Measuring Inefficiency in Public Utilities: Does the Distribution Matter?

In this paper we describe two of the most used distributions for the inefficiency term in a stochastic frontier: the Half Normal and the Exponential distributions. We then show that both distributions affect the skewness of the composite error term in different ways, which make the Exponential distribution more inclined to identifying a larger number of efficient firms. In order to check these results we perform MOLS and ML, using four databases already used in previous work. We find that the mean efficiency is sensitive to the assumed distribution. In all cases we corroborate that the Exponential distribution identifies a larger number of efficient firms than the Half Normal, as the theory predicted. However, the rankings of firms by efficiency scores were not affected by the choice of distribution.