ECON601 Spring, 2015 UBC Li, Hao Lecture 9. Moral Hazard
Transcription
ECON601 Spring, 2015 UBC Li, Hao Lecture 9. Moral Hazard
ECON601 Spring, 2015 UBC Li, Hao Lecture 9. Moral Hazard Motivation • Origin of “moral hazard:” insurance on babies • General characteristics of moral hazard problems – Unobserved actions of one party, called agent, exert externality on the payoff of the other, called principal. – Principal can write an enforceable contract on verifiable outcomes that depend on agent’s action. – Agent enters the contract on a voluntary basis. 1 • Hidden action problems vs hidden information problems – Solution to hidden information problems is self-selection menus; solution to hidden action problems is contracts. – Incentive condition is truthful reporting of hidden information; and obeying with recommended action. General setup of moral hazard model in employer-employee relationship • Employee chooses effort e, not observed by employer. – Employee’s payoff is U (w, e) = u(w) − c(e), where w is wage paid by employer, with u0 > 0 and u00 < 0 (employee is risk averse), and c0 > 0 and c00 ≥ 0 (effort cost function is convex). – Employee’s reservation utility is U. • Employer’s payoff is V (y). – y is employer’s profit, with V 0 > 0 and V 00 ≤ 0 (employer may be risk neutral). 2 • Wage contract takes form of sharing of output x. – w( x ) is wage, and x − w( x ) is profit when output is x. • Modeling stochastic dependence of output x on employee’s effort e. – Direct approach: postulate a function x (e, θ ), where θ is a random variable that also affects output x, with some known distribution function—for example, x (e, θ ) = e + θ. – Parameterized approach: suppose output x ∈ R takes value from some set X, which is either an interval [ x, x ] or finite (common support); write as f ( x |e) density function (or probability function) of output when effort is e, and as F ( x |e) the corresponding distribution function. • Timing of game – Employer makes a take-it-or-leave-it offer of a contract to employee. – Game ends if employee rejects the offer; otherwise, employee chooses effort e, output x is realized, and wage is paid according to contract and game ends. 3 • Implicit assumptions: – Symmetric information at contracting stage. – Effort e is not observable to employer. – Output is verifiable and contractible. – No renegotiation of contract—both employee and employer commit to contract. Optimal contract design problem • Formulate contract design problem as a constrained optimization problem. – Employer chooses wage function w( x ) and effort e to maximize payoff given R by x∈X V ( x − w( x )) f ( x |e)dx, subject to R R – (IC) x∈X u(w( x )) f ( x |e)dx − c(e) ≥ x∈X u(w( x )) f ( x |e0 )dx − c(e0 ) for all e0 . R – (IR) x∈X u(w( x )) f ( x |e)dx − c(e) ≥ U. – We may think of the choice e in the maximization problem as “recommended effort;” (IC) is then interpreted as “obedience.” 4 • A two-step procedure for solving for optimal contract – Step 1: For each recommended effort e, choose a wage function w( x ) in order R to maximize x∈X V ( x − w( x )) f ( x |e)dx subject to (IC) and (IR), yielding the expected profit as a function of e. – Step 2: maximize the expected profit function by choosing e. Benchmark: Observed actions • Suppose that efforts are observable and contractible, so contract can be written on actions directly. – Still need to condition wage on output—employee is risk averse and values insurance. – But there is no (IC) constraint. – Optimal contract problem is then to choose wage function w( x ) and effort e to R maximize x∈X V ( x − w( x )) f ( x |e)dx subject to (IR) only. 5 • Use the two-step procedure to solve for the benchmark. – For any fixed e, in first step, (IR) binds at solution: if not could reduce w( x ) for each x and increase profit. – Let λ be the Lagrangian multiplier, and take derivative of objective with respect to w( x ) for each x ∈ X: first order condition is V 0 ( x − w( x ))/u0 (w( x )) = λ. – Can solve for each w( x ) in terms of λ and substitute in (IR) to get λ, and then each w( x ), as functions of e. – Optimal e can be found in second step. • Optimal risk-sharing: ratio of marginal utilities is independent across output. – In special case with V 0 constant, optimal risk-sharing implies full insurance for employee: w( x ) is constant. – (IR) becomes u(w) − c(e) = U, implying w = u−1 (c(e) + U ). – In second step, optimal effort e∗ , called “first best,” maximizes total “surplus” R −1 x ∈ X x f ( x | e )dx − u ( c ( e ) + U ). 