PublicDisclosureAuthorizedPublicDisclosureAuthorized

PolicyResearchWorkingPaper10945

IdentifyingGrowthAccelerations

BramGootjes

JakobdeHaan

KerstenStamm

ShuYu

WORLDBANKGROUP

DevelopmentEconomicsProspectsGroup

October2024

PolicyResearchWorkingPaper10945

Abstract

Thispaperintroducesanewmethodtoidentifyoutput growthaccelerationsthatintegrateselementsofboththe“criteria-based”and“break-testing”approaches,whichare prevalentintheliterature.Theproposedcriteriadonot imposeafixedlengthongrowthaccelerations,thusenabling durationanalyseswithoutrelyingonquestionablestatistical techniquesfortheidentificationoftheseaccelerations.The

findingsshowthatgrowthaccelerationslastanaverage13.4years,albeitwithsignificantvariationsindurationacrossregions.Initialconditionsandcontemporaneousdomesticandexternaleconomicconditionsallmatterforthecontin-uationofanacceleration,andchangesinanysinglepolicyconditionhavelessofanimpact.

ThispaperisaproductoftheProspectsGroup,DevelopmentEconomics.ItispartofalargereffortbytheWorldBanktoprovideopenaccesstoitsresearchandmakeacontributiontodevelopmentpolicydiscussionsaroundtheworld.PolicyResearchWorkingPapersarealsopostedontheWebat

/prwp.Theauthorsmaybe

contactedatbgootjes@;jakob.de.haan@rug.nl;kstamm@;andsyu2@.

ThePolicyResearchWorkingPaperSeriesdisseminatesthefindingsofworkinprogresstoencouragetheexchangeofideasaboutdevelopmentissues.Anobjectiveoftheseriesistogetthefindingsoutquickly,evenifthepresentationsarelessthanfullypolished.Thepaperscarrythenamesoftheauthorsandshouldbecitedaccordingly.Thefindings,interpretations,andconclusionsexpressedinthispaperareentirelythoseoftheauthors.TheydonotnecessarilyrepresenttheviewsoftheInternationalBankforReconstructionandDevelopment/WorldBankanditsaffiliatedorganizations,orthoseoftheExecutiveDirectorsoftheWorldBankorthegovernmentstheyrepresent.

ProducedbytheResearchSupportTeam

IdentifyingGrowthAccelerations

BramGootjes,JakobdeHaan,KerstenStamm,andShuY

u*

Keywords:outputgrowthaccelerations;identification;outputgrowthvolatility

JELcodes:O11;O47;O57

Declarationofinterests:none.

*BramGootjes:WorldBank,ProspectsGroup,bgootjes@;JakobdeHaan:UniversityofGroningenandCESifo,jakob.de.haan@rug.nl;KerstenStamm:WorldBank,ProspectsGroup,kstamm@;ShuYu:WorldBank,EFIChiefEconomistOffice,syu2@.Theviewsexpressedherearetheauthors’,anddonotreflecttheofficialviewsoftheWorldBank,itsExecutiveDirectors,orthecountriestheyrepresent.WearethankfultoRafaelaMartinhoHenriquesforherexceptionalresearchassistance,andtoAntonioFatás,AyhanKose,JohnBaffes,andallparticipantsoftheWorldBank’sProspectGroupseminarsfortheirvaluablecommentsandsuggestions.

1.Introduction

Promotingsustainedeconomicgrowthisprobablythemostimportantgoalofeconomicpolicy.However,studyinggrowthisacomplexandoftencontentiousissue,wherescholarshaveusedvariouseconometrictechniquestouncoverhowcountriesachieveandmaintainrobustgrowth.Traditionally,scholarshaveanalyzedlong-termoutputtrendstounderstandthefactorsbehinddifferinggrowthpatternsacrosscountries(see,forexample,BarroandSala-i-Martin,1991;Mankiwetal.,1992)

.1

Focusingonlong-termgrowthaverages,however,assumesastablelinearrelationshipbetweengrowthanditsfundamentalssuitableforallcountriesinalleconomicconditions.Thisassumptionfaltersgiventherarityofcountriesmaintainingconstantgrowthratesoverlongperiods(Easterlyetal.,1993;Ben-DavidandPapel,1998;Ayaretal.,2018).Instead,growthpatternsaretypicallyvolatile,withcountriesexperiencingphasesofprogress,stagnation,andsetbacks(Pritchett,2000).

