Kurnosau, I. and Altahhan, A. orcid.org/0000-0003-1133-7744 (2026) Direct Log-Utility Optimization for Portfolio Allocation under Exogenous Market Dynamics. In: 2026 International Joint Conference on Neural Networks (IJCNN). 2026 International Joint Conference on Neural Networks (IJCNN), 22-26 Jun 2026, Maastricht, Netherlands. Institute of Electrical and Electronics Engineers (IEEE). ISBN: 979-8-3195-3136-0. ISSN: 2161-4393. EISSN: 2161-4407.
Abstract
We propose an efficient learning framework for long–short portfolio allocation that combines an objective-driven optimization criterion with the simplicity of supervised training. Under exogenous market dynamics, where market transitions are independent of the agent’s actions under the price-taker assumption, maximising log-utility reduces to a single-step objective; hence, trajectory rollouts and multi-step credit assignment used by policy-gradient trading agents (e.g., AlphaStock, DeepTrader) often provide limited benefit. We therefore train the allocation policy end-to-end by direct log-utility optimization of a regularised objective using standard mini-batch backpropagation on observed returns. We evaluate against both trajectory-based reinforcement-learning agents and supervised portfolio-allocation baselines, including Decision by Supervised Learning (DSL), and find that our training setup consistently delivers stronger performance in our evaluation. Empirically, our method achieves substantially higher average annual return and Sharpe ratio than the evaluated baselines, while converging in fewer training iterations and with improved stability. Overall, direct log-utility optimization under exogenous dynamics provides a practical middle ground between reinforcement learning and supervised portfolio allocation.
Metadata
| Item Type: | Proceedings Paper |
|---|---|
| Authors/Creators: |
|
| Copyright, Publisher and Additional Information: | © 2026 IEEE. This is an author produced version of a conference paper published in 2026 International Joint Conference on Neural Networks (IJCNN) and made available under the terms of the Creative Commons Attribution License (CC-BY), which permits unrestricted use, distribution and reproduction in any medium, provided the original work is properly cited. |
| Keywords: | Optimization, Portfolios, Resource management, Equations, Modeling, Printing, Dynamics, Trajectory, Timing, Training |
| Dates: |
|
| Institution: | The University of Leeds |
| Academic Units: | The University of Leeds > Faculty of Engineering & Physical Sciences (Leeds) > School of Computing (Leeds) > Computing Teaching |
| Date Deposited: | 12 Nov 2026 15:43 |
| Last Modified: | 12 Nov 2026 15:43 |
| Published Version: | https://ieeexplore.ieee.org/document/11729518 |
| Status: | Published |
| Publisher: | Institute of Electrical and Electronics Engineers (IEEE) |
| Identification Number: | 10.1109/ijcnn69854.2026.11729518 |
| Open Archives Initiative ID (OAI ID): | oai:eprints.whiterose.ac.uk:247499 |

CORE (COnnecting REpositories)
CORE (COnnecting REpositories)