This was a major research paper conducted by my group member (Shadi Chamseddine) and I for our DATA 5000 Data Science Seminar course.
Since the inception of the first stock market, various models, techniques and methods have been developed to try and gain an edge in stock investing. Yet, the ever changing and complex nature of the stock markets have rendered these approaches unreliable. With the tremendous growth in computing power today, a new approach to cracking the code for stock price prediction has started to take shape. Traders have begun to explore computational solutions in an attempt to predict movements in stock prices. However, to date, there has been a lack of innovation in the application of deep learning techniques that make accurate predictions feasible. We propose a Long Short-Term Memory (LSTM) model approach to address this innovation gap. The LSTM is developed to capture pertinent underlying data related to stock price trends based on historical stock movements, as well as key features drawn from correlated assets and technical indicators. We show how this deep learning recurrent neural network model is able to accurately predict the trends in future stock price movements. We obtain a forecast error of just 3.1% for our LSTM predictions, which is considerably more accurate than the forecast error of the current standard benchmark ARIMA model which comes in at 6.8%.