Comparison of variance reduction techniques for steady-state simulation analysis

The primary purpose of many simulation studies is the estimation of performance measures for the interesting system parameters. The estimations of steady-state measures suffer from autocorrelation and initial bias of output data, causing inaccuracy and inefficiency estimations. This paper concerns statistical methods for computing efficiency estimations for system performance measures. We introduce variance reduction techniques (VRTs) into the batch means (BM) of steady-state simulation when the initial bias exists. Experimental studies consider 4 common stand-alone VRTs and 3 combined VRTs in queueing simulation systems under various experimental conditions. Empirical results illustrate significant improvements for performance measures in mean square error (MSE), and indicate the validity of the proposed method.