Forecasting the volatility of crude oil futures using HAR-type models with structural breaks

Abstract

We introduce sixteen HAR-type volatility models with structural breaks and estimate their parameters by applying 5-min high-frequency transaction data for WTI crude oil futures. We find significant structural breaks in the volatility of crude oil futures. Additionally, the historical realized volatility, continuous sample path variation, negative realized semivariance, signed jump, signed semi-jump and leverage components contain substantial and salient information for forecasting the volatility of crude oil futures. Then, we use loss functions to assess the forecasting performance of these sixteen new models, and finally, rank these models using the PROMETHEE II method. Our results indicate that different models exhibit different predictive power in forecasting the 1-day, 1-week and 1-month volatility of crude oil futures. Of the new HAR-type models, the new HAR-RSV model performs best at forecasting the 1-day and 1-month volatilities, whereas the new HAR-CJ best forecasts the 1-week volatility.