Microeconomics IV

Section 4
Aleksei Beliaev
April 6, 2023

1 Exam 2015, Q3
A principal (P) contracts with an agent (A) to perform a task from which P will derive a
benefit B. After the contract is signed, A privately learns whether his cost of performing the
task is L or H, where L < H < B. The prior probability that the cost will be H is q, where
0 < q < 1. After learning his cost, A (strategically) chooses whether to report that his cost is
L or H. The contract specifies, as a function of A’s report, whether or not A performs the task
and, if he does, what he is paid. (If he does not, then he is paid nothing.) After A learns his
cost, he can if he wishes simply quit, receiving a utility of zero.

(a) For a contract that causes the task to be performed if and only if A’s cost is L, what is
the lowest payment the contract can offer to A if he reports L? (If A reports H, he will receive
a payment of zero.) Call such a contract with this payment “separate”.

Solution. The “separate” contract prescribes that, if the report is L, the task is performed
for a payment of T , and if the report is H, the task is not performed.
Suppose c = L. If he reports L, he receives u = T − L, and if he reports H, he receives
u = 0. Thus, A with cost c = L performs the task if and only if T − L ≥ 0, or T ≥ L.
Suppose c = H. If he reports L, he receives u = T − H, and he reports H, he receives
u = 0. Thus, A with cost c = H does not perform the task if and only if T − H ≤ 0, or T ≥ H.
Therefore, the sought condition is L ≤ T ≤ H, and the lowest value is Ts = L.

(b) For a contract that causes the task to be performed whatever A’s cost, what is the
lowest payment P can offer? Call such a contract with this payment “pool”.

Solution. The “pool” contract prescribes that, for any report, the task is performed for a
payment of T . An agent with cost c will perform the task if and only if T −c ≥ 0, or T ≥ c. Thus,
for both c = L and c = H agents to perform the task, the condition is T ≥ max{L, H} = H,
so the lowest value is Tp = H.

(c) Show that P’s expected profits are greater from contract “pool” than from contract
“separate” if and only if (B − H)/(B − L) > 1 − q. Explain why. Denote by q ∗ the solution to
(B − H)/(B − L) = 1 − q.

Solution. The expected profit from contract “pool” is Eπp = B − H, and from contract
“separate” is Eπs = q · 0 + (1 − q)(B − L). Thus, Eπp > Eπs if and only if
B−H H −L
B − H > (1 − q)(B − L) ⇔ > 1 − q, and q ∗ = .
B−L B−L

1
 Now consider the following two-stage contract, signed by P and A at the start of Stage 1:

• Stage 1: A (strategically) reports to P whether his cost, c1 of performing the specified

task is L or H. If he reports L, he performs the task, is paid an amount T , and the game
ends. If he reports H, the game proceeds to Stage 2.
• Stage 2: P costlessly selects a new task, also yielding him benefit B. A’s privately-observed
cost, c2 , of performing the new task is H with probability q and L with probability 1 − q,
independently of c1 . The contract specifies, as a function of A’s (strategic) report about
c2 , whether A performs the task and, if he does, what he will be paid. At any point, A
can quit, receiving utility zero.

(d) Consider two possible Stage-2 (continuation) contracts: contract “separate” and contract
“pool”. For each case,
i. what is A’s expected payoff, computed at Stage 1, of reporting that c1 is H?
ii. if c1 is actually L, what is the lowest value of T that induces A to reveal this?
iii. with T set equal to this value, what is P’s expected profit from the two-stage contract?

Solution. Consider the two contracts:

1. Contract “separate”
i. With probability 1 − q, A draws c2 = L, performs for a pay of L, and receives zero
utility. With probability q, he draws c2 = H, does not work and receives zero utility.
Thus, A’s continuation payoff of reporting c1 = H is zero.
ii. Suppose actually c1 = L. If he reports L, he performs for a pay of T , and receives
utility T − L. If he reports H, he receives the continuation payoff zero. Thus,
T − L ≥ 0, so the lowest value is T ∗ = L.
iii. Stage 2:
With probability 1 − q, c2 = L, A performs for a pay of L, and P receives B − L.
With probability q, c2 = H, A does not work, and P receives zero. Thus, P’s Stage-2
continuation payoff is (1 − q)(B − L).
Stage 1:
With probability 1−q, c1 = L, A performs for a pay of L, and P receives B −L. With
probability q, c2 = H, A reports H and Stage 2 begins: P receives the continuation
payoff of (1 − q)(B − L).
Thus, P’s ex ante payoff is

Eπs = (1 − q)(B − L) + q(1 − q)(B − L) = (1 − q 2 )(B − L).

