈ X, i (x) = [coPi (x)] ∩ Ai (x) = ∅ and therefore, Pi (x) ∩ Ai (x) = ∅. Hence, condition (ii) of the theorem holds. To show that (i) is also fulfilled, we define the correspondence A : X → 2X by A(x) = i ∈ I clAi (x). Since each clAi is upper semi-continuous, closed-valued, convex-valued and nonempty-valued, so is A. By the Kakutani fixed point theorem there exists x∗ ∈ X such that x∗ ∈ A(x∗ ), which implies that xi∗ ∈ clAi (x∗ ) for all i ∈ I. Thus (i) also holds. We turn to case (b). For each i with Ui = / ∅, we denote by i |Ui the restriction of i to Ui , i.e., i |Ui : Ui → 2Xi . Since Ui is 

open in X, it is also paracompact.20 By applying Theorem 3.1 of Michael (1956, p. 368) to function fi : Ui → Xi such that fi (x) ∈ i (x) for all x ∈ Ui . For each i ∈ I, define gi : X → 2Xi by



gi (x) =

{fi (x)}, clAi (x),

i |Ui ,

there exists a continuous

if x ∈ Ui , if x ∈ / Ui .

By Lemma 6.1 in Yannelis and Prabhakar (1983), gi is upper semi-continuous and it is clearly convex-valued, nonempty-valued and closed-valued. Define g : X → 2Xi by g(x) = i ∈ I gi (x). Then g is upper semi-continuous, convex-valued, nonemptyvalued, and closed-valued. By the Kakutani fixed point theorem, there exists x∗ ∈ X such that x∗ ∈ g(x∗ ), i.e. xi∗ ∈ gi (x∗ ) for all i ∈ I. If x∗ ∈ Ui for some i ∈ I, then xi∗ = fi (x∗ ) ∈ coPi (x∗ ) ∩ Ai (x∗ ) ⊂ coPi (x∗ ) which contradicts A5. Hence, for all i ∈ I, x∗ ∈ / Ui , i.e.,

19 Let T and Y be any topological spaces, and : T → 2Y and : T → 2Y be correspondences. Then Lemma 4.2 of Yannelis (1987) shows that if has an 1 2 1 open graph and 2 is lower semi-continuous, the correspondence : T → 2Y defined by (t) = 1 (t) ∩ 2 (t) is lower semi-continuous. 20 X is metrizable because it is a countable product of metric spaces. It is also well-known (Stone’s Theorem) that metrizable spaces are paracompact.

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1277

xi∗ ∈ clAi (x∗ ) and coPi (x∗ ) ∩ Ai (x∗ ) = ∅, which implies Pi (x∗ ) ∩ Ai (x∗ ) = ∅ for each i ∈ I. Thus, we can conclude that Pi (x∗ ) ∩ Ai (x∗ ) = ∅, i.e., x∗ is a quasi-equilibrium for .  If each Pi has open upper sections, Pi (x∗ ) ∩ Ai (x∗ ) = ∅ implies that Pi (x∗ ) ∩ clAi (x∗ ) = ∅. Thus, the following corollary is immediate from Theorem A.1. Corollary A.1.

Suppose that satisfies A1–A5 of Theorem A.1.If each Pi has open upper sections, then has an equilibrium.

Remark A.1. Corollary A.1 does not follow from Borglin and Keiding (1976); Shafer and Sonnenschein (1975) or Yannelis and Prabhakar (1983) because the assumptions on the correspondences Pi are weaker here than those papers. In particular, Yannelis and Prabhakar (1983) assume that Pi must have open lower sections which implies that Pi is lower semi-continuous but the reverse is not true. Shafer and Sonnenschein (1975) and Borglin and Keiding (1976) assume that preference correspondences must have an open graph which implies that both sections (upper and lower) must be open. Remark A.2. The assumptions on the constraint correspondences in Theorem A.1, however, are slightly stronger than those of Shafer and Sonnenschein (1975). Nonetheless, they are automatically fulfilled by the standard exchange economy, and as we will see in the next section Theorem A.1 will enable us to provide a more general Walrasian equilibrium existence result than that of Shafer (1976). The condition A5 can be replaced by a more tractable condition without losing any generality. Since Pi is lower semicontinuous, by Proposition 2.6 in Michael (1956), the convex hull correspondence coPi is lower semi-continuous. Now we consider the economy ˜ = (Xi , Ai , P˜ i )i ∈ I where P˜ i is defined as follows: for all xi ∈ Xi , P˜ i (x) = {(1 − ˛)xi + ˛xi : 0 < ˛ ≤ 1, xi ∈ coPi (x)}. For each x ∈ X, P˜ i (x) is convex. Clearly, Pi (x) ⊂ coPi (x) ⊂ P˜ i (x), and xi is in the boundary of P˜ i (x) for all x ∈ X and all i ∈ I. By the lower semi-continuity of coPi , P˜ i is lower semi-continuous (for details, see Gale and Mas-Colell (1975, 1979) or Allouch (2002)). For each i ∈ I, we consider the following condition. 

