ECON2112 Game Theory and Business

Strategy
Chapter 1 Static Games of Complete Information
1.1 Basic Theory: Normal-Form Games and Nash Equilibrium
1.1A A Normal-Form Representation of Games
Game Theory
 Game theory studies multi-person decision problems and hence deals with strategic situations
 A game (strategic situation or multi-person decision problem) is a situation in which players (two or
more individuals) interact and thereby jointly determine the outcome
o In a normal form game, players take decisions simultaneously
o In an extensive form game, players take decisions sequentially
 (Typically) every player can modify the outcome but (typically) no player has full control (cannot
determine the entire outcome)
 The aim of game theory is to provide at least one solution for every game
o A solution is a set of recommendations, one for each player, about how to play the game,
such that no player would have an incentive not to follow this recommendation
o Hence, the solution must be stable
 In static or simultaneous-move games players simultaneously choose actions and then receive
payoffs that depend on the combination of actions chosen
 In dynamic or sequential move games one player chooses their action before the other players
choose their action
o Can use normal-form or extensive-form representations
o Central issue in dynamic games is credibility
o In games of complete information, each player's payoff function is common knowledge
among all players

Normal Form Representation of Games: General Case

Elements of a Game:
 The players in the game (decision-making agents)
 The strategy set for each player (the strategies available to each player)
 The players’ preferences (payof) over possible outcomes (strategy profiles)
Definition (Finite Normal Form Game)
A n-player finite normal form game G=(S 1 , … , Sn ; u1 ,… , un ) consists of:
 For each player i=1, … , n , a set of pure strategies S i available to player i
 For each player i=1, … , n , a utility or payoff function ui that, for each pure strategy profile
( s1 , … , sn ) ∈ S 1 × …× S n specifies a real number ui ( s1 , … , s n) ϵ R

Game Theory Assumptions

 Players have common knowledge about the structure of the game.
 Players choose their strategies independently
 If players can costlessly communicate, then the messages they send are part of their strategies and
the game specifies what messages can be sent

1.1B Iterated Elimination of Strictly Dominated Strategies

Dominated Strategies and Admissibility
Definition (Strictly Dominated Strategy)
Given a game G, the (mixed) strategy pi is strictly dominated by p'i ≠ pi if for every pure strategy

profile of the opponents (s 1 , … , si−1 , s i+1 ,… , sn ) we have:

U i ( s 1 , … , si−1 , pi , s i+1 , … , sn ) < U i (s 1 ,… , si−1 , p'i , s i+1 , … , s n)
'
i.e. pi is strictly dominated by strategy pi if for each feasible combination of the other player’s
'
strategies, i’s payoff from playing pi is strictly less than i’s playoff from playing pi

Definition (Weakly Dominated Strategy)

'
Given a game G we say that strategy pi is weakly dominated by strategy pi if for every pure strategy

profile of the opponents s−i=(s 1 , … , si−1 , si +1 , … , s n )

'
U i (s−i , pi ) ≤U i ( s−i , p i)
'
and there exists at least one pure strategy s− ⅈ such that:
U i ( s−' ⅈ , pi ) <U i ( s−ⅈ
'
, p 'i)
' '
i.e. The strategy pi is weakly dominated by pi if, whatever the other players play, pi is always at

least as good as pi and sometimes strictly better

Definition (Admissibility)
The principle that players should not use either strictly or weakly dominated strategies (special NE)

Iterated Elimination of Dominated Strategies

 The iterated elimination of dominated strategies is a common technique for solving games that
involves iteratively removing dominated strategies
 Condition in which a strategy is unequivocally better (always gives a larger payoff) than another no
matter what the other players do
 If the objective is to compute Nash equilibria, it is not safe to apply iterated deletion of weakly
dominated strategies
 Drawbacks:
o Each step requires a further assumption about what players know about each other’s
rationality e.g. assume that it is common knowledge that the players are rational
 Further assume not only that all the players are rational, but all the player know that
all the players are rational, and all the players know that all the players are rational
(ad infinitum)
o (assume all players are rational, and all players know all players are rational etc.)
o Process produces an imprecise prediction about the play of the game (in some games all
strategies survive the iterated elimination)
 If a player has a strictly dominated strategy, it will never be used in a Nash equilibrium
Theorem
 Let G be an n-player normal form game such that player i has a strictly dominated strategy in G.
 Let G′ be the n-player normal form game that is obtained from G by removing such a strictly
dominated pure strategy.
 Then G and G′ have the same set of Nash equilibria

1.1C Motivation and Definition of Nash Equilibrium

Nash Equilibrium
 A Nash equilibrium is a strategy profile such that no player has an incentive to unilaterally deviate
from their predicted strategy
o Thus, each player’s predicted strategy must be that player’s best response to the predicted
strategies of the other players
o This prediction, Nash Equilibrium is called strategically stable or self-enforcing
Formal Definition (Nash Equilibrium)
Denote as S the set of all possible strategy profiles S 1 × … × Sn and consider an n-player normal
¿ ¿ ¿
form game G= { s1 , … , s n ; u 1 , … , un } . The strategy profile s =( s1 , … , sn ) ϵ S is a Nash Equilibrium of

G if for each i=1, … , n (every s i ∈ S i ):

(s 1 , ⋯ , s ⅈ−1 , si , s i+1 , … , sn )
¿ ¿ ¿ ¿

¿ ¿ ¿ ¿ ¿
ui ( s1 ,⋯ , s ⅈ−1 , s i , s i+1 , … , sn )≥ ui ¿

Definition (Best Response):

Consider an n-player normal form game G= { S1 , … , S n ; u 1 , … , un } and a strategy profile ( s1 , … , sn ) .

