On the performance of binomial and beta-binomial models of demand forecasting for multiple slow-moving inventory items

The paper deals with the lead-time demand forecasting for inventory management of multiple slow-moving items in the case when the available demand history is very short. Two stochastic models of demand are compared: (i) the first based on the “population-averaged” binomial distribution of requests (the traditional approach); and (ii) the second based on the beta-binomial probability distribution that assumes that the demand probabilities for the inventory items follow the beta distribution, and employs the Bayesian framework to forecast the lead-time demand (longitudinal statistical approach). The conducted simulation study shows that using the latter model leads to the significant decrease of the holding cost and higher inventory system reliability. Besides, as follows from the simulation results, the beta-binomial demand model is especially useful when the demand probabilities are low and have the U-shaped distribution.