Screening and Adverse Selection in Frictional Markets Benjamin Lester Philadelphia Fed Ali Shourideh Wharton Venky Venkateswaran Ariel Zetlin-Jones NYU Stern Carnegie-Mellon March 2015 1 / 31 Introduction Many markets feature adverse selection and imperfect competition • Examples: insurance, loans, financial securities • Screening, e.g. through differential coverage, loan amounts, trade sizes 2 / 31 Introduction Many markets feature adverse selection and imperfect competition • Examples: insurance, loans, financial securities • Screening, e.g. through differential coverage, loan amounts, trade sizes A unified theoretical framework is lacking • Large empirical literature (and some theory) • But typically restricts contracts and/or assumes perfect competition 2 / 31 Introduction Many markets feature adverse selection and imperfect competition • Examples: insurance, loans, financial securities • Screening, e.g. through differential coverage, loan amounts, trade sizes A unified theoretical framework is lacking • Large empirical literature (and some theory) • But typically restricts contracts and/or assumes perfect competition This project: A tractable model that combines all three elements • Complete characterization of the unique equilibrium • Testable predictions for distribution of contracts • Policy experiments: changes in competition, transparency 2 / 31 Environment A seller with 1 divisible good, type i ∈ {h, `} with prob. µi • privately observed value ci , • payoff from an allocation (x, t): ch > c` ci (1 − x) + t 3 / 31 Environment A seller with 1 divisible good, type i ∈ {h, `} with prob. µi • privately observed value ci , • payoff from an allocation (x, t): ch > c` ci (1 − x) + t Two buyers with value vi • payoff from (x, t): vi x − t • gains from trade for both types: vi > ci • ‘lemons’ assumption: vl < ch 3 / 31 Environment A seller with 1 divisible good, type i ∈ {h, `} with prob. µi • privately observed value ci , • payoff from an allocation (x, t): ch > c` ci (1 − x) + t Two buyers with value vi • payoff from (x, t): vi x − t • gains from trade for both types: vi > ci • ‘lemons’ assumption: vl < ch Search frictions • buyers post arbitrary menus of exclusive contracts • seller receives 1 offer w.p. 1 − π and both w.p. π 3 / 31 Strategies buyers: by the revelation principle, sufficient to consider • menus with two contracts (ICj ) : z ≡ {(xl , tl ), (xh , th )} tj + cj (1 − xj ) ≥ t−j + cj (1 − x−j ) j ∈ {h, l} 4 / 31 Strategies buyers: by the revelation principle, sufficient to consider • menus with two contracts (ICj ) : z ≡ {(xl , tl ), (xh , th )} tj + cj (1 − xj ) ≥ t−j + cj (1 − x−j ) j ∈ {h, l} seller: chooses a contract from available menus • optimality with 2 menus χi (z, z0 ) = 0 1 2 1 if < ti + ci (1 − xi ) ti0 + ci (1 − xi0 ). = > 4 / 31 Equilibrium definition A symmetric equilibrium is a distribution Φ(z) such that almost all z satisfy, 1 Incentive compatibility: tj + cj (1 − xj ) ≥ t−j + cj (1 − x−j ) 2 j ∈ {h, l} Seller optimality: χi (z, z0 ) maximizes her utility 3 Buyers optimality: z ∈ arg max X z i∈{l,h} χi (z, z )Φ(dz ) , Z 0 µi (vi xi − ti ) 1 − π + π 0 (1) z0 5 / 31 Characterization No pure strategy eq. for π ∈ (0, 1) → mixing over multiple dimensions ! 