Fashion Forecasting for a Mail Order Company Using a Bayesian Approach

Traditional forecasting methods assume a large amount of product history. New product launches take place in the absence of any product sales statistics and initially it is necessary to formulate beliefs as to the product future and then to combine these with sales data as it becomes available. The particular situation considered in this paper concerns a mail order company which sells a different group of ladies dresses on each of its catalogues. The life of a catalogue is of the order of 6 months and material ordering decisions tend to be relevant only at the beginning and during the first few weeks of the catalogue. The task of the Distribution Department is not helped by the presence of a large number of returns from customers. This paper describes a Bayesian approach to the forecasting problem and the "live" performance of the method on one catalogue is discussed.