Recent Advances in Generalized Matching Theory

Transcription

1 Recent Advances in Generalized Matching Theory John William Hatfield Stanford Graduate School of Business Scott Duke Kominers Becker Friedman Institute, University of Chicago Matching Problems: Economics meets Mathematics Conference Becker Friedman Institute & Stevanovich Center, University of Chicago June 4, 2012 Hatfield & Kominers June 4,

2 Stable Marriage The Marriage Problem (Gale Shapley, 1962) Question In a society with a set of men M and a set of women W, how can we arrange marriages so that no agent wishes for a divorce? Hatfield & Kominers June 4,

3 Stable Marriage The Marriage Problem (Gale Shapley, 1962) Question In a society with a set of men M and a set of women W, how can we arrange marriages so that no agent wishes for a divorce? Assumptions 1 Bilateral relationships: only pairs (and possibly singles). 2 Two-sided: men only desire women; women only desire men. 3 Preferences are fully known. Hatfield & Kominers June 4,

4 Stable Marriage The Deferred Acceptance Algorithm Step 1 1 Each man proposes to his first-choice woman. 2 Each woman holds onto her most-preferred acceptable proposal (if any) and rejects all others. Hatfield & Kominers June 4,

5 Stable Marriage The Deferred Acceptance Algorithm Step 1 1 Each man proposes to his first-choice woman. 2 Each woman holds onto her most-preferred acceptable proposal (if any) and rejects all others. Step t 2 1 Each rejected man proposes to the his favorite woman who has not rejected him. 2 Each woman holds onto her most-preferred acceptable proposal (if any) and rejects all others. Hatfield & Kominers June 4,

6 Stable Marriage The Deferred Acceptance Algorithm Step 1 1 Each man proposes to his first-choice woman. 2 Each woman holds onto her most-preferred acceptable proposal (if any) and rejects all others. Step t 2 1 Each rejected man proposes to the his favorite woman who has not rejected him. 2 Each woman holds onto her most-preferred acceptable proposal (if any) and rejects all others. At termination, no agent wants a divorce! Hatfield & Kominers June 4,

7 Stable Marriage Stability Definition A matching µ is a one-to-one correspondence on M W such that µ(m) W {m} for each m M, µ(w) M {w} for each w W, and µ 2 (i) = i for all i M W. Definition A marriage matching µ is stable if no agent wants a divorce. Hatfield & Kominers June 4,

8 Stable Marriage Stability Definition A matching µ is a one-to-one correspondence on M W such that µ(m) W {m} for each m M, µ(w) M {w} for each w W, and µ 2 (i) = i for all i M W. Definition A marriage matching µ is stable if no agent wants a divorce: Individually Rational: All agents i find their matches µ(i) acceptable. Unblocked: There do not exist m, w such that both m w µ(w) and w m µ(m). Hatfield & Kominers June 4,

9 Stable Marriage Existence and Lattice Structure Theorem (Gale Shapley, 1962) A stable marriage matching exists. Hatfield & Kominers June 4,

10 Stable Marriage Existence and Lattice Structure Theorem (Gale Shapley, 1962) A stable marriage matching exists. Theorem (Conway, 1976) Given two stable matchings µ, ν, there is a stable match µ ν (µ ν) which every man likes weakly more (less) than µ and ν. Hatfield & Kominers June 4,

11 Stable Marriage Existence and Lattice Structure Theorem (Gale Shapley, 1962) A stable marriage matching exists. Theorem (Conway, 1976) Given two stable matchings µ, ν, there is a stable match µ ν (µ ν) which every man likes weakly more (less) than µ and ν. If all men (weakly) prefer stable match µ to stable match ν, then all women (weakly) prefer ν to µ. Hatfield & Kominers June 4,

12 Stable Marriage Existence and Lattice Structure Theorem (Gale Shapley, 1962) A stable marriage matching exists. Theorem (Conway, 1976) Given two stable matchings µ, ν, there is a stable match µ ν (µ ν) which every man likes weakly more (less) than µ and ν. If all men (weakly) prefer stable match µ to stable match ν, then all women (weakly) prefer ν to µ. The man- and woman-proposing deferred acceptance algorithms respectively find the man- and woman-optimal stable matches. Hatfield & Kominers June 4,

13 Stable Marriage Opposition of Interests: A Simple Example m1 : w 1 w 2 m2 : w 2 w 1 w1 : m 2 m 1 w2 : m 1 m 2 Hatfield & Kominers June 4,

