Realized volatility forecasting of agricultural commodity futures using the HAR model with time-varying sparsity

We develop a time-varying HAR model where both the predictors and the regression coefficients are allowed to change over time, and use it to forecast the realized volatility in the fast-growing agricultural commodity futures markets of China. The proposed model is constructed by incorporating all potential predictors in a time-varying HAR framework, and giving the independent normal-gamma autoregressive (NGAR) process priors to the regression coefficients. The out-of-sample forecast results show that the proposed HAR model with time-varying sparsity improves the forecast performances substantially relative to both the simple HAR model and more sophisticated HAR-type models in almost all cases. Finally, the forecast performance of the proposed model is robust to the alternative proxies of volatility.