文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > Regularized estimation of large covariance matrices
侵权投诉

Regularized estimation of large covariance matrices

This paper considers estimating a covariance matrix of p variables from n observations by either banding the sample covariance matrix or estimating a banded version of the inverse of the covariance. We show that these estimates are consistent in the operat

Regularizedestimationoflargecovariancematrices

PeterJ.BickelandElizavetaLevina

UniversityofCalifornia,BerkeleyandUniversityofMichigan

September14,2006

Abstract

Thispaperconsidersestimatingacovariancematrixofpvariablesfromnobservationsbyeitherbandingthesamplecovariancematrixorestimatingabandedversionoftheinverseofthecovariance.Weshowthattheseestimatesareconsistentintheoperatornormaslongas(logp)2/n→0,andobtainexplicitrates.Theresultsareuniformoversomefairlynaturalwell-conditionedfamiliesofcovariancematrices.WealsointroduceananalogueoftheGaussianwhitenoisemodelandshowthatifthepopulationcovarianceisembeddableinthatmodelandwell-conditionedthenthebandedapproximationsproduceconsistentestimatesoftheeigen-valuesandassociatedeigenvectorsofthecovariancematrix.Theresultscanbeextendedtosmoothversionsofbandingandtonon-Gaussiandistributionswithsu cientlyshorttails.Aresamplingapproachisproposedforchoosingthebandingparameterinpractice.Thisapproachisillustratednumericallyonbothsimulatedandrealdata.

Keywords:covarianceestimation,regularization,bandedestimators,Choleskydecomposition.

1Introduction

Estimationofpopulationcovariancematricesfromsamplesofmultivariatedatahasalwaysbeenimportantforanumberofreasons.Principalamongtheseare:(1)estimationofprincipalcompo-nentsandeigenvaluesinordertogetaninterpretablelow-dimensionaldatarepresentation(principalcomponentanalysis,orPCA);(2)constructionoflineardiscriminantfunctionsforclassi cationofGaussiandata(lineardiscriminantanalysis,orLDA);(3)establishingindependenceandcondi-tionalindependencerelationsbetweencomponentsusingexploratorydataanalysisandtesting;and(4)settingcon denceintervalsonlinearfunctionsofthemeansofthecomponents.Notethat(1)requiresestimationoftheeigenstructureofthecovariancematrixwhile(2)and(3)requireestimationoftheinverse.

Thetheoryofmultivariateanalysisfornormalvariableshasbeenwellworkedout–seeAnderson(1958),themajormonograph.However,itbecameapparentthatexactexpressionswerecumber-some,evenforsmalldimensionsandsamplesizes,andthatmultivariatedatawererarelyGaussian.Theremedywasasymptotictheoryforlargesamplesand xedrelativelysmalldimensions.Inrecentyears,datasetsthatdonot tintothisframeworkhavebecomeverycommon–thedataareveryhigh-dimensionalandsamplesizescanbeverysmallrelativetodimension.Examplesincludegeneexpressionarrays,fMRIdata,spectroscopicimaging,numericalweatherforecasting,andmanyothers.

第1页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email