Assessing models in frontier analysis through dynamic graphics

The use of stochastic frontier models to perform the empirical analysis of efficiency and productivity has recently become very widespread, developing into a very large literature. Within this literature, the development of the body of techniques known as model validation has remained somehow marginal. This is unfortunate, because in most cases stochastic frontier analysis cannot rely on the assumption of normality of residuals, making usual specification tests inappropriate, at least in small samples. In this paper, we first provide an overview of graphical methods which are particularly useful for the purpose of model validation. Particular emphasis is bestowed upon a relatively novel procedure, the model checking plot. We discuss then how dynamic graphics can shed light in specification issues related to a representative stochastic frontier model, and present some empirical examples of the application of these procedures.