14 Stable Marriage Opposition of Interests: A Simple Example m1 : w 1 w 2 m2 : w 2 w 1 w1 : m 2 m 1 w2 : m 1 m 2 man-optimal stable match Hatfield & Kominers June 4,

15 Stable Marriage Opposition of Interests: A Simple Example m1 : w 1 w 2 m2 : w 2 w 1 w1 : m 2 m 1 w2 : m 1 m 2 man-optimal stable match woman-optimal stable match Hatfield & Kominers June 4,

16 Stable Marriage Opposition of Interests: A Simple Example m1 : w 1 w 2 m2 : w 2 w 1 w1 : m 2 m 1 w2 : m 1 m 2 man-optimal stable match woman-optimal stable match This opposition of interests result also implies that there is no mechanism which is strategy-proof for both men and women. Hatfield & Kominers June 4,

17 Stable Marriage The Lone Wolf Theorem (McVitie Wilson, 1970) Theorem The set of matched men (women) is invariant across stable matches. Proof Hatfield & Kominers June 4,

18 Stable Marriage The Lone Wolf Theorem (McVitie Wilson, 1970) Theorem The set of matched men (women) is invariant across stable matches. Proof Let µ be the man-optimal stable match and µ be any stable match. Hatfield & Kominers June 4,

19 Stable Marriage The Lone Wolf Theorem (McVitie Wilson, 1970) Theorem The set of matched men (women) is invariant across stable matches. Proof Let µ be the man-optimal stable match and µ be any stable match. Every man weakly prefers µ; the number of married men under µ is weakly greater than under µ. Hatfield & Kominers June 4,

20 Stable Marriage The Lone Wolf Theorem (McVitie Wilson, 1970) Theorem The set of matched men (women) is invariant across stable matches. Proof Let µ be the man-optimal stable match and µ be any stable match. Every man weakly prefers µ; the number of married men under µ is weakly greater than under µ. Every woman weakly prefers µ; the number of married women under µ is weakly greater than under µ. Hatfield & Kominers June 4,

21 Applications Generalized Matching Theory Stable Marriage Hatfield & Kominers June 4,

22 Applications Generalized Matching Theory Stable Marriage National Residency Matching Program Men are the medical students and women are the hospitals. Hatfield & Kominers June 4,

23 Applications Generalized Matching Theory Stable Marriage National Residency Matching Program Men are the medical students and women are the hospitals. School choice Men are the students and women are the schools. Hatfield & Kominers June 4,

24 Applications Generalized Matching Theory Stable Marriage National Residency Matching Program Men are the medical students and women are the hospitals. School choice Men are the students and women are the schools. Labor markets Men are the workers and women are the firms. Hatfield & Kominers June 4,

25 Applications Generalized Matching Theory Stable Marriage National Residency Matching Program Men are the medical students and women are the hospitals. School choice Men are the students and women are the schools. Labor markets Men are the workers and women are the firms. Auctions Men are the bidders and the woman is the auctioneer. Hatfield & Kominers June 4,

26 Applications Generalized Matching Theory Stable Marriage National Residency Matching Program Men are the medical students and women are the hospitals. School choice Men are the students and women are the schools. Labor markets Men are the workers and women are the firms. Auctions Men are the bidders and the woman is the auctioneer. But in general these applications require that women take on multiple partners and that relationships take on many forms. Hatfield & Kominers June 4,

27 Matching with Contracts Matching with Contracts Hatfield & Kominers June 4,

28 Matching with Contracts Matching with Contracts A set of doctors D: each doctor has a strict preference order over contracts involving him, Hatfield & Kominers June 4,

29 Matching with Contracts Matching with Contracts A set of doctors D: each doctor has a strict preference order over contracts involving him, A set of hospitals H: each hospital has a strict preferences over subsets of contracts involving it, and Hatfield & Kominers June 4,

30 Matching with Contracts Matching with Contracts A set of doctors D: each doctor has a strict preference order over contracts involving him, A set of hospitals H: each hospital has a strict preferences over subsets of contracts involving it, and A set of contracts X D H T, where T is a finite set of terms such as wages, hours, etc. Hatfield & Kominers June 4,

31 Matching with Contracts Matching with Contracts A set of doctors D: each doctor has a strict preference order over contracts involving him, A set of hospitals H: each hospital has a strict preferences over subsets of contracts involving it, and A set of contracts X D H T, where T is a finite set of terms such as wages, hours, etc. x D identifies the doctor of contract x; Hatfield & Kominers June 4,

