Topic 4: Static Games of Incomplete Information

In this chapter, we will study games of incomplete information or Bayesian games. In

games of incomplete information, at least one player is uncertain about another
player’s payoff function. Same as before, we have static and dynamic versions. The
equilibrium concept related to the static version is called Bayesian Nash equilibrium.
Now, let’s look at the differences between games on complete and incomplete
information.

In games with complete information, each game has the following elements:
1. Players
2. Actions
3. Payoffs

For games with incomplete information, we have two sub elements for players:
1. Players
 Types, e.g., Player 1 could be aggressive or progressive.
 Common prior beliefs about the types, e.g., Player 2 believes that 60% chance
that Player 1 is aggressive and 40% chance that Player 1 is progressive.
2. Actions
3. Payoffs

4.1 A Static Game with Incomplete Information

Suppose Player 1 has two types: type A with a probability 𝒑 and type B with a
probability 𝟏 − 𝒑 . Player 2 has one type only. Player 1 knows his own type, but
Player 2 does not know Player 1’s type. This results in the following payoff matrices,
one for each type of Player 1.
Player 2
L R L R
Player 1 U (3, 4) (1, 0) U (6, 2) (0, 4)
D (4, 3) (2, 0) D (5, 1) (-1, 4)
Type A (𝒑) Type B (𝟏 − 𝒑)
We first consider Player 1 since he knows his types:
 For Type A, we can eliminate U since it is strictly dominated by D.
 For Type B, we can eliminate D since it is strictly dominated by U.
 Therefore, Player 1 chooses D if he is type A and U if he is type B.

Player 2
L R L R
Player 1 U (3, 4) (1, 0) U (6, 2) (0, 4)
D (4, 3) (2, 0) D (5, 1) (-1, 4)
Type A (𝒑) Type B (𝟏 − 𝒑)

Let’s consider Player 2 who faces uncertainty about the type of Player 1. But Player 2
knows Player 1 chooses D if he is type A and U if he is type B. Calculate the expected
payoff of Player 2’s moves.

The expected payoff of choosing L:

3𝑝 + 2(1 − 𝑝) = 2 + 𝑝

The expected payoff of choosing R:

0𝑝 + 4(1 − 𝑝) = 4 − 4𝑝

Player 2 will choose L if:

2
2 + 𝑝 > 4 − 4𝑝 ⟹ 𝑝 >
5
Player 2 will choose R if:
2
𝑝<
5
Player 2 is indifferent between L and R if:
2
𝑝=
5

The Bayesian Nash equilibrium:

 Player 1, Type A chooses D.
 Player 1, Type B chooses U.
2 2 2
 Player 2 chooses L if 𝑝 > 5, R if 𝑝 < 5, and can mix freely if 𝑝 = 5.
4.2 Equilibrium Concept 4: Bayesian Nash Equilibrium
The Bayesian Nash Equilibrium is applied to static game with incomplete information.
In a Bayesian Nash Equilibrium, each player's strategy is optimal given their own type
and their beliefs about the types of other players. No player wants to change his
strategy.

4.2.1 Application: First-Price Sealed-Bid Auction

 There are two bidders who want to own an object, e.g., a vase.
 The value of the object for each bidder is 𝒗𝒊 where 𝑖 = 1,2. In other words, 𝑣𝑖
is the willingness to pay for the object of bidder 𝑖.
 𝒗𝒊 is privately known by bidder 𝒊 That is, 𝑣1 is known by bidder 1 but not
bidder 2, and 𝑣2 is known by bidder 2 but not bidder 1.
 𝒗𝟏 is a random variable from bidder 2’s perspective. 𝒗𝟐 is a random variable
from bidder 1’s perspective.
 Bidder 1’s belief is that 𝒗𝟐 follows a continuous uniform distribution on [0, 100],
or 𝑣2 ~𝑈[0,100].
 Bidder 2’s belief is that 𝒗𝟏 follows a continuous uniform distribution on [0, 100],
or 𝑣1 ~𝑈[0,100].
 This means that 𝑣1 and 𝑣2 can be any number between 0 and 100 with equal
probability.
𝒙−𝒂
 A property of a uniform distribution 𝑣~𝑈[𝑎, 𝑏] is that 𝒑𝒓𝒐𝒃(𝒗 ≤ 𝒙) = 𝒃−𝒂 .

