Lecture 02

December 1, 2022

1 Static games with incomplete information

1.1 Examples
Example 1.1. A variant of BoS (from Osborne’s textbook)

1
Example 1.2. A simpler variant of BoS (also from Osborne’s textbook)

1.2 Theory
1.2.1 Representation of a Bayesian game
• A Bayesian game is a strategic form game Γ = N, Ω, p, (Θi , ti , Ai , ui )i∈N with incomplete
information that consists of
– A set of players N , e.g. N = {1, 2, . . . , n}
– A set of states: Ω, e.g. Ω = {ω1 , . . . , ωl }
– A common prior belief p ∈ ∆(Ω) defined by p(ω) = Pr(ω) for any ω ∈ Ω
– Sets of types (signals): Θi , for each player i, Θ = ×i∈N Θi
– Type functions: ti : Ω → Θi , which specifies player i’s type in the state ω ∈ Ω
– Action sets: Ai for each player i, (A = ×i∈N Ai )
– State-dependent payoff functions ui : A × Ω → R

1.2.2 Belief updating

• The belief is updated via Bayes rule.
– Denote p(ω|θi ) i’s belief about the state given own type
– For specific θ̂i ∈ Θi , and ω̂ ∈ Ω, applying Bayes rule,

Pr(θi = θ̂i , ω = ω̂)

p(ω̂|θ̂i ) = Pr(ω = ω̂|θi = θ̂i ) = ,
Pr(θi = θ̂i )

where (
p(ω̂) ti (ω̂) = θ̂i
Pr(θi = θ̂i , ω = ω̂) =
0 otherwise
X X
Pr(θi = θ̂i ) = Pr(θi = θ̂i , ω) = Pr(θi = θ̂i , ω)
ω∈Ω ω∈t−1
i (θ̂i )
n o
with t−1
i (θ̂i ) = ω ∈ Ω|ti (ω) = θ̂i .

2
1.2.3 Strategy and equilibrium
• A pure strategy for player i is a function that maps player i’s type into her action set,

σi : Θi → Ai

• Given a belief p and strategy profile (σ1 , . . . , σn ), the expected payoff of player i of type θi is
X
ui (σi , σ−i |θi ) = [p(ω|θi )ui [σi (θi ), σ−i (t−i (ω)); ω]]
ω∈Ω

where σ−i (θ−i ) = (σj (θj ))j̸=i and t−i (ω) = (tj (ω))j̸=i .

Definition 1.1. A Bayesian-Nash equilibrium, or Bayesian equilibrium, in a Bayesian game Γ =

N, Ω, p, (Θi , ti , Ai , ui )i∈N is a strategy profile σ ∗ ∈ Σ such that, for any player i ∈ N of any type
θi ∈ Θi and for any strategy σi ∈ Σi of player i,

ui (σi∗ , σ−i
∗ ∗
|θi ) ≥ ui (σi , σ−i |θi ).

Proposition 1.1. A strategy profile σ ∗ ∈ Σ is a Bayesian equilibrium in a Bayesian game Γ =

N, Ω, p, (Θi , ti , Ai , ui )i∈N if and only if a profile â∗ = (σi∗ (θi ))i∈N,θi ∈Θi is a Nash equilibrium of a
   
strategic form game Γ̂ = N̂ , Âî , ûî where
î∈N̂

• N̂ = {(i, θi )|i ∈ N, θi ∈ Θi },
• Â(i,θi ) = Ai for all player (i, θi ) ∈ N̂

• and û(i,θi ) (aî , a−î ) = ω∈t−1 (θi ) p(ω|θi )ui a(i,θi ) , a−(i,θi ) ; ω .
P  
i

1.2.4 Mixed strategy

• A mixed strategy is simply defined by, for each player i,

αi (·|·) : Θi → ∆(Ai ),

a function maps from type space to action space. That is, αi (·|θi ) ∈ ∆(Ai ).
• Suppose that the types θ and state ω is know, then
X X
ui (αi , α−i ; ω) = αi (ai |ti (ω))α−i (a−i |t−i (ω))ui (a, a−i ; ω)
ai ∈Ai a−i ∈A−i

