This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
School of AI and Advanced Computing, XJTLU Entrepreneur College (Taicang)
Liverpool University
Department of Electrical Engineering and Electronic, University of Liverpool , Liverpool , United Kingdom
Financial portfolio management aims to allocate resources for optimal returns while mitigating risk. Traditional methods based on assumptions like normal return distributions and simplistic risk measures have limitations. Deep reinforcement learning (DRL) combines deep learning for feature representation and reinforcement learning for optimal decision-making. It offers advantages like reduced assumptions, handling complex environments, automated feature learning, improved sample efficiency, online learning, and scenario simulation. However, DRL for portfolio management faces challenges regarding financial risk management, overfitting, data quality, reward design, and generalizing to unseen scenarios. Potential improvements include enhanced algorithms and neural architectures, richer state/action spaces, multi-objective reward functions, regularization, and incorporating real-world constraints. Other goals are improving generalization through sufficient representative data, managing training anomalies, and bridging gaps between simulated and live environments. Verifying state-of-the-art performance lacks established benchmarks. Common metrics used are cumulative returns, risk-adjusted returns like Sharpe/Sortino/Omega ratios, and comparisons to prior work reproduced in-house. Isolating the impact of individual model components through theoretical analysis, ablation studies, module replacement, and intermediate validation is difficult. Improving interpretability by integrating traditional financial principles and extracting trading logics is important for investor acceptance. Despite the challenges, DRL shows promise for enhanced portfolio management
The statements, opinions and data contained in the journal are solely those of the individual authors and contributors and not of the publisher and the editor(s). We stay neutral with regard to jurisdictional claims in published maps and institutional affiliations.
