Semiparametric Efficient Estimation of Partially Linear Quantile Regression Models

Lee (2003) develops a √ n-consistent estimator of the parametric component of a partially linear quantile regression model, which is used to obtain his one-step semiparametric efficient estimator. As a result, how well the efficient estimator performs depends on the quality of the initial √ n-consistent estimator. In this paper, we aim to improve the small sample performance of the one-step efficient estimator by proposing a new √ n-consistent initial estimator, which does not require any trimming procedure and is less sensitive to data outliers and the choice of bandwidth than Lee’s (2003) initial consistent estimator. Monte Carlo simulation results confirm that the proposed estimator and the one-step efficient estimator derived from it have more desirable empirical features than Lee’s estimators. c © 2005 Peking University Press