The Neglect of Correlation in Allocation Decisions

We study the effect of variation in correlation on investment decision in an experimental two asset application. Comparison of allocations across problems suggests that subjects neglect probabilistic information on the joint distribution of returns and base their allocations on the observed return levels for the two assets. When asked to predict future returns, subjects try to replicate the historical distribution, thereby falling into the probability-matching bias. Predictions drastically vary when correlations become negative, while allocations are not significantly affected by changes in sign of correlation. The observed allocation patterns contradict the predictions of standard models of choice; the inconsistency is attributed to common behavioral bias in financial decision. Field implications of the results are discussed.