Accepted Paper at ECML PKDD 2022

Authors

David Winkel, Niklas Strauß, Matthias Schubert, Thomas Seidl

European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2022),
19–23 September 2022, Grenoble, France

Abstract

The task of portfolio management is the selection of portfolio allocations for every single time step during an investment period while adjusting the risk-return profile of the portfolio to the investor’s individual level of risk preference. In practice, it can be hard for an investor to quantify his individual risk preference. As an alternative, approximating the risk-return Pareto front allows for the comparison of different optimized portfolio allocations and hence for the selection of the most suitable risk level. Furthermore, the Pareto front allows the analysis of the overall risk sensitivity of various investment policies. In this paper, we propose a deep reinforcement learning (RL) based approach, in which a single meta agent generates optimized portfolio allocation policies for any level of risk preference in a given interval. Our method is more efficient
than previous approaches, as it only requires training of a single agent for the full approximate risk-return Pareto front. Additionally, it is more stable in training and only requires per time step market risk estimations
independent of the policy. Such risk control per time step is a common regulatory requirement for e.g., insurance companies. We benchmark our meta agent against other state-of-the-art risk-aware RL methods using a
realistic environment based on real-world Nasdaq-100 data. Our evaluation shows that the proposed meta agent outperforms various benchmark approaches by generating strategies with better risk-return profiles.