32 Matching with Contracts Matching with Contracts A set of doctors D: each doctor has a strict preference order over contracts involving him, A set of hospitals H: each hospital has a strict preferences over subsets of contracts involving it, and A set of contracts X D H T, where T is a finite set of terms such as wages, hours, etc. x D identifies the doctor of contract x; x H identifies the hospital of contract x. Hatfield & Kominers June 4,

33 Matching with Contracts Matching with Contracts A set of doctors D: each doctor has a strict preference order over contracts involving him, A set of hospitals H: each hospital has a strict preferences over subsets of contracts involving it, and A set of contracts X D H T, where T is a finite set of terms such as wages, hours, etc. x D identifies the doctor of contract x; x H identifies the hospital of contract x. An outcome is a set of contracts Y X such that if x, z Y and x D = z D, then x = z. Hatfield & Kominers June 4,

34 Choice Functions Generalized Matching Theory Matching with Contracts Hatfield & Kominers June 4,

35 Choice Functions Generalized Matching Theory Matching with Contracts C d (Y ) max P d {x Y : x D = d}. Hatfield & Kominers June 4,

36 Matching with Contracts Choice Functions C d (Y ) max P d {x Y : x D = d}. C h (Y ) max P h{z Y : Z H = {h}}. Hatfield & Kominers June 4,

37 Matching with Contracts Choice Functions C d (Y ) max P d {x Y : x D = d}. C h (Y ) max P h{z Y : Z H = {h}}. We define the rejection functions R D (Y ) Y d D C d (Y ), R H (Y ) Y h H C h (Y ). Hatfield & Kominers June 4,

38 Matching with Contracts Choice Functions C d (Y ) max P d {x Y : x D = d}. C h (Y ) max P h{z Y : Z H = {h}}. We define the rejection functions R D (Y ) Y d D C d (Y ), R H (Y ) Y h H C h (Y ). Definition The preferences of hospital h are substitutable if for all Y X, if z / C h (Y ), then z / C h ({x} Y ) for all x z. Hatfield & Kominers June 4,

39 Matching with Contracts Equilibrium Definition An outcome A is stable if it is 1 Individually rational: for all d D, if x A and x D = d, then x d, for all h H, C h (A) = A H. 2 Unblocked: There does not exist a nonempty blocking set Z X A and hospital h such that Z C h (A Z) and Z C D (A Z). Hatfield & Kominers June 4,

40 Matching with Contracts Equilibrium Definition An outcome A is stable if it is 1 Individually rational: for all d D, if x A and x D = d, then x d, for all h H, C h (A) = A H. 2 Unblocked: There does not exist a nonempty blocking set Z X A and hospital h such that Z C h (A Z) and Z C D (A Z). Stability is a price-theoretic notion: Every contract not taken is available to some agent who does not choose it. Hatfield & Kominers June 4,

41 Matching with Contracts Characterization of Stable Outcomes Consider the operator ( Φ ) ( H X D X R ) D X D ( Φ ) ( D X H X R ) H X H Φ ( X D, X H) = ( ( Φ ) ( D X H, Φ )) H X D Hatfield & Kominers June 4,

42 Matching with Contracts Characterization of Stable Outcomes Consider the operator ( Φ ) ( H X D X R ) D X D ( Φ ) ( D X H X R ) H X H Φ ( X D, X H) = ( ( Φ ) ( D X H, Φ )) H X D Theorem Suppose that the preferences of hospitals are substitutable. Then if Φ ( X D, X H) = ( X D, X H), the outcome X D X H is stable. Conversely, if A is a stable outcome, there exist X D, X H X such that Φ ( X D, X H) = ( X D, X H) and X D X H = A. Hatfield & Kominers June 4,

43 Existence of Stable Allocations Matching with Contracts Theorem Suppose that hospitals preferences are substitutable. Then there exists a nonempty finite lattice of fixed points ( X D, X H) of Φ which correspond to stable outcomes A = X D X H. Hatfield & Kominers June 4,

44 Existence of Stable Allocations Matching with Contracts Theorem Suppose that hospitals preferences are substitutable. Then there exists a nonempty finite lattice of fixed points ( X D, X H) of Φ which correspond to stable outcomes A = X D X H. The proof follows from the isotonicity of the operator Φ. Hatfield & Kominers June 4,

45 Matching with Contracts Existence of Stable Allocations Theorem Suppose that hospitals preferences are substitutable. Then there exists a nonempty finite lattice of fixed points ( X D, X H) of Φ which correspond to stable outcomes A = X D X H. The proof follows from the isotonicity of the operator Φ. The lattice result implies opposition of interests. Hatfield & Kominers June 4,

