Industrial Organization1 : Game Theory

Andrey Tkachenko

ECON 415,515
Nazarbayev University

1
I am grateful to Chiara Fumagalli (Bocconi University) for the materials
Outline

1 Normal Form Games

2 Extensive Form Games

3 Repeated Games

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 2 / 52

Normal Form Games

A game in normal form is a list Γ = ({1, 2}, S1 , S2 , u1 , u2 ) with:

A set of players {1, 2}. Given a player i, we will write −i to denote the other player.
For each player i, a set of strategies Si . We define a set of strategy profiles as
S := S1 × S2 .
For each player i, a payoff function ui : S → R.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 3 / 52

 Example
{1, 2}
S1 ={U,M,D}
S2 ={L,R}
S = S1 × S2 = {(U, L), (U, R), (M, L), (M, R), (D, L), (D, R)}
u1 = {u1 (U, L) = 5, u1 (U, R) = 4, u1 (M, L) = 6, u1 (M, R) = 3, u1 (D, L) = 6, u1 (D, R) = 4}
u2 = {u2 (U, L) = 1, u2 (U, R) = 0, u2 (M, L) = 0, u2 (M, R) = 1, u2 (D, L) = 4, u2 (D, R) = 4}

L R L R

U 1,-1 -1,1 U 5,1 4,0

M -1,1 1,-1 M 6,0 3,1

D -2,5 -3,2 D 6,4 4,4

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 4 / 52

Games in Normal Form

Finite game in normal form: set of strategies of each player is finite.

Mixed strategy of player i: probability distribution over Si .
Cristiano Ronaldo (CR7) scoring a penalty. SCR7 = {L, C , R}
σCR7 = {σCR7 (L) = 13 , σCR7 (C ) = 13 , σCR7 (R) = 31 }
Pure strategy of player i: degenerate mixed strategy of player i (mixed strategy that
assigns probability one to a single element of Si ).
Σi = ∆(Si ): set of all mixed strategies of player i.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 5 / 52

Best Response

A basic assumption of game theory is that each player maximizes her payoff, given her belief
about what the other player does.
A strategy si ∈ Si is a best response (or best reply) to σ−i if ui (si , σ−i ) ≥ ui (si0 , σ−i )
for all si0 ∈ Si .

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 6 / 52

Dominated Strategies

In some cases, a strategy of i does strictly better than another, regardless of what player −i
does.
A strategy si ∈ Si is strictly dominated if there exists σi ∈ Σi such that
ui (σi , s−i ) > ui (si , s−i ) for all s−i ∈ S−i .
In this case, we say σi strictly dominates si .
A strategy si ∈ Si is weakly dominated if there exists σi ∈ Σi such that
ui (σi , s−i ) ≥ ui (si , s−i ) for all s−i ∈ S−i , with strict inequality for some s−i .

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 7 / 52

Example of Dominated Strategies

L R L R

U 1,-1 -1,1 U 5,1 4,0

M -1,1 1,-1 M 6,0 3,1

D -2,5 -3,2 D 6,4 4,4

Game on the left: Strategy D is strictly dominated.

Game on the right: Strategies U and M are weakly dominated.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 8 / 52

Pure Strategy dominated by a mixed strategy

L R

U 10,1 0,4

M 4,2 4,3

D 0,5 10,2

U is an excellent strategy for player 1 when the other player plays L, but it is a poor one against R.
Strategy D is excellent against R but poor against L.
Strategy M is good but not a great strategy against both L and R.
None of these pure strategies is strictly dominated by any of the others.
But if we allow player 1 to randomize, then playing U and D each with probability 1/2, M will be strictly
dominated.
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 9 / 52
Iterated Dominance/Rationalizability

