We examine whether attribution bias leads managers who have experienced short-term forecasting success to become overconfident in their ability to forecast future earnings. Importantly, this form of overconfidence is endogenous and dynamic. We also examine the effect of this cognitive bias on the managerial credibility. Consistent with the existence of dynamic overconfidence, managers who have predicted earnings accurately in the previous four quarters are less accurate in their subsequent earnings predictions. These managers also display greater divergence from the analyst consensus and are more precise. Lastly, investors and analysts react less strongly to forecasts issued by overconfident managers.