Belief Formation Under Signal Correlation

This paper demonstrates the importance of simultaneously considering two behavioral biases, correlation neglect and overprecision, in characterizing belief formation. Our laboratory experiments reveal that, relative to independent signals, subjects overvalue moderately or strongly correlated signals, but undervalue weakly correlated signals. We explain these seemingly inconsistent findings by developing a model that concurrently incorporates both biases and separately identifies their effects. Estimated model parameters suggest that subjects show a nearly complete level of correlation neglect and a high level of overprecision. Moreover, they do not fully appreciate the variance reducing power of averages and undervalue aggregated market information