The Bias in Analysts' Earnings Forecasts and the Cross Section of Stock Returns: An Empirical Investigation

We find that the bias in analysts' earnings forecasts for the current year and the next year comes down by 50% when special items are removed from "earnings before discontinued operations and extraordinary items" before comparing them with corresponding forecasts, i.e., when forecasts are compared with Compustat data item 20 - (data item 17)(1 - tax rate) and not with data item 20 as is commonly done in the literature. The systematic relation between some firm characteristics like size, price momentum, accruals, and trading volume and the bias in analysts' earnings forecasts is also substantially reduced with our definition of earnings. However, the cross section of forecast errors can still be predicted using firm characteristics even after controlling for persistence in forecast errors over time. Past error in forecasts and long run growth forecast, which are correlated with the cross section of corresponding future forecast errors, do not help predict the cross section one year ahead stock returns, whereas firm characteristics do. In contrast, firms with higher special items earn significantly lower returns when compared to firms with lower special items in the cross section, after controlling for other firm characteristics. Our findings are consistent with the view that stock prices take into account the biases in analysts' earnings forecasts to some extent, and investors' learning about the information in special items relevant for valuing a firm over time.