Testing for overconfidence statistically: A moment inequality approach

We propose a moment inequality approach to test for the presence of overconfidence using data from ranking experiments where subjects rank themselves relative to other experimental participants. Although a ranking experiment is a typical way to collect data for the analysis of overconfidence, recent studies show that the resulting data may apparently indicate overconfidence even if participants are purely rational Bayesian updaters, in which case a set of inequalities hold. We apply state-of-the-art tests of moment inequalities to test such a set of inequalities. We examine the data from a traditional ranking experiment as well as those from more sophisticated designs.