Virtual Measures: A Variance Reduction Technique for Simulation

Abstract

A variance reduction technique is presented that is effective when the system characteristic to be estimated is strongly influenced by rare events. Applications to simulations of a simple inventory system and of the New York City Fire Department, where the rare events are back orders and serious fires respectively, are described. In these simulations, virtual measures reduced computer time needed to achieve a given statistical reliability to 1/2 to 1/20 of that needed with the usual direct estimators. Relationships to other variance reduction techniques are indicated.