Peer-induced Fairness in Games Teck H. Ho University of California, Berkeley (Joint Work with Xuanming Su) October, 2009 Teck H. Ho 1 Outline Motivation Distributive versus Peer-induced Fairness The Model Equilibrium Analysis and Hypotheses Experiments and Results March, 2009 Teck H. Ho 2 Dual Pillars of Economic Analysis Specification of Utility Only final allocation matters Self-interest Exponential discounting Solution Method Nash equilibrium and its refinements (instant equilibration) March, 2009 Teck H. Ho 3 Motivation: Utility Specification Reference point matters: People care both about the final allocation as well as the changes with respect to a target level Fairness: John cares about Mary’s payoff. In addition, the marginal utility of John with respect to an increase in Mary’s income increases when Mary is kind to John and decreases when Mary is unkind Hyperbolic discounting: People are impatient and prefer instant gratification March, 2009 Teck H. Ho 4 Motivation: Solution Method Nash equilibrium and its refinements: Dominant theories in marketing for predicting behaviors in non-cooperative games. 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 and infinitely repeated games). Modeling subject heterogeneity really matters in games. March, 2009 Teck H. Ho 5 Bounded Rationality in Markets: Revised Utility Function Behavioral Regularities Standard Assumption New Model Specification Reference Example Marketing Application Example 1. Revised Utility Function - Reference point and loss aversion - Expected Utility Theory - Prospect Theory - Ho and Zhang (2008) Kahneman and Tversky (1979) - Fairness - Self-interested - Inequality aversion Fehr and Schmidt (1999) - Cui, Raju, and Zhang (2007) - Impatience - Exponential discounting - Hyperbolic Discounting Ainslie (1975) - Della Vigna and Malmendier (2004) Ho, Lim, and Camerer (JMR, 2006) March, 2009 Teck H. Ho 6 Bounded Rationality in Markets: Alternative Solution Methods Behavioral Regularities New Model Specification Example Standard Assumption Marketing Application Example 2. Bounded Computation Ability - Nosiy Best Response - Best Response - Quantal Best Response McKelvey and Palfrey (1995) - Lim and Ho (2008) - Limited Thinking Steps - Rational expectation - Cognitive hierarchy Camerer, Ho, Chong (2004) - Goldfrad and Yang (2007) - Myopic and learn - Instant equilibration - Experience weighted attraction Camerer and Ho (1999) - Amaldoss and Jain (2005) March, 2009 Teck H. Ho 7 Modeling Philosophy Simple General Precise Empirically disciplined (Economics) (Economics) (Economics) (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) March, 2009 Teck H. Ho 8 Outline Motivation Distributive versus Peer-induced Fairness The Model Equilibrium Analysis and Hypotheses Experiments and Results March, 2009 Teck H. Ho 9 Distributive Fairness March, 2009 Teck H. Ho 10 Ultimatum Game Yes? No? Split pie accordingly Both get nothing March, 2009 Teck H. Ho 11 Empirical Regularities in Ultimatum Game Proposer offers division of $10; responder accepts or rejects Empirical Regularities: There are very few offers above $5 Between 60-80% of the offers are between $4 and $5 There are almost no offers below $2 Low offers are frequently rejected and the probability of rejection decreases with the offer Self-interest predicts that the proposer would offer 10 cents to the respondent and that the latter would accept March, 2009 Teck H. Ho 12 Ultimatum Experimental Sites Henrich et. al (2001; 2005) March, 2009 Teck H. Ho 13 Ultimatum Offers Across 16 Small Societies (Mean Shaded, Mode is Largest Circle…) Mean offers Range 26%-58% March, 2009 Teck H. Ho 14 Modeling Challenges & Classes of Theories The challenge is to have a general, precise, psychologically plausible model of social preferences Three