Spatial stochastic frontier models

Stochastic frontier models are one of the most frequently used approaches for estimating production function parameters and individual levels of inefficiency. It is a parametric approach and therefore depends heavily on the distribution assumptions of errors in the model. One of the main assumptions in that regard is the assumption of the independence put on the error components (random shock and inefficiency) as well as between individual inefficiencies. This allows for a simple derivation of the model likelihood and its estimation, but potentially ignores possible correlations that may happen in real life applications. In this paper I try to summarize different approaches that attempt to relax this assumption, allowing some sort of correlation between individual production units