Comparing Nonparametric Regression Quantiles

This paper investigates how conditional quantiles of a given distribution relate to each other. Given two conditional quantiles estimated nonparametrically, we investigate their relation by linking them through a parametric transformation. Asymptotic normality of the associated parameter vector is established, and the method is illustrated with data from the Family Expenditure Survey (FES) of UK households. The FES records expenditures of households on six broad categories of goods (alcohol, clothing, food, fuel, transport, and "other goods"), and the methodology is applied by estimating and comparing the conditional quantiles of the Engel relation. The only category for which expenditure can explain the shift in the quantile curves is for "other goods" relationship, indicating an increase in heterogeneity for better off households, suggesting a "taste for variety" effect as the expenditure level increases. For the remaining categories one cannot reject the null of a parallel shift of the quantile curves