The set of player i’s best responses against ( s1 , … , sn ) (denotes as Ri ( s 1 , … , s n ) ) is the set of player

i’s strategies that solve:

max
u (s , … , s i−1 , s'i , s i+1 , … , sn )
s 'i ∈ S i i 1

Theorem
¿ ¿
The strategy profile ( s1 , … , s n) is a Nash equilibrium if and only if for every i=1, .. . , n we have :
¿ ¿ ¿
s i ∈ Ri (s 1 , … , s n)

1.2 Applications
1.2A A Cournot Model of Duopoly – 1838 Model of Imperfect Competition
Assumptions
 There are two firms, 1 and 2 competing in a duopolistic market
 These firms produce an identical (homogeneous) good
 They to choose the quantities that they are going to produce simultaneously
 The market clearing price when Q is the total amount in the market is:

{
p= a−Q if Q<a
0 if Q ≥ a

o P(Q)=a−Q is the market-clearing price when the aggregate quantity is Q=q 1 +q2

 Producing qi units of the good costs c qi dollars to firm I (assume c <a )

o Constant marginal cost of production is c ∴ symmetric game

The Game (Normal Form Representation)

 Players: Firms 1 and 2
 Strategy: qi =[ 0, ∞ ) for i=1, 2
 Payofs: determined by a function (incorporates each strategy profile)

{
π i ( q1 , q2 )= p q i−c qi= ( a−q 1−q 2 ) qi−c q i
−c q i
if q 1+ q2 <a
if q 1 +q2 ≥ a
  Face the same price as firms are competing in a market of undistinguishable goods

Solution
¿ ¿
A strategy profile (q 1 , q 2) is a Nash equilibrium if:
π 1 ( q¿1 , q¿2 ) ≥ π 1 ( q1 , q¿2 ) for every q 1 ∈ ¿

π 2 ( q¿1 , q¿2 ) ≥ π 2 ( q ¿1 , q 2) for every q 2 ∈ ¿

 The best strategy for firm 1, is to choose the q1 that maximises:

max π q , q = a−q −q q −c q
( ) ( 2) 1
q1 1 1 2 1 1

δπ1
=0=a−c−q2−2 q 1
δ q1

1
q1 =R1 ( q2 )= (a−c−q 2)
2

 The best response is a function as the strategy set is continuous

¿ ¿
 In the Cournot duopoly model, the analogous statement is that the quantity profile ( q 1 , q 2 ) is a
¿
NE if, for each firm i, qi solves the first-order condition for firm i's optimisation problem
o By solving the following equations simultaneously, we satisfy both best response functions
¿ ¿
2 q =(a−c−q 2)
1

2 q¿2=(a−c−q ¿1)

o Ensures that both firm 1 and 2 are best responding to the other
¿ ¿ a−c
q1 =q2 =
3

Less than a−c as assumed.

 Rather than solving for the Nash equilibrium in the Cournot game algebraically, one could instead
proceed graphically by finding the intersection of the best response functions:

 Solution, applying iterated deletion of dominated strategies:

o S 1=[ 0, ∞ ) ; S2 =¿

a−c q2
q1 = −
2 2
a−c
 Assume q ≥ 0 , ∴ R1 (0)= 2

a−c
 2 becomes the maximum quantity ( q m ) as firms can never play a negative

quantity
 a−c
o The monopoly quantity 2 strictly dominates any larger quantity

[
S 11= 0,
a−c
2 ] [
; S12= 0,
a−c
2 ]
 R1 ( a−c2 )= a−c2 − 12 ( a−c2 )= a−c4
a−c a−c
 4 becomes the minimum quantity as firms will never produce above 2
o (infinite iterations) … Process ends at Nash equilibrium
a−c a−c
S ∞1 ={ }; S∞2 ={ }
3 3

1.2B Bertrand Model of Duopoly – 1883 Model of Imperfect Competition

Assumptions
 There are two firms, 1 and 2 competing in a duopolistic market
 These firms produce an identical (homogeneous) good
 Demand absorbs exactly Q units of the good at all prices
 Each firm i chooses a price in the interval [ 0, ∞ ) given total demand Q
 Buyers will always buy from the firm with the lowest price
Q
 If both firms set the same price, each firm will sell 2 units

The Game
 Players: Firms 1 and 2
 Strategy: pi=[ 0, ∞ ) for i=1,2
 Payoffs for firms:

{
( p1−c ) Q if p1 < p2 F 1 undercuts F 2( consumers buy all their products)
π 1 ( p1 , p 2) = 1 ( p 1−c ) Q if p1= p 2 Same price(firms receive half each)
2 F 1 sells nothing , price above competitors
if p1 > p2
0

Solution
¿ ¿
 Nash equilibrium: Prices ( p1 , p2 ) such that:
π 1 ( p¿1 , p ¿2 ) ≥ π 1 ( p 1 , p¿2 ) for every p1 ∈¿

π 2 ( p¿1 , p ¿2 ) ≥ π 2 ( p ¿1 , p2 ) for every p 2 ∈¿

¿ ¿
 The unique Nash equilibrium is p1= p2=c

o Undercutting will result in a loss i.e. producing below MC

o Cannot steal demand by producing at a price above MC
o Choosing a price equal to c is a weakly dominated strategy

Bertrand Model of Duopoly with Differentiated Products

 Assumptions – if firms choose prices pi and p j , firm i’s demand curve is given by:
 qi ( p i , p j )=a− pi +b p j

o b>0 reflects the extent to which firm i's product is a substitute for firm's j product
o Unrealistic demand function because demand for firm i's product is positive even when firm
i charges an arbitrarily high price, provided firm j also charges a high enough pric
o We assume b<2 to ensure the problem makes sense
 The game is the same as the normal Bertrand Model of Duopoly
 Solutions:
o Given p2 firm one chooses a price p1 so that:
max π p , p = p −c q =( p −c ) a− p +b p
( ) ( 1 ) 1 ( 2)
p1 1 1 2 1 1

