Chapter 3
Strategic Choices: Game Theory
Carmen Arguedas
Catedrática de Fundamentos del Análisis Económico
Dpto. Análisis Económico: Teoría Económica e Historia Económica
Room: 10-310
E-mail: [Link]@[Link]
Website: [Link]
Twitter: @CarmenArguedasT
OUTLINE
1. Introduction
2. Games of complete information
3. Games of incomplete information
REFERENCES
Gibbons
Kreps (ch. 11 - 15)
Nicholson (ch. 8)
Varian (ch. 15)
EXPERIMENT I
(beauty contest)
1. Introduction
Game theory is a tool frequently used to analyze strategic interactions among two or
more agents. In strategic settings, each individual may not have an obvious
choice that is best for him or herself. What is best for an individual may depend on
what the others are doing and viceversa.
Many applications:
Models of imperfect competition, contract theory, international economics,
labor economics, monetary economics, financial economics, public economics,
tournaments, contests, political science, psicology, biology…
The three components of a game are:
Each decision maker is called player
Players
N = {1, 2, …, n}
Strategies The possible choices available to the players are called
strategies si Є Si
Payoffs Utilities/ profits for playing strategies (s1,s2,…,si,…,sn)
ui = ui (s1, s2, …, si, …, sn)
πi = πi (s1, s2, …, si, …, sn)
It depends on the type of game and the information
Solution concept:
players have about the other players
STATIC GAMES DYNAMIC GAMES
COMPLETE
Nash equilibrium Subgame perfect equilibrium
INFORMATION
INCOMPLETE
INFORMATION Bayesian equilibrium Perfect Bayesian equilibrium
PLAYER 2
L R
U 1, 4 2, 3
Normal form game
(bimatrix) 0, 2 0, 3
PLAYER 1 M
D 6, 1 0, 5
PLAYER 1
a b
0
PLAYER 2
Extensive form game 1
2
c e
d
(tree) PLAYER 3
PLAYER 3 4
f g 2 f g
1 0 1 0 3
1 0 0 1
1 2 0 2
2. Games of complete information
EXAMPLE 1 Prisoners’ dilemma
Two suspects are arrested and charged with a crime. The police lack sufficient evidence
to convict the suspects, unless at least one confesses. The police hold the suspects in
separate cells and explain the consequences that will follow from the actions they could
take:
1) If neither confess, then both will be convicted of a minor offense and sentenced to a 1
month jail.
2) If both confess, then both will be sentenced to jail for 6 months
3) If one confesses but the other does not, then the confessor will be released
immediately but the other will be sentenced to 9 months to jail (6 for the crime and a
further 3 for obstructing justice)
Questions:
1) Model the strategic situation into a simple normal form game
2) What is your prediction of what will happen? (solution of the game)
SUSPECT 2
NC C
NC -1, -1 -9, 0
SUSPECT 1
C 0, -9 -6, -6
Strategy NC is dominated by strategy C for suspect 1
Strategy NC is dominated by strategy C for suspect 2
The unique solution of the game is (C,C): each player will choose C
independently of what the other player does. (C,C) is an equilibrium in
dominant strategies
Very often, this method of eliminating dominated strategies has no solution
EXAMPLE 2
PLAYER 2
L R
U 1, 4 2, 3
Only M for player 1
PLAYER 1 M 0, 2 0, 3 can be eliminated,
so then what?
D 6, 1 0, 5
Nash equilibrium in pure and mixed strategies
Best response: In a n-player game, strategy si is player i’s best response to
the strategies specified by the other n-1 players, s-i, if player i cannot
obtain a strictly higher payoff with any other possible strategy, si’, given that
the other players are playing s-i.
Thus, si Є BRi (s-i), if ui (si, s-i ) ≥ ui (si‘, s-i), for all si’ Є Si.
A Nash equilibrium is a strategy profile (s1*,s2*,…sn*) such that for each
player i, si* Є BRi (s-i*)
The notion of Nash equilibrium is weaker than the notion of equilibrium
in dominant strategies.
An equilibrium in dominant strategies is also a Nash equilibrium
(for example, the outcome C,C in the Prisoners’ dilemma is a Nash equilibrium)
However, a Nash equilibrium need not be an equilibrium in dominant
strategies.