6 / 31 Characterization No pure strategy eq. for π ∈ (0, 1) → mixing over multiple dimensions ! Proceed in 4 steps 1. Show that menus can be summarized by a pair of utilities (uh , ul ) 2. Prove that equilibrium is strictly rank-preserving (SRP) 3. Construct SRP equilibrium 4. Show that constructed equilibrium is unique 6 / 31 A utility representation Result In all menus offered in equilibrium, • the low types trades everything: • ICl binds: xl = 1 tl = th + cl (1 − xh ) 7 / 31 A utility representation Result In all menus offered in equilibrium, • the low types trades everything: • ICl binds: xl = 1 tl = th + cl (1 − xh ) Result Equilibrium menus can be represented by (uh , ul ) with corresponding allocations tl = ul xh = 1 − uh − ul ch − cl th = ul ch − uh cl ch − cl 7 / 31 A utility representation Result In all menus offered in equilibrium, • the low types trades everything: • ICl binds: xl = 1 tl = th + cl (1 − xh ) Result Equilibrium menus can be represented by (uh , ul ) with corresponding allocations tl = ul xh = 1 − uh − ul ch − cl th = ul ch − uh cl ch − cl Since we must have 0 ≤ xh ≤ 1, ch − cl ≥ uh − ul ≥ 0 7 / 31 A utility representation Marginal distributions Z Fj (uj ) = 1 tj0 + cj 1 − xj0 ≤ uj dΦ z0 j ∈ {h, l} z0 8 / 31 A utility representation Marginal distributions Z Fj (uj ) = 1 tj0 + cj 1 − xj0 ≤ uj dΦ z0 j ∈ {h, l} z0 Then, each buyer solves Π(uh , ul ) = max ul ≥cl , uh ≥ch X µj [1 − π + πFj (uj )] Πj (uh , ul ) j∈{l,h} ch − cl ≥ uh − ul ≥ 0 s. t. with Πl (uh , ul ) ≡ vl xl − tl = vl − ul Πh (uh , ul ) ≡ vh xh − th = vh − uh vh − ch vh − cl + ul ch − cl ch − cl 8 / 31 A utility representation Marginal distributions Z Fj (uj ) = 1 tj0 + cj 1 − xj0 ≤ uj dΦ z0 j ∈ {h, l} z0 Then, each buyer solves Π(uh , ul ) = max ul ≥cl , uh ≥ch X µj [1 − π + πFj (uj )] Πj (uh , ul ) j∈{l,h} ch − cl ≥ uh − ul ≥ 0 s. t. with Πl (uh , ul ) ≡ vl xl − tl = vl − ul Πh (uh , ul ) ≡ vh xh − th = vh − uh vh − ch vh − cl + ul ch − cl ch − cl Need to characterize the two linked distributions Fl and Fh ! 8 / 31 Some useful results Result Fl and Fh have connected support and are continuous. • Except for a knife-edge case (see paper) • Proof more involved than standard case because of interdependencies 9 / 31 Some useful results Result Fl and Fh have connected support and are continuous. • Except for a knife-edge case (see paper) • Proof more involved than standard case because of interdependencies Result The profit function Π(uh , ul ) is strictly supermodular. • Intuition: ul ↑ ⇒ Πh ↑ ⇒ stronger incentives to attract high types • ⇒ Uh (ul ) ≡ argmaxuh Π(uh , ul ) is weakly increasing 9 / 31 Strict Rank Preserving Violates no−mass−points in Fh Uh Uh Violates no−mass−points in Fl ul ul Theorem Uh (ul ) is a strictly increasing function. 10 / 31 Strict Rank Preserving Violates no−mass−points in Fh Uh Uh Violates no−mass−points in Fl ul ul Theorem Uh (ul ) is a strictly increasing function. • Rank ordering of equilibrium menus identical across types • Menus attract same fraction of both types Fl (ul ) = Fh (Uh (ul )) 10 / 31 Strict Rank Preserving Theorem Uh (ul ) is a strictly increasing function. Fl (ul ) = Fh (Uh (ul )) 11 / 31 Equilibria: The two limit cases Monopsony: π = 0 • µh < µ̄h ⇒ Sep. with xh = 0 • µh ≥ µ̄h ⇒ and Πl > Πh = 0 Pooling with xh = xl = 1 and Πh > 0 > Πl 12 / 31 Equilibria: The two limit cases Monopsony: π = 0 • µh < µ̄h ⇒ Sep. with xh = 0 • µh ≥ µ̄h ⇒ and Πl > Πh = 0 Pooling with xh = xl = 1 and Πh > 0 > Πl Bertrand: π = 1 • µh < µ̄h ⇒ Sep. with xh < 1, Πh = Πl = 0 • µh < µ̄h ⇒ Sep. with xh < 1, Π = 0, but Πh > 0 > Πl Intuition: Higher µh ⇒ Relaxing IC l more attractive 12 / 31 Types of equilibria Cross-subsidization No cross-subsidization 13 / 31 Types of equilibria High µh Cross-subsidization • Πh > 0 > Πl • All separating, all pooling or a mix Low µh • Πl , Πh ≥ 0 No cross-subsidization • All separating, Uh (ul ) 6= ul 13 / 