EC319 Economic Theory and Its Applications,
Part II: Lecture 3
Leonardo Felli
NAB.2.14
30 January 2014
Optimal Auction Design
I Consider a seller who wishes to design an auction to maximize
his/her expected revenue.
I Big task: to specify the many different auctions the seller
would consider.
I The bidder might have to pay an entry fee, or the losing
bidders might have to pay money as a proportion of their bids.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 2 / 40
Optimal Auction Design (contd)
I Alternatively the seller may set a reservation price: a floor
below which bids will not be accepted.
I Fortunately, the seller might use the revelation principle to
simplify this problem dramatically.
I In particular, the seller may restrict attention to the following
class of games.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 3 / 40
Direct Mechanism
These are called direct mechanisms:
I the bidders simultaneously make possibly dishonest claims
about their types (valuations). Player i of type ti may claim
to be of type i Ti , i 6= ti .
I given each bidders claims (1 , . . . , I ), bidder i pays
xi (i , . . . , I ) and receives the good with probability
qi (i , . . . , I ) such that
I
X
qi (i , . . . , I ) = 1.
i=1
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 4 / 40
Direct Mechanism (contd)
I The second way the seller can use the revelation principle to
simplify his problem is to restrict attention to the direct
mechanisms in which it is a Bayesian Nash equilibrium for
each bidder to report the truth.
I In every direct mechanism a players strategy is the
announcement i that player i of type ti chooses: i (ti ).
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 5 / 40
The Revelation Mechanism
A direct mechanism in which truth-telling is a Bayesian Nash
equilibrium is called incentive-compatible and is such that in
equilibrium: i (ti ) = ti for all ti Ti
The revelation principle is a result that shows that there is no loss
in generality in restricting attention to incentive-compatible direct
mechanisms when solving the sellers problem.
Theorem (Revelation Principle)
Any Bayesian Nash equilibrium of any Bayesian game
corresponds to the Bayesian Nash Equilibrium of an
incentive-compatible direct mechanism.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 6 / 40
Revelation Principle (contd)
I This is the Bayesian Nash equilibrium of the direct mechanism
in which each player tells the truth.
I The incentive-compatible direct revelation mechanism is the
one that is optimal for the seller.
I To understand how this principle works consider the auction
game we solved before:
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 7 / 40
Example Auction
I Two bidders N = {1, 2};
I Action spaces:
A1 = {b1 0} A2 = {b2 0}
I Type spaces:
T1 = {0 v1 1} T2 = {0 v2 1}
I Beliefs of both players:
1 = 1, 2 = 1
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 8 / 40
Example Auction (contd)
I Payoffs:
v1 b1 if b1 > b2
1
u1 (b1 , b2 ) = (v 1 b 1 ) if b1 = b2
2
0 if b1 < b2
and
v2 b2 if b2 > b1
1
u2 (b1 , b2 ) = (v 2 b 2 ) if b2 = b1
2
0 if b2 b1
I Strategies in this game are:
b1 = b(v1 ), b2 = b(v2 )
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 9 / 40
Example Auction (contd)
Revelation principle tells us that we can re-write this game as a
direct mechanism with the following features.
I Two bidders: N = {1, 2};
I Action spaces:
A01 = T1 = {0 v1 1} A02 = T2 = {0 v2 1}
I Type spaces:
T1 = {0 v1 1} T2 = {0 v2 1}
I Beliefs of both players: 1 = 1, 2 = 1
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 10 / 40
Example Auction (contd)
I Payoffs: 1 (v1 , v2 ) = u1 (b(v1 ), b(v2 )) or
v b(v1 ) if b(v1 ) > b(v2 )
1
1
1 (v1 , v2 ) =
2 (v1 b(v1 )) if b(v1 ) = b(v2 )
0 if b(v1 ) < b(v2 )
and 2 (v1 , v2 ) = u2 (b(v1 ), b(v2 )) or
v b(v2 ) if b(v2 ) > b(v1 )
2
1
2 (v1 , v2 ) =
2 (v2 b(v2 )) if b(v2 ) = b(v1 )
0 if b(v2 ) < b(v1 )
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 11 / 40
Example Auction (contd)
I Everything is as if there exists an individual, the mechanism
designer, that decides how to reinterpret each player
announcement in terms of a bid: bi = b(vi ).
I Recall that the Bayesian Nash equilibrium of the original game
is such that
1 1
b(v1 ) = v1 , b(v2 ) = v2
2 2
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 12 / 40
Example Auction (contd)
I By definition of Bayesian Nash equilibrium this implies that
the expected payoff to player 1 is:
(v1 b(v1 )) Pr{b(v1 ) > b(v2 )} = (v1 b(v1 )) b 1 (b1 )
I and the expected payoff to player 2 is:
(v2 b(v2 )) Pr{b(v2 ) > b(v1 )} = (v2 b(v2 )) b 1 (b2 )
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 13 / 40
Example Auction (contd)
1
I Neither player wants to deviate from the bids b1 = 2 v1 , and
b2 = 12 v2 .
