人民币实际汇率购买力平价实证研究

华中科技大学博士学位论文人民币实际汇率购买力平价实证研究姓名：徐国希申请学位级别：博士专业：西方经济学指导教师：王少平20060509ARRogoffPPPAR1994120045Boostrap2.1PPP24%ARRogoffRogoff199695%AR95%ARGrangerTerasirta(1993)STARI19801200510LSTARLSTARLSTARARLSTARLSTARAR1AR1LSTARlowerregimeupperregimeTAR19801181%48Rogoff1995102005101012LSTARARLSTARLSTAR;;;;IIAbstractThePurchasingPowerParity(PPP)theoryisthebasisoftheExchangeRatedecisiontheory,soit’simportanttogiveanempiricalstudyonit,especiallyundertheglobaleconomybackground.BecauseoftheadvantagesofstraightandeasytounderstandofPPP,peoplealwayswishtodoempiricalstudyonit,althoughthehypothesisofitisverystrict.Almosteveryneweconometricmethodappears,itwillbeusedtotheempiricalstudyonPPP,becausePPPtheoryisveryimportanttotherealeconomy,justlikeforecastingfinancecrisis,usingasexchangeratenormalanchorandjudgingexchangeratemismatch.WiththedevelopingofChineseeconomy,it’simportanttootogiveanempiricalstudyonthePPPofChineseRMB’srealexchangerate.SincePPPtheoryhadbeentakenout,thereweremanypapersabouttestingit,buttheresultsalwayswereblowhotandcold.ThispaperwillgiveanempiricalstudyonthePPPofChineseRMB’srealexchangeratethroughtowwaysasfollow:first,wewillcalculatethehalf-liveofunitshock(HLS)ofRMB’srealexchangeratebasedonthelinearARmodel,usingthemethodofmedianunbiasestimationtotestiftheRogoffPuzzleofChineseRMBexistornot;afterthatwewilltestifthereverttoPPPoftherealexchangerateofChineseRMBexistornotbasedonthenonlinearmodel.ToARmodel,theOLSmethodhasanunderestimation,themedianunbiasestimationisaverygoodwaytocorrectthebias.BasedontheexchangerateofChineseRMBfromJanuary1994toMay2004,wecalculatetheHLSofChineserealexchangerate,usingthemedianunbiasestimationmethodandtheBoostrapsimulation,andtheresultisthatChineserealexchangerate’sHLSis2.1years,andtheratethatitconvergestoPPPis24%peryearwhichindicatesthattheproblemofRogoffPuzzlemaybenotsosieriousasRogoff(1996)hadsaid.AnotherresultistheprobabilitythattheChineserealexchangerateisstationaryis95%,whichindicatestheprobabilitythatthePPPofChineserealexchangerateisstandis95%inthesampleperiodbasedonthelinearARmodel.ButtherearemanyshortagestocharacterizethebehaviorsoftherealexchangeratebasedonthelinearARmodel.Becauseoftheexistenceofthetradecost,itwillbemoreacceptabletocharacterizethebehaviorsofrealexchangeratebasedonthenonlinearmodel.GrangerandTerasirta(1993)broughtforwardanonlinearSTARmodelwhichhasbeenusedwidelyintheexchangerateresearch.ThispaperwilldotheempiricalstudyonthePPPofChineseexchangerateontheSTARmodel,andgiveanimprovementtothetransitionfunctionofit.TheresultofmystudybasedontheexchangerateofChineseRMBfromIIIJanuary1980toOctober2005indicatesthatthebehaviorsofChineseRMB’srealexchangerateobeystotheLSTARmodel,whichindicatesthatChineseRMB’srealexchangerateisgloblelystationaryandconvergestothePPP,andthedynamicbehaviorofChineseRMB’srealexchangerateisasymmetry,thebehaviorthatChineseRMB’srealexchangerateconvergestoPPPnotonlydependenceonthesizeofthedeparturetothePPP,butalsoonthedirectionofthedeparture.Throughthecompareonsimulation,thenonlinearLSTARmodelismoreexcellentthanthelinearARmodeltocharacterizethebehaviorsoftherealexchangerateofChineseRMB,andtheimprovedLSTARmodelisbetterthantheunimprovedmodel.Duringthesampleperiod,therealexchangerateofChineseRMBisintheupperregimeinmosttime,inwhichChineseRMB’srealexchangerateisastationaryprocessofAR(1)model,andisinthelowerregimeinsometime,inwhichChineseRMB’srealexchangerateisaexpandprocessofAR(1)model,thismeansChineseRMB’srealexchangerateisstationaryinthelongrunandunstationaryintheshortrun.ThetransitionfunctionofChineseRMB’sLSTARmodelisverysteep,thismeansthetransitionbetweenlowerregimeandupperregimeisveryquick,whichclosetotheTARmodel.TheimpulseresponsefunctionoftheestimatedLSTARmodeldescribesthatwiththestockbecominglargertheimpulseresponsefunctionbecomessteeper,thismeanswiththedeparturebecominglargertherevertingoftherealexchangerateofChineseRMBtoPPPbecomessooner.Whenthedirectionofthestockisnotthesame,thefunctionofthepositivestockislargerthanthatofthenegativeshock,thismeansthattheadjustmenttoPPPwhentherealexchangerateofChineseRMBisoverateissoonerthanwhenitisunderrate.Atthesametime,theHLSofthenonlinearmodelisnotonlyrelatedtothesizeoftheshock,butalsotothehistoryandthedirectionoftheshock.UsingJanuary1980asthestarttime,wecalculatetheHLSoftherealexchangerateofChineseRMB,whichis18months,whilewhentheshockis1%theHLSoftherealexchangerateofChineseRMBis48months,indicatesthattheproblemofRogoffPuzzleofChineseRMBexistpartially.UsingthedatafromOctober1995toOctober2005,wedoatwelvestepaheadforecastingandtheresultshowsthattheeffectoftheforecastingofthenonlinearLSTARmodelisbetterthanthatofthelinearARmodel,andtheimprovedLSTARmodelisbetterthanthemodelhasn’tbeenimproved.Intheend,accordingtothestudyresultofthispaper,weputforwardsomeadvicesaboutChineseexchangeratepolicies,andgiveaprospecttothefuturestudyofourwork.Keywords:PurchasingPowerParity;ChineseRMB’srealexchangerate;Medianunbiasestimation;Nonlinearmodel;EmpiricalstudyIV20065102006510200651011.1PPPPPPPPPPPPEngel,19961Rogoff(1996)RogoffHLS3-513-20%Ro