2. Contract “pool”
i. With probability 1 − q, A draws c2 = L, performs for a pay of H and receives utility
H − L. With probability q, he draws c2 = H, performs for a pay of H and receives
utility zero. Thus, A’s continuation payoff of reporting c1 = H is (1 − q)(H − L).
ii. Suppose actually c1 = L. If he reports L, he performs for a pay of T , and receives
utility T − L. If he reports H, he receives the continuation payoff (1 − q)(H − L).
Thus, T − L ≥ (1 − q)(H − L), so the lowest value is T ∗ = (1 − q)H + qL.

2
 iii. Stage 2:
No matter the cost, A performs for a pay of H, and P receives B − H.
Stage 1:
With probability 1 − q, c1 = L, A performs for a pay of (1 − q)H + qL, and P receives
B − (1 − q)H − qL. With probability q, c1 = H, A reports H and Stage 2 begins: P
receives the continuation payoff of B − H.
Thus, P’s ex ante payoff is
Eπp = (1 − q)(B − (1 − q)H − qL) + q(B − H).

(e) Explain why, when q = q ∗ , P’s expected profit is greater from the two-stage contract
whose Stage-2 component is “separate” than from the one whose Stage-2 component is “pool”.

Solution. If q = q ∗ , we have Eπs > Eπp . Note that P’s Stage-2 continuation payoffs after
c1 = H are the same: (1 − q ∗ )(B − L) = (B − H). However, if the Stage-2 component is
“pool”, A has a non-zero continuation payoff after reporting c1 = H, so the payment in Stage
1 must be larger: (1 − q ∗ )H + q ∗ L vs simply L. Thus, P pays an additional cost to support
truthtelling when the Stage-2 component is “pool”, and A has a positive continuation payoff –
which translates into higher opportunity cost of a truthful report in Stage 1.

2 14.C.9 (MWG): Monopolistic insurance screening

Consider a risk-averse individual who is an expected utility maximizer with a Bernoulli
utility function over wealth u(·). The individual has initial wealth W and faces a probability θ
of suffering a loss of size L, where W > L > 0.
An insurance contract may be described by a pair (c1 , c2 ), where c1 is the amount of wealth
the individual has in the event of no loss and c2 is the amount the individual has if a loss is
suffered. That is, in the event no loss occurs the individual pays the insurance company an
amount (W − c1 ), whereas if a loss occurs the individual receives a payment [c2 − (W − L)]
from the company.

(a) Suppose that the individual’s only source of insurance is a risk-neutral monopolist
(i.e., the monopolist seeks to maximize its expected profits). Characterize the contract the
monopolist will offer the individual in the case in which the individual’s probability of loss, θ,
is observable.

Solution. First, note that the monopolist’s expected profit is

Eπ = (1 − θ)(W − c1 ) − θ(c2 − (W − L)) = −(1 − θ)c1 − θc2 + W − θL,
and thus, expected profit maximization is equivalent to the individual’s expected wealth mini-
mization, where the expected wealth is
EW = (1 − θ)c1 + θc2 .
Let u0 = (1 − θ)u(W ) + θu(W − L) denote the individual’s utility in the outside option if
he does not purchase any insurance. Then the monopolist’s problem is as follows:
(1 − θ)c1 + θc2 → min
(c1 ,c2 )

s.t. (1 − θ)u(c1 ) + θu(c2 ) ≥ u0 . (PC)

3
 It is straightforward to see that the optimal first-best contract is given by c∗1 = c∗2 = u−1 (u0 ).
That is, the individual has full insurance, and the monopolist extracts full surplus.

(b) Suppose, instead, that θ is not observable by the insurance company (the individual
knows θ). The parameter θ can take one of two values {θL , θH }, where θH > θL > 0 and
P(θL ) = λ. Characterize the optimal contract offers of the monopolist. Can one speak of one
type of insured individual being “rationed” in his purchases of insurance (i.e. he would want to
purchase more insurance if allowed to at fair odds)? Intuitively, why does this rationing occur?