I In other words, for every function b() player 1s expected
payoff is such that:
1
1 v1
(v1 v1 ) 2 1 (v1 b(v1 )) b 1 (b(v1 ))
2 2
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 14 / 40
Example Auction (contd)
I or (1/2)(v1 )2 (v1 b(v1 )) b 1 (b(v1 ));
I while for every function b() player 2s expected payoffs are
such that:
1
1 v2
(v2 v2 ) 2 1 (v2 b(v2 )) b 1 (b(v2 ));
2 2
I or (1/2)(v2 )2 (v2 b(v2 )) b 1 (b(v2 ))
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 15 / 40
Example Auction (contd)
I The mechanism designer can now make sure that the direct
mechanism is incentive compatible if he reinterprets each
player announcement in terms of a bid in the following way:
bi = (1/2) vi .
I In this case player 1 expected payoff in the direct mechanism
is:
1 1 1
v1 v1 Pr{v1 > v2 } + v1 v1 Pr{v1 = v2 }
2 2 2
1
= v1 v1 v1
2
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 16 / 40
Example Auction (contd)
I Player 1s best reply is:
1
max v1 v1 v1
v1 2
I The first order conditions are: v1 = v1
I Player 2s best reply is:
1
max v2 v2 v2
v2 2
I and v2 = v2
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 17 / 40
Example Auction (contd)
I Telling the truth, vi = vi is a Bayesian Nash equilibrium of
the direct mechanism.
I The direct mechanism that is optimal for the seller is the one
in which each player tells the truth.
I As shown this is the bid function chosen by the mechanism
designer (the seller) that corresponds to:
1
bi (vi ) = vi
2
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 18 / 40
The Revelation Principle
I In general, let s be a Bayesian Nash equilibrium of the game
of incomplete information:
= {N, Ai , Ti , i , ui }
I Define the direct mechanism d to be the new game of
incomplete information such that:
d = {N, Ti , Ti , i , vi }
I The mechanism designer associates a given announcement
i Ti for every player to a strategy choice si (i )
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 19 / 40
The Revelation Principle (contd)
I The payoffs are:
vi (i , i | ti ) = ui [(si (i ), si (i )) | ti ]
I Notice now that if s is a Bayesian Nash equilibrium of the
game it must be the case that for every ti and every si Ai :
Eti ui [(si (ti ), si
(ti )) | ti ]
Eti ui [(si , si (ti )) | ti ]
I The mechanism designer obtains an incentive-compatible
direct mechanism d by choosing the mapping si (i ).
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 20 / 40
The Revelation Principle (contd)
I Indeed, for truth-telling to be a Bayesian Nash equilibrium of
the game d we need that for every i Ti :
Eti ui [(si (ti ), si
(ti )) | ti ]
Eti ui [(si (i ), si
(ti )) | ti ]
I but if we label si (i ) = si Ai we get:
Eti ui [(si (ti ), si
(ti )) | ti ]
Eti ui [(si , si (ti )) | ti ]
I which is satisfied since s is a Bayesian Nash equilibrium of
the game .
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 21 / 40
Centralize the Mechanism
I We move from the initial game to a game where all agents
choose a message mi and then the principal associates the
allocation associated with the strategy s(m) to the message
profile m = (m1 , . . . , mn ).
Definition
A communication mechanism is a game:
c = {N, Mi , Ti , i , ui (s(m) | ti )}
Y
where Mi is a message space for every player: M = Mi .
iN
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 22 / 40
Timing of Revelation Game
I The principal selects the communication game c ,
I The agents simultaneously decide whether to participate,
I The agent simultaneously send their message mi ,
I The principal implements the allocation associated with the
message profile s(m).
Assume that c has a BNE m (t) with allocation s(m (t)) and
assume that all agents participate in stage 2.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 23 / 40
Revelation Principle (Myerson 1979)
Theorem (Revelation Principle for BNE)
For every BNE m (t) of c , there exists a direct revelation
mechanism,
d = {N, Mi = Ti , Ti , , ui (s(t) | t)}
with strategies ti Mi = Ti such that:
I truth-telling is a BNE equilibrium t (t) of d :
ti = ti , i N
I d implements the same allocation as m (t):
s(t) = s(m (t)).
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 24 / 40
Static Adverse Selection
I Static Adverse Selection problem: consider the simple
monopolist pricing model.
I A monopolist (the principal) is choosing the pricing scheme of
a commodity for a consumer (the agent) who has private
information on his preferences for the commodity (his type).
I The seller sets the terms of the contract (take-it-or-leave-it
from the principal to the agent).
I The seller does not know how much the buyer is willing to pay
for the commodity.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 25 / 40
Static Adverse Selection (contd)
I The buyers preferences are represented by:
U(q, T , i ) = i u(q) T
I T total transfer from the buyer to the seller,
I i buyer is marginal valuation for the commodity,
I u(q) common component of the buyers preferences:
u 0 () > 0, u 00 () < 0, u(0) = 0, lim u 0 (q) = +.
q0
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 26 / 40
Static Adverse Selection (contd)
I The seller chooses (T , q) so as to maximize its profit:
=T cq
I Assume that:
i {L , H } and = Pr{i = L }
I Let U = 0 be the buyers outside option (we consider a
monopoly hence there exists no alternative seller).
