Huber Behavior of ML Estimates

# Huber Behavior of - THE BEHAVIOR OF MAXIMUM LIKELIHOOD ESTIMATES UNDER NONSTANDARD CONDITIONS PETER J HUBER SWISS FEDERAL INSTITUTE OF TECHNOLOGY

This preview shows pages 1–3. Sign up to view the full content.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: THE BEHAVIOR OF MAXIMUM LIKELIHOOD ESTIMATES UNDER NONSTANDARD CONDITIONS PETER J. HUBER SWISS FEDERAL INSTITUTE OF TECHNOLOGY 1.Introductionandsummary This paperproves consistencyand asymptoticnormalityof maximum like- lihood (ML) estimatorsunderweakerconditionsthanusual. In particular, (i)itisnot assumedthatthetruedistributionunderlyingthe observations belongs to theparametricfamily defining theML estimator,and (ii)theregularityconditionsdo not involvethesecondandhigherderivatives ofthelikelihoodfunction. The needfortheoremsonasymptoticnormalityofML estimatorssubjectto (i)and (ii)becomes apparentinconnectionwithrobust estimation problems; forinstance,ifone triestoextendtheauthor'sresultson robustestimationof a location parameter [4]to multivariate and other more general estimation problems. Wald's classical consistency proof [6] satisfies (ii) and can easily bemodified to showthattheML estimatorisconsistentalsoincase (i), thatis,itconverges to the 0o characterizedbytheproperty E(logf(x, 0)- log f(x,O0)) < 0for0. Oo, wheretheexpectationistakenwithrespecttothetrueunderlyingdistribution. Asymptotic normality ismore troublesome. Daniels [1]proved asymptotic normalitysubjectto (ii),butunfortunatelyheoverlookedthatacrucialstepin hisproof (theuse ofthe centrallimittheorem in (4.4)) isincorrect without condition (2.2)of Linnik [5]; thisconditionseemstobetoorestrictiveformany purposes. In section 4 we shallprove asymptotic normality, assuming that the ML estimatorisconsistent.Forthesakeofcompleteness,sections2and 3contain, therefore, two differentsetsofsufficientconditionsforconsistency. Otherwise, thesesectionsareindependentofeachother.Section5 presentstwo examples. 2. Consistency:caseA Throughoutthissection,whichrephrasesWald'sresultsonconsistencyofthe ML estimatorin aslightlymoregeneralsetup,theparameterset0 isalocally compactspacewithacountablebase, (X, 2I, P)isaprobabilityspace,andp(x,0) issomereal-valuedfunctiononX X 0. 221 222 FIFTH BERKELEY SYMPOSIUM: HUBER Assume that xI, X2,- **are independent random variables with values in I having thecommon probabilitydistributionP. Let Tn(xI, *- , xn)beany se- quence offunctions Tn:X'ne* 0, measurableor not, suchthat in I n (1)- E p(xi,T.)- inf- E p(xi,0)-°0 almost surely (orin probability-more precisely,outerprobability). We want to givesufficientconditionsensuringthateverysuchsequenceconvergesalmost surely(orinprobability)towardsomeconstant0o. IfdP = f(x,O0)d,and p(x,0) =-logf(x,0)forsome measure uion (X,9) and some family ofprobabilitydensitiesf(x,0),then the ML estimatorof 0o evidentlysatisfiescondition(1). Convergence of Tn shallbe proved under thefollowingsetofassumptions. ASSUMPTIONS. (A-1).Foreachfixed0 E 0, p(x, 0)is91-measurable,andp(x,0)isseparable in thesense ofDoob: thereisa P-null setN and a countable subset 0' C 0 such thatforeveryopensetU C 0 and every closedintervalA, thesets (2) {xlp(x, 0) E A, VO E U}, {xlp(x, 0) E A,VO E U n et} differ by atmosta subsetofN....
View Full Document

## This note was uploaded on 12/26/2011 for the course ECON 245a taught by Professor Staff during the Fall '08 term at UCSB.

### Page1 / 13

Huber Behavior of - THE BEHAVIOR OF MAXIMUM LIKELIHOOD ESTIMATES UNDER NONSTANDARD CONDITIONS PETER J HUBER SWISS FEDERAL INSTITUTE OF TECHNOLOGY

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online