Recognizingtheinadequacyoftraditionallong-termtrendanalysisowingtothevolatilenatureofgrowthrates,researchershaveshiftedtowardsinvestigatingspecificepisodesofrapidandsustainedgrowth,alsoknownasgrowthaccelerationsorgrowthspells(Hausmannetal.,2005;TimmeranddeVries,2009).Thesignificanceofstudyinggrowthaccelerationscanbemotivatedfromvariouseconomicperspectives.Forinstance,countriesmightundergoarapidupsurgeinoutputgrowthonlytoreverttotheirpriorgrowthtrajectories,oftenduetotransientshocksthattemporarilybolstergrowthperformance(asperneoclassicalgrowththeory).Conversely,somecountriesmaytransitiontowardsapermanentlyhighergrowthpath,drivenbyenhancedeconomicpolicies,forexample(asviewedfromanendogenousgrowthmodelperspective).Regardlessofthetheoreticalframeworkused,episodesofacceleratedgrowthreflectthemostinterestingvariationingrowthdata,whichwouldbeobscuredwhenconsideringlong-termaverages.Bylinkingthetimingoftheseepisodestodrivingforces,scholarscangaindeeperinsightsintothecausalmechanismsbehindvariationsingrowthperformance.

Hausmannetal.(2005)laidthegroundworkfortheempiricalexaminationofgrowthaccelerations.Toidentifysuchaccelerations,theyproposeasetofeconomiccriteriathatfilteroutyearscharacterizedbybothahighlevelofgrowthandasignificantsurgeofgrowth.Forinstance,averagegrowthisrequiredtoexceed3.5percentperannumoveraperiodofeightyearsandtosurpassthegrowthrateoftheprecedingeightyearsbyatleasttwopercentagepoints.These

1Onestrandofliterature,followingseminalstudiesbyBarroandSala-i-Martin(1991)andMankiwetal.(1992),examinedthedeterminantsofcross-countrydifferencesinaveragelong-termeconomicgrowthrates.Anotherstrand,pioneeredbyIslam(1995)andCasellietal.(1996),useddynamicpanelsinstead,arrangingcountry-leveldataintofive-yearaveragesorotherintervals.

criteriahavegarneredwidespreadadoptioninempiricalstudiesasausefulbenchmarkforidentifyinggrowthaccelerations(cf.,ImamandSalinas,2008;TimmeranddeVries,2009;Eichengreenetal.,2012;Diaoetal.,2019;Grussetal.,2020;Avometal.,2021;KoopmanandWacker,2023).Althoughtheresultsinthisliteraturehavebeenmixed,acommonfindingisthataccelerationsarepredominantlyexplainedbystrongerinstitutionsandidiosyncraticfactors.

Toexpeditethecatching-upprocesstomoreadvancedeconomies,developingcountriesrequirenotonlyrobusteconomicgrowth,butalsosustainedgrowthoveranextendedperiod.Addressingthisissue,Bergetal.(2012)investigatethefactorsthatsustaingrowthaccelerationsbyusingstructuralbreaktestinginoutputgrowthseriestoidentifythestartofgrowthaccelerations(followingBaiandPerron,1998,2003)alongsideadhoccriteriatoidentify‘desirable’spells.Theirresultsindicatethatwhileregionsdonotdiffersignificantlyinthefrequencyofgrowthaccelerations,theseepisodestendtobeshorterinAfricanandLatinAmericancountriescomparedtothelongerdurationsobservedinindustrializedcountriesandemergingAsia.

