When the dependent variables are limited in their range in nature, the ordinary least squares(OLS) method gives inconsistent estimates due to the problem of the incomplete data. Following the work of Tobin(1958) and Amemiya(1973), statistical techniques were developed to consistently estimate the parameters of these models. One potential drawback to the application of these techniques is the sensitivity to the assumed parametric distributions. As an alternative to the parametric estimation methods, semiparametric methods have been developed.
In this study, we apply the semiparametric two-stage estimation method suggested by Lee(1994) to a censored simultaneous equations model. With the Monte Carlo simulation, we compare the semiparametric estimation results with the parametric ones. The results indicate that although our semiparametric method produces relatively inefficient estimates compared to the parametric ones when the error terms are normal and the sample size is large, its asymptotic bias sharply reduces as the sample size increases when the error terms are not normal. And, under the normality assumption, the small samples results show that the parametric method does not always outperform our semiparametric method.