ÿÈÕ³Ô¹Ï

We test the theory put forth by Compte and Postlewaite (2004) that overconfidence might persist because it is welfare improving. They argue that because confidence enhances performance, some overconfidence is optimal in spite of its negative effect on decision-making. One implication of their model is that while an agent's bias (first moment of prediction error) may not change as she gains experience in an activity, her predictive accuracy (second moment of prediction error) should improve. We test this implication by comparing predictions of success by university debaters with outcomes in debate rounds and evaluating how the first and second moments of their prediction errors change with experience. As predicted by the theory, we find that while debaters remain overconfident in spite of experience, they become more accurate in their predictions. These findings support the view that overconfidence may persist because it is welfare improving..