Verifying Payo Security in the Mixed Extension of Discontinuous Games

Transcription

1 Verifying Payo Security in the Mixed Extension of Discontinuous Games Blake A. Allison and Jason J. Lepore ;y October 25, 2014 Abstract We introduce the concept of disjoint payo matching which can be used to show that the mixed extension of a compact game is payo secure. By putting minor structure on the discontinuities, we need only check payo s at each strategy rather than in neighborhoods of each strategy pro le, placing minimal restriction on the payo s at points of discontinuity. The results are used to verify existence of equilibrium in a general model of Bertrand-Edgeworth oligopoly. Keywords: Discontinuous games, Nash equilibrium, Disjoint payo matching, Existence. Allison : Department of Economics, University of California, Irvine; Lepore : Department of Economics, Orfalea College of Business, California Polytechnic State University, San Luis Obispo. y We would like to thank Adib Bagh and Aric Shafran for their helpful comments. We would also like to thank the editor Xavier Vives and the two excellent referees for helping us improve the article. 1

2 2 1 Introduction We provide a new su cient condition for payo security that generalizes the set of games in which existence of mixed strategy equilibrium can be readily veri ed. eny (1999) introduced the class of better reply secure games and showed existence of pure strategy Nash equilibrium in such games. 1 In addition, eny provided two conditions that together are su cient for a game to be better reply secure: payo security and reciprocal upper semicontinuity. 2 The results apply to both the normal form of a game and its mixed extension. In the context of mixed strategies, the su cient conditions of eny are di cult to verify since they are based on the mixed extension of a game. 3 Our new su cient condition is relatively straightforward to verify for a large class of games in which other su cient conditions have not been readily applicable. We introduce the concept of disjoint payo matching, which imposes minor structure on the discontinuities of the game instead of solely on the payo s. A game satis es disjoint payo matching if, given any strategy of a player, that player possesses a sequence of deviations that are at least as good in the limit and whose discontinuity sets are su ciently disjoint. Su ciently disjoint in this context means that there is no strategy pro le by the other players that constitutes a discontinuity for a subsequence of these deviations. 4 The advantage that disjoint payo matching has over other su cient conditions for existence of mixed strategy equilibrium is that it is easily veri able, owing to the fact that other conditions appeal to arbitrary probability measures or neighborhoods of strategies. Disjoint payo matching can replace payo security of the mixed extension of the game in eny s or Bagh and Jofre s (2006) theorems that additionally require (weak) reciprocal upper semicontinuity to guarantee better reply security and thus existence of equilibrium in mixed strategies. 5 Most closely related to the concept of disjoint payo matching is a su cient condition introduced in Bagh (2010). He introduces the notion of variational convergence of nite approximations of games. A result of this analysis is a su cient condition for existence that involves computation of limits of mixed strategies of the nite approximations. To alleviate the di culty of this computation, he establishes a stronger su cient condition on the set of all mixed strategies of the game. Disjoint payo matching places less restriction on the 1 A game is better reply secure if for every nonequilibrium strategy pro le x and every limiting payo vector u at x, there is a player i that has a strategy that gives payo strictly higher than u i even when other players deviate slightly from x. 2 A game is payo secure if at any strategy pro le x, each player has a strategy that earns a payo close to that of x against slight deviations from x by the other players. oughly speaking, a game is reciprocally upper semicontinuous, if whenever some player s payo jumps down, some other player s payo jumps up. 3 ecent papers by Tian (2010), Nessah and Tian (2010), and McLennan et al. (2011), have worked to generalize the work of eny (1999) and made great pushes toward a better understanding of Nash equilibria. 4 We require that the limit superior of the discontinuity sets of the deviations be empty. 5 Bagh and Jofre (2006) show that reciprocal upper semicontinuity can be replaced with the weaker concept of weak reciprocal upper semicontinuity.

