This thesis explores the application of Deep Learning and Reinforcement Learning in cryptocurrency trading, focusing on the performance of these models in the volatile cryptocurrency markets and the integration of price direction predictions as external predictive signals into market-making strategies. The thesis adapts a DL architecture, initially developed for predicting the stock price directions, to cryptocurrency data, examining its robustness and efficiency in these markets. The results show performance decreases with longer prediction horizons, aligning with prior research. A significant part of the research investigates the inclusion of price direction predictions into the observation space of RL agents. These predictions, sourced externally, are used to enhance the learning process, with the objective of improving the profitability of market-making strategies. The results reveal that the integration of these specific external predictive signals yields mixed outcomes: performance improvements are evident under certain conditions, highlighting the potential of combining novel predictive techniques with state-of-the-art RL models. The thesis concludes by emphasizing the adaptability of DL models in volatile cryptocurrency markets and the potential benefits of enhancing RL agents with external predictive signals. It provides insights for future studies to further explore and refine these methods, especially focusing on the effectiveness of novel DL-based predictive signals.     «

This thesis explores the application of Deep Learning and Reinforcement Learning in cryptocurrency trading, focusing on the performance of these models in the volatile cryptocurrency markets and the integration of price direction predictions as external predictive signals into market-making strategies. The thesis adapts a DL architecture, initially developed for predicting the stock price directions, to cryptocurrency data, examining its robustness and efficiency in these markets. The results sh...     »