Business Cycles: Theory, History, Indicators, and …

1 This PDF is a selection from an out-of-print volume from the National Bureauof Economic ResearchVolume Title: Business Cycles: theory , history , Indicators, and ForecastingVolume Author/Editor: Victor ZarnowitzVolume Publisher: University of Chicago PressVolume ISBN: 0-226-97890-7 Volume URL: Date: n/aPublication Date: January 1992 Chapter Title: Composite Indexes of Leading, Coincident, and Lagging IndicatorsChapter Author: Victor ZarnowitzChapter URL: pages in book: (p. 316 - 356)11 CompositeIndexesofLeading,Coincident, ,Standards, , ,therearea numberofplausibleandnotmutuallyexclusive hypothesesaboutwhat cancausedownturnsandcontractions, , ,therearea 'sstructureandinstitutionsandthegovernme nt' hasprovedsodifficulttomakeprogresstoward a unifiedtheoryofbusinesscycles, ,forthesamereasons, ,someleadingindicatorstumouttobemostoper ativeandusefulinonesetofconditions,andot hersina ,itisthereforeadvisabletorelyona basedinpartonZamowitzandBoschan1975a, ,Coincident, ,themeasurementerrorsinindividualindicat orsareoftenlarge, ,theriskofbeingmisledcanbereducedbyevalu atingthesignals,notfromanyoneseriesviewe dinisolation,butfroma ,however, "noise.

2 "Ingeneral,indicatorstendtoreactnotonlyt osustainedcyclicalfluctuationsbutalsotof requentdisturbancesofallkinds,forexample , ,themonth-to-monthchangesintheseseries(a ftereliminationofanyseasonalelements) ,someofthatnoiseis eliminated;thatis,a withcommontimingpatternsis thata ,sucha failuremerelyimpairsand,ifre-peated, ,theproblemreducestogettinga betterrepresentationforit ,evena failureona singleoccasionwouldhavestrongnegative implicationsfortheindicatorapproachifit extendedtoa , theroleinbusinesscyclesofthevariablesrep resentedbythedata?(Thejudgmentonthisisqu antifiedinthescoreforeconomicsignificanc e.) (statisticaladequacy)? (orcoincidedorlagged)atbusinesscyclepeak sandtroughs(timingatrecessionsandrevival s)? (conformitytohistoricalbusinesscycles)? cyclicaltumintheseriesbedistinguishedfro mdirectionalchangeassociatedwithshorter, irregularmovements(smoothness,whichisinv erselyrelatedtothedegreeofstatisticalnoi se)?

3 (currencyortimeliness)?Aformal, (1975a,1975b,1975c),a ;hencemuchofit washandledbypreselection,withtheminimuma ccept-ablescoresetat70%. ,thequalityofthereportingsystemisassesse daccordingtowhetherit is setupdirectlyforstatisticalpurposes,isa by-productofanadministrativesystem,orisn onexistent(asforseriesesti-matedindirect lyfromrelatedvariables).Otheraspectsofst atisticaladequacyincludethecoverageofpro cess(fullenumeration,probabilitysample,o ther)andoftimeperiod(fullmonth,or1 week,or1 daypermonth,etc.);theavailabilityofestim atesofsamplingandreportingerrors;theleng thoftheseriesandcomparabilityovertime(br eaksarepenalized);andthefrequencyofrevis ions(none,oncea reportingperiod,ormoreoften).Morerecentl y,therevisionsweregivena moreelaboratetreatmentanda separate smallerweight(ofabout20%ofthetotaltiming score). ,thefirstfourcriteriareceivedweightsof20 %each;thelasttworeceivedweightsof10% (asfarbackasthedatawereavailablebutwitha heavypreponderanceoftheevidencefromthepo st-WorldWarII years).

4 ,criteria1,2,and4 receivedweightsofY6each,timingo/1s,smoot hness Is, , ,andZamowitzandBoschan1975a, ,errorsandrevisions,smoothnessandcurrenc y, ,ontheseandtheotherscoresforthecom-ponen tsofthe1966and1975indexes, ,Coincident, (expansionsandcontrac-tions) "extra"movementsintheindicatorsthatdonot matchthephasesofgeneralbusinessfluctuati onsandcanresultinmisleading"falsesignals ."Theamplitudesofcyclicalchangesintheser iesarealsoaccountedforhere(thelargerandm oredistinctmovementsscorehigher). ,trailingmovingav-eragescanbeadvantageou sforsomeerraticseriesthathavelongleadsan dhighcurrencyscores( ,arecompiledfrequentlyandreleasedpromptl y).Toputit differently,trade-offrelationshipsexistb etweenthesmoothness,cur-rency,andtimingc haracteristics( ,movingaveragesincreasesmoothnessbutredu cecurrencyandpossiblytheleadtimes). ( )andbyeconomicprocess( ).ListedinpartAarealsothescoresofthecomp ositeindexesofleading,coincident, ( ) is alsotobenotedthatthedifferencesbetweenth ecomponentscoresaregenerallylargeandofte nsignificantbutthattheyoffseteachotherto a ,theleadersasa grouprankwellbelowthecoinciderswithrespe cttosmoothness,butthereverseobtainsforre visions(seelines1-2 and5-6).

