Asset Clustering–Driven Risk–Return Trade-Offs Using AGE-MOEA in Multi-Objective Portfolio Optimization: Evidence From the EuroStoxx 50

Portfolio construction in financial markets is a complex decision-making problem in which multiple conflicting objectives, such as return maximization, risk control, and diversification, must be considered simultaneously. Classical single-objective portfolio optimization approaches are often inadequate for capturing this multidimensional structure. This study proposes an asset clustering based multi-objective portfolio optimization (MOPO) framework for stocks included in the EuroStoxx 50 index. In the proposed approach, assets are grouped using K-means, fuzzy C-means, and hierarchical clustering based on a multidimensional feature set that reflects risk, return, and performance characteristics. The resulting clustering structures are evaluated through a dynamic validity assessment mechanism that jointly considers the silhouette score, the Davies–Bouldin index, the Calinski–Harabasz index, and average intracluster correlation, allowing the most appropriate clustering structure to be selected for each period of analysis. The clustering structure selected is directly embedded into the MOPO formulation through cluster-based weight constraints and optimized using the adaptive geometry estimation–based multi-objective evolutionary algorithm (AGE-MOEA). The optimization simultaneously maximizes expected portfolio returns and the Sharpe ratio while minimizing the maximum drawdown (MaxDD). By explicitly exploiting the geometric structure of the Pareto-optimal solution set, AGE-MOEA effectively handles the nonlinear and ratio-based nature of the objective functions and generates a diverse set of Pareto-optimal portfolios. Our empirical results demonstrate that the proposed clustering-enhanced MOPO framework consistently outperforms traditional benchmark strategies, including the Markowitz and equal-weighted portfolios, in terms of risk-adjusted performance, drawdown control, and structural diversification. Our findings highlight that dynamically selected asset clustering structures serve not merely as a preprocessing step but as an active component that shapes the feasible solution space and improves portfolio optimization outcomes.