0% found this document useful (0 votes)
33 views24 pages

Game Theory and Strategic Management: A. Nicol' o

This document discusses game theory and strategic management. It covers topics such as games with incomplete information, Bayesian games with private values, and examples of games with incomplete information including a Cournot duopoly and a first price auction. Key concepts explained include games of complete versus incomplete information, Bayesian Nash equilibrium, and modeling games as Bayesian games where players have private information known only to themselves.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
33 views24 pages

Game Theory and Strategic Management: A. Nicol' o

This document discusses game theory and strategic management. It covers topics such as games with incomplete information, Bayesian games with private values, and examples of games with incomplete information including a Cournot duopoly and a first price auction. Key concepts explained include games of complete versus incomplete information, Bayesian Nash equilibrium, and modeling games as Bayesian games where players have private information known only to themselves.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 24

Game Theory and Strategic Management

A. Nicolò

(University of Padova)

May 2020

(U. Padova) May 2020 1 / 20


Games with incomplete information

Up to now we assumed that players know all the relevant information


about each other.

(U. Padova) May 2020 2 / 20


Games with incomplete information

Up to now we assumed that players know all the relevant information


about each other.
In game theory a game whose structure is common knowledge is
called a game of complete information (or with a roughly equivalent
modern economic term, “symmetric information”).

(U. Padova) May 2020 2 / 20


Games with incomplete information

Up to now we assumed that players know all the relevant information


about each other.
In game theory a game whose structure is common knowledge is
called a game of complete information (or with a roughly equivalent
modern economic term, “symmetric information”).
Recall that a game of complete information need not have perfect
information: There may still be simultaneous decisions, so that a
player making a decision cannot always observe all previous decisions.

(U. Padova) May 2020 2 / 20


A game of incomplete information (roughly, “asymmetric
information”) allows both simultaneous decisions and players to have
private information about the structure of the game.

(U. Padova) May 2020 3 / 20


A game of incomplete information (roughly, “asymmetric
information”) allows both simultaneous decisions and players to have
private information about the structure of the game.
Example: consider two firms competing a la Cournot in a mkt. Firm
1, the incumbent firm has marginal cost equal to c1 and is common
knowledge, while the marginal cost of firm 2 is private information
and can be either c2H or c2L with equal probability.

(U. Padova) May 2020 3 / 20


Bayesian Games: private values

In a Bayesian game with private values each player has a payoff


function ui (si , s−i , θ i ) where θ i ∈ Θ is a r.v. chosen by nature and
observed by player i only. The joint distribution for all i ∈ N of θ i is
F (θ 1 , ..., θ n )

(U. Padova) May 2020 4 / 20


Bayesian Games: private values

In a Bayesian game with private values each player has a payoff


function ui (si , s−i , θ i ) where θ i ∈ Θ is a r.v. chosen by nature and
observed by player i only. The joint distribution for all i ∈ N of θ i is
F (θ 1 , ..., θ n )
A pure strategy for player i ∈ N in a Bayesian game, is a function
si (θ i ) ∈ Si which assigns a strategy choice to each realization of the
type θ i .

(U. Padova) May 2020 4 / 20


Bayesian Games: private values

In a Bayesian game with private values each player has a payoff


function ui (si , s−i , θ i ) where θ i ∈ Θ is a r.v. chosen by nature and
observed by player i only. The joint distribution for all i ∈ N of θ i is
F (θ 1 , ..., θ n )
A pure strategy for player i ∈ N in a Bayesian game, is a function
si (θ i ) ∈ Si which assigns a strategy choice to each realization of the
type θ i .
Player i 0 s expected payoff given a strategy profile s1 (·), ...sn (·) is

(U. Padova) May 2020 4 / 20


Bayesian Games: private values

In a Bayesian game with private values each player has a payoff


function ui (si , s−i , θ i ) where θ i ∈ Θ is a r.v. chosen by nature and
observed by player i only. The joint distribution for all i ∈ N of θ i is
F (θ 1 , ..., θ n )
A pure strategy for player i ∈ N in a Bayesian game, is a function
si (θ i ) ∈ Si which assigns a strategy choice to each realization of the
type θ i .
Player i 0 s expected payoff given a strategy profile s1 (·), ...sn (·) is

Eθ [ui (s1 (θ 1 ), ..., sn (θ n ), θ i )] = ũi (s1 (·) , ..., sn (·))

(U. Padova) May 2020 4 / 20


Bayesian Games: private values

In a Bayesian game with private values each player has a payoff


function ui (si , s−i , θ i ) where θ i ∈ Θ is a r.v. chosen by nature and
observed by player i only. The joint distribution for all i ∈ N of θ i is
F (θ 1 , ..., θ n )
A pure strategy for player i ∈ N in a Bayesian game, is a function
si (θ i ) ∈ Si which assigns a strategy choice to each realization of the
type θ i .
Player i 0 s expected payoff given a strategy profile s1 (·), ...sn (·) is

Eθ [ui (s1 (θ 1 ), ..., sn (θ n ), θ i )] = ũi (s1 (·) , ..., sn (·))

For each type profile θ = (θ i , θ −i ) we compute the outcome


generated by the strategy profile s (θ ) and the VNM expected utility
function of each player i is computed as the sum of the payoffs
ui (s (θ )) weighted by the probability that θ occurs.

