Notes on Auction Theory

April 18, 2021

1
 1 Introduction
Auctions are widely used methods of allocating resources. A typical auction involves two
types of agents: (a) sellers, those who own the resources and want to sell them; (b) buyers,
those who do not own the resources but want to buy them. Depending on the number of
buyers and the number of sellers, one can roughly classify the settings as follows.

• The most commonly studied auctions are those where there is a single seller, who is
interested in selling a single object (or multiple objects) to a set of buyers. We will be
mainly interested in analyzing such auctions. Examples of such auction setting include:
Govt selling rights to mine to various companies; Google selling advertisement slots on
search pages; Used cars sold on various websites (cars24.com) using auction. In these
examples, there is a single seller (for instance, the owner of the used car) who is selling
the object she owns to a set of buyers (those who logged in to the website to buy the
car). In some of these cases, the seller need not be the auctioneer, but an intermediary
agent conducts the auction.

• An analogous theory of auctions can also be developed for settings where there is a
single buyer who is interested to buy an object and there are many sellers who can
sell or supply the object. These are procurement auctions. Procurement auctions
are mainly used by firms to procure raw materials for manufacturing. They are also
used by Governments and other organizations to procure vaccines and other medical
supplies. The analysis of auctions for single seller and multiple buyers can be straight-
forwardly adapted to the procurement settings. However, there are other concerns in
a procurement auction, which separates it from normal auction setting. For instance,
consider procurement of a raw material (say, spare parts of a car) by a firm (a car
manufacturer). Several suppliers (sellers) can supply the raw material. The firm is
interested in two dimensions of the raw material: (a) price and (b) quality. Each sup-
plier can supply the raw material at different (price, quality) pairs. The buyer (firm)
has to choose a supplier by considering offers of suppliers in both the dimensions. A
standard method to aggregate these offers is through scoring rule, where a weight is
given to each dimension and the aggregated score of each supplier is used to select the
final supplier.

2
 • There are settings where multiple sellers simultaneously sell their objects to a set
of buyers. These are double auctions, where sellers post ask prices and buyers put
offers/bids, and a market-clearing mechanism matches buyers and sellers. While used
in many settings, we will not cover such auctions. Double auctions usually have the
additional requirement that trade has to be budget-balanced: payments received by
the sellers must equal the buyers’ payments.

Unless stated explicitly, we will only be discussing settings with a single seller and mul-
tiple buyers, and that too in a single object model.

Why auction? The most prominent procedure for selling products is the posted-price mech-
anism. The posted-price mechanism is an excellent procedure when (a) the seller has a good
idea about the willingness to pay of buyers; (b) the buyers cannot come together to an auc-
tion. If the seller does not have a good idea of the willingness to pay of buyers, then the
seller can potentially get low revenues from posted-price mechanism: too low a posted price
generates low revenue and too high a posted price reduces the probability of winning. On
the other hand, auction allows us a discovery of willingness to pay.

Tools for analysis. The analysis of auctions is based on game theory. The willingness to pay
information is private to individual buyers. Hence, an auction setting induces a Bayesian
game of incomplete information.

2 Standard auction formats

We see various auction formats in practice (for selling a single object). Broadly, these
auctions can be classified into two categories:

(a) sealed-bid auctions; These are auctions where bidders submit a one-time bid and winner
and payments are decided based on these bids.

(b) open-cry auctions; These are auctions where prices are announced iteratively and de-
mands of bidders at these prices are elicited. The auction ends when demands of
bidders equal supply.

