Repeated Games

Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

Repeated Games

Abhinash Borah

March 28, 2023


Key questions

Cooperation is vital for many social and economic relationships


What sustains cooperation amongst self-interested agents?
Long-lived relationships may be one answer to this question

Repeated Games 2 / 22
Repeated Games

A repeated game is an extensive game in which players face the


same “stage game” or “constituent game” in every period
t = 1, 2, . . . , T .
The number of periods T may be finite or infinite
We assume that a player’s overall payoff in the repeated game is a
discounted sum of her payoffs in each stage

Repeated Games 3 / 22
Repeated Games with Observable Actions

A repeated game with observable actions or perfect


monitoring is one in which players’ stage game actions in any
period are publicly observed before play of the stage game in the
next period
Critical Insight: Since at each stage players’ past actions are
observable, it becomes possible for them to condition their play on
past play. This may lead to equilibrium outcomes that do not
arise when the game is played only once
Other monitoring structures
Public monitoring
Private monitoring

Repeated Games 4 / 22
Example—Repeated Game

Suppose the following stage game is repeated, say, a finite number of


times
L M R

U 0, 0 3, 4 6, 0

M 4, 3 0, 0 0, 0

D 0, 6 0, 0 5, 5

Repeated Games 5 / 22
Strategies and Solution Concept
A period t history in a repeated game is a specification of all the
action profiles that have been played in the previous t − 1
iterations of the stage game. We will denote a period t history by
ht
There is one history in period 1. We refer to it as the null history
as no play has taken place thus far. We often denote it by φ
Each history of a repeated game with observable actions or perfect
monitoring can be identified with an information set. Why?
What is a strategy?

Repeated Games 6 / 22
Strategies and Solution Concept
A period t history in a repeated game is a specification of all the
action profiles that have been played in the previous t − 1
iterations of the stage game. We will denote a period t history by
ht
There is one history in period 1. We refer to it as the null history
as no play has taken place thus far. We often denote it by φ
Each history of a repeated game with observable actions or perfect
monitoring can be identified with an information set. Why?
What is a strategy?
A strategy for player i is a mapping from the set of possible
histories to the set of actions of player i in the stage game
Solution concept?

Repeated Games 6 / 22
Strategies and Solution Concept
A period t history in a repeated game is a specification of all the
action profiles that have been played in the previous t − 1
iterations of the stage game. We will denote a period t history by
ht
There is one history in period 1. We refer to it as the null history
as no play has taken place thus far. We often denote it by φ
Each history of a repeated game with observable actions or perfect
monitoring can be identified with an information set. Why?
What is a strategy?
A strategy for player i is a mapping from the set of possible
histories to the set of actions of player i in the stage game
Solution concept?
Subgame perfection
Repeated Games 6 / 22
Example—Repeated Game

Suppose the following stage game is repeated twice:

L M R

U 0, 0 3, 4 6, 0

M 4, 3 0, 0 0, 0

D 0, 6 0, 0 5, 5

What are the set of histories?

Repeated Games 7 / 22
Suppose the following stage game is repeated twice:
L M R

U 0, 0 3, 4 6, 0

M 4, 3 0, 0 0, 0

D 0, 6 0, 0 5, 5
Consider the following strategy profile s = (s1 , s2 ):

s1 (∅) = D
(
M if h2 = (D, R)
s1 (h2 ) =
U if h2 6= (D, R)
s2 (∅) = R
(
L if h2 = (D, R)
s2 (h2 ) =
M if h2 6= (D, R)
Repeated Games 8 / 22
Example—Strategies and Payoffs
Suppose the following stage game is repeated twice:
L M R

U 0, 0 3, 4 6, 0

M 4, 3 0, 0 0, 0

D 0, 6 0, 0 5, 5

The outcome induced by the strategy profile s = (s1 , s2 ) is

Repeated Games 9 / 22
Example—Strategies and Payoffs
Suppose the following stage game is repeated twice:
L M R

