Microeconomics II

Lecture 4: Bayesian Games

Mohammad Vesal
Graduate School of Management and Economics
Sharif University of Technology

44706
Spring 2025

1 / 31
Motivation

• So far we considered static and dynamic games of complete

information.
Payoffs of given outcomes are common knowledge
• Real world is fraught with situations of incomplete
information.
harsh / lenient teacher, aggressive / tolerant rival, able /
unable worker ...
• Players often do not know the type of the person they are
dealing with.
Beliefs: What does player i believe about the preferences of
other players?
What does player i believe about the beliefs of other players
about his preferences?

2 / 31
Outline

Introduction

Bayesian Games

Bayesian Nash Equilibrium

Reference: MWG Ch 8.E (excluding 8.F), Osborne Ch 9.

3 / 31
Outline

Introduction

Bayesian Games

Bayesian Nash Equilibrium

4 / 31
Bayesian games: Harsanyi’s construction

• Players’ preferences are determined by realizations of

random variables.
p.d.f. of these random variables are common knowledge
(common prior).
The specific realization of the random variables are only
known to the relevant players.
• Timing:
1. Nature moves first and determines the realizations of
random variables.
2. Each individual learns its own type (random variable).
3. Individuals simultaneously choose their strategies.
4. Payoffs materialize.

5 / 31
Example 1: Prisoner’s Dilemma

Prisoner 2
C DC
C (−5, −5 − 6θ) (−1, −10)
Prisoner 1
DC (−10, −1 − 6θ) (−2, −2)

• P1’s preferences are common knowledge.

• P2’s type θ ∈ {0, 1}
θ = 0: An ordinary ally (usual prisoner’s dilemma)
θ = 1: A reliable friend (incurs a psychic cost equal to 6
years in prison if he confesses!)
θ = 0 with probability µ (nature).
• P2 knows his type but
P1 only knows the ex ante probabilities of P2’s type.

6 / 31
Example 1: Extensive form

Nature

µ 1−µ

P1

C DC C DC

P2 (θ = 0) P2 (θ = 1)

C DC C DC C DC C DC

−5  −1  −10 −2  −5   −1  −10 −2

−5 −10 −1 −2 −11 −10 −7 −2

Could think of the two types as two distinct players...

7 / 31
Example 1: strategies

• Strategies depend on game’s history and observed types.

• P1’s pure strategies
C: confess
DC: don’t confess
• P2’s pure strategies
C if θ = 0; C if θ = 1
C if θ = 0; DC if θ = 1
DC if θ = 0; C if θ = 1
DC if θ = 0; DC if θ = 1
• What are the equilibrium strategy profiles?

8 / 31
Example 1: Equilibrium
(
C if θ = 0
• P2’s dominant strategy is
DC if θ = 1
Rationality implies P2 plays his dominant strategy.
• Expected payoff of P1 if he
plays C
µ × −5 + (1 − µ) × −1 = −1 − 4µ
plays DC

µ × −10 + (1 − µ) × −2 = −2 − 8µ

• P1’s best response is to always Confess (regardless of µ)

because
C if −1 − 4µ ≥ −2 − 8µ ⇒ µ ≥ −1/4
DC if −1 − 4µ ≤ −2 − 8µ ⇒ µ ≤ −1/4

9 / 31
Bayesian games: formal definition

A Bayesian game is ΓN = {I, {Si } , {ui (.)} , Θ, F (Θ)}

• θi ∈ Θi : random variable (type) observed by player i
θ = (θ1 , . . . , θI ) ∈ Θ = Θ1 × · · · × ΘI
F (θ1 , . . . , θI ) gives the joint density of random variables
• Pure strategy for player i is si (θi )
What the player will do if his type is θi .
Note: Given simultaneous play the only contingency is the
observed type of the player.
• Player i’s payoff function ui (si , s−i , θi , θ−i )
Often payoff depends (directly) only on own type
ui (si , s−i , θi )

10 / 31
Outline

Introduction

Bayesian Games

Bayesian Nash Equilibrium

Example: Public good provision
Example: War of attrition
Example: Market for lemons
Example: Auctions

11 / 31
Bayesian Nash Equilibrium (BNE)
• Profile s∗ is a pure strategy Bayesian Nash Equilibrium
(BNE) if ∀i, ∀θi
 
 X  
∗ ∗


si (θi ) ∈ arg max p(θ−i | θi )ui si , s−i (θ−i ), θ
si ∈Si  
θ−i ∈Θ−i

p(θ−i | θi ): conditional probability that other players have

types θ−i .
The summation is to calculate expected utility.
If types are independent then p(θ−i | θi ) = p(θ−i )
• In the continuous case
Z
∗
ui si (θi ), s∗−i (θ−i ), θ f (θ−i | θi ) dθ−i


si (θi ) ∈ arg max
si ∈Si

12 / 31
Example 2: Public good provision

• Two individuals deciding on whether to produce a public

good
Decision: si ∈ {0, 1}
Cost (private information): θi = ci ∈ [c, c] = Θi
ci i.i.d. across individuals with CDF P (ci )
• Payoffs
ui = max (s1 , s2 ) − ci si
public good: if anyone produces the good, everyone enjoys
the utility of 1.
• What is the BNE of this game?

