1155. Reinforcement Learning Methods for Risk-Sensitive Investment Management
Invited abstract in session MB-31: Machine Learning for Optimization under uncertainty 1, stream Stochastic and Robust optimization.
Monday, 10:30-12:00Room: Maurice Keyworth 1.06
Authors (first author is the speaker)
| 1. | Sebastien Lleo
|
| Finance Department, NEOMA Business School | |
| 2. | Wolfgang Runggaldier
|
| Mathematics, University of Padova |
Abstract
We investigate the benefits of relating reinforcement learning (RL) with risk-sensitive control. Our starting point is the duality between free energy and relative entropy, which establishes an equivalence between risk-sensitive control and standard stochastic control problems with an entropy regularization term; see, e.g., Dai Pra et al. (1996). A major advantage of this approach is that it rewrites the initial linear-exponential-of-quadratic-Gaussian problem as an equivalent (penalized) linear-quadratic-Gaussian problem. In so doing, it does not require a preliminary change of measure as is being done in the seminal work of Kuroda & Nagai (2002). Adapting the policy gradient method proposed by Jia
& Zhou (2022b), we examine four distinct cases, categorized by the level of parameter knowledge (known or unknown) and state process observability (observable or not). Cases with known parameters connect directly to the existing literature on risk-sensitive investment management (Kuroda & Nagai, 2002; Davis & Lleo, 2008, 2020, 2021b) and can be solved analytically. Cases with unknown parameters showcase the advantages of reinforcement learning methods.
Keywords
- Stochastic Optimization
- Financial Modelling
- Machine Learning
Status: accepted
Back to the list of papers