Abstract
Mechanisms used to derive optimal allocations are typically designed assuming agents fully know their own preferences. It is often impossible to duplicate optimal allocations when agents imperfectly observe object characteristics. I present a crowdsourcing mechanism to approximate optimal allocations under imperfect observations. To ensure truthtelling, agents are punished when their reports differ from the “wisdom of the crowd.” Under mild conditions, this crowdsourcing-with-punishment mechanism replicates the full-information optimal allocation with probability exponentially converging to one in the market size, with small waste. No alternative mechanism meaningfully does better. The proposed mechanism can be applied in many settings, including matching markets.
