Daouia, Abdelaati
[University of Toulouse]
Florens, Jean-Pierre
[University of Toulouse]
Simar, Léopold
[UCL]
The production/econometric frontier is the locus of the optimal combinations of inputs and outputs. From a statistical point of view, it can be viewed as the upper surface of the support of a random vector under shape constraints. In this paper we investigate the problem of nonparametric monotone frontier estimation from an extreme-values theory perspective. This allows to revisit the asymptotic theory of the popular FDH estimator in a general setup, to derive new and asymptotically Gaussian estimators and to provide useful asymptotic confidence bands for the monotone boundary function. The study of the asymptotic properties of the resulting frontier estimators is carried out by relating them to an original dimensionless random sample and then applying standard extreme-values theory. The finite sample behavior of the suggested estimators is explored through Monte-Carlo experiments. We also apply our approach to a real data set.


Bibliographic reference |
Daouia, Abdelaati ; Florens, Jean-Pierre ; Simar, Léopold. Frontier estimation and extreme values theory. STAT Discussion Papers ; 0804 (2008) 34 pages |
Permanent URL |
http://hdl.handle.net/2078.1/123114 |