1.IntroductionLargedevaluationsaregenerallyassociatedwithlargedeclines - PDF document

1.IntroductionLargedevaluationsaregenerallyassociatedwithlargedeclinesintherealexchangerate(RER)andconcomitantlowratesofination.Inthispaper,wearguethattheprimaryforcethatinducestheselowratesofinationisslowadjustmentinthepriceofnontradablegoodsandservices,notslowadjustmentinthepriceofgoodsthatareimportedorexported.Atbothshortandlonghorizons,therateofpassthroughfromexchangeratestopricesismuchlowerfornontradablegoodsthanitisforgoodsthatareactuallytraded.Ourndingsareconsistentwiththeimplicationsofalargetheoreticalliteraturethatemphasizestheimportanceofnontradable-goodspricesinexplainingRERmovementsindevelopingcountries.Ourresultsarealsoconsistentwithanempiricalliteraturethatarguesthatinationislowerthantherateofdevaluationbecausepricesweretoohightostartwith(i.e.thecurrencywas“overvalued”).Wedonottakeastandonwhethercurrencieswereovervaluedbeforelargedevaluations.Instead,wefocusonthequestionofwhichpriceschangeafteralargedevaluation.Ourndingscastdoubtontheviewthatlargedevaluationsareassociatedwithlargedeviationsfromrelativepurchasingpowerparity(PPP)forimportedandexportedgoods.TostudythesourcesofmovementsintheRERwedecomposetheconsumerpriceindex(CPI)intotradableandnontradablegoods.Traditionaldecompositionsclassifygoodsastradablesandservicesasnontradables.Thepricesoftradablegoodsaregenerallymea-suredattheretaillevel.ThisapproachleadsustoconcludethattheprimarysourceofRERmovementsafterlargedevaluationsarechangesinthepricesoftradablegoodsacrosscountries.Wearguethatthisinferenceismisleading.Thebasicproblemisthattheretailpriceoftradablegoodscomprisestwoimportantnontradablecomponents,distributioncostsandlocalgoods.Wedenedistributioncostsaswholesaleandretailservices,marketingandadvertising,andlocaltransportationservices.Wedenelocalgoodsasgoodsthatareproducedsolelyforthedomesticmarket.Thus,theretailpriceoftradablegoodsdoesnotaccuratelyreectthepriceofpure-tradedgoodsatthedock,i.e.,thepriceofgoodsthatareactuallytraded,exclusiveofdistributioncosts.Todealwiththisproblem,weadoptanalternativedecompositionthatdistinguishes EarlytheoreticalworkinthisliteratureincludesMeade(1956),Salter(1959),Swan(1960),andCorden(1960).FormorerecentdiscussionsseeCalvoandVegh(1999)andthereferencestherein.SeeGoldfajnandValdes(1999)andthereferencestherein. Althoughtheratesofinationimpliedbyourprice-accountingexercisearesimilartoactualratesofination,theyarenotidentical.Therearetwofactorsthatmighteliminatetheremainingdiscrepancies.Therstfactorisaspecicationerrorassociatedwithoursimplifyingassumptionthatthedistributionsectorisperfectlycompetitive.Toassesstheimpactofthisassumption,wededucethetimeseriesofmarkupratesinthedistributionsectorthatarerequiredtoreconcileactualandimpliedratesofination.Asitturnsout,movementsinthemarkuprateareobservationallyequivalenttomeasurementerrorinprices.Undereitherinterpretation,therequiredmovementsinthemarkuparesmall.Incontrast,ifweabstractfromdistributionservicesandlocalgoodsthentherequiredmark-upmovementsareimplausiblylarge.ThesecondfactoristhemeasurementerrorintheCPIthatstemsfroma“ightfromquality,”whichwedeneasthesubstitutionbyhouseholdstolower-qualitygoodsintheaftermathoflargecontractionarydevaluations.Inpractice,CPIpriceinspectorsreplaceindividualgoodsintheCPIbaskettoreectchangesindemandandproductturnover.Unlesstheseinspectorsmakeappropriatecorrectionsforqualitychanges,theightfromqualitycaninduceadownwardbiasinCPIinationrates.Ingeneral,itisdiculttoobtaindirectevidenceofthisbias.However,weprovideindirectevidenceontheimportanceofthisbiasbyusingpricedatafromtheEconomistIntelligenceUnit(EIU).Afterdiscussingourprice-accountingexercises,weuseastrategyproposedbyEngel(1999)tounderstandthesourcesofRERmovements.WedecomposeRERmovementsintouctuationsinthepriceofpure-tradedgoodsacrosscountriesandmovementsinthepriceofnontradablesrelativetopure-tradedgoods.WecorroborateourbasicndingthatthemostimportantsourcesofRERmovementsintheaftermathoflargedevaluationsaremovementsinthepriceofnontradablesrelativetopure-tradedgoods.Wecomplementouranalysisofthevelargedevaluationepisodeswithanin-depthanalysisoftheArgentiniancase.Here,weusefourdierentdatasets:disaggregatedCPIdata,datafromourownsurveyofpricesinBuenosAires,scannerdatafromBuenosAiressupermarkets,compiledbyCCR,anArgentinemarketingresearchrm;andourownsurveyofproductoriginwhichclassiesgoodsintheCCRdatasetintoimported,exported,orlocal.Thegeneralpatternthatemergesfromthesedatasupportsourndingsfromtheaggregatedata.Weusethescannerandsurveydatatoprovidesomedirectevidenceofightfromquality 2.LargeDevaluationsInthissectionwesummarizethebehaviorofvariouspricesandexchangeratesinourlargedevaluationepisodes.Throughout,wedenetheCPI-basedrealexchangerateas:RER .(2.1)Inequation(2.1)denotesthetrade-weightednominalexchangerateofthehomecountrywhichwedeneasunitsoflocalcurrencyperunitoftrade-weightedforeigncurrency.ThevariabledenotestheleveloftheCPIinthehomecountry.representsthetrade-weightedCPIofforeigncountries.Table1presentsdataoncumulativelogarithmicchanges,relativetoonemonthpriortothedevaluation,intrade-weightednominalandrealexchangerates,andpricelevelsforvelargedevaluationepisodes:Argentina(2001),Brazil(1999),Korea(1997),Mexico(1994),andThailand(1997).Wefocusontheselargedevaluationcountriesbecauseofdataavailabilityissues.IneveryepisodethereisaconsiderabledropintheRER,reectingthefactthatthehome-countryinationismuchlowerthantherateofdevaluation.Sinceexportandimportpriceindicesplayasignicantroleinouranalysis,webriedigresstodiscusshowtheseindicesareconstructed.Thetwomostcommonmethodsaretobuildindicesbasedonpricedatacollectedbysurveyingimportersandexporters,andunitvalueindices(UVIs)computedfromtradestatisticsastheratioofthelocalcurrencyvalueofexportsorimportstovolume(weightorquantity).UVIsarecomputedforindividualgoodcategoriesandthenaggregated.Countriesthatconstructsurvey-basedindicesincludeAr-gentina(importprices),Korea(bothimportandexportprices),andMexico(exportprices).CountriesthatuseUVI-basedimportandexportindicesincludeBrazil(seeGuimarãesetal.(1997)andMarkwaldetal.(1998))andThailand.MexicodiersfromothercountriesinthatthegovernmentconstructsitsimportpriceindexusinginformationonproducerandexportpricesintheU.S.Consequently,weshouldtreattheMexicanimportpriceindexwith