Computational aspects of risk-averse optimizationin two-stage stochastic models

Abstract

Computational studies on two-stage stochastic programming problems indicate that aggregate models have better scale-up properties than disaggregate ones, though the threshold of breaking even may be high. In this paper we attempt to explain this phenomenon, and to lower this threshold. We present the on-demand accuracy approach of Oliveira and Sagastizabal in a form which shows that this approach, when applied to two-stage stochastic programming problems, combines the advantages of the disaggregate and the aggregate models. Moreover, we generalize the on-demand accuracy approach to constrained convex problems, and show how to apply it to risk-averse two-stage stochastic programming problems.