ThefindingsofBergetal.(2012)questiononeoftheimplicitassumptionsinthecriteriaintroducedbyHausmannetal.(2005):thatgrowthaccelerationshaveafixeddurationofeightyears.Additionally,theuseofstructuralbreaktestingtoidentifytheonsetofgrowthaccelerationssuggeststhatcountry-specificgrowthcharacteristicsshouldbeconsideredratherthanapplyinga‘one-size-fits-all’approach.Despitetheseinsights,themethodproposedbyBergetal.(2012)foridentifyinggrowthaccelerationshasnotgainedasmuchtractionasthecanonicalfilterofHausmannetal.(2005).OnepossiblereasonisthelowpoweroftheBai–Perrontest,whichmayleadtotherejectionoftruebreaksintheunderlyingGDPpercapitaseries(Karetal.,2013).

However,thisalsoholdsfortheapproachofHausmannetal.(2005).Forinstance,astheircriteriaarebasedonperiodicaverages,thefilterrisksidentifyingperiodswithsporadichighgrowthinterspersedwithlowornegativegrowth,contradictingtheaimofidentifyingperiodswithsustainedhighoutputgrowth.Thisissuebecomesmorepronouncedwhenepisodesstartamidstnegativegrowthyears(Jong-A-PinanddeHaan,2011)

.2

Moreover,theuseofadhoccriteriamayleadtoerrorsinassessinggrowthperformanceduetovariationsinoutputgrowthvolatility,potentiallyresultinginunder-identificationofgrowthaccelerationsinlessvolatileeconomiesandover-identificationinmorevolatileones

.3

2Jong-A-PinanddeHaan(2011)showedthattheapproachofHausmannetal.(2005)occasionallyproduceslessplausiblestartingyearsofoutputaccelerationswherethegrowthrateisnegativeorlimited.Todealwiththis,theyintroduceanadditionalcriterion,requiringthateconomicgrowthinthefirstyearofthegrowthaccelerationmustbehigherthanintheprecedingyear.

3Theliteratureonidentifyingfiscaladjustments,aspioneeredbyAlesinaandPerotti(1995),suffersfromthesameproblemaspointedoutbyWieseetal.(2018).A‘one-size-fits-all’approachislikelytoidentifymorefiscaladjustments

Toexaminetheissueswiththeuseofuniformeconomiccriteriatoidentifygrowthaccelerationsinmoredetail,

Figure1

showsrealGDPpercapitaforIndiaandZimbabwesincethe1950s.Indiaisoneoftheworld’sfastest-growingeconomiesinrecentdecades.Instarkcontrast,Zimbabwe’seconomyhasbeenfraughtwithchallenges,includinghyperinflation,currencycrises,debtdistress,andpoliticalturmoil.By2023,India’srealGDPpercapitawasovertwoandahalftimesthatofZimbabwe($7,379versus$2,811),eventhoughbothcountrieshadcomparablepercapitaincomesintheearly1990s.Clearly,India’sgrowthexperiencehassignificantlyoutperformedthatofZimbabweovertheperiodunderconsideration.

Tounderstandeconomicgrowth,itiscrucialtoexplorewhatfactorssignificantlycontributedtoIndia’ssuccesscomparedtowhathinderedZimbabwe’sprogress.However,thecriteriacommonlyusedintheliteraturetoidentifygrowthaccelerationsfallshortinthiscontext.DespiteIndia’simpressivegrowthperformancesincetheearly1990s,thefilterproposedbyHausmannetal.(2005).Incontrast,thefilteridentifiestwogrowthaccelerationsforZimbabwe,startingin1968and2008.AlthoughtheseperiodssawsharpincreasesinGDPpercapita,thesesurgesturnedouttobeshort-livedandwereprecededbysignificanteconomicdownturns.Asaresult,realGDPpercapitadidnotshowsubstantialincreasesrelativetoearlierperiods.Moreso,thesefindingsreflecttheeconomicvolatilityexperiencedbyZimbabweratherthangenuinegrowthacceleration.

