Top-down or bottom-up: Aggregate versus disaggregate extrapolations

Abstract

Two approaches have been suggested for forecasting items in a product line. The top-down (TD) approach uses an aggregate forecast model to develop a summary forecast, which is then allocated to individual items on the basis of their historical relative frequency. The bottom-up (BU) approach employs an individual forecast model for each of the items in the family. The present study compares these two approaches by using over 15,000 aggregate series constructed by combining individual series from the M-competition database. The effects of correlation between individual items and the relative frequency of individual items in the family are examined. In most situations, BU forecasting of family items produces more accurate forecasts.