Game Theory

Game Theory
A mathematical theory designed to model: how rational individuals should behave when individual outcomes are determined by collective behavior strategic behavior Rational usually means selfish --- but not always Traditionally viewed as a subject of economics Subsequently applied by many fields evolutionary biology, social psychology

An example: Big Monkey and Little Monkey

Monkeys usually eat ground-level fruit Occasionally climb a tree to get a coconut (1 per tree) A Coconut yields 10 Calories Big Monkey expends 2 Calories climbing the tree. Little Monkey expends 0 Calories climbing the tree.

An example: Big Monkey and Little Monkey

If BM climbs the tree
BM gets 6 C, LM gets 4 C LM eats some before BM gets down

If LM climbs the tree

BM gets 9 C, LM gets 1 C BM eats almost all before LM gets down

If both climb the tree

BM gets 7 C, LM gets 3 C BM hogs coconut

How should the monkeys each act so as to maximize their own calorie gain?

An example: Big Monkey and Little Monkey

Assume BM decides first
Two choices: wait or climb

LM has four choices:

Always wait, always climb, same as BM, opposite of BM.

These choices are called actions

A sequence of actions is called a strategy

An example: Big Monkey and Little Monkey

It can often be easier to analyze a game through a different representation, called normal form
Little Monkey c Big Monkey c v 5,3 9,1 v 4,4 0,0

Choosing Strategies
In the simultaneous game, its harder to see what each monkey should do
Mixed strategy is optimal.

Trick: How can a monkey maximize its payoff, given that it knows the other monkeys will play a Nash strategy? Oftentimes, other techniques can be used to prune the number of possible actions.

Eliminating Dominated Strategies

We can also use this technique in normalform games:
Column a a Row 9,1 b 4,4

5,3

0,0

Eliminating Dominated Strategies

We can also use this technique in normalform games:
a a 9,1 b 4,4

5,3

0,0

For any column action, row will prefer a.

Eliminating Dominated Strategies

We can also use this technique in normalform games:
a a 9,1 b 4,4

5,3

0,0

Given that row will pick a, column will pick b. (a,b) is the unique Nash equilibrium.

Dominant Firm Game Two firms, one large and one small Either firm can announce an output level (lead) or else wait to see what the rival does and then produce an amount that does not saturate the market.

Dominant Firm Game

Subordin ate
Lead Dominant Lead

Follow

( 0.5,

4)

( 3, 2)

Follow

( 1, 8)

( 0.5, 1)

Dominant Firm Game

Dominant

Subordin ate
Lead

Lead

Follow

( 0.5,

4)

( 3, 2)

Follow

( 1, 8)

( 0.5, 1)

Dominant Firm Game

Dominant

Subordin ate
Lead

Lead

Follow

( 0.5,

4)

( 3, 2)

Follow

( 1, 8)

( 0.5, 1)

Dominant Firm Game

Conclusion:
Dominant Firm will always lead..

But what about the Subordinate firm?

Dominant Firm Game

Subordin ate
Lead Dominant Lead

Follow

( 0.5,

4)

( 3, 2)

Follow

( 1, 8)

( 0.5, 1)

Dominant Firm Game

Subordin ate
Lead Dominant Lead

Follow

( 0.5,

4)

( 3, 2)

Follow

( 1, 8)

( 0.5, 1)

Dominant Firm Game

Conclusion:
No dominant strategy for the Subordinate firm. Does this mean we cannot predict what they will do?

Dominant Firm Game

Subordin ate
Lead Dominant Lead

Follow

( 0.5,

4)

( 3, 2)

Follow

( 1, 8 )

( 0.5, 1)

Dominant Firm Game

Conclusion:
Subordinate firm will always follow, because dominant firm will always lead.

Definition of Nash Equilibrium

A game has n players. Each player i has a strategy set Si
This is his possible actions

Each player has a payoff function

pI: S ! R

A strategy ti 2 Si is a best response if there is no other strategy in Si that produces a higher payoff, given the opponents strategies.

Definition of Nash Equilibrium

A strategy profile is a list (s1, s2, , sn) of the strategies each player is using. If each strategy is a best response given the other strategies in the profile, the profile is a Nash equilibrium. Why is this important?
If we assume players are rational, they will play Nash strategies. Even less-than-rational play will often converge to Nash in repeated settings.

An Example of a Nash Equilibrium

Column a a Row b 2,1 1,0 1,2 b 0,1

(b,a) is a Nash equilibrium. To prove this: Given that column is playing a, rows best response is b. Given that row is playing b, columns best response is a.

Finding Nash Equilibria Dominated Strategies

What to do when its not obvious what the equilibrium is? In some cases, we can eliminate dominated strategies.
These are strategies that are inferior for every opponent action.

In the previous example, row = a is dominated.

