Advanced Topics in Microeconomics

Markets with Asymmetric Information

Duozhe Li
Department of Economics
Chinese University of Hong Kong

Spring 2023
1 Introduction

• An important premise behind the neoclassical theory

of consumer and firm behavior is that the agents
have perfect and symmetric information.

• However, in reality, information is often asymmetri-

cally held by market participants. Consider the fol-
lowing three examples:

1. In the used-car market, the seller usually has

much better information about the quality of his
car than a potential buyer does.

2. When hiring a worker, a firm knows less about

the worker’s ability than the worker does.

3. When an automobile insurance company insures

a customer, the customer knows more than the
company about his driving skill and hence about
his probability of having an accident.
 • Questions to investigate:

1. How to characterize the market equilibria in the

presence of asymmetric information?
effineut
↓
秘 2. Are the equilibria efficient? (Recall fundamental
welfare theorems.)

3. If the equilibria are inefficient, how do market

participants adapt?

• These questions have attracted attentions of many

great economists since 1970s. Among them, George
Akerlof, Michael Spence and Joseph Stiglitz made
the most important contributions, and they were
awarded the Nobel Prize in Economics in 2001.
2 Adverse Selection

2.1 Market Failure

• Market for Lemons: Akerlof (QJE 1970)

• A simple model of the used car market:

— There are many sellers, and each owns a used car

of different quality . tandomrariable

— There are many identical buyers, and each is will-

ing to buy at most one used car.

— The quality  of each car is a random variable,

with a cumulative distribution function  (·).

— A seller has a monetary value of  from a car of

quality . 合 5 i
U looo ooo soo

$1 00 $2 0 $ 3 ooo
alur
01 $ 2 40 $ 3 60
Duyeld
0

$
.

。
↑
[Link]
d) 1
 — A buyer has a monetary value of  (  1)
from a car of quality .

— If a car of quality  is traded at a price , the

seller’s payoff is  − , and the buyer’s payoff is
 − .

— If no trade, each receives a payoff of 0.

— Assume there are more buyers than sellers, and

thus the entire trade surplus goes to the seller.

Aeune
• Complete Information

— The quality of each car is observable by both

seller and buyer.

— All cars are traded. The competitive equilibrium

price of a car of quality  is .
Ǔ
Set Rd
— Pareto efficient outcome
 • Asymmetric Information

— The quality  of each car is a random variable,

with a cumulative distribution function  (·).

— Each seller knows the quality of his car.

— The buyers do not observe the quality, but know

the distribution  (·).
私吵
↓

• At market price , only sellers with  ≤  are willing

to sell. Thus, the expected quality of the cars on
market is:

̄ () =  ( |  ≤ ) 
For buyer, they are willing to purchase at  iff
Zfd 21.4 , $ 4 0 0  ≤ $2 8 0 $ [Link]
 · ̄ () 
0

Tyne Rceker

[Link] $ 1 6 0 $3 [Link]
,
$ 4N 胀 了⼼ 》 。
20 0

⼈ Typecusen
• Definition A competitive equilibrium is a price ∗
○
such that ∗ =  ( |  ≤ ∗) 
wi 成 ⼤ 2000 [Link]

[Link]? [Link]
EP 共 3
、

⼈
Equihinmipnzo 、

0 0
selu 1
0
Zoo : 0 0 0
nm ↑

huyesi Exmtdvalw
-1800C 2000
[Link]
[Link] Ml
• Example Let  be uniformly distributed on [0 1] and

exnect [Link] lerakedoiknowquali.y

 = 32. From
3
∗ =  ( |  ≤ ∗) = ∗
4
we find that the only competitive equilibrium price
is ∗ = 0. No trade at all!

Compared to the complete information market, the

equilibrium outcome here is clearly inefficient. This
phenomenon is called “market failure”.

• Incomplete but symmetric information ,

— The quality of each car is unobservable by both

seller and buyer.

— In the competitive equilibrium, all cars are traded

at the same price  ().

wtinompletereyii [Link]
— It is information asymmetry that causes market
failure.

200 2400 360 。

4
p = 2 40 0 .
• Akerlof (QJE 1970) analyzed the used car market as
an example to illustrate the effects of information
asymmetry on market outcomes. The same insights
can be obtained in any market with asymmetric in-
formation, for example, labor market and insurance
market.

• How would market participants react to market fail-

ure caused by information asymmetry? Next, we
consider two types of models:

— Signaling model: Informed parties find ways to

signal to the uninformed parties his private infor-
mation through observable actions.

— Screening model: Uninformed parties may design

mechanism to distinguish, or screen the informed
parties with differing information.
2.2 Job Market Signaling
• Spence (QJE 1973)

• A simple model of job market:

— Competitive market: many firms are hiring.

