Optimal efficiency-wage contracts with subjective evaluation Jimmy Chan
Bingyong Zheng
Shanghai University of Finance & Economics
Feb. 03, 2010
Introduction
1
Only 1–5% of US workers receive performance pay in the form of piece rate and commissions (McLeod and Parent 1999)
Bingyong Zheng
Optimal contracts
1/30
Introduction
1
Only 1–5% of US workers receive performance pay in the form of piece rate and commissions (McLeod and Parent 1999)
2
More common are incentives (e.g., bonus, promotion and termination) that are based on subjective evaluations and embedded in a long-term relationship.
Bingyong Zheng
Optimal contracts
1/30
Why subjective evaluation
1
Hard to separate individual contribution
2
Dysfunctional incentive from objective performance measures: rewarding A while hoping for B (Kerr 1975, AMJ) Examples (Baker, Gibbons and Murphy 1994, QJE) At H.J. Heinz company, division manager received bonus only if earning increased from the prior year. The managers delivered consistent earning growth by manipulating the timing of shipments to customers and by prepaying for services not yet received, both at some cost to the firm. In 1992, Sears abolished the commission plan in its auto-repair shop.... Mechanics misled customers into authorizing unnecessary repairs...
Bingyong Zheng
Optimal contracts
2/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003)
Bingyong Zheng
Optimal contracts
3/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003) Principal pays a bonus when performance is good.
Bingyong Zheng
Optimal contracts
3/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003) Principal pays a bonus when performance is good. Agent harms Principal when he thinks own performance is good but receives no bonus.
Bingyong Zheng
Optimal contracts
3/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003) Principal pays a bonus when performance is good. Agent harms Principal when he thinks own performance is good but receives no bonus. Efficiency loss occurs when the two disagree on performance: no bonus and sabotage.
Bingyong Zheng
Optimal contracts
3/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003) Principal pays a bonus when performance is good. Agent harms Principal when he thinks own performance is good but receives no bonus. Efficiency loss occurs when the two disagree on performance: no bonus and sabotage.
2
Efficiency wage (Levin 2003, Fuchs 2007)
Bingyong Zheng
Optimal contracts
3/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003) Principal pays a bonus when performance is good. Agent harms Principal when he thinks own performance is good but receives no bonus. Efficiency loss occurs when the two disagree on performance: no bonus and sabotage.
2
Efficiency wage (Levin 2003, Fuchs 2007) Principal pays a fixed wage.
Bingyong Zheng
Optimal contracts
3/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003) Principal pays a bonus when performance is good. Agent harms Principal when he thinks own performance is good but receives no bonus. Efficiency loss occurs when the two disagree on performance: no bonus and sabotage.
2
Efficiency wage (Levin 2003, Fuchs 2007) Principal pays a fixed wage. Agent is fired when evaluation is bad.
Bingyong Zheng
Optimal contracts
3/30
Handling the incentive problems
1
Bonus + Sabotage (MacLeod 2003) Principal pays a bonus when performance is good. Agent harms Principal when he thinks own performance is good but receives no bonus. Efficiency loss occurs when the two disagree on performance: no bonus and sabotage.
2
Efficiency wage (Levin 2003, Fuchs 2007) Principal pays a fixed wage. Agent is fired when evaluation is bad. Efficiency loss occurs when Principal fires Agent by mistakes.
Bingyong Zheng
Optimal contracts
3/30
Brief introduction of the model
Study a T-period contracting problem between risk-neutral Principal and Agent. Each period the Principal privately evaluates Agent’s performance. Agent’s self-evaluation may be positively correlated with Principal’s Incentives are provided through money burning at the end T periods. Principal wants to motivate Agent to work and to minimize money-burning cost.
Bingyong Zheng
Optimal contracts
4/30
Main findings
It is optimal to link punishment across periods When correlation is very low, punish only when bad evaluation observed every period When correlation not very low, punishment is asymmetric: Punish deteriorating performance: punish if future performance is bad even if performance were good in the past Reward improvement: Agent can make up for poor performance in the past by performing better in future
Punishment cost increases in the correlation between evaluations.
Bingyong Zheng
Optimal contracts
5/30
Contribution
Show the Abreut et al (1991) rule is optimal for low correlation case: Fuchs (2007) works only the zero correlation case; Determine the lower bound of efficiency loss for any maximum-effort contract and for any correlation of evaluations; Find an optimal T-period contract.
