Imposing monotonicity in stochastic frontier models: an iterative nonlinear least squares procedure


Despite its importance, the monotonicity condition is typically overlooked in stochastic frontier analysis. This article illustrates a straightforward and useful method for the estimation of semiparametric stochastic frontier models imposing such constraint and incorporating exogenous inefficiency effects exploiting the scaling property. An iterative estimation algorithm based on nonlinear least squares is developed and the behavior of the proposed procedure is investigated through a set of Monte Carlo experiments comparing its finite sample properties with those of available alternatives. The simulation results highlight very good performance of the new algorithm which outperforms the competitors in small samples and in presence of small signal-to-noise ratios. Our results also show that the fraction of observations for which monotonicity naturally holds is generally quite small if this condition is not imposed. An application based on FADN agricultural data illustrates the usefulness of the proposed algorithm.

Work in progress