A5 . For all x ∈ X with Pi (x) = / ∅, Pi (x) is convex, xi ∈ / Pi (x), and for each yi ∈ Pi (x), (xi , yi ] is in Pi (x).21 

This condition is stronger than A5 but the following result shows that instead of A5, A5 can be used together with the other assumptions to prove Theorem A.1. Proposition A.1. If satisfies the assumptions A1–A5, then ˜ satisfies the assumptions A1–A4 and A5 . Moreover, if x ∈ X is an equilibrium (a quasi-equilibrium) of ˜ , it is also an equilibrium (a quasi-equilibrium, resp.) of . Proof. Suppose that satisfies the assumptions A1–A5. As discussed above, P˜ i is lower semi-continuous and convex-valued. / ∅. By A5 we immediately see xi ∈ / P˜ i (x) and by construction, xi is in the relative boundary of Let x be a point in X with Pi (x) =  P˜ i (x). Thus, ˜ satisfies A1–A4 and A5 . Suppose that x ∈ X is an equilibrium of ˜ . Then x ∈ clAi (x) and P˜ i (x) ∩ clAi (x) = ∅ for all i ∈ I. Since Pi (x) ⊂ P˜ i (x), we trivially see that x ∈ clAi (x) and Pi (x) ∩ clAi (x) = ∅ for each i. Thus x is also an equilibrium of . Similarly, we can show that if x ∈ X is a quasi-equilibrium of ˜ , then it is a quasi-equilibrium of .  Lemma A.1. Suppose that Pi satisfies B6. Let x be a point in H. Then for each zi ∈ P˜ i (x) and vi ∈ Xi , there exists  ∈ (0, 1) such that zi + (1 − )vi ∈ P˜ i (x). ˜ Then there exist ˛ ∈ (0, 1] and xi ∈ coPi (x) such that zi = (1 − ˛)xi + ˛xi . Let vi be a point in Let zi be a point in P(x).  ˜ = /(˛ − ˛ + ). It follows that Xi . Since x ∈ coPi (x), by B6 there exists  ∈ (0, 1) such that x + (1 − )vi ∈ coPi (x). We set 

Proof.

i

˜ ∈ (0, 1) and  ˜ i ˜ i + (1 − )v z ˜ ˜ i = [(1 − ˛)xi + ˛xi ] + (1 − )v

i

˜ ˜ ˛ 1− x + v ˜ − ˛) i 1 − (1 ˜ − ˛) i 1 − (1  ˜ − ˛)xi + [1 − (1 ˜ − ˛)][x + (1 − )vi ] ∈ P˜ i (x). = (1 i ˜ − ˛)xi + [1 − (1 ˜ − ˛)] = (1



Appendix B. Proofs of the results of the main text Proof of Auxiliary Theorem 2.3.1. By Proposition A.1 and Lemma A.1, without loss of generality we may assume that E   satisfies B1–B3, B5 and instead of B4 and B6, the following conditions B4 and B6 .

21

For two vectors x and y in R , we denote by (x, y] the set {z ∈ R : z = (1 − )x + y for some  ∈ (0, 1]}.

1278

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283



B4 . For all x ∈ X with Pi (x) = / ∅, Pi (x) is convex, xi ∈ / Pi (x), and for each yi ∈ Pi (x), (xi , yi ] is in Pi (x).  B6 . Let x be a point in H. Then for each zi ∈ Pi (x) and vi ∈ Xi , there exists  ∈ (0, 1) such that zi + (1 − )vi ∈ Pi (x). For each i ∈ I, let XiH denote the projection of clH onto Xi . Since H is bounded, so is clH and therefore, XiH is bounded for all

i ∈ I. Hence, we can choose a closed and bounded ball K centered at the origin in R which contains XiH and ei in its interior for all i ∈ I. We introduce the truncated economy Eˆ = (Xˆ i , ei , Pˆ i ) where for all i ∈ I, Xˆ i = Xi ∩ K, Xˆ =



