Research On Stock Price Prediction Based On Deep Learning

Artificial intelligence technology is widely used in various fields,and artificial intelligence algorithms have been applied in the financial field.Stock price prediction has always been the focus of financial research.Traditional statistics and econometrics pay more attention to the causal relationship of model prediction,but can not portray more characteristic factors.The deep learning model has always had strong advantages,but there are not many researches on the stock price prediction directly using the deep learning model in the market,and the accuracy is not high.The main reason is that there are many influencing factors in the financial market,and the signal-to-noise ratio of the stock price time series data is too low.It is often not effective to directly use the stock market data to make predictions.This paper starts from this issue and obtaines four kinds of financial data,including market data,fundamental data,macroeconomic indicators data and stock news data.For the four types of data,this paper proposes the feature extraction of four kinds of financial data using the deep neural network model.In addition,because the deep learning model does not have the interpretability of stock price prediction,this paper constructs a deep learning model that can describe multiple market interactions from the logical point of multi-market correlation,and finally uses the extracted feature data to predict stock prices.The main research contents of this paper include the following aspects.Feature extraction of structured information and unstructured information.This paper obtains market data,fundamental data,macroeconomic indicator data and news text data through web crawlers and financial data interfaces.For the text data of the news category,the convolutional neural network model is used for feature extraction.For the multi-dimensional mixing time series data synthesized by market data,fundamental data and macroeconomic indicator data,this paper extracts the features based on SCNN(Significance Convolutional Neural Network)network architecture,which solves the problem of frequency inconsistency of multi-dimensional mixing time series data.Analysis of cross-market correlation impacts.In this paper,by modifying the gate structure of LSTM(Long Short-Term Memory),a multi-dimensional LSTM network model is constructed,and the feature data extracted by the deep neural network model is used for model training.The experimental results show that the model can not only describe the impact of the correlation between different financial markets on the A share price,but also improve the accuracy of the stock price prediction.Comprehensive evaluation of the pros and cons of the model.This paper introduces the commonly used stock return evaluation indicators,develops a quantitative backtest platform system for the stock prediction results,and designs a quantitative trading strategy model to conduct the backtesting close to the real market.Based on the backtest results,the stock price prediction model proposed in this paper is comprehensively evaluated and analyzed.