Belief Formation Under Signal Correlation

Abstract

Using a set of incentivized laboratory experiments, we characterize how people form beliefs about a random variable based on independent and correlated signals. First, we theoretically show that, while pure correlation neglect always leads to overvaluing of correlated signals, that may not happen if people also exhibit overprecision perceiving signals to be more precise than they actually are. Our experimental results reveal that, while subjects do overvalue moderately or strongly correlated signals, they undervalue weakly correlated signals, suggesting concurrent presence of correlation neglect and overprecision. Estimated parameters of our model suggest that subjects show a nearly complete level of correlation neglect and also suffer from a high level of overprecision. Additionally, we find that subjects do not fully benefit from wisdom of the crowd-they undervalue aggregated information about othersi¯ actions in favor of their private information. This is consistent with models of overprecision where people do not properly incorporate the variance reducing power of averages.