Intermittent Demand Forecasting

5Compound Demand Distributions

5.1 Introduction

In Chapter 4, we reviewed the Poisson demand distribution and saw how it requires only one parameter to be forecasted, namely the mean demand over the protection interval. For those stock keeping units (SKUs) where the Poisson is a good fit to the demand, this will be sufficient. However, it was highlighted that the Poisson may not always be a good fit to real data because it is restricted to having its variance equalling its mean. In this chapter, we review some empirical evidence that shows the Poisson to be a good fit for a minority of SKUs only, albeit a significant minority. So, although the Poisson is useful, a stock control system for intermittent demand items should not be based on the Poisson distribution alone. Other distributions are needed.

In Chapter 6, we shall see how the forecasting of the mean demand of an intermittent series can be achieved by decomposing demand into two components: demand sizes (when demand occurs) and demand intervals (Croston 1972). This decomposition is also helpful for the characterisation of demand patterns using compound distributions, as will become evident in this chapter.

In Chapter 4, some evidence was reviewed showing that, even when the Poisson is not a good representation of demand, it may still be a good representation of demand incidence. This motivates the focus, in this chapter, on compound Poisson distributions, where demand incidence continues to be represented by the Poisson ...