Empirical Analysis Of Stock Price Predictability Test Based On Deep Learning

The prediction of stock price is one of the most concerned problems in the field of finance.However,the stock market is often affected by various aspects of information and operations,and there is huge volatility in the rise and fall of stock prices,making accurate prediction very difficult.In recent years,big data analysis based on machine learning algorithms has developed rapidly,and has been applied to machine translation,image recognition and other fields with good results.Therefore,scholars in the field of finance try to apply these algorithms to the stock market.Whether it is to classify stocks or to predict stock price trends,machine learning algorithm technology is much more accurate than traditional theoretical methods.In machine learning algorithm,LSTM neural network can process financial data more effectively because of its special gate structure.XGBoost integrated learning algorithm is fast and accurate.Therefore,this paper chooses these two algorithms to predict the stock price and provide reference and suggestions for investors.Based on the csi 300 index and the csi 500 index,selects the LSTM neural network and XGBoost integrated learning algorithm stock price prediction model is established,according to the optimal investment time window,the relationship between quantity and price,the high frequency data consider price factors,which affect the yield,the three parts the main research work is as follows:(1)the selection of investment cycle time window for 21,34,55 day,explore profit biggest which cycle;(2)in the LSTM model,the "closing price" and "closing price + trading volume" are taken as inputs to compare their strategic returns and random returns;(3)the XGBoost integrated learning algorithm is used to predict which dimension data is more effective when measuring the strategic return on the five-minute high frequency data and the daily low frequency data.The analysis shows that in the stock market,the optimal investment window period is 55 trading days.In the LSTM model,higher returns can be obtained if the input volume is "closing price + trading volume".The XGBoost model USES high-frequency data to predict stock prices and thus yields better results than low-frequency data.It can be seen that deep learning algorithm has certain guiding significance for stock market prediction and investment,and provides a convenient and feasible scheme for investors to make decisions.