International oil shocks and the volatility forecasting of Chinese stock market based on machine learning combination models

Abstract

This paper aims to forecast the volatility of Chinese stock market under the effects of international crude oil shocks. Eight individual models, including multiple linear regression (MLR), least absolute shrinkage and selection operator (LASSO), support vector regression (SVR), artificial neural network (ANN), recurrent neural network (RNN), long short-term memory (LSTM) network, gated recurrent unit (GRU) and bidirectional gated recurrent unit (BiGRU) are constructed. The realized volatilities of the CSI 300 index and ten primary sector indices are taken as explained variables, respectively. Four oil shock indicators and the autoregressive terms of the realized volatilities are taken as explanatory variables. The SHAP method is used to analyze their effects on the stock indices. Based on eight individual models, four kinds of combination models, i.e., a mean combination (Mean), a median combination (Median), a trimmed mean combination (Trimmed Mean), and two discount mean squared forecasting error combinations (DMSPE (1) and DMSPE (0.9)) are proposed. We compare forecasting performance between combination and individual ones. Empirical results show that the effects of international crude oil shocks on Chinese stock market are significant and have strong predictability. The effects on the energy, industry, optional consumption, and public sectors are greater than those on the CSI 300 and other sectors. Most of the combination models can effectively improve forecasting accuracy. In addition, by changing the benchmark model, the lengths of the rolling window, and the historical lengths of oil shock indicators, we find that most of the combination models are robust in volatility forecasting. This study is of guiding significance for individual and institutional investors to understand the operating mechanism of Chinese stock markets.