Bingyong Zheng
Optimal contracts
6/30
Model
There are T periods. In period 0, the Principal offers the Agent a contract. If the Agent rejects, then the game ends, and each party gets zero. If the Agent accepts, then he is hired for the next T periods. Each period, the Agent chooses et ∈ {0, 1} 1 > c = c(1) > c(0) = 0 Output is stochastic, with expected output equals et .
Bingyong Zheng
Optimal contracts
7/30
Model (cont.)
Principal observes yt ∈ {H, L}, Agent observes st ∈ {G , B} History: ht ≡ e t−1 × s t−1 Strategy: σ ≡ (σ1 , . . . , σT ), σt : H t → {0, 1} Assumption 1: π(H|1) = p > π(H|0) = q Assumption 2: π(H|1, G ) > max{π(H|1, B), π(H|0, G ), π(H|0, B)}
Bingyong Zheng
Optimal contracts
8/30
Model (cont.) An efficiency wage contract contains three parts: (B, W , Z (y T )) Fixed payment: B, W Punishment: Z (y T ) ≤ W
Payoffs for the Principal and Agent T
F (y , B) = −B − W +
T X
δ t−1 et
t=1
v (W , Z , σ) = B + W − Z (y T ) −
T X
δ t−1 c(et )|σ
t=1
B, W and Z are present value evaluated at t = 1.
Bingyong Zheng
Optimal contracts
9/30
Optimality
Principal’s problem max E
B,W ,Z ,σ
s.t.
−B − W +
T X
δ
t−1
!
et |σ ,
t=1
σ ∈ arg max v (B, W , Z T , σ), v (B, W , Z T , σ) ≥ 0.
Contract induces maximum effort if σt (ht ) = 1. Optimal maximum-effort contract is the one that minimizes E [Z (y T )] subject to σt (ht ) = 1.
Bingyong Zheng
Optimal contracts
10/30
Optimal Punishment Rule: T=1 1
One period contract
Bingyong Zheng
Optimal contracts
11/30
Optimal Punishment Rule: T=1 1
One period contract
Proposition 1 When T = 1, any contract that motivates the Agent to work has a minimum money-burning cost C (Z 1 ) = (1−p)c p−q .
Bingyong Zheng
Optimal contracts
11/30
Optimal Punishment Rule: T=1 1
One period contract
Proposition 1 When T = 1, any contract that motivates the Agent to work has a minimum money-burning cost C (Z 1 ) = (1−p)c p−q . 2
Minimizing the expected money-burning loss, C (Z 1 ) ≡ pZ 1 (H) + (1 − p)Z 1 (L),
Bingyong Zheng
Optimal contracts
11/30
Optimal Punishment Rule: T=1 1
One period contract
Proposition 1 When T = 1, any contract that motivates the Agent to work has a minimum money-burning cost C (Z 1 ) = (1−p)c p−q . 2
Minimizing the expected money-burning loss, C (Z 1 ) ≡ pZ 1 (H) + (1 − p)Z 1 (L),
3
Subject to constraint: −[pZ 1 (H) + (1 − p)Z 1 (L)] − c ≥ −[qZ 1 (H) + (1 − q)Z 1 (L)]
Bingyong Zheng
Optimal contracts
11/30
Optimal Punishment Rule: T=1 1
One period contract
Proposition 1 When T = 1, any contract that motivates the Agent to work has a minimum money-burning cost C (Z 1 ) = (1−p)c p−q . 2
Minimizing the expected money-burning loss, C (Z 1 ) ≡ pZ 1 (H) + (1 − p)Z 1 (L),
3
Subject to constraint: −[pZ 1 (H) + (1 − p)Z 1 (L)] − c ≥ −[qZ 1 (H) + (1 − q)Z 1 (L)]
4
Optimal rule Z 1 (H) = 0 and Z 1 (L) =
Bingyong Zheng
Optimal contracts
c . p−q 11/30
Low correlation
1
Let ρ = ρ = 0)
π(L|1)−π(L|1,G ) , π(L|1)
correlation coefficient (Fuchs 2007,
Bingyong Zheng
Optimal contracts
12/30
Low correlation
π(L|1)−π(L|1,G ) , π(L|1)
correlation coefficient (Fuchs 2007,
1
Let ρ = ρ = 0)
2
Optimal contract: low correlation
Bingyong Zheng
Optimal contracts
12/30
Low correlation
π(L|1)−π(L|1,G ) , π(L|1)
correlation coefficient (Fuchs 2007,
1
Let ρ = ρ = 0)
2
Optimal contract: low correlation
Proposition 2 When ρ ≤ 1 − δ, it is efficient to induce maximum effort through the strategy( 1 c if ∀ t, ytT = L p−q (1−p)T −1 b T (y T ) = Z 0 otherwise T b with money-burning cost C (Z ) = (1−p)c . p−q
Bingyong Zheng
Optimal contracts
12/30
Low correlation (cont.)