3 3 discontinuity sets along with more restriction on the payo s than does Bagh s condition, facilitating greater ease of veri cation. Other su cient conditions in the literature that have attempted to alleviate the burden of computations of payo s in the mixed extensions of games are uniform payo security due to Monteiro and Page (2007) and uniform diagonal security due to Prokopovych and Yannelis (2012). Uniform payo security, a condition on the set of pure strategies, is a su cient condition for payo security of the mixed extension of a compact game. Uniform diagonal security is similarly a condition on the set of pure strategies which under certain conditions is a generalization of uniform payo security, but with the advantage of being a su cient condition for existence of equilibrium rather than just payo security of the mixed extension. 6 The rest of the paper is formatted as follows. In Section 2, we introduce the necessary preliminaries. Section 3 de nes disjoint payo matching and proves that it implies payo security for the mixed extension of the game. In Section 4, we use our results to show existence of equilibrium for a Bertrand-Edgeworth price setting oligopoly and we use an example from Sion and Wolfe (1957) to demonstrate how an equilibrium may not exist when disjoint payo matching does not hold. 2 Preliminaries An N-player compact game is a 2N-tuple G = (X i ; u i ) N i=1, where the strategy space of each player i is a compact Hausdor space X i and the payo of each player u i : X 1 :::X N 7! is bounded and measurable. The mixed extension of the game is G = (M i ; U i ) N i=1 where the strategy space of each player i is M i, the set of regular probability measures on X i, which is compact and convex, and the payo function of player i is U i = u i d, 2 M = Q N i=1 M i. Note that, as de ned, G and G are also the graphs of the games. The closure of the graph is denoted clg. Finally, the frontier of G, denoted FrG, is de ned to be the elements of the closure that are not in the graph, that is FrG =clg r G. The closure and frontier of the mixed extension are de ned analogously. Our condition and proofs will make reference to the sets of discontinuities of each player. Speci cally, we reference the points at which a player s payo is discontinuous in the other players strategies. These are given by the discontinuity map D i : X i 7! P (X i ), where P (X i ) is the power set of X i, de ned for all x i 2 X i as D i (x i ) = fx i 2 X i : u i is discontinuous in x i at (x i ; x i )g : 6 Prokopovych and Yannelis (2012) also adapt the concept of hospitality from Duggan (2007) to the domain of nonzero sum games. This condition involves the veri cation of deviations to a speci c subset of the set of mixed strategies.

4 4 3 Disjoint payo matching and payo security We now introduce disjoint payo matching. The condition has two parts: the rst is that any player can deviate from any strategy and remain almost as well o, while the second imposes that the discontinuity sets of each deviation have limited intersection. The name disjoint payo matching for our condition follows from the existence of a sequence of strategies which match the payo of the original strategy and for which the discontinuity sets are su ciently disjoint. De nition 1 The game G satis es disjoint payo matching if for all x i 2 X i, there exists a sequence of deviations x k i X i such that the following holds: (i) lim inf k u i x k i ; x i ui (x i ; x i ) for all x i 2 X i ; (ii) lim sup k D i x k i = ;. 7 emark 1 It follows that one need only check this condition for x i such that D i (x i ) is nonempty. That is, if u i (x i ; x i ) is continuous in x i at x i, then the conditions of disjoint payo matching are trivially satis ed by the constant sequence x k i = x i. The second condition is clearly satis ed when D i x k i \ Di xi l = ; for all k 6= l. This stronger empty intersection condition holds for the prominent examples in the literature. Unlike security concepts, there is no reference here to neighborhoods of the opponents strategies. Further, the payo s at the discontinuity points of the deviations are irrelevant since they are completely avoided in the limit, making the condition easy to verify and unrestrictive. It is worth noting that games in which discontinuities do not satisfy part (ii) of DPM often share best responses with a game that does satisfy DPM. That is, if u i (x) is player i s utility function in the game of interest and does not satisfy DPM, then there is often some strategically equivalent game for which player i s utility is of the form v i (x) = u i (x) + f (x i ), where v i satis es condition (ii) of DPM. We need one more de nition before we prove the main result. The following concept was introduced by eny (1999). De nition 2 The game G satis es payo security if for all " > 0 and all x 2 X, there exists for each player i a deviation x 0 i 2 X i and a neighborhood N (x i ) of x i such that u i (x 0 i; z) u i (x) " for all z 2 N (x i ). 7 Given a sequence of sets E n, the limit superior is lim sup n E n = T 1 N=1 S 1 n=n E n. This is equivalently all points x 2 X such that x 2 E n for in nitely many n.

