Game Theory

71
Game Theory and Strategy

description

a

Transcript of Game Theory

  • Game Theory and Strategy

  • ContentTwo-persons Zero-Sum GamesTwo-Persons Non-Zero-Sum GamesN-Persons Games

  • IntroductionAt least 2 playersStrategiesOutcomePayoffs

  • Two-persons Zero-Sum GamesPayoffs of each outcome add to zeroPure conflict between 2 players

  • Two-persons Zero-Sum Games

    Colin

    A

    B

    C

    D

    Rose

    A

    7, -7

    -1, 1

    1, -1

    0, 0

    B

    5, -5

    1, -1

    6, -6

    -9, 9

    C

    3, -3

    2, -2

    4, -4

    3, -3

    D

    -8, 8

    0, 0

    0, 0

    8, -8

  • Two-persons Zero-Sum Games

    Colin

    A

    B

    C

    D

    Rose

    A

    7

    -1

    1

    0

    B

    5

    1

    6

    -9

    C

    3

    2

    4

    3

    D

    -8

    0

    0

    8

  • Dominance and Dominance Principle Definition: A strategy S dominates a strategy T if every outcome in S is at least as good as the corresponding outcome in T, and at least one outcome in S is strictly better than the corresponding outcome in T.Dominance Principle: A rational player would never play a dominated strategy.

  • Saddle Points and Saddle Points PrincipleDefinition: An outcome in a matrix game is called a Saddle Point if the entry at that outcome is both less than or equal to any in its row, and greater than or equal to any entry in its column.Saddle Point Principle: If a matrix game has a saddle point, both players should play a strategy which contains it.

  • Value Definition: For a matrix game, if there is a number such that player A has a strategy which guarantees that he will win at least v and player B has a strategy which guarantees player A will win no more than v, then v is called the value of the game.

  • Two-persons Zero-Sum Games

    Colin

    A

    B

    C

    D

    Rose

    A

    7

    -1

    1

    0

    B

    5

    1

    6

    -9

    C

    3

    2

    4

    3

    D

    -8

    0

    0

    8

  • Saddle Points

    Minimax

    Colin

    A

    B

    C

    D

    Row minimum

    Rose

    A

    4

    3

    2

    5

    2

    Maximin

    B

    -10

    2

    0

    1

    -10

    C

    7

    5

    2

    3

    2

    Maximin

    D

    0

    8

    -4

    -5

    -5

    Column Maximum

    7

    8

    2

    5

  • Saddle Points0 saddle point1 saddle pointmore than 1 saddle points

  • Mixed Strategy

    Colin

    A

    B

    Rose

    A

    2

    -3

    B

    0

    3

  • Mixed StrategyColin plays with probability x for A, (1-x) for BRose A: x(2) + (1-x)(-3) = -3 + 5xRose B: x(0) + (1-x)(3) = 3 - 3xif -3 + 5x = 3 - 3x => x = 0.75Rose A: 0.75(2) + 0.25(-3) = 0.75Rose B: 0.75(0) + 0.25(3) = 0.75

  • Mixed StrategyRose plays with probability x for A, (1-x) for BColin A: x(2) + (1-x)(0) = 2xColin B: x(-3) + (1-x)(3) = 3 - 6xif 2x = 3 - 6x => x = 0.375Colin A: 0.375(2) + 0.625(0) = 0.75Colin B: 0.375(-3) + 0.625(3) = 0.75

  • Mixed Strategy0.75 as the value of the game0.75A, 0.25B as Colins optimal strategy0.375A. 0.625B as Roses optimal strategy

  • Mixed Strategy

    Colin

    Row difference

    Rose oddments

    Rose probabilities

    A

    B

    Rose

    A

    2

    -3

    2 - (-3) = 5

    3

    3/8

    B

    0

    3

    0 3 = -3

    5

    5/8

    Column difference

    2 0 = 2

    -3 3 = -6

    Colin oddments

    6

    2

    Colin probabilities

    6/8

    2/8

  • Minimax TheoremEvery m x n matrix game has a solution. There is a unique number v, called the value of game, and optimal strategy for the players such thati) player As expected payoff is no less that v, no matter what player B does, andii) player Bs expected payoff is no more that v, no matter what player A doesThe solution can always be found in k x k subgame of the original game

  • Minimax Theorem (example)

    Colin

    A

    B

    C

    D

    E

    Rose

    A

    1

    12

    13

    9

    10

    B

    11

    2

    8

    14

    5

    C

    6

    7

    3

    4

    15

  • Minimax Theorem (example)There is no dominance in the above exampleFrom arrows in the graph, Colin will only choose A, B or C, but not D or E.So the game is reduced into a 3 x 3 subgame

