Identifying consensus analysts' earnings change forecasts with correct and incorrect signs

This paper examines whether users of consensus analysts' earnings forecasts can exploit information in (a) the nature of analysts' historical forecast accuracy and firms' earnings predictability, and (b) fundamental analysis of firms' earnings growth to assess the likelihood that the consensus forecast correctly predicts the direction of the forthcoming annual earnings change. I make this assessment by developing two models to condition the consensus forecast. First, I develop a model that indicates the likelihood that the consensus analyst forecast correctly predicts the directional earnings change based on analyst characteristics associated with forecast accuracy and characteristics of the firm's information environment. Second, I develop a model to indicate the likely sign of the forthcoming earnings change using fundamental analysis. I find the conditioned forecasts identified as more likely correct are significantly more likely to correctly predict the sign of the annual change in earnings while the conditioned forecasts identified as less likely correct are significantly less likely to correctly predict the future sign of the earnings change. I also examine the abnormal returns generated by trading strategies designed to exploit the information in the likelihood assessment. I find that implementing trading strategies based on the conditioned forecasts generates significant abnormal returns, which indicates the economic significance of relying on conditioned forecasts.