5 Productionandincomeseries(lines20-21) ,probabilitygetsa maximumof50,extraturns30, ,1947-80 StatisticalEconomicTimingConformitySmoot hnessCurrencyAdequacyRevisionsSignifican ceTotalLine(1)(2)(3)(4)(5)(6)(7)(8) (108)AllLeadingSeries(47)176685757787274 71211182729122276 ComponentsoftheLeadingIndex(12)379726880 7565747441016201083095 AllRoughlyCoincidentSeries(18)5867787578 052827861871528818118 ComponentsoftheCoincidentIndex(4)7958295 7477408882854101351656 AllLaggingSeries(26)98172856275827877101 6151732132056 ComponentsoftheLaggingIndex(6)1186718775 73708078126181618152468 Unclassified(17)137577787274757675141816 2226141888 CompositeIndexes(3) (l08) (15) (10) ,Trade,Orders,andDeliveries(13) (18) (9)267764715077647770271517381191756321 CompositeIndexesofLeading,Coincident, (1)(2)(3)(4)(5)(6)(7)(8) ,Costs,andProfits(17) (26)3077686570828278753114193330171957 ,BureauofEconomicAnalysis1984,table7, :Ineachsection,exceptCompositeIndexes(li nes15-17),entriesinthefirstlinearemeanso fthescoresoftheindividualseries, (L),coincident(C),andlagging(Lg) (whichisheretypicallycoincident)butlowon currencyandrevisions(thedatafromGNPaccou ntsarequarterlyandsubjecttoseveralandoft enlargealterations).

6 Theseriesrelatingtoconsumptionandtradedo worseontiming,conformity,andsmoothnessbu tbetteroncurrency,statisticaladequacy,an drevisions(lines22-23).Insum, ,too, ,coincident, ,highconsistencyofprocyclicalorcountercy clicalbehavior,thatis,highconformityorco herence(correlationwithbusinesscycles,al lowingforanysystematicleadsorlags),isa , , beginningwiththe1975list, , ,ofcourse,thecentralmeasureoftotaloutput , (notablyMitchell1913,1927)ascribea majorroletoprofits;also,thereis evidenceofa ,conformity,and,partic-ularly, ,thisispartlypaidforbylowerscoresonrevis ionsand(toa lesserextent) ,theoveralladvantageoftheindexcomponents eriesismodest( and3).Similarstatementscanbemadeaboutthe coincidentandlag-gingseries( and7,and9 and11).Notethatlargerdifferenceswouldbes hownhadtheindexcomponentsbeenexcludedfro mthe"allseries" ,conformity,andcurrencybutmuchlargeryetf orsmoothness, ,untilveryrecently,theindexesearnedzeroo rverylowscoresforrevisions, ,sothata monthlater,whenthelaggingdatafirstappear ed, ,theadvantageselsewherewereagainlargelyd issipatedhere( and15,7 and16,and11and17).

7 Thetwotardycomponentswereeliminatedfromt heDepartmentofCommerceleadingin-dexin198 9withtheintentionofradicallyreducingthes izeoftherevisionsinthatindex(Hertzbergan dBeckman1989;fordetail, ).Clearly,moreisrequiredofa goodindexthanofa , , , ,theimplementationofthisprincipleinvolve dsomecostsintermsoflowerscoresthanwouldb eavailableotherwise( ,fromcollectionsofserieslessdiversifieda nd/orlessaggregative).Theexistingindexes aredesignedtolead(orcoincideorlag) restrictivesinceit excludesserieswithmixedtimingpatterns,wh ichprevailinsomeareasforreasonsthatarewe llunderstood(seechapter10, ).Suchseries( ,thosethatleadatpeaksandlagattroughsorvi ceversa)areincludedinthe"unclassified" (lines13-14). ,butalsomuchmorevariable,at peaksthanat ,thetimingscoresaveragedconsiderablyless at peaksthanat , :L=leading;I=index;C=roughlycoincident;L g= ,andoftheeconomicprocessgroupsonlythemon eyandcreditseries(VII) ,thetimingperformanceis muchbetterandmorebalancedbetweenpeaksand troughsforthecompositeindexesthanforthec orresponding"allseries" rulecontaina varietyoferrors, ,thegreaterarethechancesofsubstantialdef ects,withconceptual,procedural, , ,thedatavarygreatlyincover-ageandquality , , ,asdemandedforcurrentbusinessanalysisand forecasting , ,they ,asa resultofbenchmarkrevisionsanddefinitiona lchangesthatcause"breaks,"thatis, (averageworkweek,layoffrate,unem-ploymen tduration)ornorevisionsatall(stockprices ,vendorperformance,theprimerate).

