Chapter 10

Introduction to Mechanism
Design1

10.1 A Simple Screening Example

How should a seller sell to a buyer with an unknown marginal will-
ingness to pay? Specifically, suppose the buyer has preferences

✓q p,

where ✓ > 0 is the buyer’s marginal value, q 0 is the quantity

purchased, and p is the total price paid. The seller has costs given
by the increasing convex C 2 function

c(q),

satisfying c 0 (0) = 0, c 00 0, and limq!1 c 0 (q) = 1.

If the seller knew the value of ✓, then he would set p = ✓q and
choose q to maximize

p c(q) = ✓q c(q),

that is, choose q so that marginal cost c 0 (q) equals ✓ (this is also
the efficient quantity).
But what if the seller does not know the value of ✓? Suppose the
seller assigns probability ↵H 2 (0, 1) to ✓ = ✓H and complementary
1
I have shamelessly stolen the idea of introducing mechanism design in this way
from Börgers (2015).

289
290 Chapter 10. Introduction to Mechanism Design

probability ↵L = 1 ↵H to ✓L < ✓H . Such a situation is said to ex-

hibit adverse selection: a relationship is subject to adverse selection
if one of the participants has private information, and uninformed
participants are disadvantaged by this hidden information (another
common term).
If the seller could price discriminate, how would he do so? If the
seller price discriminates, he charges different prices for different
quantities. Since there are two types, it is natural to think of the
seller as offering two quantities, q1 and q2 and associated prices, p1
and p2 , so that price is a function of quantity. But, in choosing these
two quantities (and the associated prices), the seller is targeting the
buyer type, so in what follows we think of the seller as offering a
pair of screening contracts {(qL , pL ), (qH , pH )}, with the intent that
the buyer of type ✓t chooses the contract (qt , pt ).
The seller chooses the pair of contracts to maximize

↵L (pL c(qL )) + ↵H (pH c(qH ))

subject to

✓L qL pL ✓L qH pH , (ICL )
✓H qH pH ✓H qL pL , (ICH )
✓L qL pL 0, (IRL )
and ✓H qH pH 0. (IRH )

Inequalities (ICL ) and (ICH ) are the incentive compatibility condi-

tions that ensure the seller’s targeting of buyer types is successful.
Conditions (IRL ) and (IRH ) are the individual rationality or partici-
pation constraints that ensure that both buyer types at least weakly
prefer to purchase. These last two constraints are illustrated in Fig-
ure 10.1.1.
It is immediate (both from the figure and algebraic manipulation
of the IC and IR constraints) that (IRL ) and (ICH ) jointly imply (IRH ).
Moreover, (ICL ) and (ICH ) imply qH qL . Finally, if qL î qH , only
one of (ICL ) and (ICH ) can bind.
The Langrangean for this maximization is

L = ↵L (pL c(qL )) + ↵H (pH c(qH ))

+ L [✓L qL pL (✓L qH pH )] + H [✓H qH pH (✓H qL pL )]
+ µL [✓L qL pL ] + µH [✓H qH pH ].
10.1. A Simple Screening Example 291

p
✓H indiff curve

✓L indiff curve
buyer inc pref

p0

q0 q

Figure 10.1.1: The buyer’s indifference curves for different values of ✓.

Inequality (IRL ) binds for the contract (q0 , p 0 ), since that
buyer’s indifference curve passes through the origin (and so
the buyer is indifferent between the contract and not pur-
chasing). Buyer ✓H strictly prefers the contract (q0 , p 0 ) to
not purchasing, and so (IRH ) holds as a strict inequality.

The first order conditions for an interior solution are

↵L c 0 (qL ) + L ✓L H ✓H + µL ✓L = 0, (10.1.1)
↵L L + H µL = 0, (10.1.2)
0
↵H c (qH ) L ✓L + H ✓H + µH ✓H = 0, and (10.1.3)
↵H + L H µH = 0, (10.1.4)

together with the complementary slackness conditions on the mul-

tipliers.
From above, since (IRH ) does not bind, µH = 0. Moreover, since
only one of the two incentive constraints can bind, either L or H
equal zero. But (10.1.4), µH = 0, and L 0 imply H > 0, so L = 0.
Then, (10.1.4) implies H = ↵H , and so (10.1.3) implies qH satisfies

c 0 (qH ) = ✓H ,
292 Chapter 10. Introduction to Mechanism Design

1
p = ⇡0 + 2 cq2
p ✓H
✓H
00
pH ✓L
0
pH

pL0

pL00

q
qL00 qL0 0
qH 00
= qH

Figure 10.1.2: Optimal screening contracts for two different values of ↵,

with ↵00 0
H > ↵H , with quadratic costs. The two dashed curved
lines are isoprofit loci for the seller.

