Fusion Machine Learning Based Stock Market Prediction

In recent years, due to the development of online trading, the great importance has been given to the prediction of future trends and values on financial markets, development effective trading strategies based on these predictions. Nowadays, the most perspective method of forecasting is an fusion technique, combining the predictions of the most powerful technical analysis models. However, this technique has not been used to predict stock prices of European and American markets. Thus, the daily values of Russian index MICEX O&G and the US Dow Jones Oil and Gas index was used since 2009. In addition, no research was found, dedicated to developing a trading strategy based on predictions, i.e., making profit from the forecasts. Thus, the objects of the diploma research are: 1) To investigate the accuracy of the fusion approach as compared to individual models; 2) To develop a trading strategy based on the the predictions, and compare it with a simple buy-and-hold. Neural networks and integrated autoregressive moving average model were used for the construction of individual forecasts. For the fusion predictions, we used a machine learning method, namely, a random forest. As for trading strategies, two systems were proposed, with buy and sell signals respectively. As a result of the empirical research, the following conclusions were obtained: 1) Using a fusion of machine learning techniaue significantly improved the accuracy of the forecasts for the available data; 2) The proposed trading strategies were profitable in comparison with the buy-and-hold only for the Russian index. For the US index, buy-and-hold and the proposed strategies turned ineffective, presumably because of the strong downward trend during the entire test period.