Bothmethodsforidentifyinggrowthaccelerations(filtersandbreakpoints)haveclearbenefitsbutalsolimitations.Toovercometheseissues,weintroduceanovelfilteringmethodthatbuildsuponpreviousconcepts.Ourapproachcombineselementsofthe‘criteria-based’approachseeninHausmannetal.(2005)butleveragescountry-specificgrowthcharacteristics,akintotheunderlyingprincipleofthe‘break-testing’approachoutlinedbyBergetal.(2012).Specifically,weproposeafilterbasedontheweightedaverageofthelong-termtrendandvolatilityofacountry’soutputgrowth,capturinguniquegrowthpatternstoidentifygrowthaccelerations.Furthermore,ourcriteriadonotimposeafixedlengthoftheacceleration,allowingforanalysesofthedurationofgrowthaccelerations.Byintegratingtheseelementsintoastraightforwardfilter,wesidesteptheneedforcomplexandpotentiallyunreliablestatisticaltechniquestoidentifygrowthaccelerations.

Basedonasampleof181countriesbetween1951and2023,weidentified134growthaccelerationsacross110countries.Duringtheseperiods,realGDPpercapitagrowthaverages5.9percentperannum,morethansixtimeshigherthaninotheryears.Theaveragedurationof

forcountrieswithvolatilebudgetoutcomes,simplybecausethechangeofthebudgetbalanceisthekeycriteriontoidentifyafiscaladjustment.Thisisaso-calledtypeIerror.Bythesametoken,thesefiltersarelesslikelytodetectsignificantchangesinfiscalpolicyincountrieswherethebudgetaryprocessleadstolessvolatilepolicyoutcomes.Inthatcase,thesefilterssufferfromtypeIIerrors.

growthaccelerationsis13.4years,significantlylongerthanthecommonlyassumedeight-yeardurationintheliterature.Moreover,thedurationofgrowthaccelerationsvariessignificantlyacrossregionsandcountrygroups,consistentwiththefindingsofBergetal.(2012).

Arangeofstatisticsdemonstratesthebenefitsofourmethodtoidentifygrowthaccelerations.Forinstance,theepisodesarecharacterizedbyasignificantupswinginmedianrealGDPpercapitagrowthatthebeginningoftheidentifiedgrowthaccelerationsandacleardipattheend.Thispatternisnotclearinthegrowthaccelerationsidentifiedusingothermethods.Throughaseriesofcountrycasestudies,wefindthatthemethodweproposealignscloselywithanecdotalevidenceinaccelerationpatterns.Forinstance,inthecaseofIndia,weidentifyaprolongedaccelerationfrom2003to2019.Weshowthatthemethodisrobusttotheuseofalternativeweights,variousminimum-lengthrequirements,rollingwindows,andreducedsamplingtocalculatethethresholds.

Tofurtherillustratetheadvantagesofourproposedmethod,weconductaseriesofsurvivalanalysestostudythefactorsthatarerelatedtosustaininggrowthaccelerations.Weshowthatinitialconditions,aswellasdomesticandexternaleconomicconditions,playasignificantroleinthepersistenceofanacceleration.Ontheotherhand,changesinpolicyconditionshavelessimpact.Thesefindingssuggestthatmaintainingastablemacroeconomicenvironmenthelpstoprolonggrowthaccelerations.Theseresultsarerobusttoemployingalternativepolicymeasures,includingfrailty,andusingalternativesamples.

Therestofthepaperisorganizedasfollows.Section2providesdetailsontheproposednewfilterandpresentstheoutcomes.Sections3and4comparetheoutcomesofthenewfilterwiththoseofpreviousstudies,examineitsrobustnesstoalternativeparameters,anddiscussfeaturesoftheidentifiedgrowthaccelerations.Section5presentsseveralcountrycasestudiestodemonstratetheadvantagesofusingthenewidentificationapproach.Section6analyzeswhichfactorssustainagrowthacceleration.Section7concludes.