EXAMPLE 3 Battle of the sexes
PLAYER 2
(husband)
Ballet Boxing
Ballet 2, 1 0, 0
PLAYER 1
(wife)
Boxing 0, 0 1, 2
This game has two Nash equilibria: (Ballet, Ballet) and (Boxing, Boxing)
EXAMPLE 4 Matching pennies
PLAYER 2
Heads Tails
Heads 1, -1 -1, 1
PLAYER 1
Tails -1, 1 1, -1
This game does not have any Nash equilibrium (in pure strategies)
Mixed strategy: Given a set of strategies for player i, Si = {si1, si2, …, siK} ,
a mixed strategy for player i is a probability distribution over the set
of strategies, pi = {pi1, pi2, …, piK} , where 0 ≤ pik ≤ 1 for k = 1, 2, .., K and
pi1 + pi2+ …+ piK = 1.
Mixed strategy Nash equilibrium is a set of probability distributions
over the sets of strategies of the N players, {p1*, p2*, …, pn*} such that for each
player i, pi* Є BRi (p-i*)
EXAMPLE 5
Compute the mixed-strategy Nash equilbrium in the Matching pennies game
PLAYER 2
Heads Tails
(r) (1-r)
Heads
(q)
1, -1 -1, 1
PLAYER 1
Tails
(1-q)
-1, 1 1, -1
Solution: the mixed-strategy Nash equilibrium is (q* = 1/2, r*=1/2).
Properties of Nash equilibria
A Nash equilibrium always exists (possibly in mixed strategies)
(Almost) all the games have an odd number of Nash equilibria (considering
both pure and mixed-strategy equilibria)
Examples
Prisoners’ dilema: one pure strategy equilibrium
Battle of the sexes: two pure-strategy equilibria + one mixed-strategy equilibria
Matching pennies: one mixed-strategy equilibrium
Sequential games of complete information
The main difference is that now players who move later on the game can observe
how other players have played up to that moment.
Example: sequential battle of the sexes
Ballet 2, 1
Decision node
Ballet 2 Boxing
(husband)
0, 0
1
(wife)
Ballet 0, 0
Boxing
Boxing
2
(husband) 1, 2
We now use the concept of Subgame
Perfect Equilibrium
Subgame Part of the extensive form beginning with a decision node
and including everything that branches out to the right of it
(the above example contains three subgames)
Backward induction Find the Nash equilbria on the subgames
by working backward from the end of the
game to the beginning
Ballet 2, 1
Equilibrium path
Ballet 2 Boxing
(husband)
0, 0
1
(wife)
Ballet 0, 0
Boxing
Boxing
2
(husband) 1, 2
Only one out of the three Nash equilibria is a subgame perfect equilibrium.
Subgame perfection is a refinement of Nash equilibrium.
EXPERIMENT II
(divide and choose)
Repeated games
It consists of playing the same game over and over again (for example,
playing the Prisoner’s dilemma several times)
Players Players Players
play play play
prisoners’ prisoners’ prisoners’
dilemma dilemma dilemma
STAGE 1 STAGE 2 STAGE T
Can repeated play of the same game open the possibility of cooperation?
Finitely repeated games
Repeating a game for a finite number of periods does not increase the
possibility for cooperation
Example: think of the prisoners’ dilema played for T periods and solve the game
by backward induction to find the Subgame Perfect Equilibrium.
Result (Selten): For any stage game with a unique Nash equilbrium, the unique
Subgame Perfect equilibrium of the finitely repeated game involves
playing the Nash equilbrium every period.
Infinitely repeated games
PLAYER 2
A B
A 1, 1 3, 0
(A,A) is the unique Nash
PLAYER 1
equilibrium of the stage game
B 0, 3 2, 2
Can agents sustain the ccoperative outcome (B,B) if the game is played an
infinite number of periods?
Trigger strategy: For example, continue to cooperate as long as all have cooperated
up to that point, but revert to playing the Nash equilibrium if anyone deviates from
cooperation
Use the following trigger strategy: select B in period t+1 if you have observed your
opponent choosing B up to period t. If you observe your opponent deviating,
then choose A for ever. Your opponent does also the same.
Can this strategy sustain the ccoperative outcome (B,B)?