— There are two types of workers with high or low

productivity:  =  or  (    0). The
fraction of type  workers is  ∈ (0 1).

— Productivity (type) is worker’s private informa-

tion, unobservable by firms.

— Worker’s reservation value is zero, that is, he

would choose to work at any positive wage.

• Remark The productivity of a worker can be viewed

as the units of output he can produce each day, where
the price of the output is normalized to 1. This way,
we are assuming that firms use labor as the only input
and the technology has constant return to scale.
 • Education as a signaling device:

— Although a worker’s productivity is not observ-

able by firms, his education level is. Hence, a
worker can “signal” his productivity by choosing
a certain level of education.
laeast ,

( (e) ↓— Education is costly. The cost function  ( ) is

"
assumed to be twice continuously differentiable,

↑
""

with  (0 ) = 0,  ( )  0,  ( ) ≥ 0,

摩
 ( )  0 and  ( )  0. The crucial

ghi-jepnoducti -[Link]. n
part is that both the cost and the marginal cost
of education is lower for type  worker.

— For simplicity, we also assume that education

does not affect productivity. (Not crucial.)
-7
noneededucation
• Job Market Signaling Game

— Nature selects a worker. With probability , the

worker is type , and with probability 1 − , the
worker is type .

— The worker chooses the level of education .

— Two firms, after observing , simultaneously offer

wage 1 and 2 for the worker to choose.

— The worker accepts one wage offer, the higher

one obviously. In case of indifference (1 = 2),
the worker randomizes with equal probability.
Nature

• Payoff function
YM
— worker: y※! 週
煎 恐
 (  | ) =  −  (家
⼀

) , where  ∈ { } ;

— firm:  (  | ) =  [ | ]−, where  [ | ]

is the firm’s expectation of  after observing .
Thus, firm’s belief is relevant.
• Worker’s indifference curve [Link]
ū

w Direction of
[Link]
eikā
Increasing Utility OEEHB
Rguie
[Link] ompeusation I

[Link] w*

Type L
Type H

www. e*
e

— The indifference curves of the two types has one

intersection point, at which the type ’s indiffer-
ence curve has a smaller slope. This property of
preferences, known as the single crossing prop-
erty, plays an important role in the analysis of
signaling model and in models of asymmetric in-
formation more generally.

— It arises because the worker’s marginal rate of

substitution between wage and education at given
( ) is ()̄ =  ( ) which is decreas-
ing with  because  ( )  0 by assumption.
• Perfect Bayesian equilibrium in pure strategies:

— Assessment: strategy profile and belief system.

— WPBE: sequential rationality, weak consistency.

sameìnfomationy
— Extra condition: firms have the same belief about
the type of a worker who has chosen .
⼈
[Link]
lf yonau v1
,

• Separating equilibria: the two types of workers

choose different levels of education.

— In equilibrium wage is equal to productivity. (Why?)

— Type  worker chooses zero education. (Why?)

• Pooling equilibria: the two types choose the same

education level and thus receive the same equilibrium
wage:  + (1 − ) 

• In the rest of the analysis, we use the following simple

form of cost function:  ( ) = ,  ∈ { }.
 • Separating equilibria

PUM — Worker’s strategy:  = 0 and  = ∗  0.

ii
⼋
— Firms’ belief: a worker is type  with probabil-
ity 1 if  ≥ ∗, and a worker is type  with
probability 1 if   ∗.
i.

iEtyre
— Firms’ strategy: each firm offers a wage  to a
worker with education  ≥ ∗, and offers a wage
 to a worker with education   ∗.
Butwhatdt ? !!
nianyeelg
nowge
— It should be optimal for type  to choose  = 0: ìuweua

[Link]#needthehighenoyhtpreventdeviionlLgeY
-
deùatio 、

 ≥  − ∗ ⇔ ∗ ≥  ( − ) 
where  is type  worker’s equilibrium payoff,
and ( − ∗) would be his payoff if he devi-
ates by choosing ∗ (i.e., to mimic type ). In
other words, the required education ∗ for receiv-
ing high wage  cannot be too low; otherwise,
type  would deviate by mimicking type .
 — It should be optimal for type  to choose ∗:
Hightech
denf Hudg (e)

 − ∗ ≥  ⇔ ∗ ≤  ( − ) 
where ( − ∗) is type ’s equilibrium pay-
off and  would be his payoff if he deviates by
choosing  = 0 (i.e., to mimic type ). In other
words, the required education ∗ for receiving
high wage  cannot be too high; otherwise, type
 would not find it optimal to distinguish them-
selves from type . If 1 L soìnt 999

[Link]
— Thus, the above assessment is an equilibrium for

 ( − ) ≤ ∗ ≤  ( − ) 

[Link]
1 wce ,

[Link]
• Pooling equilibria
ii)
i

o_o
— Worker’s strategy:  =  = ∗ ≥ 0. ↑

[Link]
— Firms’ belief: a worker is type  ( resp.) with
probability  (1− resp.) if  ≥ ∗, and a worker
is type  with probability 1 if   ∗.