• The expected payoff of of mixed strategy profile α = (α1 , . . . , αn ) of i with type θi :

X
ui (αi , α−i |θi ) = p(ω|θi )ui (αi , α−i ; ω)
ω∈Ω
X
= p(ω|θi )ui (αi , α−i ; ω)
ω∈t−1
i (θi )

Definition 1.2. A mixed strategy Bayesian-Nash equilibrium, or mixed strategy Bayesian equilibrium,
in a Bayesian game Γ = N, Ω, p, (Θi , ti , Ai , ui )i∈N is a mixed strategy profile α∗ such that, for any
player i ∈ N of any type θi ∈ Θi and for any strategy αi of player i,

ui (αi∗ , α−i
∗ ∗
|θi ) ≥ ui (αi , α−i |θi ).

Corollary 1.1. Every finite Bayesian game has a mixed strategy Bayesian-Nash equilibrium.

3
2 Applications
2.1 Cournot duopoloy with incomplete information
• The game:
– The inverse demand function is
p=2−Q
where Q = q1 + q2 .
– Firm 1 has one type with marginal cost c1 = 1
∗ Thus, firm 1’s profit is (p − c1 )q1 = (1 − q1 − q2 )q1
– Firm 2 has two possible types with marginal cost either ch2 = 5/4 or cl2 = 3/4
∗ type h’s profit is (p − ch2 )q2 = (3/4 − q1 − q2 )q2
∗ type l’s profit is (p − cl2 )q2 = (5/4 − q1 − q2 )q2

2.2 More information may hurt

• Consider the following (pp. 282-283 in Osborne)

– Let ϵ = 1/2
– Case 1: both player don’t the states, the unique NE is (B, L)
– Case 2: player 2 know the states, the unique NE is (T, R.M )

2.3 Public goods provision

• The game:
– A group of n people decide whether or not to provide a public good
– The public good is provided if at least one of them pays the cost c
– The benefit of the public good is vi to each person i is observed only by her
– Everyone knows that vi ∼ F on [v, v̄] with v < c < v̄.
• Solution
– We consider a symmetric cutoff equilibrium such that a person pays the cost if vi ≥ v ∗ .
– For person i, what is probability that one of others pays the cost?
n−1
Pr (∃j ̸= i, vj ≥ v ∗ ) = 1 − Pr (∀j ̸= i, vj < v ∗ ) = 1 − (F (v ∗ ))

– For no one of any type having incentive to deviate,

h i
n−1
∗ if vi ≥ v ∗ , vi − c ≥ vi 1 − (F (v ∗ ))
h i
n−1
∗ if vi < v ∗ , vi − c ≤ vi 1 − (F (v ∗ ))
– So, in equilibrium,
n−1
v ∗ (F (v ∗ )) =c

4
 – The probability of a public good being provided

Pr (∃i ∈ N, vi ≥ v ∗ ) = 1 − Pr (∀i ∈ N, vi < v ∗ )

n
= 1 − (F (v ∗ )) = 1 − cF (v ∗ )/v ∗

is decreasing in v ∗ if and only if F (v ∗ )/v ∗ is increasing in v ∗ .

2.4 Second-price sealed auction

• The game:
– A sealed auction with n participants:
∗ Set of players (bidders), N = {1, . . . , n}
∗ For each i, a set of types (valuations) Θi = [v, v̄]
∗ The set of state, Ω = ×i∈N Θi = [v, v̄]n and ti : Ω → Θi is simply defined as

ti (θ1 , . . . , θn ) = θi

∗ For prior belief, we define the cdf G on Ω:

h i
G(ω̂) = Pr(ω ≤ ω̂) = Pr (θ1 , . . . , θn ) ≤ (θ̂1 , . . . , θ̂n )
  Y Y
= Pr θ1 ≤ θ̂1 , . . . , θn ≤ θ̂n = Pr(θi ≤ θ̂i ) = F (θ̂i ),
i∈N i∈N

that is, valuations are independent.

∗ For each i, an action set Ai = R+ (any nonnegative bids are admissible)
– Payoff function for bidder i, denoting a∗ = maxi∈N ai ,
( v −P (a)
i
♯{j∈N |aj =a∗ } if ai = a∗ ,
ui (ai , a−i ; ω) =
0 otherwise,

where P (a) is the price paid by winner given the action profile a ∈ A.
– P (a) equals to the second highest bid.