1
p1=R1 ( p 2 )= (a+b p2 +c )
2
¿ ¿
o The profile ( p1 , p2 ) is a Nash equilibrium if it solves:
¿ 1 ¿ ¿ 1 ¿
p1= ( a+ b p 2+ c )∧ p2= (a+b p1 +c )
2 2

1. 1.2D The Problem of the Commons

Assumptions – Bucolic example of farmers
 There are n farmers in a village who each must decide how many goats to own
 Goats will have to share the same limited space, hence, the benefit of having a goat will decrease on
the total number of goats
2
 Per goat benefit is equal to A−G when the total number of goats is G
 The cost of buying and caring for a goat is c (constant marginal cost per goat)
 Assume A >c and that goats are perfectly divisible e.g. 3/17 goat

The Game
 Players: n farmers
 Strategy sets: gn= [ 0, ∞ ) for all n
 Payofs: given a strategy profile (g1 ,. . . , gn ) the payoff to farmer i is:
A−( g1 +…+ gn ¿2 ) −c g i
π i ( g1 , … , gn ) =gi ¿

Solution
g
 Take the first order condition using the chain rule ( g i part of (¿ ¿ 1 , … , gn ) to obtain:
¿
g1 +…+ gn ¿2−2 ( g1 +…+ g n ) g i−c=0
A−¿
 Aggregate over every farmer:
2
g1 +…+ gn ¿ −nc=0
g1 +…+g n ¿ 2−2 ¿
nA−n ¿
  Total number of goats in a Nash equilibrium:
¿
A−c ¿
¿
n
¿
2+n
G=( g1 + …+ g n )=¿

Optimality
 The optimal number of goats is the amount that maximises total payoff:
G( A−G2)−cG
 Taking the first order condition by chain rule:
2 2
A−G −2G −c=0
From where we can obtain a value for the socially optimal number of goats:
¿
A−c ¿
¿
n
¿
2+ n
1
1
[ ]
G= ( A−c ) 2 vs .Ge =¿
3

 In a NE there are always more goats than the socially optimal (as long as n>1 )
e
 As soon as there is more than one farmer, G will be greater than socially optimal

Extension: Hotelling Model of Political Competition

Assumptions
 Consider a population of voters uniformly distributed along the ideological spectrum from left (
x=0 ) to right ( x=1 )
 Consider an election with two parties, 1 and 2
 Each party must announce its political ideology (and that they only care about being elected)
o Suppose that the space of political ideologies is I i =[ 0,1] for i=1,2
 The voters observe the candidates’ choices, and then each voter votes for the candidate whose
platform is closest to the voter’s position on the spectrum
 The party with most votes wins the election (in a tie, a fair coin is tossed)

The Game
 Strategies: S 1=S2= [ 0,1 ]
 Payoffs:
1
o When I 1 < I 2 party 1 gets a share of the votes equal to (I + I )
2 1 2
1 1
 u1=1∧u2=0 when ( I 1 + I 2 ) >
2 2
1
o If I 1 =I 2 then u1=u2 =
2

Solution
 ¿ ¿
 Ideologies I 1 and I2 are a NE if no party can unilaterally increase their
chances of winning the election.
1
 The unique Nash equilibrium is I 1 =I 2=
2

1.3 Advanced Theory: Mixed Strategies And Existence of Equilibrium

1.3A Mixed Strategies
Mixed Strategies
 In any game in which each player would like to outguess the other, there is no pure NE because the
solution to such a game necessarily involves uncertainty about what the players will do.
o ∴ Mixed strategy can be interpreted in terms of one player's uncertainty about what the
players will do
 If there is no belief that play i could hold about the strategies the other players will choose such
that it would be optimal to play the strategy s i , then there exists another strategy that strictly
dominates s i
A pure strategy may be strictly dominated by a mixed strategy, even if the pure strategy is not

strictly dominated by another pure strategy
Definition (Mixed Strategy)
A mixed strategy for player i is a probability distribution pi over his set of pure strategies S i
E.g. if player 1 has two strategies, S i={T , B } , a mixed strategy could be αT + ( 1−α ) B for
0 ≤ α ≤1
Definition (Nash Equilibrium)
¿ ¿ ¿
The mixed strategy profile p =( p1 , … , pn ) is a Nash equilibrium of G if for each i=1, … , n
U i ( p ¿1 , … , p¿i−1 , p¿i , p¿i+1 , … , p ¿n ) ≥ U i ( p¿1 , … , pi¿−1 , p i , p¿ⅈ+1 , … , p¿n )

for every other mixed strategy pi available to player i

Theorem (Nash, 1950)

The n-player finite, normal form game G=(S 1 , . . ., Sn ; u1 ,. . . ,u n) has at least one Nash equilibrium,
possibly, in mixed strategies ( ∴ guarantees a Nash Equilibrium)
Note that:
Every pure strategy s i is a special case of mixed strategy that can be expressed as a mixed strategy
that gives probability one to s i and probability zero to every other strategy

If p¿ =( p¿1 ,… , p¿n ) is a Nash Equilibrium than for every s i ϵ S i :

U i ( p ¿1 , … , p¿i−1 , p¿i , p¿i+1 , … , p ¿n ) ≥ U i ( p¿1 , … , pi¿−1 , s i , p¿ⅈ+1 , … , p¿n )

Furthermore, player i is indifferent between any two pure strategies used with strictly positive probability
¿
by pi (otherwise he would play a pure strategy)

E.g. under the NE= ( 12 H + 12 T , 12 H + 12 T ) player 1 and 2 are both indifferent between H and T

Definition (Best Response)

Consider an n-player normal form game G={S 1 , . .. , Sn ; u1 ,. . ., un } and a mixed strategy profile
 ( p1 , p2 , … p n) . The set of player i’s best responses against the other player’s strategies,