Intuition for the zero correlation case (ρ = 0): Agent faces a static problem; most hard to deter single shirking; No incentive to shirk at t = 1, no incentive for single shirking; Expected punishment is convex in the number of shirking.
Per period money-burning cost: 1 (1 − p)c . T −1 1 + δ + ... + δ p−q
Bingyong Zheng
Optimal contracts
13/30
Correlation not low: ρ > 1 − δ
Punishing only when “L” is observed every period is not optimal Learning problem: what would the Agent do if he learns that he probably has passed the Principal’s test?
Bingyong Zheng
Optimal contracts
14/30
Two-period example: ρ > 1 − δ
1
The expected loss C (Z 2 ): p 2 Z 2 (HH) + p(1 − p)[Z 2 (HL) + Z 2 (LH)] + (1 − p)2 Z 2 (LL).
2
Incentive constraints p[Z 2 (LH) − Z 2 (HH)] + (1 − p)[Z 2 (LL) − Z 2 (HL)] ≥
c , p−q
π(H|1, G )[Z 2 (HL)−Z 2 (HH)]+π(L|1, G )[Z 2 (LL)−Z 2 (LH)] ≥ 3
Other constraints are not binding.
Bingyong Zheng
Optimal contracts
15/30
δc p−q
Two-period example (cont.)
1
The optimal rule π(H|1, G ) c δ+ , Z (LL) = p−q 1−p c π(L|1, G ) 2 Z (HL) = δ− , p−q 1−p 2
2
2
Z (LH) = Z (HH) = 0.
Bingyong Zheng
Optimal contracts
16/30
Two-period example (cont.)
1
The optimal rule π(H|1, G ) c δ+ , Z (LL) = p−q 1−p c π(L|1, G ) 2 Z (HL) = δ− , p−q 1−p 2
2
2
Z (LH) = Z (HH) = 0. 2
The expected efficiency loss 2
C (Z ) =
(1 − p)c (1 − p)c (δ + ρ) = δC (Z 1 ) + ρ . p−q p−q
Bingyong Zheng
Optimal contracts
16/30
A backward induction approach T=1 H
ZH
P PP P
L PPP PP P
PP
ZL
One-period contract: ZH = 0, ZL = ZL − ZH ≥
Bingyong Zheng
c , p−q
c . p−q
Optimal contracts
17/30
A backward induction approach (cont.) T=2
H
Z 1 (H) Z 2 (HH)
X HX X XX L XXXZ 1 (L) Z 2 (HL) HH H Z 1 (H) Z 2 (LH) H H L HX X X XX
L
XXXZ 1 (L) Z 2 (LL)
(1 − p)[Z 2 (LL) − Z 2 (HL)] ≥
c , p−q
δZ 1 (L) = π(H|1, G )Z 2 (HL) + π(L|1, G )Z 2 (LL).
