Signal Correlation and Belief Formation
45We characterize how people aggregate independent and correlated signals to form their belief about the characteristics of an uncertain event under a rich setting. We show that while people do understand that correlated signals are generally not as useful as independent signals, they do not optimally account for correlation in forming their beliefs. In our lab experiment, subjects observe a set of independently drawn signals and a set of signals that are correlated with each other about the realization of a random variable. They are completely aware of how these signals are generated. We elicit their belief using a scoring rule that is incentive compatible independent on a subject's utility function. We find that subjects form their beliefs in an unbiased, yet suboptimal way. Specifically, when the correlated signals are moderately or strongly correlated, they undervalue independent signal. On the other hand, when the correlation is weak, the subjects overvalue the independent signals. The same pattern is observed when subjects choose the combination of independent and correlated signals they receive. Finally, we can measure whether subjects are more confident about their ability to form beliefs more optimally than other subjects in the room. We find that subjects believe that their own predictions have lower variances than those by other subjects by a factor of more than 10.