( p1 , p2 , … , p n ) (denoted Ri ( p 1 , … , pn ) ) is the set of player i’s strategies that solve:

max '
' ui ( p1 , … , pi−1 , p i , pi+1 , … , p n)
pi

Definition (includes each player’s strategy possibly mixed)

¿ ¿ ¿
The strategy profile ( p1 , p2 , … , p n) is a Nash equilibrium if for every i=1, … , n we have
¿ ¿ ¿ ¿
pi ϵ Ri ( p1 , p 2 , … , p n)

2.1 Dynamic Games of Complete and Perfect Information

2.1A Theory – Extensive Form Games: Backwards Induction

Extensive Form Game Representation (specifies):
1. The players in the game
2. When each player has the move
3. If relevant, Nature’s moves and the corresponding probabilities attached to each move of Nature,
4. What each player can do at each of her opportunities to move
5. What each player knows at each of her opportunities to move
6. The payoff received by each player for each combination of moves that could be chosen by the
players

Perfect vs. Imperfect Information

 A game of perfect information is an extensive form game where each player has the same
information that would be available at the end of the game
o Each player can always observe and remember other players’ past actions
o The player with the move knows the full history of the game thus every information set is a
singleton
 A game of imperfect information is an extensive-form game in which each agent's choice nodes are
partitioned into information sets where a player does not know all the past history of moves
o At least one nonsingleton info set

 The game starts at an open circle called the initial node

 The black circles and the open circle are called decision nodes
 A payof vector is specified in each final (terminal) node
 N represents Nature
 An information set contains the decision nodes that the player who moves at that information set
cannot differentiate
 o Each information set is labelled by the player who moves at that information set
o Information sets can be singletons (i.e. consist of just one element)
o At every information set the extensive form specifies the set of choices available
o Note that decision nodes contained in the same information have the same set of choices
available.
 There is a unique history of choices from the initial node to each decision node and ending node
 Example 1: L, R, a, b, c, d are called choices (or actions)

Information Sets
Definition (Information Sets)
 Information sets tell us when players are moving.
 An information set contains all the information that a player possesses when he must move.
 When a player moves at some information set, this player cannot distinguish between decision
nodes (i.e. the histories leading to the decision nodes) contained in that information set
 Every decision node inside the same information set must have the same set of choices available.

Perfect Recall
Definition (Perfect Recall)
When a player has to move he remembers all his previous moves and whatever he knew before. Not:

Pure, Mixed and Behavioural Strategies

In an extensive form game there are three different concepts of strategies:
Pure strategies
In an extensive form game, a player’s pure strategy is a plan of action for the entire game, that tells the
player which particularly choice to take at:
1. Every juncture (node) of the game where she must move (games of perfect information)
2. Every possible information set (games of imperfect information)

L , R , a , b , c ,d are the choices

Player 1’s set of pure strategies: S 1={L , R }

Player 2’s set of pure strategies: S 2={ac , ad , bc , bd }

Once we know what the set of pure strategies is for each player, we can write down the normal form
representation of an extensive form game and solve for Nash Equilibrium

A perfect information game with moves of Nature

Player 1’s set of pure strategies: S 1={L , R }
Player 2’s set of pure strategies: S 2={ac , ad , bc , bd }

Mixed strategies
In an extensive form game, a player’s mixed strategy is a probability distribution on the set of that
player’s pure strategies
Theorem (Kuhn)
In any extensive form game with perfect recall, every probability distribution on ending nodes induced by
a mixed strategy profile can also be induced by some behavioural strategy profile
 The consequence is that, although mixed strategies are more general, there is no loss of
generality in working only with behavioural strategies
Behavioural strategies
In an extensive form game, a player’s behavioural strategy specifies at:
1. Every decision node where a player must move (perfect information), or
2. Every information set where he must move (imperfect information)
a probability distribution over the choices that follow

Set of behavioural strategy profiles is the set of profiles that look like
this:
( xL+ ( 1−x ) R , ya+(1− y )b , zc +(1−z )d )

where 0 ≤ x ≤1, 0 ≤ y ≤ 1, 0≤ z ≤ 1

Example 2:
( xL+ ( 1−x ) R , ya+(1− y )b , zc +(1−z )d , wA +(1−w)B)

where: 0 ≤ x ≤1, 0 ≤ y ≤ 1, 0≤ z ≤ 1, 0 ≤ w≤ 1

A perfect information game with moves of Nature behavioural

strategy profiles:
( xL+(1−x) R , ya+(1− y)b , zc+ (1−z ) d )

where 0 ≤ x ≤1, 0 ≤ y ≤ 1, 0≤ z ≤ 1

Nash Equilibrium of an Extensive Form Game

Definition (Nash Equilibrium)
The Nash equilibrium of a perfect information game is the Nash equilibrium of its Normal form
representation (gives rise to unreasonable Nash equilibria)
 A non-credible threat where Nash equilibria in pure strategies are ( L ,a ) and ( R , b )

o Under ( R , b), player 2 is threating player 1 with playing b if he plays L

o If player 2 finds himself in a situation where he has to choose between a and b, he would
choose a (b is weakly dominated)
 Generalised through backward induction

Backwards Induction
Definition (Backwards Induction)
Players should optimise and make their choices in a way consistent with deductions about other players’
rational behaviour in the future. Implemented through use of the computational method.
Backwards Induction Computation in perfect information games
1. Find optimising behaviour at the last decision nodes in the game
 2. Substitute those decision nodes by the expected payoff induced by such optimising behaviour
3. Repeat the same procedure in the resulting perfect information game
The result of this procedure is the behavioural strategy

o This example produces the behavioural strategy profile: ( L , a , d)

o Which is equivalent to the pure strategy profile: (LA , ad )