Solution. Note that the monopolist’s expected profit maximization is equivalent to the
individual’s expected wealth minimization, as in part (a). Thus, the monopolist’s problem is
given by:

λ (1 − θL )cL1 + θL cL2 + (1 − λ) (1 − θH )cH H

   
2 + θ H c2 → min
{(cL L H H
1 ,c2 ),(c1 ,c2 )}

s.t. (1 − θL )u(cL1 ) + θL u(cL2 ) ≥ uL0 (PC-L)

(1 − θH )u(cH H H
1 ) + θH u(c2 ) ≥ u0 (PC-H)
(1 − θL )u(cL1 ) + θL u(cL2 ) ≥ (1 − θL )u(cH
1 ) + θL u(cH
2 ) (IC-L)
(1 − θH )u(cH H
1 ) + θH u(c2 ) ≥ (1 − θH )u(cL1 ) + θH u(cL2 ) (IC-H)

where we define the outside options:

ui0 = (1 − θi )u(W ) + θi u(W − L), i = L, H.

If we denote ∆θ = θH − θL and

ui = (1 − θi )u(ci1 ) + θi u(ci2 ), i = L, H,

then we can rewrite the constraints in terms of utilities:

uL ≥ uL0 (PC-L)
uH ≥ uH
0 (PC-H)
u ≥ u + ∆θ u(cH
L H H
(IC-L)
 
1 ) − u(c2 )
uH ≥ uL − ∆θ u(cL1 ) − u(cL2 ) (IC-H)
 

Let us figure out which constraints are slack and which are binding.

1. (PC-H) is slack.
Proof. Using (IC-H) and (PC-L),

uH ≥ uL − ∆θ u(cL1 ) − u(cL2 )
 

≥ uL0 − ∆θ u(cL1 ) − u(cL2 )

 

≥ uL0 − ∆θ [u(W ) − u(W − L)] = uH

0 ,

where the second inequality also uses the fact that cL1 ≤ W and cL2 ≥ W − L.
2. (IC-L) is slack.
Proof. By contraction, suppose (IC-L) is binding, then it holds with equality:

uH = uL − ∆θ u(cH H
 
1 ) − u(c2 )
≥ uL − ∆θ u(cL1 ) − u(cL2 ) ,
 

4
 where the inequality holds since we must have cH
1 ≤ c1 and c2 ≥ c2 . (Suppose not, then
L H L

by necessity, both c1 > c1 and c2 < c2 since (IC-L) is binding, and the above inequality
H L H L

is reversed, contradicting (IC-H).)

The above inequality is exactly (IC-H), meaning that (IC-H) is satisfied automatically
whenever (IC-L) is binding. Combining with part 1, note that both (PC-H) and (IC-H)
are slack, and monopolist is able to increase its profits by reducing uH by a small amount
without violating the constraints. This contradicts optimality.
Thus, the remaining constraints (PC-L) and (IC-H) must be binding. Let µ, ν denote the
Lagrange multipliers on (PC-L), (IC-H), respectively. The FOCs:
(cL1 ) λ(1 − θL ) − µ(1 − θL )u′ (cL1 ) + ν(1 − θH )u′ (cL1 ) = 0
(cL2 ) λθL − µθL u′ (cL2 ) + νθH u′ (cL2 ) = 0
(cH
1 ) (1 − λ)(1 − θH ) − ν(1 − θH )u′ (cH
1 ) = 0
(cH
2 ) (1 − λ)θH − νθH u′ (cH
2 ) = 0

Combining the last two equations, we obtain u′ (cH 1 ) = u (c2 ), which implies c1 = c2 = c ,
′ H H H H

since u (·) is strictly decreasing. It remains to pin down the value of c , as well as the implied
′ H

contract for the low-risk type (cL1 , cL2 ).

Let cL0 = u−1 (uL0 ) and cH 0 = u (u0 ) denote the certainty equivalents of the two types’
−1 H

outside options. Then the problem can be viewed as choosing cH ∈ [cH 0 , c0 ] such that the
L

contract (cH , cH ) for the high-risk type and the only contract (cL1 , cL2 ) for the low-risk type
satisfying both (PC-L) and (IC-H) are optimal. Equivalently (which will be more useful), we
can choose cL1 ∈ [cL0 , W ] and restore the values of cL2 and cH from the constraints. See the
illustration below:

First note that in optimum, cL1 > cL0 .

