Beruflich Dokumente
Kultur Dokumente
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
PublishedonSTAT414/415(https://onlinecourses.science.psu.edu/stat414)
Home>MethodofMoments
MethodofMoments
Inshort,themethodofmomentsinvolvesequatingsamplemomentswiththeoretical
moments.So,let'sstartbymakingsurewerecallthedefinitionsoftheoreticalmoments,
aswellaslearnthedefinitionsofsamplemoments.
Definitions.
(1)E(X
1,2,...
isthekth(theoretical)momentofthedistribution(abouttheorigin),fork=
(3)M
(4)M
1
k
=
n
1
=
n
k
i
isthekthsamplemoment,fork=1,2,...
i=1
k
(X i X )
isthekthsamplemomentaboutthemean,fork=1,2,...
i=1
OneFormoftheMethod
Thebasicideabehindthisformofthemethodisto:
(1)EquatethefirstsamplemomentabouttheoriginM1
1
=
n
firsttheoreticalmomentE(X).
(2)EquatethesecondsamplemomentabouttheoriginM2
secondtheoreticalmomentE(X2).
Xi = X
tothe
i=1
=
n
2
i
tothe
i=1
(3)Continueequatingsamplemomentsabouttheorigin,Mk,withthe
correspondingtheoreticalmomentsE(Xk),k=3,4,...untilyouhaveasmany
equationsasyouhaveparameters.
(4)Solvefortheparameters.
Theresultingvaluesarecalledmethodofmomentsestimators.Itseemsreasonable
thatthismethodwouldprovidegoodestimates,sincetheempiricaldistributionconverges
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
1/6
4/27/2016
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
insomesensetotheprobabilitydistribution.Therefore,thecorrespondingmoments
shouldbeaboutequal.
(ATrivial)Example
LetX1,X2,...,XnbeBernoullirandomvariableswithparameterp.Whatisthemethodof
momentsestimatorofp?
Solution.Here,thefirsttheoreticalmomentabouttheoriginis:
E(Xi)=p
Wehavejustoneparameterforwhichwearetryingtoderivethemethodof
momentsestimator.Therefore,weneedjustoneequation.Equatingthefirst
theoreticalmomentabouttheoriginwiththecorrespondingsamplemoment,we
get:
p =
1
n
Xi
i=1
Now,wejusthavetosolveforp.Whoops!Inthiscase,theequationisalready
solvedforp.Ourworkisdone!Wejustneedtoputahat(^)ontheparameter
tomakeitclearthatitisanestimator.Wecanalsosubscripttheestimatorwith
an"MM"toindicatethattheestimatoristhemethodofmomentsestimator:
^
=
p
MM
1
n
Xi
i=1
So,inthiscase,themethodofmomentsestimatoristhesameasthemaximum
likelihoodestimator,namely,thesampleproportion.
Example
LetX1,X2,...,Xnbenormalrandomvariableswithmeanandvariance2.Whatarethe
methodofmomentsestimatorsofthemeanandvariance2?
Solution.Thefirstandsecondtheoreticalmomentsabouttheoriginare:
E(Xi)=andE(Xi2)=2+2
(Incidentally,incaseit'snotobvious,thatsecondmomentcanbederivedfrom
manipulatingtheshortcutformulaforthevariance.)Inthiscase,wehavetwo
parametersforwhichwearetryingtoderivemethodofmomentsestimators.
Therefore,weneedtwoequationshere.Equatingthefirsttheoreticalmoment
abouttheoriginwiththecorrespondingsamplemoment,weget:
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
2/6
4/27/2016
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
n
1
E(X) = =
n
Xi
i=1
And,equatingthesecondtheoreticalmomentabouttheoriginwiththe
correspondingsamplemoment,weget:
E(X
) =
=
n
2
i
i=1
Now,thefirstequationtellsusthatthemethodofmomentsestimatorforthe
meanisthesamplemean:
^
1
MM
=
n
Xi = X
i=1
And,substitutingthesamplemeaninforinthesecondequationandsolving
for2,wegetthatthemethodofmomentsestimatorforthevariance2is:
2
MM
=
n
2
i
1
n
i=1
2
i
i=1
whichcanberewrittenas:
2
MM
1
=
n
2
(X i X )
i=1
Again,forthisexample,themethodofmomentsestimatorsarethesameasthe
maximumlikelihoodestimators.
Insomecases,ratherthanusingthesamplemomentsabouttheorigin,itiseasiertouse
thesamplemomentsaboutthemean.Doingso,providesuswithanalternativeformof
themethodofmoments.
