Research On Stock Index Prediction Based On Deep Learning Network With Improved EWT

The stock market provides an important channel for investment entities to raise funds and allocate assets.It is an important part of the financial market,as well as a key focus of financial regulatory authorities.The stock price index measures and reflects the overall price level and change trend of the stock market.It is a key indicator that reflects the market situation.The analysis and prediction of stock index time series data is of great significance both in theory and practice.Stock index time series data has data characteristics such as non-stationary,nonlinear,high noise,and long memory.Deep learning can extract high-level and abstract features from massive raw data without relying on prior knowledge and has great advantages in the analysis and prediction of stock index data with complex data features.The basic framework of the stock index time series data prediction model constructed in this paper is as follows: Firstly,the signal decomposition algorithm is used to decompose the stock index time series data into different instrinsic mode functions.Secondly,the Long-short Term Memory(LSTM)network,which is suitable for solving long-term dependence problems,is used to extract the depth features of each instrinsic mode function and perform single-step prediction.Finally,the predicted value of each mode function is synthesized to obtain the predicted value of the original sequence.The signal decomposition algorithm used in this paper is Empirical Wavelet Transform(EWT)decomposition algorithm.According to the shortcomings of the existing EWT decomposition algorithm,the EWT decomposition algorithm is improved from two aspects,and an improved EWT(IEWT)decomposition algorithm with improved spectrum segmentation method is obtained.Firstly,the spectrum segmentation method of the EWT decomposition algorithm is improved,so that the spectrum segmentation boundaries can be determined according to the spectrum trend of the signal data and the number of instrinsic mode functions can be adaptively determined;Secondly,the auxiliary white noise is added to the original signal in the decomposition process to obtain better data decomposition and reconstruction effect.Based on the IEWT decomposition algorithm,this paper proposes a single-step forward hybrid forecasting model for stock index time series data—IEWT-LSTM model.In the empirical analysis stage,six stock indexs from three types of markets with different levels of development are selected for modeling and analysis.The result of the control experiment shows that IEWT-LSTM is superior to other models both in prediction error and accuracy for direction of stock index.The IEWT decomposition algorithm is verified to have effectiveness and advantages in stock index feature extraction.IEWT-LSTM model also shows strong robustness in the robustness test.The research has great quantitative investment value.