CORMSIS Seminar - Stability analysis for stochastic programs: Mean-risk and stochastic dominance models Event

Event details

Abstract of the talk: Measuring and managing risk has become crucial in modern decision making under stochastic uncertainty. For classes of stochastic programming models with mean-risk objectives or stochastic dominance constraints, we discuss structural properties and put them into perspective with their stability behavior with respect to perturbation of the underlying probability measure. The first part of the talk is devoted to a unified presentation of stochastic- programming related risk measures and their continuity with respect to suitable weak topologies. With atomless probability measures, this allows to extend known stability results for two-stage stochastic programs to comprehensive classes of recourse models including stochstic complementarity and bilevel models. In the second part we turn our attention to minimizing a disutility function under first-order stochastic dominance constraints. For stochastic programs with linear recourse, this leads to an optimization problem with uncountably many chance constraints. Metric regularity of the resulting constraint function is the key to stability of the solution set mapping. We focus the discussion on verifiable sufficient conditions based on local linear growth.