Since we have to sum an infinite number of payoffs (which lead to infinite),
it is convenient to use a discount factor (0 < δ < 1)
If I choose to cooperate for ever, the presented discounted value of this strategy is:
V (c) = 2 / (1 - δ)
If I choose to deviate, the presented discounted value of this strategy is:
V (nc) = 3 + δ / (1 - δ)
The trigger strategies form a subgame-perfect equilibrium if V(c) ≥ V (nc),
that is, if δ ≥1/2
Examples of trigger strategies: grim strategies, tit-for-tat strategies …
3. Games of incomplete information
In the games studied thus far, players know everything there is to be known
about the set up of the game, including each others’ strategy sets and payoffs.
This section studies the tools neded to analyze games in which not all players
know everything (this adds realism to the games but also many mathematical
complications)
Probability theory provides Bayes’ rule for making inferences about hidden
information. The relevance of this rule in this games has lead them to be called
Bayesian games
EXAMPLE 6
Consider the following simultaneous game of incomplete information:
PLAYER 2
L R
U t, 2 0, 0
PLAYER 1
D 2, 0 2, 4
Player 1 can be one of two possible types with equal probability: t = {0, 6}
Player 1 knows her own type. Player 2 only knows the probability distribution
Sequential version of the game:
6, 2
L
2
U 0, 0
R
1
t=6 L 2, 0
Chance node
p = 1/2 2
D
2, 4
R
Decision node
t=0 L 0, 2
p = 1/2 2
1 U
R 0, 0
L 2, 0
D 2
R 2, 4
The unique BN equilibrium of this game is (D/t=6, D/t=0, R).
A useful way to find the equilbrium is to write the strategies contingent on the types
in a bi-matrix, compute expected utility for player 2 and then find the Nash equilibrium
PLAYER 2
L R
U(T=6), U (t=0) 6, 0, 2 0, 0, 0
U(T=6),D (t=0) 6, 2, 1 0, 2, 2
PLAYER 1
D(T=6),U (t=0) 2, 0, 1 2, 0, 2
D(T=6),D (t=0) 2, 2, 0 2, 2, 4
A Bayesian - Nash equilibrium. In a 2-player game in which player 1 has
private information, a Bayesian-Nash game is a strategy profile (s1*(t), s2*)
such that s1*(t) is a best response for s2* for each type t Є T of player 1,
u1 (s1*(t), s2*, t) ≥ u1 (s1’, s2*, t) for all s1’ Є S1,
and such that s2* is a best response to s1*(t) given player 2’s beliefs p(tk) about
player 1’s types:
∑ p(tk) u2 (s2*, s1*(tk), tk) ≥ ∑ p(tk) u2 (s2’, s1*(tk), tk) for all s2’ Є S2,
Sequential games of incomplete information
We now move to games in which player 1 (the informed player) takes an action
that is observed by player 2 afterwards. Player 1’s action provides information
to player 2. This player can use this information (or signal) to update her beliefs
about player 1’s type (whenever possible)
We will use BAYES’ RULE to update beliefs
p(th) ∙ p(action a / th )
p (th │action a) =
∑ p(ti) ∙ p(action a / ti)
6, 2
L
2
U q 0, 0
R
1
L 0, 2
2
t=6
p = 1/2 1-q
D R 0, 0
L 2, 0
t=0 U 2
1-p = 1/2
1
r 2, 4
R
L 2, 0
D 2
1-r
R 2, 4
The unique PB equilibrium of this game is the following: SEPARATING
(i) When t = 6, (U, L, q=1) EQUILIBRIUM
(ii) When t = 0, (D, R, r = 1/2)
Consider a slight variation of the extensive form game presented before:
6, 2
L In this case, player 2 is not
2 able to update beliefs once
U q he/she observes the action of
R 0, 0
1 player 1.
L 3, 2
2
t=6
p = 1/2 1-q
D R 0, 0 In this case, the unique PB
L 2, 0 equilbrium is POOLING
t=3 U 2
1-p = 1/2 (i) When t = 6, (U, L, q=1/2)
1
r 2, 4 (ii) When t = 3, (U, L, r=1/2)
R
L 2, 0
D 2
1-r
R 2, 4