A rational player i cannot choose a strictly dominated strategy, nor can she believe player −i
would, nor can she believe player −i believes that i would, and so on. How do we formalize
this idea?
Starting from the game Γ, let us recursively delete all strictly dominated strategies of all
players.
The strategies in Γ that survive iterated elimination of strictly dominated strategies are
said to be rationalizable.
If every player has a unique rationalizable strategy, then we say the game Γ is dominance
solvable.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 10 / 52

Example 1

B1 B2 B3 B4

A1 0,7 2,5 7,0 0,1

A2 5,2 3,3 5,2 0,1

A3 7,0 2,5 0,7 0,1

A4 0,0 0,-2 0,0 10,-1

B4 is strictly dominated by σ2 = {σ2 (B1) = 1/2; σ2 (B3) = 1/2}.

Once B4 is eliminated, strategy A4 can be eliminated because it is strictly dominated by
A2.
No further strategies can be ruled out.
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 11 / 52
Example 2
Example 1

Consider the following game:

L R
T 1; 1 2; 0
M 5; 1 0; 2
B 0; 0 3; 1

Strategydominated
Strategy T is strictly T is strictlybydominated
the mixedby the mixed
strategy strategy
σ1 such that1σsuch
1 (M)that 1 .M
= 1/4 / D 1=4
and
σ1 (B) = 3/4. this gives .1=4/5 C .3=4/0 D 5=4 > 1 if player 2 chooses L, and .1=4/0 C .3=
No strategy of chooses
player 2R. Obviously,
is strictly no strategy of player 2 is strictly dominated. After elimin
dominated.
the game

L R
M 5; 1 0; 2
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 12 / 52
 B 0; 0 3; 1
Example 2
Strategy T is strictly dominated by the mixed strategy 1 such that 1 .M / D 1=4
this gives .1=4/5 C .3=4/0 D 5=4 > 1 if player 2 chooses L, and .1=4/0 C .3=
chooses R. Obviously, no strategy of player 2 is strictly dominated. After elimin
After elimination ofthe
T ,game
we are left with the game:

L R
M 5; 1 0; 2
B 0; 0 3; 1

In this game RInstrictly

this game R strictly
dominates L, dominates
whereas no whereas of
L,strategy noplayer
strategy
1 isofdominated.
player 1 is dominated
are left with the game

R
M 0; 2
B 3; 1
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 13 / 52
 M 5; 1 0; 2
Example 2 B 0; 0 3; 1

In this game R strictly dominates L, whereas no strategy of player 1 is domina

After elimination of L, are
we left
are with the game
left with the game:

R
M 0; 2
B 3; 1

In thisdominates
In this game B strictly game B strictly dominates
M. Thus, the only . Thus, the only
Mrationalizable rationalizable
strategies strategie
are B for
player 1 and R forplayer
player2.2.

For other examples see PS 1.

Example 2

Each of two people announces a number in the set f1; 2; 3; 4g. A prize of 2
Andrey Tkachenko announces the lower number.
Game Theory If they announce the415,515
ECON sameNazarbayev
number, the prize
University is
14 / 52
Discussion

All one needs to get the rationalizable outcomes is the assumption of rationality and
common belief in rationality.
However, players may have many rationalizable strategies, as in Example 1.
Rationalizability does not give sharp predictions on how rational players should play and
how much they should expect to get.
If we want to narrow our predictions further, we need to make additional assumptions
beyond common knowledge of rationality.
The solution concept we will study in the remainder of the class does so imposing
equilibrium requirements on players’ strategy choices.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 15 / 52

The Nash Equilibrium
A Nash equilibrium is a strategy profile s ∈ S such that ui (s) ≥ ui (si , s −i ) for every player i
and every si ∈ Si .

