Rationally Misplaced Confidence∗

I show that persistent underconfidence and overconfidence each arise from rational learning about one’s own abilities. If an agent chooses to exert more effort when more confident and believes that greater effort reliably improves outcomes, then the agent learns away overconfidence faster than he learns away underconfidence. The agent ends up underconfident on average. In contrast, the agent ends up overconfident on average if he believes that greater effort increases his exposure to chance. The results are consistent with evidence that underconfidence and overconfidence are both widespread in the population, and the mechanism is consistent with the modern understanding of depression.