46 Matching with Contracts The Law of Aggregate Demand Definition The preferences of h H satisfy the Law of Aggregate Demand (LoAD) if for all Y Y X, C h (Y ) C h (Y ). Hatfield & Kominers June 4,

47 Matching with Contracts The Law of Aggregate Demand Definition The preferences of h H satisfy the Law of Aggregate Demand (LoAD) if for all Y Y X, C h (Y ) C h (Y ). Hatfield & Kominers June 4,

48 Matching with Contracts The Law of Aggregate Demand Definition The preferences of h H satisfy the Law of Aggregate Demand (LoAD) if for all Y Y X, C h (Y ) C h (Y ). Intuition: When h receives new offers, he hires at least as many doctors as he did before: no doctor can do the work of two. Hatfield & Kominers June 4,

49 Matching with Contracts The Rural Hospitals Theorem and Strategy-Proofness Theorem If all hospitals preferences are substitutable and satisfy the LoAD, then each doctor and hospital signs the same number of contracts at each stable outcome. Hatfield & Kominers June 4,

50 Matching with Contracts The Rural Hospitals Theorem and Strategy-Proofness Theorem If all hospitals preferences are substitutable and satisfy the LoAD, then each doctor and hospital signs the same number of contracts at each stable outcome. Theorem If all hospitals preferences are substitutable and satisfy the LoAD, the doctor-optimal stable many-to-one matching mechanism is (group) strategy-proof. Hatfield & Kominers June 4,

51 Matching Without Substitutes Matching with Contracts Hatfield & Kominers June 4,

52 Matching Without Substitutes Matching with Contracts Substitutability is sufficient, but is it necessary? Hatfield & Kominers June 4,

53 Matching Without Substitutes Matching with Contracts Substitutability is sufficient, but is it necessary? No: Hatfield and Kojima (2010) showed that a weaker condition, bilateral substitutability, is sufficient. Hatfield & Kominers June 4,

54 Matching Without Substitutes Matching with Contracts Substitutability is sufficient, but is it necessary? No: Hatfield and Kojima (2010) showed that a weaker condition, bilateral substitutability, is sufficient. In simple many-to-one matching, substitutability is necessary. Hatfield & Kominers June 4,

55 Matching with Contracts Matching Without Substitutes Substitutability is sufficient, but is it necessary? No: Hatfield and Kojima (2010) showed that a weaker condition, bilateral substitutability, is sufficient. In simple many-to-one matching, substitutability is necessary. This has important applications: Sönmez and Switzer (2011), Sönmez (2011) consider the matching of cadets to U.S. Army branches, where preferences are not substitutable, but are unilaterally substitutable. Hatfield & Kominers June 4,

56 Matching with Contracts Matching Without Substitutes Substitutability is sufficient, but is it necessary? No: Hatfield and Kojima (2010) showed that a weaker condition, bilateral substitutability, is sufficient. In simple many-to-one matching, substitutability is necessary. This has important applications: Sönmez and Switzer (2011), Sönmez (2011) consider the matching of cadets to U.S. Army branches, where preferences are not substitutable, but are unilaterally substitutable. Open question: What is the necessary and sufficient condition for matching with contracts? Hatfield & Kominers June 4,

57 Generalized Matching Theory Supply Chain Matching Supply Chain Matching (Ostrovsky, 2008) s i b 1 b 2 Same-side contracts are substitutes. Cross-side contracts are complements. Objects are fully substitutable. Hatfield & Kominers June 4,

58 Generalized Matching Theory Supply Chain Matching Supply Chain Matching (Ostrovsky, 2008) s i b 1 b 2 Same-side contracts are substitutes. Cross-side contracts are complements. Objects are fully substitutable. Theorem Stable outcomes exist. Hatfield & Kominers June 4,

59 Supply Chain Matching Full Substitutability is Essential (Hatfield Kominers, 2012) Although (full) substitutability is not necessary for many-to-one matching with contracts, it is necessary for supply chain matching, and many-to-many matching with contracts. Hatfield & Kominers June 4,

60 Supply Chain Matching Full Substitutability is Essential (Hatfield Kominers, 2012) Although (full) substitutability is not necessary for many-to-one matching with contracts, it is necessary for supply chain matching, and many-to-many matching with contracts. This poses a problem for couples matching. Hatfield & Kominers June 4,

61 Supply Chain Matching Full Substitutability is Essential (Hatfield Kominers, 2012) Although (full) substitutability is not necessary for many-to-one matching with contracts, it is necessary for supply chain matching, and many-to-many matching with contracts. This poses a problem for couples matching. But new large-market results may provide a partial solution: Kojima Pathak Roth (2011); Ashlagi Braverman Hassidim (2011); Azevedo Weyl White (2012); Azevedo Hatfield (in preparation). Hatfield & Kominers June 4,