and so is the full information profit maximizing quantity. From

H = ↵H and (10.1.2) we conclude µL = 1. Finally, from (10.1.1), we
have
↵L c 0 (qL ) ↵H ✓H + ✓L = 0,
so that
✓L ↵H ✓H
c 0 (qL ) = .
(1 ↵H )
This only implies an interior solution if ✓L > ↵H ✓H . If ✓L  ↵H ✓H ,
qL = pL = 0, the seller is better off selling only to ✓H , and does not
sell to ✓L (see Problem 10.5.1).
The prices pL and pH are then determined from (IRL ) and (ICH )
respectively (see Figure 10.1.2 for the case of quadratic costs, c(q) =
1 2
2 cq ).
When the seller optimally sells to both types of buyer, buyer ✓H
strictly prefers to purchase rather than not purchase, and this addi-
tional payoff (over zero) is called buyer ✓H ’s information rent.
10.2. A Less Simple Screening Example 293

Note that qL & 0 as ↵H increases. As ↵H increases, it is more

valuable to the seller to sell to ✓H at a higher price. But in order to
do so, it must make the low value contract less attractive to ✓H . In-
creasing the price on qL does this but at the cost of losing sales to ✓L ,
since ✓L is already just indifferent between buying and not buying.
But since ✓L values the good less than does ✓H , by simultaneously
lowering qL and pL at just the right rate, the seller simultaneously
maintains some sales to ✓L while making the contract less attractive
to ✓H .
Rather than offering a pair of contracts, we can equivalently think
of the seller as committing to the following direct mechanism: first
the seller specifies which contract the buyer will receive as a func-
tion of the buyer’s announcement of her value of ✓, and then the
buyer chooses which value of ✓ to report. If the seller’s specifi-
cation satisfies (ICL ), (ICH ), (IRL ), and (IRH ), then the buyer finds it
optimal to both participate in the mechanism and truthfully report
her valuation. This is a trivial example of the revelation principle,
which we will return to several times (Theorems 11.2.1 and 12.1.1).

10.2 A Less Simple Screening Example

In this section, we suppose the seller has beliefs F over the buyer’s
valuation ✓, with F having support [✓, ✓] and strictly positive den-
sity f . The possibilities for price discrimination are now signifi-
cantly larger, since there are many more types. Rather than speci-
fying price as a function of quantity, it is more convenient to apply
the revelation principle (outlined at the end of the previous section)
and model the seller as choosing direct mechanism, i.e., a menu of
contracts {(q(✓), p(✓)) : ✓ 2 [✓, ✓]}, with contract (q(✓), p(✓)) in-
tended for buyer ✓ (but see Remark 10.2.1).
The seller chooses the pair of functions (q, p) : [✓, ✓] ! R2+ to
maximize Z
[p(✓) c(q(✓))]f (✓) d✓ (⇧)

subject to

✓q(✓) p(✓) ˆ
✓q(✓) ˆ
p(✓) ˆ 2 [✓, ✓]
8✓, ✓ (IC✓ )
and ✓q(✓) p(✓) 0 8✓ 2 [✓, ✓]. (IR✓ )
294 Chapter 10. Introduction to Mechanism Design

Lemma 10.2.1. Suppose a pair of functions (q, p) satisfies (IC✓ ), and

define
U (✓) := ✓q(✓) p(✓).
The function U is nondecreasing and convex, q is nondecreasing, and
Z✓
U (✓) = U (✓) + ˜ d(✓).
q(✓) ˜ (10.2.1)
✓

ˆ the function defined by

Proof. For each ✓,
ˆ
g✓ˆ(✓) := ✓q(✓) ˆ
p(✓)

is a nondecreasing affine function of ✓.

