Bias-corrected quantile regression estimation of censored regression models (Q1706470)

This paper considers a linear regression model where the response variable may be censored by a fixed cut-off point. Several quantile regression procedures have been proposed to estimate the regression coefficient in the model by generalization of classical quantile regression procedures but the finite-sample performances of these estimators are not very satisfactory. A new procedure, which relies on the linear quantile regression for the whole sample and then applies the idea of indirect inference to correct the bias, is proposed in this paper. The consistency and asymptotic normality of the proposed estimators are established and simulations show that they are preferable to existing methods in many cases when the sample sizes are small and moderate.