Forecasting of Chinese stock price using a hybrid neural network model

Against the backdrop of economic globalization, the outbreak of numerous transnational emergencies (such as geopolitical wars, trade frictions, and political conflicts) has affected the Chinese financial market and introduced significant uncertainties (e.g., monetary policy, political risk, and geopolitical risk). In addition, the stock market is a complex, nonlinear, and dynamic system. Therefore, the least absolute shrinkage and selection operator (LASSO) and principal component analysis (PCA) techniques are used to construct a comprehensive uncertainty index for the Chinese stock market. Neural networks (a gated recurrent unit (GRU), long short-term memory (LSTM), and bidirectional long short-term memory (BiLSTM)) are embedded into a generalized autoaggressive conditional heteroskedasticity-mixed-data sampling (GARCH-MIDAS) model to construct an integrated model. The empirical results show that the newly constructed uncertainty factor provided effective information for predicting Chinese stock market volatility, and that the model's predictive ability integrated model's predictive ability is significantly better than that of the traditional model both statistical and economical. A robustness test confirms these conclusions. Therefore, understanding the volatility rules and structural characteristics of the financial market plays a vital role in accurately predicting volatility and preventing financial risk.