Stock Market Forecast Using Multivariate Analysis with Bidirectional and Stacked (LSTM, GRU)

Abstract

Deep learning has recently received growing interest and attention. It has been successfully applied to many fields. Stock market time-series forecasting is one the most challenging problems for a variety of learning methodologies. In this paper, we studied the integration of deep learning methodologies into stock market forecasting. We evaluated and compared a number of variants of Deep Recurrent Neural Network based on LSTM and GRU. Both bidirectional and unidirectional stacked architectures with multivariate inputs were employed to perform short- and long-term forecasting. The deep learning architectures were also compared to shallow neural networks using S &P500 index historical data. It has been noticed that a stacked LSTM architecture has demonstrated the highest forecasting performance for both short- and long-term.