Sophisticated bidders in beauty-contest auctions

We study bidding behavior by firms in beauty-contest auctions, i.e., auctions in which the winning bid is the one that gets closest to some function (average) of all submitted bids. Using a dataset on public procurement beauty-contest auctions, we show that firms' observed bidding behavior departs from equilibrium and can be predicted by a "sophistication" index, which captures the firms' capacity of bidding close to optimality in the past. We show that our empirical evidence is consistent with a Cognitive Hierarchy model of bidders' behavior. We also investigate whether and how firms learn to bid strategically through experience.