ˆ Xˆ I and Pˆ i (x) = Pi (x) ∩ K for all x ∈ X.

i∈I

We will break the proof into several steps. Step 1. We introduce the sets and 1 in R defined by = {p ∈ R : p ≤ 1} 1 = {p ∈ R : p = 1}. To apply Theorem A.1, we need to convert Eˆ into our abstract economy = (Xˆ i , Ai , Gi )i ∈ I  where I  = I ∪ {0} by adding the agent 0 as follows; if i = 0, we set Xˆ 0 = and define

 G0 (p, x) =

 

q∈ : q

 (xi − ei )

>p

 

i∈I

 (xi − ei )

,

i∈I

ˆ A0 (p, x) = for all (p, x) ∈ × X, and if i ∈ I, for all (p, x) ∈ × Xˆ we set Gi (p, x) = Pˆ i (x), and Ai (p, x) = {xi ∈ Xi : pxi < pei + 1 − p} ∩ K.

ˆ On the other hand, clAi : Since ei is in the interior of both Xi and K, Ai (p, x) is not empty for all (p, x) ∈ × X. × Xˆ → 2K has a closed graph and K is compact. These imply that the correspondence clAi is upper semi-continuous.   Thus, satisfies A1–A4 and A5 (B4 ) of Theorem A.1, and consequently, has a quasi-equilibrium, i.e., there exists ˆ (ˆp, xˆ ) ∈ × X such that (a) (b) (c)

pˆ ∈ clA0 (ˆp, xˆ ) = and G0 (ˆp, xˆ ) ∩ = ∅ and for all i ∈ I, xˆ i ∈ clAi (ˆp, xˆ ), i.e., pˆ xˆ i ≤ pˆ ei + 1 − ˆp, and Gi (ˆp, xˆ ) ∩ Ai (ˆp, xˆ ) = ∅, i.e., Pˆ i (ˆx) ∩ Ai (ˆp, xˆ ) = ∅.

We will show that (ˆp, xˆ ) is an equilibrium for the original exchange economy E. Step 2. We show that xˆ ∈ F, i.e., (ˆx − ei ) = 0. Since pˆ ∈ clA0 (ˆp, xˆ ) and G0 (ˆp, xˆ ) ∩ = ∅, we see that ˆp ≤ 1 and for all i∈I i q ∈ ,



 

 (ˆxi − ei )

≥q

 

i∈I



i∈I

(ˆxi − ei )

.

i∈I

Suppose that xˆ ∈ / F. We set q = (

 



 (ˆxi − ei )

≥ q

 i∈I

 i∈I



(ˆxi − ei ))/



(ˆxi − ei ) = 

i∈I

(ˆxi − ei ). It follows that q ∈ and therefore,

(ˆxi − ei ) > 0.

i∈I

In particular, this implies that ˆp = 1. On the other hand, xˆ i ∈ clAi (ˆp, xˆ ) for each i ∈ I implies that pˆ (ˆxi − ei ) ≤ 1 − ˆp. Summing up over i ∈ I, we obtain pˆ ( i ∈ I (ˆxi − ei )) ≤ I(1 − ˆp) and therefore, ˆp < 1, which is impossible. Therefore, xˆ is in F.

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1279

Step 3. We claim that xˆ ∈ R(e). Suppose otherwise, i.e., ei ∈ Pi (ˆx) for some i ∈ I. Since ei ∈ intK, ei is in Pˆ i (ˆx). If ˆp < 1, then ei is in Ai (ˆp, xˆ ). This implies that Pˆ i (ˆx) ∩ Ai (ˆp, xˆ ) = / ∅, which contradicts (c). We turn to the case that ˆp = 1. Since ei is in the  interior of Xˆ i and ˆp = 1, we can pick xi ∈ Xˆ i such that pˆ xi < pˆ ei . Recalling that ei ∈ Pi (ˆx), by B6 there exists  ∈ (0, 1) such that ei + (1 − )xi ∈ Pi (ˆx) and therefore, ei + (1 − )xi ∈ Pˆ i (ˆx). Then (c) implies that pˆ [ei + (1 − )xi ] ≥ pˆ ei or pˆ xi ≥ pˆ ei , which is impossible. Thus, we conclude that xˆ ∈ R(e). Step 4. The results of Steps 2 and 3 imply that xˆ ∈ H. We want to show that ˆp = 1 and pˆ xˆ i = pˆ ei for all i ∈ I. 