A Nash equilibrium is a strategy profile such that, once each player knows every other
player is following the strategies specified in the profile, then she finds it optimal to do so
herself.
In a Nash equilibrium, each player’s strategy choice is a best response to the strategies
actually played by the other players, i.e. no one has incentives to deviate.
It is easy to see that a Nash equilibrium is a rationalizable strategy.
If s is a Nash equilibrium, then s i survives the first round of elimination of dominated
strategies, because it is a best response to the belief that −i will play s −i with probability
one. Since this is true for every player, it remains true at all subsequent rounds as well.
The Nash equilibrium requires more than rationality and common belief in rationality.
At the Nash equilibrium each player’s strategy is a best response to what she expects the
other players will be playing and players’ expectations must be correct.
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 16 / 52
Example

What is the Nash Equilibrium of this game? (Other examples in PS1.)

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 17 / 52

Example

B1 B2 B3 B4

A1 0,7 2,5 7,0 0,1

A2 5,2 3,3 5,2 0,1

A3 7,0 2,5 0,7 0,1

A4 0,0 0,-2 0,0 10,-1

In this case the Nash Equilibrium concept gives a sharper prediction than rationalizability.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 18 / 52

Multiple Nash Equilibria

The Nash equilibrium concept does not always yields a unique prediction.

Windows Linux

Windows 100,100 0,0

Linux 0,0 100,100

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 19 / 52

 A Nash Equilibrium in pure strategies may fail to exist
xistence of Nash Equilibria

Nash equilibrium may not exist, e.g. in the game matching pennies:

H T
H 1; 1 1; 1
T 1; 1 1; 1

ut if we allow mixed strategies, then it always exists (in finite games). The mixed exte
e game where the set of players is again N , the set of strategies of player i is †i D .S
ayer i from the strategy profile  2 † is ui ./.
The following is the result originally proved by Nash in 1951.
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 20 / 52
 Existence of Nash Equilibria in finite games
But if we allow for mixed strategies, the Nash equilibrium always exists in finite games.
The mixed extension of a game Γ = ({1, 2, .., N}, S1 , S2 , ..SN , u1 , u2 , ...u2 ) is the game
where the set of players is again N, the set of strategies of player i is Σi = ∆(Si ), and the
payoff to player i from the strategy profile σ ∈ Σ is ui (σ).

xistence of Nash
In 1951 Equilibria
John Nash proved that the mixed extension of any finite game has a Nash
equilibrium. (See the Notes for the formal proof.)
Nash equilibrium may not exist, e.g. in the game matching pennies:
Find Mixed Nash Equilibria of the game.
H T
H 1; 1 1; 1
T 1; 1 1; 1

ut if we allow mixed strategies, then it always

Andrey Tkachenko Game Theory
exists (in finite games). The mixed21exte
ECON 415,515 Nazarbayev University / 52
Existence of Nash Equilibria in nice games

A nice game is a game where the strategy set of each player is a closed interval
Si = [c i , c i ] ∈ R, and the payoff function of each player is continuous and strictly
quasi-concave2 in the player’s own strategy.
It is possible to to prove that a nice game always has a Nash equilibrium in pure
strategies (see Notes).

Example of a nice game: Cournot competition (to be discussed).

2
f (.) is strictly quasi-concave if ∀x, y ∈ Si and ∀λ ∈ (0, 1) we have f (λx + (1 − λ)y ) > min(f (x), f (y )).
Equivalently, f (.) is strictly quasi-concave if ∀β ∈ R the set {x ∈ Si |f (x) > β} is convex.
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 22 / 52
Outline

1 Normal Form Games

2 Extensive Form Games

3 Repeated Games

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 23 / 52

Extensive Form Games

A game in normal form specifies what the players’ strategies are, and what each player’s
payoff is, as a function of the possible profiles of strategies chosen by all players.
A game in extensive form is a more detailed description of a strategic situation, and
allows to model dynamic choice. An extensive form specifies:
the physical order of play (who moves when, and what choices are available),
the information available to a player when making a choice,
and the payoff to each player as a function of the moves selected by all players.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 24 / 52