  • Example

    9-Police

    9-0

    8-1

    7-2

    6-3

    5-4

    7-Guerrillas

    7-0

    1/2

    1/2

    1/2

    1

    1

    6-1

    1

    1/2

    1/2

    1/2

    1

    5-2

    1

    1

    1/2

    1/2

    4-3

    1

    1

    1

    1/2

    0

  • Example

    9-Police

    7-2

    6-3

    5-4

    7-Guerrillas

    7-0

    1/2

    1

    1

    6-1

    1/2

    1/2

    1

    5-2

    1/2

    1/2

    4-3

    1

    1/2

    0

  • Example

    9-Police

    7-2

    6-3

    5-4

    7-Guerrillas

    7-0

    1/2

    1

    1

    4-3

    1

    1/2

    0

  • Example

    9-Police

    7-2

    5-4

    7-Guerrillas

    7-0

    1/2

    1

    4-3

    1

    0

  • Mixed Strategy

    9-Police

    Row difference

    Guerrillas oddments

    Guerrillas probabilities

    7-2

    5-4

    7-Guerrillas

    7-0

    1/2

    1

    -1/2

    1

    2/3

    4-3

    1

    0

    1

    1/3

    Column difference

    1/2

    1

    Police oddments

    1

    1/2

    Police probabilities

    2/3

    1/3

  • Utility Theory

    Colin

    A

    B

    Rose

    A

    U

    V

    B

    W

    X

    C

    Y

    Z

  • Utility TheoryRoses order is u, w, x, z, y, vColins order is v, y, z, x, w, u

    Colin

    A

    B

    Rose

    A

    6

    1

    B

    5

    4

    C

    2

    3

  • Utility Theory

    v

    w

    x

    u

    i)

    0

    20

    40

    60

    80

    100

    ii)

    -1

    0

    1

    2

    3

    4

    iii)

    17

    19

    21

    23

    25

    27

  • Utility TheoryTransformation can be done using a positive linear function, f(x) = ax + bin this example, f(x) = 0.5(x - 17)

    -------->

    Colin

    A

    B

    Rose

    A

    27, -5

    17, 0

    B

    19, -1

    23, -3

    Colin

    A

    B

    Rose

    A

    5, -5

    0, 0

    B

    1, -1

    3, -3

  • Two-Persons Non-Zero-Sum GamesEquilibrium outcomes in non-zero-sum games ~ saddle points in zero-sum games

  • Prisoners Dilemma

    Colin

    Confess

    Dont

    Rose

    Confess

    10, 10

    0, 20

    Dont

    20, 0

    1, 1

  • Nash EquilibriumIf there is a set of strategies with the property that no player can benefit by changing her strategy while the other players keep their strategies unchanged, then that set of strategies and the corresponding payoffs constitute the Nash Equilibrium

  • Dominant Strategy EquilibriumIf every player in the game has a dominant strategy, and each player plays the dominant strategy, then that combination of strategies and the corresponding payoffs are said to constitute the dominant strategy equilibrium for that game.

  • Pareto-optimalIf an outcome cannot be improved upon, ie. no one can be made better off without making somebody else worse off, then the outcome is Pareto-optimal

  • Pareto PrincipleTo be acceptable as a solution to a game, an outcome should be Pareto-optimal.

  • Prudential Strategy, Security Level and Counter-Prudential StrategyIn a non-zero-sum game, player As optimal strategy in As game is called As prudential strategy.The value of As game is called As security levelAs counter-prudential strategy is As optimal response to his opponents prudential strategy.

  • Example

    Colin

    A

    B

    Rose

    A

    2, 4

    1, 0

    B

    3, 1

    0, 4

  • Exampleconsider only Roses strategysaddle point at AB

    Colin

    A

    B

    Rose

    A

    2

    1

    B

    3

    0

  • Exampleconsider only Colins strategy

    Colin

    A

    B

    Rose

    A

    4

    0

    B

    1

    4

    Column difference

    4-1=3

    0-4=-4

    Colin oddments

    4

    3

    Colin probabilities

    4/7

    3/7

  • Example

    Rose strategy

    Colin strategy

    Rose payoff

    Colin payoff

    prudential

    prudential

    11/7

    16/7

    prudential

    Counter-prudential

    2

    4

    Counter-prudential

    Prudential

    12/7

    16/7

    Counter-prudential

    Counter-prudential

    3

    1

  • Example

    Rose prudential

    A

    Colin Prudential

    4/7A, 3/7B

    Rose Counter-prudential

    B

    Colin Counter-prudential

    A

  • Example

    BB AA

    Equilibrium

    BA AB

  • Co-operative Solution

    Negotiation Set

  • Co-operative Solution

    Negotiation Set

  • Co-operative SolutionConcerns are Trust and Suspicion

  • N-Person GamesMore important and common in real lifen is assumed to be at least three

