Identification in Auction Models with Interdependent Costs

This paper provides a positive identification result for procurement models with asymmetric bidders, statistically dependent private information, and interdependent costs. When bidders are risk neutral, the model's payoff-relevant primitives are: (i) the joint distribution of private information and (ii) each bidder's full information expected cost--the expected cost conditional on own and competitors' information. These primitives are nonparametrically identified from the distribution of bids conditional on observable cost shifters under the following four assumptions. First, each bidder's private information can be summarized by a real-valued signal. Second, the joint distribution of bidders' signals does not depend on cost shifters. Third, each bidder's full-information cost depends on own cost shifters but not on competitors'. Fourth, conditional on covariates, the observed data are generated by the repeated play of the same equilibrium where bidders use monotone pure strategies. Using data from Highway Procurements in Michigan and exploiting variation in contractors' distance to each project, I estimate each major bidder's full-information costs. The estimates are used to evaluate alternative reserve prices and policies that reduce the severity of the winner's curse by restricting participation.