5 5 The de nition is analogous for the mixed extension of the game. eny (1999) showed that payo security combined with another condition is su cient to guarantee the existence of a pure strategy Nash equilibrium. 8 Payo security is easily veri ed in the set of pure strategies, but is particularly di cult to verify in the mixed extension of a game. Our main result shows that disjoint payo matching implies that the mixed extension of the game is payo secure. Theorem 1 Let G be a compact game. Suppose that G satis es DPM, then G is payo secure. The advantage of DPM is that it is straightforward to verify and still fairly general. The condition in the following lemma is easier to use in the proof of our main result, but more di cult to verify directly. Lemma 1 Suppose that the compact game G satis es disjoint payo matching. Then for all " > 0, x i 2 X i, and i 2 M i there exists a deviation x 0 i 2 X i and a compact set K X i r D i (x 0 i) such that the following holds: (i) u i (x 0 i; x i ) > u i (x i ; x i ) " for all x i 2 K, (ii) i (X i r K) < ". 9 Proof of Lemma 1. Assume that G satis es disjoint payo matching and consider any player i, " > 0, and i 2 M i. Take x k i to be a defection sequence from the de nition of DPM. De ne the collection of sets E k = x i 2 X i : u i x k i ; x i > ui (x) ". Then notice that lim inf k E k = X i, so i (lim inf k E k ) = Further, lim sup k D i x k i = ;, so i lim sup k D i x k i = 0. By statement (5) in Section 9 of Halmos (1974), i (lim inf k E k ) lim inf k i (E k ) and i lim sup k D i x k i lim supk i D i x k i, and so limk i (E k ) = 1 and lim k i D i x k i = 0. It follows that there exists a k such that i (E k ) > 1 ("=3) and i D i x k i < "=3. Choose such a k and by regularity of i, we may choose a closed (and thus compact) subset K E k rd i x k i such that i (K) > i E k r D i x k i ("=3). It follows that i (X i r K) < ". Now we proceed to the proof of Theorem 1 which is based on showing that the condition in Lemma 1 implies the mixed extension is payo secure. 8 Payo security along with reciprocal upper semicontinuity together imply that a game is better reply secure, which in turn guarantees existence of equilibrium in a compact, quasiconcave game. 9 These conditions are equivalent to a slight weakening of disjoint payo matching: for all players i and all x i 2 X i and i 2 M i, there exists a sequence x k i X i such that (i) lim inf k u i x k i ; x i ui (x i ; x i ) i -almost everywhere, and (ii) lim sup k D i x k i is i -measure zero. This de nition seems less useful due to its dependence on an arbitrary probability measure. 10 The limit inferior of a sequence of sets E n is lim inf n E n = S 1 N=1 T 1 n=n E n. This is equivalently the set of points are are in E n for all but nitely many n.

6 6 Proof of Theorem 1. Let " > 0 and suppose that 2 M. Note that for each player i there exists some strategy x i in the support of i such that (1) ui (x i ; x i ) d i u i (x) d: From disjoint payo matching and Lemma 1, there exists a deviation x 0 i and a set K (") X i r D i (x 0 i) such that u i (x 0 i; x i ) > u i (x i ; x i ) and i (X i r K (")) < " 6M ; where M sup ju i j. It follows that (2) Further, we have that (3) (4) De ne > K(") u i (x 0 i; x i ) d i > X i rk(") X i rk(") Combining (2) and (3) yields K(") u i (x 0 i; x i ) d i " 6 for all x i 2 K (") u i (x i ; x i ) d i " 6 : X i rk(") u i (x i ; x i ) d i (ju i (x i ; x i )j + ju i (x 0 i; x i )j) d i > 2 sup ju i j (X i r K (")) 2" > 6 : ui (x 0 i; x i ) d i > u i (x i ; x i ) d i " 2 : u i (x i ) = sup inf u i V 3x i x 0 i 2V x 0 i; x 0 i ; where the supremum is taken over all neighborhoods V of x i. As noted by eny (1999) in the proof of Theorem 3.1, u i (x i ) is lower semicontinuous. From eny s proof of Proposition 5.1, it follows that u i (x i ) d i is lower semicontinuous in i. This property implies the existence of a neighborhood N i such that for all 2 N i, (5) ui (x i ) d > u i (x i ) d i " 6 :

