서강대학교

초록/요약

This paper is aim to compare the properties of estimators under a new distribution specification in the stochastic frontier model with ones under the half-normal distribution specification. Many studies related to the stochastic frontier model used the half-normal distribution specification for the inefficiency random variable, which has only nonnegative support. Since many nonnegative random variables have long-tails and unbounded support, we consider the uniform distribution for the inefficiency random variable in this paper. With this specification, we are allowed the support of the random component to be restricted and thus the degree of inefficiency is bounded. Also, we may get additional information from the estimator, dispersion of efficient level. In the case that the model is correctly specified, the estimators derived from uniform specification are more accurate when the sample size is small. As the sample size increases, this benefit becomes smaller. When the model is not correctly specified, there is a huge distortion between constant term and inefficiency error term. This distortion is not vanished even if the sample size becomes larger. When the variance ratio is increased, the biases of constant term and inefficiency error term are more severe in the case that the model is not correctly misspecified.