Application of machine learning in capital markets trading

The thesis focuses on the application of chosen machine learning methods to daily trading in thirty randomly selected stocks. As explanatory variables, only standardized stock data and technical analysis indicators are used. The aim is mainly focused on the comparability across the approaches and different time frames. The comparisons are made between the machine learning approaches and the technical indicators, in between the machine learning approaches and across the different time frames. In the beginning chapters, the idea of trading is put into perspective with time, and a bit of a historical excursion is showcased. The work continues with the theoretical groundwork for the approaches utilized and the descriptions and definitions of their processing. Finally, the results chapters are divided, due to a larger volume of results, into two chapters. The first one goes more into detail and serves as a reference point, and the second one includes most of the important analytical work. As some summaries are done throughout the results chapters, the conclusions chapter’s main goal is to provide the answers to the selected hypotheses. With all the settings, the work concludes that machine learning methods have outperformed technical analysis indicators furthermore, that a more sophisticated approach does not necessarily always achieves better results. Finally that the time periods used for training may influence the results significantly, and the most extensive dataset might not always be the best choice for the training of the models.