(U. Padova) May 2020 4 / 20


Definition
A (pure strategy) Bayesian Nash equilibrium (BNE) in pure strategy is a
strategy profile (s1 (·) , ..., sn (·)) such that for all i ∈ N

ũi (si (·) , s−i (·)) ≥ ũi si0 (·) , s−i (·)


for all si0 ∈ Si

(U. Padova) May 2020 5 / 20


Lemma
A profile of strategies is a BNE if and only if for all i and for all θ i ∈ Θ
occurring with positive probability

Eθ −i [ui (s1 (θ 1 ), ..., sn (θ n ), θ i ) |θ i )] ≥ Eθ −i ui (si0 (θ i ), s−i (θ −i ), θ i ) |θ i )


 

Proof.
Immediate.

In essence, we can think of each type of player i as being a different


player who maximizes his payoff given his conditional probability
distribution over the strategy choices of his rivals.

(U. Padova) May 2020 6 / 20


Example 1

Consider a Cournot duopoly with inverse-demand function P (Q ) = a − Q


where Q = q1 + q2 . The marginal cost of Firm 1 is c1 = 0, and is
common knowledge. Firm 2’s marginal cost c2 is its own private
information. It can take values of c2 = cH with probability µ,and cL with
probability 1 − µ. Each firm maximizes its expected profit.

(U. Padova) May 2020 7 / 20


Here, Firm 1 has just one type, and Firm 2 has two types: cH and cL .
A strategy for Firm 1 is a real number q1 ≥ 0, while a strategy of
Firm 2 is a pair of real numbers q2 (cH ) and q2 (cL ), one for each firm
2’s type.

max (a − (q1∗ + q2 )) q2 − cH q2
q2H

FOC

a − q1∗ − cH
q2H ∗ =
2
Similarly,
a − q1∗ − cL
q2L∗ =
2

(U. Padova) May 2020 8 / 20


 
max a − (q1 + µq2H + (1 − µ)q2:L ) q1
q1

FOC
a − µq2H ∗ − (1 − µ)q2:L∗
q1∗ =
2
A Bayesian Nash equilibrium simultaneously solve the three equations:
a−q2H −cH a − q L − cL
2
a−µ( )−(1−µ)( ) a−(1−µ)cL +µcH
q1∗ = 2
2
2
⇒ q1∗ = 3
a−(1−µ)cL +µcH
a− −cH
q2H ∗ = 3
2 ⇒ q2H ∗ = 31 a − 12 cH − 61 µcH + 16 cL (1 − µ)
a−(1−µ)cL +µcH
a − − c
q2L∗ = 3
2
L
⇒ q2L∗ = 31 a − 12 cL − 16 µcH + 16 cL (1 − µ)

(U. Padova) May 2020 9 / 20


First Price Auction

There is an object to be sold. Two bidders want to buy it through an


auction. Simultaneously, each bidder i submits a bid bi ≥ 0. Then, the
highest bidder wins the object and pays her bid. If they bid the same
number, then the winner is determined by a coin toss. The value of the
object for bidder i = 1, 2 is vi , which is privately known by bidder i.
Assume that v1 and v2 are independently and identically distributed with
uniform distribution over [0, 1]. Assume that the player i 0 s bid function
(strategy) is an increasing function of his own value and find a symmetric
BNE.

(U. Padova) May 2020 10 / 20


Each realization vi is a type for player i.

(U. Padova) May 2020 11 / 20


Each realization vi is a type for player i.
Player i 0 s strategy is a function bi (vi ) : [0, 1] → R+

(U. Padova) May 2020 11 / 20


Each realization vi is a type for player i.
Player i 0 s strategy is a function bi (vi ) : [0, 1] → R+
The utility of each player i is

 vi − bi if bi > bj for j 6= i
vi −bi
ui (v1 , b1 , b2 ) = 2 if bi = bj
0 if bi < bj

(U. Padova) May 2020 11 / 20


Eui = (vi − bi )Prob(bi ≥ b (vj ))
= (vi − bi )Prob(vj ≤ b −1 (bi ))
= (vi − bi )b −1 (bi )

where the last inequality follows by the assumption of uniform distribution.


Example: if player i offers 13 , he wins with the probability that her bid is
higher than the player j’s bid, or, equivalently, that player j’s valuation is
lower than the value v 1 for which j offers 13 , and given the uniform
3
distribution assumption the probability that vj is lower than v 1 is exactly
3
v1 .
3

(U. Padova) May 2020 12 / 20


Each bidder aims to maximize her expected utility. The first order
condition of the problem is then

db −1
−b −1 (bi∗ (vi )) + (vi − bi∗ (vi )) = 0
dbi |bi =bi∗ (vi )
1
−b −1 (bi∗ (vi )) + (vi − bi∗ (vi )) 0 = 0
b (vi )|bi =bi∗ (vi )

(U. Padova) May 2020 13 / 20


Each bidder aims to maximize her expected utility. The first order
condition of the problem is then

db −1
−b −1 (bi∗ (vi )) + (vi − bi∗ (vi )) = 0
dbi |bi =bi∗ (vi )
1
−b −1 (bi∗ (vi )) + (vi − bi∗ (vi )) 0 = 0
b (vi )|bi =bi∗ (vi )

Then
1
−vi + (vi − b ∗ (vi )) = 0,
b ∗0 (vi )
b ∗0 (vi )vi + b ∗ (vi ) = vi

(U. Padova) May 2020 13 / 20


Easy computation shows

vi2
b ∗ ( vi ) vi = + const.
2
However the equality holds for all vi ≥ 0 and therefore also for vi = 0.
Therefore const = 0 and we can conclude
vi
b ∗ (vi ) = .
2

(U. Padova) May 2020 14 / 20

You might also like