Proof. Suppose by contradiction that cL1 = cL0 . The implied values by (PC-L) and (IC-H)
are cL2 = cH = cL0 . Consider a small deviation from cL1 = cL0 to cL0 + ε as ε → 0+ . Then, in
a first-order approximation, cL2 ≈ cL0 − K1 ε and cH ≈ cL0 − K2 ε, where K1 is the MRS of the
low-risk type at (cL0 , cL0 ), since we move along the constraint (PC-L):
(1 − θL )
K1 = ,
θL

5
and K2 is calculated from (IC-H):
u(cL0 − K2 ε) ≈ (1 − θH )u(cL0 + ε) + θH u(cL0 − K1 ε) ⇒ K2 ≈ −(1 − θH ) + θH K1 ,
so we can set
∆θ
K2 = −(1 − θH ) + θH K1 = .
θL
Then it is straightforward to calculate the first-order approximation of change in the agent’s
expected wealth:
∆θ
∆EW ≈ λ [(1 − θL )ε − θL K1 ε] − (1 − λ)K2 ε = −(1 − λ) < 0,
θL
which is an improvement from the monopolist’s standpoint, contradicting optimality.

Next, we claim that the optimal contracts are (cL1 , cL2 ) = (W, W −L) and (cH H H H
1 , c2 ) = (c0 , c0 )
if and only if λ is sufficiently low – namely,
1−λ θL (1 − θL )u′ (cH ′ ′
0 ) [u (W − L) − u (W )]
≥ . (1)
λ ∆θu′ (W )u′ (W − L)
(Note: this situation corresponds with an exclusion of the low-risk type and serving only the
high-risk type with a first-best contract, which occurs if and only if the proportion of the low-
risk type is sufficiently low. We can interpret (1) in this way, since the left-hand side of (1) is
strictly decreasing in λ and can take any value in [0, +∞).)

Proof. Start with the contracts described above, and consider a small deviation from cL1 = W
to W − ε as ε → 0+ . As before, we use a first-order approximation to calculate the implied
changes in the contracts: cL2 ≈ W − L + K1 ε and cH ≈ cH 0 + K2 ε, where

(1 − θL )u′ (W )
K1 =
θL u′ (W − L)
and from u(cH
0 + K2 ε) ≈ (1 − θH )u(W − ε) + θH u(W − L + K1 ε),

u′ (W ) ∆θ
u′ (cH ′ ′
0 )K2 ≈ −(1 − θH )u (W ) + θH u (W − L)K1 ⇒ K2 = .
u′ (cH
0 ) θL

The first-order approximation of change in the agent’s expected wealth:

∆EW ≈ λ [−(1 − θL )ε + θL K1 ε] + (1 − λ)K2 ε
u′ (W ) − u′ (W − L) u′ (W ) ∆θ
 
= λ(1 − θL ) + (1 − λ) ′ H ε.
u′ (W − L) u (c0 ) θL
This deviation is an improvement from the monopolist’s standpoint if and only if ∆EW < 0,
thus, the initial contract is optimal if and only if (1) holds.

Finally, we characterize the optimal contracts which serve both types, under the assumption
that (1) does not hold. The unique pair of contracts is (cL1 , cL2 ), (cH , cH ) such that cL1 , cL2 , cH
solve
θL (1 − θL )u′ (cH ) u′ (cL2 ) − u′ (cL1 )
 
1−λ
= ; (2)
λ ∆θu′ (cL1 )u′ (cL2 )
(1 − θL )u(cL1 ) + θL u(cL2 ) = uL0 ; (3)
H −1 L L
(4)
 
c =u (1 − θH )u(c1 ) + θH u(c2 ) .

6
 Proof. Start with the contract described above, and consider a small deviation (as a first-
order approximation) to (cL1 + ε, c2 − K1 ε), (cH − K2 ε, cH − K2 ε) as ε → 0 (both from the left
and from the right), where the constants K1 , K2 are calculated analogously from the constraints
and equal
(1 − θL )u′ (cL1 ) u′ (cL1 ) ∆θ
K1 = , K2 = .
θL u′ (cL2 ) u′ (cH ) θL
The first-order approximation of change in the agent’s expected wealth:
u′ (cL1 ) − u′ (cL2 ) u′ (cL1 ) ∆θ
 
∆EW ≈ λ(1 − θL ) + (1 − λ) ′ H ε=0
u′ (cL2 ) u (c ) θL
(optimality condition) if and only if (2) holds. Equations (3) and (4) simply correspond to
(PC-L) and (IC-H).

