A spreading method to improve efficiency prediction.

In efficiency analysis by means of a stochastic frontier production function, the composite error variable includes the inefficiency component. For this reason, individual prediction cannot be made directly from an estimation of the error in the model. In order to solve this problem, Jondrow et al (1982), and Battese and Coelli (1988) separately developed two different procedures, based on the expectation operator of the conditional distributions. Although the two predictors are different, each suffers from a shrinkage effect with respect to the distribution of theoretical efficiency. Our study of the behaviour of these two predictors leads us to conclude that the value of the gamma parameter has a great influence on the above-mentioned effect, producing a truncation of the distribution that could be more than 50%, so that the extreme values of the efficiency can never be estimated by the predictors considered. We also propose a method that spreads out the predicted efficiencies in order to minimise the shrinkage effect. The Monte Carlo results demonstrate that the corrected predictions have a better behaviour than the original predictors.