Multi-Indicator Joint Stock Trend Prediction Based On LSTM

With the development of our country’s economy and society,people’s living standards have gradually improved,and they have more idle funds,so they turn their attention to some financial products.Stocks have become the choice of most people because of their high-yield characteristics.However,while stocks have high returns,they also have high risks.The stock market is complicated and unpredictable.Many factors have an important impact on stock prices.Thanks to the support of behavioral finance theory,people are gradually paying attention to the impact of investors’ psychological state and emotions on the stock market.In recent years,the rapid development of machine learning and natural language processing technology has made it possible to measure investor sentiment through text information in online media,and then it is possible to mine the relationship between investor sentiment reflected in financial texts and stock price changes.A large body of research literature shows that investor sentiment behind financial texts is indeed closely related to stock price movements.Based on this point of view,this paper collects a large number of financial texts,and fully extracts its feature information through sentiment dictionary and machine learning methods to fully tap the investor sentiment hidden behind financial texts.Through the current mainstream LSTM,random forest,XGBoost methods,set up a comparison experiment,and verified the conjecture that adding the factor of financial text will help to improve the effect of stock price prediction.This article is based on financial texts and historical trading data publicly available in the stock market.Using LSTM to build a stock price prediction model,the following three conclusions are obtained: First,The effect of the LSTM prediction model incorporating financial text features is better than the LSTM prediction model not incorporating financial text features,indicating that in the study of stock price prediction,the feature of financial text cannot be ignored.Second,the model incorporating financial text features outperforms both the random forest and XGBoost classification tasks than the models not incorporating financial text features.Third,this paper uses the results of random forest and XGBoost classification prediction to revise the results of LSTM regression prediction,proposes an LSTM interval correction algorithm,applies this method to investment stock selection strategies,and compares it with other methods,and algorithm proposed in the paper performs better in terms of return on investment.