Online Portfolio Selection

Online portfolio selection aims to develop algorithms that dynamically allocate wealth across multiple assets to maximize returns over time. Current research focuses on improving the efficiency and robustness of algorithms, including gradient-based methods like exponentiated gradient variants and deep reinforcement learning approaches, while addressing challenges like transaction costs and non-smooth loss functions. A key area of investigation involves achieving low regret—minimizing the difference in performance compared to the best fixed strategy in hindsight—with computationally efficient algorithms. These advancements have implications for both theoretical understanding of online convex optimization and practical applications in financial markets and related domains.