Game Theory and its Applications

Game Theory and its Applications

(BITS F314)

Paras Sachdeva

Department of Economics and Finance

BITS Pilani, Pilani Campus

www.parassachdevaeco.com

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

A strategic form game with incomplete information (Bayesian Game)

is defined as a tuple Γ = ⟨N, (Θi ) , (Si ) , (pi ) , (ui )⟩ where

N = {1, 2, . . . , n} is the set of players.

Θi is the set of types of player i where i = 1, 2, . . . , n.

Si is the set of actions or pure strategies of player i where

i = 1, 2, . . . , n.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

The belief function pi is a mapping from Θi into ∆ (Θ−i ), the set of

probability distributions over Θ−i .

For any possible type θi ∈ Θi , pi specifies a probability distribution

pi (. | θi ) over the set Θ−i representing player i ’s beliefs about the
types of the other players if his own type were θi

The payoff function ui : Θ1 × . . . × Θn × S1 × . . . × Sn → R assigns

to each profile of types and each profile of actions, a payoff that
player i would get.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Assumptions

The belief function pi is a mapping from Θi into ∆ (Θ−i ), the set of

probability distributions over Θ−i .
Each player i knows the entire structure of the game as defined above.
Each player i knows his own type θi ∈ Θi .
The above facts are common knowledge among all the players in N.
The belief functions pi describe these conditional probabilities.
The belief functions pi are also common knowledge among the
players.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Consistency of Beliefs

We say beliefs (pi )i∈N are consistent if there is some common prior
distribution P ∈ ∆(Θ) over the set of type profiles Θ such that each
player’s beliefs given his type are just the conditional probability
distributions that can be computed from the prior distribution.

P(θi ,θ−i )
pi (θ−i | θi ) = P
P(θi ,t−i ) ∀θi ∈ Θi ; ∀θ−i ∈ Θ−i ; ∀i ∈ N
t−i ∈Θ−i

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Example

There are two players, player 1 (seller) and player 2 (buyer).

Player 1 wishes to sell an indivisible item and player 2 is interested in
buying this item.
Each player knows what the object is worth to himself but thinks that
its value to the other player may be any integer from 1 to 100 with
1
probability 100 .

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Example

The type of seller has the natural interpretation of being the

willingness to sell (minimum price at which the seller is prepared to
sell the item)
The type of the buyer has the natural interpretation of being the
willingness to pay (the maximum price the buyer is prepared to pay
for the item).
‘If the buyer’s bid is greater than or equal to the seller’s bid they will
trade the object at a price equal to the average of their bids;
otherwise no trade occurs.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games
N = {1, 2}
Θ1 = Θ2 = {1, 2, . . . , 100}
S1 = S2 = {0, 1, 2, . . . , 100}
1
p1 (θ2 | θ1 ) = ∀θ1 ∈ Θ1 ; ∀θ2 ∈ Θ2
100
1
p2 (θ1 | θ2 ) = ∀θ1 ∈ Θ1 ; ∀θ2 ∈ Θ2
100
s1 + s2
u1 (θ1 , θ2 , s1 , s2 ) = − θ1 if s2 ≥ s1
2
= 0 if s2 < s1
s1 + s2
u2 (θ1 , θ2 , s1 , s2 ) = θ2 − if s2 ≥ s1
2
= 0 if s2 < s1 .

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

1
P (θ1 , θ2 ) = ∀θ1 ∈ Θ1 ∀θ2 ∈ Θ2
10000
where
Θ1 × Θ2 = {1, . . . , 100} × {1, . . . , 100}.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Type Agent Representation and the Selten Game

This is a representation of Bayesian games that enables a Bayesian

game to be transformed to a strategic form game (with complete
information). Given a Bayesian game

Γ = ⟨N, (Θi ) , (Si ) , (pi ) , (ui )⟩

the Selten game is an equivalent strategic form game

 
s s
Γ = N , (Sθi )θi ∈Θi , (Uθi )θi ∈Θi .
i∈N i∈N

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Type Agent Representation and the Selten Game

Each player in the original Bayesian game is now replaced with a

number of type agents.
We assume that the type sets of the players are mutually disjoint.
The set of players in the Selten game is given by:
[
Ns = Θi .
i∈N

Note that each type agent of a particular player can play precisely the
same actions as the player himself. This means that for every θi ∈ Θi ,

Sθi = Si .

