Meta-t Portfolio Optimization: Flexible Marginal Tails and Joint Dependence

We study expected-shortfall portfolio optimization under a meta-t return model: a multivariate distribution formed by joining separately specified asset distributions through a Student-t copula. In our specification, each asset has a Student-t marginal distribution with its own location, scale and degrees of freedom, while the copula specifies dependence between their probability ranks, including joint extreme events. Individual tail thickness and joint dependence can therefore be parameterized separately. A standard multivariate Student-t model instead imposes the same degrees of freedom on all margins and on the copula. We connect this flexibility to conditional stress, marginal risk contributions, convex optimization and local allocation sensitivity. Covariance-controlled experiments demonstrate an allocation effect beyond second moments under known heterogeneous laws. A chronological CRSP study from February 2007 through December 2025 separates this effect from estimation, simulation and trading frictions. At 10 basis points per dollar traded, the baseline meta-t rule earns 7.30% CAGR, compared with 8.12% for an analytic multivariate Student benchmark and 10.31% for historical expected-shortfall optimization. Approximating the same Student benchmark by Monte Carlo reduces its CAGR to 6.69% and increases annualized two-way turnover from 2.18 to 6.23, demonstrating that numerical implementation can alter model comparisons. Nested scenario experiments, a common dynamic scale overlay and a turnover-penalized rule provide additional controls. The unconditional forecasts are poorly calibrated. The findings establish flexible marginal-tail specification as a meaningful modeling and allocation capability, while showing why that capability alone does not establish a reliable investment advantage.