Abstract.
In present days it is commonly recognized that firm production datasets are affected by some level of random perturbation, and that consequently production frontiers have a stochastic nature. Mathematical programming methods, traditionally employed for frontier evaluation, are then reputed capable of mistaking errors for technical (in)efficiency. Therefore, recent literature is oriented towards a statistical view: frontiers are designed by enveloping data that have been preliminarly filtered from noise.
In this paper a nonparametric smoother for filtering panel production data is presented. We pursue a recent approach of Kneip and Simar (1996), and frame it into a more general formulation whose a setting constitutes our specific proposal. The major feature of the method is that noise reduction and outlier detection are faced separately: i) a high order local polynomial fit is used as smoother; and ii) data are weighted by robustness scores. An extensive numerical study on some common production models yields encouraging results from a competition with Kneip and Simar’s filter.
Similar content being viewed by others
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Di Marzio, M. On panel data filtering in technical efficiency estimation. Statistical Methods & Applications 12, 319–329 (2004). https://doi.org/10.1007/s10260-003-0071-1
Issue Date:
DOI: https://doi.org/10.1007/s10260-003-0071-1