p.744

Optimal Aggregation in Multi-Equation Prediction Models

Walter D. Fisher

Abstract

A theory of how best to aggregate or simplify a given detailed model of simultaneous equations in the reduced form is proposed, assuming that the model is to be used for prediction purposes. Results are obtained for aggregating exogenous or endogenous variables, or both, including the cases where the aggregation is to be in partitioning form, and where variables are to be eliminated from the model. A Bayes decision theory approach is taken.