Coffee Stochastic Frontier Model with Maximum Entropy

In the classical maximum likelihood estimation of stochastic frontier model, a strong assumption on two error components, namely symmetric noise ($V_{j}$) and the non-negative inefficiency ($U_{j}$), are required. This could lead to non-reliable and erroneous interpretations when we misspecify the probability distribution of the error components. To overcome this problem, we apply the generalized maximum entropy (GME) approach to estimate the stochastic frontier model which allows us to avoid the need for making an ad hoc assumption about the distribution of the noise and inefficiency components. In this study, we investigate the technical efficiency of coffee production using generalized maximum entropy. The results show that the technical efficiency scores obtained from GME estimator are much smaller than ones from the maximum likelihood method, even though the estimated parameters are quite indifferent. In addition, we also find that the wider support value of the inefficiency component, the lower score of the estimated technical efficiency.