3
 Under sealed-bid auctions, there are many variants. The two most common variants are
(a) first-price auction and (b) second-price auction. In both the auctions, bidders (buyers)
place bids and the bidder with the highest bid wins the auction. In both the auctions, a
bidder pays only if she wins the object. The auctions differ in their payment rule: (a) in
the first-price auction, the winner pays her own bid; (b) in the second-price auction, the
winner pays the second-highest bid. While these are two popular sealed-bid auctions, there
are other sealed-bid auctions which are studied in the auction theory literature. One such
auction format is called the all-pay auction. As the name suggests, in an all-pay auction,
the highest bidder wins the object but every bidder (including losers) pay their bid. Such
auction are used to model contests, where the effort level works as a proxy for bid, which is
paid by every bidder.
In open-cry auctions, there are two popular auction formats: (a) ascending price auction
(English auction) and (b) descending price auction (Dutch auction). While various imple-
mentations of these auctions are present, it is convenient to think of the continuous clock
implementation. In this implementation, the seller keeps a continuous price clock. In the
ascending price auction, this price clock starts at a low (zero) price and the price keeps in-
creasing continuously. Bidders can decide to exit the auction at any time during the auction.
Once a bidder exits the auction, she may not come back. The price clock stops as soon as
there is exactly one bidder remaining in the auction.1 At that point, the only other bidder
remaining wins the auction and pays the price in the auction clock.
There are practical benefits of each auction. For instance, price-based auctions, like the
English and the Dutch auctions are transparent procedures with a lot of privacy preserving
features. On the other hand, they require presence of bidders when auction takes place and
can become complex in terms of communication. The sealed-bid auctions can allow bidders
to send bids by communicating them beforehand. A sealed-bid auction is a centralized
algorithm where inputs are processed centrally by the seller. On the other hand, ascending
and descending price auctions are decentralized iterative communication procedures. We
will see that there are differences in theoretical properties of these auction formats.
1
There is an implicit tie-breaking used here. If two bidders exit the auction at the same time, the auction
may order the bidders and allow them to exit one after the other. In particular, if the last exit results in no
bidder in the auction, the auction picks one of the bidders at random and allows everyone else to exit.

4
 3 Modeling auctions
The willingness to pay for the object of a bidder determines her strategy in any auction.
The willingness to pay of a bidder is the maximum amount a bidder is willing to pay such
that she is indifferent between buying the object and not buying. This is referred to as the
valuation of the bidder. Models of auctions differ in the way they model valuations of the
bidders.

3.1 An example: auction in 19th century Gujarat guilds

To understand models of auctions, let us consider an example of auction conducted in the
guilds of Gujarat in the 19th century. These auctions are studied in Sen and Swamy (2004).
The guilds of Gujarat were trading associations involving traders doing similar trades. Like
any professional association, such guilds needed money to do various community activities
and provide public goods. They had a unique procedure to raise funds for the guild. Sen
and Swamy (2004) quote the following from the Gazetter of the city of Surat:

A favorite device for raising money is for men of the craft or trade to agree, on
a certain day, to shut all their shops but one. The right to keep open this one
shop is then put up to auction, and the amount bid is credited to the guild fund.

While it is not easy to analyze such auctions because the winning bid in this auction
is used by guild (bidders themselves), let us make the simplifying assumption that winning
bid is used to provide a public good, which does not change the payoff of the bidders. For
instance, the public good is provided irrespective of which bidder wins, but the winning
bidder gets the additional benefit of keeping its shop open. So, the valuation of a bidder is
its valuation for keeping the shop open.
What is the valuation of keeping a shop open? This valuation will depend on the demand
on the shop on the day. We consider three models with n bidders.

(i) Suppose all shops trade the same good (medicine). Then, by keeping its shop open, a
bidder captures the aggregate demand of all the shops in the guild. If we write di as
the demand (no of customers) to shop i, then the valuation of a shop is a function of
P
i di . If the prices are the same across all the shops, then it is reasonable to assume

5
 P
that the valuation is some function v : R+ → R+ , where v( i di ) is the valuation of
any bidder which keeps the shop open. We observe that the valuation is the same for
all the bidders in this model. However, each bidder i only observes his own demand
di . So, even though the bidders know that everyone has the same valuation, they do
not this value ex-ante.

Such a model of valuation is called the common values model. In common values model,
each bidder receives a signal (demand for the shop) which is her private information,
and the signals of all bidders determine a common valuation for the object. Common
value models were first analyzed theoretically in Wilson (1967, 1969). Common value
models are used to analyze sale of oil tracts, sale of goods in the resale market (for
instance, most car buyers in the used car market are dealers who resell the car).

(ii) Suppose all shops trade in different goods. Then, by keeping its shop open, a bidder
captures the demand of her own shop. Since she know the demand of her shop, she
knows the valuation. However, this demand (and hence, valuation) information is
private to her – each shop only knows its own demand. Such a model of valuation is
called the the private values model. Private values models are used to analyze sale of
art, procurement auctions, sale of real estate by auction. The first study of auctions
in private values model is Vickrey (1961).

(iii) In reality, most practical models of auction are somewhere between the private values
and the common values. To understand this, suppose half the shops in the guild trade
medicines and the other half trade books. Each shop only observes her own demand
but cares about demand of shops which trade the same good as hers. So, the valuation
of a medicine shop will depend on the aggregate demand of all medicine shops, but it
will not depend on the the demand of book shops.

Such a model of valuation is called the interdependent values model. Interdependent

values model is general enough to capture the common values and the private values
as special cases. These models were first studied in Milgrom and Weber (1982).

