A Root n Bandwidth Selector in Hazard Estimation.

The data consists of multivariate failure times under random censorship. By the kernel smoothing techniques, convolutions of integrated multivariated hazard functions provide some estimators of the so-called multivariate hazard functions (Fermanian (1995)). We adopt the method of Jones, Marron and Park (1991) to study a bandwidth selector: without using high order kernels, the estimator tends to the asymptotically optimal bandwidth at the very fast relative rate of square root n.