A or B. The fact
Example 1 that 2 does not know whether 1 chose L or R is reflected by the fact that the tw
uccessors
Player 1of the initial
chooses L or node
R andare in the
then same
player information
2, without set,what
knowing denoted by a dotted
the choice line.1 has
of player Note tha
been,
t the twochooses A or B.2 has the same actions available.
nodes player
1
L R
2
A B A B

1; 0 2; 3 0; 1 "1; 0
Figure 1: An extensive form game.
Note that player 2 does not observe what has happened before it is her turn to move and does not know
whether 1 chose L or R.
HereTheis another
fact that 2example:
does not knowPlayer 1 chooses
whether L Roris R
1 chose L or and then
reflected by theplayer
fact that2,the
having observed
two successors of th
the initial node are in the same information set, denoted by a dotted line.
hoice ofA player
game like1,this
chooses
is calledAa or if heimperfect
B;with
game choosesinformation.
B, then it is 1’s turn again, and he has to choos
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 25 / 52
 Figure 1: An extensive form game.
Example 2
Here is another example: Player 1 chooses L or R and then player 2, having observed the
iceHere is an example
of player of A
1, chooses a game
or B;with
if heperfect
choosesinformation,
B, then it is where theagain,
1’s turn information
and hesets
has are
to choose
simply the decision nodes.
ween l and r.
1
L R
2 2
A B C D
1
1; 0 2; 3 0; 1 l r

0; 1 "1; 0
Figure 2: An extensive form game where player 1 moves twice.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 26 / 52

Strategies

A strategy of a player in a dynamic game is a function that maps each information set of
the player into one of the actions available at that information set.
Example 1: S1 = {L, R}, S2 = {A, B}
Example 2: S1 = {L`, Lr , R`, Rr }, S2 = {AC , AD, BC , BD}

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 27 / 52

 if z 2 Z x .
Unreasonable Nash Equilibria: Example 3
3.3. Subgame Perfect Equilibrium
The notion of Nash
The notion equilibrium
of Nash veryvery
equilibrium often leads
often totounreasonable
leads predictionsin in
unreasonable predictions extensive
extensive formform
games.
games. Consider the following extensive form game and its associated game in normal form.
1
L R AC AD BC BD
L 0; 0 0; 0 0; "1 0; "1
A B C D R 0; 0 1; 1 0; 0 1; 1

0; 0 0; "1 0; 0 1; 1

The profile .L; AC / is a Nash equilibrium, yet it is unreasonable. Indeed, player 2 is supposed
There exists three Nash Equilibria: (L, AC), (R,AD), (R,BD).
to choose
Both if player
(L,AC)C and 1 chooses
(R,BD) R (which he does
are unreasonable. not), whereas
Player wouldplay
2 wouldDnever give C
more.
afterSimilarly,
R and
is a Nash equilibrium,
would never play B after L.
.R; BD/ but it is unreasonable because it requires player 2 to play B after
L, whereas A would give more.
The idea of subgame perfect equilibrium
Andrey Tkachenko Game(Selten,
Theory 1965) is that behavior in parts
ECON 415,515 of the
Nazarbayev game
University 28 / 52
the competitor. Fighting hurts not only the competitor but also the incumbent himself (both get
Unreasonable Nash Equilibria: Example 4
1). Behaving in an accommodating way yields both players a payo↵ of 1.
InBelow
a market there is both
we provide a single firm (M).
the game A potential
tree for competitor
the extensive (E) must decide
form representation whether
and the normalenter
(e) or not (ne) in the market. If E enters, the monopolist must decide whether to
form representation of the game.
accommodate (a) or fight (f ).