Ntl 肌 ⼀

[Link]
— Firms’ strategy: each firm offers a wage  +
(1 − )  to a worker with education  ≥ ∗,
and offers a wage  to those with   ∗

— It should be optimal for type  to choose ∗:

-
 + (1 − )  − ∗ ≥ 
[Link]
⇔ ∗ ≤  ( − )  nnothetoohigh
,

where [ + (1 − )  − ∗] is type ’s equi-

librium payoff, and  is his payoff if he deviates
by choosing  = 0.

— Similarly, it should also be optimal for type  to

choose ∗:
 + (1 − )  − ∗ ≥ 
⇔ ∗ ≤  ( − ) 
where [ + (1 − )  − ∗] is type ’s equi-
librium payoff, and  is his payoff if he deviates
by choosing  = 0.

— Thus, the above assessment is an equilibrium for

any ∗ ≤  ( − ) 
• Graphical Illustration
Indifference Curve
of Type L worker
w Hwill mot study

lontandaayatihum
⼀

,
⽣⼀
Em λ
-
andget L .

H … @
…
, ⼀
…
λH ε ( l -
λ 1⼉ ⼀

…
Indifference Curve
of Type H worker ( flatter )
 H  1   L

*
e

0 |
e
: L(H-L)
1
H(H-L)

Separating Equilibria
Indifference Curve
of Type L worker
w

Indifference Curve
of Type H worker

H  1   L

L 000000

0 e
L H  L 
asteter
edu
f max
set
,
-

e
you car
.

Pooling Equilibria
• Refinement by Intuitive Criterion (Optional)

箭
…ff
。

w
" …

… *
H P (Llere ) 1

“
=

㟡 ant
⼆
satisfy -

H  1   L

L a

e
0
e’ e*
[C
f dont satisfy ,

Consider a separating equilibrium (as in the graph):

— Strategy profile:

∗ Worker:  = 0,  = ∗;

∗ Firms:  () =  iff   ∗.

— Belief system:  ( | ) = 1 iff   ∗.

— Equilibrium payoffs: ∗ = , ∗ =  − ∗.

In this equilibrium, if a worker chooses an education
level 0 ( ( − )  0  ∗), the firms’ common
belief is that he is a low type with probability 1, and
thus a wage of  will be offered.

Is such a belief reasonable?

Intuitive Criterion By deviating to 0, the low type

worker’s payoff is at most 0 =  − 0 (if he
receives a wage of ), which is lower than his equi-
librium payoff ∗. In other words, a low type worker
would definitely be worse off by making such a devi-
ation. However, a high type worker could be better
off from this deviation provided that he could receive
a wage of . Therefore, after observing 0, a rea-
sonable belief should assign probability 0 to the fact
that the worker is a low type.

Any separating equilibrium with ∗   ( − )

can be eliminated by intuitive criterion. The sepa-
rating equilibrium with ∗ =  ( − ) satisfies the
intuitive criterion.
 • Similarly, all the pooling equilibria can be eliminated
by the intuitive criterion.
w

P( L 1 e) = 1λ H
…
Ve 三 o H  1   L

e
0
LH  L  e’

Orcan I hehette . offf deviate !

Consider the pooling equilibrium in which both types

choose ∗ = 0. After observing a deviation to 0 (see
graph), a reasonable belief for the firms is that the
worker is a high type with probability 1 because a low
type worker would definitely be worse off (compared
with equilibrium payoff) by making such a deviation.
2.3 Screening

• Rothschild and Stiglitz (QJE 1976)

• A simple model of auto insurance market:

— Competitive market: many companies.

— Each customer has an initial wealth  , and with

positive probability an accident may occur, in
which case his wealth becomes  − .

— There are two types of customers, namely, low-

risk and high-risk customers, with accident prob-
abilities  and  (0      1).

— The fraction of low risk customers is  ∈ (0 1).

— An insurance contract  = ( ) specifies the

premium  that a customer has to pay, and the
payment  that he receives when accident oc-
curs.
— By purchasing the contract ( ), a customer’s
wealth is 0 = ( − ) without an accident,
and 1 = ( −  −  + ) with an accident.

