What is the leading expert opinion on simulation input modeling?



We conduct a series of in-depth semi-structured interviews with leading experts in simulation input modelling to establish an understanding of what they recommend for this modeling. This research is positivist whilst the recommendations from experts interviewed are normative. Several notable findings stand out from these expert interviews. Correlation and related methods are considered only if direct modelling of causal dependence is impossible. We find a strong preference for theoretically grounded models for distribution selection over empirical data. This extends to causal dependence models that are deemed superior to correlation modelling. Moreover, there is a universal level of awareness of cognitive biases among experts. Sophisticated as well as pragmatic, de-biasing strategies are used for model parameterization. Further, a relatively small number of distributions is deemed sufficiently realistic for most purposes. Finally, we recognize that Mixed Methods of model parameterization where multiple data sources are combined to create estimates, are widely supported amongst the experts in our sample.

Bayesian updating for simulation input modeling


Precise and forward looking simulation input modelling is pivotal to achieve accurate simulation modelling results and therefore sound decision making. We present an input modelling method that allows the aggregation of quantitative empirical data and quantified expert opinions via the process of Bayesian updating of prior distributions with expert input. Through this combination of input sources, the entropy reducing information of each input source is utilized through a formal method to reduce the uncertainty about the unknown estimated input parameters. While various methods exist to parameterize simulation models, many face limitations under realistic assumptions. Simulation modelers use empirical data for simulation model parameterization, yet this method faces limitations if the modeled process undergoes changes or when there are various viable sources of data on the modelled process.

By using input from forward looking experts for model parameterization, modelers can attempt to overcome such limitations though at the cost of risking other biases. The simulation model parameterization method presented here seeks to overcome the downsides of both these methods. For this method, we draw on a large existing body of research from academic fields as diverse as actuarial sciences, reliability engineering and Kalman filtering. We derive exemplary applications challenges in simulation input modelling to demonstrate the method. In a final benchmark to alternative model parameterization and data aggregation methods, we demonstrate the impact of the varying methods.