The results show that housing prices, GDP and the population in the economy have a positive effect on household borrowing.. Meanwhile, interest rates, the unemployment rate, the number o
Trang 1The determinants of Australian household debt: A macro level
study
UNE Business School, University of New England, Armidale, NSW 2351, Australia
1 Introduction
Therapidincreaseinhouseholddebtinthelasttwentyyearshasbeenaninternationalphenomenon,whichhasalso occurredinAustralia.AsTable1shows,Australia’saccumulatedhouseholddebtlevelincreasedfromA$187billionin1990
toA$905billionin2005.Thedebt-incomeratiojumpedfrom70.6%in1990to162.8%in2005,eventhoughthegearingratio (total liability aspercentage of totalassets) was only 18.6%in dueto arapid increasein totalassets To putthis into perspective,theaverageAustralianhouseholdwouldhavetoworkformorethanoneandahalfyearsjusttopayofftheir debt
This accelerated growth of Australian household debt has generated serious concerns The Economist (2003, p 70)
reported:
‘The profligacyof American and British households is legendary,but Australians havebeen even more reckless, pushingtheirborrowingtoaround125percentofdisposableincome therearenowconcernsthatunsustainable ratesofborrowingwillsoonerorlaterendintears.’
A R T I C L E I N F O
Article history:
Received 13 August 2012
Received in revised form 22 August 2013
Accepted 30 August 2013
Available online 12 September 2013
JEL classification:
E24
C32
H60
Keywords:
Australian household debt
Cointegrated VAR modelling
Housing market
Housing prices
Interest rates
A B S T R A C T
This paper employs a cointegrated Vector Autoregression (CVAR) model to explore the determinants of Australian household debt The results show that housing prices, GDP and the population in the economy have a positive effect on household borrowing Meanwhile, interest rates, the unemployment rate, the number of new dwellings and inflation are found to have a negative effect on Australian household debt Of these, interest rates are the most significant Based on these results, it is judicious to rein in household debt during economic booms through monitoring and intervening in the assets market and using monetary policy in a timely, comprehensive, and careful manner.
ß 2013 Elsevier Inc All rights reserved.
* Corresponding author Tel.: +61 2 6773 2682; fax: +61 2 6773 3596.
E-mail address: nam.hoang@une.edu.au (N.T Hoang).
1 Tel.: +61 2 6773 2046; fax: +61 2 6773 3596.
2 Tel.: +61 2 6773 2501; fax: +61 2 6773 3596.
ContentslistsavailableatScienceDirect
1049-0078/$ – see front matter ß 2013 Elsevier Inc All rights reserved.
Trang 2claimedthatthefinancialpositionofAustralianhouseholdswaswithinaserviceablerange.However,thisperspectivemay
be over-optimisticbecausea low gearingratio doesnotguaranteethe safety of Australian householdsthat have large amountsofhousingmortgagedebt
Fig.1disaggregatesAustralianhouseholddebtintothreecategories:owner-occupiedhousingdebt;investmenthousing debt;andotherpersonalloans.Whatisimmediatelynoticeableisthatowner-occupiedhousingaccountsformorethanhalf
oftotalhouseholddebtintheperiodunderconsideration.Sinceitisnotpracticaltopaybackthisdebtbyselling owner-occupiedhousing(householdmembersmustliveinthishouse),thegearingratioislessrelevantthanthedebt-incomeratio when the household repayment burden is considered The other distinguishing feature is that debt associated with investmenthousingincreasesdramaticallyduringtheperiod,fromlessthan5%ofdisposableincomein1990tomorethan 40%in2008.Thisincreaseininvestmenthousingdebtislargelydrivenbyrisinghousingpricessincethelate1990s.Inturn,it
isalsoasignificantfactorininflatinghousingprices.Thisdebtbreakdownexplainsthehighdebt-incomeratioandthelow gearingratio in theAustralian household sector.Largeamountsof debtlead toahigh debt-incomeratio,but the high percentageofmortgagedebtsecuredonhousingassetsreducesthegearingratio.Sincehousingassets,asthedominant assets inAustralianhousehold balancesheets,havebeenhighlyinflatedinrecentyears,adeclinein housingprice will significantlydecreasehouseholdassetvalue,andthusincreasethegearingratio.Consequently,thelowgearingratiodoes notguaranteethatAustralianhouseholddebtisoflowrisk
ThankstotheGlobalFinancialCrisis(GFC),thegrowthofAustralianhouseholddebtisnowslowingdownandappears lessofaconcern.However,inordertoavoidaneventsimilartotheUnitedStates(US)sub-primecrisis,itisnecessaryto identifythefactorsthataffectAustralianhouseholddebt
Theremainderofthepaperisorganisedasfollows.Section2reviewspreviousstudiesonhouseholddebt.Section3
discusses the construction of the dataset.Section 4is aimed at testingand estimating the empirical model.Section 5
interprets anddiscusses the main findingsfrom theempirical model.Section 6summarises the mainconclusions and providessomepolicysuggestions
Table 1
Household liabilities, GDP and disposable income, A$ billion.
