Quantile regression in the consumption of luxury goods

Abstract

The multiple linear regression estimates an average trend of the observed phenomena and so it does not provide a complete illustration of relations among phenomena, while the use of quantile regression gives this aspect. The quantile regression presents the evaluation of regression models of different parts, quantiles of the conditional distribution of some series. The quantile regression is widely used in social policy, in determination of maximal production, consumption analysis, etc. This thesis presents some basic principles and the quantile regression model, which is being estimated by minimization of total absolute deviations using linear programming algorithms. The aim of this thesis is to present how the use of quantile regression provides a more complete understanding of the moving of alcoholic beverages and tobacco consumption (i.e. consumption of luxury goods). The research subject of this thesis is the linear dependence between income and alcoholic beverages and tobacco consumption of 200 households, which were randomly selected from three territories: the city of Belgrade, Central Serbia and Vojvodina in 2010. Based on these data, we estimated simple linear regression and the quantile regression in order to compare their advantages and disadvantages. It was concluded that if the consumption of luxury goods is higher, household income has no effect on the decision about the level of spending.