2.1B Stackleberg Model of Duopoly Competition

Stackleberg (1934) Model of Duopoly Competition
 Stackleberg proposed a dynamic model of duopoly in which a dominant firm moves first, and a
subordinate firm moves second
 We develop the model under the assumption that firms choose quantities, as in Cournot’s model
 Two firms, 1 and 2, choose quantities
 Continuous action spaces
 The unit cost of production for both firms is constant and equal to c (constant MC)
 The inverse demand function is:
P=P(Q)=a−Q

 Contrasted to Cournot Competition where quantities are set simultaneously,

 Firm 1 chooses quantity first, then firm 2 observes the quantity chosen by firm 1 and then
chooses how much to produce (sequential move game involving computation of the SPE)
Stackleberg competition
 Start analysing the last subgames

Last stages
 We first compute firm 2's reaction to an arbitrary quantity by firm 1
 Suppose firm 1 chose q1 =q́1
o ( a−Q ) q2−c q 2

o Q=q́ 1 +q2 (fixed + decided)

 Firm 2 observes this and maximises:

 π 2 ( q́1 , q2 )=( a−( q́1 +q 2) ) q2−c q 2

 Therefore, upon observing q́1 , firm 2 will choose:

dπ
=0=a−q́1 q2−2 q 2−c
d q2
a−q́1−c
q ¿2 ( q́ 1 )=
2
¿
 Note that a strategy for firm 2 is a function q 2 :¿ →¿ such that, given a quantity produced by
¿
firm 1, q1 , tells how much to produce, q 2 ( q 1 )
 Here, R2 (q 1) is truly firm 2's reaction to firm 1's observed quantity

 Whereas in the Cournot analysis, R2 (q 1) is firm 2's best response to a hypothesized quantity
to be simultaneously chosen by firm 1
 Firm 1 will anticipate firm 2’s best response function
First stage
¿
 Firm 1 knows that if he produces q1 firm 2 will produce according to q 2 ( q 1 )
 Firm 1 will choose q1 to maximise:
π 1 ( q 1 )=( a−( q 1+ q¿2 ( q 1) ) ) q 1−c q1

( (
π 1 ( q 1 )= a− q1 +
a−q1−c
2 ))
q1−c q 1

d π 1 ( q1 ) a−c
=0 → q¿1 =
d q1 2

Subgame perfect equilibrium strategies:

a−c ¿ a−q 1−c
q1¿ = ,q 2 ( q1 ) =
2 2
Subgame perfect equilibrium outcome:
a−c ¿ ¿ a−c
q1¿ = ,q 2 ( q1 ) =
2 4
The market price is lower in this game, so aggregate profits are lower, so the fact that firm 1 is better off
implies that firm 2 is worse off in the Stackelberg than in the Cournot game

2.1D Sequential Bargaining

General Case
 Players 1 and 2 are bargaining over one dollar
 They alternate in making offers – first player 1 makes a proposal that player 2 can accept or reject
 If 2 rejects that 2 makes a proposal that player 2 can accept or reject
 This continues so on
 Once an offer has been rejected, it ceases to be binding and is irrelevant to the play of the game
 Each offer takes one period, and the players are impatient: the discount payoffs received in later
periods by the factor δ per period, where 0< δ<1

Three-Period Bargaining Game

  At the beginning of the first period, player 1 proposes to take a share of s 1 of the dollar, leaving
1−s1 for player 2

 Player 2 either accepts the offer (in which case the game ends and the payoffs are immediately
received) or rejects the offer in which case play continues to the second period
 At the beginning of the second period, player 2 proposes that player 1 take a share of s 2 of the
dollar, leaving 1−s2 for player 2 (convention that s t always goes to player 1)
 Player 1 either accepts the offer or rejects the offer
 At the beginning of the third period, player 1 receives a share s of the dollar, leaving 1−s for
player 2 where 0< s<1 (this is given exogenously)

Solution by backwards induction

 t=3: payoffs are (s ,1−s )
 t=2: player 1 accepts any split (s 2 ,1−s 2) such that s 2 ≥ δs
o If player 2 offers s 2=δs , player 1 accepts and payoffs are (δs , 1−δs)
o If player 2 offers s 2 <δs ,, player 1 rejects and payoffs are (δs , δ (1−s ))
o Therefore, player 2 proposes a split (δs , 1−δs)
 t=1: player 2 accepts any split (s 1 ,1−s 1) such that 1−s1 ≥ δ (1−δs)
o If player 1 offers 1−s1=δ (1−δs) , player 2 accepts and payoffs are
(1−δ (1−δs) , δ (1−δs))

o If player 1 offers 1−s1=δ (1−δs) , player 1 rejects and payoffs are (δ 2 s , δ(1−δs))
 Therefore, player 1 proposes a split (1−δ(1−δs) , δ (1−δs)) In the subgame perfect equilibrium,
payoffs are (1−δ(1−δs), δ (1−δs))

2.2 Two-Stage Games of Complete but Imperfect Information

2.2A Theory: Subgame Perfection
Subgames
 Subgame (general): the piece of a game that remains to be played beginning at any point at which
the complete history of the game thus far is common knowledge among the players
 A subgame in an extensive-form game:
o Begins at a decision node n that is a single info set (not the initial node)
o Includes all decision and terminal nodes following n in the game tree (but no nodes that do
not follow n), and
o Does not cut any information sets
 A subgame with an info set is not relevant to the original game because the player does not know
the move the player before made
Definition (Subgames)
A subgame is a part of an extensive form game that can be considered as a separate game. Particularly:
 If the subgame contains a decision node x then it also contains the whole information set that
x belongs to

 If the subgame contains a decision node x then it also contains every choice, decision node,
 and final node that comes after x
Note: Every extensive form is also a subgame of itself
Definition (Proper subgame)
A proper subgame is a subgame that is different from the entire game that contains it
 A game with three subgames vs. a game without proper subgames:

