Predicting The Price Volatility Of CSI 300 Index Futures Based On Deep Learning Algorithm

Volatility plays crucial roles in financial markets,such as in derivative pricing,portfolio risk management,and hedging strategies.Therefore,accurate prediction of volatility is critical.Nowadays,many scholars have applied the deep neural network algorithm in the field of artificial intelligence to the analysis of financial time series and achieved success.Based on the previous research,this paper firstly applies the AE(Auto-Encoder)automatic encoder to the input of the deep neural network,and forms a hybrid model with the deep neural network model to predict the volatility of the index futures.By applying the AE automatic encoder,it is possible to extract more necessary abstract information from the high-dimensional data at the input of the deep network model,thereby improving the learning efficiency of the network.In order to verify the effectiveness of AE auto-encoder in high-dimensional input data feature extraction,this paper also introduces PCA principal component analysis method and compares it with AE auto-encoder.Secondly,this paper innovatively combines two nonlinear algorithms(ANN,LSTM)and two feature extraction methods(PCA,AE)in the field of artificial intelligence into four hybrid models: ANN-PCA,ANN-AE,LSTM-PCA,LSTM-AE.In order to obtain a more comprehensive empirical research results,this paper applies the above seven volatility forecasting models to the index futures market in China and the United States in the long-term and short-term time spans.Through empirical research,it is found that two single nonlinear models and four nonlinear models with feature extraction methods are superior to the EGARCH model in the prediction of long-term and short-term realized intraday volatility.By comparing the improvement of nonlinear models by different feature extraction methods,it is found that the AE automatic encoder improves the LSTM model in the long-term and short-term empirical studies more than the PCA principal component analysis.Taking China's short-term empirical study as an example,the NMSE value of the LSTM-AE model is 0.924,the LSTM model is 0.96,and the increase rate is 3.82%;while the LSTM-PCA model has a short-term NMSE value of 0.969,which has a 1.8% increase compared with the LSTM model.t is worth noting that the LSTM-AE hybrid model performs best in both China and American empirical studies in the prediction of long-term and short-term realized intraday volatility.Taking China's CSI 300 Index Futures as an example of the intraday volatility forecast,the NMSE values of the test set in the short-term and long-term are 0.924 and 0.940,while the HMSE values of the EGARCH model are 1.112 and 1.141,respectively.The improvement rate of prediction accuracy is 16.94% and 21.38%,and this increase rate is the largest among all other models,indicating that the LSTM-AE model is the optimal model in the seven intraday volatility prediction models.The contribution of this paper is to demonstrate the feasibility of the hybrid model composed of AE automatic encoder and deep neural network model in the forecast of intraday volatility of index futures through empirical research on China and US stock index futures markets;It is also demonstrated that auto-encoder,as a feature extraction method,can effectively enhance the prediction efficiency of deep neural network model for volatility,and proposes a deep neural network volatility prediction model with feature extraction : LSTM-AE model.