The decentralization of information processing in the presence of interactions

We propose a model of organizational decision making, in which information processing is decentralized. Our model incorporates two features of many actual organizations: aggregation entails a loss of useful information, and the decision problems of different agents interact. We assume that an organization forms a portfolio of risky assets, following a hierarchical procedure. Agents' decision rules and the organization's hierarchical structure are derived endogenously. Typically, in the optimal hierarchical structure, all agents have one subordinate, and returns to ability are at least as high at the bottom as at the top. However, these results can be reversed in the presence of returns to specialization.