The conditions (IC✓ ) imply (since U (✓) = g✓ (✓))

U (✓) = max g✓ˆ(✓) 8✓.

ˆ
✓2[✓,✓]

That is, U is the pointwise maximum (upper envelope) of the col-

lection of nondecreasing affine functions {g✓ˆ : ✓ ˆ 2 [✓, ✓]}. It is,
therefore, nondecreasing and convex (see Figure 10.2.1).2
Every convex function on an open interval is absolutely continu-
ous (Royden and Fitzpatrick, 2010, page 132, Corollary 17) and so
differentiable almost everywhere. From the envelope characteriza-
tion, the derivative of U at ✓ (when it exists) is given by q(✓) (Prob-
lem 10.5.5), and since it is the derivative of a convex function, q is
nondecreasing when it is the derivative.3 To prove that q is nonde-
creasing everywhere, add (IC✓ ) to the same constraint reversing the
roles of ✓ and ✓ 0 and rearrange to conclude

(✓ ✓ 0 )(q(✓) q(✓ 0 )) 0.

Finally, since U is absolutely continuous on (✓, ✓), it is equal to

the integral of its derivative.4 That is, for all ✓ 0 < ✓ 00 2 (✓, ✓), we
2
If each f↵ is convex and f (x) := sup↵ f↵ (x), then f (tx + (1 t)x 0 ) =
sup↵ f↵ (tx + (1 t)x 0 )  sup↵ {tf↵ (x) + (1 t)f↵ (x 0 )}  t sup↵ f↵ (x) +
(1 t) sup↵ f↵ (x 0 ) = tf (x) + (1 t)f (x 0 ).
3
A more direct and intuitive (but less elegant) argument that implicitly assumes
continuity of q(·) (as well as ignoring the issue of footnote 4 on page 294) is in
Mas-Colell et al. (1995, page 888). See also the proof of Lemma 12.2.1.
4
See Royden and Fitzpatrick (2010, Theorem 10, page 124). Absolute continu-
ity is needed here, as there are functions for which the fundamental theorem of
calculus fails. For an example of a strictly increasing function with zero derivative
almost everywhere, see Billingsley (1995, Example 31.1).
10.2. A Less Simple Screening Example 295

g✓ˆ00

g✓ˆ00

g✓ˆ0

g✓ˆ

Figure 10.2.1: The upper envelope of affine functions is convex.

have
Z ✓00
00
U (✓ ) = U (✓ ) + 0 ˜ d(✓).
q(✓) ˜ (10.2.2)
✓0

Since U satisfies (IC✓ ) and is nondecreasing, we have that for all

✓ † < ✓ ‡ 2 [✓, ✓],

U (✓ ‡ ) (✓ ‡ ✓ † )q(✓ ‡ )  U (✓ † )  U (✓ ‡ ), (10.2.3)

and so U is continuous on the closed interval [✓, ✓].5 Consequently,

(10.2.2) holds with ✓ 0 = ✓ and ✓ 00 = ✓.

It is immediate from (10.2.1) that U is continuous in ✓, but as

(10.2.3) reveals, continuity is a direct implication of (IC✓ ).
Lemma 10.2.1 implies that for any pair of functions (p, q) satis-
fying (IC✓ ), the price function is pinned down (up to a constant U (✓))
5
Continuity on the open interval follows from convexity, with continuity at ✓
then being an implication of U being nondecreasing. There are nondecreasing
convex functions that are discontinuous at ✓, such as the function obtained from
a continuous nondecreasing convex function by adding 1 to the function at ✓.
296 Chapter 10. Introduction to Mechanism Design

by the quantity function, since U (✓) = ✓q(✓) p(✓) and (10.2.1) im-
plies
Z✓
p(✓) = ✓q(✓) U (✓) = ✓q(✓) ˜ d✓
q(✓) ˜ U (✓).
✓

Lemma 10.2.1 implies that U (✓) 0 is sufficient to imply partic-

ipation for all ✓. Since the seller’s profits are decreasing in U (✓),
the seller sets U (✓) = 0 (being a constant, the value of U (✓) has no
implications for (IC✓ )).
It turns out that no further restrictions beyond Lemma 10.2.1 are
implied by (IC✓ ) and (IR✓ ):

Lemma 10.2.2. Suppose q : [✓, ✓] ! R+ is a nondecreasing function.

