Abstract

Demand forecasting with regard to stock control is a central issue of inventory management. Serious difficulties arise for intermittent demands, that is, if there are slow-moving items demanded only sporadically. Prevalent methods then usually perform poorly as they do not properly take the stochastic nature of intermittent demand patterns into account. They often rely on theoretically unfounded heuristic assumptions and apply inappropriate deterministic smoothing techniques. We overcome these weaknesses by means of systematically built and validated stochastic models that properly fit to real (industrial) data. Initially, no assumptions are made but statistical methods are invoked for model fitting. Reasonable model classes are found by summary statistics and correlation analysis. Specific models are obtained by parameter estimation and validated by goodness-of-fit tests. Finally, based on the stochastic models, stock control strategies are proposed to facilitate service levels guarantees in terms of probability bounds for being out of stock.