Game Theory III: Extensive Form Games and Repeated Games

RECAP:
- Last session alleviated the assumption that players act simultaneously. – if they act
one after the other then they can engage in an extensive form game and act
accordingly
- This session also alleviates the assumption that players only act once – REPEATED
GAMES – in the market firms don’t leave the market after one period

Extensive form game recap;

- Definition of extensive form games

- Introduction of game trees and subgames

- Definition of subgame-perfect equilibria as solution concept for extensive form games

Start with a normal form game and matrix eg:

Here, 2 NASH EQUILIBRIUM

- A PREFERS D,L
- B PREFERS U,R
- Two pure-strategy Nash equilibria (U,R) and (D,L) One mixed-strategy Nash
equilibrium
- Coordination problem since each player prefers a different pure strategy Nash
equilibrium

Both players prefer different Nash equilibrium so we have the coordination problem
Can solve with sequential moves so allowing one to observe

Eg:
Explaining this:
A goes U :
- B would rather go R as 8>7
A goes D:
- B would rather go L as 5>3
-
When we let A go first, B chooses L if N3 and R if N2

End up with D,L with payoff 7,5 that A prefers and solves coordination problem

SUBGAME PERFECT EQUIULIBRIUM IS U,R

- WHOEVER STARTS PICKS THE NASH EQUILIBROUM THEY PREFER IN THE GAME

Summary:
 - In simultaneous-move game, we have 2 Nash equilibria (U,R) and (D,L)

- Coordination problem since A prefers (D,L) and B prefers (U,R)

- A starts: SPE is (D,L) with (7,5)

- B starts: SPE is (U,R) with (6,8)

- SPE is unique but may depend on who starts

- First-mover advantage: Difference between when a player moves first to when player
moves second

- First-mover advantage A: 7 − 6 = 1

- First-mover advantage B: 8 − 5 = 3

Example 2:
In a normal form game, even if you choose simultaneously, they will always choose r
So independently of who stars you always end up with the same subgame perfect
equilibrium (U,R)

However, no coordination problem since both players prefer (U,R) over (D,L)
Same concept as before but left with one answer both ways

Summary:
- Normal form game: (U,R) and (D,L)

- But no coordination problem since both players prefer (U,R) with (8,9)

- In this extensive form game (U,R) is the SPE independent of who starts

- Therefore, no first-mover advantage

NO FIRST MOVER ADVANTAGE HERE

Example 3: Prisoners Dilemma

- Dominant strategy leads to an equilibrium payoff lower than if they worked together
(exactly like in the real market world)

- Both players have a dominant strategy

Unique Nash equilibrium in dominant strategies
- (U,L) with (9,9).
However, (D, R) with (18, 18) would leave both
- players better off
Summary:

- Normal form game: (U,L) as equilibrium in dominant strategies.

- Independent of who moves first, the SPE is the equilibrium in dominant strategies
(U,L)

- Therefore, there is no first-mover advantage

- But, Pareto-optimal outcome (D,R) is not reached

IV: Matching Pennies:

- Randomise between U and D, then see

Summary:
- Subgaime N1: A is indifferent between U and D
- A will randomise between U and D; σA = (0.5, 0.5)
- B moving second knows whether s/he is in N2 or N3 and acts accordingly.
- No SPE in pure strategies

Example V: NOT COMING UP IN EXAM TOO HARD

Summary:
- Extensive form games need a refinement of the Nash equilibrium — subgame perfect
equilibrium
- SPE in pure strategies always corresponds to a Nash equilibrium in a normal form
game

- SPE in pure strategies is always unique

- First-mover advantage if there are multiple Nash equilibria in pure strategies in the
normal form game and players prefer different equilibria

- Extensive form games are multi-stage games (not all decisions are made at once)

- Here, multi-stage but still only one period

- There are multi-stage, multi-period extensive form games

Repeated games: (applies for both simultaneously and sequential games)

- Repeated games are a special form of multi-stage games

- A simultaneous-move one-shot game is played repeatedly

- Finite and infinite horizon games

- (Elaborate) conditional strategies become available

Horizon:
 - Horizon refers to the number of rounds played

- Finite horizon: last round T is known

- Infinite horizon:

o There is no last round

- Last round is unknown

- But discounting takes place

Conditional strategies

- Conditional strategies means that players’ actions in t > 1 depends on opponents’

actions in t − 1 and potentially earlier.

- Large number of conditional strategies can be designed

- Two famous examples for infinite horizon

o Carrot-and-Stick strategy: “cooperate in t = 1 and continue to cooperate if the

opponents’ have done so in t − 1 otherwise defect forever”

o Tit-for-Tat strategy: “cooperate in t = 1 and then do what the opponents’

choose in t − 1.” LEARN FOR MCQ EXAM

THESE REPEATED GAMES IN IND ECON HELP EXAPLIN COLLUSION

WHY DO THEY NOT WANT TO COMOETE AND COOPERATE – EXPLAINED WITH
REPEATED GAMES IN CARROT AND STICK AND TIT FOR TAT STRATEGY

Solution concept:

- Since repeated games are (special) multi-stage games, SPE can be used for finite
horizon games

Prisoners dilemma:

One-shot game
 - A has dominant strategy U, B has dominant strategy L

- Equilibrium in dominant strategies (U,L)

Once repeated

- A repeated game is a multi-stage (and multi-period) game

- Therefore, subgame-perfect equilibrium is applied.
- Identify the subgames, solve from bottom up

Stylised game tree

- PD1 (N1)  PD2 (N2)

- Two subgames, N1 and N2

- In each subgame, players move simultaneously

- When players are at N2, both players know what happened at N1

- When players are in N1, both players anticipate what will happen in N2

Subgame N2

- Doesn’t matter how players arrived at N2.

- Dominant strategies are still U for player A and L for player B
- Nash equilibrium for N2 will be (U,L)

Subgame N1

- Players anticipate equilibrium (U,L) in N2

- In N1, players will choose their dominant strategies
- SPE is ((U,L), (U,L))

We do not get to 14,14 – no trust, no sense of justice – just rational beings

Summary prisoners’ dilemma

- Players will choose the dominant strategy in both periods (2 periods)

 - (D,R) with higher payoff for both cannot be achieved
- This will be the outcome for all finite horizon versions of the prisoners’ dilemma
- At the end of the day same outcome in any time period

Summary finite horizon repeated games

- In repeated games, punishment is possible

- However, Pareto optimum will not be achieved in finite horizon games

- This is not replicated in experiments for various reasons

- Our rational players do have no use for sentiments (fairness, trust etc)

- We assume subjects have infinite abstraction power

In Ind econ – infinite horizon games more accurate as resembles what happens with
firms – so much more important

BUT NEED TO KNOW FINITE FOR EXAMS