Then, (q, p) satisfies (IC✓ ) and (IR✓ ), where
Z✓
p(✓) = ✓q(✓) ˜ d✓
q(✓) ˜ (10.2.4)
✓
Z✓
= ✓q(✓) + [q(✓) ˜ d✓.
q(✓)] ˜
✓

Proof. Define
Z✓
⇤
U (✓) := ˜ d✓
q(✓) ˜ = ✓q(✓) p(✓).
✓

Then, U ⇤ is a convex function with U ⇤ (✓) = 0, and (IR✓ ) is satisfied.

ˆ
Moreover, since it is convex, for all ✓,
⇤ ˆ
U ⇤ (✓) ˆ + (✓
U ⇤ (✓) ˆ dU (✓)
✓)
d✓
ˆ + (✓
= U ⇤ (✓) ˆ ˆ
✓)q(✓)
ˆ
= ✓q(✓) ˆ
p(✓),

proving (IC✓ ).

Thus, to solve the seller’s problem we need only maximize

Z✓( Z✓ )
✓q(✓) ˜ d✓
q(✓) ˜ c(q(✓)) f (✓) d✓
✓ ✓
10.2. A Less Simple Screening Example 297

over all nondecreasing functions q. R R

We first apply integration by parts (i.e., uv 0 = uv u0 v) to
the double integral:6

Z✓Z✓ Z✓ ! ✓ Z✓
˜ d✓
q(✓) ˜ f (✓) d✓ = F (✓) ˜ d✓
q(✓) ˜ F (✓)q(✓) d✓
✓ ✓ ✓ ✓=✓ ✓
Z✓ Z✓
= ˜ d✓
q(✓) ˜ F (✓)q(✓) d✓
✓ ✓
Z✓
= (1 F (✓))q(✓) d✓.
✓

Substituting into the seller’s objective function, we have

Z✓( ! )
(1 F (✓))
✓ q(✓) c(q(✓)) f (✓) d✓.
✓ f (✓)

Consider maximizing the integrand pointwise. Since the cost

function c is convex, the first order condition is sufficient for an
interior maximum. If q = q(✓) maximizes the integrand, it satisfies
the first order condition

1 F (✓)
✓ = c 0 (q). (10.2.5)
f (✓)

There remains the possibility that there may not be a strictly posi-
tive q solving (10.2.5), as would occur if the left side were negative.
6
An alternative derivation is via Fubini’s theorem (where ˜ ✓) = 1 if ✓
(✓, ˜ ✓
and 0 otherwise):
Z✓ Z✓ Z✓ Z✓
˜ d✓
q(✓) ˜ f (✓) d✓ = ˜ ✓)q(✓)f
(✓, ˜ (✓) d✓d✓
˜
✓ ✓ ✓ ✓
Z✓ Z✓
= ˜ ✓)q(✓)f
(✓, ˜ (✓) d✓d✓
˜
✓ ✓
Z✓ Z✓
= ˜ (✓) d✓d✓
q(✓)f ˜
✓ ˜
✓
Z✓ Z✓ Z✓
= ˜
q(✓) ˜=
f (✓) d✓d✓ (1 ˜
F (✓))q( ˜ d✓.
✓) ˜
✓ ˜
✓ ✓
298 Chapter 10. Introduction to Mechanism Design

Theorem 10.2.1. Suppose the function

1 F (✓)
(✓) := ✓ (10.2.6)
f (✓)
is nondecreasing in ✓. For all ✓, denote by q⇤ (✓) the quantity given
by (
⇤ 0, (✓)  0,
q (✓) :=
q(✓), (✓) > 0,
where q(✓) is the unique quantity q satisfying (10.2.5), and denote
by p ⇤ (✓) the right side of (10.2.4), i.e.,
Z✓
⇤ ⇤
p (✓) := ✓q (✓) ˜ d✓.
q⇤ (✓) ˜
✓