U 0, 0 3, 4 6, 0

M 4, 3 0, 0 0, 0

D 0, 6 0, 0 5, 5

The outcome induced by the strategy profile s = (s1 , s2 ) is (D, R)


in pd. 1 and (M, L) in pd. 2
Then, players’ payoffs are:
U1 (s1 , s2 ) = 5 + 4δ1 and U2 (s1 , s2 ) = 5 + 3δ2

Repeated Games 9 / 22
Example: SPNE

L M R

U 0, 0 3, 4 6, 0

M 4, 3 0, 0 0, 0

D 0, 6 0, 0 5, 5

Three Nash Equilibria

Repeated Games 10 / 22
Example: SPNE

L M R

U 0, 0 3, 4 6, 0

M 4, 3 0, 0 0, 0

D 0, 6 0, 0 5, 5

Three Nash Equilibria


Two pure strategy Nash equilibria, (M, L), (U, M ), and one mixed
strategy Nash equilibrium, (( 37 U, 47 M ), ( 73 L, 47 M )), with payoffs
(4, 3), (3, 4) and ( 12 12
7 , 7 ), respectively
(D, R) is not a Nash equilibrium of the stage game
Repeated Games 10 / 22
7 7
If the stage game specified above is repeated twice, then for δ1 ≥ 16
and δ2 ≥ 9
the
following strategy profile is a SPNE:

s∗1 (∅) = D
(
M if h2 = (D, R)
s∗1 (h2 ) =
( 73 U, 47 M ) if h2 6= (D, R)
s∗2 (∅) = R
(
L if h2 = (D, R)
s∗2 (h2 ) =
( 73 L, 47 M ) if h2 6= (D, R)

If h2 = (D, R), then (s∗1 (h2 ), s∗2 (h2 )) = (M, L) is a Nash eq. of the
stage game.

If h2 6= (D, R), then (s∗1 (h2 ), s∗2 (h2 )) = (( 73 U, 74 M ), ( 37 L, 74 M )) is a


Nash eq. of the stage game.

Therefore, for the subgame starting at any history h2 in pd. 2, the


strategy profile (s∗1 , s∗2 ) restricted to that subgame is a Nash eq.
Repeated Games 11 / 22
7 7
If the stage game specified above is repeated twice, then for δ1 ≥ 16
and δ2 ≥ 9
the
following strategy profile is a SPNE:

s∗1 (∅) = D
(
M if h2 = (D, R)
s∗1 (h2 ) =
( 73 U, 47 M ) if h2 6= (D, R)
s∗2 (∅) = R
(
L if h2 = (D, R)
s∗2 (h2 ) =
( 73 L, 47 M ) if h2 6= (D, R)

Consider player 1’s choice at the initial history ∅ holding player 2’s
strategy fixed at s∗2 . If she follows her strategy s∗1 and plays D, her
payoff (in the overall game) is 5+4δ1 . On the other hand, if she
deviates from D and plays U , her most profitable deviation against s∗2 ,
then her payoff is 6+ 12
7 δ1 . So, no profitable deviation exists for 1 if:
12 7
5 + 4δ1 ≥ 6 + 7 δ1 ⇔ δ1 ≥ 16
Repeated Games 12 / 22
7 7
If the stage game specified above is repeated twice, then for δ1 ≥ 16
and δ2 ≥ 9
the
following strategy profile is a SPNE:

s∗1 (∅) = D
(
M if h2 = (D, R)
s∗1 (h2 ) =
( 73 U, 47 M ) if h2 6= (D, R)
s∗2 (∅) = R
(
L if h2 = (D, R)
s∗2 (h2 ) =
( 73 L, 47 M ) if h2 6= (D, R)

Next, consider player 2’s choice at the initial history ∅ holding player
1’s strategy fixed at s∗1 . If she follows her strategy s∗2 and plays R, her
payoff (in the overall game) is 5+3δ2 . On the other hand, if she
deviates from R and plays L, her most profitable deviation against s∗1 ,
then her payoff is 6+ 12
7 δ2 . So, no profitable deviation exists for 2 if:
12 7
5 + 3δ2 ≥ 6 + 7 δ2 ⇔ δ2 ≥ 9
Repeated Games 13 / 22
Example: Finitely Repeated Prisoners’ Dilemma

Consider the finitely repeated game with 2 repetitions of the prisoner’s


dilemma

Don’t Confess (N) Confess (Y)

Don’t Confess (N) 3, 3 0, 4

Confess (Y) 4, 0 1, 1

What is the set of SPNE for this game?

