Forecasting realized volatility of agricultural commodity futures with infinite Hidden Markov HAR models

We construct a set of HAR models with three types of infinite Hidden Markov regime-switching structures. In particular, jumps, leverage effects, and speculation effects are all taken into account in the realized volatility modeling. We forecast five agricultural commodity futures (Corn, Cotton, Indica rice, Palm oil and Soybeans) based on high-frequency data from Chinese futures markets, and evaluate the forecast performances using both statistical and economic evaluation measures. The statistical evaluation results suggest that HAR models with infinite Hidden Markov regime-switching structures have better precision than the benchmark HAR models based on the MZ- R2, MAFE, and MCS results. The economic evaluation results suggest that portfolios constructed with infinite Hidden Markov regime-switching HARs also achieve higher portfolio returns for risk-averse investors than benchmark HAR model for short-term volatility forecasts.