Estimating simulation metamodels using correlated control variates

Abstract

This paper develops a variance reduction technique that applies correlated control variates in a simulation experiment to estimate linear regression metamodels. A decision rule is proposed to determine whether the induced correlation of the control variates is positive or negative. Under specified conditions, the proposed technique is shown to be superior to a conventional variance reduction technique that applies independent control variates in a simulation experiment.

Keywords:

• metamodel
• control variable
• antithetic variable
• common random numbers