Application of artificial intelligence methods for trading on currency markets

In this thesis, we focused on whether profitable trading strategies can be developed in highly competitive and efficient environments such as financial markets. For this task, we applied the most well-known artificial intelligence models (LSTM, MLP, RNN, GRU, CNN) and machine learning models (k-NN, Random Forest, logistic regression). We solved the classification problem and estimated the direction of movement based on volatility. In one option we used only past returns and in the other a high number of technical and macroeconomic indicators as predictors. For currency pairs, we showed that the use of technical indicators helped to improve the observed t-statistics across the board, while in the cryptocurrency market we register better results using a simpler model. Artificial intelligence models have not shown a significant advantage especially over logistic regression, which competed with these advanced approaches, even outperforming them in certain cases. Also, this work confirmed that markets are not perfectly efficient and profitable opportunities can be found.