Title: Bayesian stochastic frontiers using transformation to normal
Authors: Reza Hajargasht - Swinbuurne University of Technology (Australia) [presenting]
Abstract: The normal distribution is very convenient. In particular, one can transparently allow for correlation between normally distributed variables and also conditional distributions can be easily defined for multivariate normal distributions. One cannot, however, directly specify a normal distribution for an efficiency effect in stochastic frontier models due to the fact the distribution of the effects must be one-sided. But, it is always possible to transform a one-sided random variable with a known distribution to another that is normally distributed. The purpose is to show how using these simple facts (i.e. by transforming the efficiency effects ``ui'' to a normally distributed variable and allowing for correlation between transformed variables), one is able to handle some difficult problems in stochastic frontier analysis relatively easily. We consider problems such as stochastic frontiers with endogeneity, Stochastic frontiers with serially correlated errors and Stochastic frontier models with factor error structure and show how they can be estimated using either Bayesian or maximum simulated likelihood approaches.