Vector-autoregressive inference for equally spaced, time-averaged, multiple queue length processes

This paper investigates the performance of the vector-autoregressive method of analyzing multivariate output data (numbers in subsystem) from queueing network models vis-a-vis three other methods of multivariate analysis-Bonferroni batch means, multivariate batch means, and spectral analysis. Differences in performance for all methods are found when time averages of numbers in subsystem are used rather than discretized observations taken at equally spaced points in simulated time. Further investigation is made into the effect of varying the spacing of averaging times for the methods. The results show that the analysis of time averages rather than discretized observations leads to slightly improved performance for all methods considered but that there is little difference in the relative performance of the methods considered.