Forecasting oil price volatility using high-frequency data: New evidence

Abstract

In this article, we account for the conditional time-varying volatility of realized volatility to model and forecast oil futures price volatility based on the HAR-RV model and its various extensions. Our empirical results reveal several noteworthy observations. First, the in-sample results indicate that the residuals of the HAR-RV-type models exhibit a significant ARCH effect. Second, the out-of-sample results demonstrate that compared to the linear HAR-RV-type models, the HAR-RV-type models, including the FIGARCH structure models, can generally generate a higher forecast accuracy when forecasting short-term horizon volatility. Third, when predicting middle-term and long-term volatilities, the proposed model, i.e., HAR-S-RV-J-FIGARCH, can exhibit a higher predictive ability.