Boundary and Bias Correction in Kernel Hazard Estimation

A new class of local linear azard estimators based on weig ted least square kernel estimation is considered. The class includes the kernel Hazard estimator of Ramlau-Hansen (1983), which as tHe same boundary correction property as the local linear regression estimator (see Fan and Gijbels, 1996). It is shown that all the local linear estimators in the class have the same pointwise asymp-totic properties. We derive the multiplicative bias correction of the local linear estimator. In addition we propose a new bias correction technique based on bootstrap estimation of additive bias. This latter method as nice theoretical properties. Based on an extensive simulation study where we compare the per-formance of competing estimators we also recommend the use of the additive bias correction in applied work.