Game theory

Phyllis Machio
 Definition
• Game theory is the study of interacting
decision makers where several players must
make choices that potentially affect the
interests of the other players.
• Game theory applies where actions of several
agents are interdependent.
• These agents may be individuals, groups,
firms, or any combination of these.
 Objective of game theory
• A key objective of game theory is to determine
the optimal strategy for each player.
• An optimal strategy for a player is the one that
maximizes his or her expected pay off.
 Nash equilibrium
• It is defined as a set of strategies or actions
such that each player is doing the best it can
given the actions of its opponents.
 Example 1
• Mary and John are discussing whether to take
micro or macro this sem.
• Mary gets utility 4 and John gets utility 2 if
they both take micro.
• The payoffs are reversed if they both take
macro.
• If they take different courses they both get 0
utility.
Example 1
 Refinement of Nash equilibrium
• In example 1 we have 2 Nash equlibria: top
left and bottom right
• Sometimes, like in this case, the Nash
equilibrium is not unique.
• We need refinement of Nash equilibrium
• We do this in 2 ways
 i) Dominant strategy equilibrium
• Dominant strategy equilibrium is a special
case of Nash equilibrium where every player
has a dominant strategy.
• Under DSE a strategy is optimal for a player no
matter what the other player does
 Example 2
• Aumann (1987) used a simple example in
which each player can simply announce to a
referee give me 1,000 or give the other player
3000.
 Example 2

• We have a unique equilibrium of defect

ii) Elimination of dominated strategies

• When there is no dominant strategy

equilibrium, we have to resort to the idea of
Nash equilibrium.
• But typically there will be more than one Nash
equilibrium.
• One problem is to try to eliminate some of the
Nash equilibria as being unreasonable.
 Cont
• One sensible belief to have about players’s behavior us
that it would be unreasonable for them to play
strategies that are dominated by other strategies.
• This suggests that when given a game, we should
eliminate all strategies that are dominated and then
calculate the Nash equilibria of the remaining game.
• This procedure is called elimination of dominated
strategies. It can sometimes result in a significant
reduction in the number of Nash equilibria. Consider
the game
 Cont
• We say r1 strictly dominates r2 for Row if the
payoff from strategy r1is strictly larger than
the payoff for r2 no matter what choice
column makes.
• The strategy r1 weakly dominates r2 if the
payoff from r1 is at least as large for all the
choices column might make and strictly larger
for some choice.
 Example 3
• We eliminate top and left and end up with a
unique Nash equilibrium of Bottom right
 Repeated games
• Prisoners dilemma
• Two prisoners have been accused of committing a crime.
• They are put in separate jails and cannot communicate with
each other.
• Each has been asked to confess. If both prisoners confess,
each will receive a jail term of five years.
• If neither confesses, the prosecution’s case will be difficult
to make and so the prisoners each receive terms of two
years after bargaining.
• If one prisoner confesses and the other does not, the one
who confesses gets 1 year term while the one who does
not confess gets a 10 year term.
• If you were one of the prisoners would you confess or not?
The payoff summarizes the possible outcomes
 Repeated games
• The Nash equilibrium here is to confess and
each gets a 5 year jail term.
• If the prisoners were to come together and
agree, they would choose not confess because
each gets 2 years.
• This is the cooperative outcome
Prisoner 2
Prisoner 1 Confess Don’t confess
Confess -5, -5 -1, -10
Don’t confess -10, -1 -2, -2
 Repeated games
• Suppose this game is repeated, what will be the outcome of
the game?
• The outcome depends on the number of times the game is
repeated
• Finite repetitions:
• Suppose the game is repeated a finite number of times say
N months.
• If the two individuals are rational, then they will reason as
follows, the other individual will be playing the cooperative
outcome until the last month when they have an incentive
to undercut and make huge profit and the other individual
has no opportunity to retaliate since at that point the game
ends.
 Repeated games
• The other individual however can figure out this
and do the same.
• They will reason the same way for the last but
one month.
• And since the same reasoning applies to each
preceding month, the only rational outcome is for
both individuals to choose the Nash equilibrium
of confess.
• The equilibrium is therefore for both individuals
to confess and get a 5 year jail term
 Repeated games
• Infinite repetitions
• Suppose two individuals repeatedly make decisions each period
infinitely, cooperative behavior (not confess) is the outcome of game
• Players use tit – for – Tat strategy where they respond in kind to an
opponent’s previous play cooperating with cooperative opponents
and retaliating against uncooperative ones.
• To see why, suppose one individual decides to undercut the other by
choosing not confess, in that period the individual gets only 1 year
jail term while the other gets 10 years.
• However in the subsequent period, the other individual will also
chose not confess so that both get a higher jail term of 5 years
forever forfeiting 2 years.
• Because the game is repeated infinitely the cumulative loss that
result (3 more years for each period indefinitely) outweigh any short
term gain (1 less year) that accrue during the first period of
undercutting. The equilibrium will be for both individuals not to
confess and get 2 year jail term
 Mixed strategies
• In the games we have examined so far, we have
considered strategies in which players make a
specific choice (advertise, not to advertise, invest or
not invest and so on).
• Strategies of this kind are called pure strategies.
• There are games, however, in which a pure strategy
is not the best way to play.
• In this case players play mixed strategies.
• Mixed strategy is one in which a player plays his
available pure strategies with certain probabilities.
 Mixed strategies
• This game has no equilibrium in pure
strategies
• It however has equilibrium in mixed strategies
• We use probabilities to determine mixed
strategy equilibrium

