Abstract: Daily volatility of Shanghai Stock Exchange Composite Index and Shenzhen Stock Exchange Component Index are decomposed into three components, which are the continuous sample-path variation, the discontinuous variation due to jumps and the overnight variance. Then HAR-CJN model is proposed to study the interaction of the three components and their impact on forecasting. The results show that the continuous variation has positive impact on each of the three components and contributes the most in forecasting, while the impact from jump variation is generally weaker than that from continuous variation and varies in direction and size as the length of lag-period changes. The out-of-sample forecast results show that HAR-CJN model outperforms traditional GARCH model considerably, and also outperforms the popular realized volatility model HAR-RV in the one-day-ahead and one-month ahead forecast.

Key words: realized volatility, jump, overnight variance, forecast