|
|
Reinforcement Learning for Automated Stock Portfolio Allocation
Lapeš, Zdeněk ; Andriushchenko, Roman (oponent) ; Češka, Milan (vedoucí práce)
This thesis is focused on the topic of reinforcement learning applied to a task of portfolio allocation. To accomplish this objective, the thesis first presents an overview of the fundamental theory, which includes the latest value-based and policy-based methods. Following that, the thesis describes the Stock portfolio environment, and finally, the experimental and implementation details are presented. The creation of datasets is discussed in detail, along with the rationale and methodology behind it. The RL agent is then trained and tested on three datasets, and the results obtained are promising and outperform common benchmarks. However, it was discovered that the annual return of the agent is still not better than the returns generated by the world’s top investors. The pipeline was implemented in Python 3.10, and technology from Weights & Biases was used to monitor all datasets, models, and hyperparameters. In conclusion, this work represents a significant step forward in the development of more effective RL agents for financial investments, with the potential to exceed even the performance of the world’s greatest investors.
|
host ::
RSS.