In the present study, we have investigated several competing stochastic frontier models which differ in terms of the form of the production function (Cobb-Douglas or translog), inefficiency distribution (half-normal or exponential distribution) and type of prior distribution for the parameters (hierarchical or non-hierarchical from the Bayesian point of view). This last distinction corresponds to a difference between random coefficients and fixed coefficients models. Consequently, this study aims to examine to what extent inferences about estimates of farms' efficiency depend on the above assumptions. Moreover, the study intends to investigate how far the production function's characteristics are affected by the choice of the type of prior distribution for the parameters. First of all, it was found that the form of the production function does not impact the efficiency scores. Secondly, we found that measures of technical efficiency are sensitive to distributional assumptions about the inefficiency term. Finally, we have revealed that estimates of technical efficiency are reasonably robust to the prior information about the parameters of crop farms' production technology. There is also a resemblance in the elasticity of output with respect to inputs between the models considered in this paper. Additionally, the measurement of returns to scale is not sensitive to model specification.
Nonparametric Tests of Tail Behavior in Stochastic Frontier Models
