An experimental study of estimation and bidding in common-value auctions with public information

Abstract

We study how subjects with identical public data first make estimates and then bid in common-value environments. The data presented rows of numbers and values associated with them by our (undisclosed) rule. Subjects were asked to estimate the missing value in the last row with only the numbers given, and then bid for that value in a second-price auction. There is no presumption of commonly-known distributions, yet we derive necessary conditions for equilibrium. The strong winner's curse that we observe in our data results from the dispersion of the value estimates and the poorly-chosen bid-strategies. We find that bidding is lower when bidders are more uncertain about the estimates they have made. Finally, the k-nearest-neighbor method does well in explaining the estimates of the (common) value.