Modelling Scale Effect in Cross-Section Data: The Case of Hedonic Price Regression

Abstract

An innovative and simple experiment with cross-section data ordering is carried out to exploit a basic feature between many economic variables – nonlinear scale dependence. The experiment is tried on hedonic price models using two data sets: automobiles and computers. Our key findings are: (a) Economic knowledge based data ordering offers considerable potential to filter scale-dependent information between the modelled variable and key explanatory variables from cross-section samples; (b) The filtering can be easily carried out by systematic adoption of dynamic modelling methods developed originally for time-series data, once cross-section data have been ordered; (c) The consequent information gain is much greater than that gained by semi-parametric estimation or quadratic specification of the traditional hedonic model; (d) The hedonic price indices derived from our experiment deviate significantly from those conventionally constructed indices, indicating the latter being systematically biased due to mis-specified scale-dependence of the traditional model.