Random forest, gradient boosted machines and deep neural network for stock price forecasting: A comparative analysis on South Korean companies

Predicting the final closing price of a stock is a challenging task and even modest improvements in predictive outcome can be very profitable. Many computer-aided techniques based on either machine learning or statistical models have been adopted to estimate price changes in the stock market. One of the major challenges with traditional machine learning models is the feature extraction process. Indeed, extracting relevant features from data and identifying hidden nonlinear relationships without relying on econometric assumptions and human expertise is extremely complex and makes deep learning particularly attractive. In this paper, we propose a deep neural network-based approach to predict if the stock price will increase by 25% for the following year, same quarter or not. We also compare our deep learning method against 'shallow' approaches, random forest and gradient boosted machines. To test the proposed methods, KIS-VALUE database consisting of the Korea Composite Stock Price Index (KOSPI) of companies for the period 2007 to 2015 was considered. All the methods yielded satisfactory performance, namely, deep neural network achieved an AUC of 0.806. 'Shallow' approaches, random forest and gradient boosted machines have been used for comparisons.