6
 4 Objectives of an auction
There are two reasons to analyze auctions. First, we would like to understand the behavior of
bidders. For this, we will adopt an appropriate notion of equilibrium and analyze equilibrium
behavior of bidders. Second, we would like to compare auction formats in terms of their
equilibrium outcomes. We will carry out these exercises in all the models we will study: (a)
private values model and (b) interdependent values model. The private values model is a
special case of the interdependent values model, but it is analyzed separately because it is
more tractable and simpler than the general interdependent values model.
When comparing auction formats, we usually use two parameters: (a) expected revenue
to seller; (b) efficiency. Efficiency is the standard notion of Pareto efficiency here and
boils down to the following simple notion: an auction is efficient if the bidder with highest
valuation of the object wins the object. This is an ex-post notion of efficiency. Expected
revenue reflects an ex-ante objective of the seller to maximize expected revenue across auction
formats. Under reasonable conditions, we will be able to rank standard auctions in terms of
expected revenue and efficiency.2

5 Private values model and prior-free auctions

We now formally define a private values model. There is a single object for sale by a seller.
There are n bidders and the set of bidders is denoted by N = {1, . . . , n}. The valuation of
each bidder is a random variable denoted by Vi , and its realization is denoted by vi . Each
bidder privately observes the realization of her valuation before entering the auction: this is
the private values model.
We assume that the support of the distribution of this random variable is a set Ti , which
we refer to as the type set of bidder i. The utility of not winning the object is normalized
to zero. If the probability of winning the object with a payment pi is qi , utility from this
outcome is given by qi (vi − pi ). This form of utility function is consistent with a risk neutral
bidder, and we will study extensions to other forms of utility functions later.
2
Most popular criticism of an auction is by looking at the revenue of one instance. But, revenue to a
seller is a random variable, and observed revenue is just one realization of that random variable. Hence, the
right criticism of an auction format should be based on expected revenue it can generate.

7
 5.1 Second-price auction
In a second-price auction, the strategy of a bidder is a map si : Ti → R+ . If bidders bid
b ≡ (b1 , . . . , bn ), let qi (b) denote the winning probability of bidder i and pi (b) denote the
payment of bidder i. Note that

1 if bi > maxj6=i bj
qi (b) =
0 if b < max b
i j6=i j

Further, pi (b) = 0 if qi (b) = 0 and pi (b) = maxj6=i bj otherwise.

Definition 1 Bidding strategy si of bidder i is weakly dominant if for every (vi , v−i ) and
for every s−i ,
h i h i
qi (si (vi ), s−i (v−i )) vi − pi (si (vi ), s−i (v−i )) ≥ qi (bi , s−i (v−i )) vi − pi (bi , s−i (v−i )) ∀ bi .

Strategy si is truthful for bidder i if si (vi ) = vi for all vi ∈ Ti .

Theorem 1 (Vickrey (1961)) In the Vickrey auction, truthful strategy is a weakly domi-
nant strategy for every bidder.

Proof : Fix a profile of valuations v ≡ (vi , v−i ). Fix a buyer i and suppose each of the other
bidder j 6= i bids bj – so, we have fixed an arbitrary profile of bids of other bidders {bj }j6=i .
This profile of bids is generated due to some arbitrary strategy profile of other bidders. We
will argue whatever this bid profile may be, bidder i weakly prefers to bid vi to every other
bid.
Before proceeding with the proof, consider Figure 1. It plots the payoff of a buyer i along
the Y -axis and bid of the buyer i along the X-axis. The payoff of the buyer i is zero if it
bids below max bj . Otherwise (if he bids above max bj ),
j6=i j6=i

• if the value of the buyer i is above max bj , then its payoff of the buyer is given by the
j6=i
blue line (line above Y -axis),

• if the value of the buyer i is below max bj , then its payoff of the buyer is given by the
j6=i
red line (line below Y -axis).

8
 Hence, each bidder i, independent of its value can partition its strategies into two sets:
(i) below max bj and (ii) above max bj . It gets the same payoff by bidding anything in each
j6=i j6=i
of these sets. A buyer whose value is above max bj prefers the blue part to the orange part
j6=i
in Figure 1, but a buyer whose value is below max bj prefers the orange part to the blue part
j6=i
in Figure 1.

if value is vi0
vi0 − max bj
j6=i

vi vi0
ui (bi ; b−i ) 0
bi
max bj
j6=i

vi − max bj
j6=i
if value is vi

Figure 1: Weakly dominant strategy in Vickrey auction

Figure 1 gives an idea on why bidding value maximizes payoff of any buyer. Below, we
formally show that it is indeed a weakly dominant strategy. Suppose buyer i has value vi .
We consider two cases.

