Title: Creating a safe space for bias correction in kernel hazard estimation
Authors: Vali Asimit - City University London (United Kingdom)
Maria Luz Gamiz - University of Granada (Spain)
Maria-Dolores Martinez-Miranda - Universidad de Granada (Spain) [presenting]
Jens Perch Nielsen - Cass Business School (United Kingdom)
Abstract: Bias-correction methods in kernel estimation are often knocked down by an explosion in variance. Considering the simplest situation of kernel hazard estimation we propose a new algorithm to reduce the bias with no cost in terms of variance. We start the algorithm in a not so complex model (rather large bias but small variance) and move with a safety net towards a more complex model (reducing the bias). We move in a safe way smoothing the slope of the ride. The end result is that we get the variance from our starting place and the bias from our ending place. The algorithm is therefore a safe ride from a non-complex world towards complexity where we get the best of both worlds. Simulation experiments and asymptotic theory support our proposal.