Lecture 3: Applications of Screening

ECON 813

Shota Ichihashi
 Applications of Screening

Many important applications of optimal screening.

We cover applications to:
I Auctions (Myerson 1981)
I Insurance (Stiglitz 1971)
I Taxation (Mirrlees 1971)
We don’t have time to cover other applications:
I Credit rationing (Stiglitz and Weiss 1981)
I Regulation (Laffont and Tirole 1986)
 Optimal Auctions (Myerson 1981)

Auctioneer has a single, indivisible good to sell.

n ≥ 1 bidders.

Bidder i’s value for winning the good is θi , distributed on [θ i , θ i ] with

density fi (θi ).

Bidders’ values are independent, but may not be identically distributed.

What auction format maximizes auctioneer’s expected revenue, subject to

giving each bidder a non-negative payoff from participation in the auction?
 Connection to Screening

Problem similar to last lecture

Last lecture: Allocation is quantity qi , utility θi v(qi ), any non-negative

allocation possible.

Today: Allocation to bidder i is a probability qi of receiving good, utility θi qi

(i.e., v(qi ) = qi ), any non-negative allocation where probabilities sum to
less than 1 possible.

Only difference is feasibility constraint:

X
qi (θ) ≤ 1 and qi (θ) ∈ [0, 1], ∀i, ∀θ = (θ1 , . . . , θn ).
i
 Auctioneer’s Problem
From bidder i’s view, what matters is the expected quality and transfer
(because of risk neutrality)
Given allocation rule q(θ), transfer rule t(θ), define

Qi (θi ) := Eθ−i [qi (θi , θ−i )] (1)

Ti (θi ) := Eθ−i [ti (θi , θ−i )] (2)

Thanks to Revelation Principle, auctioneer’s problem is

Z θi X
max Ti (θi )fi (θi )dθi
q(θ),t(θ) θ
i i

subject to
(ICθi ) θi Qi (θi ) − Ti (θi ) ≥ θi Qi (θi0 ) − Ti (θi0 ), ∀θi , θi0
(IRθi ) θi Qi (θi ) − Ti (θi ) ≥ 0, ∀θi
X
(feasibility) qi (θ) ≤ 1.
i
 Solution

Exactly as in the last lecture, we can show that IC is equivalent to

monotonicity and local IC
Z θi
Ui (θi ) = Ui (θi ) + Qi (x)dx.
θi

Therefore, Z θi
Ti (θi ) = θi Qi (θi ) − Qi (x)dx, (3)
θi

where we use Ui (θi ) = 0 at the optimum.

 Solution
Substitute out for Ti in objective and integrate by parts (like last lecture).
Find that (q(θ), t(θ)) is optimal auction if q solves

θi  
XZ 1 − Fi (θi )
max θi − Qi (θi )fi (θi )dθi
q(θ) θi fi (θi )
i
R
or equivalently (recall Qi (θi ) = θ qi (θi , θ−i )f−i (θ−i )dθ−i )
−i

Z X   
1 − Fi (θi )
max θi − qi (θ) f (θ)dθ
q(θ) Θ fi (θi )
i

subject to monotonicity of Qi (θi ) and feasibility, and t satisfies

Z θi
Ti (θi ) = θi Qi (θi ) − Qi (x)dx, ∀θi . (4)
θi
 Interpreting Solution

Z X 
1 − Fi (θi )
max θi − qi (θ)f (θ)dθ
q(θ) Θ fi (θi )
i

As before, we can solve the problem point-wise: For each θ , what

R
(q1 (θ), . . . , qn (θ)) maximizes the integrand (inside of · · · dθ)?
Given
h θ, the inside h (ignoring f (θ)
i of the integral i is
1−F1 (θ1 ) 1−Fn (θn )
θ1 − f1 (θ1 ) q1 (θ) + · · · + θn − fn (θn ) qn (θ).
Optimal to allocate good with probability 1 to bidder with highest virtual
value, so long as this is positive.

If no bidder has positive virtual value, optimal for seller to keep good
(reserve price)

As in last lecture, solution only applies in regular case, i.e., every bidder
has non-decreasing virtual values
 Computing Optimal Auctions

We now the optimal (q1 (θ), . . . , qn (θ)). Transfer?