Year Total liabilities Total assets Disposable income Liabilities as % of total assets Liabilities as % of disposable income
Source: Reserve Bank of Australia, Table B20 http://www.rba.gov.au/statistics/tables/index.html Australian Bureau of Statistics, cat, no 5204.0 http:// www.abs.gov.au/AusStats/ABS@.nsf/MF/5204.0
Fig 1 Composition of Australian household debt 3
3
Trang 32 Previousstudies
Therapidriseofhouseholddebtisarelativelynewphenomenon,andthusitisraretofindstudiesonhouseholddebt before the 1990s However, booming household debt of the 1990s triggered considerable academic interest, and consequently a substantial amount of literature has been published As the purpose of this paper is to explore the determinantsofAustralianhouseholddebt,onlypapersconcerningthefactorsaffectinghouseholddebtarereviewedhere
AnumberofempiricalstudiesontheeffectsofhouseholddebthavebeencompletedforcountriesotherthanAustralia.In South Africa,Aron andMuellbauer (2000)utilised adjustedSouthAfricanReserveBank datatoestimate the impactof financialliberalisationonhousehold consumptionand householddebt Theyconcludedthat financialliberalisationand fluctuationinasset valueshaveimportant implicationsforconsumerspending andincreasinghousehold debtinSouth Africa.IntheUS,BarnesandYoung(2003)employedacalibratedpartialequilibriumoverlappinggeneration(OLG)modelto explainhouseholddebtintermsofconsumption-incomeandhousing-financemotivations.Theyfoundthatthesubstantial riseofhouseholddebtin the1990scouldbeexplainedby realinterest rate,incomegrowth expectations,demographic changes,andtheremovalofcreditconstraint.TudelaandYoung(2005)alsousedtheOLGmodeltoanalysehouseholddebt
intheUKandclaimsthatchangesininterestrates,houseprices,preferences,andretirementincomeaffecthouseholddebt
theeffectsofvariousfactorsonhouseholddebt,Manyfactorswerefoundtoinfluencehouseholddebt,includinghousing stock, interestrates, thenumberof house sales,wageincome, housingprices, unemployment rate,and thenumberof students.MartinsandVillanueva(2003)constructedadatasetcombininghouseholdsurveydataandadministrativerecord
ofdebttoestimatetheresponsivenessoflong-termhouseholddebttotheinterestratechangeinPortugal.Theyfoundthat theelasticityoftheprobabilityofmortgageborrowingtoachangeintheinterestratewaslargeandnegative.Magri(2002)
analysedthedeterminantsofItalianhouseholdparticipationinthedebtmarket,usingdatafromtheBankofItaly’sSurveyof HouseholdIncomeandWealth.Theresultssuggestedthatage,income,livingarea,andtheenforcementcostofbanks,have importanteffectsonhouseholddebt.Kearns(2003) employedhousehold-level datatoexplorewhyhouseholdsfellinto mortgagearrearsduring the1990sin Ireland.Hisstudysuggestedthatamodestriseofinterestrateswouldresultin a substantialrepaymentburdenforasignificantnumberofnewlymortgagedhouseholds.Heconcludedthatthecontinuing strong growth of mortgage lending, causedby relaxed lending criteriaand by household willingness toaccept higher repaymentburdens,couldleadtoahigherrateofmortgagearrearsamonghouseholds.Thaicharoen,Ariyapruchya,and
Thaihouseholds.TheysuggestedthatcurrentdebtlevelsinThailanddidnotposeathreattofinancialstabilityandthe macro-economy
Somestudieshaveintegrateddatafrommultiplecountries.Debelle(2004)analysedthepossibledeterminantsand the macroeconomic implications of rising household debt He argued that the rise of household debt reflected the responseofhouseholdstolowerinterestratesandaneasingofliquidityconstraints.Increasedhouseholddebtitselfwas notlikelytobethesourceofnegativeshockstotheeconomy,howeverhesuggesteditcouldamplifyshocksfromother sources.Crook(2003)comparedstudiesoftheeffectsofhouseholddebtacrossanumberofcountries.Hefoundthatthe debtholdingbyagefollowedthelifecyclepatterninallcountriesobserved,thattherewereconsiderablevariationsinthe determinantsofdesiredlevelsofdebt,andthattherewasintra-andinter-nationalvariationinthemarginaleffectsof householddebt
InAustralia,promptedbyrisinghouseholddebtlevels,someinstitutionshaveincludedhouseholddebtinformationin theirsurveys,orinitiatedsurveysonthistopic TheAustralianBureauofStatistics(ABS)hasconductedseveralwave surveys on Household Expenditure (HES) Melbourne University commenced the Household, Income and Labour DynamicsinAustralia(HILDA)Surveyin2001,fundedbytheFederalgovernment,andhadcompleted6wavesatthetime
ofwriting TheReserveBankofAustralia(RBA)completedasurveyin2006onHouseholdBehaviourAroundHousing Equity
Anumberofstudiesatthehouseholdsectorlevelhavebeenbasedonthesesurveys,andprovideinformationatthemicro level.Bray(2001)studiedfinancialstressinAustraliausingthe1998–99HESsurvey.LaCavaandSimon(2003)employeda logit modelusing thedata from the HESand HILDA surveystostudy the factorsaffectingthe financial constraintson Australianhouseholds.Schwartz,Hampton,Lewis,andNorman(2006)usedbivariateandlogitmodelstoanalysethedata obtainedbyRBAsurveysonHouseholdBehaviourAroundHousingEquity
Atthemacrolevel,theRBAandotherinstitutionshavepublishedanumberofpapersandpresentationsonhousehold debt.Stevens(1997)emphasisedthepositiveeffectoflowinflationonhouseholdborrowings.TheRBA(1999)attributedthe rapidgrowthofpersonalcreditto:innovationsinproductsofferedbybanks;theincreasinghouseholdpreferencetowards theuseofcreditcards;andcontinuingeconomicexpansionwithlowinflationandlowinterestrates.TheRBA(2003)also illustratedthecompositionanddistributionofhouseholddebt,suggestingthatlowinterestrates,lowinflationrates,and financialderegulationmayhavecontributedtoariseinhouseholddebt.Anumberofstudiessupportthisview.TheTreasury
and interest rates, and partly reflected the improved product choice and reductions in borrowing costs caused by deregulationofthefinancialsectorinthe1980sand1990s.TheANZbank(2005)claimedthatrisinghouseholddebtlevel wascausedbyasustainedboominhouseprices,andthatthesustainabilityofhouseholdcreditdependedonthegrowthof householddisposableincomeandemployment