— Assume that customers are risk averse. A type 

( ∈ { }) customer’s expected utility from a
contract ( ) is:

 ( ) =  ·  ( −  −  + )
+ (1 − ) ·  ( − ) 
where  (·) is assumed to be continuous, differ-
entiable and strictly concave.
f visk averse

— Assume that the insurance companies are risk

neutral, and that they are only concerned with
expected profits. A contract ( ) sold to a
type  customer brings a company an expected
profit of:

 (  | ) =  − 
• Indifference curves
ß Direction of
Increasing Utility

等:
^
df =
preminm ^
High-risk customer ( pa )
β … …

fr = oupenlation δ ,

(eeie )
Low-risk customer

,
a

d β↑ 10
Single Crossing If
[ ld ^ ). cusbms
Re need finaeau
→
-

High .r

• Break-even (zero-profit) lines ketter

tha β is e
u ,

ß
Low-risk consum er High-risk consum er
45º

a
0

 −  = 0 for  ∈ { }

β -

&
[ d : sbe 31

istre .
• Insurance Screening Game

— Two companies announce simultaneously a list

(menu) of insurance contracts. Without loss of
generality, we may restrict companies to lists with
at most two contracts.

— Nature chooses a customer. With probability ,

the chosen customer has low risk, and with prob-
ability 1 − , he has high risk.

— The customer then chooses whether to purchase

a contract, and if yes, which one. In case of in-
difference between contracts, the customer ran-
domizes with equal probabilities.

• Subgame perfect equilibrium in pure strategies

— In an equilibrium, each customer purchases the

best contract available, and the chosen contract
makes the customer better off than no insurance.
 — No company has a profitable deviation, say, to
withdraw a contract that is losing money or to
offer a new contract that can improve profit.

• If the customer’s type is observable by firms, the

firms’ contract offers can be conditioned on the type.

• Proposition If type is observable, the game has a

unique SPE, in which both companies offer the ac-
tuarially fair contract ( ) to a type  ∈ { }
customer, which will be accepted.

— All companies make zero profits.

↓
— All customers are completely insured, i.e.,  = 
and thus 0 = 1.
bulfore pile
ß
onperiatibn
Same
ß

宫
y
D ∵

tyneij
D

unobsevable

a a
0 0

PcD PiiD

( ( de )
detle
w

Pe ulw p 1 Pe )
-

t u
-

max
-

d -

c ,
☆

s -

t .

+.
d
=µ Be ,

→ morx
e u (r -

p -

l tl - pe )
utneeifr Pe β )
e

βt
w -是
] =
F , 1 - pe ) Pe in' ( pe ] pe - Pe [ -pe ) a
.
. Btl 1
)

'
-

O -
1

c (
r

Be Icoi kw PeBeJ
'
- Be Lhpe )
(
'
ow
u
-

β eas .
• If the customer’s type is unobservable by firms,
the two actuarially fair complete insurance contracts
cannot be sustained in an equilibrium because the
high-risk type will purchase the contract designed
for the low-risk type.

• We consider two types of equilibria:

— Pooling equilibria: both types of customers pur-

chase the same contract  ∗.

— Separating equilibria: different types purchase

different contracts,  and  respectively. In
the equilibrium, the following incentive com-
patibility constraint must be satisfied:

 () ≥  ( ) 
 ( ) ≥  ()  mala
C
high riik
peepe choose high
Priedicxrimination risk schame hinsalf .

• Claim 1 In any equilibrium, pooling or separating,

both companies must make zero profit.
 • Claim 2 No pooling equilibria exist.

Proof. By Claim 1, a pooling equilibrium contract

 ∗ lies on the pooling break-even line  − ̄ = 0,
where ̄ =  + (1 − )  .

Then, there is a deviating contract  0, which at-

tracts only the low-risk type and yields positive profit.

ß Low-risk break-even line

β 2
o 。

Pooling break-even line

beak ee lie
Hishaisl
.

lae -

β ctoQ _

*
dlcpp
戏

(d
β
C* λ -

1: β ) + |1 λ ) (d
-
-
Pa β ) = 0

( rositie
I pufit. . pB
P =λ P Q
=

( 1 . λ] PH Deviating contract C’ a
1

-
,

λ Fractionof
qx 3 k
Ton -

1 i onsumer

The low-risk type prefers  0 to  ∗, and the high-risk

,

type prefers  ∗ to  0. As  0 is under the low-risk

break-even line, such a contract sold to a low-risk
type yields positive profit.
 Clain 1 : ovevall 2 ero
protit
• Claim 3 If  = (  ) and  = (    )
are the contracts purchased by low-risk and high-
risk customers in a separating equilibrium, then both
contracts yield zero profit, i.e.,  =  .

