You are on page 1of 21

University of Geneva

Asymmetric Information

1 / 21
University of Geneva

Asymmetric Information
A situation in which one party of a transaction has more information than
the other.
Adverse Selection : Information about “exogenous” (fixed)
characteristic. Examples:
Insurance: Insurer knows better than the insurance company about
whether she is a bad driver or sick.
Labor Markets: Job seeker knows more about own ability than
potential employer.
Corporate Finance: Managers have more information than investors
regarding the future cash flows of the firm.
Moral Hazard : Information about “endogenous” (choice variable)
characteristic. Examples:
Insurance: Insurer does not drive carefully or carry out preventative
care.
Labor Markets: Workers shirk.
Corporate Finance: Managers do not exert effort or take lot of risk.
2 / 21
University of Geneva

Moral Hazard - The Principal-Agent Problem


Basic problem in corporate finance: separation of ownership and control
The owners (shareholders) of the firm are typically not the ones who
manage it on a daily basis.
The owners (principal) delegate tasks to managers (agent)
Yet, managers have their own objective function. They may not exert
much effort, e.g. because it is costly for them.
The way to solve the problem would be to write a contract that
compensates the manager on the basis of his effort.
⇒ Unfortunately, effort is typically unobservable (moral hazard)
Hence, we write contracts that compensate the manager based on
performance, which is a noisy signal of manager’s effort.
This might be costly when the manager is risk averse, since extra
compensation is needed for the risk taken.

3 / 21
University of Geneva

Moral Hazard - Outline


Basic Setup:
A principal (e.g. shareholders) and an agent (e.g. manager) are
involved in a bilateral relationship.
The principal wants the agent to perform some task.
The agent can choose how much effort to devote to this task.
The outcome of the task is determined by a mix of effort and luck.
The principal cannot observe the agent’s effort and can only motivate
him by linking his salary to the outcome of the task (e.g. firm’s
profitability).
Timing:
1 Pricipal chooses a wage schedule which depends on the outcome.

2 Agent chooses how much effort to devote to the task.

3 Agent’s effort and luck determine the outcome.

4 Payments are made according to the proposed wage schedule.

4 / 21
University of Geneva

A simple Pricipal-Agent Model


A task has two possible monetary outcomes q ∈ {q, q̄} with q < q̄.
Agent can choose one of two effort levels: {eL , eH } with eL < eH .
Exerting effort provides a disutility to the agent.
The probability of the high outcome given effort ei for i ∈ {L, H} is:
pi = P r (q = q̄|ei )
with pL < pH , i.e. more effort ⇒ better outcomes
Principal chooses a wage schedule based on the observed outcome:

w if q = q
w (q) =
w̄ if q = q̄
Agent is risk averse and his preferences are:
U (w, e) = E [u (w, e)]
Principal is risk neutral and his preferences are:
V (w, q) = E [q − w]
5 / 21
University of Geneva

Principal-Agent Model
Suppose the principal prefers the agent to exert high effort eH .
The problem of the principal who wants the agent to exert effort eH is to
choose w (q) in order to maximize his expected utility

maximize E [q − w (q) |eH ] = pH (q̄ − w̄) + (1 − pH ) q − w
{w̄,w}

subject to:
1 Agent’s participation constraint, i.e. the agent prefers to exert high

effort than to resign:


U (w (q) , eH ) = pH u (w̄, eH ) + (1 − pH ) u (w, eH ) ≥ u (PC)
¯
where u is the agent’s reservation utility, what he gets if he resigns.
¯
2 Agent’s incentive constraint, i.e. the agent prefers to exert high
effort than low effort.
U (w (q) , eH ) ≥ U (w (q) , eL ) (IC)
6 / 21
University of Geneva

Principal-Agent Model: Complete Information Benchmark


Let’s look at the complete info benchmark first:
The principal can perfectly observe the effort level chosen by the
agent.
If principal wants to induce effort level ei , his problem becomes:

maximize pi (q̄ − w̄) + (1 − pi ) q − w
{w̄,w}

subject to pi u (w̄, ei ) + (1 − pi ) u (w, ei ) ≥ u

Under complete info, only the agent’s participation constraint matters


for the principal. Why?
Participation constraint must be binding at the optimum. Why?
Solution to this problem often referred to as first-best.

