A Corrected Plug-in Method for Quantile Interval Construction through a Transformed Regression

We propose a corrected plug-in method for constructing confidence intervals of the conditional quantiles of an original response variable through a transformed regression with heteroscedastic errors. The interval is easy to compute. Factors affecting the magnitude of the correction are examined analytically through the special case of Box-Cox regression. Monte Carlo simulations show that the new method works well in general and is superior over the commonly used delta method and the quantile regression method. An empirical application is presented. [PUBLICATION ABSTRACT]