major theories that capture distributive fairness Fehr-Schmidt (1999) Bolton-Ockenfels (2000) Charness-Rabin (2002) March, 2009 Teck H. Ho 15 A Model of Social Preference (Charness and Rabin, 2002) Blow is a general model that captures both classes of theories. Player B’s utility is given as: U B ( A , B ) ( r s ) A (1 r s) B where r 1 if B A , and r 0 otherwise; s 1 if B A , and s 0 otherwise. B’s utility is a weighted sum of her own monetary payoff and A’s payoff, where the weight places on A’s payoff depend on whether A is getting a higher or lower payoff than B. March, 2009 Teck H. Ho 16 Peer-induced Fairness March, 2009 Teck H. Ho 17 Distributional and Peer-Induced Fairness peer-induced fairness March, 2009 Teck H. Ho 18 A Market Interpretation posted price SELLER posted price take it or leave it? March, 2009 BUYER peer-induced fairness BUYER 19 Teck H. Ho Examples of Peer-Induced Fairness Price discrimination (e.g., iPhone) Employee compensation (e.g., your peers’ pay) Parents and children (favoritism) CEO compensation (O’Reily, Main, and Crystal, 1988) Labor union negotiation (Babcock, Wang, and Loewenstein, 1996) March, 2009 Teck H. Ho 20 Social Comparison Theory of social comparison: Festinger (1954) One of the earliest subfields within social psychology Handbook of Social Comparison (Suls and Wheeler, 2000) WIKIPEDIA: http://en.wikipedia.org/wiki/Social_comparison_theory March, 2009 Teck H. Ho 21 Outline Motivation Distributive versus Peer-induced Fairness The Model Equilibrium Analysis and Hypotheses Experiments and Results March, 2009 Teck H. Ho 22 Modeling Differences between Distributional and Peer-induced Fairness 2-person versus 3-person Reference point in peer-induced fairness is derived from how a peer is treated in a similar situation 1-kink versus 2-kink in utility function specification People have a drive to look to their peers to evaluate their endowments March, 2009 Teck H. Ho 23 The Model Setup 3 Players, 1 leader and 2 followers Two independent ultimatum games played in sequence The leader and the first follower play the ultimatum game first. The second follower receives a noisy signal about what the first follower receives. The leader and the second follower then play the second ultimatum game. Leader receives payoff from both games. Each follower receives only payoff in their respective game. March, 2009 Teck H. Ho 24 Revised Utility Function: Follower 1 The leader divides the pie: ( s1 , s1 ) Follower 1’s utility is: s1 max{ 0, ( s1 ) s1} if a1 1. U F1 ( s1 , a1 ) if a1 0. 0, Follower 1 does not like to be behind the leader (B > 0) March, 2009 Teck H. Ho 25 Revised Utility Function: Follower 2 Follower 2 believes that Follower 1 receives ŝ1 The leader divides the pie: ( s2 , s2 ) Follower 2’s utility is: ˆ ˆ U F 2 (s2 , a2 | z ) s2 max{ 0, ( s2 ) s2 } - p( z ) max{0, s1 (z) - s 2 } if a2 1. if a2 0. 0, Follower 2 does not like to be behind the leader ( > 0) and does not like to receive a worse offer than Follower 1 ( > 0) March, 2009 Teck H. Ho 26 Revised Utility Function: The Leader The leader receives utilities from both games In the second ultimatum game: s2 max{ 0, s2 ( s2 )} U L, II ( s2 , a2 | z ) 0, if a2 1. if a2 0. In the first ultimatum game: s1 max{ 0, s1 ( s1 )} U L, I ( s1 , a1 ) 0, if a1 1. if a1 0. Leader does not like to be behind both followers March, 2009 Teck H. Ho 27 Hypotheses Hypothesis 1: Follower 2 exhibits peer-induced fairness. That is, > 0. Hypothesis 2: If > 0, The leader’s offer to the second follower depends on Follower 2’s expectation of what the first offer is. That is, s2* f (sˆ1 | 0) March, 2009 Teck H. Ho 28 Economic Experiments Standard experimental economics methodology: Subjects’ decisions are consequential 75 undergraduates, 4 experimental sessions. Subjects were told the