One transfer rule that satisfies

Z θi
Ti (θi ) = θi Qi (θi ) − Qi (x)dx, ∀θi . (5)
θi

is Z θi
ti (θ) = θi qi (θ) − qi (x, θ−i )dx, ∀θ. (6)
θi

That is, have equality (5) hold for each type profile θ−i of other bidders, not
just in expectation over θ−i

Can use this to compute optimal auctions in examples

(together with optimal allocation rule)
 Example
Suppose θi distributed uniformly on [0, 1].
1−F(θi ) 1−θi
Then θi − f (θi ) = θi − 1 = 2θi − 1.
Virtual values are monotone, so optimal to allocate good with probability 1 to
1
bidder with highest value, so long as value exceeds 2

Recall one possible payment rule is

Z θi
ti (θ) = θi qi (θ) − qi (x, θ−i )dx, ∀θ. (7)
θi

∗
Winning bidder ( ⇐⇒ bidder i with θi > max(1/2, θ−i ) ) pays
Z θi  
1 ∗
ti (θ) = θi · 1 − 1dx = max , θ−i ,
max{ 12 ,θ−i
∗
} 2
∗
where θ−i is the second-highest value

That is, optimal auction is 2nd price auction with reserve price of 12 .
 Revenue Equivalence Theorem
We’ve essentially prove the following:

Theorem ((Traditional) Revenue Equivalance Theorem)

Suppose bidder i’s type is drawn from an interval [θ i , θ i ] with strictly
positive density fi , independently across bidders. Then any auction format
and an equilibrium in which
1. The good is always allocated to the bidder with the highest value, and
2. A bidder with her lowest possible type (θ i ) gets payoff 0,
the seller’s expected revenue is the same.

Proof.
Expected revenue in any such auction is
Z X   
1 − Fi (θi ) ∗
θi − qi (θ) f (θ)dθ
Θ fi (θi )
i
 Revenue/Payoff Equivalence Theorem: Remarks

1. More general result: any two auction formats with same allocation rule
q(θ) and same values of Ui (θi ) are equivalent in terms of seller’s expected
revenue and expected payoff of each bidder type (payoff equivalence
theorem)

2. Auctions covered by (traditional) revenue equivalence theorem may not

be optimal; not optimal to allocate good to bidder with highest value if the
virtual value is negative.
 Optimal Insurance (Stiglitz 1977)

Risk-neutral monopoly insurance company faces a risk-averse consumer

with utility function u(c) satisfying u0 > 0, u00 < 0.

Consumer has initial wealth W , has probability θ of suffering loss of size L.

θ ∈ {θL , θH }, 0 < θL < θH < 1, Pr(θ = θL ) = β .

Company offers insurance contract specifying payment from consumer to

company when loss does not occur (premium), payment from company to
consumer when loss does occur (L minus deductible).

Equivalently, offers consumption for consumer when loss does/doesn’t

occur.

What insurance contract maximizes company’s expected profit when θ is

observable? When θ is unobservable?
 Complete Information

max(1 − θ)(W − c1 ) + θ(W − c2 )

c1 ,c2

subject to
IRθ : (1 − θ)u(c1 ) + θu(c2 ) ≥ (1 − θ)u(W) + θu(W − L).
 Complete Information
Can rewrite objective as

min(1 − θ)c1 + θc2

c1 ,c2

subject to IRθ : (1 − θ)u(c1 ) + θu(c2 ) ≥ U(θ)

where U(θ) := (1 − θ)u(W) + θu(W − L).

By the “certainty equivalent” argument we saw in Moral Hazard part

implies the optimal (c1 , c2 ) is c1 = c2 = c∗ that solves u(c∗ ) = U(θ).

Alternatively, IR binds at the optimum, so the problem becomes

  
−1 U(θ) − (1 − θ)u(c1 )
min(1 − θ)c1 + θ u .
c1 ,c2 θ

The first-order condition implies u0 (c1 ) = u0 (c2 ), or c1 = c2 .

