An Empirical Research on the Effectiveness of Different LSTM Architectures on Vietnamese Stock Market

Abstract

Stock price prediction is a challenging financial time-series forecasting problem. In recent years, on account of the rapid progression of deep learning, researchers have developed highly accurate, state-of-the-art time-series models. Long short-term memory (LSTM) stands out as one of the most reliable architecture at capturing long-time temporal dependences. In Vietnam, there is a lack of research papers that solely focused on the effectiveness of deep-learning in stock price prediction. This paper surveys three different variations of LSTM (Vanilla, Stacked, Bidirectional) when applied to 20 companies’ stock prices over a period of 5 years from 2015 to 2020 in the VN-index stock exchange. The results show that Bidirectional LSTM is the most accurate model.