Case 1. vi > maxj6=i bj . In this case, the payoff of buyer i from bidding vi is vi −maxj6=i bj >
0. As long as he bids more than maxj6=i bj , buyer i’s payoff remains the same: she still wins
the object and pays the same. By bidding strictly less than maxj6=i bj she does not win the
object and gets a payoff of zero. By bidding equal to maxj6=i bj , she gets the object but with
some probability q ≤ 1 and pays maxj6=i bj . Hence, her payoff is q(vi − maxj6=i bj ), which is
not more than what she was getting by bidding vi .

Case 2. vi ≤ maxj6=i bj . In this case, the payoff of buyer i from bidding vi is zero. This
is because either she is not getting the object (in which case his payoff is zero) or she is
sharing the object in which case she is paying maxj6=i bj = vi . This is the case for all bids
strictly less than maxj6=i bj . If she bids greater than or equal to maxj6=i bj , she wins (with
some probability) but pays maxj6=i bj ≥ vi . Hence, her payoff is non-positive. Hence, bidding

9
vi is at least as good as bidding anything else.3 

The weak dominance is a very strong strategic requirement. It states that the truthful
strategy is better than every other strategy (a) in every state of the world and (b) for any
strategy of other players. Thus, it is independent of the distributional assumptions. The
English auction shares similar properties.

5.2 Ascending price auction

The ascending price auction induces an extensive form game. Strategy in an extensive form
game is more complicated. Remember, we modelled the ascending price auction using a
continuous price clock. At every price p, denote the history at p as hp . This will include all
the bidders who have dropped out and at what prices they have dropped out. Let H be the
set of all possible histories. A strategy of bidder i is a map

si : Ti × R+ × H → [0, 1]

with the requirement that si (·, p, ·) = 0 implies si (·, p0 , ·) = 0 for all p0 > p (i.e., once you exit
an auction, you cannot come back). So, si (vi , p, hp ) = 1 denotes that bidder i with value vi
stays in auction at price p with history hp . With this definition of strategy, Definition 1 also
works for ascending price auction to define a weakly dominant strategy.
Strategy si is truthful for bidder i in ascending price auction if si (vi , p, hp ) = 1 for all
p ≤ vi and for all hp , and si (vi , p, hp ) = 0 otherwise.

Theorem 2 In the ascending price auction, truthful strategy is a weakly dominant strategy
for each bidder.

Proof : The proof is quite simple and does not require any notation. Fix the strategies of
other players, and consider any other strategy in which bidder i is not truthful. Consider an
arbitrary valuation profile. Then, there are couple of cases to consider.

3
To show that bidding vi is weakly dominant, we must also show that vi is strictly better than any other
bid for some bid vector of other players. For this, fix vi and some strategy bi 6= vi . As we saw from the two
cases, if bi > vi , then when vi < maxj6=i bi < bi , it is strictly better for buyer i to bid vi . Similarly, if bi < vi ,
then when bi < maxj6=i bi < vi , it is strictly better for buyer i to bid vi .

10
Case 1. Suppose bidder i wins the auction at price pi by being truthful. In that case,
the only action performed by bidder i is 1 at each price p ≤ pi . By following any other
strategy if bidder i wins then also the only action performed by bidder i is 1 at each price in
the auction. Since other bidders are following the same strategy, the history in the auction
remains the same. As a result, the auction again ends at price pi .
By not being truthful, if she does not win the auction, then she gets zero payoff, which
is weakly worse than following the truthful strategy and winning.

Case 2. Suppose bidder i does not win the auction by being truthful. By not being truthful,
if she still does not win the auction, then her payoff remains the same.
By not being truthful, if she wins the auction, then the actions taken by other bidders
remain the same till price hits vi . In that case, the auction must end at a price ≥ vi . So,
she will win the auction at price above vi , which gives lower payoff than zero. 

Though the truthful strategy is weakly dominant in both second-price and ascending
price auctions, the definition of truthful strategy is different in both the auction formats.
Ascending price auction is a more complex extensive form game, and we established that
truthful strategy is a weakly dominant strategy. A recent paper by Li (2017) studies a
stronger equilibrium concept than weakly dominant strategies in extensive games called
obviously strategy-proof, and shows that the ascending price auction is obviously strategy-
proof but the second-price auction is not.
Since truthful strategy is weakly dominant, the payment of winning bidder is equal to
the second highest value of all bidders. This is exactly the payment of the winning bidder in
the second-price auction. Further, both the auctions are efficient, i.e., the bidder with the
highest value wins the object. Thus, we have established the following corollary.