 Incomplete Information

Problem is

min β[(1 − θL )cL1 + θL cL2 ] + (1 − β)[(1 − θH )cH H

1 + θH c2 ]
cL1 ,cL2 ,cH H
1 ,c2

subject to
ICθH : (1 − θH )u(cH H L L
1 ) + θH u(c2 ) ≥ (1 − θH )u(c1 ) + θH u(c2 )

ICθL : (1 − θL )u(cL1 ) + θL u(cL2 ) ≥ (1 − θL )u(cH H

1 ) + θL u(c2 )

IRθH : (1 − θH )u(cH H
1 ) + θH u(c2 ) ≥ (1 − θH )u(W) + θH u(W − L)

IRθL : (1 − θL )u(cL1 ) + θL u(cL2 ) ≥ (1 − θL )u(W) + θL u(W − L).

 Simplifying Constraints

What constraints do we expect to bind?

First-best solution: type θi consumer is fully insured and gets consumption

c(θi ) that solves

u(c(θi )) = (1 − θi )u(W) + θi u(W − L). (8)

Low risk type θL gets more consumption than high risk type θH

Type θL (low risk) is not tempted to take contract intended for type θH

Type θH is tempted to take contract intended for type θL

Solve problem assuming that ICθL and IRθH are slack

→ Check (on your own) that ICθL and IRθH hold.
 Analysis

After dropping ICθL and IRθH , relaxed problem is

min β[(1 − θL )cL1 + θL cL2 ] + (1 − β)[(1 − θH )cH H

1 + θH c2 ]
cL1 ,cL2 ,cH H
1 ,c2

subject to
ICθH : (1 − θH )u(cH H L L
1 ) + θH u(c2 ) ≥ (1 − θH )u(c1 ) + θH u(c2 )

IRθL : (1 − θL )u(cL1 ) + θL u(cL2 ) ≥ (1 − θL )u(W) + θL u(W − L).

cH H
1 , c2 do not appear in IRθL , so be chosen to minimize objective subject
to ICθH

By the same argument as in complete information case, implies that

cH H H
1 = c2 = c at optimum.
 Solution

min β[(1 − θL )cL1 + θL cL2 ] + (1 − β)cH

cL1 ,cL2 ,cH

subject to
ICθH : u(cH ) ≥ (1 − θH )u(cL1 ) + θH u(cL2 )
IRθL : (1 − θL )u(cL1 ) + θL u(cL2 ) ≥ U(θL ).

Both constraints bind. We then have

U(θL ) − (1 − θL )u(cL1 )
 
−1
Binding IRθL : =u cL2
θL
 
H −1 θH θH − θL L
Binding ICθH : c = u U(θL ) − u(c1 ) .
θL θL
 Solution

Substitute out for cH , cL2 . Problem becomes

U(θL ) − (1 − θL )u(cL1 )
  
−1
minβ (1 − θL )cL1
+ θL u
cL1 θL
 
−1 θH θ H − θL L
+(1 − β)u U(θL ) − u(c1 ) .
θL θL

The FOC wrt cL1 gives (recall (u−1 )0 = u10 )

u0 (cL1 ) 1−β θ H − θL u0 (cL1 )

= 1 − · · < 1. (9)
u0 (cL2 ) β θL (1 − θL ) u0 (cH )
 Interpreting Solution
u0 (cL1 )
u0 (cL2 )
< 1 ⇐⇒ cL1 > cL2 .

Low-risk type θL is underinsured at optimum.

Optimal to underinsure low-risk type to make low-risk type’s contract less

appealing for high-risk type, reduce high-risk type’s information rent.

θH θ H − θL L
u(cH ) = U(θL ) − u(c1 ). (10)
θL θL

For given utility level of low-risk type, high-risk type’s information rent is
decreasing in cL1 (increasing in level of low-type’s insurance).
1−β
More underinsurance when high-risk types more prevalent ( β larger),
risk difference between types greater (θH − θL larger), welfare gains from
insuring low-risk type smaller θL close to 0 or 1.
 Optimal Taxation (Mirrlees 1971)

Utilitarian government faces continuum of individuals.

