Accounting for input model and parameter uncertainty in simulation

Taking into account input-model, input-parameter, and stochastic uncertainties inherent in many simulations, our Bayesian approach to input modeling yields valid point and confidence-interval estimators for a selected posterior mean response. Exploiting prior information to specify the prior plausibility of each candidate input model and to construct prior distributions on the model's parameters, we combine this information with the likelihood function of sample data to compute posterior model probabilities and parameter distributions. Our Bayesian Simulation Replication Algorithm involves: (a) estimating parameter uncertainty by sampling from the posterior parameter distributions on selected runs; (b) estimating stochastic uncertainty by multiple independent replications of those runs; and (c) estimating model uncertainty by weighting the results of (a) and (b) using the corresponding posterior model probabilities. We allocate runs in (a) and (b) to minimize final estimator variance subject to a computing-budget constraint. An experimental performance evaluation demonstrates the advantages of this approach.