following: Subjects were told their cash earnings depend on their and others’ decisions 15-21 subjects per session; divided into groups of 3 Subjects were randomly assigned either as Leader or Follower 1, or Follower 2 The game was repeated 24 times The game lasted for 1.5 hours and the average earning per subject was $19. March, 2009 Teck H. Ho 29 29 Sequence of Events Ultimatum Game 2 Leader : Follower 2 Ultimatum Game 1 Leader : Follower 1 Noise Generation Uniform Noise March, 2009 Teck H. Ho 30 Subjects’ Decisions Leader s1 to Follower 1 s2 to Follower 2 after observing the random draw X (-20, - 10, 0, 10, 20) Follower 1 Accept or reject a1 Follower 2 ŝ1 (i.e., a guess of what s1 is after observing s1 X) Accept or reject a 2 Respective payoff outcomes are revealed at the end of both games March, 2009 Teck H. Ho 31 Hypotheses Hypothesis 1: Follower 2 exhibits peer-induced fairness. That is, > 0. Hypothesis 2: If > 0, The leader’s offer to the second follower depends on Follower 2’s expectation of what the first offer is. That is, s2* f (sˆ1 | 0) (Proposition 1) March, 2009 Teck H. Ho 32 Tests of Hypothesis 1: Follower 2’s Decision Being Ahead On Par Being Behind N Number of Rejection N Number of Rejection N Number of Rejection 165 ? 110 ? 179 ? March, 2009 Teck H. Ho 33 Tests of Hypothesis 1: Follower 2’s Decision Being Ahead On Par Being Behind N Number of Rejection N Number of Rejection N Number of Rejection 165 6 (3.6%) 110 5 (4.5%) 179 42 (23.5%) March, 2009 Teck H. Ho 34 Tests of Hypothesis 1: Logistic Regression Follower 2’s utility is: ˆ ˆ U F 2 (s2 , a2 | z ) s2 max{ 0, ( s2 ) s2 } - p( z ) max{0, s1 (z) - s 2 } if a2 1. if a2 0. 0, Probability of accepting is: ˆ2 0.024 ( p 0.05) March, 2009 Teck H. Ho 35 Test of Hypothesis 2: Second Offer vis-à-vis the Expectation of the First Offer On Par Being Behind March, 2009 Teck H. Ho Being Ahead 36 Tests of Hypothesis 2: Simple Regression The theory predicts that That is, we have s2 is piecewise linear in ŝ1 1 0 ˆ1 0.09 ( p 0.01) March, 2009 Teck H. Ho 37 Implication of Proposition 1: S2* > S1* Method 1: Each game outcome involving a triplet in a round as an independent observation Wilcoxon signed-rank test (p-value = 0.03) Method 2: Each subject’s average offer across rounds as an independent observation Compare the average first and second offers Wilcoxon signed-rank test (p-value = 0.04) March, 2009 Teck H. Ho 38 Structural Estimation The target outlets are economics journals We want to estimate how large is compared to (important for field applications) Is self-interested assumption a reasonable approximation? Understand the degree of heterogeneity March, 2009 Teck H. Ho 39 Is Self-Interested Assumption a Reasonable Approximation? No March, 2009 Teck H. Ho 40 Is Peer-Induced Fairness Important? YES March, 2009 Teck H. Ho 41 Latent-Class Model The population consists of 2 groups of players: Self-interested and fairness-minded players The proportion of fairness-minded * s See paper for Propositions 5 and 6: 2 depends on March, 2009 Teck H. Ho 42 Is Subject Pool Heterogeneous? 50% of Subjects are Fairness-minded 43 Teck H. Ho Model Applications Price discrimination Executive compensation Union negotiation March, 2009 Teck H. Ho 44 Price Discrimination Di ( pi ) Ai pi March, 2009 Teck H. Ho i L, H 45 Summary Peer-induced fairness exists in games Leader is strategic enough to exploit the phenomenon Peer-induced fairness parameter is 2 to 3 times larger than distributional fairness parameter 50% of the subjects are fairness-minded 46 Teck H. Ho Standard Assumptions in Equilibrium Analysis Assumptions Nash Equilbirum Cognitive Hierarchy QRE EWA Learning Strategic Thinking X X X X Best Response X X Mutual Consistency X Solution Method Instant Convergence March, 2009 X X X Teck H. Ho X 47