AnotherFormoftheMethod
Thebasicideabehindthisformofthemethodisto:
(1)EquatethefirstsamplemomentabouttheoriginM1
firsttheoreticalmomentE(X).
(2)EquatethesecondsamplemomentaboutthemeanM2
Xi = X
tothe
i=1
1
n
2
(X i X )
i=1
tothesecondtheoreticalmomentaboutthemeanE[(X ) ].
2
(3)ContinueequatingsamplemomentsaboutthemeanMk withthe
correspondingtheoreticalmomentsaboutthemeanE[(X )k ],k=3,4,...
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
3/6
4/27/2016
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
untilyouhaveasmanyequationsasyouhaveparameters.
(4)Solvefortheparameters.
Again,theresultingvaluesarecalledmethodofmomentsestimators.
Example
LetX1,X2,...,Xnbegammarandomvariableswithparametersand,sothatthe
probabilitydensityfunctionis:
1
f ( xi ) =
x/
()
forx>0.Therefore,thelikelihoodfunction:
L(, ) = (
()
) ( x1 x2 xn )
exp [
xi ]
isdifficulttodifferentiatebecauseofthegammafunction().So,ratherthanfindingthe
maximumlikelihoodestimators,whatarethemethodofmomentsestimatorsofand?
Solution.Thefirsttheoreticalmomentabouttheoriginis:
E(Xi)=
Andthesecondtheoreticalmomentaboutthemeanis:
Var(Xi)=E(Xi)2=2
Again,sincewehavetwoparametersforwhichwearetryingtoderivemethodof
momentsestimators,weneedtwoequations.Equatingthefirsttheoreticalmoment
abouttheoriginwiththecorrespondingsamplemoment,weget:
1
E(X) = =
n
Xi = X
i=1
And,equatingthesecondtheoreticalmomentaboutthemeanwiththecorresponding
samplemoment,weget:
V ar(X) =
1
=
n
2
(X i X )
i=1
Now,wejusthavetosolveforthetwoparametersand.Let'sstartbysolvingfor
inthefirstequation(E(X)).Doingso,weget:
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
4/6
4/27/2016
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
X
=
Now,substituting
X
=
intothesecondequation(Var(X)),weget:
X
= (
= X =
1
n
2
(X i X )
i=1
Now,solvingforinthatlastequation,andputtingonitshat,wegetthatthemethodof
momentestimatorforis:
^
MM =
2
(X i X )
nX
i=1
And,substitutingthatvalueofbackintotheequationwehavefor,andputtingonits
hat,wegetthatthemethodofmomentestimatorforis:
^MM =
X
=
^
MM
2
(1/nX ) (X i X )
i=1
nX
=
2
(X i X )
i=1
(AnotherTrivial)Example
Let'sreturntotheexampleinwhichX1,X2,...,Xnarenormalrandomvariableswith
meanandvariance2.Whatarethemethodofmomentsestimatorsofthemeanand
variance2?
Solution.Thefirsttheoreticalmomentabouttheoriginis:
E(Xi)=
And,thesecondtheoreticalmomentaboutthemeanis:
Var(Xi)=E(Xi)2=2
Again,sincewehavetwoparametersforwhichwearetryingtoderivemethodof
momentsestimators,weneedtwoequations.Equatingthefirsttheoreticalmoment
abouttheoriginwiththecorrespondingsamplemoment,weget:
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
5/6
4/27/2016
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
n
Xi
E(X) = =
n
i=1
And,equatingthesecondtheoreticalmomentaboutthemeanwiththecorresponding
samplemoment,weget:
1
=
n
2
(X i X )
i=1
Now,wejusthavetosolveforthetwoparameters.Oh!Well,inthiscase,the
equationsarealreadysolvedforand2.Ourworkisdone!Wejustneedtoputahat
(^)ontheparameterstomakeitclearthattheyareestimators.Doingso,wegetthatthe
methodofmomentsestimatorofis:
= X
MM
(whichweknow,fromourpreviouswork,isunbiased).Themethodofmoments
estimatorof2is:
2
=
MM
1
n
2
(X i X )
i=1
(whichweknow,fromourpreviouswork,isbiased).Thisexample,inconjunctionwith
thesecondexample,illustrateshowthetwodifferentformsofthemethodcanrequire
varyingamountsofworkdependingonthesituation.
SourceURL:https://onlinecourses.science.psu.edu/stat414/node/193
https://onlinecourses.science.psu.edu/stat414/print/book/export/html/193
6/6