6 Hidden actions: When both employer and employee are risk neutral • First best can be achieved even though employee’s effort is unobservable. – Contract cannot be written on effort. • Without loss of generality, assume u(y) = V (y) = y. – Key observation: when employee is risk neutral, the first best e∗ is incentive compatible if he is residual claimant to output—employee pays a constant fee to employer (franchise fee, tenancy). – After paying the fee, employee chooses e to maximize R x∈X x f ( x |e)dx − c(e), leading to e∗ (for the case of risk neutral employee). • Optimal contract for employer: choose a fee equal to R x∈X f ( x |e∗ )dx − (c(e∗ ) + U ) to bind (IR). – Interpretation: employer “sells the firm” to employee. – Employee is willing to bear all output risk, because of risk-neutrality. 7 Two-effort model • Suppose set of effort choices of employee is {e H , e L }, with e L < e H . – A representative model demonstrating how to apply the two-step procedure. – The central idea of moral hazard models: incentives, which require wage to increase when higher output is realized, versus insurance, which requires riskneutral employer to insure risk-averse employee against inherent output risks. • Assumption on output distributions: F ( x |e H ) first-order stochastically dominates F ( x |e L ), that is, F ( x |e H ) ≤ F ( x |e L ) for all x ∈ X. – e H leads to a higher output than e L , but only stochastically. • Step 1: if e L is recommended effort. – Choose w to bind (IR): u(w) − c(e L ) = U. – (IC) is satisfied under constant wage, because c(e H ) > c(e L ): no need to provide incentives through wage contract if employer wants to implement e L . 8 • Step 1, continued: if e H is recommended effort. – Consider the constrained maximization problem of choosing a wage contract R w( x ) to maximize the expected profit x∈X ( x − w( x )) f ( x |e H )dx, subject to R R (IC) x∈X u(w( x )) f ( x |e H )dx − c(e H ) ≥ x∈X u(w( x )) f ( x |e L )dx − c(e L ), and (IR) R x ∈ X u ( w ( x )) f ( x | e H )dx − c ( e H ) ≥ U. – Let µ be multiplier for (IC) and λ multiplier for (IR), and write first order condition with respect to w( x ) as: 1/u0 (w( x )) = λ + µ(1 − f ( x |e L )/ f ( x |e H )). – µ > 0 at any solution w( x ) so that (IC) binds: otherwise, first-order condition implies full insurance and constant wage, but then (IC) cannot hold because c ( e H ) > c ( e L ). – λ > 0 at any solution w( x ) so that (IR) binds: otherwise, for small positive e, can define new contract w˜ ( x ) such that u(w˜ ( x )) = u(w( x )) − e for all x ∈ X and (IR) remains satisfied; then (IC) is unaffected but value of objective is increased; alternatively, if λ = 0, then first-order condition implies that f ( x |e H ) > f ( x |e L ) for all x ∈ X, an impossibility since they are density functions. 9 • Step 2: does employer want to implement e H ? – The answer may be no, if incentive cost of implementing e H is higher than the benefit relative to first best solution to e L . • Interpreting first order condition: tradeoff between incentive and insurance. – Full insurance is not compatible with effort incentives: to provide incentives for employee to choose e H instead of shirking by choosing e L , wage w must vary with output x. – Incomplete insurance is costly to employer: concavity of u implies that certainty R equivalent of w( x ) to employee, wˆ such that u(wˆ ) = x∈X u(w( x ))dF ( x |e H ), is R less than expected cost of the wage contract to employer, x∈X w( x )dF ( x |e H ) (Jensen’s inequality). • Under what condition optimal contract exhibits wage monotonicity? – Answer is: if f ( x |e H )/ f ( x |e L ) is increasing in x, from first order condition with respect to w( x ) in step 1 when recommended effort is e H . 