By B5, we have Pi (ˆx) = / ∅ for all i ∈ I. Let yi be a point in Pi (ˆx). Then by B4 , (ˆxi , yi ] is in Pi (ˆx). Since xˆ i is in the interior  / ∅ and thus, Pˆ i (ˆx) = / ∅. Then we can choose ti ∈ Pˆ i (ˆx). By B4 , ˛ti + (1 − ˛)ˆxi is in Pˆ i (ˆx) for any of K, we have K ∩ (ˆxi , yi ] = ˆ ˛ ∈ (0, 1]. By (c), we have Pi (ˆx) ∩ Ai (ˆp, xˆ ) = ∅, which implies that for all ˛ ∈ (0, 1], pˆ (˛ti + (1 − ˛)ˆxi ) ≥ pˆ ei + 1 − ˆp. By letting ˛ → 0, we have pˆ xˆ i ≥ pˆ ei + 1 − ˆp. On the other hand, (b) gives pˆ xˆ i ≤ pˆ ei + 1 − ˆp. Hence, for all i ∈ I, pˆ xˆ i = pˆ ei + 1 − ˆp. Summing it over I, we see that



pˆ (ˆxi − ei ) =

i∈I

Since

 i∈I



(1 − ˆp).

i∈I

(ˆxi − ei ) = 0, we obtain ˆp = 1. Moreover, we can conclude that pˆ xˆ i = pˆ ei for all i ∈ I.

ˆ Now we show that (ˆp, xˆ ) is Step 5. What is proved up to Step 4 is that (ˆp, xˆ ) is a quasi-equilibrium for the truncated economy E. an equilibrium of E by verifying that Pi (ˆx) ∩ Bi (ˆp) = ∅ for all i ∈ I. By Steps 3 and 4, xˆ is in clH. Since XiH ⊂ intK, it implies / ∅ for some i ∈ I. Let zi be a point in Pi (ˆx) ∩ ˇi (ˆp). Then we can choose that xˆ i ∈ intK for all i ∈ I. Suppose that Pi (ˆx) ∩ ˇi (ˆp) =  ˛ ∈ (0, 1) such that ˛ xˆ i + (1 − ˛ )zi ∈ K. It follows from B4 that ˛ xˆ i + (1 − ˛ )zi ∈ Pˆ i (ˆx). Recalling that ˆp = 1, pˆ xˆ i = pˆ ei   ˆ ˆ ˆ ˆ ˆ ˆ and pzi < pei , we have ˛ pxi + (1 − ˛ )ˆpzi < pei = pei + 1 − ˆp, and therefore, ˛ xˆ i + (1 − ˛ )zi ∈ Pˆ i (ˆx) ∩ Ai (ˆp, xˆ ), which contradicts (c). Thus, we have Pi (ˆx) ∩ ˇi (ˆp) = ∅ for each i ∈ I. We claim that pˆ xi > pˆ ei for all xi ∈ Pi (ˆx). Suppose that there exists xi ∈ Pi (ˆx) such that pˆ xi = pˆ ei . Since ei ∈ intXi , 

we can pick vi ∈ Xi such that pˆ vi < pˆ ei . By B6 , there exists  ∈ (0, 1) such that xi + (1 − )vi ∈ Pi (ˆx). On the other hand, we have pˆ [xi + (1 − )vi ] < pˆ ei , and therefore, xi + (1 − )vi ∈ ˇi (ˆp). This contradicts the fact that Pi (ˆx) ∩ ˇi (ˆp) = ∅. We conclude that Pi (ˆx) ∩ Bi (ˆp) = ∅ for all i ∈ I and therefore, (ˆp, xˆ ) ∈ 1 × X is an equilibrium for E. 

Proof of Lemma 3.3.1. For a point x ∈ H, suppose that G(x) = / R . Since G(x) is a convex cone, by the separating hyperplane  theorem there exists a nonzero p ∈ R such that for all z ∈ G(x), 0 ≤ pz. Recalling that xi ∈ clPi (x) for all i ∈ I, we have xi − ei ∈ cl[Pi (x) − {ei }] and therefore, xi − ei ∈ clcon[Pi (x) − {ei }] for all i ∈ I. Since x ∈ H, it follows that clcon[Pi (x) − {ei }] − {xi − ei } = clcon[Pi (x) − {ei }] +



xj − ej



⊂ G(x).

j= / i

This implies that for all zi ∈ clcon[Pi (x) − {ei }], 0 ≤ p[−(xi − ei ) + zi ].

