Stochastic Frontier Models for Discrete Output Variables.

This chapter reviews recent contributions to the area of stochastic frontiers models (SFM) for the analysis of discrete outcomes. More specifically, we discuss models for binary indicators (probit-SFM), ordered categorical data (ordered logit SMF) and discrete outcomes (Poisson SFM).

All these models are mixtures of a standard distribution with an asymmetric distribution. This allows us to frame the discussion within a general framework from which most SFM can be derived. Because many of these models might lack a closed form likelihood function, we suggest the use of Maximum Simulated Likelihoods to estimate the parameters of each model. The latter method is easy to implement in a modern computer and the unknown likelihood can be approximated with arbitrary accuracy using low discrepancy sequences such as Halton sequences.