Chronic Bias in Earnings Forecasts

Whatever the source or explanation of bias in forecasts of company earnings, if such bias persists, it is potentially discoverable and exploitable by investors. This research addresses (1) whether characterizing forecasts as if they were a homogeneous group with respect to bias is accurate or useful and (2) whether a long-term record of forecast errors contains information useful in predicting subsequent errors. We found that earnings forecasts are heterogeneous with respect to direction and degree of bias. We also found evidence of extremes in optimism and pessimism and that extreme errors tend to persist in the same direction, which suggests certain potentially profitable trading strategies. Whatever the source or explanation of bias in forecasts of a company's earnings, if such bias persists long enough, it is potentially discoverable and exploitable by investors. “Exploitation” in this context implies that investors, through examination of historical forecasting performance, can more or less reliably estimate the direction and extent of any bias and impute unbiased estimates for themselves. The absence of persistence in forecast errors would suggest that analysts' behavior ultimately self-corrects within a time frame that eliminates the possibility that the patterns can be exploited by investors.We addressed two primary issues related to bias in forecasts: (1) whether characterizing forecasts as if they were a homogeneous group with respect to bias is accurate or useful and (2) whether a long-term record of forecast errors contains information that would be useful in predicting subsequent errors.We devised a method for using historical data on forecast errors to separate earnings forecasts into classes (portfolios) based on “chronic” bias—the presence of a pattern of forecast errors that pervades long series of forecasts. The methodology for formation of portfolios used 20 quarters of past consensus forecast errors to identify the forecasts that have been at the extremes of optimism and pessimism. The portfolio formation algorithm, called the “mean—frequency forecast error” (MFFE) method, was based on mean forecast errors and the weighted-average frequency of negative errors.We applied several parametric and nonparametric tests to determine whether historical bias is persistent (i.e., whether it has predictive power with respect to subsequent forecast errors). The MFFE portfolio formation method resulted in a range of portfolios, with the observations classified as optimistic (pessimistic) having reliably negative (positive) mean contemporaneous forecast errors and a reliably higher percentage of negative (positive) forecast errors than the full sample. Furthermore, we found that the predictive power was incremental to that of the prior-period forecast error and persisted even when we controlled for the effects of unexpected earnings shocks related to time period, industry, or exchange listing and for company-specific factors.On the whole, our findings imply that analysts' behavior in both the optimistic and the pessimistic extremes does not entirely self-correct, leaving open the possibility that investors may find historical forecast errors useful in making inferences about current forecasts. Tests of a simple announcement-period trading strategy support the conclusion that the forecast bias identified by using the historical data on consensus forecasts may already be reflected in stock prices prior to earnings-announcement dates. Postannouncement drift, however, appears to be positive for optimistically biased forecasts and negative for pessimistically biased forecasts.