National and Kapodistrian University of Athens
Goodness–of–fit tests in stochastic frontier models
Abstract
dc:descriptionStochastic Frontier Models (SFMs) are widely used in efficiency analysis, where they help to estimate the efficiency of decision-making units. In contrast with conventional zero-mean error term models, the fundamental concept behind SFMs is the existence of an asymmetric, non zero-mean composed error term. This composed error term consists of two random variables – representing random noise and inefficiency – and the identification of a SFM relies heavily on the distribution assumptions for each of these variables. In the literature goodness-of-fit (GoF) tests have been developed for ensuring that the estimated models accurately represent the data, and consequently providing reliable efficiency measures. However, most of these tests are not omnibus, i.e. they may have negligible power against certain alternatives. Our approach is to define omnibus tests for the parametric assumptions, that is procedures which, being based on consistent tests, enjoy non-negligible power for arbitrary deviations from the null model, and not just for directive alternatives. Specifically, we propose GoF tests formulated as a weighted integral of properly standardized data for the distribution of the composed error term. Instead of the classical tools of moments, densities or cumulative distribution functions, our approach is based on the distributional transforms, such as the characteristic function, the moment generating function and the Laplace transform. To this end, since the composed error term in the SFM framework is always defined as the sum of two random variables, we capitalize on the property that for these transforms, the transform of the sum of independent components is equal to the product of the transforms of the individual components. A variety of models are considered in this work, including the normal/exponential, normal/gamma, and stable/gamma frontier models. We examine the commonly assumed models of normal/exponential and normal/gamma with fixed gamma shape parameter. The new test statistics are shown to be consistent and computationally convenient, and bootstrap versions of the tests are shown to have competitive power compared to the classical Kolmogorov-Smirnov and Cramer-von Mises tests in finite samples. Following, we examine the more complicated – in terms of the underlying distribution function – cases, those of normal/gamma with both gamma parameters unknown and the stable/gamma models. Despite their complex functional form, the proposed tests are computationally efficient as they avoid numerical integration, making them easier to implement compared to traditional distribution function-based tests. The proposed normal/gamma test is consistent, and it is shown to have an intrinsic relation to moment-based tests, while the proposed stable/gamma test provides a method through which allows to distinguish between a medium-tailed normal/gamma and a purely heavy-tailed stable/gamma SFM. The application of our methods on real-data serves the purpose of investigating the presence of heavy–tails in the empirical modelling of production, where our results lead to non-rejection of a normal/gamma SFM null hypothesis. In summary, our approach provides significant advancements in the form of new statistical tests for SFMs.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papadimitriou, Christos
- Παπαδημητρίου, Χρήστος
Subjects
dc:subject × 24- Έλεγχοι καλής προσαρμογής
- Χαρακτηριστική συνάρτηση
- Ροπογεννήτρια συνάρτηση
- Παραμετρική bootstrap
- Εκτίμηση μέγιστης πιθανοφάνειας
- Κανονική/Εκθετική κατανομή
- Κανονική/Γάμμα κατανομή
- Συμμετρική κατανομή
- Goodness–of–fit tests
- Characteristic function
- Moment generating function
- Parametric bootstrap
- Maximum likelihood estimation
- Normal/exponential distribution
- Normal/gamma distribution
- Stable distribution
- Οικονομικά και Επιχειρήσεις
- Κοινωνικές Επιστήμες
- Οικονομετρία
- Στατιστικές, Πιθανότητες και Αβεβαιότητα
- Economics and Business
- Social Sciences
- Econometrics
- Statistics, Probability and Uncertainty
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58356