Eb
@@
ne @Re
@
@
@
r @@Mr
0, 2 @@
f @Ra
@
@
@
r @r
@
1, 1 1,1

Figure 4: Game tree of the market entry game

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 29 / 52
 @
4r @r
@
Unreasonable Nash Equilibria: Example
1, 1 1,1

Figure 4: Game tree of the market entry game

The The associated
associated game
game in normal
in normal form
form is:is

f f a a
Last modified at 11:57 on ne
January
0; 219,0,2009
ne 2 0; 2 0, 2
e e 1 1, 11; 1 1, 1
1;

[Link]; underlining
fare also
/ and of
.e;in arepayo↵s
a/the subgame, in the
Nash equilibria, 2 Xy
[Link]
x but form
theimplies Hy .x/theDplayers’
firstindicates
is unreasonable, H.x/ !
becausebestyresponses. From
X; it prescribes f a
Both (ne, f ) best
these and (e, a) are Nash
responses we canequilibria,
see that butgame
the the first
admitsis unreasonable,
two NE: (ne, because
f ) and (e,ita).
prescribes
In the game
is irrational at the node following e.
f after e, which" u
tree the
is
ytwo y
irrational
i W equilibria
Z ! R isare at the node
therepresented following
restrictionbyofthe e. y
to Z for
uicolored each(red
arrows player i . f ) and blue for (e, a)).
for (ne,
The idea of subgame perfect equilibrium is to rule out these “not credible threats” (or p
requiring In that
the ‘red equilibrium’,
behavior in parts of (ne,
thef ),game
player
thatMcanadopts a strategy
be regarded as that
games specifies that he would
in themselves shoul
Nash fight Obviously,
equilibria every
the newofcompetitor
them. A extensive
should he
subgame form
[Link]
perfect Givenisthis
equilibrium a subgame
threat of aoffight,
is a strategy itself.
it isThe
profile thengame
such in Figure
optimal
that, for for thes
every
continuation
player
subgame strategy
E not profile
oftoitself.
enter in thefrom
The firstthe
game ininitial
place. Thenode
Figure fact of
thattheEsubgame
2 instead stays out isinamore
has three Nash
turn equilibrium
implies
subgames that M of the sub
never
besides has
its
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 30 / 52
Subgame Perfect Nash Equilibria

The idea of subgame perfection is to rule out these “not credible threats” (or promises) by
requiring that the behavior prescribed by an equilibrium strategy is a Nash equilibrium in every
part of the game that can be regarded as a game in itself (subgame).

DEFINITION:
A subgame perfect Nash equilibrium is a strategy profile such that, for every subgame, the
continuation strategy profile from the initial node of the subgame is a Nash equilibrium of the
subgame.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 31 / 52

 are also in the subgame, i.e. x 2 Xy implies H
y .x/ D H.x/ ! X;
y
Definition of subgames
y ! R is the restriction of ui to Z
yi W Z
" u y for each player i .

A subgame is a subset
Obviously, of aextensive
every game starting withisan
form game information
a subgame set that
of itself. Thecontains a single1 is the onl
game in Figure
decision node.
subgame of itself. The game in Figure 2 instead has three more subgames besides itself, depicte
Every extensive form game is a subgame of itself.
IninExample
Figure 3.1, the entire game above is the only subgame of itself.
In Example 2, the game has three more subgames besides itself:
2
2 1 C D
A B l r 1
0; 1 l r
1; 0 2; 3 0; 1 #1; 0
0; 1 #1; 0
Figure 3: The subgames (other than itself) of the game in Figure 2

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 32 / 52

Backward Induction

In finite extensive games with perfect information subgame perfect Nash Equilibria can be
found using the backward induction method:
solving first for optimal behavior at the “end” of the game,
then determining what optimal behavior is earlier in the game given the anticipation of
this later behavior.

In PS1 you will have to find the SPNE in Example 2, 3 and 4.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 33 / 52

Zermelo’s Theorem

Every finite game of perfect information has at least one pure strategy SPNE (that can be
derived through backward induction).

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 34 / 52

 choosing R) would deliver a payo↵ of 0. The backward induction method therefore delivers
the solution (LL0 , ll0 ). And the payo↵s that the player earn by choosing these strategies are 1
The centipede game
for P1 and 0 for P2.

