On extreme regression quantiles (Q5926462)

\textit{R. Koenker} and \textit{G. Basset} [Econometrica 46, 33-50, (1978; Zbl 0373.62038)] introduced regression quantiles to generalize the notion of order statistic from the case of a single sample to the linear regression setting. In linear models regression quantiles estimate the conditional quantile of the response at each value of the independent variable. In earlier research most of the one-sample properties of quantiles have found natural analogues in the regression quantile situation. One property only recently considered in depth is that of the extreme regression quantiles: analogues of the smallest and largest order statistics. The focus in this paper is on using these extreme regression quantiles to generalize extreme value theory beyond the usual one-sample case. Two sets of results are presented. The first generalizes results of \textit{R.L. Smith} [Biometrika 81, No.1, 173-183 (1994; Zbl 0803.62056)] on the asymptotic distribution theory for extreme regression quantiles. Smith considered algebraically (''heavy'' tailed) error distributions. In this paper, some further analogues and also some results exhibiting behavior different from the one-sample case are provided. A class of exponentially tailed error distributions is considered. The second set of results addresses questions of robustness and sensitivity properties of extreme regression quantiles. In particular, the measure of ''tail behavior'' of estimators is extended to extreme quantiles and to the context of extreme regression quantiles.