62 Supply Chain Matching Cyclic Contract Sets g y f 1 x 1 f 2 x 2 P f 1 : {y, x 2 } {x 1, x 2 } P f 2 : {x 2, x 1 } P g : {y} Hatfield & Kominers June 4,

63 Supply Chain Matching Cyclic Contract Sets g y f 1 x 1 f 2 x 2 P f 1 : {y, x 2 } {x 1, x 2 } P f 2 : {x 2, x 1 } P g : {y} Theorem Acyclicity is necessary for stability. Hatfield & Kominers June 4,

64 Supply Chain Matching The Rural Hospitals Theorem Theorem (two-sided) In many-to-one (or -many) matching with contracts, if all preferences are substitutable and satisfy the LoAD, then each doctor and hospital signs the same number of contracts at each stable outcome. Hatfield & Kominers June 4,

65 Generalized Matching Theory The Rural Hospitals Theorem Supply Chain Matching Theorem (two-sided) In many-to-one (or -many) matching with contracts, if all preferences are substitutable and satisfy the LoAD, then each doctor and hospital signs the same number of contracts at each stable outcome. What happens in supply chains? s x i y b z P s : {x} {z} P i : {x, y} P b : {z} {x} Hatfield & Kominers June 4,

66 Supply Chain Matching The Rural Hospitals Theorem Theorem (two-sided) In many-to-one (or -many) matching with contracts, if all preferences are substitutable and satisfy the LoAD, then each doctor and hospital signs the same number of contracts at each stable outcome. Theorem (supply chain) Suppose that X is acyclic and that all preferences are fully substitutable and satisfy LoAD (and LoAS). Then, for each agent f F, the difference between the number of contracts the f buys and the number of contracts f sells is invariant across stable outcomes. Hatfield & Kominers June 4,

67 The Assignment Problem The Model (Koopmans Beckmann, 1957; Gale, 1960; Shapley Shubik, 1972) ζ m,w total surplus of marriage of man m and woman w Hatfield & Kominers June 4,

68 The Assignment Problem The Model (Koopmans Beckmann, 1957; Gale, 1960; Shapley Shubik, 1972) ζ m,w total surplus of marriage of man m and woman w assignment indicators: a m,w { 1 m, w married 0 otherwise Hatfield & Kominers June 4,

69 The Assignment Problem The Model (Koopmans Beckmann, 1957; Gale, 1960; Shapley Shubik, 1972) ζ m,w total surplus of marriage of man m and woman w assignment indicators: a m,w { 1 m, w married 0 otherwise Stable assignment (ã m,w ) solves the integer program max m w a m,w ζ m,w 0 w a m,w 1 m 0 m a m,w 1 w Hatfield & Kominers June 4,

70 Efficient Mating Generalized Matching Theory The Assignment Problem z m,w ζ m,w ζ m, ζ,w marital surplus max m ( a m,w ζ m,w = max a m,w z m,w + w m w m ζ m, + w ζ,w ) Theorem Stable assignment maximizes aggregate marriage output. Note Even with a m,w [0, 1], the optimum is always an integer solution. Hatfield & Kominers June 4,

71 Other Notes Generalized Matching Theory The Assignment Problem Dual problem shows us shadow prices which describe the social cost of removing an agent from the pool of singles. If ζ m,w = h(x m, y w ), then complementarity (substitution) in traits leads to positive (negative) assortative mating. (Becker, 1973) Matches stable in the presence of transfers need not be stable if transfers are not allowed, and vice versa. (Jaffe Kominers, tomorrow) Hatfield & Kominers June 4,

72 Stability and Competitive Equilibrium in Trading Networks Generalization to Networks Main Results In arbitrary trading networks with 1 bilateral contracts, 2 transferable utility, and 3 fully substitutable preferences, competitive equilibria exist and coincide with stable outcomes. Hatfield & Kominers June 4,

73 Stability and Competitive Equilibrium in Trading Networks Generalization to Networks Main Results In arbitrary trading networks with 1 bilateral contracts, 2 transferable utility, and 3 fully substitutable preferences, competitive equilibria exist and coincide with stable outcomes. Full substitutability is necessary for these results. Correspondence results extend to other solutions concepts. Hatfield & Kominers June 4,

