Application of the ML Hausman approach to the demand of water for residential use: heterogeneity vs two-error specification

This paper presents an application of two ML models to the analysis of residential demand of water - the heterogeneity and the two-error model, both apt to model demand in presence of a kinked budget constraint. The heterogeneity model is especially suitable when the distribution is characterized by a strong clustering around the kinks. Since in practice observations can be very close, but not exactly at the kink, its application may require the definition of an interval of data around the kink, so that the observations falling inside this interval are attributed to the kink. We propose a procedure, based upon the estimates obtained from the twoerror model, to define this interval. In this application we find that the heterogeneity model allows to obtain more efficient estimates than the two-error model for the parameter of principal interest, i.e. the coefficient of the price variable.