Do Analysts’ Forecasts Predict their own Forecast Error? Evidence from the Horizon of Forecasted Earnings

• In Event: Trading, Valuation and Market Efficiency

Abstract

We show that analysts incorporate less negative news into long horizon forecasts than short- horizon forecasts, leading the precision of long-horizon forecasts relative to short-horizon forecasts to vary with the sign of news. Specifically, we show analysts revise longer horizon forecasts less than short-horizon forecasts in response to negative news, and conditional on a revision analysts’ long-horizon forecasts under-react more to negative news. We show this under-reaction drives the optimistic bias and inaccuracy in long-horizon forecasts, explaining the curious phenomenon that analyst forecasts at longer horizons are actually less accurate than simple time-series models. We propose a regression model to re-weight analyst forecasts at each horizon as a flexible function of the information precision – short-horizon forecasts receive more weight, particularly when past news is bad. We show this method achieves accuracy gains on the order of 14%. Adjusting forecasts for the predictable errors we identify generates forecasts of earnings more in line with market expectations than published forecasts, suggesting our methodology helps to measure market expectations of earnings. We contribute to the literature by linking the biases in the flow of information in forecasts (i.e. under-reaction) to biases in the level of forecasts (i.e. over-optimism).

Authors

• Charles Ham, Washington University

• Zachary R Kaplan, Washington University

• Zawadi Rehema Lemayian, Washington University