74 Stability and Competitive Equilibrium in Trading Networks Cyclic Contract Sets g y f 1 x 1 f 2 x 2 P f 1 : {y, x 2 } {x 1, x 2 } P f 2 : {x 2, x 1 } P g : {y} Theorem Acyclicity is necessary for stability! Hatfield & Kominers June 4,

75 Related Literature Generalized Matching Theory Stability and Competitive Equilibrium in Trading Networks Matching: Kelso Crawford (1982): Many-to-one (with transfers); (GS) Ostrovsky (2008): Supply chain networks; (SSS) and (CSC) Hatfield Kominers (2012): Trading networks (sans transfers) Exchange economies with indivisibilities: Koopmans Beckmann (1957); Shapley Shubik (1972) Gul Stachetti (1999): (GS) Sun Yang (2006, 2009): (GSC) Hatfield & Kominers June 4,

76 The Setting: Trades and Contracts Stability and Competitive Equilibrium in Trading Networks Finite set of agents I Hatfield & Kominers June 4,

77 Stability and Competitive Equilibrium in Trading Networks The Setting: Trades and Contracts Finite set of agents I Finite set of bilateral trades Ω each trade ω Ω has a seller s(ω) I and a buyer b(ω) I An arrangement is a pair [Ψ; p], where Ψ Ω and p R Ω. Hatfield & Kominers June 4,

78 Stability and Competitive Equilibrium in Trading Networks The Setting: Trades and Contracts Finite set of agents I Finite set of bilateral trades Ω each trade ω Ω has a seller s(ω) I and a buyer b(ω) I An arrangement is a pair [Ψ; p], where Ψ Ω and p R Ω. Set of contracts X := Ω R each contract x X is a pair (ω, p ω ) τ(y ) Ω set of trades in contract set Y X A (feasible) outcome is a set of contracts A X which uniquely prices each trade in A. Hatfield & Kominers June 4,

79 The Setting: Demand Stability and Competitive Equilibrium in Trading Networks Each agent i has quasilinear utility over arrangements: U i ([Ψ; p]) = u i (Ψ i ) + p ψ p ψ. ψ Ψ i ψ Ψ i U i extends naturally to (feasible) outcomes. Hatfield & Kominers June 4,

80 The Setting: Demand Stability and Competitive Equilibrium in Trading Networks Each agent i has quasilinear utility over arrangements: U i ([Ψ; p]) = u i (Ψ i ) + p ψ p ψ. ψ Ψ i ψ Ψ i U i extends naturally to (feasible) outcomes. For any price vector p R Ω, the demand of i is D i (p) = argmax Ψ Ωi U i ([Ψ; p]). For any set of contracts Y X, the choice of i is C i (Y ) = argmax Z Yi U i (Z). Hatfield & Kominers June 4,

81 Assumptions on Preferences Stability and Competitive Equilibrium in Trading Networks 1 u i (Ψ) R { }. Hatfield & Kominers June 4,

82 Assumptions on Preferences Stability and Competitive Equilibrium in Trading Networks 1 u i (Ψ) R { }. 2 u i ( ) R. Hatfield & Kominers June 4,

83 Assumptions on Preferences Stability and Competitive Equilibrium in Trading Networks 1 u i (Ψ) R { }. 2 u i ( ) R. 3 Full substitutability... Hatfield & Kominers June 4,

84 Stability and Competitive Equilibrium in Trading Networks Full Substitutability (I) Definition The preferences of agent i are fully substitutable (in choice language) if 1 same-side contracts are substitutes for i, and 2 cross-side contracts are complements for i. Hatfield & Kominers June 4,

85 Full Substitutability (I) Stability and Competitive Equilibrium in Trading Networks Definition The preferences of agent i are fully substitutable (in choice language) if for all sets of contracts Y, Z X i such that C i (Z) = C i (Y ) = 1, 1 if Y i = Z i, and Y i Z i, then for Y C i (Y ) and Z C i (Z), we have (Y i Y i ) (Z i Z i ) and Yi Z i ; 2 if Y i = Z i, and Y i Z i, then for Y C i (Y ) and Z C i (Z), we have (Y i Yi ) (Z i Zi ) and Y i Z i. Hatfield & Kominers June 4,

86 Stability and Competitive Equilibrium in Trading Networks Full Substitutability (II) Theorem Choice-language full substitutability 1 has equivalents in demand and indicator languages; 2 holds if and only if the indirect utility function V i (p) := max Ψ Ω i U i ([Ψ; p]) is submodular (V i (p q) + V i (p q) V i (p) + V i (q)). Hatfield & Kominers June 4,

