Variance reduction for sequential sampling in stochastic programming

This paper investigates the variance reduction techniques Antithetic Variates (AV) and Latin Hypercube Sampling (LHS) when used for sequential sampling in stochastic programming presents a comparative computational study. It shows conditions under which with AV LHS satisfy finite stopping guarantees are asymptotically valid, discussing detail. computationally compares their use both non-sequential settings through collection of two-stage linear programs different characteristics. The numerical results show that while can be preferable to random either setting, typically dominates setting performing well sequentially gains some advantages setting. These imply that, given ease implementation these techniques, armed same theoretical properties improved empirical performance relative sampling, procedures present attractive alternatives practice class programs.