Bad Reputation Under Bounded and Fading Memory

Abstract

I relax the full memory assumption in Ely and Valimaki's (2003) mechanic game, where reputation is bad for all players. First I consider "bounded memory," where only finitely many recent periods are observed. For long memory, reputation is still bad. Shortening memory avoids bad reputation but only by making it "useless." There is no "happy middle:" reputation is either useless or reduces equilibrium payoffs for any memory length. I find a qualitatively different result for "fading memory," where players randomly sample past periods with probabilities "fading" toward zero. Unlike bounded memory, reputation is not bad but remains useful under sufficiently fast fading. This result extends to a more general class of both good and bad reputation games, suggesting reputation leaves long-run player behavior unaffected in some realistic word-of-mouth environments.