87 Stability and Competitive Equilibrium in Trading Networks Solution Concepts Definition An outcome A is stable if it is 1 Individually rational: for each i I, A i C i (A); 2 Unblocked: There is no nonempty, feasible Z X such that Z A = and for each i, and for each Y i C i (Z A), we have Z i Y i. Definition Arrangement [Ψ; p] is a competitive equilibrium (CE) if for each i, Ψ i D i (p). Hatfield & Kominers June 4,

88 Existence of Competitive Equilibria Stability and Competitive Equilibrium in Trading Networks Theorem If preferences are fully substitutable, then a CE exists. Proof 1 Modify: Transform potentially unbounded u i to û i. 2 Associate: Construct a two-sided one-to-many matching market: i firm : valuation ũ i (Ψ) := û i (Ψ i (Ω Ψ) i ); ω worker : wants high wages; p wage. 3 A CE exists in the associated market (Kelso Crawford, 1982). 4 CE associated CE modified = CE original. Hatfield & Kominers June 4,

89 Stability and Competitive Equilibrium in Trading Networks Structure of Competitive Equilibria Theorem (First Welfare Theorem) Let [Ψ; p] be a CE. Then Ψ is efficient. Hatfield & Kominers June 4,

90 Stability and Competitive Equilibrium in Trading Networks Structure of Competitive Equilibria Theorem (First Welfare Theorem) Let [Ψ; p] be a CE. Then Ψ is efficient. Theorem (Second Welfare Theorem) Suppose agents preferences are fully substitutable. Then, for any CE [Ξ; p] and efficient set of trades Ψ, [Ψ; p] is a CE. Hatfield & Kominers June 4,

91 Stability and Competitive Equilibrium in Trading Networks Structure of Competitive Equilibria Theorem (First Welfare Theorem) Let [Ψ; p] be a CE. Then Ψ is efficient. Theorem (Second Welfare Theorem) Suppose agents preferences are fully substitutable. Then, for any CE [Ξ; p] and efficient set of trades Ψ, [Ψ; p] is a CE. Theorem (Lattice Structure) The set of CE price vectors is a lattice. Hatfield & Kominers June 4,

92 Stability and Competitive Equilibrium in Trading Networks The Relationship Between Stability and CE Theorem If [Ψ; p] is a CE, then A ψ Ψ {(ψ, p ψ )} is stable. The reverse implication is not true in general. Hatfield & Kominers June 4,

93 Stability and Competitive Equilibrium in Trading Networks The Relationship Between Stability and CE Theorem If [Ψ; p] is a CE, then A ψ Ψ {(ψ, p ψ )} is stable. The reverse implication is not true in general. Theorem Suppose that agents preferences are fully substitutable and A is stable. Then, there exists a price vector p R Ω such that 1 [τ(a); p] is a CE, and 2 if (ω, p ω ) A, then p ω = p ω. Hatfield & Kominers June 4,

94 Full Substitutability is Necessary Stability and Competitive Equilibrium in Trading Networks Theorem Suppose that there exist at least four agents and that the set of trades is exhaustive. Then, if the preferences of some agent i are not fully substitutable, there exist simple preferences for all agents j i such that no stable outcome exists. Hatfield & Kominers June 4,

95 Full Substitutability is Necessary Stability and Competitive Equilibrium in Trading Networks Theorem Suppose that there exist at least four agents and that the set of trades is exhaustive. Then, if the preferences of some agent i are not fully substitutable, there exist simple preferences for all agents j i such that no stable outcome exists. Corollary Under the conditions of the above theorem, there exist simple preferences for all agents j i such that no CE exists. Hatfield & Kominers June 4,

96 Alternative Solution Concepts Stability and Competitive Equilibrium in Trading Networks Definition An outcome A is in the core if there is no group deviation Z such that U i (Z) > U i (A) for all i associated with Z. Hatfield & Kominers June 4,

97 Alternative Solution Concepts Stability and Competitive Equilibrium in Trading Networks Definition An outcome A is in the core if there is no group deviation Z such that U i (Z) > U i (A) for all i associated with Z. Definition A set of contracts Z is a chain if its elements can be arranged in some order y 1,..., y Z such that s(y l+1 ) = b(y l ) for all l < Z. Hatfield & Kominers June 4,

98 Alternative Solution Concepts Stability and Competitive Equilibrium in Trading Networks Definition An outcome A is in the core if there is no group deviation Z such that U i (Z) > U i (A) for all i associated with Z. Definition A set of contracts Z is a chain if its elements can be arranged in some order y 1,..., y Z such that s(y l+1 ) = b(y l ) for all l < Z. Definition Outcome A is stable if it is individually rational and Unblocked: There is no nonempty, feasible Z X such that Z A = and for each i, and for each Y i C i (Z A), we have Z i Y i. Hatfield & Kominers June 4,