1b
@
L @R
@
r @2r
@
1, 0 @
l @r
@
r @1r
@
0, 3 @
L0 @ R0
@
r @@2r
3, 2 @
l0 @ r0
@
r @@r
2, 5 5, 4

The only SPNE is (LL0Figure

, ``0 ).3: Solving a game by backward induction
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 35 / 52
 Example 5
ons leading to the highest payoffs for the player whose turn it is, using possibly different colors
ordingly.
1
L R
2 2
A B C D
1
1; 0 2; 3 3; 1 l r

0; 1 !1; 0
Figure 5: A variant of the game in Figure 2.

1
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 36 / 52
Example 5 0; 1 !1; 0
Figure 5: A variant of the game in Figure 2.

1
L R
2 2
A B C D
1
1; 0 2; 3 3; 1 l r

0; 1 !1; 0
Figure 6: Backward Induction in the game of Figure 5: Step I.

1
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 37 / 52
 Example 5 0; 1 !1; 0
Figure 6: Backward Induction in the game of Figure 5: Step I.

1
L R
2 2
A B C D
1
1; 0 2; 3 3; 1 l r

0; 1 !1; 0
Figure 7: Backward Induction in the game of Figure 5: Step II.

This allows us to find all pure strategy subgame perfect equilibria of the game. Indeed, a strat-
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 38 / 52
 Example 5
t modified at 11:57 on January 19, 2009

1
L R
2 2
A B C D
1
1; 0 2; 3 3; 1 l r

0; 1 !1; 0
Figure 8: Backward Induction in the game of Figure 5: Step III.
There are then two subgame perfect equilibria, (L`, BD) and (R`, BC ).

game perfect. Finally, every .!1 ; !2 / such that !2 .BC / D 2=3 and !2 .BD/ D 1 ! !2 .BC / is
Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 39 / 52
Outline

1 Normal Form Games

2 Extensive Form Games

3 Repeated Games

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 40 / 52

Repeated Games

A repeated game consists in the repetition of a normal form game ({1, 2}, A1 , A2 , u1 , u2 ),
called the stage game.
We write Ai instead of Si to denote the stage game strategy set of player i.
This is because we reserve the symbol Si to denote i’s set of strategies in the repeated
game.
A strategy of the stage game is instead called an action.
We assume that A1 , A2 are finite.
We also assume perfect monitoring, i.e. we assume that at each stage the two players
choose their actions simultaneously and, once actions have been chosen, they are
observed before the game moves to the following stage.
The total payoff of the dynamic game is a sum (potentially discounted) of the payoffs in the
stage games.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 41 / 52

Histories

A history is either the empty set or a finite sequence of action profiles, representing what has
happened so far in the game.
The empty set ∅ is the initial history, i.e. nothing has happened yet.
The set of initial histories is H 1 := {∅}.
For all t ≥ 2, the set of period t histories is H t := At−1 .
Let H denote the set of all histories, that is,

H = H1 ∪ H2 ∪ H3 ∪ · · · .

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 42 / 52

Strategies

A strategy for player i is a function si : H → Ai .

A strategy for player i specifies an action si (∅) to take at the first stage and, for each
t ≥ 2 and for each possible period t history H t , an action si (H t ) to take in stage t.
The set of all strategies for player i is denoted Si and as usual we define S := S1 × S2 .

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 43 / 52

Finitely Repeated Games

L M R

A 1,1 5,0 0,0

B 0,5 4,4 0,0

C 0,0 0,0 3,3

This stage game has two NE: (A, L) and (C , R).

If we repeat the stage game twice, it is a SPE to play (A, L) (or (C , R)) in the first stage,
and then (A, L) (or (C , R)) in the second stage, regardless of what happened in the first
stage. Why?