  • N-Person Games

    Larry A

    Colin

    A

    B

    Rose

    A

    1, 1, -2

    -4, 3, 1

    B

    2, -4, 2

    -5, -5, 10

    Larry B

    Colin

    A

    B

    Rose

    A

    3, -2, -1

    -6, -6, 12

    B

    2, 2, -4

    -2, 3, 1

  • N-Person Games

    Colin and Larry

    AA

    BA

    AB

    BB

    Rose

    A

    1

    -4

    3

    -6

    B

    2

    -5

    2

    -2

  • N-Person Games

    Rose and Larry

    AA

    BA

    AB

    BB

    Colin

    A

    1

    -4

    -2

    2

    B

    3

    -5

    -6

    3

  • N-Person Games

    Rose and Colin

    AA

    BA

    AB

    BB

    Larry

    A

    -2

    2

    1

    10

    B

    -1

    -4

    12

    -1

  • N-Person Games

    Colin and Larry

    BA

    BB

    Rose optimal

    Rose

    A

    -4

    -6

    3/5

    B

    -5

    -2

    2/5

    Colin and Larry optimal

    4/5

    1/5

    Value = -4.4

  • N-Person Games

    Rose and Larry

    BA

    AB

    Colin optimal

    Colin

    A

    -4

    -2

    1

    B

    -5

    -6

    0

    Rose and Larry optimal

    1

    0

    Value = -4

  • N-Person Games

    Rose and Colin

    AA

    BA

    Larry optimal

    Larry

    A

    -2

    2

    3/7

    B

    -1

    -4

    4/7

    Rose and Colin optimal

    6/7

    1/7

    Value = -1.43

  • N-Person GamesThe result is

    Rose v.s. Colin and Larry

    -4.4, -0.64, 5.04

    Colin v.s. Rose and Larry

    2, -4, 2

    Larry v.s. Rose and Colin

    2.12, -0.69, -1.43

  • SuperaddictiveA characteristic function form game (N, v) is called superadditiveif v(S, T) >= v(S) + v(T) for any two coalitions S and T

  • N-Person Prisoners Dilemma

    Number of others choosing C

    0

    1

    2

    3

    4

    Player chooses

    C

    -2

    -1

    0

    1

    2

    D

    -1

    0

    1

    2

    3

  • N-Person Prisoners DilemmaGeneral form of N-Person Prisoners Dilemmaeach of n players has two strategies, C and Dfor every player, D is a dominant strategyif all players choose D, add will be worse off than if all players had chosen C

  • Example

    Blues

    Red

    A

    B

    B

    C

    C

    D

    D

    A

  • ExampleSincere choice

    Blues

    Red

    1st round

    A

    B

    B

    C

    2nd round

    C

    D

    D

    A

  • ExampleOptimal choice

    Blues

    Red

    2nd round

    A

    B

    1st round

    B

    C

    C

    D

    D

    A

  • From the bottom up algorithmi) under optimal play, the Reds choice in last round will be the player who is last on the Blues preference list. Mark that player as the Reds last round choice and cross him off both teams listsii) the Blues choice in last round will be the player who is last on the Reds reduced list. Mark the player as Blues and cross him off both teams listsiii) continue like this, finding the choices in the next-to-last round, and on up to the first round

  • ExampleSincere choice

    Blues

    Red

    A

    B

    B

    C

    2nd round

    C

    D

    D

    A

  • ExampleSincere choice

    Blues

    Red

    A

    B

    B

    C

    2nd round

    C

    D

    D

    A

  • ExampleSincere choice

    Blues

    Red

    A

    B

    1st round

    B

    C

    2nd round

    C

    D

    D

    A

  • ExampleSincere choice

    Blues

    Red

    A

    B

    1st round

    B

    C

    2nd round

    C

    D

    D

    A

  • Example of N-Person Prisoners Dilemma

    Blues

    Red

    Green

    A

    E

    C

    B

    F

    F

    C

    B

    E

    D

    A

    D

    E

    D

    A

    F

    C

    B

  • Example of N-Person Prisoners DilemmaSincere Choice

    Blues

    Red

    Green

    1st round

    A

    E

    C

    2nd round

    B

    F

    F

    C

    B

    E

    D

    A

    D

    E

    D

    A

    F

    C

    B

  • Example of N-Person Prisoners DilemmaAfter Greens optimal Choice

    Blues

    Red

    Green

    1st round

    A

    E

    C

    2nd round

    B

    F

    F

    C

    B

    E

    D

    A

    D

    E

    D

    A

    F

    C

    B

  • Example of N-Person Prisoners DilemmaAfter Reds optimal Choice

    Blues

    Red

    Green

    1st round

    A

    E

    C

    B

    F

    F

    C

    B

    E

    2nd round

    D

    A

    D

    E

    D

    A

    F

    C

    B

  • Example of N-Person Prisoners DilemmaAfter Blues optimal Choice

    Blues

    Red

    Green

    2nd round

    A

    E

    C

    B

    F

    F

    1st round

    C

    B

    E

    D

    A

    D

    E

    D

    A

    F

    C

    B

  • END