Man vs. Machine Learning: The Term Structure of Earnings Expectations and Conditional Biases

We use machine learning to construct a statistically optimal and unbiased benchmark for firms' earnings expectations. We show that analyst expectations are on average biased upwards, and that this bias exhibits substantial time-series and cross-sectional variation. On average, the bias increases in the forecast horizon, and analysts revise their expectations downwards as earnings announcement dates approach. We find that analysts' biases are associated with negative cross-sectional return predictability, and the short legs of many anomalies consist of firms for which the analysts' forecasts are excessively optimistic relative to our benchmark. Managers of companies with the greatest upward biased earnings forecasts are more likely to issue stocks.