Statistical Techniques for Simulation Model Validation

Statistically sound approaches to making interfrences regarding simulation model validation, rather than reliance upon subjective appeal, are needed. Fundamentally, the question of interest is whether a model reflects reality to the required degree of accuracy. The utility (and hence validity) of a simulation model relies on how well it captures those aspects of the phenonmenon under study that are relevant application of the simulation. For this reason, procedures for validating simulation models must be developed in a context specific to the properties of the phenomena intended to be capptured by the model. The work proposed herein is concerned with the development of multivariate, nonparametric statistical techniques for validation of simulation models. In particular, we will consider situations in which one or more empirical realizations (i.e., time series) of the phenomena to be modeled are availabble, as well as one or more realizations produced by the simulation model. Given two such sets of time series, we wish to determine whether the simulated series accurately capture the behavior of the empirical series. That is, we wish to perform a test of homogeneity. Two general statistical approaches to this problem will be considered: nonparametric homogeneity testing and change point detection.