10 • Monotone likelihood ratio property (MLRP): statistics – f ( x |e H )/ f ( x |e L ) is likelihood ratio: for any given x, the ratio gives likelihood of e H is chosen relative to likelihood e L is chosen, so under MLRP, the higher the output x, the more likely that e H was chosen instead of e L . – MLRP implies first order stochastic dominance: take any x ∈ [ x, x ]; by MLRP f ( x1 |e H ) f ( x2 |e L ) ≤ f ( x1 |e L ) f ( x2 |e H ) for any x1 , x2 ∈ X and x1 ≤ x ≤ x2 ; integrate both sides first from x1 = x to x1 = x, and then from x2 = x to x2 = x to get F ( x |e H )(1 − F ( x |e L )) ≤ F ( x |e L )(1 − F ( x |e H )), so F ( x |e H ) ≤ F ( x |e L ). • Monotone likelihood ratio property (MLRP): economics – Wage monotonicity requires MLRP; first order stochastic dominance is not enough. – Wage monotonicity at optimal wage contract arises from incentive provision, not from inference about effort; indeed, at the optimal wage contract, employer knows employee has chosen e H . 11 • Value of information – Suppose contractible information is richer: besides output x, wage can also depend on another consequence y of e. – Same first order condition: 1/u0 (w( x, y)) = λ + µ(1 − f ( x, y|e L )/ f ( x, y|e H )). – Optimal wage contract w( x, y) depends also on y if and only if the likelihood ratio f ( x, y|e H )/ f ( x, y|e L ) depends on y. – Statistics: x is “sufficient statistic” for ( x, y) in inference problem with regards to e H versus e L , if f ( x, y|e H )/ f ( x, y|e L ) does not depend on y. – Economics: contractible information should be used in the optimal contract together with output only if it affects likelihood ratio; otherwise, dependence would reduce employee’s certainty equivalent and hence profit. – Application: is relative performance evaluation optimal when the outputs are separately produced and observed? Answer is “no” if output distributions are independent—“competition” is not useful per se. 12 • Renegotiation? – Suppose original wage contract is w( x ), to implement e H . – After e H is chosen and before output x is realized, employee’s cost of effort is sunk but there is output risk due to incomplete insurance provided by the employer in order to provide incentives. – Employer would have incentive to renegotiate the wage contract at this stage, by offering the certainty equivalent w˜ of original contract w( x ) under e H , that R is, u(w˜ ) = x∈X u(w( x )) f ( x |e H )dx. – Employee is happy to accept new offer (is indifferent), but employer is better off R R by Jensen’s inequality: u(w˜ ) = x∈X u(w( x )) f ( x |e H )dx < u x∈X w( x ) f ( x |e H )dx , R implying that w˜ < x∈X w( x ) f ( x |e H )dx. – Of course, if such renegotiation is anticipated, employer would not be able to implement e H in the first place: inability to commit to not to renegotiate hurts principal. 13 Multiple levels of effort • Grossman-Hart (ECMA83) setup. – Finitely many effort levels, and finitely many outputs, x1 < x2 < . . . < xn , with probability f ( xi |e) > 0 of output level xi given effort level e for each i. • Two-step procedure again: first find the least costly way to implement each effort level, and then find the effort that leads to maximum profit. – For given e to be implemented, principal chooses wi , i = 1, . . . , n, to minimize ∑in=1 f ( xi |e)wi , subject to (IR) ∑in=1 f ( xi |e)u(wi ) − c(e) ≥ u, and (IC) that e solves maxe˜ ∑in=1 f ( xi |e˜)u(wi ) − c(e˜). – Consider change of variables: choose ui , i = 1, . . . , n. – Objective is convex in choice variables, as wi = u−1 (ui ); (IR) is linear (with multiplier λ); (IC) is linear constraints (with multiplier µ j for each e j 6= e); and first order conditions 1/u0 (ui ) = λ + ∑ j:e j 6=e µ j (1 − f ( xi |e j )/ f ( xi |e)) are both necessary and sufficient. 14 Continuous effort • Mirrlees’ setup – Effort e is chosen from some compact interval, and f ( x |e) has common support [ x, x ] for all feasible e. • Main issue: use the same two-step procedure but in first step, there is a continuum of IC for recommended effort. • Rogerson (ECMA85)’s first order approach. – Replace (IC) by agent’s first order condition with respect to e, solve resulting relaxed problem, and impose conditions such that solution satisfies original (IC) constraint. – In last step, ensure agent’s effort choice problem is concave. – For fixed e, if principal chooses w( x ), agent’s first order condition in effort Rx choice is x u(w( x )) f e ( x |e)dx = c0 (e), where f e ( x |e) ≡ ∂ f ( x |e)/∂e. 