GMM Estimation for Tobit Models with Endogenous Regressors
Haizheng Li,Jeffrey Zax,Jushan Bai,Shinichi Sakata,Whitney Newey,Donald Waldman,Robert McNown,Nicholas Flores,B. J. Lee,Songnian Chen,Buhong Zheng,Thomas Dee,Lynn Yang
2003-01-01
Abstract:In this study, a new semi-parametric estimation method for censored models, symmetrically censored generalized method of moments (SCGMM) estimator, is proposed. The SCGMM estimator can be applied to censored models when some regressors are correlated with the regression error. Moreover, it does not impose the independence assumption for data, and thus can be applied to time-series and panel data models with dependent and heterogeneous observations. Furthermore, the SCGMM does not require any non-parametric techniques for estimation, and is relatively simple to implement. The SCGMM estimator is also robust to a fairly wide variety of non-normal and non-identical but symmetric disturbance distributions. This estimator adopts a new approach to identifying the true parameter by restricting the parameter space in the estimation. The SCGMM estimator is shown to be consistent and asymptotically normal for near epoch dependent functions of mixing processes. Thus, it can be applied to models with mixing processes, and models in which the disturbance term is an ARMA process of finite order (with roots outside the unit circle), or an infinite non-stationary MA process. A heteroskedasticity and autocorrelation consistent estimator of the covariance matrices is also provided. Simulation results show that the SCGMM estimator performs very well.