Cross-Sectional Reliability of Portfolio Selection Metrics in a Multi-Asset Universe

This paper investigates the cross-sectional reliability and persistence of commonly used portfolio performance indicators when employed as portfolio selection metrics over a broad multi-asset investment universe. Using 14 years of daily data (2012–2025) spanning a diverse multi-asset universe, we construct the full cross-section of feasible equally weighted portfolios and assess how consistently different selection metrics identify top-performing portfolios out of sample. At each decision date, portfolios are selected by maximizing an in-sample decision criterion, after which the same criterion is re-evaluated out of sample across the entire portfolio universe and the selected portfolio is ranked based on its out-of-sample metric value. This cross-sectional, rank-based framework is designed to mitigate spurious predictability driven by correlation, overfitting, and look-ahead bias by evaluating each selected portfolio relative to the full cross-sectional opportunity set using strictly out-of-sample rankings. Across multiple time scales, we document a robust separation between decision accuracy and decision stability. While the Omega ratio maximizes short- and medium-horizon selection accuracy, the Calmar ratio exhibits superior tail robustness and substantially higher decision persistence. At longer horizons, average selection accuracy converges across the metrics, while Calmar retains a substantial stability advantage. The main contribution of this study is the introduction of a cross-sectional decision-based evaluation framework that assesses performance measures according to the persistence of their portfolio selection decisions rather than solely their realized return characteristics. The results indicate that different selection metrics emphasize distinct dimensions of decision quality, including tail dominance, stability, and persistence, rather than converging to a single notion of optimality. Our findings are further supported by robustness checks based on alternative portfolio sizes and bootstrap-based statistical validation.