Model Research On Chinese Stock Trend Prediction Based On LSTM-CNN

Financial market is an important part of modern financial system.The function of financial market not only effects the process of economic construction,but also relates to the social development.Analyzing and predicting the behavior of financial market can help investors avoid risk and raise benefit.In the past,it was generally assumed that financial time series were generated by linear process,and models such as ARIMA were often used to predict stock price.With the development of machine learning,models such as SVM and RF are increasingly used in stock price and trend prediction.In recent years,deep learning has made great achievements in computer vision and natural language processing.At the same time,it also begins to use in financial time series problem.In this paper,the market structure of Chinese stock market is analyzed based on the historical date of large-scale and representative constituent stocks of SSE 50 Index.And we propose the hypothesis that "the historical price of a stock affects not only its own price trend,but also the future price trend of other related stocks".The convolutional layer in CNN can extract features efficiently through local connection and weight sharing.With gate structures,LSTM can capture the dependencies of time series and remain long-term information.Combining the advantages of these two networks,we propose LSTM-CNN model,and verify the prediction ability of this model for stock trend prediction through comparative experiments with RF,CNN and LSTM.Based on the predicted trend,the T+1 trading strategy is used to simulate the trading in the test interval.We compare the proposed model LSTM-CNN with W-CNN model,LSTM integration model and multi-filters neural network MFNN on their dataset.The accuracy is improved by 6.0%,3.4%and 0.8%(almost the same with MFNN)respectively by LSTM-CNN.Overall,our model performs better than the existing models.