Evaluating the semi-nonparametric Fourier, AIM, and neural networks cost functions

This study compares how well three semi-nonparametric functions, the Fourie r flexible form, asymptotically ideal model, and neural networks, approxima te simulated production data. Results show that higher order series expansi ons better approximate the true technology for data sets that have little o r no measurement error. For highly nonlinear technologies and much measurem ent error, lower order expansions may be appropriate. (C) 2000 Elsevier Sci ence S.A. All rights reserved. JEL classification: C14; D12.