7 / 21
University of Geneva

Principal-Agent Model: Complete Information Benchmark


Letting λ be the Lagrange multiplier, F.O.C with respect to {w̄, w, λ}
require that:
∂L
= 0 ⇔ −pi + λpi uw (w̄, ei ) = 0 (1)
∂ w̄
∂L
= 0 ⇔ − (1 − pi ) + λ (1 − pi ) uw (w, ei ) = 0 (2)
∂w
∂L
= 0 ⇔ pi u (w̄, ei ) + (1 − pi ) u (w, ei ) = u (3)
∂λ
From (1) and (2) we derive that:
1 1
λ= = ⇒ w̄ = w = w∗
uw (w̄, ei ) uw (w, ei )
So the risk-neutral principal offers full insurance to the risk-averse agent.
From (3), the lowest wage that induces effort ei from the agent solves:
u (wi∗ , ei ) = u
8 / 21
University of Geneva

Principal - Agent Model: Complete Information Benchmark


The optimal wage for each choice of e is wi∗ such that u (wi∗ , ei ) = u.
But what is the optimal level of effort being implemented?
The principal chooses to induce effort e∗ = eH if and only if:

pH q̄ + (1 − pH ) q − wH∗ > pL q̄ + (1 − pL ) q − wL∗


and effort e∗ = eL if and only if

pL q̄ + (1 − pL ) q − wL∗ > pH q̄ + (1 − pH ) q − wH∗

Proposition
In the principal-agent model with observable managerial effort, the
uniquely optimal contract specifies that the manager chooses the effort e∗
that maximizes E [q|ei ] − wi∗ and pays the manager a fixed wage w∗ such
that u (w∗ , e∗ ) = u.
9 / 21
University of Geneva

Principal-Agent Model: Incomplete Information


Under incomplete information, the Lagrangian of the principal’s problem is:
L = E [q − w (q) |eH ] + . . .
λ (U (w (q) , eH ) − u) + . . .
µ (U (w (q) , eH ) − U (w (q) , eL ))
F.O.C with respect to {w̄, w} requires that:
pH [−1 + λuw (w̄, eH ) + µuw (w̄, eH )] − pL µuw (w̄, eL ) = 0
(1 − pH ) [−1 + λuw (w, eH ) + µuw (w, eH )] − (1 − pL ) µuw (w, eL ) = 0

By rearranging it is possible to show that:


Both µ and λ are positive if u is increasing and concave.
Since µ > 0 the incentive constraint binds.
Since λ > 0 the participation constraint binds.
Wages w̄, w are found by solving the two constraints.
10 / 21
University of Geneva

Optimization with Inequality Constraints

maximize f (x)
x
subject to g1 (x) ≤ c1
g2 (x) ≤ c2
Form the Lagrangian:

L = f (x) + λ1 (c1 − g1 (x)) + λ2 (c2 − g2 (x))


Necessary conditions for maximum (Kuhn-Tucker conditions)

∂L ∂L
≤0 x≥0 x =0
∂x ∂x
g1 (x) ≤ c1 λ1 ≥ 0 λ1 (c1 − g1 (x)) = 0
g2 (x) ≤ c2 λ2 ≥ 0 λ2 (c2 − g2 (x)) = 0

11 / 21
University of Geneva

Principal-Agent Model: Incomplete Information


Letting u be additively separable in w and e, i.e. u (w, e) = v (w) + c (e),
the F.O.C. become:

pH [−1 + λvw (w̄) + µvw (w̄)] − pL µvw (w̄) = 0


(1 − pH ) [−1 + λvw (w) + µvw (w)] − (1 − pL ) µvw (w) = 0
Solving we find that λ, µ > 0
pH 1 − pH
λ= + >0
vw (w̄) vw (w)
   
pL 1 1
µ 1− = (1 − pH ) − >0
pH vw (w̄) vw (w)

The wages are no longer constant ⇒ Agent is not fully insured


Principal pays more when output reveals that an action more
favorable to him is likely to have been chosen by the agent.
12 / 21
University of Geneva

Principal - Agent Model: Incomplete Information


Main conclusions with Moral Hazard:
Compared to complete info benchmark, the principal:
Pays agent more when output is high
Pays agent less when output is low
no longer provides full insurance to the agent on the risky output.

He does so to provide incentives for agent to exert effort, since a fully


insured agent would have no incentives to exert effort.
These conclusions rely on the information problem of the principal
and would hold even if the principal were risk averse.
They always hold as long as the agent is risk averse.