13 / 31
Example 2: BNE
• (s∗1 , s∗2 ) is a BE iff

s∗1 (c1 ) ∈ arg max Ec2 [max (s1 , s∗2 (c2 )) − c1 s1 ]

s1
s∗2 (c2 ) ∈ arg max Ec1 [max (s∗1 (c1 ), s2 ) − c2 s2 ]
s2

• BE equilibrium is a decision rule.

• But how do we solve for the BE?
h i
• Define zj = Pr cj : s∗j (cj ) = 1
Probability that cj is s.th. player j chooses to produce the
public good in the equilibrium.
• Player i expected payoff is

zj + (1 − zj )si − ci si

14 / 31
Example 2: Solving for the BNE
• Player i’s decision problem
max zj + (1 − zj )si − ci si
si

• Fixing zj , there are two possible values: si ∈ {0, 1}.

Choose si = 1 if
1 − ci ≥ zj ⇔ ci ≤ c∗i ≜ 1 − zj
• In equilibrium
Player 1 contributes iff c1 ≤ c∗1
Player 2 contributes iff c2 ≤ c∗2
• Best responses take the form of a cut-off rule.
∗ ∗
 
Therefore zj = Pr cj ≤ cj = P (cj )
• BE is characterized by c∗1 , c∗2 that solve
c∗1 = 1 − z2 = 1 − P (c∗2 )
c∗2 = 1 − z1 = 1 − P (c∗1 )

15 / 31
Example 2: Unique, symmetric BE
Suppose P (c) = 1 − 1/4 (c − 3/2)2 if c ∈ [−1/2, 3/2]
c∗1 = 1/4 (c∗
2 − 3/2)2 (1)
2
c∗2 = 1/4 (c∗
1 − 3/2) (2)

c∗2
(1):P1 best response
1.5 (2):P2 best response

0.5

−0.5 0.5 1 1.5 c∗1

−0.5
16 / 31
Example 2: Multiple BE
Suppose P (c) = 1 − (c − 1)2 if c ∈ [0, 1]
c∗1 = (c∗2 − 1)2 (3)
2
c∗2 = (c∗1 − 1) (4)

c∗2
1 (3):P1 best response
(4):P2 best response
0.8

0.6

0.4

0.2

0.2 0.4 0.6 0.8 1 c∗1

17 / 31
Example 3: War of attrition
• Two animals fighting for a prey
• Value of the prey is private information: θi ∈ [0, ∞)
θ1 and θ2 have i.i.d. distributions with CDF P (θ)
• Strategy of player i is the length of fighting: si (θi )
• Payoff of player i
(
−si if sj ≥ si
ui =
θi − sj if sj < si

• Expected payoff

E [ui ] = −si Pr [sj (θj ) > si ] +

E [θi − sj (θj ) | sj (θj ) ≤ si ] Pr [sj (θj ) ≤ si ]

• Static game: players pick s1 and s2 at the beginning

18 / 31
Example 3: BNE
• If si (θi ) is strictly increasing and continuous then ∃ϕi s.th.

si (θi ) ≤ s ⇔ θi ≤ ϕi (s)
• Expected payoff could be written as

E [ui ] = −si [1 − P (ϕj (si ))] +

"Z #
ϕj (si )
p(θj )
(θi − sj (θj )) dθj P (ϕj (si ))
0 P (ϕj (si ))
• Change variable of integration from θj to sj using
ϕj (sj ) = θj and sj (θj ) =identity.
E [ui ] = −si [1 − P (ϕj (si ))] +
Z si
(θi − sj ) p(ϕj (sj ))ϕ′j (sj )dsj
0
• Differentiate this w.r.t. to si to get best response of player
i.
19 / 31
Example 3: F.O.C.

• F.O.C.

0 = − [1 − P (ϕj (si ))] + si p (ϕj (si )) ϕ′j (si ) +

(θi − si ) p(ϕj (si ))ϕ′j (si )

• Simplify
1 − P (ϕj (si )) = θi p(ϕj (si ))ϕ′j (si )
| {z } | {z }
marginal cost marginal benefit

• Fighting for dsi more

marginal cost: fighting more and losing
marginal benefit: increased probability of winning×value of
the prey

20 / 31
Example 3: Symmetric BNE
• Let’s search for a symmetric BNE

s1 (θ) = s2 (θ) = s(θ)

ϕ1 (s) = ϕ2 (s) = ϕ(s)
• Simplify F.O.C.

