An Empirical Study Of Recurrent Neural Network In Chinese Stock Market

With the development of science and technology and the improvement of computer power,artificial intelligence can solve more and more complex problems,including face recognition,speech recognition,semantic analysis and so on.The stock price is influenced by many factors,so the trend of stock price is complex and difficult to predict.In order to predict the trend of stock price accurately.In this paper,the machine learning models,such as RNN,LSTM and GRU,are introduced to train the factors of stock market.The purpose is to describe the trend of stock market returns and find the optimal portfolio.the paper studies compared with the traditional linear regression model whether the neural network can perform better in the stock market.This paper chooses many factors,such as finance,technology,momentum,market sentiment and so on.In order to describe the stock market environment completely and train the model to describe the stock trend more accurately.The model trains the multi-factor index and uses the "rise" and "fall" as labels to forecast the stock's rise and fall.This is a typical binary classification problem.It is also a problem that machine learning is good at solving.Select the data from 2007 to 2017,divide the test interval,and construct the model after determining the training set,validation set and test set,and optimize the parameters of the model.In order to obtain the optimal stock portfolio of market performance.After training,The model are divided into industry neutrality of China Securities 500,industry neutrality of Shanghai and Shenzhen 300 and non-industry neutrality.And the portfolio are adjusted respectively to find the optimal market portfolio in different situations.Get the calculate the excess return,maximum withdrawal,information ratio and Calmar ratio of the portfolios.Compare the performance of different models in different situations and analyze the optimal market portfolio.Improving the generalization ability of the neural network.Summing up the advantages of the neural network for the construction of stock portfolio.Provide new ideas for the construction of stock portfolio in the market.