有关失业率的时间序列分析和回归分析.docx
《有关失业率的时间序列分析和回归分析.docx》由会员分享,可在线阅读,更多相关《有关失业率的时间序列分析和回归分析.docx(25页珍藏版)》请在冰点文库上搜索。
有关失业率的时间序列分析和回归分析
TheUnemploymentRate
Summary
Unemploymentratereflectstheemploymentsituationofacountryoradistrict.Itispossiblethatsomecountrieshavesimilartimeseriesinunemploymentrate.Wewanttodividethesecountriesintoservalclasses,andthenwecanpredicttheemploymentrateofspecialclassandstudythefactorsinfluencingunemploymentrate.
Wecomputemeanandvarianeeofunemploymentrateindifferentregions,andsortthemindependently.Fromtheaboveresults,weknowthatSpain,PolandandBulgariahavebiggervaluesinabovetwostatisticsthantheothers,whichrevealsthesegovernmentsperformedbadlyinemployment.
WeusetheSystemClustermethodtodivide35countriesintofourlevelsaccordingtopseudoFstatistic.Thesameclassdataillustratethatthesecountriesarecommoninunemploymentrate,especially,Spainisaclassalone.
TakingunemploymentdataofChinaasexample,wesolvethisproblemwithtimeseriesaftermakingoriginaldatastationary.WeselecttheappropriatemodelthroughAICandSBCstatistics,andthenwegetthetrendequation.Whentestingthequalityoffitting,weobtaintheMPEAstatisticwhichis3.51%,thuswethinktheequationperformswell.SowepredictURin2011and2012,comparingwiththemeasuringvalues,itissurprisedthatpredictingvaluesissameasmeasuringvalues.Atend,werepeattheaboveprocesswiththedataofJapanandAustralia.
Forstrugglingwithmulticollinearityandnonlinearityalwaysexistingineconomicdata,weuseRFR(randomforestsregression)method.BycomparingR-Square,MSEandMPEA,weobtainthattheRFRismoreaccuracythanOLSR.Inordertoillustratetheimportaneeofindependentvariables,wedefineastatisticasacriterion.
WeuseClusterAnalysis,TimeSeriesAnalysisandtheRandomForestsRegressiontoanalyzetheunemploymentrateamongdifferentregions.
Keywords:
SystemCluster;TimeSeries;RFR;OLSR;
1.Introduction
Twoyearsafterwewillfindajob,atthattime,theunemploymentratewillbeassociatedwithus.Nowletusexploretheunemploymentrate.
Unemploymentrateisanimportantindexofthecapitalmarket,isalaggingindicatorcategory.Theincreaseinunemploymentisaweakeconomysignal,canstimulateeconomicgrowth.Onthecontrary,theunemploymentratedroppedwillbetheformationofinflation.
Weanalyzethesituationofallcountries,thefurtherresearchwillgetthelevelofunemploymentrateinallcountries,andthenwepredictthetendencyofunemploymentrate
2.Notations
Table1indicators
Indicator
meanings
GDP
GrossDomesticProduct
FS
PublicFinanceExpenditure
M2
CurrencySupply
EP
TheEconomicActivity
PC
PeopleFinalConsumption
CPI
ConsumerPriceIndex
EG
EnergyConsumption
UR
RateofUnemployment
3.MeanandVarianceofUnemploymentRate
Themeanofunemploymentratereflectsthelevelofeconomicdevelopment,whengettingthemeanofURamong35countries,wecandiscoverthefactthatthemeanofURchangefromlessthan2%tomorethan14%,whichillustratesclearlythatlevelsofunemploymentratearedifferentamongthesecountries.Thefigure1canhelpusseethatwell.WearesurprisedatSpainwhoseunemploymentrateachieves15.39%.
TheMeanofUR
in
country
Figurelthemeaningsofindicators
Inreality,differentcountriesowndifferentlevelofdevelopmentsothatthevalueofURisnotaconstant.Therefore,wewantfurthertoknowtheURvarianeewhichrevealtheeconomicstability.Thefigure2reflexesthatthevolatilityofvarianeeofdifferentcountries.
TheVarianceofUR
ChinaBulgariaGermanyIsraelNorwayRussiaBritain
country
Figure2VarianeeofUR
Atthelast,wegainthetopthreeofURmeanandvarianee,asfollow:
Table2Topthreeoftwoindicators.
country
Spain
Poland
Bulgaria
mean
15.39
14.22
13.76
country
Spain
Bulgaria
Poland
varianee
29.66
18.76
13.04
Fromtable2,weseethatSpain,PolandandBulgariaarethetopthreebothofthetwoindicators,that'stlwseyjovernmentsdidbadlyinthefieldofemployment,andtheireconomicenvironmentisunstable.
4.SystemClusterAnalysis
Nowweanalyzetheunemploymentrate,whethertheyareatasimilarlevel,weutilizeClusterAnalysistoclassify[1].
4.1Expressthedistancebetweencountries
ThefollowformuladescribesthedistaneebetweentwopointsusingEuclideandistanee,
i
(p2
dj=Z(诲-Xjk)
(1)
Ik」丿
AdvantageofEuclideandistaneeiswhentheaxisorthogonalrotation,theEuclideandistanceismaintained.
4.2Thedistanceamongclasses
Hereweselecttheaveragelinkagemethodtoexpressthedistanceforbothofclasses.Usingaveragelinkagemethodisagoodwayofallthesamplesbetweeninformation
Where%andnLarerespectivelystandforthenumberofsamplesinclassesGkandGL.Theindexd0isthedistancebetweenthesamplesiinGkandthesamplejinGk.
