Abstract for: Bayesian Parameter Estimation of System Dynamics Models Using Markov Chain Monte Carlo Methods: An Informal Introduction

While calibration is an important element of the System Dynamics modeling process, traditional calibration techniques exhibit significant limitations. Many such techniques are limited to providing point estimates of calibrated values, sometimes together with information on uncertainty around such estimates. Such techniques also impose assumptions concerning the error distributions and privilege a specific dynamic model structure. Markov Chain Monte Carlo (MCMC) techniques offer a powerful, general, and versatile alternative approach. Bayesian MCMC approaches eschew point estimates, and instead provide a means of sampling from a full (“posterior”) distribution of parameter vectors. Such techniques can further express the relative likelihood of different model structures. Finally, MCMC approaches allow a modeler to explicitly specify a general probabilistic model giving the likelihood that observed empirical data would be produced by a certain parameter vector. While MCMC approaches offer strong benefits, it can be daunting for System Dynamics modelers to secure even a basic understanding of the MCMC process, and there is only a small extant literature concerning applications of MCMC to simulation models, largely using language unfamiliar to most System Dynamics practitioners. Within this paper, we seek to provide a gentle introduction to the use of Bayesian MCMC techniques for System Dynamics parameter estimation.