𝒙−𝟎 𝒙
Since 𝑣𝑖 ~𝑈[0,100] in our case, 𝒑𝒓𝒐𝒃(𝒗𝒊 ≤ 𝒙) = 𝟏𝟎𝟎−𝟎 = 𝟏𝟎𝟎 . For example,

50
the probability of 𝑣𝑖 ≤ 50 is = 0.5 or 50%.
100

 Types of bidder 𝑖: 𝑣𝑖 .
 Beliefs: 𝑣𝑖 ~𝑈[0,100].
 Each bidder simultaneously submits a sealed-bid, bidder 1 submits 𝒃𝟏 and
bidder 2 submits 𝒃𝟐 . 𝑏1 , 𝑏2 ∈ (0, ∞].
 The winner gets the object and pays his bid. The loser has 0 payoff. Therefore, if
bidder 1 wins, his payoff is 𝑣1 − 𝑏1 and bidder 2’s payoff is 0.
 For simplicity, we will find a Bayesian Nash Equilibrium that is:
1. Linear: 𝑏𝑖 = 𝑐𝑣𝑖 where 𝑐 > 0
2. Symmetric: The equilibrium strategies for each player are the same. This
means we only need to find the equilibrium strategies for one of the bidders.
 Let’s consider bidder 1. His payoff 𝑢1 (𝑏1 , 𝑏2 ) is given by:
𝑣1 − 𝑏1 𝑖𝑓 𝑏1 ≥ 𝑏2
𝑢1 (𝑏1 , 𝑏2 ) = {
0 𝑖𝑓 𝑏1 < 𝑏2
(One may think 𝑏1 > 𝑏2 is more reasonable. Note that because 𝑣𝑖 is continuously
Ask
distributed and 𝑏𝑖 = 𝑐𝑣𝑖 is also continuously distributed. There are an infinite
yourself:
number of possible values of 𝑏𝑖 in a continuous distribution. The probability of
How many
randomly selecting the exact same value for both 𝑏1 and 𝑏2 is effectively zero.
numbers
Therefore, 𝑏1 = 𝑏2 is a zero-probability event. Basically, it doesn’t matter if we write
are there
𝑏1 ≥ 𝑏2 or 𝑏1 > 𝑏2 )
between 0
and 100?
We can see that bidder 1’s payoff depends on 𝒃𝟐 . If 𝑏1 ≥ 𝑏2 , he gets 𝑣1 − 𝑏1 , and if
𝑏1 < 𝑏2 , he gets 0. However, from bidder 1’s perspective, 𝒃𝟐 is unknown. His payoff
is uncertain.

Bidder 1’s expected payoff is given by:

𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏(𝑏1 ≥ 𝑏2 ) × (𝑣1 − 𝑏1 ) + 𝑝𝑟𝑜𝑏(𝑏1 < 𝑏2 ) × 0
𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏(𝑏1 ≥ 𝑏2 ) × (𝑣1 − 𝑏1 )

Given 𝒃𝟐 = 𝒄𝒗𝟐 , we can rewrite it as:

𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏(𝑏1 ≥ 𝒄𝒗𝟐 ) × (𝑣1 − 𝑏1 )
𝑏1
𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏 (𝑣2 ≤ ) × (𝑣1 − 𝑏1 )
𝑐
𝒙
Using the property of the uniform distribution 𝒑𝒓𝒐𝒃(𝒗𝒊 ≤ 𝒙) = 𝟏𝟎𝟎, then:

