Input uncertainty in stochastic simulation

Stochastic simulation requires input probability distributions to model systems with random dynamic behavior. Given the input distributions, random behavior is simulated using Monte Carlo techniques. This randomness means that statistical characterizations of system behavior based on finite-length simulation runs have Monte Carlo error. Simulation output analysis and optimization methods that account for Monte Carlo error have been in place for many years. But there is a second source of uncertainty in characterizing system behavior that results from error in estimating the input probability distributions. When the input distributions represent real-world phenomena but are determined based on finite samples of real-world data, sampling error gives imperfect characterization of these distributions. This estimation error propagates to simulated system behavior causing what we call input uncertainty. This chapter summarizes the relatively recent development of methods for simulation output analysis and optimization that take both input uncertainty and Monte Carlo error into account.