1 University of Michigan, Ann Arbor Teck H. Ho A Cognitive Hierarchy (CH) Model of Games Teck H. Ho Haas School of Business University of California, Berkeley Joint work with Colin Camerer, Caltech Juin-Kuan Chong, NUS

2 University of Michigan, Ann Arbor Teck H. Ho Motivation  Nash equilibrium and its refinements: Dominant theories in economics and marketing for predicting behaviors in competitive situations.  Subjects do not play Nash in many one-shot games.  Behaviors do not converge to Nash with repeated interactions in some games.  Multiplicity problem (e.g., coordination games).  Modeling heterogeneity really matters in games.

3 University of Michigan, Ann Arbor Teck H. Ho Main Goals  Provide a behavioral theory to explain and predict behaviors in any one-shot game  Normal-form games (e.g., zero-sum game, p- beauty contest)  Extensive-form games (e.g., centipede)  Provide an empirical alternative to Nash equilibrium (Camerer, Ho, and Chong, QJE, 2004) and backward induction principle (Ho, Camerer, and Chong, 2005)

4 University of Michigan, Ann Arbor Teck H. Ho Modeling Principles PrincipleNash CH Strategic Thinking   Best Response   Mutual Consistency 

5 University of Michigan, Ann Arbor Teck H. Ho Modeling Philosophy Simple(Economics) General(Economics) Precise(Economics) Empirically disciplined(Psychology) “the empirical background of economic science is definitely inadequate...it would have been absurd in physics to expect Kepler and Newton without Tycho Brahe” (von Neumann & Morgenstern ‘44) “Without having a broad set of facts on which to theorize, there is a certain danger of spending too much time on models that are mathematically elegant, yet have little connection to actual behavior. At present our empirical knowledge is inadequate...” (Eric Van Damme ‘95)

6 University of Michigan, Ann Arbor Teck H. Ho Example 1: “zero-sum game” Messick(1965), Behavioral Science

7 University of Michigan, Ann Arbor Teck H. Ho Nash Prediction: “zero-sum game”

8 University of Michigan, Ann Arbor Teck H. Ho CH Prediction: “zero-sum game”

9 University of Michigan, Ann Arbor Teck H. Ho Empirical Frequency: “zero-sum game”

10 University of Michigan, Ann Arbor Teck H. Ho The Cognitive Hierarchy (CH) Model  People are different and have different decision rules  Modeling heterogeneity (i.e., distribution of types of players). Types of players are denoted by levels 0, 1, 2, 3,…,  Modeling decision rule of each type

11 University of Michigan, Ann Arbor Teck H. Ho Modeling Decision Rule  Proportion of k-step is f(k)  Step 0 choose randomly  k-step thinkers know proportions f(0),...f(k-1)  Form beliefs and best-respond based on beliefs  Iterative and no need to solve a fixed point

12 University of Michigan, Ann Arbor Teck H. Ho

13 University of Michigan, Ann Arbor Teck H. Ho Theoretical Implications  Exhibits “increasingly rational expectations”   ∞  Normalized g K (h) approximates f(h) more closely as k  ∞ (i.e., highest level types are “sophisticated” (or "worldly") and earn the most  ∞  Highest level type actions converge as k  ∞   0  marginal benefit of thinking harder  0

14 University of Michigan, Ann Arbor Teck H. Ho Modeling Heterogeneity, f(k)  A1:  sharp drop-off due to increasing difficulty in simulating others’ behaviors  A2: f(0) + f(1) = 2f(2)

15 University of Michigan, Ann Arbor Teck H. Ho Implications  A1  Poisson distribution with mean and variance =   A1,A2  Poisson,  golden ratio Φ)

