Microfoundations for stochastic frontiers

The purpose of the paper is to propose microfoundations for stochastic frontier models. Previous work shows that a simple Bayesian learning model supports gamma distributions for technical inefficiency in stochastic frontier models. The conclusion depends on how the problem is formulated and what assumptions are made about the sampling process and the prior. After the new formulation of the problem it turns out that the distribution of the one-sided error component does not belong to a known family. Moreover, we find that without specifying a utility function or even the cost inefficiency function, the relative effectiveness of managerial input can be determined using only cost data and estimates of the returns to scale. The point of this construction is that features of the inefficiency function u(z) can be recovered from the data, based on the solid microfoundation of expected utility of profit maximization but the model does not make a prediction about the distribution.