On the Persistence of Overconfidence: Evidence from Multi-Unit Auctions

Abstract

We analyze pre and post-task confidence in an experiment in which subjects bid in multi-unit common value auctions. Subjects return for a second session, so we are able to assess how performance affects the evolution of confidence. Those with low confidence prior to the first session underestimate performance while those with high confidence overestimate performance. Although the change in pre-experiment confidence from session one to session two is close to zero, the dispersion in confidence increases. For those with moderate initial confidence, the change in confidence depends significantly on performance in session one. For those with high initial confidence, the change in confidence does not depend on performance, and the correlation between confidence prior to session two and confidence prior to session one is significantly higher than for those with neutral or low confidence. Subjects with high initial confidence also base their perception of post-experiment relative performance primarily on pre-experiment confidence: an effect not present in the moderate and low confidence groups. Based on a pre-experiment survey, we also find that those with high initial confidence are more likely to have prior experience trading stocks or options.

Keywords: Overconfidence, better than average effect, perceptions of performance, miscalibration, divisible good auctions

JEL Classification: C91, C92, D44, D82