Reducing bias in nonparametric density estimation via bandwidth dependent kernels view.pdfVIP

Statistics and Probability Letters 123 (2017) 17–22 Contents lists available at ScienceDirect Statistics and Probability Letters journal homepage: /locate/stapro Reducing bias in nonparametric density estimation via bandwidth dependent kernels: L1 view? Kairat Mynbaev a, Carlos Martins-Filho b,c,? a International School of Economics, Kazakh-British Technical University, Tolebi 59, Almaty 050000, Kazakhstan b Department of Economics, University of Colorado, Boulder, CO 80309-0256, USA c IFPRI, 2033 K Street NW, Washington, DC 20006-1002, USA article info Article history: Received 24 February 2016 Received in revised form 19 November 2016 Accepted 20 November 2016 Available online 5 December 2016 MSC: 62G07 62G10 62G20 Keywords: Kernel density estimation Higher order kernels Bias reduction abstract We define a new bandwidth-dependent kernel density estimator that improves existing convergence rates for the bias, and preserves that of the variation, when the error is measured in L1. No additional assumptions are imposed to the extant literature. ? 2016 Elsevier B.V. All rights reserved. 1. Introduction Given a sequence of n ∈ N independent realizations {Xj}jn=1 of the random variable X , having density f on R, the Rosenblatt–Parzen kernel estimator (Rosenblatt, 1956; Parzen, 1962) of f is given by 1 n ? fn(x) = n (Shn K )(x ? Xj), j=1 (1.1) where Shn is an operator defined by 1 ?x? (Shn K )(x) = K hn hn , (1.2) K is a kernel, i.e., a function on R such that ? K (x)dx = 1 and hn 0 is a non-stochastic bandwidth such that hn → 0 as n → ∞.1 ? We thank an anonymous referee and an Associate Editor for excellent comments that improved this note significantly. ? Corresponding author at: Department of Economics, University of Colorado, Boulder, CO 80309-0256, USA. E-mail addresses: kairat_mynbayev@ (K. Mynbaev), carlos.martins@, c.martins-filho@ (C. Martins-Filho). 1 Throughout this note, integrals are over R, unless otherwise specified. /10.1016/j.spl.2016.11.019 0167-7152/?