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Type Agent Representation and the Selten Game

The payoff function Uθi for each θi ∈ Θi is the conditional expected

utility to player i in the Bayesian game given that θi is his actual type.
 
× ×
Uθi : S →R
i ∈ N θi ∈ Θi i

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games
Example

Company 1 produces a product x1 whereas company 2 produces

either product x2 or product y2 .
The product x2 is somewhat similar to product x1 while the product
y2 is a different line of product.
The product to be produced by company 2 is a closely guarded secret,
so it can be taken as private information of company 2 .
We thus have N = {1, 2}, Θ1 = {x1 }, and Θ2 = {x2 , y2 }.
Each firm has to choose a price for the product it produces.
Company 1 has the choice of choosing a low price a1 or a high price
b1
Company 2 has the choice of choosing a low price a2 or a high price
b2 .
We therefore have S1 = {a1 , b1 } and S2 = {a2 , b2 }.
Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Example

The type of company 1 is common knowledge since Θ1 is a singleton.

The belief probabilities of company 2 about company 1 are given by
p2 (x1 | x2 ) = 1 and p2 (x1 | y2 ) = 1.
The belief probabilities of company 1 about company 2 to be
p1 (x2 | x1 ) = 0.6 and p1 (y2 | x1 ) = 0.4.
Let the utility functions for the two possible type profiles
(θ1 = x1 , θ2 = x2 ) and (θ1 = x1 , θ2 = y2 ) be given as in Tables.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Example

2
1
a2 b2
u and u2 for θ1 = x1 ; θ2 = x2
a1 1,2 0,1 1
b1 0,4 1,3
2
1
a2 b2
u and u2 for θ1 = x1 ; θ2 = y2
a1 1,3 0,4 1
b1 0,1 1,2

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

N s = Θ1 ∪ Θ2 = {x1 , x2 , y2 }
Sx1 = S1 = {a1 , b1 }
Sx2 = Sy2 = S2 = {a2 , b2 } .

Uθi : S1 × S2 × S2 → R ∀θi ∈ Θi , ∀i ∈ N

S1 × S2 × S2 = {(a1 , a2 , a2 ) , (a1 , a2 , b2 ) , (a1 , b2 , a2 ) , (a1 , b2 , b2 ) ,

(b1 , a2 , a2 ) , (b1 , a2 , b2 ) , (b1 , b2 , a2 ) , (b1 , b2 , b2 )} .

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

A typical strategy profile can be represented as (sx1 , sx2 , sy2 ).

This could also be represented as (s1 (.), s2 (.)) where the strategy s1 is
a mapping from Θ1 to S1 , and the strategy s2 is a mapping from Θ2
to S2 .

In general, for an n player Bayesian game, a pure strategy profile is of

the form  
(sθ1 )θ1 ∈Θ1 , (sθ2 )θ2 ∈Θ2 , . . . , (sθn )θn ∈Θn .

An equivalent way to write this would be (s1 (.), s2 (.), . . . , sn (.)),

where si is a mapping from Θi to Si for i = 1, 2, . . . , n.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

The following conditional expectation gives the required payoff.

