Semiparametric Estimation of Stochastic Frontiers A Bayesian Penalized Approach

Almost all previous approaches to estimating semiparametric frontier models, where the functional form for the production (cost) function is unknown, have been local nonparametric (ie. kernel) approaches. In this paper we use a penalized (ie. spline) approach. We show how this approach can be applied to a variety of frontier models, including panel models with fixed and random effects, within a Bayesian framework. We also apply our approach to different multivariate settings, including additive and additive with interaction models. The latter is a promising model because it is very flexible and does not suffer the severe curse of dimensionality problem common with fully nonparametric functions. We illustrate our method using a simulated example.