Bingyong Zheng
Optimal contracts
18/30
A backward induction approach (cont.) The backward induction gives the same result as Lagrangian method The efficiency loss: 2
C (Z ) = δC (Z 1 ) + ρ
(1 − p)c , p−q
Optimal rule: δZ 1 (L) =
δc p−q 2
2
=π(H|1, G )Z (HL) + π(L|1, G )Z (LL) cπ(H|1, G ) π(L|1, G ) cπ(L|1, G ) π(H|1, G ) = δ− + δ+ . p−q 1−p p−q 1−p
Bingyong Zheng
Optimal contracts
19/30
Main result: δ > 1 − ρ Efficiency-loss Lemma 1 Suppose Z T induces maximum effort in a T -period contracting game. Then C (Z T ) ≥ δC (Z T −1 ) +
ρ(1 − p)c . p−q
As C (Z 1 ) = (1 − p)c/(p − q), applying the rule repeatedly: ! T −1 X (1 − p)c δ t−1 . C (Z T ) = δ T −1 + ρ p−q t=1
Bingyong Zheng
Optimal contracts
20/30
Proof 1
For all y T −1 (Here y T −1 ≡ (y2 , . . . , yT )) Z T −1 (y T −1 ) ≡
1 [π(H|1, G )Z T (H ◦y T −1 )+π(L|1, G )Z T (L◦y T −1 )] δ (*)
Bingyong Zheng
Optimal contracts
21/30
Proof 1
For all y T −1 (Here y T −1 ≡ (y2 , . . . , yT )) Z T −1 (y T −1 ) ≡
2
1 [π(H|1, G )Z T (H ◦y T −1 )+π(L|1, G )Z T (L◦y T −1 )] δ (*)
Z T must satisfy IC at t = 1 X µ ˜(y T −1 )[Z T (L ◦ y T −1 ) − Z T (H ◦ y T −1 )] ≥ y T −1
Bingyong Zheng
Optimal contracts
21/30
c . p−q
(**)
Proof 1
For all y T −1 (Here y T −1 ≡ (y2 , . . . , yT )) Z T −1 (y T −1 ) ≡
2
1 [π(H|1, G )Z T (H ◦y T −1 )+π(L|1, G )Z T (L◦y T −1 )] δ (*)
Z T must satisfy IC at t = 1 X µ ˜(y T −1 )[Z T (L ◦ y T −1 ) − Z T (H ◦ y T −1 )] ≥ y T −1
3
c . p−q
(**)
Combining conditions (*) and (**) X µ ˜(y T −1 ) π(H|1)Z T (H ◦ y T −1 ) + π(L|1)Z T (L ◦ y T −1 ) C (Z T ) = y T −1
≥ δC (Z T −1 ) + ρ
(1 − p)c . p−q
Bingyong Zheng
Optimal contracts
21/30
Optimal rule Corollary 1 When ρ > 1 − δ, it is efficient to induce maximum effort through: 1 1 Z (L) = c/(p − q), Z (H) = 0; and T
Z (y T ) ≡ π(H|1, G ) c T −1 T −1 δZ (L )+ if y T = L ◦ LT −1 , (1 − p)T −1 p − q c π(L|1, G ) T −1 T −1 (L )− δZ if y T = H ◦ LT −1 , T −1 (1 − p) p − q T −1 T T 6= LT −1 , (y−1 ) if y−1 δZ
T The money burning cost C Z = C (Z T ). Bingyong Zheng
Optimal contracts
22/30
Implications 1
Z
T
depends only on the last time H is observed.
T
Z (y T ) = 0
if yT = H T −˜t
X δ T −1−t (1 − p)c Z (y ) = (δ + ρ − 1) , ˜t ≡ max t|yt = H p−q (1 − p)t t=1 " # T −1 T −1−t X δ 1 (1 − p)c T T + (δ + ρ − 1) Z (L ) = p−q (1 − p)T (1 − p)t T
T
t=1
Bingyong Zheng
Optimal contracts
23/30
Implications 1
Z
T
depends only on the last time H is observed.
T
Z (y T ) = 0
if yT = H T −˜t
X δ T −1−t (1 − p)c Z (y ) = (δ + ρ − 1) , ˜t ≡ max t|yt = H p−q (1 − p)t t=1 " # T −1 T −1−t X δ 1 (1 − p)c T T + (δ + ρ − 1) Z (L ) = p−q (1 − p)T (1 − p)t T
T
t=1
2
Z
T
b T as ρ → 1 − δ. converges to Z
Bingyong Zheng
Optimal contracts
23/30
Implications 1
Z
T
depends only on the last time H is observed.
T
Z (y T ) = 0
if yT = H T −˜t
X δ T −1−t (1 − p)c Z (y ) = (δ + ρ − 1) , ˜t ≡ max t|yt = H p−q (1 − p)t t=1 " # T −1 T −1−t X δ 1 (1 − p)c T T + (δ + ρ − 1) Z (L ) = p−q (1 − p)T (1 − p)t T
T
t=1
2 3
T b T as ρ → 1 − δ. Z converges to Z Minimum efficiency loss increasing in ρ,
Bingyong Zheng
Optimal contracts
23/30
Implications 1
Z
T
depends only on the last time H is observed.