Player 2
Player 1 L R
U 2,2 3,1
D 3,-1 0,0
 Mixed strategies
• Suppose player 1 plays up with probability p and
down with probability 1-p
• And player 2 plays left with probability q and
right with probability 1-q
• To determine the expected payoff for one player,
we use the other player’s probabilities
• Expected payoff for player 1 playing up is
• 2q+3(1-q)=3-q
• Expected payoff for player 1 playing down is
• 3q+0(1-q)=3q
 Mixed strategies
• At equilibrium we equate the payoffs
• 3-q=3q
• Q=3/4 and 1-q=1/4
• Expected payoff for up becomes 3-q=3-3/4=9/3
• Expected payoff for down is 3p=3(3/4)=9/4
• Expected payoff of player 2 playing left is
• 2p+-1(1-p)=2p-1+p=3p-1
• Expected payoff from player 2 playing right is
• p+0(1-p)=p
 Mixed strategies
• At equilibrium we equate
• 3p-1=p
• P=1/2 and 1-p=1/2
• Expected payoff for left is 3p-1=3(1/2)=1=0.5
• Expected payoff for right is p=0.5
• Player 1 plays up ½ of time and down ½ of time
and gets an expected payoff of 9/4
• Player 2 plays left ¾ of time and right ¼ of time
and gets an expected payoff of 0.5
 Mixed strategies
• Example 2
• Two firms are making a decision on pricing
decision
• This game has no equilibrium in pure
strategies
Firm 2
Firm 1 High pricing Low pricing
High pricing 50,50 80,20
Low pricing 90,10 20,80
 Mixed strategies
• Firm 1 choses high pricing with probability p and
low pricing with probability 1-p
• Firm 2 choses high pricing with probability q and
low pricing with probability 1-q
• We use firm 1’s probability to determine
expected profit for firm 2 and vice versa
• Expected profit for firm 1 from high pricing
• 50q+80(1-p)=50q+80-80p=80-30p
• Expected profit for firm 1 from low pricing
• 90p+20(1-p)=90p+20-20p=20+70p
 Mixed strategies
• Equilibrium
• 80-30p=20+70p
• P=0.6
• Expected profit firm 1 HP=80-30(0.6)=62
• Expected profit firm 1 LP=20+70(0.6)=62
• Expected profit for firm 2 from high pricing
• 50q+10(1-p)=50q+10-10p=10+40p
• Expected profit for firm 2 from low pricing
• 20p+80(1-p)=20p+80-80p=80-60p
• Equilibrium
• 10+40p=80-60p
• P=0.7
 Mixed strategies
• Expected profit firm 2 HP=10+40(0.7)=38
• Expected profit firm 2 HP=80-60(0.7)=38
• Thus firm 1 choses high pricing 70% of the
time and low pricing 30% of the time and
earns an expected profit of 62
• Firm 2 choses high pricing 60% of the time
and low pricing 40% of the time and earns an
expected profit of 38
 Dynamic/sequential games
• In the games we have looked at, both players move at the
same time.
• In sequential games players move in turn.
• Sequential games are easier to visualize if the possible
moves are represented in form of a decision tree.
• This representation is called the extensive form of a game.
• Decision tree is represented by a collection of nodes and
branches.
• Nodes represent either decision point or outcome.
• Branches represent available decisions.
• The initial node indicates a decision point for a player while
bottom nodes represent outcome
 Dynamic/sequential games
• Consider the following game:
Firm 2
Firm 1 High pricing Low pricing
High pricing 50,50 80,20
Low pricing 90,10 20,80

• To solve the game we use backward induction.

