Simulated maximum likelihood estimation of spatial stochastic frontier model and its application

This paper considers a spatial stochastic frontier model that accounts for possible unknown geographical variation of the outputs. The stochastic frontier model is augmented with a spatial autoregressive structure for the two-sided part of the disturbance, and the time-varying technical inefficiency is not imposed a rigorous function structure. Because of the spatial effect and the asymmetry composed error structure, it is intractable to employ maximum likelihood method directly to estimate the proposed model. Simulated maximum likelihood estimation is used instead. We derive the simulated likelihood function of the model, and present an application of the estimation method on China province-level panel data from 2000 to 2007. The results show that the spatial effect is highly significant, and the ignorance of the spatial effect produces significantly different rankings of technical efficiencies across production units.