8 However,formostoftheindexcomponentsthefi rstprelimi-naryfigureAI(issuedinthecurre ntmonthforthepreviousone)isrevised1-4 timesinasmanysuccessivemonths(Ai'i=2,3,4 ).Thefirsttwochangesgenerallyaccountfora ,boththeaveragestakenwithoutregardtosign andthestan-darddeviationsofEittendtoincr easewithi, ,suggest-ingunderestimationoflevelsinthe earlydata, andbi=1 intheregressionequationsAit=Qi+biAIt+ ,theyaresoforsomeseriesthatarebasedonind irectestimationand,intheirearlyversion, ,verylargeandfrequentrevisionsarefoundin bothsetsoftimeseriesforsuchimportantvari ablesasinflation,monetarychanges,invento ryinvest-ment, , ,seeZamowitz1982c, ,throughthoseinJulyofeachofthethreesucce ssiveyears, ; ,seriesforvariousmonetaryaggregateshaveb eenpublishedatmonthlyandweeklyintervalsa ndrevisedalmostasfrequently(see, , ).Ontheimpor-tanceofrevisionsintheGNPimp licitpricedeflator,seeKeaneandRunkle1990 , ,seeZamowitz1982c, ,Coincident,andLaggingIndicatorstheearly releaseshaslongpresenteda , ,weights,orothertechnicalaspectsoftheind exes.

9 (Aboutallofthismorewillbesaidlaterinthis chapter.)Measurementerrorsareverycommoni neconomicdataandtheyaffecteco-nomicbehav ior,analysis,andforecasting; ,considertestsofthepredictivevalueofthel eadingindex(l)basedonforecastsofrealGNP( q) "truth"tobepredicted?Toansweryestobothqu estionsimpliesthattheforecastsareexpecte dtoeliminatecor-rectlythecumulativefutur eerrorsintheseriesonrealGNPgrowth(asre-v ealedbythewholestringofstatisticalandcon ceptualrevisions).Buttheseerrorsmaybebot hhighlysignificantandtoa , ,testsofthistypecouldbeseverelybiasedaga instfindingla spuriousrejectionofthehypothesisthatthef orecastsarera-tionalorunbiased( ).However,theerrorsinthepredictedvariabl ecanhavedifferentconsequences, , ,theyshouldoptimallybetakenintoaccountby theforecaster(seechapter13).Ontheassumpt ionthattherevisions,asintended,cumulativ elyimprovethedata,11theyprovideimportant information:thelarger,thelessstable,andt hemorestretchedoutintimetheyare, ,thisinformationis ,orthecessationoffurtherrevisions, helpassessthequalityofa , methodanalogoustothatdescribedinthetextp aragraphabove,buttheirtargetvariableis industrialproduction,whichis notalwaystrue:it is notunusualtofindsomerevisions,ina chainofseveral,whichincrease,ratherthanr educe, (outof100),fellheavilyinthe60-79range, minimumlossoftimelinessandprovidemonthly estimatesthesamemonthorearlyinthenextmon th;67aremonthlyseriesavailablewithlagsof 1 month(48),2 months(15),and3 month(4).

10 Seriesthatiscollectedatleastweekly,20for a quarterlyseries,andintermediateformonthl yseries-thelowerthescore, a seriesisavailablepromptly,itsmonth-to-mo nthmovementmaybesoobscuredbyeitherseason alchangeorirregularvariation(noise)astos hedlittlelightonthelonger, ,independent,andstable,theycanbereasonab lywellmeasuredandeliminated(mostindicato rsarepresentlyre-portedinseasonallyadjus tedform).Thenoiseelementvariesgreatlyacr osstheindicators, ;comparestheirregulartothetrend-cyclecom ponentofthegivenseries, ,C;buildsupwhileI;showslittle(andnosyste matic)change, ,inmonths,forwhichllc<1 is calledMCD(monthsforcyclicaldominance).Th esmoothera series, setof81weeklyandmonthlyindicators,30seri eshadMCD=1,21serieshadMCD=2,and30seriesh adMCD~3(37%,26%,and37%,respectively).For 29quarterlyindicators,thecorrespondingfr equencieswere4 (14%),13(45%),and12(41%).Across-classifi cationoftheseriesbyscoresforcurrencyands moothnessshowsnotendencyforthetwoordinal scalestoeitheragreeordisagreewitheachoth ersystematically(seeZamowitz1982c,table4 ).