Sequential Procedure for Simultaneous Estimation of Several Percentiles

Percentiles are convenient indices to characterize the entire range of the values of simulation outputs. However, they have only seldom been used in simulation studies. One of the reason has been the lack of sequential estimation procedures, which are needed to obtain estimates of predefined accuracy. In this paper we introduce a sequential procedure for estimating several percentiles simultaneously. The procedure uses the extended P 2 algorithm to estimate the percentiles. The variances of the percentile-estimates are estimated using a spectral method. Since the method applies a Bonferroni inequality, the covariances between the percentile-estimates are not needed. The procedure is shown to produce estimates having the predefined accuracy in eight queueing network models, representing multiprogrammed and time-shared computer systems. Key words: Percentile Estimation, Variance Estimation, Run Length Control 1. Introduction In most simulation studies the mean values of output data hav..