𝒃𝟏
𝐸𝑢1 (𝑏1 , 𝑏2 ) = × (𝑣1 − 𝑏1 )
𝟏𝟎𝟎𝒄
Bidder 1 maximizes her expected payoff by choosing 𝒃𝟏 , the FOC is:
𝜕𝐸𝑢1 (𝑏1 , 𝑏2 ) 1
= (𝑣 − 2𝑏1 ) = 0
𝜕𝑏1 100𝑐 1
Solve for 𝑏1 :
𝑣1
𝑏1 =
2
By symmetric:
𝑣2
𝑏2 =
2
𝑣1 𝑣2
Therefore, the Bayesian Nash Equilibrium is 𝑏1 = and 𝑏2 = .
2 2

For example, if 𝑣1 = 80 then 𝑏1 = 40. If 𝑣2 = 50 then 𝑏2 = 25.

4.2.2 ICE

There are two bidders who want to own a piece of land. The value of the land for each
bidder 𝑖 is 𝑣𝑖 where 𝑖 = 1,2. 𝑣𝑖 is independently and identically distributed with
uniform distribution on [0, 1]. The value 𝑣𝑖 is privately known by bidder 𝑖 .
Simultaneously, each bidder 𝑖 bids a real number 𝑏𝑖 . The player who bids the higher
number wins the piece of land; if the bids are identical, we toss a coin to determine
the winner. The winner gets the land and pays his bid. The loser gets 0 payoff.
Therefore, if bidder 𝑖 wins, then her payoff is 𝑣𝑖 − 𝑏𝑖 and the other bidder’s get 0.
Find a symmetric, linear Bayesian Nash equilibrium, where 𝑏𝑖 (𝑣𝑖 ) = 𝑐𝑣𝑖 and 𝑐 > 0.

Bidder 1’s expected payoff is given by:

𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏(𝑏1 ≥ 𝑏2 ) × (𝑣1 − 𝑏1 ) + 𝑝𝑟𝑜𝑏(𝑏1 < 𝑏2 ) × 0
𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏(𝑏1 ≥ 𝑏2 ) × (𝑣1 − 𝑏1 )

Given 𝒃𝟐 = 𝒄𝒗𝟐 , we can rewrite it as:

𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏(𝑏1 ≥ 𝒄𝒗𝟐 ) × (𝑣1 − 𝑏1 )
𝑏1
𝐸𝑢1 (𝑏1 , 𝑏2 ) = 𝑝𝑟𝑜𝑏 (𝑣2 ≤ ) × (𝑣1 − 𝑏1 )
𝑐
Using the property of the uniform distribution 𝒑𝒓𝒐𝒃(𝒗𝒊 ≤ 𝒙) = 𝒙, then:
𝒃𝟏
𝐸𝑢1 (𝑏1 , 𝑏2 ) =× (𝑣1 − 𝑏1 )
𝒄
Bidder 1 maximizes her expected payoff by choosing 𝒃𝟏 , the FOC is:
𝜕𝐸𝑢1 (𝑏1 , 𝑏2 ) 1
= (𝑣1 − 2𝑏1 ) = 0
𝜕𝑏1 𝑐
Solve for 𝑏1 :
𝑣1
𝑏1 =
2
By symmetric:
𝑣2
𝑏2 =
2
𝑣1 𝑣2
Therefore, the Bayesian Nash Equilibrium is 𝑏1 = and 𝑏2 = .
2 2
4.2.3 Application: Cournot Game under Asymmetric Information

Two firms produce a homogeneous good and choose outputs 𝑞1 and 𝑞2 . The market
demand is given by
𝑃 = 𝑎 − 𝑞1 − 𝑞2
The cost functions of the firms are different. The cost function of Firm 1 is
𝐶1 = 𝑐𝑞1
Firm 2 has two types: either high marginal cost or low marginal cost. Let 𝑐𝐿 < 𝑐𝐻 .
Low type:
𝐶2 = 𝑐𝐿 𝑞2
High type:
𝐶2 = 𝑐𝐻 𝑞2

Firm 1’s cost function is common knowledge. However, Firm 1 only knows Firm 2 can
be either high or low type. Firm 1 has “beliefs” that Firm 2 is high type with a
probability of 𝜽 and low type with a probability of 𝟏 − 𝜽.