13 / 21
University of Geneva

Principal-Agent Model - Example 1


eL = 0, eH = 1
u (w, e) = 2w1/2 − e
u = 1, q̄ = 4, q = 0, pH = 3/4, pL = 1/4
¯
What is the optimal wage schedule w∗ and the optimal effort level e∗
under complete info?
Wages wH∗ , wL∗ are found by agent’s participation constraint.
1/2
2wH∗ − 1 = 1 ⇒ wH∗ = 1
1/2
2wL∗ − 0 = 1 ⇒ wL∗ = 1/4
Optimal effort e∗ = 1 is found by comparing principal’s expected
profits:
3 1 1 3 3
q̄ + q − wH∗ = 2 > q̄ + q − wL∗ =
4 4 4 4 4
Thus the agent is fully insured by the principal.
14 / 21
University of Geneva

Principal-Agent Model - Example 2


What is the optimal wage schedule w∗ under incomplete info?
25 1
If principal wants e∗ = 1, wages w̄H∗ = 16 and wH∗ = 16 are found
by solving IC and PC constraints:
3  1/2  1 
2w̄ − 1 + 2w1/2 − 1 = 1
4 4
3  1/2  1  1  3 
2w̄ − 1 + 2w1/2 − 1 = 2w̄1/2 + 2w1/2
4 4 4 4
If principal wants e∗ = 0, optimal contract is fixed w¯L∗ = wL∗ = wL∗
such that PC binds (IC does not bind):
1/2
2wL∗ − 0 = 1 ⇒ wL∗ = 1/4
The principal prefers e∗ = 1 since
3 1  29 1 3 3
(q̄ − w̄H∗ ) + q − wH∗ = > q̄ + q − wL∗ =
4 4 16 4 4 4
The principal cannot fully insure the agent since it would undermine
the incentives to exert effort.
15 / 21
University of Geneva

Adverse Selection - Used Cars Example (Akerlof 1970)


Buyers can’t tell the quality of a used car ⇒ they are willing to pay
an “average” price
Average price attractive for low (but not high) quality sellers (adverse
selection!)
Higher proportion of low quality products (i.e., lemons) offered
Buyers anticipate it and are willing to pay less than the average price
The proportion of good products on offer falls further...
This process may lead to a complete break down of the market

16 / 21
University of Geneva

Adverse Selection (Lemons problem) - Setup


Usual Setup:

Entrepreneur needs investment F > 0 to finance a project.

Entrepreneur has wealth W = 0.

Project yields cash flows C H = C̃ in case of success or C L = 0 in


case of failure.

Entrepreneur and investors are both risk neutral.

Entrepreneur is protected by limited liability (no punishment for


failure).
Investors are competitive (many, make zero profit)

17 / 21
University of Geneva

Adverse Selection (Lemons problem) - Setup


Differences from usual setup:
Entrepreneur can be one of two types, with different probability of
success:  
For good entrepreneur, p := P C = C̃
 
For bad entrepreneur, q := P C = C̃ , with q < p.

Information structure:
Entrepreneur has private info about type.
Investors believe a(ny) entrepreneur has probability α (1 − α) to be
good (bad).
Define m := αp + (1 − α) q, the probability of success for the investors.

Credit-worthiness:
Only the good type is credit-worthy, i.e., pC̃ > F > q C̃.
Both types are credit-worthy, i.e., pC̃ > q C̃ > F .
18 / 21
University of Geneva

Adverse Selection (Lemons problem) - Symmetric Benchmark

As a benchmark, consider case where investors know type.


The good entrepreneur would obtain financing and one optimal
contract for him would be
to get max payoff CbG in the case of success such that investors break
even  
p C̃ − CbG = F

The bad entrepreneur gets financed if (i) q C̃ > F but not if (ii)
q C̃ < F and
in (i) he would get max payoff CbB such that
 
q C̃ − CbB = F

Clearly, CbB < CbG

19 / 21
University of Geneva

Adverse Selection (Lemons problem) - Asymmetric Information

Suppose the only feasible financial contract is a ”pooling contract”


that gives both entrepreneurs a wage Cb ≥ 0 in case of success and 0
in case of failure.
Investors’ expected profits are:
   
[αp + (1 − α) q] C̃ − Cb − F = m C̃ − Cb − F

Lending not possible if mC̃ − F < 0


This can happen only if bad type is not credit-worthy, i.e. q C̃ < F .
Then, lending not possible when α < α∗ (probability of good type is
low) where α∗ solves:
[α∗ p + (1 − α∗ ) q] C̃ − F = 0
Market Breakdown!
Under-investment: good entrepreneurs not financed, even though they
have profitable projects.
Good entrepreneurs hurt by potential presence of bad entrepreneurs.
20 / 21
University of Geneva

Adverse Selection (Lemons problem) - Asymmetric Information

Suppose now that lending is possible, i.e. mC̃ − F > 0


(i.e. either (i) q C̃ > F or (ii) q C̃ < F and α ≥ α∗ )
Then Cb is such that investors break even on average:
 
m C̃ − Cb = F

This implies that, ex post, investors make


 moneyon the good
 types 
and lose money on the bad types, i.e. p C̃ − Cb > F > q C̃ − Cb .

Cross-subsidization and potential over-investment:


Good borrowers are again hurt by the presence of bad borrowers:

Cb < CbG

If q C̃ > F bad projects are now financed!

21 / 21

You might also like