Proof. Suppose that  yields positive profit, then

a company can deviate to offering a single contract
 0 = (   + ), which will attract the high-
 
risk type and yield positive profit for sufficiently small
  0. This contradicts Claim 1.

Suppose that  yields positive profit. Then 

must lie on or above the high-risk type’s indiffer-
ence curve passing through . There is a deviating
contract  0 that attracts only the low-risk type and
yields positive profit.
Low-risk break-even line
UH ( ( r 1 ) ÷ V ( )
ß (

High-risk break-even line

CH

CL

Deviating contract C’

a
• Claim 4 In any separating equilibrium, the high-
risk type purchases the complete insurance contract
 ∗ = (  ), the same as when type is the

observable.

Proof. In an separating equilibrium, the contract for

high-risk type lies on the high-risk break-even line.
If the equilibrium contract is some  6=  ∗ , then
the high-risk break-even line intersects the high-risk
type’s indifference curve (red dashed). Then there
is a deviating contract  0, which attracts only the
high-risk type and lies below the high-risk break-even
line (yielding positive profit).
Low-risk break-even line
ß
High-risk break-even line

D C H* ( mnusttheal he

CH Deviating contract C’

' s cm (+ ici
i .

a
 * *
Ue ( ( vi ) Uv . ( ( i
= 1

uc ( (c ) Uu ( cn 1 *

• Claim 5 In any separating equilibrium, the low-risk

type purchases a contract  ∗ = (   ), where
 
∗ and  ∗ .
the high-risk type is indifferent between  
That is,  (  ) =  (  ) 
Proof. In a separating equilibrium, the high-risk
type purchases  ∗ , and the low-risk type’s contract
must lie on the low-risk break-even line and on or
below the high-risk type’s indifference curve. If the
0 ∗ , there exists
low-risk type’s contract is  6= 
a deviating contract that attracts only low-risk type
and lies below the low-risk break-even line (i.e., yield-
ing positive profit). Thus, the only possible equilib-
rium contract for the low-risk type is  ∗.
) *
lu ( ( )
Ur ( ( ut = c

ß
( Butmn nod randomice
and my Cntl
*
C H
D

”
CL*

Deviating contract pufit )

(posifie ,

CL' lonly affacf

a
)
lon oinkcustmes
Lou nople
fully insured ,

Imefficieand becanse

risk avere risk

peepe taliy ( de > d )
• Proposition In any equilibrium, the high-risk type
∗ , the low-risk type purchase  ∗ , and
purchases  
the companies make zero profit.

• An equilibrium may not exist.

oflow If
lowpItfaotion fraction of
ß
is high ↓ow
d enogh
ho equilihrinm
hae equilituim
*
,

C H
D
⼩
*
C L

Deviating contract

If the pooling break-even line intersects the low-risk

type’s indifference curve, then there is a deviating
contract that attracts both types and lies below the
pooling break-even line (yielding positive profit). This
is possible when the fraction of the low-risk cus-
tomers is sufficiently high.
• Welfare analysis

— Equilibrium outcome is Pareto inefficient.

— The presence of the high-risk type exerts a neg-

ative externality on the low-risk type.

• Circumvent the nonexistence problem

— One approach is to establish existence by allowing

mixed strategies.

— Another approach is to consider a different equi-

librium concept that takes into consideration the
firms’ dynamic reaction to new contracts. The
key idea is that when a new contract is offered,
if an existing contract becomes unprofitable, it
will be withdrawn from the market, and then the
new contract may not be profitable any more.
3 Principal-Agent Problem

• Difference on Timing In the adverse selection model,

including signaling and screening, asymmetry of in-
formation exists at the time of contracting. In the
principal-agent problem, asymmetry of information
develops after the signing of the contract.

• A broad range of economic relationships can fit into

the principal-agent framework.

— The owner of a factory, after hiring a manager,

may be unable to observe how much effort the
manager puts into the job.

— The insurance company cannot observe how

much care that the insured customers exercise.

— The bank cannot observe whether the borrower

uses the loaned funds for the purpose for which
the loan was granted.
 — The manufacturer cannot observe the market
conditions faced by the distributor.

• Anticipating the development of such information

asymmetries, the contracting parties seek to design
a contract that mitigates the difficulties they cause.

• The literature distinguished between two types of

informational problems, namely, hidden action (or,
moral hazard) and hidden information. Take the
owner-manager relationship as an example,

1. In the hidden action case, the owner cannot ob-

serve how hard the manager is working.

2. In the hidden information case, the manager pos-

sesses superior information about the firm’s op-
portunity.

