A Generalized Winner's Curse: An Experimental Investigation of Complexity and Adverse Selection

We investigate experimentally a possible Winner’s Curse in a generalized environment with pure common values. In our design, participants are given examples in which parameters lead to prize values (using an undisclosed and typically complex formula). They are then asked to estimate the prize value with a final set of parameter values and then bid for this prize value in a second-price auction. Unlike previous studies, two main ingredients ‘responsible’ for the creation of the winner’s curse are absent: 1) there is no presumption of common knowledge of a joint distribution, and 2) all signals are identical and public. Thus, it is difficult to discuss or compute a Bayes-Nash Equilibrium, let alone to measure individual departures from it. It is unclear what ought to be the behavioral norm, how to compute it and/or measure individual departures from it. It is also difficult, if not impossible, to talk about a posterior distribution conditional on winning as we don’t induce any priors. Assuming common knowledge in this context is patently implausible. In this generalized environment, we find strong evidence of the winner’s curse that is entirely consistent with previous findings from standard environments. We also find that the winner’s curse results from both dispersion in the estimates made and poorly-chosen bid-strategies. Finally, the nearest-neighbors method explains well how participants make estimates of the (common) value.