On the performance of the Crémer-McLean auction : an experiment

The paradoxical full-surplus-extraction (FE) result, which can impair the mechanism design paradigm, is a long-standing concern in the literature. We tackle this problem by experimentally testing the performance of an FE auction, which is a second-price (2P) auction with lotteries. In the FE treatment, overbid amounts given entry increased and entry rates decreased through rounds, thus FE failed. By contrast, most subjects learned value bidding in the 2P treatment. To identify the causes of failure in the FE, we take an evolutionary-game approach. The FE auction with risk-neutral bidders has exactly two symmetric equilibria, either value bidding with full or partial entry, and only the partial-entry equilibrium is (evolutionarily or asymptotically) stable. Replicator dynamics with vanishing trends well explain observed dynamic bidding patterns. Together, these findings suggest that the FE outcome is not robust to trial-and-error learning by bidders.