Anxiety and Pro-cyclical Risk Taking with Bayesian Agents

We provide a model that can explain empirically relevant variations in confidence and risk taking by combining horizon-dependent risk aversion (“anxiety”) and selective memory in a Bayesian intrapersonal game. In the time series, overconfidence is more prevalent when actual risk levels are high, while underconfidence occurs when risks are low. In the cross section, more anxious agents are more prone to biased confidence and their beliefs fluctuate more. This systematic variation in confidence levels can lead to objectively excessive risk taking by “insiders” with the potential to amplify boom-bust cycles