(8)

In particular, for any  > 0 we have −(xi − ei ) + (xi − ei ) ∈ G(x). This implies that for each i ∈ I, 0 ≤ p[−(xi − ei ) + (xi − ei )] = ( − 1)p(xi − ei ). If  > 1, then 0 ≤ p(xi − ei ), and if  < 1, then 0 ≥ p(xi − ei ). Thus, we have pxi = pei for all i ∈ I. Let zi ∈ clPi (x). Then there exist  ≥ 0 and zi ∈ clcon[Pi (x) − {ei }] such that zi = (zi − ei ). Since pxi = pei , it follows from (8) that 0 ≤ pzi = p(zi − ei ) or pei ≤ pzi . Suppose that G(x) = R . Then it is easy to see that 0 ≤ pz for all z ∈ G(x) implies p = 0, which is impossible. Suppose that B7a holds for agent h. Let {xn } be a sequence in H where r



(xn ) increases to infinity. Then

Proof of Lemma 3.3.2. h there exist a subsequence {xnk } and a sequence {ynk } convergent to a point y ∈ clH which satisfy (1) and (2). Since rh (xnk ) → ∞

1280

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

and {ynk } is bounded, we have Pi (ynk ) ∩ C(0, rh (xnk )) = / ∅ and by (2), Pi (xnk ) ∩ C(0, rh (xnk )) = / ∅ for all i = / h and for sufficiently / ∅ for sufficiently large nk . large nk . Thus, we have Gh (xnk ) = It follows from (1) and (2) that for sufficiently large nk , Ph (ynk ) − {xhnk }



con [Ph (xnk ) − {eh }] − {xhnk − eh }

=

con [Ph (xnk ) − {eh }] +



Gh (xnk ),



xink − ei



i= / h

and for each j = / h, Pj (ynk ) ∩ C(0, rh (xnk )) − {xjnk }



con[Pj (xnk ) ∩ C(0, rh (xnk )) − {ej }] − {xjnk − ej }

=

con[Pj (xnk ) ∩ C(0, rh (xnk ))] +



Gh (xnk ).

Therefore, we conclude that B7a implies B7.



xink − ei



i= / j

 

Proof of Theorem 3.3.1. As mentioned earlier in this section, without loss of generality we can assume that E satisfies B4 instead of B4. (For details, we refer the reader to the proof of the Auxiliary Theorem 2.3.1) Then Pi is convex-valued, and for all x ∈ X, xi ∈ clPi (x). Let {K n } denote a sequence of increasing closed balls centered at the origin in R such that n R  ⊂ ∪∞ n=1 K .

We can take K 1 to be a sufficiently large ball that ei is contained in the interior of K 1 for all i ∈ I. Since K n is increasing, all ei ’s are contained in the interior of K n for all n. Let h be an agent which satisfies Assumption B7. Since the following arguments do not rely on the choice of h, without loss of generality, we can assume that h = 1. For each n, we set X1n = X1 ,

Xjn = Xj ∩ K n

for all j = / 1,

and

Xn =



Xin .

i∈I

Let

x ∈ Xn

and p ∈ . For all n and i ∈ I, we set

Pin (x) = Pi (x) ∩ Xin , ˇin (p) = {xi ∈ Xin : pxi

< pei }.

Note that P1n (x) = P1 (x) and ˇ1n (p) = ˇ1 (p) for all n. For each n, let E n = {(Xin , ei , Pin ) : i ∈ I} denote the truncated economy of E. Remarks B.1. For each n, the first agent’s choices are not restricted in any E n while the other agents’ choices are restricted to K n in E n for all n. Since Xin ’s are bounded except for the first agent, it is easy to check that H ∩ X n is bounded for all n. Step 1. Since E n satisfies B1–B4 and H ∩ X n is bounded for all n, by the same arguments made in Steps 1–3 of the proof of the Auxiliary Theorem 2.3.1, there exists a pair (pn , xn ) ∈ × X n for each n such that xn ∈ H, and for all i ∈ I, (a) pn xin ≤ pn ei + 1 − pn  and (b) Pin (xn ) ∩ {xi ∈ Xin : pn xi < pn ei + 1 − pn } = ∅. For each i = / 1, xin ∈ K n and therefore, C(0, r1 (xn )) ⊂ K n . First, we consider the case that there exists n such that xin 

