| This project evaluates two different Deep Learning models for Asset Management using Deep Reinforcement Learning. While it’s been shown that predicting the market is unreliable, there are still indicators that can produce accurate descriptions of the future trends of assets in the Stock Market. However, new forms of assets such as cryptocurrencies and Defi tokens pose a new challenge for investors and a unique opportunity for higher-risk strategies. Furthermore, little research is done in the High-Frequency market. The goal is to test approaching the problem as a Computer Vision problem instead and using only the most fundamental elements of Q-learning and Vision Transformers in order to train a model to trade in the High-Frequency Trading crypto market and unveil its strategies. This project also tests state-of-the-art computer vision algorithms (Vision Transformers) against traditional Convolutional Neural Networks. I found that Vision Transformers tend to perform better in the evaluation dataset compared to CNNs and also take less volatile strategies, contrary to CNN-based models which either take high-risk or very neutral strategies. |