16 University of Michigan, Ann Arbor Teck H. Ho Poisson Distribution  f(k) with mean step of thinking  :

17 University of Michigan, Ann Arbor Teck H. Ho

18 University of Michigan, Ann Arbor Teck H. Ho Theoretical Properties of CH Model  Advantages over Nash equilibrium  Can “solve” multiplicity problem (picks one statistical distribution)  Sensible interpretation of mixed strategies (de facto purification)  Theory:  τ  ∞ converges to Nash equilibrium in (weakly) dominance solvable games

19 University of Michigan, Ann Arbor Teck H. Ho Estimates of Mean Thinking Step 

20 University of Michigan, Ann Arbor Teck H. Ho Nash: Theory vs. Data

21 University of Michigan, Ann Arbor Teck H. Ho CH Model: Theory vs. Data

22 University of Michigan, Ann Arbor Teck H. Ho Economic Value  Evaluate models based on their value-added rather than statistical fit (Camerer and Ho, 2000)  Treat models like consultants  If players were to hire Mr. Nash and Ms. CH as consultants and listen to their advice (i.e., use the model to forecast what others will do and best-respond), would they have made a higher payoff?

23 University of Michigan, Ann Arbor Teck H. Ho Nash versus CH Model: Economic Value

24 University of Michigan, Ann Arbor Teck H. Ho Application: Strategic IQ  A battery of 30 "well-known" games  Measure a subject's strategic IQ by how much money she makes (matched against a defined pool of subjects)  Factor analysis + fMRI to figure out whether certain brain region accounts for superior performance in "similar" games  Specialized subject pools  Soliders  Writers  Chess players  Patients with brain damages

25 University of Michigan, Ann Arbor Teck H. Ho Example 2: P-Beauty Contest  n players  Every player simultaneously chooses a number from 0 to 100  Compute the group average  Define Target Number to be 0.7 times the group average  The winner is the player whose number is the closet to the Target Number  The prize to the winner is US$20 Ho, Camerer, and Weigelt (AER, 1998)

26 University of Michigan, Ann Arbor Teck H. Ho A Sample of CEOs  David Baltimore President California Institute of Technology  Donald L. Bren Chairman of the Board The Irvine Company Eli Broad Chairman SunAmerica Inc. Lounette M. Dyer Chairman Silk Route Technology David D. Ho Director The Aaron Diamond AIDS Research Center Gordon E. Moore Chairman Emeritus Intel Corporation Stephen A. Ross Co-Chairman, Roll and Ross Asset Mgt Corp Sally K. Ride President Imaginary Lines, Inc., and Hibben Professor of Physics

27 University of Michigan, Ann Arbor Teck H. Ho Results in various p-BC games

28 University of Michigan, Ann Arbor Teck H. Ho Summary  CH Model:  Discrete thinking steps  Frequency Poisson distributed  One-shot games  Fits better than Nash and adds more economic value  Sensible interpretation of mixed strategies  Can “solve” multiplicity problem  Application: Measurement of Strategic IQ

29 University of Michigan, Ann Arbor Teck H. Ho Research Agenda  Bounded Rationality in Markets  Revised Utility Functions  Empirical Alternatives to Nash Equilibrium (Ho, Lim, and Camerer, JMR, forthcoming)  A New Taxonomy of Games  Neural Foundation of Game Theory

30 University of Michigan, Ann Arbor Teck H. Ho Bounded Rationality in Markets: Revised Utility Function

31 University of Michigan, Ann Arbor Teck H. Ho Bounded Rationality in Markets: Alternative Solution Concepts

32 University of Michigan, Ann Arbor Teck H. Ho Neural Foundations of Game Theory  Neural foundation of game theory

33 University of Michigan, Ann Arbor Teck H. Ho Strategic IQ: A New Taxonomy of Games

34 University of Michigan, Ann Arbor Teck H. Ho

35 University of Michigan, Ann Arbor Teck H. Ho Nash versus CH Model: LL and MSD (in-sample)

36 University of Michigan, Ann Arbor Teck H. Ho Economic Value: Definition and Motivation  “A normative model must produce strategies that are at least as good as what people can do without them.” (Schelling, 1960)  A measure of degree of disequilibrium, in dollars.  If players are in equilibrium, then an equilibrium theory will advise them to make the same choices they would make anyway, and hence will have zero economic value  If players are not in equilibrium, then players are mis-forecasting what others will do. A theory with more accurate beliefs will have positive economic value (and an equilibrium theory can have negative economic value if it misleads players)

