Robust maximum likelihood estimation of stochastic frontier models

When analysing the efficiency of decision-making units, robustness scores to changes in data is desirable, especially context managerial or regulatory benchmarking. However, maximum likelihood estimation stochastic frontier models remains underexplored. We examine behaviour influence function estimator a context, and derive some sufficient conditions for robust terms properties marginal distributions error components and, cases where they are dependent, copula density. find that canonical distributional assumptions do not satisfy these conditions. The Student’s t noise distribution found have particularly attractive which means it can be paired with broad class inefficiency while still satisfying our under independence. show parameter estimates predictions from specifications significantly less sensitive contaminating observations than those non-robust specifications.