2.Identifyinggrowthaccelerations

2.1Thenewmethod

Weproposeidentifyinggrowthaccelerationsbuildinguponpreviousfilters.Ourmethodisdesignedtopinpointperiodsofsustainedgrowth,effectivelymitigatingtheinfluenceofsuddenspikesinrealGDPpercapitagrowthratesbyincorporatingmeasuresofvolatility.

Thefilteringtechniquestartswithcalculatingcountry-specificmetrics(denotedasθiforcountryi)basedonthemean(μ)andstandarddeviation(σ)ofacountry’srealGDPpercapitagrowthoverthefullsampleperiod

.4

Aswillbeexplainedbelow,thismetricisusedtoimposerequirementsonrealper-capitaoutputgrowthtoidentifyaccelerationepisodes.Themean(l)controlsforlong-termtrendsinrealGDPpercapitagrowth,whilethestandarddeviation(σGDPPCi)capturesitsvolatility:

θi(μ,σ)=θ(l)+(1?θ)(σGDPPCi)(1)

where0≤θ≤1.Fornow,weassignequalweights(θ=0.5)tothelong-runtrendandvolatilityofrealGDPpercapitagrowth(referredtoasthe“baseline”inthefollowingsections)

.5

Tocalculatethemeanandstandarddeviationofpercapitaoutputgrowth,weexcludetheminimumandmaximumobservationsforeachcountrytopreventthesemeasuresfrombeingskewedbyoutliers.

Figure2

presentstheaveragecountry-specificmetricsacrossregions.Weobservethatthemeasuresdifferconsiderablybothacrossandwithinregions.Inemergingmarketsanddevelopingeconomies(EMDEs)acrossLatinAmericaandtheCaribbean(LAC),Sub-SaharanAfrica(SSA),theMiddleEastandNorthAfrica(MNA),andEuropeandCentralAsia(ECA),metricsareprimarilydeterminedbythestandarddeviationofgrowth,reflectingthehighereconomicvolatilityprevalentintheseregions.Inadvancedeconomies(AEs),aswellasintheSouthAsiaregion(SAR)andEastAsiaandthePacific(EAP),metricsaremorestronglydrivenbylong-termaveragegrowthrates,capturingtheirgenerallystrongereconomicperformance.Overall,lowermetricsaregenerallyobservedinLAC,wheretheaveragemeasurestandsat2.9,whilehighermeasuresarefoundinECA,averaging4.3.

4FollowingHausmannetal.(2005),realGDPpercapitagrowthisusedtoavoidtheresultsbeingdrivenbypopulationgrowth.

5InSection4,wedemonstratehowadjustingtheseweightsimpactsthenumberofgrowthaccelerationsthatareidentified.

Afterhavingcalculatedthemetricforeachcountry,weidentifyaccelerationepisodesusingthefollowingcriteria:

?Startandendyears.AnepisodestartswhenrealGDPpercapitagrowthexceedsthecountry-specificmetricandendswhenitfallsbelowthismetric.

?Duration.TheidentifiedaccelerationmustencompassatleasteightyearsofrealGDPpercapitagrowthabovetheestablishedmetric

.6

?Adjustingfortemporarydips.ShouldrealGDPpercapitagrowthfallbelowthecountry-specificmetric,theepisodeisstillconsideredongoingifaveragegrowthduringtheperiodsbeforeandafterthedip,alongwiththeyearwhenittemporarilyfallsbelowthisthreshold,staysabovethecountry-specificmetric.

?Excludingcyclicalrebounds.EpisodesareexcludedifthelevelofrealGDPpercapitaattheendoftheepisodeislowerthaninanyyearbeforethestartoftheepisode.

Thesecriteriaarenecessarytoensurethattheidentifiedepisodesaresustainedandthattheycaptureacceleratedrealpercapitaoutputgrowth.Whilethefirstcriterionallowstheepisodelengthtovary,thesecondonemakessurethattheidentifiedepisodesaresustainedandlessinfluencedbycyclicalmovements.ThethirdcriterioniscrucialtoavoidtheprematureendingofagrowthaccelerationwhenrealGDPpercapitagrowthtemporarilydipswhilethetrendisstillstrong.Forexample,innearlyhalfofthecaseswithatemporarydip,realGDPpercapitagrowthdoesnotdeviatemorethan25percentfromthecalculatedthresholdintheyearitfallsbelowthethreshold

.7

Finally,thelastcriterionisneededtominimizetheinclusionofpurecyclicalrebounds(ChristianoandFitzgerald,2003;BarroandSala-i-Martin,1992).