Trang 4All of thesestudies onAustralian household debt areinstructive However,they present only financialanalyses of Australianhouseholddebt.ToprovideamorecomprehensivepictureofAustralianhouseholddebt,thisstudyidentifiesthe determinantsofAustralianhouseholddebtbyemployingempiricalmodels.Themodelsusedatafromhouseholdaccounts, microeconomicdatafromsurveys,andmacroeconomicdata
3 Dataset
Thehousehold debt level isjointly determinedby supply and demand That is, the availabilityof funding,and the household’sdecisiontotakeondebt.However,themacroeconomicenvironmentultimatelydeterminesbothsupplyand demand.Consequently,thedeterminantsofhouseholddebtmustlieinthesemacroeconomicfactors.Byanalysingfactors affectingborrowingand/orlending,wecanthereforeidentifypotentialdeterminantsofhouseholddebt
Withregard todemand,householddesireforborrowingis subjecttothelevelof householddisposableincomeas wellasthepurpose/sforborrowing.Householddisposableincomeiscomprisedofhouseholdgrossincomeplussocial transfer,lessincometaxpayableandotheroutlays.Wefocushereonhouseholdgrossincome,sinceincometaxratesin Australiahaveundergoneminimalchangeinrecentdecades,whilesocialtransferandotheroutlaysaresmallrelative
to household gross income Household gross income includes wage income and gross mixed income, as well as domestic andoverseas investmentincome.At themacro level,these factorscanbe approximatedbyGrossDomestic Product(GDP)
Thepurposesofhouseholdborrowingincludesmoothingconsumptionandinvesting.Thelevelofconsumptionisclosely relatedtotheAustralianpopulationandthepricelevel(ConsumerPriceIndex–CPI).Itmayalsoberelatedalsotothemacros affectingconsumerconfidence,suchasunemploymentrateandGDP.Investmentdecisionsaretypicallyrelatedtointerest rates.Moreover,fromthedisaggregationofAustralianhouseholddebt,wefoundhousingtobethemaininvestmentvehicle forAustralianhouseholds.Inconsideringthelargeamountofhousingdebt,housingpricesandthenumberofnewhouses enteringthemarketareimportantfactors
Withregardtosupply,theavailabilityoffundingandeaseofobtainingfinancearelargelyindicatedbyinterestrates However,toreducecreditrisk,lendersmaytakeintoaccounthouseholdincomelevelandothermacroeconomicvariables, includingunemploymentrate,inflationrateandGDP.Amongthesefactors,thehouseholdincomelevelcanbeapproximated
by GDPandthe inflationrateis amonotonictransformationofCPI Includingallpossiblevariables affectingAustralian householddebtyieldsthefollowingdataset:
X¼ðDEBT;GDP;NDWELL;HPI;R;U;CPI;POPÞ (1)
whereDEBT–accumulatedhouseholddebt;GDP–grossdomesticproduct;NDWELL–numberofnewdwellingsapproved; HPI–housingpriceindex;R–interestrate;U–unemploymentrate;CPI–consumerpriceindex;POP–population
It may beargued that real percapitadata arepreferable because theycanreduce heteroscedasticity in the model Nonetheless,thisstudyuses nominalvaluesforanumberofreasons.First,theWhitetestsshowthe heteroscedasticity probleminthemodelisnotseriouswhennominalvaluesareused.Second,measurementerrorsareoftenaproblemin macrotimeseries,andtransformingnominalvaluetorealand/orpercapitavalue(dividedbyCPIorPOP)maymagnifythese errors.Finally,someinternationalstudiesshowthatpopulationandinflationhaveasignificantinfluenceonhouseholddebt Thus,anominalvalueisrequiredtotestifthisholdstrueforAustraliaaswell
Thequarterlytimeseriesdatafrom1988Q2to2011Q2forthevariablesintheinformationsetwerecollectedfromthe AustralianBureauofStatistics(ABS,2011),theReserveBankofAustralia(RBA,2011)andotherinstitutions.Themajorityof datawereobtainedfromtheABS,includingpopulation,unemploymentrates,housingprices,GDP,andtheCPI.Thedatafor newdwellingscommencementscame fromtheHousingIndustryAssociation(HIA,2011).Dataforhouseholddebtand officialinterestrateswereprovidedbytheRBA.Duetodifferentsourcesanddifferentmeasurementofdata,somedata sequenceswereadjustedbeforeuse.Specifically,thedatasetinthisstudyisdescribedasfollows:
Householddebt(DEBT):seasonallyadjustedquarterlydata,measuredinbillionAustraliandollars(A$billion),attheendof quarter
GDP:seasonallyadjustedquarterlydata,measuredinbillionAustraliandollars(A$billion)
Consumerpriceindex(CPI):baseyear1989/90=100
Housingpriceindex(HPI):CPIonhousing1989/90=100
Interestrate(R):officialinterestrate,quarterlyaveragedmonthlydata
Unemploymentrate(U):quarterlyaveragedmonthlydata
Population(POP): measured in thousandpersons, ABS estimated quarterlydata Sincethe data were availableonly fromJuneof1989onwards,theannualpopulationin1988providedbyABShasbeenadoptedasthedatainJuneof1988 ThedatafromSeptember1988toMarch1989havebeencalculatedassumingthatpopulationgrowthratewasstablein thisperiod
NumberofNewdwellingapprovals(NDWELL):measuredinthousanddwellings,includingalltypesofhousingsuchasunits andhouses
Trang 54 Theempiricalmodel
Since macroeconomic variables are notorious for non-stationarity, cointegrated VAR analysis is used to study the relationshipbetweenAustralianhouseholddebtandothermacroeconomicfactors
ThelikelihoodestimationofacointegratedVARmodelfortimeseriesdataintegratedoforderoneI(1)wasdevelopedby
canbeexpressedasfollows:(2)DXt¼ab0Xt1þXk1
i¼1
GiDXtiþet; t¼1; ;Ta;barePrmatrices
BytheGrangerrepresentationtheorem,themovingaverage(MA)formoftheVECMmodel(2)isgivenby:
Xt¼CXt
i¼1
whereC¼b?ða0
?G b?Þ1a0
?
G ¼IXk1
i¼1
Gi
C*(L)isaninfiniteorderpolynomiallagoperatorwhichresultsCðLÞet isthestationarypartoftheprocess.a?;b? are orthogonalcomplementmatricesofa;bwhichsatisfiesa0a? ¼0;b0b? ¼0.X0containstheinitialvaluesoftheprocessXt Premultiplying(2)withanon-singular p pmatrixB,wehavethestructuralmodelwithcontemporaneouseffects:
BDXt¼Bab0Xt1þBXk1
i¼1
ut¼Bet,whichisthestructuralshock
TheMArepresentationof(4)canbeobtainedbysubstitutinget¼B1ut in(3):
Xt¼CB1Xt
i¼1
MatrixC¼b?ða0
?G b?Þ1a0
? hasrank(pr),implyingthatmatrix ˜C¼CB1has,atmost,rcolumnsofzero.Torecover the structural shocks, wehavetodecidethe matrix Bsothat the residuals etcanbe transformedinto permanentand transitoryshocksut¼ðuP
t;uT
tÞbythetransformationut ¼Bet whereutINð0;IÞ
WefollowtheapproachbyJueslius(2006)andDennis,Hansen,Johansen,andJuselius(2005)todefinethematrix
B¼H1KG; G¼ a0V1
a1
?
