Forecasting Earnings Variance : Quantile-Based Vs. Residuals-Based Variance Proxies

The aim of this study is to analyze different forecasting approaches for the variance of future earnings, compare the respective forecast accuracy and test whether the forecasted information are relevant to equity or debt markets. The results, in line with former research, indicate that quantile-based variance proxies outperform resiudal-based variance proxies in industry-level tests. However, a residual-based variance proxy outperforms the established quantile-based variance proxies in terms of forecast accuracy in firm-level tests. This study finds that the opposing performance outcome appears to be driven by the proneness of quantile-based variance proxies to produce comparably large forecasts in the extreme percentiles whose influence is masked in the industry-, but not in the firm-level test. Further, this study finds the book-value of equity to be an important predictor for the future earnings variance overlooked in previous studies. Finally, it confirms findings from former studies that equity prices are increasing in the variance of future earnings, whereas there is no significant relationship between bond ratings and the variance of future earnings