Asymptotically optimal estimators of general regression functionals (Q689366)

The estimation of general regression functionals is considered. These functionals are a generalization of the mean value functional and include in addition to the regression function the conditional distribution function, the conditional median and conditional quantiles as particular examples. For the mean value and quantile functional it is well known that the optimal rate of convergence, locally and globally, is achieved by kernel type estimators. The author states conditions under which kernel estimators of these general regression functionals achieve also this optimal asymptotic accuracy. The derivation of the optimality properties is based on a Poissonization argument, and the Poisson process approach leads to the proof of the weak convergence of the maximum error of the estimate over a compact interval.