Selecting control variates to estimate multiresponse simulation metamodels

In this paper the authors discuss the optimal selection of control variates for simulation experiments in which the objective is estimation of a multiresponse metamodel-that is, a linear regression model for an output vector of simulation response variables expressed in terms of an input vector of design variables for the target system. They consider the control-variate selection problem in the context of some specific covariance structures for the responses and the candidate controls that commonly occur in certain types of econometric and psychometric simulation studies. The authors conclude that in these situations, the optimal set of controls is frequently larger than would be obtained by certain conventional control-variate selection procedures; moreover as a function of the number of selected controls, the efficiency of the controlled metamodel point estimator is often relatively insensitive in the neighborhood of the optimal number of controls.