Corollary 1 The outcome of the second-price auction and the ascending price auction are
identical and efficient in weakly dominant strategies.

5.3 Symmetric Bayesian equilibria in first-price auctions

Let Ti = [0, a] for all i ∈ N . We will assume that the values of bidders are independently
and identically distributed. The cummulative distribution function of values will be denoted

11
by F . We will assume F is differentiable with a positive density f . Hence, bidders are
symmetric ex-ante.
In any sealed-bid auction, a strategy of bidder i is a map: si : [0, a] → R+ . A strategy
profile s ≡ (s1 , . . . , sn ) is symmetric if s1 = . . . , sn . In that case we will denote the strategy
of each bidder as s. A strategy s is monotone if s(x) > s(y) for each x, y ∈ [0, a] with
x > y.
Given a monotone strategy s, the value s(x) denotes the bid amount of any bidder with
valuation x ∈ [0, a]. In first-price auction, given a bid b of bidder i and given all the other
bidders are following symmetric monotone strategy s, bidder i wins if b > s(vj ) for all j 6= i.
The probability of this event is [F (s−1 (b))]n−1 , where we use s−1 (b) = a if b > s(a) and
s−1 (b) = 0 if b < s(0). Denote this as

Q(b; s) := [F (s−1 (b))]n−1

.
We formally define a Bayesian equilibrium using this notation. The definition accounts
for the fact that if a bidder does not win the object she pays zero and her payoff is zero.

Definition 2 A symmetric strategy profile s ≡ (s, . . . , s) is a Bayesian equilibrium of first-

price auction if for every bidder i, for every value vi ∈ [0, a]

   
Q(s(vi ); s) vi − s(vi ) ≥ Q(b; s) vi − b ∀ b ∈ R+ (1)

The following lemma shows that only a particular kind of incentive constraints must hold
for a symmetric strategy profile to be a Bayesian equilibrium.

Lemma 1 (Imitation lemma) A symmetric strategy profile s ≡ (s, . . . , s), where s is

monotone, is a Bayesian equilibrium of first-price auction if and only if for every bidder
i, for every value vi ∈ [0, a]

Q(s(vi ); s) vi − s(vi ) ≥ Q(s(vi0 ); s) vi − s(vi0 ) ∀ vi0 ∈ [0, a]

   
(2)

Proof : Constraints in (1) clearly imply (2). For the other direction, suppose for every
bidder i, for every value vi ∈ [0, a], (2) holds. Note that Q(s(0); s) = 0. This is because if

12
other bidders follow s, whenever one other bidder has value x > 0, she bids s(x) > s(0) and
the bidder bidding s(0) does not win. As a result, the probability of winning by bidding s(0)
is the probability that all the other bidders have value zero, which is zero.
Now, for every  > 0, we have Q(s(); s) > 0. This is because if other bidders have
value less than , then they will bid less than s(), and the bidder bidding s() wins. The
probability that (n − 1) bidders have value less than  is positive since density f is positive.
Using (2) with vi =  and vi0 = 0 implies that Q(s(); s)( − s()) ≥ 0 or  ≥ s(). Using
monotonicity of s, we get  > s(0). Hence, s(0) <  for all , which means s(0) = 0.
Now, pick some b ∈ R+ . Since other bidders follows s, they never bid more that s(a).
Hence, by bidding b > s(a), bidder i always wins. Hence, if b > s(a), then Q(b; s) = 1.
But Q(s(a); s) = 1 too. This is because the only event when bidder i does not win with
probability 1 is when one of the other bidders have value equal to a. This has zero probability.
Hence, we have for every vi ∈ [0, a]

     
Q(s(vi ); s) vi − s(vi ) ≥ Q(s(a); s) vi − s(a) ≥ Q(b; s) vi − b ,

where the first inequality follows from (2). Hence, (1) holds for all b > s(a).
Now, using s(0) = 0, we only need to show (1) holds for any b ∈ [s(0), s(a)]. Since s
is strictly increasing, for every b ∈ [s(0), s(a)], there exists a unique vi0 ∈ [0, a] such that
s(vi0 ) = b. Then, (2) implies (1). 

Theorem 3 Suppose s ≡ (s, . . . , s) is a symmetric strategy profile, where s is a monotone

and differentiable strategy in the first-price auction. Then, the following are equivalent.