37 University of Michigan, Ann Arbor Teck H. Ho Alternative Specifications  Overconfidence:  k-steps think others are all one step lower (k-1) (Stahl, GEB, 1995; Nagel, AER, 1995; Ho, Camerer and Weigelt, AER, 1998)  “Increasingly irrational expectations” as K  ∞  Has some odd properties (e.g., cycles in entry games)  Self-conscious:  k-steps think there are other k-step thinkers  Similar to Quantal Response Equilibrium/Nash  Fits worse

38 University of Michigan, Ann Arbor Teck H. Ho

39 University of Michigan, Ann Arbor Teck H. Ho Example 3: Centipede Game Figure 1: Six-move Centipede Game

40 University of Michigan, Ann Arbor Teck H. Ho CH vs. Backward Induction Principle (BIP)  Is extensive CH (xCH) a sensible empirical alternative to BIP in predicting behavior in an extensive-form game like the Centipede?  Is there a difference between steps of thinking and look-ahead (planning)?

41 University of Michigan, Ann Arbor Teck H. Ho BIP consists of three premises  Rationality: Given a choice between two alternatives, a player chooses the most preferred.  Truncation consistency: Replacing a subgame with its equilibrium payoffs does not affect play elsewhere in the game.  Subgame consistency: Play in a subgame is independent of the subgame’s position in a larger game. Binmore, McCarthy, Ponti, and Samuelson (JET, 2002) show violations of both truncation and subgame consistencies.

42 University of Michigan, Ann Arbor Teck H. Ho Truncation Consistency VS Figure 1: Six-move Centipede game Figure 2: Four-move Centipede game (Low-Stake)

43 University of Michigan, Ann Arbor Teck H. Ho Subgame Consistency VS Figure 1: Six-move Centipede game Figure 3: Four-move Centipede game (High-Stake (x4))

44 University of Michigan, Ann Arbor Teck H. Ho Implied Take Probability  Implied take probability at each stage, p j  Truncation consistency: For a given j, p j is identical in both 4-move (low-stake) and 6-move games.  Subgame consistency: For a given j, p n-j (n=4 or 6) is identical in both 4-move (high-stake) and 6-move games.

45 University of Michigan, Ann Arbor Teck H. Ho Prediction on Implied Take Probability  Implied take probability at each stage, p j  Truncation consistency: For a given j, p j is identical in both 4-move (low-stake) and 6-move games.  Subgame consistency: For a given j, p n-j (n=4 or 6) is identical in both 4-move (high-stake) and 6-move games.

46 University of Michigan, Ann Arbor Teck H. Ho Data: Truncation & Subgame Consistencies

47 University of Michigan, Ann Arbor Teck H. Ho K-Step Look-ahead (Planning) V1V2V1V2 Example: 1-step look-ahead

48 University of Michigan, Ann Arbor Teck H. Ho Limited thinking and Planning  X k ( k ), k=1,2,3 follow independent Poisson distributions  X 3 =common thinking/planning; X 1 =extra thinking, X 2 =extra planning  X (thinking) =X 1 +X 3 ; Y (planning) =X 2 +X 3 follow jointly a bivariate Poisson distribution BP( 1, 2, 3 )

49 University of Michigan, Ann Arbor Teck H. Ho Estimation Results  Thinking steps and steps of planning are perfectly correlated

50 University of Michigan, Ann Arbor Teck H. Ho Data and xCH Prediction: Truncation & Subgame Consistencies