Empirical evidence suggests that consumers' mutual fund purchase decisions depend on past performance. It is not clear, however, whether this behavior is rational. We take the experimental approach to analyze investors' behavior without confronting the measurement problems that empirical study faces. Our results show that investors respond irrationally to information on past performance. We detect two behavioral anomalies: ``Absolute Performance Effect'' asserting that investors' tendency to delegate money to a fund manager increases with past performance , even in cases where past performance conveys no information regarding ability; and ``Relative Performance Effect'' claiming that investors' tendency to delegate money to the manager is negatively correlated with the performance of the other manager, even in cases where the other managers' performance are due to luck per-se. We formally show that subjects' behavior is inconsistent with expected utility theory and suggest two alternative models: ``Subjective Conditional Probability'' stipulating that the probabilities assigned by decision makers to future events are different from those implies by Bayesian updating, and ``Subjective Risk Aversion'' asserting that the utility function is history dependent. Using GMM we show that the absolute and the relative performance effects are significant in both models.

Co-author Doron Kliger