2.3 Repeated Games

2.3A Theory: Two-Stage Repeated Games
 Repeated games: a fixed group of players play a given simultaneous-move game repeatedly, with
the outcomes of all previous plays observed before the next play begins
o E.g. Prisoner's dilemma example

2.3B Theory: Infinitely Repeated Games

2.4 Dynamic Games of Complete but Imperfect Information

2.4A Extensive Form Representation of Games
 The extensive-form representation of a game specifies:
o (1) The players in the game
o (2a) Ehen each player has to move
o (2b) What each player can do at each of his or her opportunities to move
o (2c) What each player knows at each of his or her opportunities to move
o (3) The payoff received by each player for each combination of moves that could be chosen
by the players
 A strategy for a player is a complete plan of action – it specifies a feasible action for the player in
every contingency in which the player might be called to act
o Necessary as we could not apply the notion of NE to dynamic games if we allowed a player's
strategy to leave the actions in some contingencies unspecified
o For player j to compute a best response to player i's strategy, j may need to consider how i
would act in every contingency, not just contingencies i or j things likely to arise
 An information set for a player is a collection of decision nodes satisfying:
o The player has the move at every node in the information set and
o When the play of the game reaches a node in the information set, the player with the move
does not know which node in the info set has or has not been reached
 Implies the player must have the same set of feasible actions at each decision node
in an information set
 Otherwise each player would be able to infer from the set of actions available that
some node(s) had or had not been reached

2.4B Subgame-Perfect Nash Equilibrium

 To find the Nash equilibrium of an extensive form game to:
1. Compute the normal form representation
2. Compute the set of Nash equilibria of that normal form.
 When an extensive-form game has subgames, there is a (stricter) equilibrium concept that can be
used to find a more convincing equilibrium which requires a NE in every subgame of the original
o In games of complete and perfect information, backwards induction eliminates noncredible
threats
  Perfect information: every information set is a singleton  each decision node in the
tree represents a contingency in which a player could be called upon to act.
 The process of backwards induction, node by node, amounts to forcing each player
to consider carrying out each threat the player might make
o Some games have multiple Nash equilibria but have one equilibrium that stands out as the
compelling solution to the game
 NE vs SP outcome: an NE is a collection of strategies whereas a SP outcome will happen only in the
contingencies that are expected to arise, not in every contingency that might arise
Definition (Subgame Perfect Equilibrium)
A strategy profile is a Subgame Perfect Equilibrium of an extensive form game if it induces a Nash
equilibrium in every subgame – in any finite dynamic game of complete information, perhaps in MS
Subgame Perfect Equilibrium in games of perfect information, it is the same as backwards induction i.e.
the subgame-perfect outcome of games is the natural extension of backwards induction to these games
Properties
 Every subgame perfect equilibrium is a Nash equilibrium
 Not every Nash equilibrium is subgame perfect
 However, subgame perfection has no bite if the extensive form game does not have proper
subgames

Chapter 3 Static Games of Incomplete Information

3.1 Theory: Static Bayesian Games and Bayesian Nash Equilibrium
3.1A An Example: Cournot Competition under Asymmetric Information
 Consider a Cournot duopoly model with:
o Inverse demand is P(Q)=a−Q , where Q=q 1 +q2 is the aggregate market quantity
o Firm 1’s cost function is C1 ( q1)=c q1
o Firm 2’s cost function, however, is C2 (q2)=c H q 2 with probability θ and
C2 (q2)=C L q 2 with probability 1−θ , where c L <c H
 Information is asymmetric: firm 2 knows its cost function and firm 1’s, but firm 1 knows its cost
function and only the probabilities of firm 2’s marginal costs being C L or C H
o This is common knowledge – firm 1 knows that firm 2 has superior information, firm 2 knows
that firm 1 knows this, and so on
 Firm 2 may want to choose a different (and presumably lower) quantity if its marginal cost is high
than if it is low
o Firm 1, for its part, should anticipate that firm 2 may tailor its quantity to its cost in this way.
¿ ¿ ¿
 Let q 2( c H ) and q 2(c L ) denote firm 2’s quantity choices as a function of its cost, and q1 let
denote firm 1’s single quantity choice.
¿
 If firm 2’s cost is high, it will choose q 2 ( c H ) to solve:
max [ ( a−q ¿1−q 2 )−c H ] q2
q2

¿
 Similarly, if firm 2’s cost is low, q 2 ( c L ) will solve:
max [ ( a−q 1−q 2 )−c L ] q 2
¿

q2

 Finally, firm 1 knows that firm 2’s cost is high with probability θ and should anticipate that firm
¿ ¿
2’s quantity choice will be q 2(c H ) or q 2(c L ) depending on firm 2’s cost.
 ¿
 Thus, firm 1 chooses q1 to solve (so as to maximise expected profit):
[ ] [ ]
max θ ( a−q1−q ¿2 (c H ) ) −c q1 +(1−θ) ( a−q 1−q ¿2( c L ) ) −c q1
q1

 The first-order conditions for these three optimisation problems are:

¿ a−q¿1−c H
q 2 ( c H )=
2
a−q ¿1−c L
q ¿2 ( c L ) =
2

q=¿ [ ] [
θ ( a−q1 −q2 (c H ) ) −c +(1−θ) ( a−q 1−q 2 (c L ) ) −c
¿ ¿
]
1
2
 The solutions to the three-first order conditions are:
a−2c H + c 1−θ
q ¿2 ( c H )= + (c H −c L )
3 6
a−2 c L +c θ
q ¿2 ( c L ) = − (c H −c L )
3 6
a−2 c +θ c H + ( 1−θ ) c L
q1¿ =
3

3.1B Normal-Form Representation of Static Bayesian Games

Games of Incomplete Information
 Games of incomplete information – also called Bayesian games
o Static: I don’t know who you really are, but I have an idea, the timing being;
 The players simultaneously choose actions (player i chooses ai from the feasible
set A i ), and then;

