Confidence intervals from a single simulation using the delta method

Abstract

This paper describes a method for the calculation of confidence intervals of simulation throughputs and utilizations. The method is based on the delta method and uses only a single simulation, where the variation of the underlying means is used to determine the variation of the performance function by using the first derivative of the performance function. While the method has some limitations, it can be used frequently in practice. In addition, the method can also be used for short simulations or rare event applications, where methods based on batch means fail. This method can easily be implemented into existing simulation software.