Root--consistent estimation of fixed-effect panel data transformation models with censoring

This paper considers semiparametric n -consistent estimation of the parameters of the generalized panel data transformation model with fixed effects under various forms of censoring, without parametric specification for the transformation function or the error distribution. While the approach in Abrevaya (1999) is n -consistent, it is not applicable when censoring is present. For the case with fixed censoring, existing approaches such as those of Manski (1987) and Abrevaya (2000) apply, but their estimators converge at rates slower than n , thus possessing zero efficiency compared with n -consistent estimators. While the approaches by Honoré (1992) and Ridder and Tunali (1999) do produce n -consistent estimators under fixed and independent censoring respectively, they require either the error distribution or the transformation function to be completely known. Our n -consistent estimator for the fixed censoring case could be extended to the cases with independent and dependent censoring. Under dependent censoring, in contrast to our method, the existing approaches (e.g., Horowitz and Lee (2003), Lee (2008) and Das and Ying (2005)) require parametric specification for the transformation function or the error distribution. Large sample properties of the proposed estimators are presented. We also provide a simulation study to illustrate our estimation methods in finite samples.