7 7 Since M bounds u i as well as u i, we have that X i rk(") (u i (x i ) u i (x 0 i; x i )) d i X i rk(") > 2M (X i K (")) 2" = 6 : (ju i (x i )j + ju i (x 0 i; x i )j) d i Further, since u i (x 0 i; x i ) is continuous in x i at all x i 2 K ("), then u i (x i ) = u i (x 0 i; x i ) on K ("). 11 Therefore, (6) ui (x i ) d i = u i (x 0 i; x i ) d i + > u i (x 0 i; x i ) d i 2" 6 : X i rk(") (u i (x i ) u i (x 0 i; x i )) d i Using the fact that u i (x 0 i; x i ) u i (x i ) and combining (5) and (6), we have that for all 2 N i, u i (x 0 i; x i ) d u i (x i ) d > " u i (x i ) d i 6 (7) > u i (x 0 " i; x i ) d i 2 : Lastly, we combine (1), (4), and (7) and nd that for all 2 N i, ui (x 0 i; x i ) d > u i (x 0 i; x i ) d i " 2 > u i (x i ; x i ) d i " u i d ": Therefore, the mixed extension G is payo secure. 4 Examples In the rst part of this section, we use Theorem 1 to prove existence of mixed strategy equilibrium for a Bertrand-Edgeworth price-setting oligopoly with general speci cations of costs, residual demand rationing, and tie breaking rules. In the second part of this section, 11 The continuity here is with respect to the topology on X i, not to be confused with the subspace topology on K ("). Otherwise, if the function were only continuous with respect to the subspace topology, it might be that u i > u i on the boundary of K (").

8 8 an example from Sion and Wolfe (1957) is used to demonstrate that equilibrium may not exist if disjoint payo matching does not hold. 4.1 Bertrand-Edgeworth oligopoly A Bertrand-Edgeworth (BE) price-setting oligopoly is a competition between producers of homogenous products where prices are the only strategic variables. We apply disjoint payo matching to a BE oligopoly speci cation that subsumes much of the large literature and o ers a basis to generalize the analysis in these games. 12 Existence of equilibrium in such games has been examined by Dixon (1984), Allen and Hellwig (1986), Dasgupta and Maskin (1986a&b), Maskin (1986), Deneckere and Kovenock (1996), and Bagh (2010). With the exception of Allen and Hellwig, which studies a symmetric oligopoly with constant marginal cost, these papers only demonstrate existence for a BE duopoly. 13 Most of these results rely upon Dasgupta and Maskin (1986a) to guarantee existence, while Deneckere and Kovenock construct an equilibrium and Bagh (2010) develops and applies the concept of variational convergence to show existence. In addition to extending existence results to an oligopoly setting, our formulation greatly generalizes the set of rationing rules which are permitted. 14 Consider a homogeneous product industry with a set of rms N, with jnj = n. All rms simultaneously announce prices, then production decisions are made after demand is realized. Each rm i has a continuous, nondecreasing cost of production c i with c i (0) = The market demand F : 7! is continuous and nonincreasing in x with F (0) > 0. Further, assume that there exists a x > 0 such that F (x) = 0 for all x x. Note that any price x > x is weakly dominated by x 0 = x, so we may restrict the strategy space to X = [0; x] n. We denote by p i the price of any rm i and by p the vector of all rms prices. Each rm i has a capacity k i, which serves as an upper bound on the quantity that it can produce. Thus, the production problem faced by the rm at a price p i is max i (p i ; z) = p i z c i (z). z2[0;k i ] We refer to the solution to this problem as s i (p i ). 16 We assume that s i (p i ) is a continuous 12 The literature on BE games includes: Kreps and Scheinkman (1983), Davidson and Deneckere (1986), Osborne and Pitchick (1986), Deneckere and Kovenock (1992), Allen and Hellwig (1993), Deneckere and Kovenock (1996), Allen et al.(2000), Boccard and Wauthy (2000) and Lepore (2009)). 13 The result of Allen and Hellwig has been used to study BE oligopoly in other settings. Vives (1986) studies the an BE oligopoly as the number of rms gets large with e cient rationing and constant marginal cost up to capacity. Two recent articles Hirata (2009) and De Francesco and Salvadori (2010) characterize equilibria of a BE triopoly with e cient rationing and constant marginal cost up to capacity. 14 Most of the literature focuses on either e cient or proportional rationing. Maskin (1986) and Bagh (2010) consider a larger class of rationing rules, although many reasonable rules are excluded from their frameworks. 15 It is well known that equilibrium may not exist if c i is discontinuous or c i (0) > The speci cation of s i (p i ) follows from Dixon (1984), Maskin (1986) and Bagh (2010).