Wecomputetrade-weightednominalandrealexchangeratesusingasetoftradingpartnersthatrepre-sent,onaverage,80percentofexportsandimportsforeachcountry.Foreachcountry,wechoosethelargesttradingpartnersforwhichwehavepricedata.SomestudiesusetheProducerPriceIndex(PPI)ortheWholesalePriceIndex(WPI)asameasureofthepriceofpure-tradedgoods.Wethinkthatthesearepoormeasureofpuretradedgoodspricesfortworeasons.First,sincethePPItargetspriceschargedbydomesticproducers,importpricesaregenerallyexcluded.Second,thecomposition,coverageandavailabilityofthePPIandWPIvarieswidelyacrosscountries(seeMaitland-Smith(2000)). priceofimportsandexports(seealsoFigure1).Table1alsoshowsthatthepriceofnontradablegoodsandservicesmovesbymuchlessthantherateofdevaluation.Althoughtheretailpricesoftradablegoodsmovemorethanpricesofnontradablegoodsandservices,thedierencesaresmallrelativetooverallmovementsinthenominalexchangerate.Inprinciple,theslowresponseofnontradable-goodspricescouldreecttheimportanceofgovernment-controlledprices.Table1indicatesthatwiththeexceptionofArgentina,therateofinationforpublicgoodsisroughlythesameastheoverallrateofinationfornon-tradables.EveninArgentina,thedierencebetweenpublic-goodsinationandnontradablesationissmallrelativetotheoverallrateofdevaluation.Weconcludethatthereisastarkdierenceinthebehaviorofthetwoalternativemeasuresoftradableprices,importandexportprices,andretailprices.RelativePPPisareasonabledescriptionofthebehaviorofpure-traded-goodsprices.However,itisaverypoordescriptionofthebehavioroftradable-goodsretailprices.3.PriceAccountingInprinciple,therearemanypossibleexplanationsforthedeclinesintheRERafterlargedevaluations.SupposethatmostgoodsintheCPIbundlearetradableandrelativePPPisapoordescriptionofthepriceofpure-tradedgoods.Then,thefallintheRERwouldprimarilyectabigchangeintherelativepriceofpure-tradedgoodsacrosscountries.Alternatively,supposethatmostgoodsintheCPIarenontradableandrelativePPPisagooddescriptionofpure-tradedgoodprices.Then,thefallintheRERprimarilyreectsalargechangeinthepriceofnontradablegoodsrelativetopure-tradedgoods.Ofcourse,therearemanyintermediatepossibilities.Toassesstheempiricalrelevanceofthedierentpossibilities,thissectiondiscussesourestimateoftheweightofnontradablesintheCPI.Wethendescribeaseriesofprice-accountingexercisesthatarehelpfulinisolatingthedeterminantsofCPIationandRERmovements. ThisisconsistentwithFrankel,Parsley,andWei’s(2004)ndingthatexchange-ratepassthroughishigherforimportpricesatthedockthanforretailprices.TheyndweakerevidencethanwedothatrelativePPPholdsatthedock.Thisdierencemayrefectthefactthat,unlikethem,wefocusonlargedevaluationepisodes. contentofconsumptionforourvelargedevaluationcountries.Ourcalculationsuseinput-outputtables,exceptforBrazilandThailandforwhichweusedatafromtheNationalIncomeandProductsAccounts.Table3reportstwomeasuresoftheimportanceofpure-tradedgoodsinconsumption.rstmeasureisthefractionofimportednalgoodsintotalconsumptionexclusiveofdis-tributionservices.Werefertothismeasureasthe“direct”importcontentofconsumption.Thesecondmeasureisthedirectimportcontentplusthevalueofimportedintermediateinputsusedtoproducenalconsumptiongoodsasafractionoftotalconsumptionexpendi-tures.Werefertothismeasureasthe“total”importcontentofconsumption.Table3showsthat,forallofourlargedevaluationcountries,theimportanceofpure-tradedgoodsinconsumptionissmall.Theaveragedirectimportcontentofconsumptioninthesecountriesispercent,whiletheaveragetotalimportcontentofconsumptionispercent.Wewouldliketoestimatethefractionofexportablegoodsinconsumption.Unfortu-nately,wecannotuseinput-outputmatricestoconstructtheseestimates,becausetheydonotcontaininformationonthefractionofexportablegoodsthatisconsumeddomestically.Inlightofthis,weusetwoboundsfortheimportanceofpuretradablegoodsinconsumption.Thelowerbound,whichabstractsfromexportables,isequaltothetotalimportcontentofconsumption.Theupperbound,whichabstractsfromlocalgoods,isequaltooneminustheconventionalestimateoftheshareofnontradablegoodsplusdistributionservicesinconsumption.3.2.Price-AccountingResultsWethinkoftheCPIasageometricaverageoftheretailpriceoftradablegoods()andthepriceofnontradablegoodsandservices(.