A Semiparametric Derivative Estimator in Log Transformation Models

This paper considers a regression model with a log-transformed dependent variable. The log transformed model is estimated by simple least squares, but computing the conditional mean of the dependent variable on the original scale given the explanatory variables analytically requires knowing the conditional distribution of the error term in the transformed model. We show how to obtain a consistent estimator for the conditional mean and its derivatives without specifying the conditional distribution of the error term. The asymptotic distribution of the estimator is derived. The proposed procedure is then illustrated with health expenditure data from the Medical Expenditure Panel Survey.