 Payoffs ui ( a1 , … , an ) are received

o Dynamic: I don’t know who you really are, but I have an idea and I see what you did
o At least one player is uncertain about another player’s payoff function

Normal Form Representation

 We start by considering how each player knows his or her own payoff function but may be uncertain
about the other players’ payoff functions
o Let player i's possible payoff functions be represented by ui (a 1 , … , an ; t i ) , where t i is
called player i's type and belongs to a set of possible types (type space) T i
o Saying that a player i may be uncertain about the other players’ payoff functions is
equivalent to saying that player i may be uncertain about the type of the other players
 This is denoted by: t −i =(t 1 , … , t i−1 , t i +1 , … , t n)
 We use t −i to denote the set of all possible values of t −i
 We use the probability distribution pi (t −i ∨t i ) to denote player i’s belief about the
other players’ types, t −i , given player i’s knowledge of his or her own type, t i
Definition (Normal form representation)
The normal-form representation of a n-player static Bayesian game specifies the players’ action spaces
 A 1 , … A n , their type spaces T 1 , … T n , their beliefs p1 , … , pn , and their payoff functions
u1 , … ,u n . Player i's type t i is privately known by player i, determines player i's payoff function,

ui (a 1 , … an ; t i) , and is a member of the set of possible types, T i . Player i's belief pi (t −i ∨t i )

describe i's uncertainty about the n-1 other players’ possible types, t −i given i's own type, t i . We
denote this game by G=( A , T , p ,u ) .
 Following Harsanyi (1967), we will assume that the timing of a static Bayesian game is:
o Nature draws a type vector t=(t 1 , … ,t n) , where t i is drawn from the set of possible
types T i (assume this is done according to the prior probability distribution p(t)
 When nature then reveals t i to player i, he or she can compute the belief
pi (t −i ∨t i ) using Bayes’ rule:

p (t −i , t i ) Pr ( t i|t −i ) p(t −i)

pi ( t −i|t i )= =
p(t i) ∑ Pr ( t i|z ) p( z )
zϵT −i

 Frequently assume that the players’ types are independent, in which case pi (t −i )

does not depend on t i but is still derived from the prior distribution p(t)
o Nature reveals t i to player i but not to any other player
o The players simultaneously choose actions, player i choosing ai from the feasible set
Ai

o Then (4) payoffs ui (a 1 , … , an ; t i ) are received

 Sometimes the informed firm’s type enters the uninformed firm’s payoff function,
this is denoted by ui (a 1 , … , an ; t 1 , … , t n )
 We need:
o Players
o Actions (choices) they can take
o Types for each player, representing all possible different information they can privately know
and affects payoffs (this way included “states”)
o Payoffs that depend on actions and types (potentially on the type of other players)
o A prior probability distribution over types
 We need to specify beliefs that players have
o A belief is a probability distribution over all the nodes in an information set
o In static games, this arises only if my type tells me something about the other player’s type
o Call the prior distribution p e.g. p (1 ) =Pr ⁡( t=1)
o By Bayes’ rule the belief pi (t −i ∨t i ) that player i of type t i has about -i’s type is given
by:
p (t −i , t i ) Pr ( t i|t −i ) p(t −i)
pi ( t −i|t i )= =
p(t i) ∑ Pr ( t i|z ) p( z )
zϵT −i

o But if types are independent Pr ( t i|t −i ) = p(t i) and so pi ( t −i|t i )= p(t −i) : beliefs = prior

Definition (Normal form representation)

A normal form representation of an n-player game of incomplete information is:
 An action space A ≡ A1 ×… × A n
 Type space T ≡T 1 × … ×T n
 Payoff functions u ≡u1 , … ,u n defined over A ×T . That is, ui ( a|t ) , a ϵ A , t ϵ T

 Beliefs p≡ p1 , … , pn

We denote the game by G=( A ,T , p , u)

Definition (strategies)
In the static Bayesian game G=( A ,T , p , u) , a strategy for player i is a function s i (t i) which, for
each type t i ∈T i , specifies the action from the feasible set A i that type t i would choose if
drawn by Nature
 That is, we reduce the extensive form and define different strategies for each different player-type
o Each type is treated as a different player
 A rational player-type chooses her strategies to maximise:
o Her expected payoff
o Where the expectation is taken with respect to the set of types of all other players
o Each profile of types is “weighted” by beliefs pi (t −i ∨t i )
E[ui (a i , s−i (t −i)∨t )∨t i]≡ ∑ ui ( s 1 ( t1 ) , … , s i−1 ( t i−1 ) ,a i , si+1 ( t i+1 )|t ) p i (t−i ∨ti )
t−i ϵT −i

3.1C Definition of Bayesian Nash Equilibrium

 Given the timing of a static Bayesian game, in which nature begins the game by drawing the players’
types, a (pure) strategy for player i must specify a feasible action for each of i’s types
Definition (Strategy)
In the static Bayesian game G= { A 1 , … , A n ; T 1 , …T n ; p 1 , … , pn ;u 1 , … un } , a strategy for player i is a
function S i (t i) , where for each type t i in T i , s i (t i) specifies the action from the feasible set
that type t i would choose if drawn by nature
 Unlike (both static and dynamic) games of complete information, in a Bayesian game the strategy
spaces are not given in the normal-form representation of the game
 Instead, in a static Bayesian game the strategy spaces are constructed from the type and action
spaces: player i’s set of possible (pure) strategies, S i , is the set of all possible functions with
domain T i , and range Ai

o In a separating strategy, each type t i in T i chooses a different action ai from Ai

o In a pooling strategy, all types choose the same action

 Necessary to state strategies in order to apply the concept of Nash equilibrium
Definition (Bayes Nash Equilibrium)
In the static Bayesian game G= { A 1 , … , A n ; T 1 , …T n ; p 1 , … , pn ;u 1 , … un } , the strategy profile
s ¿1 ,… , s¿n
s ¿=¿ is a (pure-strategy) Bayes Nash equilibrium if for each player i and each of 1’s types
¿
t i ∈T i s i ( t i ) solves:

