Optimal variance estimation without estimating the mean function

Tiejun Tong, Yanyuan Ma, Yuedong Wang

Research output: Contribution to journalArticle

13 Citations (Scopus)

Abstract

We study the least squares estimator in the residual variance estimation context. We show that the mean squared differences of paired observations are asymptotically normally distributed. We further establish that, by regressing the mean squared differences of these paired observations on the squared distances between paired covariates via a simple least squares procedure, the resulting variance estimator is not only asymptotically normal and root-n consistent, but also reaches the optimal bound in terms of estimation variance. We also demonstrate the advantage of the least squares estimator in comparison with existing methods in terms of the second order asymptotic properties.

Original languageEnglish (US)
Pages (from-to)1839-1854
Number of pages16
JournalBernoulli
Volume19
Issue number5 A
DOIs
StatePublished - Dec 2 2013

Fingerprint

Optimal Estimation
Variance Estimation
Least Squares Estimator
Second-order Asymptotics
Optimal Bound
Variance Estimator
Asymptotic Properties
Least Squares
Covariates
Roots
Demonstrate
Observation
Context

All Science Journal Classification (ASJC) codes

  • Statistics and Probability

Cite this

Tong, Tiejun ; Ma, Yanyuan ; Wang, Yuedong. / Optimal variance estimation without estimating the mean function. In: Bernoulli. 2013 ; Vol. 19, No. 5 A. pp. 1839-1854.
@article{1716f00902cb47a98d2ccd02176ac1a4,
title = "Optimal variance estimation without estimating the mean function",
abstract = "We study the least squares estimator in the residual variance estimation context. We show that the mean squared differences of paired observations are asymptotically normally distributed. We further establish that, by regressing the mean squared differences of these paired observations on the squared distances between paired covariates via a simple least squares procedure, the resulting variance estimator is not only asymptotically normal and root-n consistent, but also reaches the optimal bound in terms of estimation variance. We also demonstrate the advantage of the least squares estimator in comparison with existing methods in terms of the second order asymptotic properties.",
author = "Tiejun Tong and Yanyuan Ma and Yuedong Wang",
year = "2013",
month = "12",
day = "2",
doi = "10.3150/12-BEJ432",
language = "English (US)",
volume = "19",
pages = "1839--1854",
journal = "Bernoulli",
issn = "1350-7265",
publisher = "International Statistical Institute",
number = "5 A",

}

Optimal variance estimation without estimating the mean function. / Tong, Tiejun; Ma, Yanyuan; Wang, Yuedong.

In: Bernoulli, Vol. 19, No. 5 A, 02.12.2013, p. 1839-1854.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Optimal variance estimation without estimating the mean function

AU - Tong, Tiejun

AU - Ma, Yanyuan

AU - Wang, Yuedong

PY - 2013/12/2

Y1 - 2013/12/2

N2 - We study the least squares estimator in the residual variance estimation context. We show that the mean squared differences of paired observations are asymptotically normally distributed. We further establish that, by regressing the mean squared differences of these paired observations on the squared distances between paired covariates via a simple least squares procedure, the resulting variance estimator is not only asymptotically normal and root-n consistent, but also reaches the optimal bound in terms of estimation variance. We also demonstrate the advantage of the least squares estimator in comparison with existing methods in terms of the second order asymptotic properties.

AB - We study the least squares estimator in the residual variance estimation context. We show that the mean squared differences of paired observations are asymptotically normally distributed. We further establish that, by regressing the mean squared differences of these paired observations on the squared distances between paired covariates via a simple least squares procedure, the resulting variance estimator is not only asymptotically normal and root-n consistent, but also reaches the optimal bound in terms of estimation variance. We also demonstrate the advantage of the least squares estimator in comparison with existing methods in terms of the second order asymptotic properties.

UR - http://www.scopus.com/inward/record.url?scp=84888316047&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84888316047&partnerID=8YFLogxK

U2 - 10.3150/12-BEJ432

DO - 10.3150/12-BEJ432

M3 - Article

AN - SCOPUS:84888316047

VL - 19

SP - 1839

EP - 1854

JO - Bernoulli

JF - Bernoulli

SN - 1350-7265

IS - 5 A

ER -