The pair of functions (q⇤ , p ⇤ ) maximizes the seller’s profits (⇧) sub-
ject to (IC✓ ) and (IR✓ ). Moreover, any pair of functions (q̂⇤ , p̂ ⇤ ) that
maximizes the seller’s profits (⇧) subject to (IC✓ ) and (IR✓ ) equals
(q⇤ , p ⇤ ) almost surely.
Proof. If (✓) > 0, since the marginal cost is unbounded above, con-
tinuous with value 0 at 0, there exists a unique q solving (10.2.5).7
Since is nondecreasing in ✓, the function q⇤ is also nondecreasing
and so by Lemma 10.2.2, (q⇤ , p ⇤ ) is an admissible solution. Since
q⇤ (✓) uniquely pointwise maximizes the integrand, it is the opti-
mal quantity schedule. Since the optimal choices for the seller must
leave the bottom buyer indifferent between accepting and rejecting
the contract, U (✓) = 0 and the optimal price schedule is given by
(10.2.4).

The function defined in (10.2.6) is called the virtual type or

virtual value, and that assumption that is increasing in ✓ is called
the regular case. Note that there is no distortion at the top (since
F (✓) = 1). Moreover, every type ✓ earns an information rent of
Z✓
˜ d✓
q⇤ (✓) ˜
✓

(which is zero for ✓ = ✓ ).

7
If c 0 is bounded above, (10.2.5) may not have a solution. In particular, if (✓)
exceeds c 0 (q) for all q, the integrand is unbounded at that ✓, precluding the exis-
tence of a maximum over all q 0. This is the only place the assumption that c 0 is
unbounded above is used.
10.3. The Take-It-or-Leave-It Mechanism 299

Remark 10.2.1 (The Taxation Principle). In practice, it is more nat-

ural to think of the buyer as choosing a quantity q while facing a
nonlinear pricing schedule, P (as in Section 10.1). Such a nonlin-
ear schedule is said to screen buyers. The observation that one can
always implement an optimal direct mechanism by an equivalent
nonlinear pricing schedule is called the taxation principle. Problem
10.5.2 goes through the steps for the current setting. The taxation
principle is an “inverse” of the revelation principle.

10.3 The Take-It-or-Leave-It Mechanism

We continue with the previous section’s example, but now consider
the case of a seller selling one unit of a good to a buyer with un-
known valuation. The preceding analysis applies once we interpret
q as the probability of trade, which of course requires q 2 [0, 1].
Conditional on a report ✓,ˆ there is a probability of trade q(✓)
ˆ and
ˆ
a payment p(✓). The seller has an opportunity cost of provision of
c < ✓, so the expected cost is

c(q) = cq.

Suppose we are in the regular case, so that

1 F (✓)
(✓) = ✓
f (✓)
is increasing in ✓. Note that E (✓) = ✓.
The seller chooses a nondecreasing probability of trade to maxi-
mize
Z✓( )
1 F (✓)
✓ c q(✓) f (✓) d✓.
✓ f (✓)
Denote by ✓ ⇤ a value of ✓ that solves
1 F (✓)
✓ = c;
f (✓)
note that ✓ ⇤ > c. The maximum is clearly achieved by choosing
(
⇤ 1, if ✓ ✓ ⇤ ,
q (✓) =
0, if ✓ < ✓ ⇤ .
300 Chapter 10. Introduction to Mechanism Design

From (10.2.4), it is immediate that

(
✓⇤, if ✓ ✓ ⇤ ,
p ⇤ (✓) =
0, if ✓ < ✓ ⇤ .

In other words, the seller optimally chooses a price p = ✓ ⇤ , every

buyer with a valuation above ✓ ⇤ buys, and every buyer with a valu-
ation below ✓ ⇤ does not.
The conclusion that the take-it-or-leave-it mechanism is optimal
does not require being in the regular case (though the proof does
require more advanced techniques, see Section 2.2 of Börgers, 2015).