is in the interior of K n for each i = / 1, or C(0, r1 (xn )) is in the interior of K n . Then by B4 , Pin (xn ) = Pi (xn ) ∩ K n is not / ∅ for all i ∈ I. Thus, empty for each i = / 1. Recalling that the choices of agent 1 are not restricted, we have Pin (xn ) = by applying the arguments of Steps 1–4 of the proof of the Auxiliary Theorem 2.3.1 to the truncated economy E n , we see that (pn , xn ) is an equilibrium for E n . Since P1 (xn ) = P1n (xn ) and xin is in the interior of K n for each i = / 1, it follows by applying the arguments of Step 5 of the proof of the Auxiliary Theorem 2.3.1 for each i = / 1 that (pn , xn ) is an equilibrium for E, and in this case, we are done. Thus, we only need to examine the case that for each n there exists some in = / 1 such that xin is in the boundary n n n n n of K . In this case, we have C(0, r1 (x )) = K for all n. This implies that r1 (x ) increases to infinity and therefore, {xn } has no bounded subsequences, and for all n and i = / 1, Pin (xn ) = Pi (xn ) ∩ C(0, r1 (xn )).

(9)

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1281

By Assumption B7, there exist a subsequence {xnk } of {xn }, and a sequence {ynk } convergent to a point y ∈ clH such that for all nk , P1nk (ynk ) = P1 (ynk ) ⊂ G1 (xnk ) + {xhnk }

(10)

and for all i = / 1, Pink (ynk ) = Pi (ynk ) ∩ C(0, r1 (xnk )) ⊂ G1 (xnk ) + {xink }.

(11)

(xnk )

(xnk ) ∩ C(0, r

(xnk ))

In particular, (10) implies that G1 is not empty for all nk . Therefore, Pi = / ∅, and by (9), 1 Pink (xnk ) = / ∅ for all nk and for all i = / 1. Since P1nk (xnk ) = P1 (xnk ), P1nk (xnk ) is also not empty for all nk . Step 2. We show that (pnk , xnk ) is an equilibrium of the economy E nk (xnk ). First, we claim that xink is in the relative boundary 



/ 1, we choose a point yi in Pink (xnk ). By B4 , (xink , yi ] is in of Pink (xnk ). By B4 , this trivially holds for i = 1. For each i = Pi (xnk ). Since xink and yi are in K nk , we have (xink , yi ] ∈ Pink (xnk ) and therefore, xink is in the relative boundary of Pink (xnk ). 

For a given nk , we choose ti ∈ Pink (xnk ) for each i ∈ I. By B4 , ˛ti + (1 − ˛)xink is in Pink (xnk ) for any ˛ ∈ (0, 1]. It follows from (b) that for all ˛ ∈ (0, 1], pnk (˛ti + (1 − ˛)xink ) ≥ pnk ei + 1 − pnk . By letting ˛ → 0, we have pnk xink ≥ pnk ei + 1 − pnk . On the other hand, (a) gives pnk xink ≤ pnk ei + 1 − pnk . Hence, for all i ∈ I, pnk xink = pnk ei + 1 − pnk . Summing it over I, we see that



pnk (xink − ei ) =

i∈I



(1 − pnk ).

i∈I



Since (xnk − ei ) = 0, we obtain pnk  = 1. Moreover, we can conclude that pnk xink = pnk ei for all i ∈ I. i∈I i Thus, it follows from the conditions (a) and (b) that (pnk , xnk ) is an equilibrium of the economy E nk (xnk ), or (A) pnk  = 1, and pnk xink = pnk ei for all i ∈ I, (B) pnk e1 ≤ pnk z1 for all z1 ∈ P1 (xnk ), and (C) pnk ei ≤ pnk zi for all zi ∈ Pi (xnk ) ∩ C(0, r1 (xnk )) and each i ∈ I \ {1}. Since G1 (xnk ) has the form G1 (xnk ) = clcon[P1 (xnk ) − {e1 }] +



clcon[Pj (xnk ) ∩ C(0, r1 (xnk )) − {ej }],

j= / 1

the last two outcomes (B) and (C) lead to the following relation. pnk z ≥ 0

for all z ∈ G1 (xnk ).

(12)

Step 3. It follows from (10) and (12) that for all z1 ∈ P1 (ynk ), p(z1 − x1nk ) ≥ 0.

(13)

Since yink is in the interior of C(0, r1 (xnk )) for sufficiently large nk and for all i = / 1, Pink (ynk ) is not empty for sufficiently large nk and for all i = / 1. Thus, it follows from (11) and (12) that for all i = / 1 and for all zi ∈ Pi (ynk ) ∩ C(0, r1 (xnk )), p(zi − xink ) ≥ 0. yink

Since belongs to the relative boundary of Pi nk and for all i ∈ I, pnk (yink − xink ) ≥ 0.