Ux1 (a1 , a2 , b2 ) =p1 (x2 | x1 ) u1 (x1 , x2 , a1 , a2 )

+ p1 (y2 | x1 ) u1 (x1 , y2 , a1 , b2 )
=(0.6)(1) + (0.4)(0)
=0.6

Ux1 (b1 , a2 , b2 ) = 0.4

Ux2 (a1 , a2 , b2 ) = 2
Ux2 (a1 , b2 , b2 ) = 1
Uy2 (a1 , a2 , b2 ) = 4
Uy2 (a1 , a2 , a2 ) = 3.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Payoff Computation in Selten Game

Given a Bayesian game Γ = ⟨N, (Θi ) , (Si ) , (pi ) , (ui )⟩, suppose
(s1 , . . . , sn ) is a strategy profile where for i = 1, . . . , n, si is a mapping
from Θi to Si . Assume the current type of player i to be θi . Then the
expected utility to player i is given by

ui ((si , s−i ) | θi ) = Eθ−i [(ui (θi , θ−i , si (θi ) , s−i (θ−i ))]

For a finite Bayesian game, the above immediately translates to

X
ui ((si , s−i ) | θi ) = pi (t−i | θi ) (ui (θi , t−i , si (θi ) , s−i (t−i ))
t−i ∈Θ−i

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Pure Strategy Bayesian Nash Equilibrium

A pure strategy Bayesian Nash equilibrium in a Bayesian game

Γ = ⟨N, (Θi ) , (Si ) , (pi ) , (ui )⟩

can be defined in a natural way as a pure strategy Nash equilibrium of

the equivalent Selten game. That is, a profile of strategies
(s1∗ , . . . , sn∗ ) is a pure strategy Bayesian Nash equilibrium if
∀i ∈ N; ∀si : Θi → Si ; ∀θi ∈ Θi ,

ui si∗ , s−i
∗ ∗





| θi ≥ ui si , s−i | θi

That is, ∀i ∈ N; ∀ai ∈ Si ; ∀θi ∈ Θi ,

Eθ−i ui θi , θ−i , si∗ (θi ) , s−i

∗ ∗

 

(θ−i ) ≥ Eθ−i ui θi , θ−i , ai , s−i (θ−i )

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Mixed Strategy Bayesian Nash Equilibrium

ui ((σi , σ−i ) | θi ) =
P P Q
t−i ∈Θ−i pi (t−i | θi ) (s1 ,...,sn )∈S σi (si | θi ) j∈N\{i} σj (sj | tj )
(ui (θi , t−i , si (θi ) , s−i (t−i ))

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Mixed Strategy Bayesian Nash Equilibrium

σi∗ , σ−i
∗ ∗





ui | θi ≥ ui σi , σ−i | θi

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

N = {1, 2}, Θ1 = {x1 , y1 }, and Θ2 = {x2 , y2 }.

S1 = {a1 , b1 } and S2 = {a2 , b2 }.
P (θ1 , θ2 )
P(x1 ) = P (x1 , x2 ) + P (x1 , y2 )
P(y1 ) = P (y1 , x2 ) + P (y1 , y2 )
P(x2 ) = P (x1 , x2 ) + P (y1 , x2 )
P(y2 ) = P (x1 , y2 ) + P (y1 , y2 )
p2 (. | x2 )= {p2 (x1 | x2 ) , p2 (y1 | x2 )}.
p2 (. | y2 )= {p2 (x1 | y2 ) , p2 (y1 | y2 )}.
p1 (. | x1 ) = {p1 (x2 | x1 ) , p1 (y2 | x1 )}.
p1 (. | y1 ) = {p1 (x2 | y1 ) , p1 (y2 | y1 )}.

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

N s = Θ1 ∪ Θ2 = {x1 , y1 , x2 , y2 }
Sx1 = Sy1 = S1 = {a1 , b1 }
Sx2 = Sy2 = S2 = {a2 , b2 } .