• Examine the last decision points (actions of the last player)
• Eliminate actions that will not be played
• Re-draw the tree diagram
• Determine equilibrium using forward looking approach
• The payoffs of the first mover comes first
 Dynamic/sequential games
• Suppose firm 1 is the first mover
• Firm 2 is the last player
• It will not play low pricing if firm 1 choses high pricing since
50>20, this we eliminate low pricing
• It will also not high pricing if firm 1 choses low pricing
because 80>10 thus we eliminate HP
• We have 2 sub-game equilibia: HP, HP and both firms earn
50 and LP,LP and firm 1 gets 20 and firm 2 gets 80
• Firm one looks forward and anticipates this reaction from
firm 2 and picks high pricing since 50>20.
• HP,HP is therefore the sub game perfect equilibrium. Bothe
firms earn 50
 Dynamic/sequential games
• If firm 2 moves first, we have to swap the
payoffs because in the normal form
representation, the payoffs of firm 1 came first
but now the payoffs of firm 2 come first
• In the sub game equilibriums are HP,LP and LP,
HP
• The sub-game perfect equilibrium is LP, HP
where firm 2 gets 20 and firm 1 gets 80
 Dynamic/sequential games
• Games of imperfect and incomplete information
• This far, the game trees we have examined are ones
which the firms move one after the other and each
firm can observe all of the earlier actions (if any) taken
by its rivals.
• A game in which some players must make a move but
is unable to observe the earlier or simultaneous move
of some other player is said to be a game of imperfect
information.
• Similarly, whenever one or more of the players has
private information not know by others (payoff, cost)
the game is said to be a game of incomplete
information.
 Dynamic/sequential games
• Consider a variant to our prisoners’ dilemma
game
• Suppose player 1 has a standard selfish
preference
• Player 2 has either selfish preference or nice
preference
• Player 1’s payoff depend on whether player 2
is nice or selfish
 Dynamic/sequential games
• The normal form representation now is:
• If player 2 was selfish, both players have a
dominant strategy of defecting
• If player 2 was nice both players have a
dominant strategy of cooperating
 Dynamic/sequential games
• We can now represent this game sequentially
taking note of incomplete information
• That player 2 has private information not known
by player 1
• That it is either nice or selfish
• When player 1 moves, it doesn’t know player 2’s
move and this creates imperfect information
• We use a dotted line to show imperfect
information
 Dynamic/sequential games
• Suppose player 1 moves first
• We delete the branches that will not be
played by player 2 which is the last mover
• Then we compute expected payoff for player 1
 Dynamic/sequential games
• Player 2 will defect if it is selfish and cooperate if it is
nice
• For player 1 we compute the expected payoff because
on imperfect information
• Expected payoff from cooperate is
• 0p+6(1-p)=6-6p
• Expected payoff from defect is
• 2p+4(1-p)=2p+4-4p=4-2p
• 6-6p=4-2p
• 4p=2
• P=1/2
 Dynamic/sequential games
• To know the decision player 1 will make we say
suppose
• P>1/2 i.e. 2/3 player 2 is nice
• Expected pay off from cooperate is 6-6(2/3)=2
• Expected payoff from defect is 4-2(2/3)=2.67
• Hence if p>1/2 player 1 choses defect since 2.67>2
• Suppose P<1/2 i.e. 1/3 player 2 is selfish
• Expected pay off from cooperate is 6-6(1/3)=4
• Expected payoff from defect is 4-2(1/3)=3.3
• Hence if p<1/2 player 1 choses cooperate since 4>3.3
 Dynamic/sequential games
• Thus the Bayesian Nash equilibrium is player 2
to defect if it is selfish and cooperate if it is
nice
• Player 1 cooperates if p<1/2 i.e. p=1/3 and
player 2 is selfish
• Player 1 defects if p>1/2 i.e. p=2/3 and player
2 is nice
 Dynamic/sequential games
• Suppose now player 2 moves first
• We have to swap the payoffs because now player 2’s
payoffs come first yet in the normal form
representation, the payoffs for player 1 come first
• We start by deleting the branches not chosen by player
2 because it has no imperfect information
• If player 2 is selfish and choses C the maximum it gets
is 4 but if it choses D the maximum it gets is 6.
• Thus it will chose D when it is selfish and we delete C
• If player 2 is nice it will chose C because the maximum
it gets 6>4 thus we delete D
 Dynamic/sequential games
• For player 1 we calculate expected payoff
because of imperfect information
• Expected payoff from cooperate is
• 0p+6(1-p)=6-6p
• Expected payoff from defect is
• 2p+4(1-p)=2p+4-4p=4-2p
• 6-6p=4-2p
• 4p=2
• P=1/2
 Dynamic/sequential games
• To know the decision player 1 will make we say
suppose
• P>1/2 i.e. 2/3 player 2 is nice
• Expected pay off from cooperate is 6-6(2/3)=2
• Expected payoff from defect is 4-2(2/3)=2.67
• Hence if p>1/2 player 1 choses defect since 2.67>2
• Suppose P<1/2 i.e. 1/3 player 2 is selfish
• Expected pay off from cooperate is 6-6(1/3)=4
• Expected payoff from defect is 4-2(1/3)=3.3
• Hence if p<1/2 player 1 choses cooperate since 4>3.3
 Dynamic/sequential games
• Thus the Bayesian Nash equilibrium is player 2
to defect if it is selfish and cooperate if it is
nice
• Player 1 cooperates if p<1/2 i.e. p=1/3 and
player 2 is selfish
• Player 1 defects if p>1/2 i.e. p=2/3 and player
2 is nice