Robustness in Stochastic Frontier Analysis

A number of recent studies have addressed the issue of robustness in the context of stochastic frontier analysis, and alternative models and estimation methods have been proposed that appear more robust to outliers. For example, several models assuming heavy-tailed noise distributions appeared in the literature, including the logistic, Laplace, and Student’s t distributions. Despite this, there has been little explicit discussion of the what is meant by ‘robustness’ and how models might be compared in terms of robustness to outliers. This chapter discusses two different aspects of robustness in stochastic frontier analysis: first, robustness of parameter estimates, by comparing the influence of outlying observations across different specifications—a familiar approach in the wider literature on robust estimation; second, the robustness of efficiency predictions to outliers across different specifications—a consideration unique to the efficiency analysis literature.