4.3TheClassingStatistic
Wesetupnasthetotalnumberofsamples,dividingoriginalsampleintokclasses,eachclasshavenisamplesandwederivethepseudoFstatistic:
厂W-R/k-1
Pk/n-k
Where
n'
W二'Xj_X]〔Xj「X
k#
Pk八Wi
i=1
ni
ThebiggerPseudoFstatisticvalueandthesmallerkvalueisthebettereffectofclassification.
Weobtaintheexaminationappealthroughclusteringanalysismethod,asshowninthedifferentnumberofclustersofStatistics.
Table3theinformationofcluster
Numbersofclusters
1
2
3
4
PseudoFstatistic
0
19.3
12.6
23.8
Numbersofclusters
5
6
7
8
PseudoFstatistic
19.3
17.3
15.1
19.5
Fromtable3,wefindthatwhenwedivideoriginalsampleintofourclasses,thepseudoFstatisticachievesthebestvalueaswellasth@valueisnotbig,thuswechoosethenumberofclassesasfour.Intheend,wegivethesystemdiagram.
Sp^ln
Poland
Bulgaria
Ireland
Russia
Turkey
Philippines
Israel
Germany
Greece
llaly
France
Finland
Romania
U.S.A
Sweden
Portugal
Hungaiy
Canads
New-Zealand
Denmark
Britain
Australia
Czech
Thailand
Norway
Holland
Iceland
ChiIna-Macao
China-Hongkong
Korea
Auslrla
Japan
China
0.00.51.D1.5
A^rag-eDigtancpBetw&e-niClusters
Figure3theclassresult
Ifclassifyingoriginalsampleintofour,thenwecometotheconclusionasfollows:
thefirstclassification:
China,Japan,Austria,SouthKorea,China
HongKong,Macao,Iceland,Holland,Norway,Thailand,Czech.Thesecondclassification:
Australia,Britain,Canada,NewZealand,Denmark,Hungary,Portugal,Sweden,Romania,Finland,American,France,Italy,Greece,Germany,Israel,Philippines,Turkey,Russia,Ireland
Thethirdclassification:
Bulgaria,Poland,Venezuela
Theforthclassification:
Spain
Weknowthattheunemploymentrateindexreflectstheoverallstateoftheeconomy,anditistheeconomicdataforeachmonthwithfirstpublished,sotheunemploymentrateindexcalledalleconomicindicatorsofthe"crownjewel".Itisforthemonthlyeconomicindicatorssensitiveonthemarket.
5.TimeSeriesModel
Atimeseries[2]isasetofobservationsxt,eachonebeingrecordataspecifictimet,andobserveddata{xt}isaspecificationofjointdistributions(orpossiblyonlymeansandcovarianee)ofasequenceofrandomvariables[Xt}ofwhich{xt}ispostulatedtoberealization.
5.1StationaryTest
Ifwewanttomakeagoodtimeseriesmodel,weshouldrecognizeitfirstly.Akeyroleintimeseriesanalysisisplayedbyprocesswhoseproperties,orsomeofthem,donotvarywithtime.
WechoosethedataofChinaasexampleandcurvethigure4
china
Figure4thescatterofUR
Formfigure4,wecanfindthattherateofunemploymentofChinahasincreasedtrendwiththetimegoing,thusweneedtomakeitstationarybeforeweconstructmodel.
5.2Stationaryprocess
Weproceedtomakethetimesequencestationarywiththefirstdifference,andwegivethescatterdiagram
dchlna
2
Figure5thescatterafterdifference
Fromthefigure5,weperceivethedatatendtostationary,whichsuggestthatwecanconstructtimeseriesmodel.
5.3TheARMAModel
WemakeanARMAtofitthedataofChinaUR,themodelisasfollows:
nm
Xt八讥」*t八jt_j
idjd(4)
TheparameteriofformulaisanautomaticregressionparameterofARMA,parameter咼isamovingaverage.ParametertisaStochasticProcesswithazeromeanandanormalwhitenoise,thats'o
2
say,tNo,;「a
Especially,whenm=0,themodelARMAn,missameasARn,theformulaisasfollows:
nXt=嘉:
卩iXt4*t
=(5)
InordertomakesuretheparametersofARMA,wegivetheautocorrelationfigure6.
Figure6thePACFandACF
Fromthefigure6,wecanfindtheautocorrelationcoefficientisfirsttruncationandthefirstpartialcorrelationcoefficientistwotimesthanstandarderror.SowepreliminarymakeitasAR
(1)orMA
(1).
Tomakesurewhichmodelisbetter,wecomputesomestatisticsoftwomodels.
TheinformationofmodelAR
(1)
Table4theARinformationofAICandBIC
ConditionalLeastSquaresEstimation
Parameter
Estimate
StandardError
T-Value
Approx
Pr>|t|
Lag
MU
2.47236
0.21209
11.66
<.0001
0
AR(1,1)
1.00000
0.04125
24.25
<.0001
1
1.11E-6
0.044984
0.212094
-3.37869
-1.38722
20
ConstantEstimateVarianeeEstimateStdErrorEstimateAICSBC
NumberofResiduals
TheMA
(1)modelparametersandAIC,SBC:
Tabie5ConditionalLeastSquaresEstimation
Parameter
Estimate
StandardError
T-Value
Approx
Pr>|t|
Lag
MU
3.34510
0.18211
18.37
<.0001
0
MA(1,1)
-0.85777
0.12761
-6.72
<.0001
1
Table6theMAinformationofAICandBIC
ConstantEstimate
3.345103
Varianee