(14) (ynk )

for all i ∈ I and nk , (13) and (14) imply that for sufficiently large

1282

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

Recalling that pnk xink = pnk ei , we see that pnk yink ≥ pnk ei for all nk and all i ∈ I. By passing to the limit, this implies that pyi ≥ pei for all i ∈ I. By summing up the inequalities over i ∈ I, we obtain p i ∈ I yi ≥ p i ∈ I ei . Since y ∈ F, this implies that for all i ∈ I, pyi = pei . Step 4. To complete the proof we must show that Pi (y) ∩ Bi (p) = ∅ for all i. First, we claim that Pi (y) ∩ ˇi (p) = ∅. Suppose otherwise. Then we can pick zi ∈ Pi (y) ∩ ˇi (p). By Lemma 4.2 of Yannelis (1987), the correspondence Pi ∩ ˇi defined by Pi (x) ∩ ˇi (p) for all (p, x) ∈ × X is lower semi-continuous.22 Thus, there exists zink ∈ Pi (ynk ) ∩ ˇi (pnk ) for each nk which converges to zi . In particular, it implies that pnk zink < pnk ei . Since z1nk ∈ P1 (ynk ) and zink ∈ Pi (ynk ) ∩ C(0, r1 (xnk )) for all i = / 1 and sufficiently large nk , by Step 3 we must have pnk ei = pnk xink ≤ pnk zink , which leads to a contradiction. Thus, we have Pi (y) ∩ ˇi (p) = ∅ for all i ∈ I. By the same arguments made in Step 5 of the proof of Theorem 2.3.1, we see that Pi (y) ∩ Bi (p) = ∅ for all i ∈ I. Step 5. By the results of Steps 1–5, we see that y ∈ F, p = 1, and yi ∈ Bi (p) and Pi (y) ∩ Bi (p) = ∅ for all i. Therefore we conclude that (p, y) is an equilibrium of E.  Proof of Theorem 3.3.1s. Let {Tn } denote the compactification procedure of the assumption B7s. For each n, we denote by En = {(Xin , ei , Pin ) : i ∈ I} the truncated economy defined such that Xin := Xi ∩ Tn

and

Pin (x) = Pi (x) ∩ Tn ,

∀x ∈ X n :=



Xin .

i∈I n

For a sufficiently large n, E satisfies B1–B4. The set clH ∩ TnI contains the set of individually rational and feasible allocations for En , and by (iii) of Definition 3.3.1, it is compact. By (iv) of Definition 3.3.1, Pin (x) = / ∅ for all x ∈ R(e) and thus, En satisfies B5. n n n n Then by Auxiliary Theorem 2.3.1, there exists an equilibrium (p , x ) of E with p ∈ 1 . In particular, {pn } has a subsequence convergent to a point p ∈ 1 . By Assumption B7s, there exist  > 0 and y ∈ clH such that







Pi (y) ∩ C(yi , ) − {ei } ⊂ Ls con Pi (xn ) ∩ Tn − {ei }

.

(15)

We claim that (p, y) is an equilibrium of the economy E. All we have to show is that if zi ∈ Pi (y), then pzi > pei for all i ∈ I. For each ˛ ∈ (0, 1), we set zi (˛) = ˛zi + (1 − ˛)yi . Then for each ˛ sufficiently close to 0, the point zi (˛) is in Pi (y) ∩ C(yi , ). It follows from (15) that there exist {zin (˛)} and n ≥ 0 such that n (zin (˛) − ei ) → zi (˛) − ei and zin (˛) ∈ Pi (xn ) ∩ Tn for each n. Since (pn , xn ) is an equilibrium of En , we have pn (zin (˛) − ei ) > 0 and by passing to the limit, pzi (˛) ≥ pei . By letting ˛ → 0, we obtain pyi ≥ pei . It follows from the market clearing condition that pyi = pei for each i ∈ I. Since zi (˛) = ˛zi + (1 − ˛)yi and pzi (˛) ≥ pei , this implies that pzi ≥ pei for each i ∈ I. By the same arguments made in Step 5 of the proof of Theorem 2.3.1, one can show that zi ∈ Pi (y) implies that pzi > pei for each i ∈ I.  References Allouch, N., 2002. An equilibrium existence result with short selling. Journal of Mathematical Economics 37, 81–94. Allouch, N., Le Van, C., 2008. Equilibrium with dividends and walras equilibrium with possibly satiated consumers. Journal of Mathematical Economics. 44, 907–918. Allouch, N., Le Van, C., Page, F.H., 2006. Arbitrage and equilibrium in unbounded exchange economies with satiation. Journal of Mathematical Economics 42, 661–676. Berg, J.E., Dickhaut, J.W., O’Brien, J.R., 1985. Preference reversal and arbitrage. Research in Experimental Economics 3, 31–72. Borglin, A., Keiding, H., 1976. Existence of equilibrium actions and of equilibrium: a note on the ‘new’ existence theorem. Journal of Mathematical Economics 3, 313–316. Boyle, P.P., Ma, C., 2006. Mean-Preserving-Spread Risk Aversion and the CAPM. Mimeo. Chichilnisky, G., 1995. Limited arbitrage is necessary and sufficient for the existence of a competitive equilibrium with or without short sales. Economic Theory 5, 79–108. Dana, R.A., Le Van, C., Magnien, F., 1999. On the different notion of arbitrage and existence of equilibrium. Journal of Economic Theory 87, 169–193. Debreu, G., 1959. Theory of Value. Wiley, New York.

