Robust distortion riskmetrics under symmetry, unimodality, and Wasserstein constraints

Abstract

We study worst-case distortion riskmetrics over symmetric distributions with prescribed mean and variance in a Wasserstein ball around a reference distribution. Our key contribution is the construction of a reference-dependent auxiliary distortion that encodes the Wasserstein constraint while preserving symmetry under the concave-envelope operation. Under suitable regularity and nondegeneracy conditions, this construction yields sharp upper bounds for distortions of bounded variation, allowing nonmonotonicity and interior discontinuities, and identifies extremal distributions when the bounds are attained. An approximation preserving symmetry and the prescribed moments exactly shows that upper semicontinuous relaxation of the symmetrized distortion leaves the supremum unchanged for budgets strictly above the feasibility threshold. With additional unimodality, symmetry fixes the quantile's concavity-convexity transition at the midpoint, yielding a characterization via $L^2$-projection onto a fixed closed convex cone for absolutely continuous distortions with square-integrable derivatives. Numerical examples for ES, RVaR, and VaR illustrate how shape restrictions and the reference distribution affect feasibility and worst-case risk. Simulations with data-dependent Wasserstein radii examine worst-case ES and feasibility under correctly specified and misspecified shape restrictions.