Identifiability of the Stochastic Frontier Models

This paper examines the identifiability of the standard single-equation stochastic frontier models with uncorrelated and correlated error components giving, inter alia, mathematical content to the notion of “near-identifiability” of a statistical model. It is seen that these models are at least locally identifiable but suffer from the “near-identifiability” problem. Our results also highlight the pivotal role played by the Signal to Noise Ratio in the “near-identifiablity” of the stochastic frontier models.