High frequency volatility of oil futures in China: Components, modeling, and prediction

This paper investigates the high-frequency volatility modeling and prediction for crude oil futures in China, a new asset class emerging in recent years. Two volatility measures, the realized variance ( $\mathrm{RV}$) and realized bi-power variations ( $\mathrm{RBV}$) are constructed at various frequencies by virtue of 1-minute crude oil futures prices. The distinctive components of these volatility estimators are further identified to exploit the information contents in the in-sample explanatory power of the realized variance dynamics and the out-of-sample prediction of realized variance across different horizons, leading to four new HAR-RV-type models. First, the empirical results show that the continuous component of the weekly realized variance, representing investors' trading behavior in the medium-term, is the dominant factor driving up volatility trends in China's crude oil futures market over a range of market conditions. Second, the monthly jump component in realized variance presents the significant in-sample explanatory power, and yet marginally improves prediction performance in realized variance during the two out-of-sample periods. Finally, these results are robust toward various market/model setups, over day- and night-trading hours, and across a range of prediction horizons and relative to prediction benchmarks.