Stable asymptotics for M-estimators

Davide La Vecchia

Research output: Contribution to journalArticleResearchpeer-review

Abstract

We review some first-order and higher-order asymptotic techniques for M-estimators, and we study their stability in the presence of data contaminations. We show that the estimating function (ψ) and its derivative with respect to the parameter (∇θTψ) play a central role. We discuss in detail the first-order Gaussian density approximation, saddlepoint density approximation, saddlepoint test, tail area approximation via the Lugannani–Rice formula and empirical saddlepoint density approximation (a technique related to the empirical likelihood method). For all these asymptotics, we show that a bounded ψ (in the Euclidean norm) and a bounded (∇θTψ) (e.g. in the Frobenius norm) yield stable inference in the presence of data contamination. We motivate and illustrate our findings by theoretical and numerical examples about the benchmark case of one-dimensional location model.
Original languageEnglish
Pages (from-to)267-290
Number of pages24
JournalInternational Statistical Review
Volume84
Issue number2
DOIs
Publication statusPublished - 1 Aug 2016

Keywords

  • Edgeworth expansion
  • empirical likelihood
  • higher-order
  • infinitesimal robustness
  • p-value
  • redescending M-estimator
  • relative error
  • saddlepoint techniques
  • von Mises expansion

Cite this

La Vecchia, Davide. / Stable asymptotics for M-estimators. In: International Statistical Review. 2016 ; Vol. 84, No. 2. pp. 267-290.
@article{08a0f2e8215c439995c02b77cebcf826,
title = "Stable asymptotics for M-estimators",
abstract = "We review some first-order and higher-order asymptotic techniques for M-estimators, and we study their stability in the presence of data contaminations. We show that the estimating function (ψ) and its derivative with respect to the parameter (∇θTψ) play a central role. We discuss in detail the first-order Gaussian density approximation, saddlepoint density approximation, saddlepoint test, tail area approximation via the Lugannani–Rice formula and empirical saddlepoint density approximation (a technique related to the empirical likelihood method). For all these asymptotics, we show that a bounded ψ (in the Euclidean norm) and a bounded (∇θTψ) (e.g. in the Frobenius norm) yield stable inference in the presence of data contamination. We motivate and illustrate our findings by theoretical and numerical examples about the benchmark case of one-dimensional location model.",
keywords = "Edgeworth expansion, empirical likelihood, higher-order, infinitesimal robustness, p-value, redescending M-estimator, relative error, saddlepoint techniques, von Mises expansion",
author = "{La Vecchia}, Davide",
year = "2016",
month = "8",
day = "1",
doi = "10.1111/insr.12102",
language = "English",
volume = "84",
pages = "267--290",
journal = "International Statistical Review",
issn = "0306-7734",
publisher = "Wiley-Blackwell",
number = "2",

}

Stable asymptotics for M-estimators. / La Vecchia, Davide.

In: International Statistical Review, Vol. 84, No. 2, 01.08.2016, p. 267-290.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Stable asymptotics for M-estimators

AU - La Vecchia, Davide

PY - 2016/8/1

Y1 - 2016/8/1

N2 - We review some first-order and higher-order asymptotic techniques for M-estimators, and we study their stability in the presence of data contaminations. We show that the estimating function (ψ) and its derivative with respect to the parameter (∇θTψ) play a central role. We discuss in detail the first-order Gaussian density approximation, saddlepoint density approximation, saddlepoint test, tail area approximation via the Lugannani–Rice formula and empirical saddlepoint density approximation (a technique related to the empirical likelihood method). For all these asymptotics, we show that a bounded ψ (in the Euclidean norm) and a bounded (∇θTψ) (e.g. in the Frobenius norm) yield stable inference in the presence of data contamination. We motivate and illustrate our findings by theoretical and numerical examples about the benchmark case of one-dimensional location model.

AB - We review some first-order and higher-order asymptotic techniques for M-estimators, and we study their stability in the presence of data contaminations. We show that the estimating function (ψ) and its derivative with respect to the parameter (∇θTψ) play a central role. We discuss in detail the first-order Gaussian density approximation, saddlepoint density approximation, saddlepoint test, tail area approximation via the Lugannani–Rice formula and empirical saddlepoint density approximation (a technique related to the empirical likelihood method). For all these asymptotics, we show that a bounded ψ (in the Euclidean norm) and a bounded (∇θTψ) (e.g. in the Frobenius norm) yield stable inference in the presence of data contamination. We motivate and illustrate our findings by theoretical and numerical examples about the benchmark case of one-dimensional location model.

KW - Edgeworth expansion

KW - empirical likelihood

KW - higher-order

KW - infinitesimal robustness

KW - p-value

KW - redescending M-estimator

KW - relative error

KW - saddlepoint techniques

KW - von Mises expansion

UR - http://www.scopus.com/inward/record.url?scp=84927634396&partnerID=8YFLogxK

U2 - 10.1111/insr.12102

DO - 10.1111/insr.12102

M3 - Article

VL - 84

SP - 267

EP - 290

JO - International Statistical Review

JF - International Statistical Review

SN - 0306-7734

IS - 2

ER -