Semiparametric stochastic frontier models: A generalized additive model approach

The choice of the functional form of the frontier into a stochastic frontier model is typically neglected in applications and canonical functions are usually considered. This paper introduces a semiparametric approach for stochastic frontier estimation that extends previous works based on pseudo-likelihood estimators allowing flexibility in model selection and capability of imposing monotonicity and concavity constraints. For these purposes the present work introduces a generalized additive framework that moreover permits to model the influence of contextual/environmental factors to the hypothesized production process by the relative extension given by generalized additive models for location, scale and shape. Through some Monte Carlo simulations and an application to European agricultural data the flexibility of the proposed framework in analyzing efficiency is illustrated.