10.4 Implementation
We begin with a simple example of implementation.

Example 10.4.1. Two children (Bruce and Sheila) have to share a pie,
which has a cherry on top. Both Bruce and Sheila would like more
of the pie, as well as the cherry, with both valuing the cherry at
✓ 2 (0, 1). A division of the pie is denoted by x 2 [0, 1], where x is
the size of the slice with the cherry, with the other slice having size
1 x (it is not possible to divide the cherry). An allocation is (x, i)
where i 2 {1, 2} is the recipient of the slice with the cherry. Payoffs
are (
✓ + x, if j = i,
vj (x, i) =
1 x, if j î i,

for i, j 2 {B, S}. If the parent knew that ✓ = 0, then the parent can
trivially achieve a fair (envy-free) division by cutting the pie in half.
But suppose the parent does not know ✓ (but that the children do).
How can the parent achieve an equitable (envy-free) division? The
parent has the children play the following game: Bruce divides the
pie into two slices, and Sheila then chooses which slice to take (with
Bruce taking the remaining slice). It is straightforward to verify that
in any subgame perfect equilibrium, the resulting division of the pie
is envy-free: both Bruce and Sheila are indifferent over which slice
they receive. ´

Let E denote a set of environments and Z the outcome space.

10.4. Implementation 301

Definition 10.4.1. A social choice function ( scf) is a function

⇠ : E ! Z.

Example 10.4.2 (Social choice). Let Z be a finite set of social alterna-

tives, {x, y, z, . . .}. Let R denote the collection of preference order-
ings on Z; R 2 Z satisfies completeness and transitivity. We write
xP y if xRy but not yRx (that is, x is strictly preferred to y). There
are n members of society, so E = Rn .
A social choice function ⇠ satisfies unanimity if for all e 2 E
satisfying xPi y for all y 2 Z and all i, we have ⇠(e) = x. ´

Example 10.4.3. A single indivisible good to be allocated to one of

n people. Then Z = {0, 1, . . . , n} ⇥ Rn , where (j, t1 , . . . , tn ) 2 Z is
the outcome where j receives the good (if j = 0, no one receives
the good), and agent i pays the seller ti (an amount that may be
negative).
If player i values the good at vi , then i’s payoffs are
(
vi ti , if i = j,
ui (j, t1 , . . . , tn ) =
ti , if i î j.
Q
The set of environments is E = i Vi , where Vi is the set of
possible valuations for player i, so that e = (v1 , . . . , vn ).
Efficient scf: ⇠(e) = (j, t1 , . . . , tn ) only if vj maxi vi .
Maximally revenue extracting and individually rational scf: ⇠(e) =
(j, t1 , . . . , tn ) only if vj maxi vi , tj = vj , and ti = 0 for i î j. ´

While the players know the details of the environment (though

some of the information may be private), the “social planner” may
not (Example 10.4.1 is a simple illustration). Nonetheless, the social
planner wishes to “implement” a social choice function. Can the
social planner arrange matters so that the players will reveal the
details of the environment? This is the implementation problem.
More formally, a game form or mechanism is described by
((Ai )n
i=1 , f ), where Ai is player i’s action set, and f : A ! Z is the
outcome function describing the induced outcome in Z for each ac-
tion profile a 2 A := A1 ⇥· · ·⇥An . If the mechanism has a nontrivial
dynamic structure, then ai is an extensive form strategy.
For each mechanism and environment e, there will be a set of
solutions or equilibria ⌃ (e) ⇢ A. In general, there are many choices
302 Chapter 10. Introduction to Mechanism Design

⇠
E Z
⌃ f
A

Figure 10.4.1: The mechanism ⌃-implements ⇠ if this diagram is a com-

mutative diagram, in the sense of Definition 10.4.2.

possible for ⌃ , such as dominant strategy, Nash equilibrium, Bayes-

Nash, subgame perfect, etc. The specification of ⌃ depends upon
the nature of informational assumptions made about the players
and the static/dynamic nature of .

Example 10.4.4. Suppose there are two players and three possibili-
ties in Example 10.4.2. Voting by veto (Example 1.1.5) is an example
of a game form, and using iterated deletion of dominated strategies
gives, for each profile of preferences, an outcome. ´

We thus have the scf ⇠ which maps E to Z, the solution corre-

spondence ⌃ , which maps E to A, and outcome function f , which
maps A to Z (see Figure 10.4.1).

Definition 10.4.2. The mechanism weakly ⌃-implements ⇠ if for all

e 2 E, there exists a solution (e) 2 ⌃ (e) such that

f( (e)) = ⇠(e).

The mechanism strongly (or fully) ⌃-implements ⇠ if for all e 2 E,

and for all solutions (e) 2 ⌃ (e),

f( (e)) = ⇠(e).

If implement is used without an adjective, the writer typically

means weakly implement.