Repeated Games 14 / 22
Some Results

Proposition
In any repeated game, whether finite or infinite, playing a stage game
Nash equilibrium in every period can be supported as a SPNE outcome
of that repeated game.

Repeated Games 15 / 22
Some Results

Proposition
In any repeated game, whether finite or infinite, playing a stage game
Nash equilibrium in every period can be supported as a SPNE outcome
of that repeated game.

Proposition
If the stage game of a finitely repeated game has a unique Nash
equilibrium, then the finitely repeated game has a unique SPNE.

Repeated Games 15 / 22
Repeated games: An important result

Proposition (One-deviation property)


A strategy profile in a repeated game is an SPNE iff no player can
increase her payoff by changing her action at the start of any subgame,
given the other player’s strategies and the rest of her own strategy

Repeated Games 16 / 22
Example: Infinitely Repeated Prisoner’s Dilemma
Consider the infinitely repeated game Γ∞ involving an infinite
repetition of the prisoners’ dilemma:

Don’t Confess (N) Confess (Y)

Don’t Confess (N) 3, 3 0, 4

Confess (Y) 4, 0 1, 1

Notation: We can partition the set of histories of Γ∞ into the following


two set:
1 HN consists of the initial history and all those histories in which
both players have played N in all previous plays of the game
2 HY consists of all other histories
Repeated Games 17 / 22
Example: Infinitely Repeated Prisoner’s Dilemma

We assume here that both players have the same discount factor of δ

For δ ≥ 31 , (s∗1 , s∗2 ) below is a SPNE of this repeated game:


(
N if h ∈ HN
s∗1 (h) =
Y if h ∈ HY
(
N if h ∈ HN
s∗2 (h) =
Y if h ∈ HY

Strategies of these type are referred to as grim-trigger strategies

Repeated Games 18 / 22
Example: Infinitely Repeated Prisoner’s Dilemma

1
For δ ≥ 3
, (s∗1 , s∗2 ) below is a SPNE of Γ∞ :
(
∗ N if h ∈ HN
s1 (h) =
Y if h ∈ HY
(
N if h ∈ HN
s∗2 (h) =
Y if h ∈ HY

Consider any h ∈ HY . Given that player j is playing Y in all histories


of the subgame following h under the strategy s∗j , player i 6= j cannot
profit by playing N instead of Y .

Repeated Games 19 / 22
Example: Infinitely Repeated Prisoner’s Dilemma
1
For δ ≥ 3
, (s∗1 , s∗2 ) below is a SPNE of Γ∞ :
(
∗ N if h ∈ HN
s1 (h) =
Y if h ∈ HY
(
N if h ∈ HN
s∗2 (h) =
Y if h ∈ HY

Next, consider any h ∈ HN . Holding j’s strategy fixed at s∗j , consider


the one-shot deviation of player i 6= j from s∗i at the history h involving
playing Y instead of s∗i (h) = N . Such a deviation is not profitable if:

3 + 3δ + 3δ2 + ... ≥ 4 + δ + δ2 + ...


3 δ
≥ 4+
1−δ 1−δ
1
δ ≥
3
Repeated Games 20 / 22
Example: Dynamic Oligopoly and Collusion

Consider an infinitely repeated Cournot Oligopoly game, where


market demand in each period is given by P = 1 − Q and both
firms produce at zero cost.
Is it possible for the firms to “collude” in equilibrium and restrict
output in the market to the monopoly level?

Repeated Games 21 / 22
Axelrod’s tournaments and tit-for-tat strategy

Consider the infinitely repeated prisoner’s dilemma. In this game,


tit-for-tat refers to the strategy of doing whatever the other player did
in the previous period. That is, for i = 1, 2 and j 6= i,
(
N if ht = ∅ or ajt−1 = N
si (ht ) =
Y if ajt−1 = Y

Repeated Games 22 / 22

You might also like