Fraction β of individuals are low-ability (θ = θL ), fraction 1 − β are
high-ability (θ = θH )
Each individual chooses how much effort e to exert, produces output
y = θe.
Individual utility u(y − t − c(e)). u0 > 0, u00 < 0, c0 > 0, c00 > 0
Government’s problem is to design tax system to maximize utilitarian
social welfare:
βu(θL eL − tL − c(eL )) + (1 − β)u(θH eH − tH − c(eH ))
Government’s budget constraint βtL + (1 − β)tH ≥ 0
 Complete Information
If government can observe ability and output (and thus effort), the problem
is

max βu(θL eL − tL − c(eL )) + (1 − β)u(θH eH − tH − c(eH ))

eH ,eL ,tH ,tL

subject to budget constraint βtL + (1 − β)tH ≥ 0.

Substitute the binding constraint to eliminate tL or tH from the objective
First-order conditions:

u0L = u0 (θL eL − tL − c(eL )) = u0 (θH eH − tH − c(eH )) = u0H

c0 (eL ) = θL
c0 (eH ) = θH .

Government has each individual exert first-best effort, redistributes income

to as to equalize marginal utilities (and hence total utilities).
 Incomplete Information
If government can only observe output θe (but not effort or ability), the
problem is

max βu(θL eL − tL − c(eL )) + (1 − β)u(θH eH − tH − c(eH )) (11)

eH ,eL ,tH ,tL

subject to (12)
 
θL
(ICH) θH eH − tH − c(eH ) ≥ θL eL − tL − c eL (13)
θH
 
θH
(ICL) θL eL − tL − c(eL ) ≥ θH eH − tH − c eH (14)
θL
(Budget) βtL + (1 − β)tH ≥ 0. (15)

Explanation: If type θH takes contract intended for type θL , must produce

output θL eL .
Requires effort θθL from her, as her ability is θH
H
 Solution
ICH violated at first-best, as at first-best we have
 
θL
θH eH − tH − c(eH ) = θL eL − tL − c (θL ) < θL eL − tL − c eL .
θH
Binding constraints are ICH and budget constraint. Implies
  
θL
tH =β θH eH − c(eH ) − θL eL + c eL
θH
  
θL
tL = − (1 − β) θH eH − c(eH ) − θL eL + c eL .
θH
Substitute our for tH , tL , take first-order conditions with respect to eH , eL :

c0 (eH ) = θH
u0L − u0H
   
0 0 θ L 0 θL
c (eL ) = θL − (1 − β) c (eL ) − c eL
βu0L + (1 − β)u0H θH θH
< θL .
 Interpreting Solution

u0L − u0H
   
θL 0 θL
c0 (eL ) = θL − (1 − β) c0 (eL ) − c eL .
βuL + (1 − β)u0H
0 θH θH

Again, distort low type’s allocation to reduce high type’s information rent.
 
High type’s information rent (in consumption) is c(eL ) − c θθL eL .
H

Intuition:
 Benefit high type gets from mimicking low type is that only costs
θL
c θH eL to produce θL , rather than c(eL ).
Convexity of c implies that a marginal
 decrease
 in eL reduces high type’s
information rent by c0 (eL ) − θθL c0 θL
θH e L >0
H

u0L −u0H
βu0L +(1−β)u0H
term measures benefit of this reduction in inequality for
utilitarian social welfare.
 Real-World Implications

Model implies that marginal income taxes should be zero for highest
income, positive for lower incomes
There is still redistribution: entire purpose of marginal tax on low types is
to make redistribution possible by getting high types to reveal themselves.
Recall with continuum of types, zero marginal tax at top only holds for very
highest income, marginal taxes just below the top depend on inverse
1−F(θ)
hazard rate f (θ)
Diamond (1998), Saez (2001) argue that with realistic skill distribution,
Mirrlees model is consistent with high marginal taxes on the rich.
 Summary

3 applications of screening.
Auctions: Optimal auction allocates good to bidder with highest
non-negative virtual value.
Explains use of reserve price in auctions. Revenue/payoff equivalence
holds for auctions with the same allocation rule.
Insurance: Optimal to underinsure low-risk types to get high-risk types
to pay high price for full insurance.
Taxation: Optimal to impose positive marginal labor tax on low-ability
types to get high-ability types to work harder.
With contiuum of types, model consistent with positive marginal taxes on
high-ability types.