Boundary Value Models for the Combination of Forecasts

Abstract

The combination of forecasts model (which is usually expressed as a linear combination of individual forecasting models) has received increasing attention in the literature. The principal advantage of forecast combination models is that they yield lower forecast errors than their individual constituent models under appropriate conditions. Two conditions of particular importance are (1) the relative accuracy of the individual forecasting models and (2) the correlations among the models’ disturbance terms. For marketing forecasting the author demonstrates that a variety of boundary value models for the forecast combination represent reasonable model alternatives to an optimal model when it is difficult to judge the relative accuracy and disturbance term intercorrelations of the individual forecasting models.