Abstract
We provide a review of the literature related to the “wrong skewness problem” in stochastic frontier analysis. We identify two distinct approaches, one treating the phenomenon as a signal from the data that the underlying structure has some special characteristics that allow inefficiency to co-exist with “wrong” skewness, the other treating it as a sample-failure problem. Each leads to different treatments, while siding with either raises certain methodological issues, and we explore them. We offer simulation evidence that the wrong skewness as a sample problem likely comes from how the noise component of the composite error term has been realized in the sample, which points towards a new way to handle the problem. We also investigate the issues that arise when attempting to use the unconstrained Normal-Half Normal (Skew Normal) likelihood.