Predicting Sell-Side Analysts' Relative Earnings Forecast Accuracy When It Matters Most

Abstract

We introduce a novel framework to predict the relative accuracy of sell-side analysts’ annual earnings forecasts out-of-sample. Prior studies only evaluate forecasts shortly before the corresponding earnings release. In contrast, our study is the first to provide long-term predictions which are of particular value for both investors and academics. Overall, we show that analysts classified as superior outperform their inferior counterparts by 8.4 percent, on average. The prediction performance is even more pronounced for longer-term forecasts and for firms with high dispersion of analysts’ forecasts, that is, when the identification of superior forecasts matters most. Moreover, we challenge the conclusion of existing literature that characteristics reflecting an analyst’s skill set are not helpful to obtain better predictions. In particular, when evaluating forecasts which draw on similar information sets, we find that a model based on analyst characteristics outperforms a model focusing simply on the forecast horizon, for example.