1. (s, . . . , s) is a Bayesian equilibrium.

2. s satisfies

Zx
1
s(x) = x − [F (y)]n−1 dy ∀ x ∈ [0, a] (3)
[F (x)]n−1
0

 
Proof : For every x ∈ [0, a], let u(x) = Q(s(x); s) x − s(x) . Since s is monotone and
highest bidder wins, Q(s(x); s) = [F (x)]n−1 , and we write q(x) ≡ Q(s(x); s). Hence, u(x) =

13
q(x)(x − s(x)). Note that if s is differentiable, u is differentiable. By Lemma 2, we know
that s is a Bayesian equilibrium if and only if

u(x) ≥ u(y) + q(y)(x − y) ∀ x, y ∈ [0, a] (4)

Necessity. Suppose s is a Bayesian equilibrium. Then, fix some x, x + δ ∈ [0, a], where δ > 0.
Using (4) we get

u(x + δ) ≥ u(x) + δq(x)

u(x) ≥ u(x + δ) − δq(x + δ)

Hence, we get
δq(x + δ) ≥ u(x + δ) − u(x) ≥ δq(x)

By continuity of q, we thus get that

d[u(x)]
= q(x) ∀ x ∈ [0, a] (5)
dx

Since u(0) = 0, (12) and the fundamental theorem of calculus implies that

Zx
u(x) = q(y)dy
0
Zx
⇒ q(x)(x − s(x)) = q(y)dy
0
Zx
1
⇒ s(x) = x − [F (y)]n−1 dy
[F (x)]n−1
0

Sufficiency. Suppose s is as defined in (10). Then, for every x ∈ [0, a], we have

Zx
u(x) = q(x)(x − s(x)) = q(y)dy
0

14
Hence, for any x, y ∈ [0, a], we have

Zx
u(x) − u(y) = q(z)dz ≥ (x − y)q(y),
y

where the inequality follows since q is non-decreasing. Thus, (4) holds, and we are done. 

By Theorem 3 in a symmetric equilibrium (with monotone and differentiable) strategies,

a bidder with value x bids according to (4). Since this is a symmetric strategy profile with
monotone strategies, for any two bidders with values x, y we see that s(x) > s(y) if and
only if x > y. Hence, the symmetric equilibrium identified in Theorem 3 is efficient: the
highest valued bidder makes the highest bid and wins. Hence, the probability of winning of
 n−1
a bidder with value x is q(x) = F (x) . Hence, using Theorem 3, the expected payment
of a bidder with value x in this symmetric equilibrium is given by

Zx
 n−1  n−1
q(x)s(x) = x F (x) − F (y) dy (6)
0

 n−1
Now, denote G(x) = F (x) for all x ∈ [0, a]. The function G is the cdf of the random
variable that is the maximum of (n−1) draws using F . Hence, the density of this distribution
 n−2
is g(x) = (n − 1) F (x) f (x) for all x ∈ [0, a]. So, an alternate way to write (6) is

Zx Zx
q(x)s(x) = xG(x) − G(y)dy = yg(y)dy (7)
0 0

Rx
The last expression yg(y)dy is the expected value of the random variable highest of
0
(n − 1) values conditional on the fact that it is less than x. Hence, conditional on winning,
the expected payment of a bidder with value x is the expected value of the second highest
valuation.
In a second-price auction, bidders have a weakly dominant strategy to bid their value.
In this equilibrium, a bidder with value x pays zero if she does not win but pays the highest
of (n − 1) other bidders’ values if she wins. Hence, her expected payment is the expected
value of the random variable highest of (n − 1) values conditional on the fact that it is less

15
than x, which is exactly (7).
Since all the bidders are symmetric, the expected revenue of a seller in the first-price and
the second-price auction (in the equilibrium described) is given by

Za Za  Zx 
n q(x)s(x)f (x)dx = n yg(y)dy f (x)dx
0 0 0
Za  Z1 
=n f (y)dy xg(x)dx
0 x
Za
=n x(1 − F (x))g(x)dx
0
Za
 n−2
= n(n − 1) x(1 − F (x)) F (x) f (x)dx
0

Thus, we have come to an important result in auction theory.

Theorem 4 (Revenue equivalence, Vickrey (1961)) The expected payment of each bid-
der in the symmetric equilibrium of the first-price auction and the weakly dominant strategy
of the second-price auction are the same. The expected revenue of the seller is identical
across these two auctions:

Za
 n−2
n(n − 1) x(1 − F (x)) F (x) f (x)dx.
0

While this is a striking result, let us remember the assumptions we have made:

• Values are private.

