[UTMD-098] Bayesian Learning When Players Misspecify Others (by Takeshi Murooka, Yuichi Yamamoto) (Revised version of UTMD-085)

Takeshi Murooka, Yuichi Yamamoto

This paper considers Bayesian learning when players are biased about the data-generating process, and are biased about the opponent’s bias about the data-generating process. Specifically, we assume that each player’s bias about others takes the form of interpersonal projection, which is a tendency to overestimate the extent to which others share the player’s own view. We show that there is a class of games in which even an arbitrarily small amount of bias can destroy correct learning of an unknown state, i.e., there is zero probability of the posterior belief staying in a neighborhood of the true state.