• Due to time constraint, in the rest of this section,

we will only study the moral hazard problem.
• The owner (the principal) of a firm wishes to hire
a manager (the agent) for a one-time project. The
project’s profits are affected by the manager’s ac-
tions, say, the levels of managerial effort.

• For simplicity, assume there are only two levels of

effort:
 ∈  = {  } 

• Assume that the profit  has finitely many possibil-

ities, i.e.,
 ∈ {1   }
with  1  2      

• Denote the probability distribution of  conditional

on the effort choice of  as

p () = (1 ()    ())

 feffert

eftorepretbabilis
τu 2 π n . ? … >π2 )π .

, 2i
π

, π . .
hu
,

pla 1
去
+
t
…

↓ ↓ pler ) ⼀ +

• Two assumptions about the distributions:

(1) For all ,  ( )  0 and  ()  0. That

is, every  can arise following any choice of effort.

(2) Monotone likelihood ratio property (MLRP):

the likelihood ratio  ()  ( ) is decreasing
with  (recall    +1). ↓
. plevi ) 8:
stronger → <
→ no af

The MLRP implies first order stochastic dominance,

100
but not vice versa. π 200 300 400

Flx ) dominae G (
x )
ec G 年年主它
FOs 1D F (x ) ≡ G ( ) Vx
x 0 ⾔的云
CM F
• The manager is an expected utility maximizer with
^
⾔产 ↑ 、

the vNM utility function  ( ) =  () −  (),

where  denotes the wage:  0 ()  0 and  00 () ≤
0, i.e., the manager is risk averse;  () denotes the
manager’s disutility of effort:  ( )   ().

• The owner is assumed to be risk neutral, and thus,

his objective is to maximize his expected return, i.e.,
the expected profit from the project less the expected
wage payment.
Observable Effort Choice Fixe .

,
• The owner makes a contract offer to the manager,
fPoustant
who can accept or reject. The optimal contract for
the owner is the solution to the following problem:

0
X
max  () ( −  ())
=1 
∈ {( )}=1
I.
X
o
heou
we by an btal pebabilidy,
s.t.  ()  ( ( )) −  () ≥ ̄
=1  f dintili所 of effort .

where ̄ is the manager’s reservation utility level.

The constraint is usually referred to as the partici-
pation constraint.

• We can tackle this problem in two steps:

Step 1. Fix  and specify the optimal wage scheme
by solving
Taiuane
X
min  ()  ( ) s.t.
=1 
{( )}=1 。

X
 ()  ( ()) −  () ≥ ̄
=1 

史
(ulal , . ,
ulanl :
, ple )
)

ulilwantrfpieliheailifey
區

=
vln ( a . ) "
∵
0 =
v wlars )
( "

≈ …
= Wlulaml 1
' EJucdlusludeuf
-

Pnle ) +γ Pnle ) v ( w ( am ) ) = O 內 witm a .

VOm
rYwhhml
' areresuagsbwttalik
.

=
|
- ik herthrnl risk
,
manafortith
The constraint must be binding; thus the FOC is ω= 1
"
(w ) 瓜
,

e
= -

e ,

u ,

− () +  ()  0 ( ()) = 0 gle <

)
e

h i−1 π . = 5π z
=5
0
 ( ()) =  宁
,
c = 1

e σt 2 =

年
where  is Lagrangian multiplier. ⼗⽂爬 - i
Impumnt -

ti 」 W 迁
)

= ψ ,

If the manager is strictly risk averse ( 00 ()  0), - ea ⽑

wn
:

:
1

Ʃ5

the FOC implies that the optimal wage scheme is a

constant. The manager receives thegiveconstant
hion
wage
wtility t

∗ such that  (∗) −  () = ̄, that is, ∗ =

=

 −1 (̄ +  ())regardless of the realized profit.

Risk sharing: The risk neutral owner fully insures
the risk averse manager. ECR ⼆是 5 +÷× 50 -
nlen
× 4
E =

| : ⽟⼀⼗六 x - 25
080505 =

The wage contract explicitly specifies the effort choice, 女

,

which is observable or verifiable. A sufficiently large

penalty can be imposed on the manager if he breaches
the contract by not exerting the required effort.
Step 2. Pin down the optimal . The owner chooses
between  and  to maximize his expected profit
less wage payments:
X
 ()  −  −1 (̄ +  ()) .
=1  
Unobservable Effort Choice yonpor onstant wage
If
arth
inderendet a

hey win dewae ,

• The outcome in the full observability case is Pareto

efficient, or, the first best.

• With unobservable effort, inefficiency is caused by

the conflict between implementing the efficient effort
choice and achieving the risk sharing between the
owner and the manager. To highlight this point, we
first study the case with a risk-neutral manager.