whereV¼Eðete0Þ;His the Choleskydecomposition ofðKGVG0K0Þ, and Kis a ðppÞblock diagonalmatrixtoimpose restrictionsonmatrixut¼Bet andmatrix ˜C0¼Cð0ÞB1,whichareneededforidentificationofthestructuralmodel.The restrictionmatrixKisconstructedsothat itimposesð1=2ÞðprÞðpr1Þzerorestrictionsin thelast(pr)non-zero columnsofmatrix ˜Ctalsoimposesað1=2Þrðr1Þdimensionaltriangularmatrixofzerosinthefirstrcolumnsofmatrix ˜C0
In the abovemodel, b0Xt1 arethe cointegrationvectors, including allvariables in the dataset,namely, DEBT, GDP, NDWELL,HPI,R,U,CPIandPOP.aistheadjustmentcoefficientsmatrix.DXtisthefirstdifferenceofX.Giisthecoefficient vectorassociatedwithDXti.SinceGirepresentsthedynamiceffects(howlaggedchangecanaffectpresentchange),itcan takeanysigndependingonhowthevariableswillbehaveintheshortrun.However,thecointegrationvectorXt1indicates thelongrunrelationship,sothesignofcoefficientsonthemshouldbeconsistentwitheconomictheory.InTable2welistthe expectedsignsofelementsinthecointegrationvectorbasedoneconomicreasoning
GDPshouldhaveapositivesignfortworeasons.First,GDPindicatesthesizeoftheeconomyandaggregatehousehold income and thus the size of debt, given the unchanged debt-income ratio Second, rapid GDP growth indicates good economictimeswhichwillencouragebothborrowingandlending.TheimpactofNDWELL(thenumberofnewdwellings)on householddebtisambiguous.Morenewdwellingsmeansmorehousingassets,iftheratioofmortgagedebttohousing assetsisunchanged,andimpliesmorehousingdebt(amajorpartofhouseholddebt).However,increasedhousingsupply mayleadtolowerhousingpricesandthuslesshousingdebt.TheeffectofHPIshouldbepositivebecauseofitsdirectpositive impactonhousingdebt.Theimpactofinterestrates(R)dependsontheperspectivesofdifferentagentsinthehousehold debtmarket.Intheviewofborrowers,ahigherinterestratemeanshigherborrowingcostandthuslessborrowing.Inthe
Table 2
Expected signs of coefficients (DEBT as a dependent variable).
Trang 6onthelendingsideiscreditrisk,sotheimpactofinterestratesonsupplysideisexpectedtobesmall(theunemployed capitalhastofindan outlayevenifinterestratesarelow) andthustheoveralleffectshouldbepositive.Theimpactof unemployment(U)andinflation(indicatedbyCPI)canbeeitherpositiveornegative.Ahigherunemploymentratemay indicate higherdemand forfinancialassistance (e.g borrowing).But,on the otherhand,it creates additionalfinancial constraintforhouseholds.Anunemployedpersonislesslikelytorepaytheirdebtintime,andthusismoreunlikelytoobtain
aloan.Highinflationcanstimulatedebt-financedpurchaseofphysicalassets(e.g.housing)butitwilldiscouragelending becauseofdiminishing (orevennegative)realreturns.Theimpactof population(POP)shouldbe positivebecauseitis positivelyrelatedtothenumberofhouseholdswithdebt,otherthingsbeingequal
Sincemost macroeconomictimeseriesdata arenon-stationary,unitroottests havetobeperformed Beforeformal testingproceduresareundertaken,thetimeseriesdataareplottedtoallowforvisualinspection.Mosttimeseriesshow apparenttrends(exceptNDWELL).TheDickey–FullertestwithGLSdetrending(DF-GLS)isperformed(theresultsofthese testsbeingavailableuponrequest).Thetestssuggestfirstorderintegrationforallvariablesata5%levelofsignificance
forGDPat2008Q4,Rat1993Q1andUat1993Q1.TheproceduredevelopedbyPerron(1989)andextendedbyZivotand
showsthatexistenceofstructuralchangedoesnotaltertheDF-GLStestsresults.Thet-valuesfora1aremuchhigherthanthe valuesgivenbyZivotandAndrews(1992)at0.05level.Thus,weconcludethatalltimeseriesinthemodelareI(1) Since we are working with non-stationary data, the co-integration test must be performed Both Trace and Max-eigenvalue tests are used In implementing the tests, a deterministic linear trend and intercept are included in the cointegrationequation.Thestructuralchangesinthemodelwereconfirmedusingtherecursivetestforstablecointegration vectorbdevelopedbyHansenandJohansen(1999).Theforwardandbackwardrecursivetestsindicatethattherearebreaks
at1993Q1and2008Q4,sothetrendchangesat1993Q1and2008Q4arealsoincludedinthecointegrationtestthroughtwo dummyvariablesDM93Q1andDM08Q4.Thetracetestsuggests6cointegrationequations,whilethemax-eigenvaluetest suggestsonly5cointegrationequations(seeTable3
However,thepresenceofinterventiondummies(DM93Q1andDM08Q4)mayaffectthedistributionofcointegration tests (Johansen,Mosconi,& Nielsen,2000andJoyeux, 2001) Usingthe methodandthe simulationresultsprovidedby
Accordingtothesecriticalvalues,thereareonly4cointegrationrelationshipsinthemodel.Moreover,basedonthestudies
propertyofthetestisnotapplicable.FollowingthesuggestionbyCheungandLai(1993),theauthorsuseT/(Tnk)(whereT
isthesamplesize,nisthenumberofvariablesinthemodelandkisthenumberoflags)toscaleupthecriticalvaluewith breaksoastoobtaintheadjustedcriticalvalueswithbreaks.ThisisshowninthelastcolumnofTable2.Accordingtothese adjustedcriticalvalueswithbreaks,thereisonlyonecointegratingrelationshipatthe0.05levelofsignificance
Next,thevectorerrorcorrectionmodel(VECM)isestimated.Aconstantandageneraldeterministictrendareincludedin the cointegration equation Two lags are chosen to minimise Schwartz criterion (SC) The estimated cointegration relationshipandadjustmentcoefficientsareshowninTable4
ThefirstpanelofTable3showstheestimatedcointegrationvector.Theestimatedresultssuggestthat,withexceptionof CPIandDM93Q1,allvariableshaveasignificantinfluenceonhouseholddebtinthelongrun.Thesecondpanelshowsthe adjustment coefficients from the VECM estimation results The adjustment coefficient for DEBT has the correct sign (negative)andissignificant,whichmeansthattheerrorcorrectionisquiteeffective
Table 3
The results of cointegration tests with structural changes.
