Abstract: Making decisions about inventory for items with intermittent demand poses a significant challenge. When deciding to discontinue sales of a product, decisions often hinge on average demand estimates, while replenishment decisions rely on intervals of demand estimates rather than single points. Addressing the modeling of intermittent demand becomes crucial in this context. Previous studies have approached this issue using generalized linear models or integer-valued autoregressive moving average (ARMA) models. However, the development of models within the state space framework has yielded varied results. In this paper, we introduce a comprehensive state space model that accommodates intermittent data, expanding the range of single-source error state space models. We demonstrate that our model shares connections with traditional non-intermittent state space models commonly used in inventory planning. Some variants of our model can be estimated using forecasting methods such as Croston’s method and the Teunter–Syntetos–Babai (TSB) method. We analyze the properties of our proposed models and illustrate how to choose among them within our framework. Additionally, we conduct a simulation experiment to empirically assess the inventory implications of our approach.

This talk is based on this paper.