22

For details, see footnote 19.

D.C. Won, N.C. Yannelis / Journal of Mathematical Economics 44 (2008) 1266–1283

1283

Gale, D., Mas-Colell, A., 1975. An equilibrium existence theorem for a general model without ordered preferences. Journal of Mathematical Economics 2, 9–15. Gale, D., Mas-Colell, A., 1979. Corrections to an equilibrium existence theorem for a general model without ordered preferences. Journal of Mathematical Economics 6, 297–298. Grether, D.M., Plott, C.R., 1979. Economic theory of choice and the preference reversal phenomenon. American Economic Review 69, 623–638. Hart, O.D., 1974. On the existence of equilibrium in a securities model. Journal of Economic Theory 9, 293–311. Hammond, P.J., 1983. Overlapping expectations and Hart’s condition for equilibrium in a securities model. Journal of Economic Theory 31, 170–175. Loomes, G.C.S., Sugden, R., 1991. Observing violations of transitivity by experimental methods. Econometrica 59, 425–439. Martins-da-Rocha, V.F., Monteiro, P.K., 2007. Unbounded Exchange Economies with Satiation: How Far Can We Go? Mimeo. Michael, E., 1956. Continuous selections I. Annals of Mathematics 63, 361–382. Monteiro, P.K., Page, F.H., Wooders Jr., M.H., 2000. Increasing cones, recession cones and global cones. Optimization 47, 211–234. Page Jr., F.H., 1987. On equilibrium in Hart’s securities exchange model. Journal of Economic Theory 41, 392–404. Page, F.H., Wooders Jr., M.H., 1996. A necessary and sufficient condition for compactness of individually rational and feasible outcomes and existence of an equilibrium. Economics Letters 52, 153–162. Page, F.H., Wooders, M.H., Monteiro Jr., P.K., 2000. Inconsequential arbitrage. Journal of Mathematical Economics 34, 439–469. Shafer, W.J., 1976. Equilibrium in economies without ordered preferences or free disposal. Journal of Mathematical Economics 3, 135–137. Shafer, W.J., Sonnenschein, H., 1975. Equilibrium in abstract economies without ordered preferences. Journal of Mathematical Economics 2, 345–348. Werner, J., 1987. Arbitrage and the existence of competitive equilibrium. Econometrica 55, 1403–1418. Won, D.C., 2001. Competitive equilibrium with non-concavifiable preferences. Seoul Journal of Economics 14 (2). Won, D.C., Hahn, G., Yannelis, N.C., 2008. Capital market equilibrium without riskless assets: heterogeneous expectations. Annals of Finance 5, 183–195. Won, D.C., Yannelis, N.C., 2002a. Equilibrium theory with unbounded consumption sets and non-ordered preferences. Part I. Non-satiation. Revised 2007. Won, D.C., Yannelis, N.C., 2002b. Equilibrium Theory with Unbounded Consumption Sets and Non-Ordered Preferences. Part II. Satiation. Revised 2007. Yannelis, N.C., 1987. Equilibria in noncooperative models of competition. Journal of Economic Theory 41, 96–111. Yannelis, N.C., Prabhakar, N.D., 1983. Existence of maximal elements and equilibria in linear topological spaces. Journal of Mathematical Economics 12, 233–245.