Optimal Allocation with Peer Information

We study allocation problems without monetary transfers where agents have correlated types, i.e., hold private information about one another. Such peer information is relevant in various settings, including science funding, allocation of targeted aid, or intra-firm allocation. Incentive compatibility requires that agents cannot improve their own allocation by misrepresenting the merits of allocating to others. We characterize optimal incentive-compatible mechanisms using techniques from the theory of perfect graphs. Optimal mechanisms improve on review panels commonly observed in practice by eliciting information directly from eligible agents and by using allocation lotteries to alleviate incentive constraints. Computational hardness results imply that exactly optimal mechanisms are impractically complex. We propose ranking-based mechanisms as a viable alternative and show that they are approximately optimal when agents are informationally small, i.e., when no single agent has information that is crucial for evaluating a large fraction of the other agents.