LSTM Prediction Model And Empirical Study Of Stock Index Based On Deep Learning

Stock market price fluctuations are the basis for the operation of the stock market,and are a comprehensive reflection of many factors such as national politics,economy,and society.Large price fluctuations in the stock market will cause unimaginable losses to the financial market,thereby affecting the optimal allocation and structural adjustment of the entire country's economic resources.The research on stock forecasting methods at home and abroad mainly focuses on statistical methods and shallow machine learning methods.However,due to the large amount of stock price data,non-linearity,and long memory characteristics,statistical forecasting methods have certain limitations.At the same time,shallow machine learning methods have fewer model parameter selections,low practicability,and forecasting consumption time length and other issues.Therefore,how to study effective deep learning methods to characterize long memory and complex relationships between data and improve the recognition efficiency of predictive models is an urgent problem to be solved in stock forecasting.Based on deep learning,this paper systematically studies the selection of stock forecast indicators,the construction of long short-term memory neural network(Long short-term memory,LSTM)forecasting models and empirical analysis.First,it analyzes the current research status of the relationship between stock forecasting methods and investor preferences and stock indexes at home and abroad.Secondly,according to the characteristics of China stock indexes,a stock index predictive index system is constructed.The random forest method is used to screen the stock index predictive indexes,and the filtered indexes are combined with the theoretical value of the prospect of measuring investor preferences.As an input item of the LSTM neural network,a stock index LSTM prediction model based on deep learning is established.Finally,the daily price data of five stock market indexes in China stock market from January 1st to December 31st,2019 are selected as research samples,and the constructed stock index LSTM prediction model is optimized and empirically analyzed.The research results show that the forecasting model constructed in this paper has a good forecasting effect and a higher accuracy rate for China stock index.The research features and innovations of this paper are mainly manifested in:?Introducing prospect theoretical value indicators that can measure investor preferences,which makes up for the lack of previous research on stock prediction indicators;?Using random forest to screen China stock index prediction indicators and constructing A predictive index system suitable for deep learning.?Established a LSTM stock prediction model based on deep learning,and the empirical results showed a good prediction effect.The research in this article not only proposes a new method of stock forecasting that uses the theoretical value of the prospect of measuring investor preferences as a predictive indicator,but also provides a theoretical basis and policy support for the supervision of relevant government departments,institutional investors and other investors' stock price analysis.