Scribd Logo
Upload

Welcome to Scribd!

  • 9 views

    Level - K Thinking

    Uploaded by

    social sites

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Level - K Thinking

    Uploaded by

    social sites
    9 views19 pages

    Original Title

    Level_k thinking

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    9 views19 pages

    Level - K Thinking

    Uploaded by

    social sites

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 19

    Level K Thinking

    Mark Dean

    ECON 1820 - Brown University - Spring 2015


    Introduction

    Game theory: The study of strategic decision making


    Your outcome depends on your own actions and the actions of
    others
    Standard tool for prediction: Nash Equilibrium
    No player has incentive to deviate given the actions of others
    But Nash Equilibrium has some problems
    Play of experimental subjects systematically violate its
    predictions
    Can be very complex to calculate
    Assumes a high degree of rationality on the part of subject
    Assumes that THEY assume a high degree of rationality on
    the part of others

    Level K model tries to deal with both of these problems


    An Example: The p Beauty Contest Game

    n players
    Each player chooses si 2 f1, 2....100g = Si
    Earn $10 if you are closes to p times average choice

    ∑nk =1 sk
    si 2 arg min jsj p j
    s1 ...sn n
    Earn zero otherwise
    Split the prize in event of the tie
    p 2 (0, 1)
    This de…nes ui (si , fs1 , ..si 1 , si +1 , ..sn g) = u ( si , s i )
    The utility if you play si and others play s i
    Nash Equilibrium

    Nash Equilibrium: A strategy pro…le fs1 , ...sn g such that no


    player has an incentive to deviate

    u ( si , s i ) u (s, s i ) 8 s 2 Si

    What is the Nash Equilibrium of the p-beauty contest game?


    Nash Equilibrium

    Nash Equilibrium: A strategy pro…le fs1 , ...sn g such that no


    player has an incentive to deviate

    u ( si , s i ) u (s, s i ) 8 s 2 Si

    What is the Nash Equilibrium of the p-beauty contest game?


    The unique Nash equilibrium is si = 1 8 i
    No gain by deviating for any player
    ∑nk =1 sk
    For any other strategy pro…le, any player with si n has
    incentive to deviate
    Nash Equilibrium

    Do people play Nash Equilibrium strategies?


    Nash Equilibrium

    Do people play Nash Equilibrium strategies?


    Makes strong rationality assumptions
    That players can …gure out what the Nash Equilibrium is
    They assume that others can …gure out what the Nash
    equilibrium is
    Nash Equilibria may also come about through a process of
    learning
    We will focus on one shot games
    Depth of Reasoning

    2
    Consider the following sequence of reasoning for the 3 beauty
    contest
    1 I think the other players will play 50, so I will play the best
    response to 50, i.e 33 13
    2 I think the other players think everyone will play 50 and so will
    play 33 13 . I will therefore play the best response to this, i.e.
    22 92
    3 I think that the other players will initially think that everyone
    will play 50, and will consider playing 33 13 . However, they will
    think that others have done the same reasoning, and will
    therefore play 22 29 . I will best respond to this and play 14 22
    27
    4 ......
    Depth of Reasoning

    More generally (in the case of two players)


    1 I assume that the other player will play s̄, so I will play
    si1 2 arg maxs 2Si ui (s, s̄ )
    2 I assume that other players will best respond to s̄ and so play
    sj1 2 arg maxs 2Sj uj (s, s̄ ). I will therefore play
    si2 2 arg maxs 2Si ui (s, sj1 )
    3 I assume that other players will best respond to sj1 and so play
    sj2 2 arg maxs 2Sj uj (s, si1 ). I will therefore play
    si3 2 arg maxs 2Si ui (s, sj2 )
    Depth of Reasoning

    Notice that a Nash equilibrium is a …xed point of this type of


    reasoning
    I assume that other players will best respond to si and so play
    sj 2 arg maxs 2Sj uj (s, si ). I will therefore play
    si 2 arg maxs 2Si ui (s, sj )
    In the case of the p-beauty contest game this type of
    reasoning will converge to the Nash Equilibrium
    This is not always true
    Level-K Thinking

    What if you are constrained in how many steps of this type of


    reasoning that you can do?
    You have a ‘type’equal to the
    Level 0: Non-strategic (play at random)
    Level 1: Best respond to level 0
    Level 2: Best respond to level 1
    Level 3: Best respond to level 2
    ...
    There is a distribution of types in the population: π i
    probability of level i
    Generally assumed that π i = 0
    ‘Anchor’for remaining levels
    Level-K Thinking

    2
    What would this imply for the data in the 3 beauty contest
    game?
    We would see a focus of responses at the following levels:
    π0 : 50
    π1 : 33 13
    π2 : 22 92
    π3 22
    : 14 27
    Nagel [1995]
    Class Data
    Class Data

    You are much more ‘rational’than the Nagel sample


    Class mean 22.6 vs 36.7 in the Nagal Data
    Many more ‘Nash’players
    Some evidence that there are a number of level 2 players
    10 players played 22
    2/3 of the mean: 15.3
    Winning guess
    Class Data

    You are much more ‘rational’than the Nagel sample


    Class mean 22.6 vs 36.7 in the Nagal Data
    Many more ‘Nash’players
    Some evidence that there are a number of level 2 players
    10 players played 22
    2/3 of the mean: 15.3
    Winning guess
    Max Deutsch
    Shuying Ni
    Joshua Herman
    Issues with Level K Model

    Lots of additional degrees of freedom


    What is level 0?
    What is the distribution of types?
    The model has low predictive power
    Consistent with any choice pattern
    Needs more (ad hoc) assumptions in order to constrain it
    Issues with Level K Model

    Are types …xed?


    Should be able to use estimated type in one game to predict
    play in others
    Georganas et al [2013] get same subject to play ‘undercutting’
    and ‘guessing games’
    Estimate type in each case
    Find no correlation in estimated type or estimated rank
    Issues with Level K Model

    Response to learning and incentives?


    There is also evidence that types change predictably
    Nagel [1995] - bidding in the p-beauty contest game falls with
    experience
    Alaoui and Penta [2013] - subjects change their level of play
    with incentives

    You might also like