Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
SCREENING UNDER ADVERSE SELECTION Some Optimal Regulation Theory • Principal-Agent Theory in the pure Adverse-Selection Case. • Principal : Agent : State Public Firm Banker Entrepreneur Firm Worker Shareholder Manager • Pure Adverse Selection : hidden characteristics of the agent only. • No hidden actions (No moral hazard). 1 • Important applications : Optimal Regulation and Optimal Taxation. • We study a basic model of the Principal-Agent relationship • We look for the optimal contract between Principal and Agent • The Principal is uninformed. 2 Must Read : • Jean-Jacques LAFFONT and David MARTIMORT (2002), The Theory of Incentives : The Principal-Agent Model, Princeton University Press, Chapter 2, pp. 28-81. Other important references : • Davis P. BARON and Roger B. MYERSON (1982), ”Regulating a Monopolist with Unknow Costs”, Econometrica, 50, pp. 911-930 • Jean-Jacques LAFFONT and Jean TIROLE (1986), ”Using Cost Observation to Regulate Firms”, Journal of Political Economy, 94, pp. 614-41. 3 Basic Principal-Agent Model • Agent : produces q ≥ 0 • Surplus due to production is S(q) • We assume that S is differentiable and S 0 > 0, S 00 < 0, S(0) = 0. • Marginal Cost of production is denoted θ ∈ {θ, θ̄} (Two types) • Cost function : C(q, θ) = θq • θ < θ̄ 4 • So, θ̄ = ”inefficient type” θ = ”efficient type” • Prior probability of types : Prob (θ = θ) = ν ν ∈ (0, 1) • Denote 4θ = θ̄ − θ (the ”spread”) 5 Definition of a Contract • A pair of functions θ 7−→ (q(θ), t(θ)) • q(θ) = production of type θ • t(θ) = transfer of money from Principal to Agent 6 • Utility of the Agent : U = t − θq • Utility of the Principal : V = S(q) − t • Social surplus : W =U +V = S(q) − θq 7 • First-Best Allocation : maximize (S(q) − θq) with respect to q, for all θ, s.t. q ≥ 0 or, maximize (S(q) − t) with respect to (q, t) subject to, t − θq ≥ Uo(θ) (participation constraint) and q ≥ 0, for all θ. 8 • First-Order Conditions for First-Best Optimality : 0 ∗ S (q ) = θ ⇒ q ∗ > q̄ ∗ S 0(q̄ ∗) = θ̄ • Transfers : to implement the first-best, ∗ ∗ t = θq + U0(θ) t̄∗ = θ̄q̄ ∗ + U (θ̄) 0 ⇒ U = U0(θ) (No rents) 9 • For simplicity, we assume that Uo(θ) = Uo = 0 (reservation utilities do not depend on type) • The timing of the model : t = 0 Agent discovers θ t = 1 Principal offers contract (q, t) t = 2 Agent accepts or refuses t = 3 Contract is executed. 10 Incentive Compatibility and Feasibility : • A contract is an array : {(t̄, q̄), (t, q)} • The contract is Incentive Compatible if and t − θq ≥ t̄ − θq̄ (IC) t̄ − θ̄q̄ ≥ t − θ̄q (IC) 11 • The contract is Individually Rational if t − θq ≥ 0 (IR) t̄ − θ̄q̄ ≥ 0 (IR) • IR + IC = feasible contract 12 Monotonicity Property : • Adding IC and IC yields, (θ̄ − θ)q ≥ (θ̄ − θ)q̄ ⇒ q ≥ q̄ • If monotonicity holds, then, there exists (t̄, t) such that IC holds : θ(q − q̄) ≤ t − t̄ ≤ θ̄(q − q̄) (IC) (IC) 13 INFORMATION RENTS : • Denote U = t̄ − θ̄q̄ ; U = t − θq. • We have, t̄ − θq̄ = t̄ − θ̄q̄ + q̄4θ = U + q̄4θ • It follows that IC is equivalent to, U ≥ U + 4θq̄ 14 • We also have, t − θ̄q = t − θq − 4θq = U − 4θq • Hence, IC is equivalent to U ≥ U − q̄4θ • And we see that U > U . The informational rent of type θ̄ is q̄4θ. 15 The Principal’s Problem : (Second-best problem) max ν(S(q) − t) + (1 − ν)(S(q̄) − t̄) (t̄,q̄,t,q) subject to IC, IC, IR, IR. Use change of variables : U = t − θq 16 The Principal’s problem rewritten : max {ν[S(q) − θq] + (1 − ν)[S(q̄) − θ̄q̄] − [νU + (1 − ν)U ]} (U ,q,Ū ,q̄) subject to, U ≥ U + q̄4θ (IC) U ≥ U − q4θ (IC) U ≥0 U ≥0 (IR) (IR) We consider contracts without ”shutdown”, i.e., q̄ > 0. 