CMStatistics 2022: Start Registration
View Submission - CFE
A0977
Title: Distributional assumptions and the sensitivity of stochastic frontier efficiency predictions Authors:  Alex Stead - University of Leeds (United Kingdom) [presenting]
Phill Wheat - University of Leeds (United Kingdom)
William Greene - New York University (United States)
Abstract: Efficiency scores may be used to inform important regulatory, managerial, or policy decisions. In these and other applied settings, it is desirable that they be robust to small changes in the sample. However, it is well known that under standard distributional assumptions, even a single observation can have a dramatic effect on efficiency predictions, e.g. if it leads to 'wrong skewness'. Recent findings show that alternative distributional assumptions can improve the robustness of parameter estimates. We derive influence functions for the conditional mean efficiency predictor, and show that efficiency predictions from robust specifications are less sensitive to contaminating observations than those from non-robust specifications. We also discuss important differences in the way the models handle outliers with respect to prediction uncertainty.