• Bidders are ex-ante identical: values are independently and identically distributed.

• Bidders are risk neutral.

But this benchmark result will serve as a template for the rest of the course. We will
relax various assumptions in this result and compare auction formats.

16
5.3.1 Descending price and ascending price auctions

The equivalence between first-price and second-price auction formats automatically induce
equivalence with the ascending price auction (1). Now, a strategy in a descending price
auction can be defined similar to an ascending price auction. But the history in a descending
auction does not change till the auction ends. Hence, strategy in a descending auction is just
a function si : [0, a] × [0, P̄ ] → {0, 1}, where P̄ is the highest possible price in the auction.
Here, si (vi , p) = 0 indicates that the bidder is not interested in the object and si (vi , p) = 1
indicates the bidder is interested in the object. Hence, si (vi , p) = 1 implies si (vi , p0 ) = 1 for
all p0 > p. Thus, there is a cutoff price p∗ such that si (vi , p) = 0 for all p < p∗ and si (p) = 1
for all p ≥ p∗ . Thus, a strategy of a bidder is to figure out for each vi , a cut-off price p∗ such
that the bidder is interested in the object below that price. Note that if the bidder wins the
auction she pays p∗ in this case.
The decision in a first-price auction is similar for a bidder. Given the value vi of bidder
i, she has to decide how much to bid. This bid is the amount she pays if she wins. This
bid is exactly similar to p∗ . Hence, the first-price auction and the descending price auction
are equivalent strategically. Thus, the equivalence in Theorem 4 extends to all standard
auctions. We formalize this below.

5.4 Symmetric equilibrium in standard auctions

There are many sealed-bid auction formats that one can think of: first-price, second-price,
third-price etc. A sealed-bid auction is a standard auction if

(a) highest bidder wins;

(b) winner pays non-negative amount;

(c) losers payment is zero;

(d) payment is non-decreasing: higher bid does not lead to lower payment to a bidder.

Each of these assumptions are trivially satisfied by first-price, second-price, and third-price
auctions. We will see the analysis of the first-price auction extends to any standard auction.
Let s be a strategy in a standard auction. If all the other bidders follow strategy s,
then the probability of winning by bidding b is given similarly: bidder i with bid b wins if

17
b > maxj6=i s(vj ) and probability of this event is [F (s−1 (b))]n−1 . As before, we will denote
this as Q(b; s). Note that if bidder i also follows the strategy s, then probability of winning
is [F (s−1 (s(vi )))]n−1 = [F (vi )]n−1 , which we will denote as q(vi ).
If all bidders play s, then let P (b; s) be the expected payment of a bidder by bidding
b. In the first-price auction, this expected payment is bQ(b; s). In the second-price auction,
this expected payment is the expected value of Ev−i :max(s(v−i ))<b max(s(v−i )).4
It is without loss of generality to denote the standard auction by (Q, P ): these are the
only things that will be required for analysis (given a strategy). Using this notation a
symmetric strategy profile s ≡ (s, . . . , s) is a Bayesian equilibrium in a standard auction if
for every i ∈ N , for every vi ∈ [0, a]

Q(s(vi ); s)vi − P (s(vi ); s) ≥ Q(b; s)vi − P (b; s) ∀ b ∈ R+ (8)

Lemma 2 (Imitation lemma) Let s be a monotone strategy in a standard auction (Q, P ).

Strategy s is a Bayesian equilibrium of a standard auction (Q, P ) if and only if for every
bidder i, for every value vi ∈ [0, 1]

Q(s(vi ); s)vi − P (s(vi ); s) ≥ Q(s(vi0 ); s)vi − P (s(vi0 ); s) ∀ vi0 ∈ [0, a] (9)

Proof : Constraints in (8) clearly imply (9). For the other direction, note that by bidding
less than s(0), a bidder always loses and pays zero. Hence, for any b < s(0), Q(b; s) = 0 and
P (b; s) = 0. Similarly, by bidding b = s(0), a bidder wins with positive probability only if all
other bidders have value 0 (in which case they bid s(0)), and this event happens with zero
probability. Hence, Q(s(0); s) = P (s(0); s) = 0. Using this for all b < s(0), we see that for
every vi ∈ [0, a], (9) implies

Q(s(vi ); s)vi − P (s(vi ); s) ≥ Q(s(0); s)vi − P (s(0); s) = 0 = Q(b; s)vi − P (b; s)

Similarly, for all b > s(a), Q(b; s) = Q(s(a); s) = 1 and P (b; s) ≥ P (s(a); s) by non-
4
We know that in the second-price auction s(vi ) = vi is a weakly dominant strategy but we do not use
this specific s here.