Prisoners Dilemma
Each player can cooperate or defect
Column
cooperate defect

cooperate
Row defect

-1,-1
0,-10

-10,0
-8,-8

Prisoners Dilemma
Each player can cooperate or defect
Column
cooperate defect

cooperate
Row defect

-1,-1
0,-10

-10,0
-8,-8

Defecting is a dominant strategy for row

Prisoners Dilemma
Each player can cooperate or defect
Column
cooperate defect

cooperate
Row defect

-1,-1
0,-10

-10,0
-8,-8

Defecting is also a dominant strategy for column

Prisoners Dilemma
Even though both players would be better off cooperating, mutual defection is the dominant strategy. What drives this?
One-shot game Inability to trust your opponent Perfect rationality

Prisoners Dilemma
Relevant to:
Arms negotiations Online Payment Product descriptions Workplace relations

How do players escape this dilemma?

Play repeatedly Find a way to guarantee cooperation Change payment structure

Coordination Games
Consider the following problem:
A supplier and a buyer need to decide whether to adopt a new purchasing system.
Buyer

new
new Supplier old 0,0 20,20

old
0,0 5,5

No dominated strategies!

Coordination Games
Buyer new new Supplier old 20,20 0,0 old 0,0 5,5

This game has two Nash equilibria (new,new) and (old,old) Real-life examples: Beta vs VHS, Mac vs Windows vs Linux, others? Each player wants to do what the other does which may be different than what they say theyll do How to choose a strategy? Nothing is dominated.

Solving Coordination Games

Coordination games turn out to be an important real-life problem
Technology/policy/strategy adoption, delegation of authority, synchronization

Human agents tend to use focal points

Solutions that seem to make natural sense
e.g. pick a number between 1 and 10

Social norms/rules are also used

Driving on the right/left side of the road

These strategies change the structure of the game

Mixed strategies
Unfortunately, not every game has a pure strategy equilibrium.
Rock-paper-scissors

However, every game has a mixed strategy Nash equilibrium. Each action is assigned a probability of play. Player is indifferent between actions, given these probabilities.

Mixed Strategies
In many games (such as coordination games) a player might not have a pure strategy. Instead, optimizing payoff might require a randomized strategy (also called a mixed strategy)
Wife
football shopping

football
Husband shopping

2,1
0,0

0,0
1,2

Strategy Selection
Wife

football football Husband

shopping

shopping

2,1
0,0

0,0
1,2

If we limit to pure strategies: Husband: U(football) = 0.5 * 2 + 0.5 * 0 = 1 U(shopping) = 0.5 * 0 + 0.5 * 1 = Wife: U(shopping) = 1, U(football) = Problem: this wont lead to coordination!

Mixed strategy
Instead, each player selects a probability associated with each action
Goal: utility of each action is equal Players are indifferent to choices at this probability

a=probability husband chooses football b=probability wife chooses shopping Since payoffs must be equal, for husband:
b*1=(1-b)*2 b=2/3

For wife:
a*1=(1-a)*2 = 2/3

In each case, expected payoff is 2/3

2/9 of time go to football, 2/9 shopping, 5/9 miscoordinate

If they could synchronize ahead of time they could do better.

Example: Rock paper scissors

Column rock rock Row paper 1,-1 -1,1 0,0 1,-1 -1,1 0,0 0,0 paper -1,1 scissors 1,-1

scissors

Setup
Player 1 plays rock with probability pr, scissors with probability ps, paper with probability 1-pr ps P2: Utility(rock) = 0*pr + 1*ps 1(1-pr ps) = 2 ps + pr -1 P2: Utility(scissors) = 0*ps + 1*(1 pr ps) 1pr = 1 2pr ps P2: Utility(paper) = 0*(1-pr ps)+ 1*pr 1ps = pr ps
Player 2 wants to choose a probability for each strategy so that the expected payoff for each strategy is the same.

Setup
qr(2 ps + pr 1) = qs(1 2pr ps) = (1-qr-qs) (pr ps) It turns out (after some algebra) that the optimal mixed strategy is to play each strategy of the time.

Intuition: What if you played rock half the time? Your opponent would then play paper half the time, and youd lose more often than you won. So youd decrease the fraction of times you played rock, until your opponent had no edge in guessing what youll do.

Repeated games
Many games get played repeatedly A common strategy for the husband-wife problem is to alternate
This leads to a payoff of 1, 2,1,2, 1.5 per week.

Requires initial synchronization, plus trust that partner will go along. Difference in formulation: we are now thinking of the game as a repeated set of interactions, rather than as a one-shot exchange.

Repeated vs Stage Games

There are two types of multiple-action games:
Stage games: players take a number of actions and then receive a payoff.
Checkers, chess, bidding in an ascending auction

Repeated games: Players repeatedly play a shorter game, receiving payoffs along the way.
Poker, blackjack, rock-paper-scissors, etc

Analyzing Stage Games

Analyzing stage games requires backward induction We start at the last action, determine what should happen there, and work backwards.
Just like a game tree with extensive form.

Strange things can happen here:

Centipede game
Players alternate can either cooperate and get $1 from nature or defect and steal $2 from your opponent Game ends when one player has $100 or one player defects.

Analyzing Repeated Games

Analyzing repeated games requires us to examine the expected utility of different actions. Assumption: game is played infinitely often
Weird endgame effects go away.

Prisoners Dilemma again:

In this case, tit-for-tat outperforms defection.

Collusion can also be explained this way.

Short-term cost of undercutting is less than long-run gains from avoiding competition.