Model selection when estimating and predicting consumer demands using international, cross section data

This paper assesses the ability of five structural demand systems to predict demands when estimated with cross sectional data spanning countries with widely varying per capita expenditure levels. Results indicate demand systems with less restrictive income responses are superior to demand systems with more restrictive income effects. Among the least restrictive demand systems considered, An Implicitly, Directly Additive Demand System (AIDADS) and Quadratic Almost Ideal Demand System (QUAIDS) seem roughly tied for best, while the Quadratic Expenditure System (QES) is a close second. Given differences in the characteristics of AIDADS and QUAIDS, it is concluded the former is better suited to instances where income exhibits wide variation and the latter to cases when prices exhibit considerable variation.