No of CE(s) Max-Eigen
statistic
Critical value (0.05)
No of CE(s) Trace
statistic
Critical value (0.05)
Critical value with breaks (0.05)
Adjusted critical value with breaks (0.05) None c 88.82414 68.81206 None c , a , b 460.0601 273.1889 357.1008 459.1296
At most 1 c 86.08198 62.75215 At most 1 c , a 371.2359 228.2979 301.2805 387.3606
At most 2 c 77.40195 56.70519 At most 2 c , a 285.1539 187.4701 251.3317 323.1408
At most 3 c 63.64656 50.59985 At most 3 c , a 207.7520 150.5585 206.1153 265.0054
At most 4 c 51.79813 44.49720 At most 4 c 144.1054 117.7082 165.1130 212.2881
At most 5 29.37749 38.33101 At most 5 c 92.30730 88.80380 128.2082 164.8391
Source: authors’ tests.
Note: 2 lags are chosen to minimise Schwartz criterion Minimising AIC will lead to more lags and more cointegration vectors when standard critical values are used But when the adjusted critical value is used, the test also indicates only one cointegration vector.
a Denotes rejection of the hypothesis at the 0.05 level according to the critical value with breaks.
b Denotes rejection of the hypothesis at the 0.05 level based on the adjusted critical value with breaks.
c Denotes rejection of the hypothesis at the 0.05 level according to the standard critical value.
Trang 7TheestimationpassesalldiagnostictestsshowninthelastpanelofTable3.TheECMequationforDEBThasareasonable R-squared(0.68)andadjustedR-squared(0.64),indicatingthattheequationexplainsthebehaviourofchangeinhousehold debtintheshortrunwell.TheLMtestsupto10thlagcouldnotfindautocorrelationatasignificantlevel:thelargep-value indicatesthatwecouldnotrejectthenullhypothesisofnoserialcorrelation.TheJarque–Beratestconfirmsthenormal distributionofresidualsfordisplayedECMwithap-valueof0.3068:evenfortheVECMsystem,theJ.B.testcouldnotreject thenullofnormalityatconventionallevel.TheWhitetestshowsnoseriousheteroscedasticityprobleminthemodel Finally,wecanestimatethestructural-contemporaneouslong-runimpactfromthestructuralmodel.Sincethereisonly onecointegrationvectorinthemodel,onetransitoryshock,andsevenpermanentshocks.Theresponseofhouseholddebtis themainconcernofthisstudy,andeconomicreasoningshowsthatallothervariablesshouldhaveaninfluenceonhousehold debt We thereforeviewthe shock ofDEBT astransitory andthe othershocksas permanent Theorderof variablesis importantinimposingrestriction,throughwhichthestructuralmodelcanbeidentified.Theorderofothervariablesisbased
onthemacroeconomicinfluenceofeachvariableandthepurposeofthisstudy.Forinstance,DEBTisincludedattheendof thelistbecauseitisinfluencedbyanyothervariables.Ontheotherhand,GDPhasprofoundimpactsonmanyaspectsofthe economy,soispositionedfirst.TheorderofvariablesotherthanDEBTisarrangedaspresentedinTable2.Thepermanent andtransitorydecompositionofstructuralshocksislistedinTable5inthenextsection,andtherotationmatrixislistedin
5 Analysisofestimationresults
AccordingtothefirstpanelofTable3(thesignoncoefficientsinthetableshouldbereadopposite,sincethedependent variableDEBTisonthesamesideofothervariablesinthecointegrationvector),factorsnegativelyaffectinghouseholddebt areinterestrate,unemploymentrate,thenumberofnewdwellingapprovals,andCPI.Thoseactingpositivelyonhousehold debtincludehousingprices,GDP,andpopulation.Inaddition,bothdummieswithtrendsaresignificant.Wediscussthese threecategoriesoffactorsinturn
5.1 Factorsnegativelyaffectinghouseholddebt
First,themodelshowsthattheinterestratehasadramaticallynegativeeffectonhouseholddebt.Theintervalestimates showthatiftheinterestrateincreasesbyonebasepoint(0.01%),householddebtlevelwoulddecreasebyA$2.57billionover
Table 4
The results of VECM estimation and diagnostic tests a
Coefficient 1.000 13.4760 257.4072 1.4928 119.4678 24.9666 40.1032 24.2793 3.8336 16.5573 S.E (0.0039) (41.6646) (0.4491) (31.4057) (18.2657) (9.8366) (3.2325) (3.5016) (2.4254) t-Value [3.4318] [6.1781] [3.3236] [3.8040] [1.3669] [4.0769] [7.5111] [1.0948] [6.8265] Adjustment
coefficient
0.0348 4.6367 0.0007 6.51E05 0.00025 0.00082 0.00017 0.0072 0.0018 0.0197 S.E (0.0071) (1.7230) (0.00038) (1.3E05) (0.00013) (0.00068) (0.0013) (0.0026) (0.0021) (0.0065) t-Value [4.9011] [2.6910] [1.8392] [4.9057] [1.8985] [1.1985] [0.1348] [2.7382] [0.8519] [3.0527] Diagnostic tests:
R-squared: 0.6820, adj R-squared: 0.6377, Schwartz (for ECM): 7.2343, Schwartz (for VECM): 42.9581
VEC residual serial correlation LM tests
10th order
Jarque–Bera residual test for normality: J.B.(2) = 2.3630 (ECM equation for DEBT)
J.B.(20) = 22.5482 (VECM system)
p = 0.3068
p = 0.0673
Source: authors’ estimation and tests.
a There is a general trend and intercept in the model, but they are not displayed in the table 2-lag is chosen to minimise Schwartz criterion.