31 No cross-subsidization: Characterization Marginal benefits vs costs of increasing ul πfl (ul ) Πl µh vh − ch + 1 − π + πFl (ul ) 1 − µh ch − cl | {z } | {z } MB of more low types = 1 |{z} MC MB of relaxing ICl 14 / 31 No cross-subsidization: Characterization Marginal benefits vs costs of increasing ul πfl (ul ) Πl µh vh − ch + 1 − π + πFl (ul ) 1 − µh ch − cl | {z } | {z } MB of more low types = 1 |{z} MC MB of relaxing ICl Boundary conditions Fl (cl ) = 0 Fl (ūl ) = 1 → Fl (ul ) Equal profit condition [1 − π + πFl (ul )] Π(Uh , ul ) = Π → Uh (ul ) 14 / 31 No cross-subsidization: Distributions, π = 0.2 15 / 31 No cross-subsidization: Distributions, π = 0.5 16 / 31 No cross-subsidization: Distributions, π = 0.9 17 / 31 No cross-subsidization: Allocations 18 / 31 No cross-subsidization: Allocations Uh (ul ) − ul ch − cl xh (ul ) = 1− xh0 (ul ) > 0 ⇔ Uh0 (ul ) > 1 19 / 31 No cross-subsidization: π = 0.2 20 / 31 No cross-subsidization: π = 0.5 21 / 31 No cross-subsidization: π = 0.9 22 / 31 No cross-subsidization: Welfare Z W = (1 − µh )(vl − cl ) + µh (vh − ch ) xh (ul )d F̂ (ul ) 23 / 31 No cross-subsidization: Welfare Z W = (1 − µh )(vl − cl ) + µh (vh − ch ) 0 π xh (ul )d F̂ (ul ) 1 23 / 31 No cross-subsidization: Welfare Z W = (1 − µh )(vl − cl ) + µh (vh − ch ) 0 π xh (ul )d F̂ (ul ) 1 Intuition: Profits change rapidly close to ul = vl ⇒ Uh0 (ul ) > 1 (because it has a bigger effect on profits) ⇒ xh0 < 0 23 / 31 No cross-subsidization: Price vs quantity (conditional) ph π = 0.2 Correlation < 0 for suff. high π π = 0.5 ph A strategy to infer competitiveness ? ph π = 0.95 xh 24 / 31 No cross-subsidization: Policies Do the following policies improve welfare ? • Allowing insurance providers to discriminate based on observables • Introducing credit scores in loan markets • Requiring OTC market participants to disclose trades 25 / 31 No cross-subsidization: Policies Do the following policies improve welfare ? • Allowing insurance providers to discriminate based on observables • Introducing credit scores in loan markets • Requiring OTC market participants to disclose trades Answer: Depends on π ! 25 / 31 No cross-subsidization: Policies Do the following policies improve welfare ? • Allowing insurance providers to discriminate based on observables • Introducing credit scores in loan markets • Requiring OTC market participants to disclose trades Answer: Depends on π ! • Can be mapped into a mean-preserving spread of µh • W is linear when π = 0 and π = 1 ⇒ no effect on welfare • W is concave when π is high ⇒ bad for welfare 25 / 31 Cross-subsidization No cross-subsidization 26 / 31 Cross-subsidization No cross-subsidization 27 / 31 Cross-subsidization: Separating All equilibrium menus • cross-subsidize Πl < 0 • separate (except the best one) Uh (ul ) > ul No cross-subsidization 28 / 31 Cross-subsidization: Pooling All equilibrium menus • cross-subsidize Πl < 0 • pool Uh (ul ) = ul No cross-subsidization 29 / 31 Cross-subsidization: Semi-Pooling Equilibrium menus • cross-subsidize Πl < 0 • separate at the high end, Uh (ul ) > ul iff ul > ûl No cross-subsidization • and pool at the low end, Uh (ul ) = ul iff ul ≤ ûl 30 / 31 Conclusion Methodological contribution • Imperfect competition and adverse selection with optimal contracts • Rich predictions for the distribution of observed trades Substantive insights • Depending on parameters, pooling and/or separating menus in equilibrium • Competition, transparency can be bad for welfare Work in progress • Generalize to N types, curved utility • Non-exclusive trading 31 / 31