Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Reinforcement learning for financial portfolio management

Vogiatzis Antonios

Full record

---

Summary

Portfolio Optimization saw a huge flood of interest in recent years, because of the rapidly growing ability of modern computers.The financial community continually seeks outstanding techniques from other fields to enhance financial market modelling. In this thesis, we propose a machine learning approach to the portfolio optimization problem, which calls for optimizingthe allocation of capital across various financial assets, such as bonds, stocks, or funds, to maximize a preferred performance metric, such as expected returns or risk-adjusted return. We use the reinforcement learning framework, which offers innovative methods of learning good decision making policies that maximizean autonomous agent’s performance in an unknown and uncertain environment. Using state-of-the-art technology based on policy gradient and deep neural networks, we developed and implemented a portfolio trading system with reinforcement learning.Subsequently, we assessed the success of our portfolio management approach and evaluated its performance using real data from the Standard & Poor’s 500 repositories of the American stock market. The results we obtained achieve about 2% more wealth in the best scenario compared to the baseline models. Webelieve that the proposed approach outlines several metrics as evaluation indices and could expand in the future for adaptive solutions to specific cases of portfolio management, delivering better performance.

Available Files

Services