Abstract: Zero inefficiency stochastic frontier model (ZISF) contains regression model and stochastic frontier model, which are with certain probability respectively. Thus ZISF can accommodate the presence of both efficient and inefficient firms. Now the theoretical researches about ZISF are rare. Especially for spatial ZISF, the existing ZISFs are with poor applicability. By incorporating spatial effects and nonparametric functions into ZISF, semiparametric spatial ZISF is constructed in this paper. The semiparametric spatial ZISF can avoid under-fitting derived from linear model and the biased and inconsistent estimators derived from neglecting spatial effects. B-splines are used to approximate nonparametric function and the model has been changed into linear spatial ZISF. The two order norm of approximate error can converge to zero quickly, so the approximate error can be neglected. Maximum likelihood method and JLMS method are used to estimate parameters and technical efficiencies respectively. The Monte Carlo simulation shows that:(i) The method in this paper is with high estimation accuracies for parameters, nonparametric functions and technical efficiencies and with high classification for technical efficiencies. With sample size increasing, the accuracies become higher. (ii) Neglecting any one of effect such as spatial effect or nonparametric effect will get lower estimation accuracies and classification accuracies. So the model in paper is necessary. (iii) The nonparametric effect in production function or in probability of occurrence has different impact on the estimation and classification accuracies. When the nonparametric effect in production function is neglected, there is only a small reduction for the estimation and classification accuracies. While the nonparametric effect in probability of occurrence is neglected, the estimation and classification accuracies have been substantially decreased.

Key words: stochastic frontier model, zero technical inefficiencies, spatial effect, nonparametric function, Monte Carlo simulation