Table 5
Structural-contemporaneous long-run impact matrix: ˜ C ¼ CB 1
u T
7;t
Source: authors’ estimation.
Trang 8time.Thedirectreasonforthisisthatinterestrateriseswillincreasetheborrowingcost,whichinturnwilldeterhouseholds fromborrowing,oratleastreducetheamounttheyareinclinedtoborrow.Moreover,forhouseholdsthat havealready incurreddebt,therepaymentburdenwillincreaseifthedebtisbasedonavariableinterestrate,asisthecaseformost Australianhousingloans.Iftherepaymentburdenisunbearable,somehouseholdsarerequiredtoselltheirpropertytopay offtheirdebt.Asaresult,householddebtwilldecrease.Increaseininterestratesalsoindirectlyaffectshouseholddebtby discouraginginvestment,andthereductionininvestmentwillslowdownthewholeeconomy.Scalingbacktheeconomy mayreducehouseholdincome,increasetheunemploymentrateandthusreducehouseholdborrowing.Ifhouseholds’newly incurreddebtislessthantheamountoftheirscheduledrepayments,thehouseholddebtlevelwillalsodecrease FromtheresultswecaninferthatrisinghouseholddebtinAustraliamustbeassociatedwithadecreaseininterestrates Thisis exactlythecase.Therising householddebtin Australiahas coincidedwiththe lowinterestratesafter financial deregulation.Duringandshortlyafterfinancialderegulationinthe1980s,theAustralianinterestratesweregenerallymore than10%.Forexample,intheearly1990s,theofficialinterestratereached17%,therateforlendingtobusinesswas20.50%, andtherateforpersonalloanswas23.50%.Thankstothedelayedeffectoffinancialderegulation,easycreditatagloballevel, andtheRBA’sadoptionofatargetinginflationpolicy,theofficialinterestratedecreasedtoabout6%inthe1990sandbelow 5%intheearly2000s.GiventhedramaticnegativeimpactofinterestratesshowninTable4,lowinterestratesinthemost recenttwodecadesmusthavearemarkableroleintheriseofhouseholddebt
Second,anotherimportantfactoraffectinghouseholddebtnegativelyistheunemploymentrate.Thisisduetothenegative effectofunemploymentonhouseholdincome.Generallyspeaking,ahighunemploymentratemeansthereislessincomeforall householdsandthusagreaterdesireforloanstofinanceconsumption.Fromthispointofview,highunemploymentrateswill leadincreasedhouseholddebt.However,lowerincomeduetounemploymentcastsdoubtonthefutureincome,withtwo implications.Thefirstimplicationisthathouseholdswithoutregularemploymentwillbediscouragedfromborrowingbecause
ofconcernsabouttheabilitytorepaytheloan,whilethesecondimplicationisthatunemploymentincreasesthepossibilityof financialconstraints.Thesetwofactorsactuallylimithouseholddemandforfinancinganddepressthegrowthofhousehold debt.Moreimportantly,therisingunemploymentrateindicatesadeterioratedeconomicsituation,inwhichinvestorsarevery cautioustolend.Theestimationresultsshowthatthenegativeeffectdominates
ThisresultisquitereasonableconsideringthedominanceofthesupplysideintheAustralianhouseholddebtmarket.For variousreasonsthedemandforcreditisstronginAustralia Amongstotherthings,thesereasonsincludetheAustralian homeownershipdream,risinghousingprices,highrentalprices,andhousinggrantsprovidedbytheAustraliangovernment However,thisdemandisregulatedbythesupplyside,whichisprimarilydeterminedbycreditrisk.Anapparentsituation relatedtorisingunemploymentisthat,whenthelenderperceivesthattheborrowerisunlikelytopaybacktheloan,thehigh chanceofmakingalossprohibitshimfromprovidingcredit,andthusthesupplyofcreditwillshrinkdramatically.Thisis whathappenedduringtheGFC
Itisnoticeablefromthe estimatedresultsthat theunemployment rateisless influentialthantheinterest rate.This findingisconsistentwithpreviousstudies.Forexample,Debelle(2004,p.57)arguesthattheunemploymentrateisless severethaninterestratebecause‘unemploymentgenerallyaffectsonly arelativelysmallsectionofpopulation,andthe degree of overlap betweenthose households with a higherrisk of unemployment and those withhigh debt levelhas historicallybeenlow’
Third,thenumberofnewdwellingapprovalsalsohasasignificantinfluence.Theeffectsofnewdwellingsaretwo-fold
Ononehand,newdwellingsincreasethetotalhousingassetsinthemarket.Giventhehighdemandforhousingassetsand thepopularityofhousingmortgageloansinAustralia,thisimpliesmorehousingdebtforhouseholds.Ontheotherhand, newdwellingsenteringthemarketmeansgreaterhousingsupply.Ifthehousingdemandisunchanged,housingpriceswill drop.Decreasedhousingpriceswillreducethemarketvalueofhousingassetsandthusreducetheamountofhousingloans Theestimatedresultsshowthatthelattereffectdominatestheformer
Finally,CPIshowsaninsignificantnegativeeffectonhouseholddebt.Inflation(thepercentageincreaseintheCPI)has differenteffectsonborrowingandlending.Withregardtoborrowing,inflationwilldevaluethedebt,providingastrong stimulusforhouseholdstoborrow.However,onthesupplyside,inflationwillerodetheprincipalanddiscouragelending TheinsignificantnegativeeffectofCPIindicatesthatthesupplysidedominates.Thatis,inthefaceofhighinflation,fewer fundsarelent,andhouseholddebtwilldecrease.Thedominanteffectofthesupplyside inthehousehold debtmarket concurswithourearlieranalysisoftheeffectofunemploymentrate.Moreover,thisfindingisalsoconsistentwithprevious research.Forexample,manystudies(RBA,2003;Debelle,2004;andTheTreasury,2005)suggestthatlowinflationcouldbea reasonforrisinghouseholddebtbecauseitmaydecreasethefinancialconstraintsonhouseholds(lowerinflationleadsto lowerinterestratesandthuslessincomeisneededforthereducedscheduledpayment)andencourageslending(lower inflationerodesprincipalmoreslowly)
