Forecasting seasonal demand : a serious limitation of Winters' forecasting procedure and the added value of product-aggregation

The well-known method for forecasting seasonal demand, Winters’ procedure, has a serious drawback: if the relative demand uncertainty increases (e.g. due to larger product assortments) or if the amount of historical demand data decreases (e.g. due to smaller product life cycles), the quality of the forecasts deteriorates. In this paper mathematical models are created to quantify these effects. These models help researchers, teachers and practitioners to better understand why and when Winters’ forecasting procedure may deteriorate. One way to improve again the performance of Winters’ procedure may be to use the concept of product-aggregation: if different products have a similar seasonal pattern, the seasonal indices from Winters’ method can be determined from the product family’s aggregate demand. Mathematical modelling as well as simulation is used to assess the added value of product-aggregation. It turns out that impressive improvements can be achieved, especially in case demand uncertainty is high or (when forecasting is applied in inventory systems) in case the lead times are large.