Stochastic Frontier Analysis with Fat-Tailed Error Models

The stochastic frontier analysis (Aigner et al. [1] and Meeusen and van den Broeck [8]) has been widely used to estimate technical efficiency of firms. The basic idea lies in the introduction of a composed error term consisting of a noise v and an inefficiency term u. From there, technical efficiency of each firm is estimated by utilizing distributional assumptions on the two error components. In the literature, v is usually assumed to be normally distributed and the distribution of u can be exponential, truncated normal or Gamma. In this study, we will consider other models which are more realistic than the existing models in accounting for heavy tail data and in allowing flexibility in the shape of the distribution of the composed error term under both cross-sectional and panel data. We apply the models to real data sets for illustration