Belief formation under signal correlation

Abstract

This paper demonstrates the importance of considering incorrect perceptions of variance in characterizing people's belief formation rules when they suffer from correlation neglect. Using laboratory experiments, we elicit posterior beliefs regarding the realization of a random variable where some signals that a subject receives are correlated. Subjects in our experiments consistently neglect correlation and also misperceive signal variance when updating their beliefs. We present behavioral models that illustrate how people who suffer from both biases overvalue strongly correlated signals and undervalue weakly correlated signals relative to Bayesian predictions. Additional experiments provide strong support for such a nuanced impact of correlation neglect on belief updating rules. The estimated model parameters are consistent with subjects showing high levels of correlation neglect and undersensitivity to increases in the true variance. Moreover, we find significant heterogeneity across subjects, with a majority exhibiting correlation neglect combined with variance misperception.