Analysis of Simulation-Generated Responses using Autoregressive Models

Abstract

Insights concerning a system are often provided by comparisons of historical versus simulated data. Such insights offer valuable assistance in the evolution of the simulation modeling process. In this article, a testing procedure is suggested for comparing historical time series records against responses generated by a simulation model. Time series models are identified and their parameters estimated for both the historical and simulated data using the techniques outlined by Box and Jenkins. The models are then tested for differences in their means, autoregressive parameters, and residual variances employing Bayesian arguments. The procedure suggested is emphasized as a diagnostic instrument useful for validation, and for suggesting iterative modifications of a simulation model. Applications of the procedure are illustrated by an example involving air traffic control communications.