15 • The relaxed problem: principal chooses w( x ) to maximize subject to (IR) Rx x Rx x ( x − w( x )) f ( x |e)dx, u(w( x )) f ( x |e)dx − c(e) ≥ u, and above first order condition of agent. – First order condition with respect to w( x ) from point-wise maximization is: 1/u0 (w( x )) = λ + µ f e ( x |e)/ f ( x |e), where λ is the multiplier for (IR) and µ the multiplier for agent’s first order condition. – Assume the continuous version of MLRP that f e ( x |e)/ f ( x |e) is nondecreasing. – Solution to relaxed problem satisfies w0 ( x ) ≥ 0. – Assume the convexity of distribution function condition (CDFC) on F ( x |e) that F ( x |λe1 + (1 − λ)e2 ) ≤ λF ( x |e1 ) + (1 − λ) F ( x |e2 ) for all x ∈ [ x, x ], λ ∈ [0, 1], and feasible e1 and e2 . – With c00 ≥ 0, agent’s effort choice problem is concave in effort choice because Rx Rx 0 0 x u ( w ( x )) f ( x | e ) dx = u ( w ( x )) − x u ( w ( x )) w ( x ) F ( x | e ) dx is concave in e due to CDFC and w0 ( x ) ≥ 0. 16 Moral hazard in teams • When multiple agents are engaged in joint production, there is also the free-rider problem when individual efforts are not observed, in addition to trade-off between insurance and incentives. • Setup of a joint production problem: – There are n agents; no principal. – Each i, i = 1, . . . , n, chooses an effort ei ≥ 0, unobserved to any other agent or the principal. – No uncertainty: output x is deterministic function of effort profile: x = g(e1 , . . . , en ), with g continuous, positively-valued, and increasing in each argument. – Agents are all risk-neutral: payoff to i is si − ei , where si is i’s share of output (effort cost function is linear). – Agents’ non-participation payoff is 0. 17 • A contract in this context is an output sharing agreement. – Suppose that agents agree beforehand on sharing contract: (s1 ( x ), . . . , sn ( x )) such that si ( x ) ≥ 0 for each i = 1, . . . , n, and ∑in=1 si ( x ) = x. • Efficient (first best) efforts – Suppose efforts are observed. – Efficient effort profile (e1∗ , . . . , en∗ ) maximizes g(e1 , . . . , en ) − ∑in=1 ei : for any other effort profile (eˆ1 , . . . , eˆn ), and for any sharing (s1 ( xˆ ), . . . , sn ( xˆ )) where xˆ = g(eˆ1 , . . . , eˆn ), if each i switches to ei∗ and receives si ( xˆ ) − eˆi + ei∗ he would be indifferent, but then we would have ∑in=1 (si ( xˆ ) − eˆi + ei∗ ) is smaller than ∑in=1 si ( g(e1∗ , . . . , en∗ )) by construction, implying the gain can be distributed to make everyone better off. – First order condition with respect to ei : ∂g(e1∗ , . . . , en∗ )/∂ei = 1. – Suppose that there is a sharing arrangement (s1∗ , . . . , s∗n ) such that each agent is willing to particapte: si∗ − ei∗ ≥ 0 for each i, and ∑in=1 si∗ = g(e1∗ , . . . , en∗ ). 18 • Can efficient efforts be implemented under some sharing contract? – Answer is “no:” simple proof restricts to differentiable sharing contracts. – Suppose (e1∗ , . . . , en∗ ) is a Nash equilibrium under some (s1 ( x ), . . . , sn ( x )). – Each i chooses ei to maximize si ( g(e1∗ , . . . , ei , . . . , en∗ ) − ei —first order condition is given by si0 ( g(e1∗ , . . . , en∗ ))(∂g(e1∗ , . . . , en∗ )/∂ei ) = 1. – Using the first order conditions for efficient profile (e1∗ , . . . , en∗ ), and summing first order conditions over i, we have ∑in=1 si0 ( g(e1∗ , . . . , en∗ )) = n, contradicting budget balance ∑in=1 si ( x ) = x. • Intuition: output sharing means that each agent gets less than 100% of gain in output as result of increase in his effort. – Efficiency requires ∂g(e1∗ , . . . , en∗ )/∂ei = 1: each i exerts effort up to marginal cost equal to marginal output. – This means si0 ( g(e1∗ , . . . , en∗ )) = 1: to implement efficient effort, each i must receive all the gain in output as his share, impossible under output sharing. 19 • Budget balance and budget breaking – A simple contract achieves the first best effort profile as a Nash equilibrium: si ( x ) = si∗ if x = g( x1∗ , . . . , xn∗ ) and 0 otherwise. – This contract violates budget-balance off the path. – The budget-breaking theory of the firm: principal is someone who has no effort decision to make and who is a participant to the sharing agreement to balance the budget off the equilibrium path in order to achieve the first best. 20