Does Past Success Lead Analysts to Become Overconfident?

This paper provides evidence that analysts who made a sequence of accurate predictions of earnings relative to the median forecast tend to be relatively more inaccurate and more out of consensus in their subsequent earnings prediction. This phenomenon is economically and statistically meaningful. The results are robust to different estimation techniques and different control variables. Our findings are consistent with an attribution bias that leads accurate analysts to become too confident in the quality of their information. However, we also offer alternative explanations that would rationally fit the data.