Asymptotic Properties of Some Confidence Interval Estimators for Simulation Output

Abstract

The classical confidence interval estimator commonly used in simulation is compared with four new estimators based on standardization of a time series presented in a previous paper. These new interval estimators are shown to have asymptotic properties that strictly dominate the classical estimator when used with data from independent simulation replications or means of batched observations from a single simulation run. Two of the new estimators also can be used with a single unbatched replication of a simulation program, a situation where the classical estimator is not defined.