Estimation in semi-parametric transformation boundary regression models
vendredi 15 mars 2019, 9h30 - 10h30
Boundary regression models naturally arise in many applications for instance when analysing auctions or records but also in production frontiers and image analysis. Before fitting a regression model it is very common to transform the response variable to gain efficiency in the statistical inference. In this talk, we will consider parametric transformations that induce independence of the error distribution from the points of measurements. In such a context, if the transformation of the response is monotone, the attractive feature is that one may recover the original functional dependence in an easy manner. The main purpose of this talk is to investigate the consistency of an estimator based on a minimum distance approach in the context of nonparametric regression models with one-sided errors. In particular, we estimate the transformation parameter and give mild model assumptions under which the estimator is consistent, for both ran- dom covariates and fixed design points. The small sample behavior will be shown in a simulation study using the so-called Yeo-Johnson transformations.
Keywords. Nonparametric regression, frontier estimation, minimum distance estimation, boundary models, local constant approximation, Yeo-Johnson transformations.