Step-level reasoning and bidding in auctions

Step-level models of reasoning (SLR) proved to be very successful in predicting behavior in the beauty contest game. Recently, a quantified version of the model was suggested as a more general model of thinking. In particular, it was found that the distribution of choices could be represented by a Poisson distribution. I test the model in stylized first- and second-price common-value sealed-bid auctions. Equilibrium, for both auction types, prescribes that players undercut each other and profits are small. The SLR prediction, on the other hand, is different for the two auctions. Nash equilibrium predicts the outcomes poorly; the SLR model predicts the outcomes well in the second-price auction. However, while bids in the first-price auction could be represented by a Poisson distribution, this could not be attributed to step-level reasoning.