9 9 nondecreasing function. Further, we assume that if q < q 0 < s i (p i ) and i (p i ; s i (p i )), then i (p i ; q 0 ) > i (p i ; q). This is necessarily true if c i is strictly convex. 17 The quantity s i (p i ) may be referred to as rm i s supply, the maximum quantity that it is willing to produce at any given price. Inherently, s i k i, so the supply functions account for the capacity constraints. For any price vector p, order the players so that p 1 p 2 ::: p n. The demand served by rm 1 is Q 1 = min ff (p 1 ) ; s 1 (p 1 )g. We make minimal assumptions as to which portion of demand is served by rm i, only that for all j > i there is a continuous function ij (p) which denotes the share of i s quantity that satiates j s demand. 18 In the event that multiple rms choose the same price, there are multiple ways to order the players such that prices are nondecreasing. In this case, some tie breaking rule is used to allocate the demand. Speci cally, serves to give each player some weighted average of the demand they would receive under each possible ordering of the prices. Most commonly in application gives a uniform weight to each possible ordering, however, this is not necessary. Let O be the collection of possible orderings of the prices. For each o 2 O, we let o denote the weight applied to the order o, o (i) the position of player i in the ordering o, and Q o i the quantity served by rm i as if the ordering under o were a strictly increasing order of prices. That is, Q o i = min n F (p i ) P o j<i Qo j ji (p) ; s i (p i ) ; where ji = 1 for all j such that p j = p i. 19 We require that P o2o o = 1, so that demand is always fully allocated, though may be any measurable function. 20 Let I denote the set of players that charge p i and J the set of players that charge a price strictly less than p i. The actual demand served by rm i is then given by Q i = min n F (p i ) P j2j Q j ji (p) P o2o P o o o(j)<o(i) Qo j; s i (p i ) : Thus, each rm i serves the minimum of its capacity, supply, and the demand left by the rms with lower prices than i. The purpose for this formulation is to allow any possible rationing between tied rms. When multiple rms are tied, this allows any order of satiation of supply, be it simultaneous, partially sequentially, or fully sequentially. 17 Notice that for symmetric constant marginal cost c 0, we can restrict the strategy space to X = [c; x] n and the supply functions satisfy our assumptions. 18 A simple way to understand the purpose of ij is to consider the case in which a continuum of consumers have unit demand. In this case, ij speci es the fraction of consumers served by rm i that have willingness to pay of at least p j. 19 One way to interpret Q o i is that o represents a strict preference order for consumers, whereby p i at rm i is strictly preferred to p j at rm j for all i < j. Thus, the demand Q o i re ects the notion that this preference induces consumers to shop at rm i before rm i Both and the functions may depend on the full vector of prices as well as the capacities. We suppress the capacity arguments for clarity. The quantities Q o i and Q i depend on capacities only through the supply functions,, and the functions.

