AUCTIONS (of a single object)

Standard procedures

English/ascending (with the “japanese” variant)

(Sealed-bid) …rst-price

Dutch/descending

(Sealed-bid) second-price (Vickrey)

Basic setups

(I.I.D) private values

Common value (“mineral rights”)

“A¢ liated” values

The Dutch auction and the …rst-price auction are strategically equivalent.

Under private values, “bidding his own value” is a weakly dominant strategy in
the English auction and the second-price auction (which are thus treated as a
single format).
Private values

Bidders i = 1; :::; n; the seller is not a player.

Nature chooses the value vi 2 [0; 1] of bidder i, with Fi (of density fi) as
marginal distribution; v = (v1; :::; vn).

In a sealed-bid auction, every bidder makes a bid bi, b =(b1; :::; bn).

Ordered bids: min1 i n bi = b(1) < < b(n 1) < b(n) = max1 i n bi

Highest bid: b(n)

Second-highest bid: b(n 1)

Payo¤ functions (neglecting possible ties):

First-price: ui(v; b) = (vi b(n))I (bi = b(n)).

Second-price: ui(v; b) = (vi b(n 1))I (bi = b(n)).

A (pure) strategy of bidder i is a mapping: i : [0; 1] ! R : vi ! i(vi).

Proposition: In a second-price auction, for every bidder i, i (v i ) = vi is a

weakly dominant strategy.

Proof: exercise (see, e.g., M1 course).

Second-price expected revenue RSP A of the seller (at the dominant strategies
equilibrium):

assuming that the vi’s are i.i.d. according to F (with density f ):

R1
RSP A = EF (v(n 1)) = 0 vg (v )dv ,
h i
where g is the density of v(n 1): g (v ) = d Pr(v(n 1) v) .
dv

Pr(v(n 1) v ) = (F (v ))n + n(1 F (v ))(F (v ))n 1

n(F (v ))n 1 (n 1)(F (v ))n.

Example: in the uniform case, RSP A = n 1.

n+1
h i
r , which is quite intuitive.
N.B.: In the uniform case, E v(r) = n+1
Symmetric equilibrium of the …rst price auction, still assuming the vi’s i.i.d.

Bidder’s strategy: : [0; 1] ! R strictly increasing, di¤erentiable, such that

(0) = 0.

Let us …nd bidder 1’s best reply to 2 = ::: = n = .

Let bidder 1’s value v1 = v ; if he bids b, his payo¤ (as a r.v.) is

(v b)I [max f (v2); :::; (vn)g b] = (v b)I [h (vi) b; i = 2; :::; n] i

(v b)I vi 1 (b); i = 2; :::; n

Using the independence of the vi’s, his expected payo¤ is

h i
(v b) F ( 1 (b)) n 1 (*)
 de…nes an equilibrium , (*) is max at b = (v ):

h i
(v (v )) [F (v )] n 1
(v b) F ( 1 (b)) n 1 for every b 2 [0; 1]

, (v (v )) [F (v )]n 1 (v (t)) [F (t)]n 1 for every t 2 [0; 1]

, (v (t)) [F (t)]n 1 is max at t = v

(can be interpreted as in the “revelation principle”).

1 Rv d n n 1
o R
F.O.C. ) (v ) = t [F (t)] dt = 01 tgv (t)dt.
[F (v)]n 1 0 dt

Proposition: In a …rst-price auction,

n with private
o values i.i.d. according to F ,
1 R v n 1
bidding (v ) = n 1 0 td [F (t)] is an equilibrium (and it is unique
[F (v)]
within the class of symmetric well-behaved equilibria).
Example: in the uniform case, (v ) = nn 1 v , which yields the seller’s expected
revenue
h i
n 1E = nn 1 n+1
n = n 1 =R
RF P A = n F v(n) n+1 SP A .

Generalization: First-price expected revenue RF P A of the seller under i.i.d.

private values, at the symmetric, increasing, equilibrium

R1
Interpretation of (v ) = 0 tgv (t)dt: gv is the density of of the max of n 1
r.v.’s i.i.d. according to F , conditionally on this max being v.
h i
(v1) = EF max2 i n vi j max2 i n vi v1

max2 i n vi v1 , max1 i n vi = v1
 h i
(v ) can be interpreted as the realization at v of the r.v. EF v(n 1) j v(n) .

h i n h io h i
RF P A = EF (v(n)) = EF EF v(n 1) j v(n) = EF v(n 1)

) RF P A = RSP A.
Auction mechanisms: general revenue equivalence theorem

General auction game between our bidders i = 1; :::; n

(the seller or auctioneer is there, but not a player)

Initial stage: as before, values vi 2 [0; 1] with marginal distribution Fi,

v = (v1; :::; vn).

