Abstract
Training a practical and effective model for stock selection has been agreatly concerned problem in the field of artificial intelligence. Even thoughsome of the models from previous works have achieved good performance in theU.S. market by using low-frequency data and features, training a suitable modelwith high-frequency stock data is still a problem worth exploring. Based on thehigh-frequency price data of the past several days, we construct two separatemodels-Convolution Neural Network and Long Short-Term Memory-which can predictthe expected return rate of stocks on the current day, and select the stockswith the highest expected yield at the opening to maximize the total return. Inour CNN model, we propose improvements on the CNNpred model presented by E.Hoseinzade and S. Haratizadeh in their paper which deals with low-frequencyfeatures. Such improvements enable our CNN model to exploit the convolutionlayer's ability to extract high-level factors and avoid excessive loss oforiginal information at the same time. Our LSTM model utilizes Recurrent NeuralNetwork'advantages in handling time series data. Despite considerabletransaction fees due to the daily changes of our stock position, annualized netrate of return is 62.27% for our CNN model, and 50.31% for our LSTM model.