• Proposition In the principal-agent model with unob-

servable managerial effort and a risk-neutral manager
(i.e.,  () = ), an optimal contract generates the
same effort choice and expected utilities for the man-
ager and the owner as when effort is observable.
Proof. Consider the contract  ( ) =  − 
where  is some constant. This contract can be
interpreted as selling the project to the manager at
the price .

If the manager accepts the contract, he chooses  to

maximize his expected utility:
X
。

X
Ionrtant
 ()  ( )− () =  ()  −− () 
 
and the optimal choice of ∗ would be the same as
in Step 2 of the observable effort case.
Ifd tolaye , manoye
The manager is willing to accept the contract iff nonl
aaept
.
X
 (∗)  −  −  (∗) ≥ ̄ (∗)
=1  

Choose ∗ such that (∗) holds with equality. Then,

with  () =  − ∗, the manager’s expected
utility is ̄, and the owner’s utility is
X
∗ = =1
 (∗)  −  (∗) − ̄


the same as in the full observability case.

v( n )
<
w
 • A risk-averse manager
The optimal contract for implementing ∗ solves (A):
X
min  (∗)  ( ) s.t.
=1  
{( )}=1

X
(i)  (∗)  ( ( )) −  (∗) ≥ ̄
=1  

X
(ii) ∗ solves max  ()  ( ()) −  () 
=1 
∈
where (ii) is known as the incentive constraint.
Case 1. Implementing . The owner optimally
offers the manager the fixed wage payment
 −1 (̄ +  ()) 
the same as in the full observability case.
Case 2. Implementing  . The incentive constraint
can be rewritten as: 瓶 * 4 _ eri 三弄⼆ 4≥
千
-
s
.
X
 ( )  ( ()) −  ( ) ≥
=1  
X
 ( )  ( ()) −  () 
=1  

denialian
higher prhent
f

hed
,

Y ( ulail .
.
leul
iulauli i µ = -
是

Rlerelwtanleh Ʃ piletulum)
- gcem
)
-
a
tM [ Pnenlu [ ulanl ) - s ( en ilelvulam egleil
)
1 )
max f (x .
… ,
xn )
) ÷
s. . ,
( xm [
g
x . …
) ,
t

gim ( x , … yixn ) ≤ cm

I
,
(
kuidii tm ) f ( x my
t "
xjlgilxitn
) j)
4 , . =
)
,

&

KKτ :
X
(

.… xni
λ
,
gbnt: o . ( i = 1 . …
^
)
.

CjOI
λ; ≥ 0
, g; ( x "
,
xn )
Ecj , λ
j[ gi ( x. . i xal -

=
 The Kuhn-Tucker condition for every  ( ):

−  ( ) +  ( )  0 ( ( )) +
h      i
0 0
  ( )  ( ( )) −  ()  ( ( )) = 0
neyuines deareasing
" #
.

Ml γ>
1  ()-
or, 0 =+ 1− 
( Bindirl  ( ( ))  ( )

Claim In any solution to (A) with  =  , both

constraints are binding, i.e.,   0 and   0.

Proof. If  = 0, then  () is a constant, under

which the manager would choose .

Suppose  = 0. For some , we have  () 

 ( ) and then  0 ( ())  0, impossible.
∴~ l .

• Under the MLRP assumption, the wage is increasing

with the realized profit, which is intuitive.

• To give the manager an expected utility of ̄, the

expected wage is strictly higher than the fixed wage
payment in the full observability case. (Why?)

CPay the risk preminml .

 Ru
Rc
灣
= 5

ec = 1 号 g ( ec )
e 1

ev 1 = 2
屰 它 glem )
n
c
4

• Welfare analysis :
*

Ui 4 wnt
:
25 Iuplemenl eaii
m ax µ w
. 是 wi

S . f ,
µ 瓜年瓜 glel 是 ( :
~

北瓜 瓜 - gleu 尽它瓜 。 市品 Sler

)
) 4

binadiq
1. When  would be the optimal effort choice if 年以 : ω
,
中
ASSmme

effort were observable, then it still is when effortElalqm K )

=

is unobservable. In this case, nonobservability ↑

causes no loss. 云( i 年( 步 5 -

1+ 0

2. In contrast, when  would be the optimal ef-

fort choice if effort were observable, then one
of the two things may happen: it may be opti-
mal to implement  using an incentive scheme
that faces the manager with risk; alternatively,
the risk-bearing costs may be so high that the
owner finds it optimal to implement . In ei-
ther case, nonobservability causes a welfare loss
to the owner (the manager’s expected utility is ̄
in either case).