Example 10.4.5. The second price auction strongly implements an

efficient scf in the setting of Example 10.4.3 in weakly undominated
strategies (Example 1.1.6), but it only implements (but not strongly
so) in Nash equilibrium (Example 2.1.4). ´
10.5. Problems 303

Example 10.4.6 (Example 10.4.1 cont.). Fair division is strongly im-

plemented by both subgame perfect equilibrium and (in the normal
form) by the iterated deletion of weakly dominated strategies. ´

Remark 10.4.1. Implementation theory is part of mechanism design

and fixes the social choice function and asks whether there is mech-
anism for which either (for weak implementation) there is an equilib-
rium or (for strong implementation) all equilibria result in outcomes
consistent with the a priori fixed social choice function. Mechanism
design does not necessarily fix the social choice function. A desired
outcome or objective is first identified (examples include efficient
trade, firm profits, social welfare, and gains from trade) and then
solves for (designs) a mechanism with an equilibrium that either
achieves or maximizes the objective (as appropriate) subject to the
relevant resource and information constraints.

10.5 Problems
10.5.1. In this problem, we explicitly take into account the nonneg-
ativity constraints in the simple screening example of Section
10.1. The Langrangean becomes (since the nonnegativity of
prices is implied by the nonnegativity of quantities and IR,
those constraints can be ignored)

L = ↵L (pL c(qL )) + ↵H (pH c(qH ))

+ L [✓L qL pL (✓L qH pH )] + H [✓H qH pH (✓H qL pL )]
+ µL (✓L qL pL ) + µH (✓H qH pH ) + ⇠L qL + ⇠H qH ,
where ⇠L and ⇠H are the multipliers on the nonnegativity con-
straints qL 0 and qH 0, respectively.
(a) Describe the first order conditions and the complementary
slackness conditions.
(b) Suppose ✓L < ↵H ✓H . Solve for the optimal pair of con-
tracts (including the values of the multipliers).
(c) Suppose now that the marginal cost at zero is strictly pos-
itive, while maintaining all the other assumptions in Sec-
tion 10.1. How does the analysis of the seller’s optimal
pair of contracts change?
304 Chapter 10. Introduction to Mechanism Design

10.5.2. Suppose (p, q) is an optimal direct mechanism for the less

simple screening example of Section 10.2 and we are in the
regular case.
(a) Prove that p is strictly increasing whenever q is (i.e., ✓1 <
✓2 and q(✓1 ) < q(✓2 ) implies p(✓1 ) < p(✓2 )).
(b) Prove that p is constant whenever q is (i.e., ✓1 < ✓2 and
q(✓1 ) = q(✓2 ) implies p(✓1 ) = p(✓2 )).
(c) (The taxation principle.) Let Q := q([✓, ✓]). Prove there
exists a nonlinear pricing function P : Q ! R+ such that
buyer ✓ optimally chooses q = q(✓) from Q.
10.5.3. Solve for the optional direct mechanism (p, q) for the less
simple screening example of Section 10.2 when costs are quadratic,
1
in particular, c(q) = 2 q2 , and ✓ has the following distributions.
(a) A uniform distribution on the interval [0, 1].
(b) The type ✓ is distributed on [1, 11] according to the fol-
lowing density:
8
> ✓ 2 [1, 2],
<0.1,
>
2
f (✓) = 0.9(✓ 1) , ✓ 2 (2, 10),
>
>
:0.1, ✓ 2 [10, 11].

10.5.4. Prove that the take-it-or-leave-it mechanism is optimal for

the sale of a single item when the seller’s beliefs are as de-
scribed in Section 10.1.
10.5.5. Give the (few) details of the argument that U 0 (✓) = q(✓) in
the proof of Lemma 10.2.1.
10.5.6. Suppose a seller wishes to sell one unit of a good to a buyer
of unknown valuation who is budget constrained. That is, as-
sume the buyer in Section 10.3 cannot pay more than his bud-
get of b > 0. Suppose we are in the regular case. If b > ✓ ⇤ , then
this additional constraint is not binding. What is the firm’s op-
timal mechanism if b < ✓ ⇤ ?
10.5.7. Verify the claim in Example 10.4.1 that the division of the
pie in any subgame perfect equilibrium is envy free (neither
child strictly prefers one of the slices).