Volatility forecast of stock indices by model averaging using high-frequency data

GARCH-class models provide good performance in volatility forecasts. In this paper, we use realized GARCH (RGARCH), HEAVY (high-frequency-based volatility), and MEM (multiplicative error model) models to forecast one-day volatility of Chinese and Japanese stock indices. Forecast series from each are computed and the results compared to see which performs the best. To explore the possibility of better predictions, we combine the models by a model-averaging technique. In the empirical analysis, the CSI 300 and the Nikkei 225 are employed. We implement rolling estimation and evaluate the forecast performance by the superior predictive ability (SPA) test. As a result, we found that the proposed combination methods provided significant improvement in the forecast performance.