Uθi : S1 × S1 × S2 × S2 → R ∀θi ∈ Θi , ∀i ∈ N

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Ux1 (a1 , b1 , a2 , b2 ) =p1 (x2 | x1 ) u1 (x1 , x2 , a1 , a2 )

+ p1 (y2 | x1 ) u1 (x1 , y2 , a1 , b2 )

Uy1 (a1 , b1 , a2 , b2 ) =p1 (x2 | y1 ) u1 (y1 , x2 , b1 , a2 )

+ p1 (y2 | y1 ) u1 (y1 , y2 , b1 , b2 )

U1 (a1 , b1 , a2 , b2 ) =p1 (x1 )Ux1 (a1 , b1 , a2 , b2 )

+ p1 (y1 )Uy1 (a1 , b1 , a2 , b2 )

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Ux2 (a1 , b1 , a2 , b2 ) =p2 (x1 | x2 ) u2 (x1 , x2 , a1 , a2 )

+ p2 (y1 | x2 ) u2 (y1 , x2 , b1 , a2 )

Uy2 (a1 , b1 , a2 , b2 ) =p2 (x1 | y2 ) u2 (x1 , y2 , a1 , b2 )

+ p2 (y1 | y2 ) u2 (y1 , y2 , b1 , b2 )

U2 (a1 , b1 , a2 , b2 ) =p2 (x2 )Ux2 (a1 , b1 , a2 , b2 )

+ p2 (y2 )Uy2 (a1 , b1 , a2 , b2 )

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

P(x1 ) = p2 (x1 | x2 ) = p2 (x1 | y2 )

P(y1 ) = p2 (y1 | x2 ) = p2 (y1 | y2 )
P(x2 ) = p1 (x2 | x1 ) = p1 (x2 | y1 )
P(y2 ) = p1 (y2 | x1 ) = p1 (y2 | y1 )

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Consider the following Bayesian game:

Player one has two types x1 and y1 . Player 2 has one type x2
p(x1 ) = p(y1 ) = 0.5
Player 1 chooses U or D, Player 2 chooses L or R (choices are made
simultaneously)

x1 and x2 y1 and x2
L R L R
U 1, 1 0, 2 U 2, 2 0, 1
D 0, 2 1, 1 D 4, 4 2, 3

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

x1 and x2 y1 and x2
L R L R
U 1, 1 0, 2 U 2, 2 0, 1
D 0, 2 1, 1 D 4, 4 2, 3

L R
3 3
UU 2, 2 0, 32
5 5
UD 2, 2 1, 52
1
DU 1,2 2, 1
3
DD 2,3 2, 2

Nash Equilibrium: (U,D,L)

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

x1 and x2 y1 and x2
y (1 − y ) y (1 − y )
L R L R
x U 1,1 0,2 z U 2,2 0,1
(1 − x) D 0,2 1,1 (1 − z) D 4,4 2,3

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Player 1 best response when his type is x1 :

Player 1 would play U instead of D(x = 1) if
1y + 0(1 − y ) > 0y + 1(1 − y ) =⇒ y > 1 − y which can be summarized
as: i)
1
y > =⇒ x = 1
2
1
y < =⇒ x = 0
2
1
y = =⇒ x ∈ [0, 1]
2

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games
Player 1 best response when his type is y1 :
Player 1 will always play D, since U is a strictly dominated strategy.
In other words z = 0 always.
Player 2 best response (recall he does not know in which Matrix he’s
playing, assigning a probability 21 to be in one or the other
Player 2 would play L instead of R(y = 1) if
1 1 1 1
2 [1x + 2(1 − x)] + 2 · 4 > 2 [2x + 1(1 − x)] + 2 · 3

6 − x > 4 + x =⇒ x < 1
which can be summarized as:
x = 1 =⇒ y ∈ [0, 1]
x < 1 =⇒ y = 1

Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)
 Game Theory and its Applications

Bayesian Games

Nash Equilibrium
y = 12 , x = 1 and z = 0
y ∈ 12 , 1 , x = 1 and z = 0


Paras Sachdeva BITS Pilani, Pilani Campus Game Theory and its Applications (F314)