Evaluation of bidirectional LSTM for short-and long-term stock market prediction

Recently, there has been a rapidly growing interest in deep learning research and their applications to real-world problems. In this paper, we aim at evaluating and comparing LSTM deep learning architectures for short-and long-term prediction of financial time series. This problem is often considered as one of the most challenging real-world applications for time-series prediction. Unlike traditional recurrent neural networks, LSTM supports time steps of arbitrary sizes and without the vanishing gradient problem. We consider both bidirectional and stacked LSTM predictive models in our experiments and also benchmark them with shallow neural networks and simple forms of LSTM networks. The evaluations are conducted using a publicly available dataset for stock market closing prices.