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 27 / 40
First best:
I Assume first that the seller is perfectly informed on each
buyers type i .
I The contract is then (Ti , qi ), for i {L, H} such that:
i u 0 (qi ) = c, i {L, H}
and
Ti = i u(qi ), i {L, H}
I The sellers total expected profit:
(TL c qL ) + (1 ) (TH c qH
)
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 28 / 40
Second best:
I If the seller cannot observe the the buyers type then she has
to offer the same contract to both types.
I In other words the seller may offer to the agent (whatever his
type) a set of choices
(T (q), q)
I In principle the contract space is potentially large: the set of
functions T (q), of all shapes and features.
I Fortunately, the Revelation Principle simplifies the search for
the best contract from the principals perspective.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 29 / 40
Revelation Principle
I If the principal has all the bargaining power he chooses the
mechanism which has the best equilibrium from her view
point (mechanism design).
I The principal is a Stackelberg leader, he selects the game the
agent will play.
I The game played by the agent is extremely simple: one player,
the agent, choosing an element in the schedule (T (q), q).
I The revelation principle identifies the set of all possible
mechanisms among which the principal selects.
I We proceed in Six Steps.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 30 / 40
The principals problem:
Step 1: By revelation principle the principal focuses on the direct
mechanisms
(Ti , qi ) = (T (q(i )), q(i )), i {L, H}
such that
I in equilibrium the agent wants to buy the commodity,
I and the agent reports the truth.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 31 / 40
The principals problem (contd)
This mechanism corresponds to the solution to the following
principals problem:
max (TL c qL ) + (1 )(TH c qH )
Ti ,qi
s.t. H u(qH ) TH H u(qL ) TL
L u(qL ) TL L u(qH ) TH
H u(qH ) TH 0
L u(qL ) TL 0
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 32 / 40
The principals problem (contd)
I The Principal maximizes her expected payoff:
(TL c qL ) + (1 )(TH c qH ).
I Subject to the individual rationality constraint (IR) that
guarantee that in equilibrium both types of agent want to
participate in the game:
H u(qH ) TH 0
L u(qL ) TL 0
I Subject to the incentive compatibility constraint (IC) that
guarantee that in equilibrium the agent will report the truth:
H u(qH ) TH H u(qL ) TL
L u(qL ) TL L u(qH ) TH
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 33 / 40
The Optimal Solution
Step 2: The individual rationality constraint of the type H will not
bind at the optimum.
Indeed since H > L :
H u(qH ) TH H u(qL ) TL > L u(qL ) TL 0
Step 3: Solve the relaxed problem that ignores the (ICL ) constraint.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 34 / 40
The Optimal Solution (contd)
To select which constraint to omit consider the two (IC)
constraints at the first best optimum:
H u(qH ) TH = 0
and
H u(qL ) TL = (H L ) u(qL ) > 0
.
while
L u(qL ) TL = 0
and
L u(qH ) TH = (L H ) u(qH
)<0
.
Therefore the key (IC) constraint is the one of the H-type.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 35 / 40
Spence-Mirrlees Condition
The reason why the (IC) constraint of only one type of agent binds
is Spence-Mirrlees Single Crossing Condition:
U/q
= u 0 (q) > 0
U/T
Marginal utility of consumption (relative to the marginal utility of
money) rises with the type of the agent .
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 36 / 40
The Optimal Solution (contd)
Step 4: Notice that the relaxed problem is such that both
constraints bind at the optimum:
max (TL c qL ) + (1 )(TH c qH )
Ti ,qi
s.t. H u(qH ) TH H u(qL ) TL
L u(qL ) TL 0
If (ICH ) does not bind then the principal can raise TH without
affecting (IRL ), while improving the maximand.
The maximand is monotonic increasing in TL while (IRL ) is
monotonic decreasing in TL .
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 37 / 40
The Optimal Solution (contd)
Step 5: Eliminate TL and TH from the binding constraints and
substitute them into the maximand.
We get:
max [L u(qL ) c qL ] +
qi
+ (1 ) [H u(qH ) (H L ) u(qL ) c qH ]
The second best contract (qi , Ti ) is then the solution to the
unconstraint maximization problem above.
To characterize the solution we distinguish two cases.
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 38 / 40
The Optimal Solution (contd)
Case 1: [ L (1 ) (H L )] 0
In this case the slope of the maximand with respect to qL is strictly
negative for every qL 0:
[ L (1 ) (H L )] u 0 (qL ) c < 0
Therefore the principal chooses qL at a corner:
qL = 0, TL = 0
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 39 / 40
The Optimal Solution (contd)
In other words the principal decides not to serve the type L of the
agent.
The principal then serves only the type H of the agent:
qH = qH , TH = TH
Recall that in this case the (ICL ) constraint we omitted is satisfied:
L u(qH ) TH < 0
Leonardo Felli (LSE) EC319 Economic Theory and Its Applications 30 January 2014 40 / 40