99 Alternative Solution Concepts Stability and Competitive Equilibrium in Trading Networks Definition An outcome A is in the core if there is no group deviation Z such that U i (Z) > U i (A) for all i associated with Z. Definition A set of contracts Z is a chain if its elements can be arranged in some order y 1,..., y Z such that s(y l+1 ) = b(y l ) for all l < Z. Definition Outcome A is chain stable if it is individually rational and Unblocked: There is no nonempty, feasible chain Z X s.t. Z A = and for each i, and for each Y i C i (Z A), we have Z i Y i. Hatfield & Kominers June 4,

100 Alternative Solution Concepts Stability and Competitive Equilibrium in Trading Networks Definition An outcome A is in the core if there is no group deviation Z such that U i (Z) > U i (A) for all i associated with Z. Definition A set of contracts Z is a chain if its elements can be arranged in some order y 1,..., y Z such that s(y l+1 ) = b(y l ) for all l < Z. Definition Outcome A is strongly group stable if it is individually rational and Unblocked: There is no nonempty, feasible Z X such that Z A = and for each i associated with Z, there exists a Y i Z A such that Z i Y i and U i (Y i ) > U i (A). Hatfield & Kominers June 4,

101 Generalized Matching Theory Relationship Between the Concepts Stability and Competitive Equilibrium in Trading Networks CE Strongly Group Stable Stable Chain Stable Core Efficient Hatfield & Kominers June 4,

102 Generalized Matching Theory Multilateral Contracts Multilateral Matching Cu Sn Bronzemaker Toymaker Full substitutability is necessary in (Discrete, Bilateral) Contract Matching with Transfers. Hatfield & Kominers June 4,

103 Generalized Matching Theory Multilateral Contracts Multilateral Matching Cu Sn (ω, r ω, s ω ) (ψ,r ψ,s ψ ) Bronzemaker (ϕ,r ϕ,s ϕ) Toymaker Full substitutability is necessary in (Discrete, Bilateral) Contract Matching with Transfers. Hatfield & Kominers June 4,

104 Generalized Matching Theory Multilateral Contracts Multilateral Matching Publisher 1 Publisher 2 (ω, r ω, s ω ) (ψ,r ψ,s ψ ) Ad Exchange (ϕ,r ϕ,s ϕ) Residual Networks Full substitutability is necessary in (Discrete, Bilateral) Contract Matching with Transfers. Hatfield & Kominers June 4,

105 Multilateral Matching Multilateral Contracts Main Results In arbitrary trading networks with 1 multilateral contracts, 2 transferable utility, 3 concave preferences, and 4 continuously divisible contracts, competitive equilibria exist and coincide with stable outcomes. = Some production complementarities work in matching! Hatfield & Kominers June 4,

106 Discussion Generalized Matching Theory Conclusion Applications of stability in absence of CE? Linear programming approach? Empirical applications? Substitutability vs. concavity? Hatfield & Kominers June 4,

107 Discussion Generalized Matching Theory Conclusion Applications of stability in absence of CE? Linear programming approach? Empirical applications? Substitutability vs. concavity? \end{talk} Hatfield & Kominers June 4,

108 Extra Slides Demand-Language Full Substitutability Definition The preferences of agent i are fully substitutable in demand language if for all p, p R Ω such that D i (p) = D i (p ) = 1, 1 if p ω = p ω for all ω Ω i, and p ω p ω for all ω Ω i, then for the unique Ψ D i (p) and Ψ D i (p ), we have Ψ i Ψ i, {ω Ψ i : p ω = p ω} Ψ i ; 2 if p ω = p ω for all ω Ω i, and p ω p ω for all ω Ω i, then for the unique Ψ D i (p) and Ψ D i (p ), we have Ψ i Ψ i, {ω Ψ i : p ω = p ω} Ψ i. Hatfield & Kominers June 4,

109 Extra Slides Indicator-Language Full Substitutability 1 ω Ψ i eω(ψ) i = 1 ω Ψ i 0 otherwise Definition The preferences of agent i are fully substitutable in indicator language if for all price vectors p, p R Ω such that D i (p) = D i (p ) = 1 and p p, for Ψ D i (p) and Ψ D i (p ), we have e i ω(ψ) e i ω(ψ ) for each ω Ω i such that p ω = p ω. Hatfield & Kominers June 4,