5.2 Factorspositivelyaffectinghouseholddebt
Withregardto thepositivefactors, the empiricalmodelshows that housingprices havea substantialinfluence on householddebt,whilebothGDPandthepopulationalsohaveaverysignificanteffect
First,thesignificantpositiveeffectofthehousingpriceindexcanbeunderstoodeasily,byconsideringtheimportanceof housing prices in housing assets and the importance of housing assets in Australian household investment portfolios Increasinghousingpriceswillscaleuphousingassetvalue.Fromthedemandperspective,thishastwoimplications.Fornew
Trang 9homebuyers,itmeanstheyhavetotakeonsubstantiallymoredebttobuyhousing,otherthingsbeingequal.Forthosewho havealreadytakenhousingloans,theincreasedhousingassetsprovidethemwithagoodopportunitytowithdrawhousing equity,toobtainmoreloansagainsttheincreasedvalueoftheirhousingassets.Fromthesupplyperspective,risinghousing assets pricesimplygoodeconomictimes.Thistendstoindicatelowcreditriskandthusstimulateslending.Asaresult, householddebtwillincreasealongwiththehousingprices
Thereareanumberoffactorscontributingtorisinghousingprices.Oneisfinancialderegulationandtargetedinflation policy,whichprovideanenvironmentinfavourofhousingdemand.Australia’sfinancialderegulationabolishedalargenumber
ofrestrictionsonlendingandborrowing,andthusfacilitateddebtfinancedhousingpurchases.Howevertosomedegree, over-simplifiedregulationofthemortgagemarketand/ornoregulationonsomemarkets(suchasthemarketsoffinancialbrokers andcomplexfinancialproducts)pusheshousingdemandtoanunsustainablelevel.Meanwhile,interestratesarelowdueto changesintheRBA’sinterestratepolicy,targetingtheinflationrate.SincehousingpricehasverylittleweightingintheCPI basket,thelowinflationofdailyconsumptionitemsmeanslowofficialinterestratesinAustraliaevenwherehousingprices skyrocket.Thelowborrowingcostandsimplifiedborrowingprocedurefacilitateandencouragehousingdemand
Another factor contributing to rising housing prices is the home ownership incentives provided by the Federal government.ToeasethefinancialburdenonhouseholdsaftertheintroductionoftheGSTin1999,thegovernmentoffered variousassociatedpolicies.Someofthesepolicieswereintendedtoreducethecostofhomeownership.Forexample,first homebuyerswereentitledtoagovernmentgrantofA$7000.Anumberofothercompensationpolicieswereenacted.Forthe owner-occupiedproperty,theimputedrentalincomeandcapitalgainsonsaleswereexemptoftax.Forinvestmenthousing, expensessuchasinterestpaymentanddepreciationaretaxdeductible.Finally,capitalgainsonsellinginvestmenthousing weretaxedathalfthetaxpayers’marginaltaxrate.Allthesepoliciestendedtostimulatehousingdemand
Thestategovernments’tightcontroloflanduseisalsoanimportantcontributorofrapidlyrisinghousingpricesbecauseit putarigidconstraintonhousingsupply.Duetothistightcontrol,thenumberofnewdwellingapprovalshasbarelychanged
inrecentdecades,fluctuatingataround38,000approvalsperquarterintheperiod1988–2012.Inotherwords,thehousing supplyinAustraliaisveryinelastic.Giventhetremendouslyincreaseddemandandaveryinelastichousingsupply,rapid increasesinhousingpriceandthusinhouseholddebtarelogicalconsequences
Second,thepositiveinfluenceofGDPonhouseholddebtmayariseintwoways.Ontheonehand,themagnitudeofGDP indicatesthesizeoftheeconomyandthusthecapacityofhouseholdborrowingandlending.AhigherGDPimplieshigher incomeforhouseholdsandhighergrossoperatingsurplusforfirms.Withhigherincome,householdswouldbeless credit-constrained.Ontheotherhand,thehigherincomeandprofitprovidericherfundingsourcesforbanks.Theotherchannel maycomefromhouseholdconfidence.ThegrowthrateofGDPisapopularindicatorofeconomicdevelopment.Robust growth of GDP makes people more confident so that they feelsafe to borrow and lend Withthe increased demand (willingnessandabilitytoborrow)andincreasedsupply(willingnesstolend),householddebtmaygrowinlinewithGDP Last,thepopulationintheeconomyalsohasasignificantpositiveaffectonhouseholddebt.Thereasoningbehindthe estimationresultsisthatthegrowthofpopulationislikelytoincreasethenumberofhouseholdswithdebtandthusthetotal householddebtlevel.Thisresultconcurswithpreviousstudies(e.g.Crook,2003;Thaicharoenetal.,2004;andTudela&
5.3 Effectsofstructuralchanges
Thestructuralchangein1993coincidedwiththepolicychangeoftheRBA.Thatis,followingthegradualderegulationof thefinancialsysteminthe1980s,theinflationtargetingpolicywasintroducedin1993.Theinsignificantpositiveestimation resultsindicatethatfinancialderegulationandRBAmonetarypolicychangemayhaveamildeffectofincreasingAustralian householddebt Practically,thismildpositiveeffectcanbeexplainedbythe easieraccesstofinancethankstofinancial deregulationandmoreconfidenceinthenewinterestrateregime
Themarkedlysignificantstructuralchangein2008isapparentlyinfluencedbytheGFC.Theestimationresultsshowthat thetrendofhouseholddebtdecreasesremarkablyafter2008Q4.Thenegativeeffectofthisstructuralchangeisconsistent withthenegativeeffectoftheGFContheeconomy.AstheGFChit,confidenceinbothconsumptionandinvestmentdeclined sharply.Facingan uncertaineconomicfuture,householdstend tosavemoreandborrowless, soit isreasonabletosee householddebtgrowingataslowerpace(althoughevenasthequarterlyhouseholdborrowingdecreases,thedebtlevel keepsrising asitis anaccumulatedvalue).However,while thenegative effectoftheGFConhousehold debthasbeen undoubted,cautionmustbetakenintheinterpretationoftheestimatedcoefficientofT*DM08Q4sinceweonlyhavetwo andahalfyearspost-GFCdata.Asthetimeseriesextendsinthefuture,thedegreeofsignificanceofT*DM08Q4maychange
