STOCK INDEX VOLATILITY PREDICTION BASED ON MIXED-FREQUENCY GARCH AND LSTM

Mixed-frequency GARCH models can establish the relationship between low-frequency macroeconomic variables and high-frequency stock index volatility, but their strong linear characteristics will reflect the limitations. In order to reduce the prediction error, a volatility prediction method based on hybrid frequency GARCH and LSTM is proposed. A single mixed-frequency GARCH and LSTM model was established and compared. The fusion model of the two was established, that was, the residual term of GARCH with mixed frequency was used as the LSTM input, and the two were superposed to form the final output. The results show that the single model fits well when the volatility changes frequently, but the offset increases when the volatility changes extremely. The multi-factor GJR-GARCH fusion model can solve this problem. The residual terms of the two models reject the null hypothesis at the 1% level in the Diebold-Mariano test, and the fusion model significantly improves the prediction accuracy.