Weighted and Two Stage Least Squares Estimation of Semiparametric Censored and Truncated Regressions

This paper provides a root-n consistent, asymptotically normal weighted least squares estimator of the coeecients in the censored or truncated regression model with a xed censoring point. The distribution of the errors is unknown and permits general forms of unknown heteroscedasticity. Also provided is an instrumental variables based two stage least squares estimator for these models, which can be used when the errors are correlated with some regressors. Estimation is based on a "special" regressor as in Lewbel (1999). A new, related binary choice estimator is also provided. for helpful comments. Any errors are our own.