2.2Theoutcomes

WeexaminerealGDPpercapitagrowthacrossanunbalancedpanelof181countriesfrom1951to2023toidentifyinstancesofgrowthaccelerations.WeusedataonrealGDPpercapita(in2017USD)providedbythePennWorldTable(PWT)10.1databaseandupdateitto2023using

6Aperiodofeightyearsiscommonlyusedinthegrowthaccelerationliterature.However,whereaspreviousstudiesapplythisnumberasthemaximum,weconsideritasaminimumrequirement.InSection4,wedemonstratethatalteringthisminimum-yearrequirementonlyimpactsthenumberofidentifiedaccelerations.

7Inaquarterofthecases,realGDPpercapitagrowthisnegativewhenitdipsbelowthethreshold.Onemayarguethatthiscriterionispickingupeconomicrecoveryinsteadofacontinuedaccelerationinthesecases.However,

conceptually,thisisveryunlikely.Whilerecoveryisprobable,forthesubsequentperiodtoqualifyaspartofa

growthacceleration,averagegrowthduringthisperiod—includingthedipyear—mustexceedthecalculated

benchmarksplustherecovery.Inotherwords,thepost-crisisgrowthmustbesubstantialenoughtoensurethatweareobservinggenuinecontinuationoftheaccelerationratherthanmerelyarecovery.

theIMFWorldEconomicOutlook(WEO)Dataset.Toqualifyasagrowthacceleration,werequireaminimumdurationofeightyears,making2016thelatestpossibleyearforwhichwecanidentifytheinitiationofagrowthacceleration.

Column(1)of

Table1

showsthedescriptivestatisticsofthegrowthaccelerationepisodesweidentifybasedonthecountry-specificthresholdsdeterminedinthebaselinescenario

.8

Overthesampleperiod,wefind134accelerationepisodesacross110countries.Duringtheaccelerations,realGDPpercapitagrowthaveragesanimpressive5.9percentperyear,morethansixtimeshigherthanaveragegrowthinotheryears.Mostofthegrowthduringtheaccelerationepisodesisnotprimarilyattributabletothebeginningorendoftheseepisodes,asevidencedbythecomparablerealGDPpercapitagrowthratesobservedinthefirstandlastyearofthesegrowthaccelerations.Intotal,growthaccelerationsareobservedin18.1percentofthecountry-yearobservations.ThisisinlinewiththeoutcomesofHausmannetal.(2005),conveyingthatexperiencingagrowthaccelerationisnotuncommon,anditisnotlimitedtojustafewfortunatecountries.Infact,somecountriesevenexperiencedmorethanoneaccelerationwithinthetimeframeconsidered.

Auniquefeatureofourproposedfilteristhatitdoesnotimposeafixeddurationforgrowthaccelerations.Asshownin

Table2,

theaveragedurationofgrowthaccelerationsis13.4years,whichissignificantlylongerthantheeight-yeardurationcommonlyassumedintheliterature.Outofthe134accelerationsweidentified,98lastedtenyearsorlonger.Aconsiderablenumberevenspannedfifteenyearsormore,suchasthoseintheRepublicofKoreaandTaiwanbetweenthe1960sandthe1990s,inseveralWesternEuropeancountriesandJapanbetweenthe1950sandthe1970s,andmorerecentlyinBangladeshandVietNam.

Weidentified39growthaccelerationsinAEsand95inEMDEs.However,whenconsideringthenumberofcountriesforwhichwehavedata,AEssignificantlyoutperformEMDEsintermsofexperiencinggrowthaccelerations.Furthermore,growthaccelerationsinAEstypicallylastedlonger,averaging14.7years,comparedto12.8yearsinEMDEs.