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 44 / 52

Finitely Repeated Games

But there is also a SPE where at the first stage player 1 chooses B and player 2 plays M,
and the second period behavior prescribed by the equilibrium strategies depends in the
following way on what happened in the first stage:
If in the first stage (B, M) was chosen, then in the second stage player 1 plays C and player
2 plays R.
If in the first stage (B, M) was not chosen, then in the second stage player 1 plays A and
player 2 plays L.

Theorem 1. Playing a NE in each stage of the finite repeated game is a SPNE.

Theorem 2. If a stage game has the unique NE (in pure strategies), then there is a unique
SPNE (in pure strategies), prescribing playing NE in each stage.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 45 / 52

Infinitely Repeated Games
Discounting

An outcome is an infinite sequence (a1 , a2 , . . .), so it is not obvious how to compute player i’s
payoff in the repeated game:
if ui (at ) > 0 for all t, then just taking the sum ui (a1 ) + ui (a2 ) + · · · will give +∞
regardless of what the sequence (a1 , a2 , . . .) actually is.

The most common solution to this problem is discounting:

It is assumed that both players evaluate streams of stage payoffs by taking the discounted
sum according to a common discount factor δ ∈ (0, 1).
Thus, player i’s payoff from the outcome (a1 , a2 , . . .) will be given by
∞
X
1 2
Ui (a , a , . . .) = δ t−1 ui (at )
t=1

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 46 / 52

One-shot deviation principle

Using the discounting criterion to evaluate payoffs is convenient because it allows us to use the
one-shot deviation principle to check whether a strategy profile is a subgame-perfect
equilibrium.

The one-shot deviation principle says that a strategy profile is subgame perfect if and only if,
at each history, no player has an incentive to deviate choosing an action different from the one
prescribed by the strategy profile at that stage and then follow the strategy profile in the
continuation, given that the co-player follows the strategy.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 47 / 52

One-shot deviation principle

The one-shot deviation principle was originally formulated by David Blackwell (1965) in the
context of dynamic programming.

The basic idea is that any change in strategy can be broken down into a sequence of
one-period changes; if there is no one-period change that can increase payoffs, then there is no
change in strategy that can increase payoffs.

We will not show a proof here; however, if you are interested you can look at the Notes
uploaded in Blackboard.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 48 / 52

Infinite repetition of the NE of the stage game

Result 1: Suppose the stage game h{1, 2}, (Ai , ui )i∈{1,2} i has at least one pure strategy Nash
equilibrium a∗ ∈ A. Then the strategy profile such that s(h) = a∗ for all h ∈ H is a subgame
perfect equilibrium of the repeated game.
Proof.
It follows from the application of the one-shot deviation principle and from the definition of
NE of the stage game.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 49 / 52

Other SPE: The Folk Theorem

However, in a repeated game there are many other SPE outcomes besides those
consisting in the repetition of a stage game NE.

“Folk Theorem”: for every feasible and individually rational profile of payoffs there exists a
discount factor δ such that, for all δ ≥ δ, there exists a SPE of the δ-discounted repeated
game that gives the players exactly those payoffs.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 50 / 52

Example: the repeated Prisoner’s Dilemma

D C

D 1,1 3,0

C 0,3 2,2

The unique NE of this game is (D, D).

In the repeated game the outcome where (C , C ) is played in every stage is a SPE outcome for δ ≥ 1/2 if
each player adopts the grim-trigger strategy:
s(∅) = (C , C ) and, for all t ≥ 2,
(
1 t−1 (C , C ) if a1 = · · · = at−1 = (C , C ),
s(a , . . . , a ) =
(D, D) otherwise

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 51 / 52

Low Discount Factors

In general, it is hard to describe the set of SPE outcomes for a fixed value of δ.
However, if the stage game has a unique NE and δ is very small, then the only SPE
outcome in the repeated game is the infinite repetition of the stage game NE outcome:
the immediate gain from deviating (which some player must have, if players are not playing a
stage game NE) cannot be compensate by the loss associated to future punishment.

Andrey Tkachenko Game Theory ECON 415,515 Nazarbayev University 52 / 52