Application of Decision Tree Based Algorithms on Russian Stock Market

Due to the rapid development of a computer science during the last decades, the new methods of the stock market's analysis have appeared. One of the most important branches of a CS that has affected the analysis of any data is machine learning and AI. This paper studies an application of a specific family of machine learning algorithms – decision tree based (decision tree itself, random forest, gradient boosting) in order to solve two problems in the stock market analysis: index movement direction prediction and trading operation recommendation. Both problems are treated as a classification task from the machine learning perspective. Some authors have already applied these methods to the specified problems. In order to reassess the effectiveness of such approach the evidence from Russian stock market is provided in this paper. Following the previous studies, common technical indicators are used as features in both cases. As a result, while most of the previous studies are optimistic about approach, this paper shows that neither MICEX index prediction using tree-based ML algorithms, nor prediction of the optimal trading decision using random forest proved to be effective. In case of the index prediction, an accuracy comparable to a random guess was obtained. What about an operation recommendation, the trading strategy based on the predicted actions provided positive returns only for 9 stocks out of 15 blue chip stocks tested. The “Buy & Hold” strategy was beaten for 7 of them.