Template-type: ReDIF-Paper 1.0 Author-Name: Daniel Santín González Author-Workplace-Name: Departamento de Economía Aplicada, Pública y Política. Universidad Complutense de Madrid. Author-Workplace-Homepage: https://www.ucm.es/eapp Author-Name: Aurelia Valiño Castro Author-Workplace-Name: Departamento de Economía Aplicada, Pública y Política. Universidad Complutense de Madrid. Author-Workplace-Homepage: https://www.ucm.es/eapp Title: Comparing neural networks and efficiency techniques in non-linear production functions Abstract: Non-linear production functions are common in economic theory and in real life, especially in cases with increasing and diminishing returns to scale but there are also contexts where an increase in one input implies a decrease in one output. The aim of this paper is to test how non-linearity affect estimations of technical efficiency obtained by ordinary and corrected least squares (OLS, COLS), data envelopment analysis with constant and variables returns to scale (DEAcrs, DEAvrs), stochastic frontier analysis (SFA) and by multilayer perceptron neural networks with backpropagation (MLP). To do this we will construct a very simple non-linear one input-one output production function and we will obtain different synthetic data with 50, 100, 200 and 300 decision-making units (DMUs). Afterwards we will add up different quantities of noise to the data and finally we will compare real efficiency with estimated values for all techniques named before among the different scenarios. Our results suggest that MLP is a flexible tool to fit production functions and a possible alternative to traditional techniques under non-linear contexts. Keywords: Análisis funcional no lineal, Non-linear production function, Technical efficiency, Artificial neural networks. Length: 12 pages Creation-Date: 2002 Number: 02-02 X-File-Ref: http://america.sim.ucm.es/repec/ucm/ref/doctra02-02.txt File-URL: https://eprints.ucm.es/id/eprint/6763/1/0202.pdf File-Format: Application/pdf Handle: RePEc:ucm:doctra:02-02