Relativelyspeaking,moregrowthaccelerationsoccurredinECA,LAC,andSAR,whilefewerwereobservedinMNAandSSA.InEAPandSAR,growthaccelerationslastedsignificantlylongercomparedtootherEMDEregions.Thesefindingsareconsistentwiththeresultsobtainedwhenwedividethesampleintostableandvolatileeconomies,withmoreandlongergrowthaccelerationsexperiencedbystableeconomiescomparedtovolatileones.Theresultsforcommodityexporters,

8

TableA1

intheAppendixprovidesalistofalltheidentifiedgrowthaccelerations.

smallstates,andfragileandconflict-affectedsituations—groupsofcountriestypicallycharacterizedbygreatereconomicvolatility—furthersupporttheseconclusions

.9

Insum,theresultspresentedin

Table2

challengetwoassumptionsoftenusedintheliterature:theimpositionofafixeddurationforgrowthaccelerationsandtheuseofadhocstandardstoidentifygrowthaccelerationsacrosscountrieswithdiverseeconomiccharacteristics.Inthefollowingsection,weexaminetheimplicationsofthesefindingsfortheaccurateidentificationofgrowthaccelerations.WecomparetheoutcomesofourapproachwiththosederivedfromthefiltersusedbyHausmannetal.(2005)andJong-A-PinanddeHaan(2011).

3.Comparisonwithotherfilters

3.1Growthduringandaroundtheaccelerationepisodes

Column(2)of

Table1

presentsthegrowthaccelerationsidentifiedusingthefilteringmethodsuggestedbyHausmannetal.(2005),whilecolumn(3)showstheresultswhentheadditionalcriterionproposedbyJong-A-PinanddeHaan(2011)isapplied.TheepisodesidentifiedwithourproposedfilterexhibitmodestlyhigheraveragerealGDPpercapitagrowthratesandlowervolatilityduringaccelerationscomparedtothoseidentifiedwiththeotherapproaches.Thissuggeststhatthemethodweproposeislesspronetoidentifyingaccelerationsdrivenbyshort-termgrowthspikes.

Table1

alsohighlightsmoregeneraldifferencesbetweenthegrowthaccelerationsidentifiedbythedifferentfilters.Forinstance,ourproposedmethodyieldsasmallernumberofgrowthaccelerationsthantheotherapproaches.However,sincemostoftheepisodesidentifiedwithourfilterlastedlongerthaneightyears,ahighershareofobservationsareclassifiedasaccelerationsusingourmethod.Additionally,

Table1

showsthataveragerealGDPpercapitagrowthratesintheepisodesidentifiedwithourproposedfilterareconsiderablyhigherinthefirstyearcomparedtotheothermethods.Moreover,growthinthefinalyearoftheaccelerationismarginallybelowaveragegrowthobservedduringtheepisodes,whereasfortheotherfilters,thelastyearoftheidentifiedepisodesisalmostonepercentagepointabovetheaverage.Overall,thissuggeststhatourapproachmaybemoreeffectiveinidentifyingthestartandendofanaccelerationepisode.

Figure3

furthersupportsthisclaim.WhiletheepisodesidentifiedbyourfilterarecharacterizedbyasignificantupswinginmedianrealGDPpercapitagrowthatthebeginning(i.e.,yeart)andasignificantdipattheend(i.e.,yearT)ofgrowthaccelerationsidentified,thispatternisnot

9Wealsoinvestigatedwhetherthedurationsofgrowthaccelerationsvariedacrossdifferentdecades.However,nocleartrendemergedfromtheanalysis.Section3.2providesamoredetailedexaminationofgrowthaccelerationtrends.

clearlyobservedusingtheothermethods

.10

Notably,growthoftenrisesmoderatelyinthefirstyearoftheidentifiedaccelerationswhenusingtheothermethods,particularlyinthecaseoftheHausmannetal.(2005)criteria.Moreover,med