Intermediate stages: messages are exchanged according to rules

(e.g., successive bids, o¤ers-countero¤ers, ...).

Final stage: the object is allocated, payments are made (positive or negative,
independently of getting the object, e.g., entry fees and subsidies are allowed).

By contrast with the four standard auction formats.

N.B.: An auction game is a dynamic game with incomplete information.

Assume = ( i)1 i n is a mixed (Bayesian) Nash equilibrium of ; i is

bidder i’s ex ante strategic plan.

Every pair ( ; ) induces (pi( ; ); ci( ; ))1 i n:

pi( ; )(v ) = Pr( ; )(bidder i gets the objectj v ), i = 1; :::; n,

Pn
such that i=1 pi( ; )(v ) 1, and

ci( ; )(v ) = E( ; )(payment from bidder i j v ), i = 1; :::; n.

Imagine that the (benevolent) auctioneer asks every bidder i to report a value
and guarantees to use (p( ; ); c( ; )) to give the object to some bidder and
to make bidders pay some amount.
We face a new auction game D( ; ), with a single intermediate stage:

The bidders are simultaneously asked to report a valuation; let r = (ri)1 i n.

Bidder i gets the object with probability pi( ; )(r) and pays ci( ; )(r),
i = 1; :::; n.

Strategy for bidder i in D( ; ): i : [0; 1] ! [0; 1] : vi ! i(vi) = ri.

Claim: If is an equilibrium of , the truthful strategies i(vi) = vi,

i = 1; :::; n, form an equilibrium of D( ; ), with the same payo¤s as
(and thus the same revenue for the seller).

This result is a version of the revelation principle.

Proof: Consider bidder i with value vi. Let i(vi) be his interim strategic
plan, which describes what to do, according to i, as a function of his informa-
tion vi, at every intermediate stage of . Taking for granted that all bidders
j 6= i truthfully report their value vj and that the auctioneer faithfully uses
(p( ; ); c( ; )), reporting ri amounts to playing according to i(ri) in .
Hence truthfully reporting his value vi is a best reply of bidder i.

Important consequence of the revelation principle: if we are interested in the

seller’s revenue, we can focus on direct selling mechanisms, like the one that is
behind the new auction game above.

De…nition: A direct selling mechanism (p; c) consists in

pi : [0; 1]n ! [0; 1] : (r1; :::rn) ! pi(r1; :::rn), i = 1; :::; n, such that
P
i pi 1
ci : [0; 1]n ! R : (r1; :::rn) ! ci(r1; :::rn), i = 1; :::; n.
Interpretation: (p; c) is associated with the following auction game:

The bidders are simultaneously asked to report a valuation; let r = (ri)1 i n.

Bidder i gets the object with probability pi(r) and pays ci(r), i = 1; :::; n.

De…nition: The direct selling mechanism (p; c) is incentive compatible (I.C.)

, truthful report of vi by every bidder i is an equilibrium of the auction game
associated with (p; c).

Revelation principle: Let be an auction game and let be an equilibrium

of . Then and induce an incentive compatible direct selling mechanism,
with the same payo¤s and revenue for the seller as .
Explicit I.C. conditions

Let (p; c) be a direct selling mechanism. Assume bidder i reports ri and bidders
j 6= i truthfully report their value vj :

pi(ri) = Ev i (pi(ri; v i)) = expected probability that bidder i gets the object

ci(ri) = Ev i (ci(ri; v i)) = expected payment to be made by bidder i

ui(ri; vi) = pi(ri)vi ci(ri) = bidder i’s interim expected payo¤ when his
type is vi and he reports ri.

(p; c) is I.C.

, for every i, vi, ri: ui(vi; vi) ui(ri; vi).

, for every i, vi: maxri2[0;1] ui(ri; vi) = ui(vi; vi).

Characterization of I.C. direct selling mechanisms

(p; c) is I.C. , for every bidder i,

pi( ) is non-decreasing and

R vi
for every vi 2 [0; 1]: ci(vi) = ci(0) + pi(vi)vi 0 pi(r )dr .

As soon as p( ) and ci(0) are determined, the remainder of ci( ) is determined

as well.
Important consequence of the characterization:

Revenue equivalence

If two I.C. direct selling mechanisms have the same probability assessment
functions p( ) and every bidder with value 0 is indi¤erent between the two
mechanisms, then the two mechanisms generate the same expected revenue for
the seller.

Application to the four standard auctions with symmetric bidders:

all of them assign the object to the bidder with the highest value, and in each
of them, a bidder with value 0 has 0 expected utility.