A novel convolutional auto-encoder networks for daily stock market prediction

Abstract

In financial markets, it is important and challenging to predict the daily stock price movement. In stock price movement prediction domain, feature learning from financial data is one of the most important problems. With the development in deep learning, convolutional neural networks (CNN) have been used for feature learning and prediction, and achieved good forecasting results. However, in predicted methods reported so far, less focus has been paid to different source of information for learning effective features. In this paper, we propose a Convolutional Auto-Encoder (CAE) networks, which can be used to process different sources data, including different stock markets. The CAE networks are applied to predict the next day’s direction of movement for the stock indices of Shanghai Stock Exchange Composite (SSEC), NASDAQ, S&P500 and DJIA based on various of initial financial and economic variables. The experimental results on different stock indices demonstrate the outstanding performance of proposed method compared to the baseline methods