|
Abstract:
|
Simple distributional models, perhaps based on a PDF, may not be sufficiently complex to adequately model the underlying mechanisms of many systems of interest. Simulation models are often more useful for studying these complex systems. A simulation model, just as a PDF, has parameters, and the statistical task in either case is to estimate those parameters using observed data. We propose estimators that are based on comparisons of the empirical CDF of the observed data and the simulated CDF, which is a function of the parameters of the simulation model. We consider various measures of discrepancy between the ECDF and the simulated CDF. The estimators of the model parameters are defined as the values that minimize the discrepancy. Simulation studies on a Normal random number generator are provided to illustrate our method.
|
