Portfolio Diversification Revisited: An Out-of-Sample Comparison of Twenty-One Strategies Across Asset Universes and Rebalancing Frequencies

This study revisits comparative portfolio optimization by extending an established framework from eleven to twenty-one strategies and evaluating them under a common out-of-sample evaluation design. The analysis covers eight asset universes with different degrees of heterogeneity, observed over a common period (2008–2025), and two rebalancing frequencies, using rolling estimation, Ledoit–Wolf covariance shrinkage, hyperparameter calibration, Sharpe ratios in excess of the Treasury bill rate, transaction-cost adjustments, and robustness diagnostics. The expanded framework incorporates established comparators and study-specific extensions, corrects an implementation inconsistency in a previous Second-Order Stochastic Dominance procedure, and clarifies that the Yager z=2 penalty is a reparameterization of Tikhonov regularization on the unit simplex. We also examine the sensitivity of results to risk-aversion calibration and rebalancing frequency. No strategy is uniformly superior across universes. The equally weighted portfolio remains competitive in homogeneous equity universes, whereas several optimization-based strategies achieve higher nominal Sharpe ratios in the heterogeneous multi-asset and commodity universes. In a joint test based on the stationary bootstrap, however, no strategy is better than the equally weighted portfolio in a statistically robust way, and with daily rebalancing the best mean ranks correspond to the risk-parity and equally weighted portfolios. Overall, the findings support a cautious interpretation of nominal rankings and emphasize model specification, calibration, transaction costs, and multiplicity-aware evaluation in portfolio comparisons.