10 10 This very general framework captures the notion that consumers shop rst at rms with lower prices. Consider two choice for the functions ij given by e ij (p) = 1 and p ij de ned iteratively as ( ) p ij = min Q i D (p i ) Pj<i Q j p ji (p); 1. The rationing rule under e ij is the well known e cient, or parallel rule, whereas the rule under p ij is the proportional rationing rule. The pro t of each rm i can then be written as u i (p) = p i Q i (p) c i (Q i (p)) : We now turn to establishing that this game satis es DPM. Proposition 1 The BE oligopoly game satis es disjoint payo matching. Proof of Proposition 1. For any rm i, u i (0; p i ) = 0 for all p i. Consequently, D i (0) = ;. Let p i > 0. Note that the set of discontinuities D i (p i ) is a subset of points where p j = p i for some i 6= j. Thus, if p i 6= p 0 i, then D i (p i ) \ D i (p 0 i) = ;. It follows that for any sequence p l i! p i with p l i < p l+1 i < p i for all l, condition (ii) of DPM is satis ed. Note that lim l Q i p l i; p i Qi (p) for all p i. Since s i is continuous, lim l Q i p l i; p i si (p i ). By de nition, u i is increasing in Q i (p) for Q i (p) s i (p i ), and since u i is continuous in Q i (p), it follows that Therefore, the game satis es DPM. lim l p i Q i p l i; p i = p i lim l Q i p l i; p i p i lim l Q i (p) c i (Q i (p)). c i Q i p l i; p i c i lim l Q i p l i; p i Since this game satis es DPM, we know from Theorem 1 that the mixed extension is payo secure. Now we establish that the game has a mixed strategy equilibrium by appealing to results from eny (1999) and Bagh and Jofre (2006). The following de nitions are necessary for our proof of existence. De nition 3 The game G is weakly reciprocal upper semicontinuous (WUSC) if for all (x ; u ) 2FrG, there exists for some player i with a deviation x i 2 X i such that u i x i ; x i > u i.

11 11 De nition 4 The game G is better reply secure if whenever x is not an equilibrium and (x ; u ) is in the closure of the graph of G, there exists for some player i a strategy x i and a neighborhood N (x i) of x i such that for all x i 2 N (x i), u i (x i ; x i ) > u i. The de nitions are analogous for the mixed extension of the game. eny (1999) showed that a compact game whose mixed extension is better reply secure possesses a Nash equilibrium. He also showed that payo security together with reciprocal upper semicontinuity implies that a game is better reply secure. Bagh and Jofre (2006) showed that the latter condition can be replaced with WUSC. To prove that the game has a mixed strategy equilibrium we only need to show that the mixed extension of the game is WUSC. Proposition 2 The BE oligopoly game has a mixed strategy equilibrium. Proof. See Appendix. 4.2 Nonexistence The following example is constructed by Sion and Wolfe (1957) as an example of a game without equilibrium. There are two players with strategy spaces X 1 = X 2 = [0; 1]. The game is zero-sum, with 8 >< u 1 (x 1 ; x 2 ) = >: 1 if x 1 > x 2 0 if x 1 = x 2 or x = x if x 1 < x 2 < x if x < x 2 2 It is easy to see that this game does not satisfy disjoint payo matching. The discontinuities occur at ties of the form x 1 = x 2 and x 1 + 1=2 = x 2. For true ties (x 1 = x 2 ), player 1 bene ts from deviating to points with x 0 1 > x 1, while at shifted ties (x 1 + 1=2 = x 2 ), player 1 bene ts from deviations to points with x 0 1 < x 1 or x 0 1 > x 1 + 1=2. If we consider x 1 = 1=2, then if x 2 = 1=2, a improvement requires x 1 > 1=2, while if x 2 = 1, then an improvement requires x 0 1 < 1=2. This tension where one discontinuity demands deviations upward while another demands deviations downward to improve is what causes DPM to fail. Indeed, given any sequence of deviations from x 1 = 1=2, each individual deviation must make player 1 discretely worse o at either x 2 = 1=2 or x 2 = 1. A lesson in general is that DPM tends to hold whenever players can always improve their payo s at all discontinuities by deviations in a single direction, as is the case with the Bertrand-Edgeworth game where rms can always lower their prices any be at least as well :

