Adapting the Bootstrap to Generate Realistic Simulation Systems *

Omer Demirel & Thomas Willemain

Monte Carlo analyses, such as a discrete event simulation of a factory typically rely on built-in input generators. These generators make unrealistic assumptions about the independence and marginal distributions. The alternative source of inputs, historical ‘trace’ data, though realistic by definition, provide only a single input stream for driving the simulation. The bootstrap provides a way to convert the one trace into multiple inputs. We report on work in progress to adapt the bootstrap to the generation of realistic inputs. We describe a quantitative measure of difference between two time series, how this measure can be used to ‘tune’ the bootstrap, and new results on how to perform base the tuning on a single observed trace.