Prediction of Chinese stock volatility: Harnessing higher-order moments information of stock and futures markets

Abstract

This paper examines the predictive capacity of higher-order moments (skewness and kurtosis) of the Chinese stock index and futures market for the realized volatility of the stock market. Owing to the model uncertainty caused by structural changes, we propose the use of data-driven combination forecasting, namely, the LASSO-weighted average windows method over forecasts of long short-term memory network (LSTM), support vector regression (SVR), or the ordinary least squares (OLS) method. Empirical findings indicate that the LSTM method outperforms both SVR and OLS. The LASSO-weighted forecasts across these three methods significantly enhance the predictive ability of individual methods. The realized higher-order moments of both markets can effectively increase the prediction accuracy of stock market volatility, with the higher-order moments in the stock market contributing more than those in index futures.