Consider a long-term auctioneer who repeatedly sells identical or similar items and who might disclose selective information about past bidding. We present a theory that yields different predictions about bidding behavior depending on the information bidders are provided with, and then test it using a lab experiment. We focus on the disclosure of all bids and of winning bids only. Our theory is based on a selection bias: some of the bidders who are presented with historical winning bids mistakenly best-respond to that distribution, failing to realize that winning bids are not representative of all bids. In the steady state, this bias results in higher bids and auction revenue in comparison to the case when all bids are presented. Our experimental test confirms the qualitative predictions of the theory. On the theory side, our findings challenge the predictive power of Bayesian Nash Equilibrium based on rational bidders. On the market design side, they underline the role of historical market information as a key design choice.
Keywords: auctions, bidding, feedback, selection bias, mechanism design.
JEL Classification: C91, C92, D44.