Augmented simulation methods for discrete stochastic optimization with recourse

Abstract

We develop an augmented simulation approach to solve discrete stochastic optimization problems by converting them into a grand simulation problem in the joint space of random and decision variables. The optimal decision is obtained via the mode of the augmented probability model, using a new multivariate extension of the classic Barker’s algorithm. Illustrations on different versions of univariate and multivariate discrete news-vendor problems with exogenous and endogenous uncertainties are detailed. We contrast our method with the Metropolis–Hastings algorithm, the nested sampling-based augmented simulation method, and traditional Monte Carlo simulation-based optimization schemes. The proposed method is shown to be computationally efficient and could serve as another tool to solve discrete stochastic optimization problems with recourse.