An IIF workshop “Supply Chain Forecasting for Operations” took place at Lancaster University on 28th and 29th of June. I have given a presentation on a topic that John Boylan and I are currently working on. We suggest a universal statistical model, that allows uniting standard methods of forecasting (for example, for fast moving products) and methods for intermittent demand. It is done via introduction of binary variable that characterises occurrence of demand. The similar approach is used in Croston’s method and TSB, however we propose a statistical model, potentially underlying many more forecasting methods. This model allows correctly estimating parameters of aforementioned Croston and TSB, constructing prediction intervals and selecting the most appropriate model for any time series.

The discussed models are already implemented in function es() of my package smooth for R. All you need to do is provide intermittent data and ask for a specific model type via intermittent parameter (either “fixed”, “croston”, “tsb” or “auto”).

And here are the slides.