On the estimation error in binned local linear regression

Authors: Fabian Hoti; Lasse Holmström

Source: Journal of Nonparametric Statistics, Volume 15, Numbers 4-5, Numbers 4-5/August-October 2003 , pp. 625-642(18)

Publisher: Taylor and Francis Ltd

Buy & download fulltext article:

Price: $56.94 plus tax (Refund Policy)

Abstract:

Kernel estimation is probably the most popular method in nonparametric estimation of functions. However, direct implementations of the basic forms of the estimators are sometimes too slow to use in applications that require repeated rapid evaluations of the estimates. Data prebinning is a common way to reduce the computational burden in kernel estimation. We consider a binned local linear regression estimator and present rigorous results on its consistency and the asymptotic mean squared error. A simulation study is used to suggest a reasonable choice for the relative size of the smoothing parameter and the bin width.

Keywords: Local linear regression; Binning; Estimation error; Simulations

Document Type: Research article

DOI: http://dx.doi.org/10.1080/10485250310001605469

Publication date: 2003-08-01