12 12 o, or in any contests, where players can increase their bids or e orts and be at least as well o. In the current example, di erent discontinuities require con icting deviations to improve, and so a single sequence of deviations cannot uniformly improve a player s positions. 5 Appendix Proof of Proposition 2. Since the strategy space X is compact and Hausdor, we need only show that the game satis es WUSC. We begin by de ning for each player i u i (p) = lim sup u i (x) x!p and u = (u 1 ; :::; u n ), noting that each u i is upper semicontinuous. Since s i (p) is continuous and Q i (p) s i (p) for all p, then lim sup x!p Q i (x) s i (p). Further, by assumption, for any Q such that Q i (p) < Q s i (p), i (p i ; Q) > i (p i ; Q i (p)). Thus, it follows that Note that u i (p) = p i lim sup x!p Thus, for any p 2 X with p i > 0; Q i (x) c i lim sup Q i (x). x!p lim Q i (x i ; p i ) = lim sup Q i (x) x i!p i x!p Q i (p). (8) lim x i!p i u i (x i ; p i ) = u i (p). Let ( ; u ) 2FrG and let l! be such that ud l! u. Note that u = lim ud l l lim sup l ud l. Since each u i is upper semicontinuous, then as eny (1999) shows in the proof of Proposition 5.1, lim sup l ud l ud : Thus, we have that u u ( ).

13 13 De ne Y i = p 2 X : Q i (p) < Q i (p) and Y = S i Y i. We consider two cases: (i) (Y ) = 0, and (ii) (Y ) > 0. (i) In this case, Q i (p) = Q i (p) -almost everywhere for all players i. Thus, as noted above, u i = u i -almost everywhere, so we conclude that u ( ) = u ( ) u. Since ( ; u ) =2 G, then it must be that u i ( ) > u i for some player i. It follows that i = i satis es the de nition of WUSC. (ii) We begin by showing that u i ( ) > u i for some player i. For any p 2 Y, at least two rms must charge the same positive price, and at least one such rm i must have Q i (p) < Q i (p) s i (p). Let I be the set of rms j with p j = p i and J the set of rms j with p j < p i. Note that (9) P j2i Q j (p) F (p i ) P j2j Q j ji (p) for any choice of. The inequality in (9) must hold with equality else there would be excess demand that rm i would be able to satiate. Let A (p) P n j=1 u j (p) and A (p) lim sup x!p A (x). We will show that A (p) < P n j=1 u j (p). Let x m! p be such that A (x m )! A (p), and for each j let e Q j (p) lim m Q j (x m ). Since P n j2i e Q j (p) = F (p i ) P n j2j e Q j ji (p), then there still exists at least one rm i 0 such that e Q i 0(p) < Q i 0 (p). By our assumption, i 0(p i 0; e Q i 0(p)) < i 0 p i 0; Q i 0 (p), so (10) P n j=1 u j (p) A (p) = P n j=1 j p i 0; Q i 0 (p) j (p i 0; e Q i 0(p)) i 0 p i 0; Q i 0 (p) i 0(p i 0; e Q i 0(p)) > 0. The inequality in (10) holds based on the facts that: (i) u j (p) lim m u j (x m ) for all players j, and (ii) based on (8), u j (p) = i 0 p i 0; Q i 0 (p) and lim m u j (x m ) = i 0(p i 0; Q e i 0(p)). Therefore for all p 2 Y, A (p) < P n i=1 u i (p).

14 14 Since A is upper semicontinuous, then P n i=1 u i = A (p) d l A (p) d < Pn i=1 u i (p). The nal line follows from the fact that (Y ) > 0. Let i be the player with u i ( ) < u i. Consider the deviation functions f m (p i ) = 1 1 p i. m We construct a sequence of measures that transfers any mass or density from each p i to f m (p i ). For each m and every measurable set E, de ne m i (E) = i (fm 1 (E)). By Theorem 39C in Halmos (1974), lim u i (p i ; p i ) d m i d i = u i (f m (p i ) ; p i ) d. m Since u i is bounded, there is a constant, integrable function which bounds u i, so the Lebesgue dominated convergence theorem states that lim u i (f m (p i ) ; p i ) d = lim u i (f m (p i ) ; p i ) d. m m As noted, for all p, lim m Q i (f m (p i ) ; p i ) = lim sup x!p Q i (x), so lim m u i (f m (p i ) ; p i ) = u i. It follows that lim u i d m i d i = u i d. m Therefore, for su ciently large m, m i satis es the de nition of WUSC. eferences [1] Allen, B.,. Deneckere, T. Faith, and D. Kovenock (2000) Capacity Precommitment as a Barrier to Entry: A Bertrand-Edgeworth Approach, Economic Theory, 15, [2] Allen. B., and M. Hellwig (1986) Bertrand-Edgeworth Oligopoly in Large Markets, eview of Economic Studies, 53, [3] Allen, B., and M. Hellwig (1993) Bertrand-Edgeworth Duopoly with Proportional esidual Demand, International Economic eview, 34,