1 − P (ϕj (si )) = θi p(ϕj (si ))ϕ′j (si )

1
⇒ 1 − P (θi ) = θi p(θi ) ′
s (θi )
• Rewrite and integrate
p(θi )
s′ (θi ) = θi
1 − P (θi )
Z θi
p(t)
s(θi ) − s(0) = t dt
0 1 − P (t)
s(0) = 0 why?
21 / 31
Example 3: exponential distribution

• To find the functional form of the equilibrium strategies

let’s assume P (θ) = 1 − e−θ then
θi
e−t θi2
Z
s(θi ) = t dt =
0 e−t 2
• BNE is characterized by the decision rule above.
θ2
Given the value of θi to player i he fights for 2i .
Note this is increasing and continuous as we assume.

22 / 31
Example 3: plot equilibrium strategy and payoff

si (θi )
E[ui (θi )]
0
0 1 2 3
θi

23 / 31
Example 4: Market for lemons

• Consider a market with a buyer and a seller.

• Seller has one car of quality v with uniform distribution
over [0, 1].
• Seller’s actions: sell, do not sell
• Buyer’s actions: buy, do not buy
Exogenous price p ∈ [0, ∞).
Car is traded if seller sells and buyer buys.
• Payoffs
( (
v if not traded 0 if not traded
πs = πb = 3
p if traded 2v −p if traded

• Will there be any equilibrium that leads to trade of the car?

24 / 31
Example 4: Symmetric information

• Will there be a trade equilibrium under symmetric

information?

1. Neither the seller nor the buyer know v

Expected value of the car
For seller: 1/2
For Buyer: 3/4
For any p ∈ [1/2, 3/4], there is a trade equilibrium.
2. Both the seller and the buyer know v
For any v if p ∈ [v, 3/2v] there exists a trade equilibrium.

25 / 31
Example 4: Asymmetric information

• Seller knows v but buyer does not.

• Will there be a trade equilibrium under asymmetric
information?
• Proposition: For all p > 0, there is no Bayesian Nash
equilibrium at which the car is traded with positive
probability.
• Proof:
Seller sells if v ≤ p
If buyer buys the car when seller is selling, she gets
expected quality 1/2p, therefore
3
πb = p − p < 0 if p > 0
4
The buyer should not buy!

26 / 31
First-price auction

• Each bidder submits bi ≥ 0

• The player with highest bid gets the object and pays her
bid.
simultaneous bidding with private valuations
strategy: bid function bi (vi )
• Assume f1 = · · · = fN = f
strictly increasing bid functions
symmetric bid functions bi (vi ) = b(vi )

27 / 31
Solution
• Payoff of bidder i if she behaves as if she has valuation r
 
u(vi , r) = Pr max vj ≤ r (vi − b(r))
j̸=i

= (F (r))N −1 (vi − b(r))

• Why is the probability in terms of v−i and not r−i ?

• This payoff should be maximized at r = vi , therefore

∂u(vi , r)
=0
∂r r=vi

∂  
(N − 1)vi f (vi ) (F (vi ))N −2 = (F (vi ))N −1 b(vi )
| {z } |∂v
marginal benefit
{z }
marginal cost

• Solve the differential equation to find the equilibrium

strategy.
28 / 31
Symmetric Bayesian equilibrium
• The first-price auction has a symmetric equilibrium in
which
Z vi
1 h
N −1
i
b(vi ) = xd (F (x))
(F (vi ))N −1 0
• Notes:
as assumed this is strictly increasing in vi
bidder i bids expected valuation of the second highest
bidder conditional on i having the highest bid.
 
b(vi ) = E max vj | vj ≤ vi
j̸=i

need to check second order condition to be sure.

this is the unique BE in pure strategies but we shall not try
the proof.

29 / 31
Solution with uniform distribution of values

• Suppose v ∼ uniform then F (v) = v, so

Z v
1 N −1
b(v) = N −1 xdxN −1 = v
v 0 N
• Bidders shade their bid
• As N increases shading decreases

30 / 31
Summary

• We have looked at static games of incomplete information.

Bayesian games (Harsanyi’s construction)
Bayesian Nash equilibrium
Four examples
▶ prisoners dilemma with unknown types
▶ public good provision with unknown cost of provision
▶ war of attrition with unknown value of the prey
▶ market for lemons
▶ first price auction

31 / 31