Claim 2.1. bidding one’s valuation is weakly dominant for everyone in the second-price sealed auction.

2.5 First-price sealed auction

• The game:
– Everything is the same as the second-price auction, except that P (a) = a∗ .
• Solution:
– Look for symmetric equilibrium βi (vi ) = β(vi ) (and β is continuous and strictly increasing)
– What is Pr(b = a∗ ), i.e. the probability of b being the maximum bid give β from perspective
of i?


n−1
Pr(b = a∗ ) = Pr (β(vj ) < b, ∀j ̸= i) = Pr vj < β −1 (b), ∀j ̸= i = F β −1 (b)

– The expected payoff of bidding b:

n−1
(v − b) Pr(b = a∗ ) = (v − b) F β −1 (b)

5
 – The FOC w.r.t. b:


n−1

n−2 ′ −1
1
− F β −1 (b) + (v − b)(n − 1) F β −1 (b) F β (b) =0
β′ (β −1 (b))

– If β is a NE, then β(v) = b must satifies the FOC, hence,

n−1 n−2 1
− (F (v)) + (v − β(v))(n − 1) (F (v)) F ′ (v) =0 (1)
β′ (v)

– (1) is a differential equation of β,

∗ Rearrane it and obtain
n−1 n−2 n−2
β ′ (v) (F (v)) + β(v)(n − 1) (F (v)) F ′ (v) = (n − 1)v (F (v)) F ′ (v)

∗ Integrate both sides

Z v
n−1 n−1
β(v) (F (v)) = ud (F (u))
v
Z v
n−1 n−1
= v (F (u)) − (F (u)) du
v

∗ Solving β, Rv n−1
v
(F (u)) du
β(v) = v − n−1
(F (v))
– If F (v) = v on [0, 1],
1 vn n−1
β(v) = v − = v
n v n−1 n

2.5.1 Order statistics and revenue equivalence theorem

• What is ithe expected revenue from first and second price auctions?
• Order statistics:
– Let v(k) be the k-th smallest from (v1 , . . . , vn ) and denote Fk its distribution
– The distribution of Fn :


Fn (v) = Pr v(n) ≤ v = Pr (v1 ≤ v, . . . , vn ≤ v)
Y n
= Pr(vi ≤ v) = [F (v)]
i∈N

– The distribution of F(n−1) :

Fn−1 (v) = Pr v(n−1) ≤ v, v(n) > v + Pr v(n) ≤ v
n n−1
= [F (v)] + n [F (v)] [1 − F (v)]

– In general,
n
X h n−h
Fk (v) = Cnh [F (v)] [1 − F (v)]
h=k

∗ v(k) ≤ v: (one partition) k elements smaller or equal to v; n − k ones greater

k n−k
∗ There can be Cnk such partitions with probability of each [F (v)] [1 − F (v)]

6
 – Hence, suppose that F (v) = v on [0, 1]
Z 1
n
vdv n =
 
E v(n) =
0 n+1

∗ The expected revenue from first price auctions is

 
n−1 n−1 n n−1
E v(n) = =
n n n+1 n+1

∗ The expected revenue from second price auctions is

Z 1
vd nv n−1 (1 − v) + v n
   
E v(n−1) =
0
1 1
n−1 n−1
Z Z
n
=n vdv n−1 − (n − 1) vdv n = n − (n − 1) =
0 0 n n+1 n+1

∗ The expected revenues are the same from first and second price auctions.

Theorem 2.1. The expected revenues are the same in any auctions with independent private values
with a common distribution, in which
1. there are same number of risk neutral bidders,

2. the object always goes to the bidder with the highest value in equilibrium, and
3. the bidder with the lowest value expects zero surplus.
Exercise 2.1 (All-pay auction). Consider a sealed auction with n participants. Only the bidder with
the highest bid wins the auction (with equal winning probability if there is a tie), and everyone need to
pay her bid. What is the equilibrium bidding strategy? What is the revenue from an all-pay auction?