On the Communication Complexity of Expected-Prof it Maximization

Models of the behavior of the firm under uncertainty typically assume that the joint probability distribution over all random variables is common knowledge, and only the realized values of some variables are observed differentially. The main result of this paper shows that if the knowledge of the relevant probabilities is initially decentralized, then expected-profit maximizing decisions can require unbounded communication, as measured by the minimal message space dimension. The result is obtained in a simple model of a two-period, one product firm with two managers who must communicate in order to determine the level of inventory that maximizes expected profit. Journal of Economic Literature Classification Numbers: D21, D82, D92.