• Remark Nonobservability leads only to downward

distortions in the manager’s effort. This is a spe-
cial feature of the two-effort-level specification. With
many possible effort choices, the distortion caused by
nonobservability can be upward as well as downward.
• More general settings

(1) The profit  is a continuous variable, distrib-

uted on [  ] by conditional density function
 ( | ), with  ( | )  0 for all  and . The
MLRP becomes

[ ( | )  ( |  )] is decreasing in .
With two choices of effort, the analysis remains the
P
same, except that in the objective function is re-
R
placed by .

(2) General specification of the effort choice set .

Break up the owner’s problem into three parts:

1. What effort levels can be induced?

2. What is the optimal contract to implement each

specific effort level  ∈ ?

3. What effort level is optimal?

 e
.
lc , es . 立 vCuclt
t: r lr 之 vlwnl
u 3 污
i vl
( wcy
)
-

π .
a 时 ! ÷ v(
ivlwalyz) +÷ ( wnl - 3 三 v ( wc
wc
) r

g ( ar )
= 4

en ⾔号 glem 1 :
3 IcctI ( 2
=
) Nlwc ) trlun 1 6 ≥ V (wc ) + V (ui )
-

6 ≡
sler ) 5
GImosnible
-

的立 1
=
-

em
.

(2a) Multiple but finitely many effort choices, for mediam

ec 号产 Contraat
.

 = {1   } 
Part 1: Not every effort level can be implemented.
infinily may Coustraint
Part 2: To implement a specific ∗ ∈ , the fol-
lowing incentive constraint is needed:
Z
∗ solves max  ( ())  ( | )  −  () 
∈
(∗)
which consists of ( − 1) constraints.
(2b)  is a continuous set, say,  = [0 ̄].
To implement a specific ∗ ∈ , we have an infin-
ity of incentive constraints contained in (∗). One
trick, the so-called the first-order approach, is to re-
place the incentive constraint (∗) with a first-order
condition:
Z
 ( ())  ( | )  −  0 () = 0
Then, we can derive the condition for  ():
" #
1  ( | )
=+ 
 0 ( ())  ( | )
 Ru = 0
km 1600 Rn =
4000
oimplement
=

Oc = (
O . evs
4 0
. O 1

en = 4
0 - 1 0 .

5 0 -

4 .
Cwhen effort is unobereblel
2
l 4
Sle ) = e o O 5 O

JWe
wm
max
-

wH wu
-
- .
.
.

= 24
S. 0 40 Tum .4 - lu :
4 to M

下
. 1 瓜 5

)
.
t

( =
v
<④= 625
w ≥0 4m + 0 5 名
W
_ . m
.

原 le
.
to 1 . u- ,

=
24 ,

a
The condition that ratio [ ( | )  ( | )] being
increasing in  is the differential version of MLRP.
CwciwmiwuijrinI - .
I
o cto Jo 0 ( , 0 . Jmy - 40 ) +µ ( . Mcto . Jom
+ 如
5

wnl .
4 1

lwe to tun . V t
5

Co . to 4
1

=
40 4 Fn .

16 -

However, the solution obtained from this substitution

.

0 . 4 瓜 。
-

Q 5 J
座 H

O .

1 Fv 1 + 1 )

is not necessarily a valid solution. This is because the 业

O 3 瓜- 15 ]
( O 3 原M
.

µ .

agent’s objective function is not necessarily concave (et uc

= T瓜
(

km =

巫
in , and thus, FOC is only a necessary but not a UNE { NH ,

sufficient condition.
Part 3: Nonobservability can lead to upward effort
distortion. Also, at the optimal contract we can have
both inefficient effort choice and inefficiencies result-
ing from managerial risk bearing.

• Extensions

— Many agents and relative performance evaluation

— Common agency: a single agent hired simultane-

ously by several principals

— Repeated relationship and long term contract

— Multidimensional effort
 wyr - o
3 e
^ M
: 0

z熟
: 0
umeoisr =
⼦ Gm

8 mvlto 4 zto 3
. =
,

0 .
. .

µ= 0

{
o . lant 6
. jumlo . 46 H
= 4

O .
} UH - Or 3 UL τ 15 ,
 The condition that ratio [ ( | )  ( | )] being
increasing in  is the differential version of MLRP.
However, the solution obtained from this substitution
is not necessarily a valid solution. This is because the
agent’s objective function is not necessarily concave
in , and thus, FOC is only a necessary but not a
sufficient condition.
Part 3: Nonobservability can lead to upward effort
distortion. Also, at the optimal contract we can have
both inefficient effort choice and inefficiencies result-
ing from managerial risk bearing.

• Extensions

— Many agents and relative performance evaluation

— Common agency: a single agent hired simultane-

ously by several principals

— Repeated relationship and long term contract

— Multidimensional effort