Dr. Dave Goldsman
School of Industrial and Systems Engineering, Georgia Tech
Talk is at 11:00 AM Central Time (calculating your local time…)
Abstract
Simulation analysts routinely estimate model parameters from data and then substitute those estimates into subsequent analyses. While this “plug-in” paradigm is nearly universal, its consequences can have significant implications for decision making. We examine these consequences in a collection of stochastic models driven by exponential distributions, where exact finite-sample analysis is possible. Using the maximum likelihood estimates of exponential rate parameters, we derive exact distributions and moments for plug-in survival estimators, generated random variables, competing exponential processes, insurance valuation models, M/M/1 queueing performance measures, and network travel times.
We show that parameter estimation can introduce bias, inflate variability, change distributional forms, and induce dependence among observations that would otherwise be independent. These effects can be especially pronounced in nonlinear performance measures such as queueing metrics, where rare estimation errors may have disproportionate impact. The exponential setting provides an analytically tractable laboratory in which the propagation of estimation uncertainty can be understood in detail.
Our results demonstrate that even seemingly innocuous parameter estimation can fundamentally alter the behavior of a stochastic model. More broadly, the paper highlights how input uncertainty propagates through simulation models and shows that, even in the remarkably basic exponential setting, plug-in analyses can yield behavior that differs qualitatively from that of the corresponding known-parameter system.
Gang of Co-Authors:
Joseph Bakhtiar, David Goldsman, Luis Herrera, Paul Horton, H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, USA.
John-Paul Clarke, Department of Aerospace Engineering and Engineering Mechanics, University of Texas at Austin, USA.
Kemal Dingec, Department of Industrial Engineering, Gebze Technical University, Turkey.
James R. Wilson, Edward P. Fitts Department of Industrial and Systems Engineering, North Carolina State University, USA.