Relaxation of stochastic dominance constraints via optimal mass transport

Abstract: Optimization problems with stochastic dominance constraints provide a possibility to shape risk by selecting a benchmark random outcome with a desired distribution. The comparison of the relevant random outcomes to the respective benchmarks requires functional inequalities between the distribution functions or their transforms. A difficulty arises when no feasible decision results in a distribution that dominates the benchmark. Our paper addresses the problem of choosing a tight relaxation of the stochastic dominance constraint by selecting a feasible distribution, which is closest to those dominating the benchmark in terms of mass transportation distance. For the second-order stochastic dominance in a standard atomless space, we obtain new explicit formulae for the Monge-Kantorovich transportation distance of a given distribution to the set of dominating distributions. We use our results to construct a numerical method for solving the relaxation problem. Under an additional assumption, we also construct the associated projection of the distribution of interest onto the set of distributions dominating the benchmark. For the stochastic dominance relations of order $r\in[1,\infty)$, we show a lower bound for the relevant mass transportation distance. Our numerical experience illustrates the efficiency of the proposed approach.