To summarize: if λ is low, or (1) holds, it is optimal to exclude the low-risk type and only
serve the high-risk type with the first-best contract; and if λ is high, or (1) does not hold,
it is optimal to serve both types by offering contracts described in (2)—(4). In either case,
the high-risk type has full insurance, and in the latter case, the low-risk type only has partial
insurance, bearing some uncertainty.
Now, let us answer the “rationing” part. Consider the situation in which the monopolist
serves both types. Keep in mind that both contracts are available for purchase for either type,
and if we allow agents to purchase contracts that are proportional to the existing ones (like
buying more or less of the same insurance plan, each plan offering a fixed rate at which one
can trade wealth across states), the high-risk type would like to purchase a scaled-up version
of the low-risk type’s insurance instead of his own insurance. See the illustration below:

In this sense, by removing the option to scale contracts, the high-risk type is being “rationed”:
he would like to purchase more of the low-risk type’s insurance, but he cannot. This “rationing”
occurs so that the monopolist is able to separate types and extract higher profits.

7
 (c) Compare your solution in (b) with the equilibrium in the model with competing insur-
ance companies.

Solution. The picture below illustrates the equilibria with competitive insurance and mo-
nopolistic insurance:

As we can see, the low-risk type is clearly better off in the competitive equilibrium. However,
it is not clear for the high-risk type. The picture shows an example in which the high risk type
is worse off in the competitive equilibrium. However, depending on the parameters, it might
be that the high-risk type is better off.
We can interpret this ambiguity for the high-risk type as follows. On the one hand, the
competitive equilibrium should allocate more resources to the customers – this would correspond
to a situation in which both types are better off. On the other hand, it should increase overall
efficiency of the market, meaning that the high-risk types would be treated worse relative to
the low-risk types – if this effect prevails, this would correspond to the high-risk type being
worse off in the competitive equilibrium.

3 Screening with random contracts

Consider a monopolist who is willing to sell one unit of some good at zero cost. There is a
single buyer who can have the following utility from acquiring the good:

θL − T

or
log(θH − T ),
where T is the price of the good, and θH − 1 > θL . We assume that the buyer’s outside option
is zero.

8
First best. Suppose the monopolist observes the buyer’s type. Then he sells the good at
TL = θL to the low type, and TH = θH − 1 to the high type, extracting full surplus in both
cases.

Second best. Suppose the monopolist does not observe the buyer’s type, and assume that the
proportion of the low type is λ. Then he either serves both types by offering a uniform price of
T = θL and obtaining the profit π = θL , or he only serves the high type by offering a uniform
price of T = θH − 1 and obtaining the profit π = (1 − λ)(θH − 1). Note that both types are
served if and only if
θL
θL ≥ (1 − λ)(θH − 1) ⇔ λ≥1− ,
θH − 1
that is, if the proportion of the low type is sufficiently high.

Random contracts. Suppose the monopolist still cannot observe the buyer’s type, but is
allowed to offer random contracts (that is, deals which are lotteries over contracts). In this
case, in fact, the monopolist can achieve the first-best profit by offering the following two
contracts:

• The good is sold at T = θH − 1 with probability 1.

• The good is sold at T = θL + γ with probability 0.5 and at T = θL − γ with probability

0.5.

Note that, if the high type takes the first contract, and the low type takes the second
contract, the monopolist achieves the same profit (in expectation) as in the first best: he
extracts θH − 1 from the high type, and 0.5(θL + γ) + 0.5(θL − γ) = θL from the low type.
Now, let us check if there exists γ such that this set of contracts is incentive compatible.
The incentive constraints:

0.5(θL − (θL + γ)) + 0.5(θL − (θL − γ)) ≥ θL − (θH − 1) (IC-H)

log(θH − (θH − 1)) ≥ 0.5 log(θH − (θL + γ)) + 0.5 log(θH − (θL − γ)) (IC-L)

Note that (IC-H) is fulfilled for any γ: the left-hand side equals 0, and the right-hand side
equals θL − (θH − 1) < 0. Next, (IC-L) can be rewritten as

log((θH − θL ) − γ) + log((θH − θL ) + γ) ≤ 0.

Use the log addition rule and exponentiate both sides to obtain:

((θH − θL ) − γ)((θH − θL ) + γ) ≤ 1 ⇔ (θH − θL )2 − γ 2 ≤ 1,

or p
γ≥ (θH − θL )2 − 1.
Thus, if γ is sufficiently high, the proposed set of offers containing a random offer does
implement the first-best profit for the monopolist in an incentive compatible way.