Each week, we publish lay abstracts of new articles from our prestigious portfolio of journals in statistics. The aim is to highlight the latest research to a broader audience in an accessible format.

The article featured today (A reinforcement learning algorithm for trading commodities by Federico Giorgi, Stefano Herzel & Paolo Pigato) is from Applied Stochastic Models in Business and Industry with the full article now available to read here.

Giorgi, F, Herzel, S, Pigato, P. A reinforcement learning algorithm for trading commodities. Appl Stochastic Models Bus Ind. 2023; 1–16. doi: 10.1002/asmb.2825

This study melds techniques of Reinforcement Learning (RL) with traditional methods to derive optimal financial investment strategies, offering a perspective on portfolio management within a framework that considers transaction costs and unpredictable market returns.

Financial literature has proposed investment strategies based on established mathematical models. While these models have their merits, they often grapple with non-linear market dynamics, transaction costs, and other non-standard, practically relevant features of market dynamics. The authors addressed this problem by turning to RL, a branch of machine learning where algorithms learn by interacting with an environment and receiving feedback in the form of rewards or penalties.

More specifically, the authors employed a tailored version of the SARSA algorithm, a type of RL algorithm that learns to predict the quality of actions through experience. The paper presents two case studies on commodities trading. The first validates the prowess of RL, demonstrating its ability to mirror, and in some instances, surpass traditional optimal solutions. The second focuses on non-linear market dynamics, where analytical optimal solutions are not available, illustrating how RL can navigate such complexities.

The results of this paper suggest that RL can be useful in addressing features of the financial markets often ignored or neglected by theoretical models.