Distinguishing Overconfidence from Rational Best-Response in Markets

This paper studies the causal effect of individuals' overconfidence and bounded rationality on asset markets. To do that, we combine a new market mechanism with an experimental design, where (1) players' interaction is centered on the inferences they make about each others' information, (2) overconfidence in private information is controlled by the experimenter (i.e., used as a treatment), and (3) natural analogs to prices, returns and volume exist. We find that in sessions where subjects are induced to be overconfident, volume and price error analogs are higher than predicted by the fully-rational model. However, qualitatively similar results are obtained in sessions where there is no aggregate overconfidence. To explain this, we suggest an alternative possibility: participants strategically respond to the errors contained in others' actions by rationally discounting the informativeness of these actions. Estimating a structural model of individuals' decisions that allows for both overconfidence and errors, we are able to separate these two channels. We find that a substantial fraction of excess volume and price error analogs is attributable to strategic response to errors, while the remaining is attributable to overconfidence. Further, we show that price analog exhibit serial autocorrelation only in the overconfidence-induced sessions.