Reliable prediction intervals and Bayesian estimation for demand rates of slow-moving inventory

Inventory having intermittent demand has infrequent sales that appear at random, with many periods that do not show any demand at all. Managing inventory with intermittent demand has received less attention in the literature than that of fast-moving products. This is due in part, perhaps, to the lack of observable historical sales figures for inventory with intermittent demand or because slow-moving inventory does not provide the bulk of sales, despite often being the bulk of inventory on hand.;Inventory management tools are proposed that provide estimation procedures for the future demand rates of inventory with intermittent demand. Prediction intervals, adapted from statistical procedures developed for software reliability, for the future demand rate of a group of products that have no sales or no more than one sale over a specified time frame are proposed. A Monte Carlo simulation study is conducted to assess the reliability of these prediction intervals across various sizes of product groups and demand rates as well as for mixtures of demand rates and identify reliable parameter ranges. Sales data from a Fortune 500 company were used to assess the performance of the proposed prediction intervals.;Inventory managers periodically update their predictions of future demand rates for products. Two models---a Bayes model, using a prior probability distribution for the demand rate and a Poisson model, using a Poisson distribution for demand---were used to obtain optimal inventory levels over several periods assuming a known cost for surplus and shortage. This procedure has been proposed in the literature. However, its performance has not been examined under various demand rates such as intermittent demand.;A Monte Carlo simulation study was used to examine the performance of the Bayes and Poisson model under moderate and intermittent demand. When the demand rates of the products are homogeneous, the inventory costs related to the Bayes model is lower than that of the Poisson model. The Poisson model is preferred under conditions of high variability among product demand rates. An improvement that optimized inventory costs for some demand rates was made to the Bayes model using a mixture of priors.