The aim of the proposal is to develop statistical methods useful for inference on systems of demand equations. Such a system describes how household expenditure (or demand) is distributed over goods and services and how the distribution varies with household income as well as household demographics. Understanding and measuring consumer behavior is essential for planning and policy making. Estimates of demand functions are necessary inputs to these kinds of analyses, as they provide the link between theoretical models of consumer behavior and their real behavior.
This project especially focuses on methods to estimate demand equations using data from Statistics Sweden's annual survey on household expenditures. The estimation is complicated by observations of households with zero expenditure on one or several goods and/or services. This is called censoring and biases standard regression estimation methods. Moreover, a system of demand equations should be estimated instead of one equation at a time since demand of a good/service is affected by the demand of other goods/services via the budget restriction.
We suggest the use of a finite mixture of multivariate Tobit models which addresses both censoring and the budget restriction. This approach will be less sensitive to distribution assumptions then earlier suggested methods. The project provides with theoretical and empirical results on the properties of the estimator, and a comparison is made with estimators earlier suggested.