Stock Index Price Forecasting Based On Long Short-Term Memory Deep Learning Network Model

Since the birth of financial markets such as the stock market,various data models and machine learning,data mining,and other methods have been used by investors to predict the future trend of stock prices and obtain generous returns.As a recognized complex dynamic system,the stock market has many influencing factors,such as nonstationarity,non-linearity,high noise,and long memory.It is difficult to explain it simply by mathematical models.Therefore,the analysis and prediction of the stock market hasIt has been a challenging job since.With the emergence of deep learning methods in various industries,financial researchers have begun to apply them to the prediction research of financial time series.This paper studies the forecasting problem of financial time series based on deep learning methods,and uses the Shanghai and Shenzhen 300 Index as the prediction object.The data of the Shanghai and Shenzhen 300 Index from 2009 to 2019 is used as the research sample,and the closing price is used as the forecast target.Time series prediction model of memory deep learning network.The model combines three modules of wavelet denoising module,stacked selfencoder module(SAE)and long-term short-term memory network module(LSTM),and then analyzes and predicts the time series of stock market index in financial markets.This article selects basic market data such as opening and closing prices(OHLC),technical indicators such as MACD,ROC,and macroeconomic indicators such as SHIBOR.A total of 21 indicators are used as input data for the model.First,in the wavelet noise reduction module,this paper selects the Sym6 wavelet and SureShrink threshold method to perform noise reduction processing on the OHLC data of the stock index.After that,the noise-reduced OHLC data is normalized with other selected indicators as the following.Input data for a module.Secondly,in stacking selfencoder modules,this paper selects a 4-layer network structure and uses the L2 regularization method to reduce the input 21-dimensional data to 10-dimensional output data as the input data of the LSTM module.Thirdly,in the LSTM module,a bidirectional LSTM network structure with better performance is selected.After the optimal parameters of the model are determined using an automatic parameter adjustment algorithm,the input data is analyzed and predicted to obtain the model's predicted structure.Finally,construct a trading strategy to test the performance of the model's profitability,and compare it with SAE + LSTM,wavelet denoising + LSTM,and a single LSTM model.The empirical research results found that:(1)the model overcomes the nonlinear and non-stationary characteristics of the financial time series and can achieve good prediction results;(2)the model proposed in this paper is profitable when the buy-sell strategy is usedThe excellent ability indicates that this model has certain practical application value.(3)The model proposed in this paper has a good prediction effect on the less mature and developing stock markets such as the Shanghai and Shenzhen 300 Index,and has certain promotion value.