A Bayesian framework for modeling demand in supply chain simulation experiments

In order to postpone production planning based on information obtained close to the time of sale, decision support systems for supply chain management often include demand forecasts based on little historical data and/or subjective information. Particularly, when simulation models for analyzing decisions related to safety inventories, lot sizing or lead times are used, it is convenient to model (daily) demand by considering historical data, as well as information (often subjective) of the near future. This presents an approach for modeling a random input (e.g., demand) in simulation experiments. Under this approach, the family of distributions proposed for modeling demand should include two types of parameters: the ones that capture information of historical data and the ones that depend on the particular scenario that is to be simulated. The approach is extended to the case where uncertainty on the appropriate family of distributions is present.