Improving the Precision of Analysts' Earnings Forecasts by Adjusting for Predictable Bias

Abstract

This research demonstrates that publicly-available information can be used to develop estimates of analysts' optimistic bias in earnings forecasts. These bias estimates can be used to produce more accurate forecasts, resulting in significant reductions of both cross-sectional mean forecast error and error variance. When bias estimates are based on past observations of forecast error alone, however, reductions in mean forecast error are smaller, and forecast precision is unimproved. Further tests provide evidence of a significant association between returns and the bias predictable from contemporaneously-available information, suggesting that predictable bias is only partially discounted by market participants. This study has significant implications for researchers and investors. The pricing of predictable bias in analysts' forecasts may add error toinferences which are based on the association between returns and analyst forecast errors, and knowledge of the market's partial discounting of predictable bias may help investors to make more efficient resource allocations.