On the approximation of production functions: a comparison of artificial neural networks frontiers and efficiency techniques

Abstract

The aim of this article is to show how Artificial Neural Networks (ANN) is a valid semi-parametric alternative for fitting empirical production functions and measuring technical efficiency. To do this a Monte-Carlo experiment is carried out on a simulated smooth production technology for assessing efficiency results of ANN compared with Data Envelopment Analysis (DEA) and Stochastic Frontier Analysis (SFA). As ANNs provides average production function estimations this article proposes a so-called thick frontier strategy for transform average estimations into a productive frontier. Main advantages of ANN are in contexts where the production function is smooth, completely unknown, contains nonlinear relationships among variables and the quantity of noise and efficiency in data is moderate. Under this scenario, the results display that an ANNs algorithm can detect, better than traditional tools, the underlying shape of the production function from observed data.

Acknowledgements

This article owes a debt to Francisco Pedraja-Chaparro, Abel Santin and Aurelia Valiño for valuable comments, help and suggestions.