We consider three methods based on advance demand information. Each of these methods ?rst forecasts total season demand in the upcoming season, denoted by M, for a group of SKUs N by scaling up the registered advance (preview) demands for those SKUs, and then divides this forecasted group demand over the individual SKUs. The scaling up factor is calculated as the ratio of ?nal demand to preview demand for a ‘comparable' group of SKUs (e.g. t-shirts), denoted by H, in one or more historical seasons.

Using notation Pn for the preview demand in the new season for SKU n 2N;H for the number of SKUs in H;R for the preview demand in the historic season(s) for SKU h 2 H, and Sh for the total demand in the historic season(s) for SKU h 2 H, this gives

So given preview demand, we forecast total demand by assuming that the ratio of total demand to preview demand will be the same as in past season(s) for a comparable group of SKUs.We remark that this forecast could be modi?ed if additional information on e.g. the economical situation or meteorological conditions were available.

All methods can be applied for any choice of grouping. Intuitively, it makes sense to group SKUs in such a way that the SKUs in N have similar product characteristics as the SKUs in H. Note that in order to obtain a decent estimate of M, it is required that SKUs in historical season(s) can be found that bear suf?cient resemblance to the SKUs in N. In our numerical investigation, we will consider several ways of grouping in line with classi?cations used by the case company. Collections change every selling season to follow the latest fashion and trends. Hence, there will generally be no overlap between H and N, although there might be a group of generic SKUs that are carried over from one season to the next. While individual SKUs change, the de?nitions of groups and the classi?cation of SKUs into these groups do not change.