Forecasting Long-Horizon Factor Volatility

Abstract

This article investigates forecasts of long-term volatility for the fast-growing field of long–short factor strategies in an extensive in-sample and out-of-sample framework. The author follows previous work by empirically comparing various forecast configurations to provide guidance for academics and practitioners on how to form accurate predictions of future volatility for various established factors. The set spans 21 factor return time series over multiple asset classes, factor styles, and a long historical data period. Both in-sample and out-of-sample results suggest monotonically increasing forecast accuracy for longer historical lookback periods, longer forecasting windows, and more-sophisticated models (considering short-term volatility clustering and external predictors motivated by the asset-pricing literature), while the findings appear less pronounced in a real-time setting than observed in-sample. Moreover, investors engaging in carry-styled factor strategies and multifactor portfolios (rather than single factors) achieve more-reliable forecasts, on average, as confirmed by the out-of-sample analysis.