15 15 [4] Bagh A. (2010) Variational convergence: Approximation and existence of equilibria in discontinuous games, Journal of Economic Theory, 145, [5] Bagh A., and A. Jofre (2006) eciprocal Upper Semicontinuity and Better eply Secure Games: a Comment, Econometrica, 74, [6] Boccard, N., and X. Wauthy (2000) Bertrand Competition and Cournot Outcomes: Further esults, Economics Letters, 68, [7] Dasgupta, P., and E. Maskin (1986) The Existence of Equilibrium in Discontinuous Economic Games, I: Theory, eview of Economic Studies, 53(1): [8] Dasgupta, P., and E. Maskin (1986) The Existence of Equilibrium in Discontinuous Economic Games, II: Applications, eview of Economic Studies, 53(1): [9] Davidson, C., and. Deneckere (1986) Long-un Competition in Capacity, Short- un Competition in Price, and the Cournot Model, AND Journal of Economics, 17, [10] De Francesco, M., and N. Salvadori (2010) Bertrand-Edgeworth Games under Oligopoly with a Complete Characterization for the Triopoly, Munich Personal epec Archive, MPA Paper No [11] Deneckere,., and D. Kovenock (1992) Price Leadership, eview of Economic Studies, 59, [12] Deneckere,., and D. Kovenock (1996) Bertrand-Edgeworth Duopoly with Unit Cost Asymmetry, Economic Theory, 8, [13] Dixon, H. (1984) Existence of Mixed Strategy Equilibria in a Price-Setting Oligopoly with Convex Costs, Economics Letters, 16, [14] Duggan, J. (2007) Equilibrium Existence for ero-sum Games and Spatial Models of Elections, Games and Economic Behavior, 60, [15] Halmos, P. (1974) Measure Theory, New York: Springer-Verlag. [16] Hirata, D., (2009) Asymmetric Bertrand-Edgeworth Oligopoly and Mergers, B.E. Journal of Theoretical Economics, 9, Article 22 (Topics). [17] Kreps, D., and J. Scheinkman (1983) Quantity Precommitment and Bertrand Competition Yield Cournot Outcomes, Bell Journal of Economics, 14, [18] Lepore, J., (2009) Consumer ationing and the Cournot Outcome, B.E. Journal of Theoretical Economics, 9, Article 28 (Topics).

16 16 [19] Maskin, E. (1986) The Existence of Equilibrium with Price-Setting Firms, American Economic eview Papers and Proceedings, 76(2): [20] McLennan, A., P. Monteiro, and. Tourkey (2011) Games with Discontinuous Payo s: a Strengthening of eny s Existence Theorem, Econometrica, 79(5), [21] Monteiro, P., and F. Page (2007) Uniform Payo Security and Nash Equilibrium in Compact Games, Journal of Economic Theory, 134, [22] Nessah., and G. Tian (2010) Existence of Equilibrium in Discontinuous Games, Mimeo, UL: [23] Osborne, M., and C. Pitchik (1986) Price Competition in a Capacity-Constrained Duopoly, Journal of Economic Theory, 38, [24] Prokopovych, P., and N. Yannelis (2012) On Uniform Conditions for the Existence of Mixed Strategy Equilibria, Discussion Papers 48, Kyiv School of Economics. [25] eny, P. (1999) On the Existence of Pure and Mixed Strategy Nash Equilibrium in Discontinuous Games, Econometrica, 67, [26] Sion, M., and P. Wolfe (1957) On a Game Without Value In: M. Dresher, A. W. Tucker, P. Wolfe (eds.) Contribution to the theory of Games III, Princeton University Press, Princeton. [27] Tian G. (2010) Existence of Equilibria in Games with Arbitrary Strategy Spaces and Preferences: A Full Characterization, Mimeo, UL: [28] Vives, X., (1986) ationing ules and Bertrand-Edgeworth Equilibria in Large Markets, Economics Letters, 21,