Research On Stock Index Prediction Based On CEEMD-CNN-LSTM

The stock market is not only an important part of a country’s financial market,but also an important way for investors to invest and manage wealth and realize wealth appreciation,as well as an important channel for operators to finance funds.The stock index is a combination of representative stocks in the stock market through weighting method,so it represents the general trend of the stock market.If it can accurately predict the trend of stock index future,on the one hand,help regulators to build a financial risk early warning system,real-time monitoring of the assets of the stock market bubble and systemic risk,promote the healthy development of the stock market,on the other hand,can provide important reference basis for investors to investment decisions,to control the investment risk,to obtain investment returns.Therefore,the prediction of stock index has always been research with important practical and theoretical significance.However,stock index data is a chaotic system with high noise,dynamic and nonlinear in nature,which is contrary to some assumptions of the classical theory,limiting the scope of application of the classical theory.Therefore,the research on the high-precision prediction model of stock index is conducive to exploring the internal operation rules of stock index,giving full play to the functions of the stock market such as financing funds,dispersing risks,and allocating resources to a greater extent,better monitoring of stock market risks and resolving financial crises.Therefore,the research of stock index prediction has great realistic value and theoretical value.In this paper,we use Complementary Ensemble Empirical Mode Decomposition(CEEMD)to decompose the original sequence and obtain 7 intrinsic Mode functions and 1residual term.The eigen modal function was reconstructed into high frequency component,low frequency component and trend component by run-length coding,and the high frequency component was optimized to further eliminate the noise in the original time series.Then,CNN-LSTM neural network model is established for each item to predict.Finally,the predicted values of each item are integrated to obtain the result.To evaluate the prediction effect of CEEMD-CNN-LSTM,representative stock indexes from China,the United States and Europe: HS 300,S&P 500 and FTSE 100 were selected as experimental data to verify the robustness of the model.To verify the feasibility and validity of the model,different direct prediction models and integrated prediction models were compared and analyzed.The empirical results show that CEEMD-CNN-LSTM model has stronger predictive performance than other models.By modeling the original sequence processed by CEEMD method,the time lag in direct prediction can be effectively eliminated and the prediction performance of the model can be significantly improved.In different data sets,CEEMD-CNN-LSTM prediction results have strong robustness.