18
decreasing payment. Hence, (9) implies that for every b > s(a) and for every vi ∈ [0, a],

Q(s(vi ); s)vi − P (s(vi ); s) ≥ Q(s(a); s)vi − P (s(a); s) ≥ Q(s(a); s)vi − P (b; s)

Now, consider b ∈ [s(0), s(a)]. Since s is monotone, there exists vi0 such that s(vi0 ) = b.
Hence, (9) implies (8) holds.
This exhausts all cases, and hence, (9) implies that s ≡ (s, . . . , s) is a Bayesian equilib-
rium. 

Once the imitation lemma is done, the equilibrium characterization proof is similar.

Theorem 5 Suppose s ≡ (s, . . . , s) is a symmetric strategy profile, where s is a monotone

and differentiable strategy in a standard auction (Q, P ). Then, the following are equivalent.

1. (s, . . . , s) is a Bayesian equilibrium.

2. s satisfies

Zx
n−1
P (s(x); s) = x[F (x)] − [F (y)]n−1 dy ∀ x ∈ [0, a] (10)
0

 
Proof : For every x ∈ [0, a], let u(x) = Q(s(x); s) x − s(x) . Since the highest bidder wins
and s is monotone, Q(s(x); s) = [F (x)]n−1 , and we write q(x) ≡ Q(s(x); s). Hence, u(x) =
xq(x) − P (s(x); s). Unline the proof of Theorem 3, we cannot assume u is differentiable
without loss of generality. However, the rest of the proof can be modified slightly. By
Lemma 2, we know that s is a Bayesian equilibrium if and only if

u(x) ≥ u(y) + q(y)(x − y) ∀ x, y ∈ [0, a] (11)

Necessity. Suppose s is a Bayesian equilibrium. Then, (11) holds. Pick x, y ∈ [0, a] and
λ ∈ [0, 1] with z = λx+(1−λ)y. Then, u(x) ≥ u(z)+q(z)(x−z) and u(y) ≥ u(z)+q(z)(y−z).
Multiplying the first inequality by λ and the second by (1−λ) gives λu(x)+(1−λ)u(y) ≥ u(z).
Hence, u is convex. A convex function is differentiable almost everywhere in the interior of
[0, a].

19
 Then, fix some x, x + δ ∈ [0, a], where δ > 0 and u is differentiable at x. Using (11) we
get

u(x + δ) ≥ u(x) + δq(x)

u(x) ≥ u(x + δ) − δq(x + δ)

Hence, we get
δq(x + δ) ≥ u(x + δ) − u(x) ≥ δq(x)

By continuity of q, we thus get that

d[u(x)]
= q(x) ∀ x ∈ [0, a] (12)
dx

Since u(0) = 0, (12) and the fundamental theorem of calculus implies that for all x ∈ [0, a]

Zx
u(x) = q(y)dy
0
Zx
⇒ q(x)x − P (s(x); s) = q(y)dy
0
Zx
n−1
⇒ P (s(x); s) = x[F (x)] − [F (y)]n−1 dy
0

Sufficiency. Suppose s is as defined in (10). Then, for every x ∈ [0, a], we have

Zx
u(x) = q(x)x − P (s(x); s) = q(y)dy
0

Note that q(x) = [F (x)]n−1 , and hence, q is increasing. Hence, for any x, y ∈ [0, a], we have

Zx
u(x) − u(y) = q(z)dz ≥ (x − y)q(y),
y

where the inequality follows since q is increasing. Thus, (11) holds, and we are done. 

20
 References
Shengwu Li. Obviously strategy-proof mechanisms. American Economic Review, 107(11):
3257–87, 2017.

Paul R Milgrom and Robert J Weber. A theory of auctions and competitive bidding. Econo-
metrica: Journal of the Econometric Society, pages 1089–1122, 1982.

Arijit Sen and Anand V Swamy. Taxation by auction: fund raising by 19th century indian
guilds. Journal of Development Economics, 74(2):411–428, 2004.

William Vickrey. Counterspeculation, auctions, and competitive sealed tenders. The Journal
of finance, 16(1):8–37, 1961.

Robert B Wilson. Competitive bidding with asymmetric information. Management Science,

13(11):816–820, 1967.

Robert B Wilson. Competitive bidding with disparate information. Management science,

15(7):446–452, 1969.

21