5.4 Responsestostructuralshocks
TheresponsestostructuralshocksareshowninTable5.Forthepurposeofthisstudy,wediscusstheresultswithan emphasisontheresponseofhouseholddebttoshocksonothervariables(thefirstrowinTable5
SincetheshockofDEBT(uT
firstcolumnofthelong-runimpactmatrixiszero.Thefirstpermanentshock,uP
1;t,isidentifiedasproductivityshockorGDP shock.Thesecondpermanentshock,uP
2;t asthepartofthedisturbancesininterestratewhichisnotexplainedbyuP
1;t.The thirdpermanentshock,uP canbeidentifiedasthepartofdisturbancesinpopulationwhichisnotexplainedbyuP anduP ,
Trang 10andsoforth.TheresponseofDEBTtoashockonhousingpricesistremendouslypositive,thatisanA$13.08billionincreasein householddebtinthelong-runinresponsetoa1unitincreaseinhousingpriceindex.Ithasalsoasubstantialpositive responsetoashockonGDP.Positiveresponsesinthelong-runarealsofoundtoshocksonunemploymentrateandon population,thoughtheyaremuchmilder.Theresponsestoshocksonothervariablesarenegative.Theinterestrateshavethe largestnegativeinfluenceinthelong-runonhouseholddebt,followedbynewdwellingapprovalsandCPI
Thelarge responses ofhousehold debt toshocks on variables likeGDP, HPI, Rand NDWELLhave important policy implications.ThepositiveresponsetoanincreaseinGDPindicatesthatanincreaseinhouseholddebtisanaturalthing.In otherwords,oneshouldnotworrytoomuchaboutamoderateincreaseinhouseholddebtingoodeconomictimes.The notableinfluenceofhousingpricesonhouseholddebtshowstheimportanceofthehousingmarketonhouseholddebt.To reduce thepaceofrisinghouseholddebt,thegovernment shouldintervenein thehouseholdmarkettoreduce housing prices.Onewaytointerveneistoreducehousingdemandbysuspendinghomeownershipincentives,whileanotherwayis
toincreasehousingsupplybyeasingrestrictionsonlanduse.Thesubstantialnegativeinfluenceofinterestratessuggests thatmonetarypolicyisausefultooltocontainhouseholddebt.Anincreaseintheofficialinterestrate(ortighteningthe moneysupply)canreducetheamountofhouseholdborrowingsubstantially
6 Concludingremarks
TheempiricalresultspresentedinthispaperrevealthattherapidlygrowingAustralianhouseholddebtismainlythe result of a favourable macroeconomic environment, a booming housing market, and a rising population The robust economicdevelopmentinAustralia(growingGDPwithalowinterestrate,lowunemploymentandalowinflationrate) causesoptimisticexpectations.Whenhouseholdsaresufficientlyoptimistictoborrowandinvestorsareconfidenttolend, householddebtsurges.ThehousingmarkethasalsoplayedasignificantroleintherapidgrowthofAustralianhousehold debt.Australianhouseholdsareinfavourofhousinginvestment.Thehighdemandforhousingpushesuphousepricesand theexpectationofrisinghousingpricesinturnencouragesinvestmentdemandforhousing.Aboominghousingmarketleads
toahighlevelofhousingmortgagedebt.TheresultsalsodemonstratethattheGFChashadamarkedlynegativeimpacton household borrowing, while the financial market deregulation prior to 1993 and the RBA targeting inflation policy introducedin1993seemtohaveencouragedhouseholdborrowing
Basedonthefindingsofthispaper,theauthorssuggestthatitisnecessarytoreininrisinghouseholddebt.Todothis, thereareanumberofinstrumentsatthegovernment’sdisposal.First,properregulationandstandardisationoffinancial marketsisnecessary.Overlyunregulatedlendingandborrowingpracticeswillgiverisetounaffordablehomeloansand causehouseholddebtbubbles.Properlevelsofregulationcanminimiseirresponsibleandirrationalbehavioursinfinancial markets,andthusreducetheriskoffinancialbubbles.Second,theofficialinterestrateisaneffectivetoolbutmustbeusedin
atimelyand carefulway.Duetothe dramaticimpactof achange ininterestratesand thetimelagofthispolicy,this instrumentmustbeuseddecisivelywhenitiscertainthattheeconomyisheadinginanunwanteddirection.Furthermore,it
isbettertochangeinterestratesgradually.Dramaticchangesininterestrateswithinappropriatetimingmaycauselarge economicfluctuations.Moreover,thedecisiontochangeanofficialinterestrateshouldbebasedoncomprehensiveanalysis Manyfactorsotherthaninflationrateshavetobetakenintoaccount,includinghousingprices,theunemploymentrate andthegrowthofhouseholddebt.Finally,thegovernmentshouldmonitortheassetsmarketclosely(especiallythehousing marketinAustralia),andintervenewhennecessary.Newdwellingapprovalisausefultoolinmanaginghousingmarkets Sincenewhousingenteringthe marketcandamphousingprices, state governmentsinAustralia mustloosenlanduse restrictionsinthefaceofrapidrisinghousingpricesandhouseholddebt.Thevariousincentivesforhomeownership(suchas the firsthomegrant), andforinvestmenthousing,haveplayedasignificantrolein encouragingdemandforhousingin Australia.Thesekindsofstimulineedtobeabolishedorsuspendedwhenahousingbubbleisaroundthecorner
AppendixA Resultsofunitroottestswithstructuralchange.*
value (0.05)**
GDP with break at 08Q4 Coefficient 1009.017 0.020157 8.374419 6521.01 817.2433
t-Value 93 0.88 5.08 0.874563 1.154451 0.200283 3.35636 3.339856
t-Value 93 0.22 5.08 3.009933 3.152300 2.073827 2.25818 2.24019
t-Value 93 0.22 5.08 4.176358 4.415341 3.067263 0.12147 2.984417 Source: authors’ estimation.
* Model for test: Yt ¼ a0 þ a1 Yt1þ a2 t þ u1 DM þ u2 DT þ X k
i¼1
bi DYtiþet
DT ¼ t Tlift > Tl; and 0 otherwise:
* Lags are chosen in order to minimise AIC.