Some Model Specification Issues in Applied Demand Analysis

Abstract

The availability of large cross-sections of household expenditure microdata has given new impetus to applied demand research. These data permit modelling detailed household characteristics effects, empirical analyses of the conditions under which micro parameters can be recovered using aggregate data, estimation of household equivalence scales, examination of the importance of labour force variables as conditioning variables in demand models, and the generalization of model specifications to rank three systems. Examples of this research include Blundell, Pashardes and Weber (1993), who find that aggregate data alone cannot yield reliable estimates of structural price and income coefficients. This is demonstrated with a demographically flexible model, using British Family Expenditure Survey (FES) data. They also find, however, that aggregate data can still provide reasonable forecasts, so long as these models contain aggregation factors, trend and seasonal components. Conditions for the identifiability of general household equivalence scales have been explored by Lewbel (1989). Deaton and Muellbauer (1986) and Nicol (1994) present theoretical and empirical analyses of equivalence scales, and discuss the inherent difficulties in estimating these. The availability of microdata has also permitted the use of nonparametric techniques to explore the shape of demand functions. These studies have found that functional forms vary from equation to equation in a demand system. This provides insights into how improved specifications of parametric models might be formulated, as in Bierens and Pott-Buter (1990). Some other work has shown that conditions for perfect aggregation are not met empirically (see Browning and Meghir 1991; Nicol 1993). In particular, models which have demographic effects interacting with income are not exactly aggregable. Models of this type are now quite commonly used in the empirical demand literature.