Ulric B. and Evelyn L. Bray Social Sciences Seminar

Abstract: In linear best-response games, players wish to minimize the distance between their action and an unknown target, a linear combination of others' actions and some exogenous uncertain state; the structure of the game is summarized by a network where players' connections are given by the influence of their action on others' targets. Imagine that, before actions are taken, players can acquire costly information to reduce the uncertainty they face. Players' incentive then is twofold: learning the payoff-relevant state and what others know. In this paper I separately characterize value of information about state and others' information. To disentangle incentive to learn the state from incentive to learn what others know, I study how players' behavior changes as more possibilities to correlate information become available. A lesson is that there is always complementarity in learning what others know, even if, for example, actions are substitutes. This complementarity, however, is weaker in diversified networks such as large networks of small players.