A note on functional form specification in random coefficients stochastic frontier models

Abstract

This study presents a random coefficients stochastic frontier model that can accommodate the flexible translog functional form without being computationally demanding and thus time consuming to estimate. This is achieved by restricting the second-order frontier parameters to be common to all firms. For comparison, random coefficients stochastic frontier models with Cobb–Douglas, semi-translog and translog specifications with all parameters being firm-specific are estimated. The models are applied to an unbalanced panel of German dairy farms, and Bayesian techniques are used for the estimation. The results suggest that the time needed for the sampler to complete in the proposed model reduces dramatically as opposed to a translog model where all parameters are firm-specific. The elasticities exhibit some differences, depending on the choice of functional form, whilst the efficiency scores are less affected. Bayes factors suggest that the proposed model fits the data best.