RL-NSGA-II-GRC Optimizes NASDAQ Portfolios with Multi-Objectives

In today's intricate financial markets, decision-makers face the challenge of balancing multiple, often conflicting objectives, such as minimizing risk while maximizing returns in portfolio management. This research proposes a new, sophisticated algorithm called RL-NSGA-II-GRC to tackle constrained multi-objective optimization problems. This method integrates a reinforcement learning (RL) agent with the Non-dominated Sorting Genetic Algorithm II (NSGA-II) and enhances it with Gray Relational Coefficients (GRC). The RL agent dynamically adjusts evolutionary parameters based on real-time feedback, including hypervolume, feasibility, and diversity metrics, to guide the search more effectively. Concurrently, a GRC-enhanced tournament operator provides a comprehensive ranking for parent selection, considering dominance, crowding distance, and proximity to an ideal solution. When applied to NASDAQ portfolio optimization, the RL-NSGA-II-GRC framework generated a smooth and densely populated efficient frontier, enabling the identification of optimal portfolios for various risk-aversion levels, including the maximum Sharpe ratio portfolio. The method demonstrated improved convergence and diversity compared to standard NSGA-II on both benchmark problems and the financial application.