17 • If IC and IR hold, then IR is always satisfied : U ≥ U + q̄4θ ≥ q̄4θ > 0. • Both IC and IR must be binding at the second-best optimum. • If not, let U = ε > 0. Choose dε < 0 and decrease U and U by dε. Contradiction since (νU + (1 − ν)U ) decreases. • If U = q̄4θ+η, η > 0, then, decrease η by dη < 0. Contradiction. • We conclude that U = q̄4θ at the second-best optimum. 18 • The Principal’s Problem becomes : max{ν[S(q) − θq] + (1 − ν)[S(q̄) − θ̄q̄] − ν4θq̄} (q,q̄) (if we substitute Ū = 0 and U = q̄4θ and we ignore IC for a while...) • The First-Order Conditions are 0 ∗∗ S (q ) = θ {S 0(q̄ ∗∗) − θ̄}(1 − ν) = ν4θ • Distortion = ν4θ/(1 − ν) • Trade-off : EFFICIENCY vs. RENT EXTRACTION 19 • We finally check that IC also holds. • From monotonicity : q̄ ∗∗ ≤ q ∗∗. The omitted IC constraint can be written U ∗∗ = 0 ≥ U ∗∗ − 4θq ∗∗ ≥ 4θq̄ ∗∗ − 4θq ∗∗ ⇔ q ∗∗ ≥ q̄ ∗∗ (true). 20 • Note : q ∗∗ = q ∗ > q̄ ∗ > q̄ ∗∗. This is because, S 0(q̄ ∗∗) = θ̄ + ν 4θ. 1−ν 21 Proposition : The second-best optimal contract is such that, S 0(q ∗∗) = θ (”no distortion at the top”) S 0(q̄ ∗∗) = θ̄ + ν 4θ (distortion) 1−ν U ∗∗ = q̄ ∗∗ 4θ (informational rent for efficient types) U ∗∗ = 0 (full rent extraction for inefficient types) • Discussion : rôle played by ν and 4θ. 1−ν 22 Some Mechanism Design. Suppose we have one Principal and n agents indexed i = 1, ..., n. Agent i’s type θi is drawn from a set Θ. Principal chooses a decision x ∈ X. (For instance x = (q, t), a contract). The utility of agent i is U (x; θi) 23 Definition 1 (Mechanism) A mechanism is a pair (M, g) where M is a message space and g : M n → X is an outcome function. x = g(m1, m2, ..., mn) • We suppose that θ = (θ1, ..., θn) is not observed by the Principal. Each θi is private information of agent i. Definition 2 (Direct Mechanism) A mechanism is direct if M = Θ. 24 Definition 3 (Equilibrium in Dominant Strategies) Let m∗i : Θ → M be agent i0s communication strategy Then, (m∗i (θi))i=1...n is an equilibrium in dominant strategies if for all i, for all (m∗j (θj ))j6=i = m∗−i(θ−i), we have U [g(m∗(θ)), θi] ≥ U [g(m∗−i(θ−i), m̂i), θi] for all m̂i ∈ M • Notation : θ−i = (θj )j6=i m(θ) = (m1(θ), ..., mn(θ)) 25 Definition 4 (Revealing Mechanism) A mechanism (M, g) is revealing (in dominant strategies) if m∗i (θi) ≡ θi is a dominant strategy for all i = 1, ..., n. Note : (M, g) is a direct and revealing mechanism if m∗i (θi) = θi is an equilibrium is dominant strategies. 26 REVELATION PRINCIPLE Theorem : If mechanism (M, g) implements decision f : Θ → X in dominant strategies, that is, g(m∗(θ)) = f (θ) for all θ, m∗ being a n−tuple of dominant strategies, then, (Θ, f ) is revealing in dominant strategies. Remark : If (M, g) chooses f (θ) for all θ, then, there exists a revealing mechanism which does the same job ; i.e., (Θ, f ). 27 Proof of the Revelation Principle : • There exists a n−tuple of dominant strategies m∗ such that g[m∗(θ)] = f (θ) for all θ, by assumption. • U [g(m∗i (θi), m−i), θi] ≥ U [g(m0i, m−i), θi], for all m0i ∈ M , all m−i ∈ M n−1. 28 • So, in particular, for all i and θi, U [g(m∗i (θi), m∗−i(θ−i)), θi] ≥ U [g(m∗i (θi0 ), m∗−i(θ−i)), θi] for all θi0 , all θ−i. • Now, g[m∗(θ)] = f (θ), then, for all i, all θi, U [f (θ), θi] ≥ U [f (θi0 , θ−i), θi] for all θi0 , θ−i 29 • We conclude that f is truthfully implementable by the direct revealing mechanism (Θ, f ). • Agents report their types θi ∈ Θ directly. • Agents have no incentive to make false reports (in a very strong sense). • There is no loss of generality in constraining optimal contracts to be incentive compatible (i.e., revealing). 30 Other equivalent interpretation : If g implements f and g:M →X is not revealing (and not direct). Then g̃ = g ◦ m∗ is a direct and revealing mechanism that also implements f . m∗ g Θ −−→ M − →X g̃(.) = g ◦ m∗(.) 31