Let’s consider Firm 2’s maximization problem. Firm 2 faces NO uncertainty.

The high-type Firm 2 will maximize:

𝜋2𝐻 = (𝑎 − 𝑞1 − 𝑞2 )𝑞2 − 𝑐𝐻 𝑞2

FOC:
𝜕𝜋2𝐻
= 𝑎 − 2𝑞1 − 𝑞2 − 𝑐𝐻 = 0
𝜕𝑞2
Solve for 𝑞2
𝑎 − 𝑞1 − 𝑐𝐻
𝑞2 (𝑐𝐻 ) =
2

The low-type Firm 2 will maximize:

𝜋2𝐿 = (𝑎 − 𝑞1 − 𝑞2 )𝑞2 − 𝑐𝐿 𝑞2

FOC:
𝜕𝜋2𝐿
= 𝑎 − 2𝑞2 − 𝑞1 − 𝑐𝐿 = 0
𝜕𝑞2
Solve for 𝑞2
𝑎 − 𝑞1 − 𝑐𝐿
𝑞2 (𝑐𝐿 ) =
2
Let’s consider Firm 1’s maximization problem. Firm 1 does not know its competitor’s
type and must there maximize its expected payoff over the two possible types of
Firm 2.

𝐸(𝜋1 ) = 𝜃{(𝑎 − 𝑞1 − 𝑞2 (𝑐2𝐻 ))𝑞1 − 𝑐𝑞1 } + (1 − 𝜃){(𝑎 − 𝑞1 − 𝑞2 (𝑐2𝐿 ))𝑞1 − 𝑐𝑞1 }

𝐸(𝜋1 ) = 𝜃{𝑎𝑞1 − 𝑞12 − 𝑞1 𝑞2 (𝑐2𝐻 ) − 𝑐𝑞1 } + (1 − 𝜃){𝑎𝑞1 − 𝑞12 − 𝑞1 𝑞2 (𝑐2𝐿 ) − 𝑐𝑞1 }

FOC:
𝜃{𝑎 − 2𝑞1 − 𝑞2 (𝑐2𝐻 ) − 𝑐} + (1 − 𝜃){𝑎 − 2𝑞1 − 𝑞2 (𝑐2𝐿 ) − 𝑐} = 0

Solve for 𝑞1
𝜃{𝑎 − 𝑞2 (𝑐2𝐻 ) − 𝑐} + (1 − 𝜃){𝑎 − 𝑞2 (𝑐2𝐿 ) − 𝑐}
𝑞1 =
2

Now, we have 3 equations and 3 unknowns.

𝑎 − 𝑞1 − 𝑐𝐻
𝑞2 (𝑐𝐻 ) =
2
𝑎 − 𝑞1 − 𝑐𝐿
𝑞2 (𝑐𝐿 ) =
2
𝜃{𝑎 − 𝑞2 (𝑐2𝐻 ) − 𝑐} + (1 − 𝜃){𝑎 − 𝑞2 (𝑐2𝐿 ) − 𝑐}
𝑞1 =
2

Suppose 𝜃 = 0.5, 𝑎 = 130, 𝑐 = 10, 𝑐𝐻 = 20, 𝑐𝐿 = 0

110 − 𝑞1
𝑞2 (𝑐𝐻 ) =
2
130 − 𝑞1
𝑞2 (𝑐𝐿 ) =
2
𝑞2 (𝑐𝐻 ) 𝑞2 (𝑐𝐿 )
𝑞1 = 60 − −
4 4
Then
4𝑞1 = 240 − 𝑞2 (𝑐𝐻 ) − 𝑞2 (𝑐𝐿 )
110 − 𝑞1 130 − 𝑞1
4𝑞1 = 240 − −
2 2
8𝑞1 = 480 − 110 + 𝑞1 − 130 + 𝑞1
𝑞1 = 40
𝑞2 (𝑐𝐻 ) = 35
𝑞2 (𝑐𝐿 ) = 45