Title: Bootstrap bandwidth selection for matching estimators in nonparametric regression
Authors: Ines Barbeito - University of A Coruna (Spain) [presenting]
Stefan Sperlich - University of Geneva (Switzerland)
Ricardo Cao - University of Coruna (Spain)
Abstract: The smoothed bootstrap method has been used in the context of prediction, in which the response variable of the target population remains unknown. Specifically, this bootstrap procedure is used for the purpose of bandwidth selection in regression estimation. The aim is to establish a new bootstrap bandwidth selector based on the exact expression of the bootstrap version of the mean average squared error of some approximation of the kernel regression estimator. This is very useful since Monte Carlo approximation is avoided for the implementation of the bootstrap selector. Furthermore, the distribution of the target population no longer needs to be estimated. The method is illustrated by applying it to a real data set which accounts for the gross annual income of men and women in Spain.