max ∑ ui ( s¿1 ( t 1) , … , s ¿i−1 ( t i−1 ) , ai , s¿i +1 ( t i+1 ) , … , s ¿n ( tn ) ; t ) p i (t−i∨ti )

ai ϵ A i t ϵ T
−i −i
That is, no player wants to change his or her strategy, even if the change involves only one action by one
type ( ∴ in a finite static Bayesian game there exists a Bayesian Nash equilibrium, perhaps in MS)
 Use Nash Equilibrium concept in an expanded game, where each different player-type has a
different strategy
o Equivalently, form conjectures about other player’s actions in each state and act optimally
given these conjectures
 Normal form representation of Bayesian games is reduced form of a simultaneous move game
where Nature chooses types (at a previous stage)
 Bayesian Nash equilibrium is a reduced form Nash equilibrium of that game with all player-types as
different players and beliefs “given” to players using Bayes’ rule
Definition (Bayes Nash Equilibrium)
s ¿1 ,… , s¿n
In the static Bayesian game G=( A ,T , p , u), the strategy profile is a (pure-strategy) Bayes
s ¿=¿

Nash equilibrium if for each player i and each of i’s types t i ∈T i

s ¿i ( t i ) ϵ arg max E[ui (a i , s−i (t −i)∨t )∨t i]

s i ϵ Si

 That is, each player-type is maximizing the expected payoff conditional on her own type

3.1 Applications
3.2A Mixed Strategies Revisited
Extending the Definition of Bayes Nash Equilibrium
 Note that if T is a continuum, then p must be defined as a density and expectations are in this form

[ ] ∫ u ( s (t ) , … , s
E ui ( a i , s−i ( t −i )|t i )|t i ≡
t −i ∈ T −i
i 1 1 i−1 ( t i−1 ) , ai , s i+1 ( t i +1 ) , sn ( t n )|t i) pi ( t −i∨t i)

3.2A An Auction
 First-price, sealed-bid auction
 Two bidders, i=1,2
 i’s valuation is v i
 v i U (0,1) (iid draws)
 Tie-breaking rule is fair coin
+¿ ≡ ¿
 Actions: bi ϵ R¿

 Types: T i =[0,1]
 Beliefs: pi ( v i|v j )= p ( vi ) , i≠ j
 Payoffs:

{
v i−b i
if b i>b j
v i−b i
ui ( bi , b j|bi )= if bi =b j
2 otherwise
0

 Strategies: we can write bi ( v i) , i.e. the bid that player 1 would choose if type i is drawn

Bayes Nash Equilibrium

  Each player-type must maximise her expected payoff
1
max ( v i−bi ) Pr ( bi >b j ( v j ) ) + ( v i −bi ) Pr ⁡( bi =b j (v j))
b i
2
 Let us focus on symmetric equilibria where players use linear strategies
 i.e. bi ( v i)=α + β v i , where α , β ∈ R are two parameters we have to find
1
max ( v i−bi ) Pr ( bi >α + β v j ) + ( v i −bi ) Pr ⁡( bi =α + β v j)
b i
2

 Since v j is uniformly distributed on (0, 1), the probability that it is exactly equal to any number is
0:
max ( v i−bi ) Pr ( bi >α + β v j )
bi

bi−α b i−α
 Notice that (
Pr ( b i> α + β v j )=Pr v j <
β
=) β
bi−α
max ( v i−bi )
bi β

2 v i+ α
f .o.c bi =
β β
 We get bi (vi) = vi2 2 + α
 How do we find α?
 Notice that no bidder should ever bid more than her valuation: bi (vi) ≤ vi
o Indeed, in a symmetric equilibrium there is always a chance of winning, hence any bid bi > vi
gives negative expected payoff
 In particular bi (0) =α 2≤ 0 (together with bi ∈ R+) ⇒ α = 0
 There is a symmetric (Bayes-Nash) equilibrium where bidders bid half their valuation
 Appendix 3.2B shows that this is the unique symmetric equilibrium with strictly increasing,
differentiable strategies.

Mixed Strategies Revisited

 Recall the Battle of the Sexes:

2 1 1 2
 Nash Equilibria: ( Opera ,Opera ) , ( Fight , Fight ) ,( 3 Opera+ 3 Fight , 3 Opera+ 3 Fight)

 A player mixed strategy might represent her opponent’s uncertainty about her choice of a pure
strategy (Harsanyi, 1973)
 Formally, let Chris’s payoff if both go to the Opera be 2+t c
 Similarly, Pat’s payoff if both go to the Fight is 2+t p
 t c and t p are uniformly distributed over [0, x ] and we will look at x → 0 (hence, a slight
perturbation of the payoffs, not a big one)

Threshold Strategies
 let us focus on (pure) strategies of the following type:
o Chris plays Opera if t c is greater than a threshold c
 o Pat plays Fight if t p is greater than a threshold p
x−c x− p
 notice that this implies that each play favourite option with probability x and x we

want to show that in equilibrium they both converge to 2/3 as x→0

Expected Payoffs
 Chris’ expected payoffs from playing Opera Fight are:
p p p
x
( ( )
2+t c ) + 1− ∙ 0= ( 2+t c )
x x

p p p
∙0+ (1− ) ∙ 1=1−
x x x
 Playing Opera is optimal if and only if:
x
t c ≥ −3=c
p
 Repeating the exercise for Pat gives:
x
p= −3
c

 This implies p=c and:

2
p + 3 p−x=0
p −3+ √9+ 4 x
Pr ( most pref . option )=1− =1−
x 2x
lim
 Taking the x→ 0 gives 2/3