Research On Stock Investment Strategy Based On Deep Learning LSTM Neural Network

Deep learning technology,the current research direction in the field of artificial intelligence,arousing wild attention of social areas over lasted years and used in all fields.Using it to build neural network models to predict and research stock prices has been a hot topic in the financial field.There is a model often used in analysis of financial time series,called long-short-term memory network(LSTM).With its unique structure imitating the human brain's memory pattern for long-term information,it can search for the law of stock price movements from longer-term historical data well.Testing whether adding more feature vectors to the neural network model will improve the predictive ability of the model or not is the main research content of this paper,those feature vectors selected can be divided into three types:basic stock price data,technical analysis indicators,and market sentiment indicators.Firstly,this article selects Baidu index daily data of keywords related to the stock market,and then uses the principal component analysis(PCA)to reduce the dimensionality of 49 keyword data,thereby obtaining 10 principal component variables and selecting from them seven variables that have significant correlation with stock prices serve as indicators of market sentiment.With the acquisition of main component indicators showing market sentiment,this paper constructs a vector autoregressive(VAR)model to verify that market sentiment indicators constructed from Baidu index daily data will have an impact on future stock price changes.Then,constructing an LSTM model and a support vector machine(SVM)model to predict the 5-day return of the Shanghai 50 Index respectively,which verified that the LSTM model added to the market sentiment index can improve the accuracy of prediction.In terms of the accuracy of the index change,the SVM model is slightly better than the LSTM model.In the end,this paper uses the constructed model to design a set of quantitative investment strategies and compares them with a buy-and-hold strategy and a fixed investment strategy by calculating various quantitative evaluation indicators.The results show that the LSTM model is superior to the SVM model and other strategies in simulating investment,which proves the feasible of stock investment strategy based on deep learning technology in practical application.The main value of this paper is:It proves that the successful processing and analysis of Baidu index data can measure the change of market sentiment to predict the stock price trend,and the feature vector as the model input can improve the prediction accuracy of LSTM network.And in this paper,by constructing an investment strategy based on the LSTM model,it finally achieved a higher return than the simple holding strategy or fixed investment strategy in the simulated trading,which will provide some inspiration for investors who intend to use deep learning technology for investment activities.The structure of this paper can be divided into five parts:the first part summarizes the domestic and foreign literatures that used deep learning algorithm to predict the financial market in the past,and introduces the research methods and significance of this paper;the second part mainly describes the principle and algorithm of the model structure used in this paper;the third part verifies that the market sentiment index and index return rate are significant through the construction of VAR model The fourth part constructs the deep learning model to predict the 5-day return of Shanghai 50 index,constructs the investment strategy and evaluates it;the fifth part draws the conclusion of this paper.