T
Z (y T ) = 0
if yT = H T −˜t
X δ T −1−t (1 − p)c Z (y ) = (δ + ρ − 1) , ˜t ≡ max t|yt = H p−q (1 − p)t t=1 " # T −1 T −1−t X δ 1 (1 − p)c T T + (δ + ρ − 1) Z (L ) = p−q (1 − p)T (1 − p)t T
T
t=1
2 3
T b T as ρ → 1 − δ. Z converges to Z Minimum efficiency loss increasing in ρ, T
When ρ → 1 − δ, C (Z ) converges to a single one-period contract loss C (Z 1 ).
Bingyong Zheng
Optimal contracts
23/30
Implications 1
Z
T
depends only on the last time H is observed.
T
Z (y T ) = 0
if yT = H T −˜t
X δ T −1−t (1 − p)c Z (y ) = (δ + ρ − 1) , ˜t ≡ max t|yt = H p−q (1 − p)t t=1 " # T −1 T −1−t X δ 1 (1 − p)c T T + (δ + ρ − 1) Z (L ) = p−q (1 − p)T (1 − p)t T
T
t=1
2 3
T b T as ρ → 1 − δ. Z converges to Z Minimum efficiency loss increasing in ρ, T
When ρ → 1 − δ, C (Z ) converges to a single one-period contract loss C (Z 1 ). T When ρ → 1, C (Z ) converges to T one-period contract loss. Bingyong Zheng
Optimal contracts
23/30
Contract with infinite horizon
1
The Principal pays a fixed wage w to the Agent every period
2
The Agent exerts effort every period
3
At the end of T periods, the Principal terminates the employment contract with probability ψ(y T )
4
If the Agent is not fired, he continues working for the Principal with a clear record from period T + 1 on
Bingyong Zheng
Optimal contracts
24/30
Infinite contracting game
First review contract
?
?
...... 1
T-1
T
......
...... T+1
Bingyong Zheng
2T
Optimal contracts
25/30
∞
Infinite horizon: main result
The optimal review length T is i. Increasing in δ; ii. Decreasing in q and c; iii. Decreasing in ρ if ρ > 1 − δ, but independent of ρ when ρ ≤ 1 − δ; iv. Equal to one if p = 1.
Bingyong Zheng
Optimal contracts
26/30
Conclusion
Show robustness of results in Abreu et al. (1991), Fuchs (2007). Show optimal to link punishment decisions across periods: pooling information The optimal rule is not “fair” Efficiency wage contracts is more costly when ρ is large May not be optimal for the Principal to induce the Agent to exert maximum effort when both ρ and c are large
Bingyong Zheng
Optimal contracts
27/30
Extension: communication 1
Allowing the Principal to communicate with the Agent does not change the outcome when correlation not very high
Bingyong Zheng
Optimal contracts
28/30
Extension: communication 1
Allowing the Principal to communicate with the Agent does not change the outcome when correlation not very high
2
When T = 1, the optimal contract is c , p−q Z (H, G ) = Z (H, B) = 0. Z (L, G ) = Z (L, B) =
Bingyong Zheng
Optimal contracts
28/30
Extension: communication 1
Allowing the Principal to communicate with the Agent does not change the outcome when correlation not very high
2
When T = 1, the optimal contract is c , p−q Z (H, G ) = Z (H, B) = 0. Z (L, G ) = Z (L, B) =
3
If the minimum efficiency loss in the T period contracting game is C (Z T ), the minimum loss in the T + 1 period game is δC (Z T ) +
Bingyong Zheng
ρ(1 − p)c . p−q
Optimal contracts
28/30
Extension: communication 1
Allowing the Principal to communicate with the Agent does not change the outcome when correlation not very high
2
When T = 1, the optimal contract is c , p−q Z (H, G ) = Z (H, B) = 0. Z (L, G ) = Z (L, B) =
3
If the minimum efficiency loss in the T period contracting game is C (Z T ), the minimum loss in the T + 1 period game is δC (Z T ) +
4
ρ(1 − p)c . p−q
For T > 1, allowing communication does not change the optimal rule if low correlation: π(L|0) > π(L|1, B). Bingyong Zheng
Optimal contracts
28/30
Communication (cont.)
Allowing the Principal to communicate with the Agent may change the result if correlation is high Effective independence to get rid of learning problem: Zheng (GEB 2008), Obara (JET 2009)
Bingyong Zheng
Optimal contracts
29/30
Thank You!