(3.1)Inequation(3.1),istheweightofnontradablegoodsintheCPI,astraditionallyde-ned.Wetaketheviewthattradablegoodsincludebothgoodsthatareactuallytraded Unfortunately,wewerenotabletoobtaindatafortheotherfourcountriesincludedinTable3.ForBrazilwemeasurethedirectandtotalimportcontentinnaldemand.Investmentgoodsgenerallyhaveahigherimportcontentthanconsumptiongoods.SeeBurstein,Neves,andRebelo(2004). whereisanarbitraryconstant.Inthiscase,wemeasuretheforeignpriceofpure-tradedgoods()usingasimplegeometricaverageofforeignimportandexportprices.Wecomputeforeignimportandexportpricesastrade-weightedgeometricaveragesoftheimportandexportpriceindicesofthetradingpartnersforthecountryinquestion.Incase2wedonotimposerelativePPP.Instead,wemeasureusinganequallyweightedgeometricaverageoftheimportandexportpriceindicesforthecountryinquestion.Weconsiderfouralternativeassumptionsabout:thatallgoodsaretraded,thatsomegoodsarenontradable,buttherearenodistributioncostsorlocalgoods;thattherearenontradablegoodsanddistributioncostsbutnolocalgoods;andthattherearenontradablegoods,distributioncosts,andlocalgoods.Foreachsetofassumptionsweuseequation(3.6)tocomputetheimpliedrateofCPIination.Wethencomparetheimpliedratewiththeactualinationrateoneyearafterthedevaluation.AssumptionA:AllGoodsAreTradedSupposethatallgoodsaretradedsothat.Table4showsthatforbothmeasuresof,theimpliedrateofinationismuchhigherthantheactualrateofinAssumptionB:NontradableGoodsbutNoDistributionCostsorLocalGoodsSupposethattherearenontradablegoodsbutnodistributioncosts(),andnolocalgoods()sothat.WesettotheconventionalmeasureoftheweightofnontradablegoodsintheCPIbasketreportedinTable3.Wenotethattheassumptionoflocalgoodsisisomorphictotheassumptionthattherearenolocalgoods,buttheirpricemovesone-to-onewiththepriceofpure-tradedgoods().Table4indicatesthatforallcountriesandbothmeasuresof,therateofinationimpliedbytheprice-accountingexercisesismuchlargerthantheactualrateofination.AssumptionC:NontradableGoodsandDistributionCostsbutnoLocalGoodsWeintroducedistributioncostsandassumeapercentdistributionmargin().WecontinuetosettothevaluesreportedinTable3.Theresultingvalueofprovidesanupperboundontheimportanceofpure-tradedgoods,sinceitcompletelyabstractsfromlocalgoods().Table4indicatesthat,forallcountriesandbothmeasuresof,the WedonotincludeThailandinouranalysisbecausewedonothavedataon pricesoftradablesare,byconstruction,thesame.Supposethatretailpricesaregivenby:Thentheretailpriceoftradablegoodsisgivenby::(¯PIt)(13φ)(13θ)(PNt)13(13φ)(13θ)].(3.8)Oneinterpretationofisthatitrepresentsmeasurementerrorinthedierentpriceindices.Amoreinterestingpossibilityisthatretailrmsarenotperfectlycompetitive.Inthiscase,correspondstothetimemarkup.Forexpositionalpurposesweadoptthelatterinterpretation.Toquantifythedierencebetweenactualandpredictedretailprices,wecomputethevaluesofrequiredforequation(3.8)toperfectlytthetimeserieson.Sincewetakefromthedata,thesevaluesofalsoimplythatthepredictedratesofinationcoincideateachpointintimewithactualratesofination.Figure3showsthetimeseriesforunderassumptionsB,C,andD.Wemeasureusinganequallyweightedgeometricaverageoftheimportandexportpriceindicesforthecountryinquestion.WelookrstatourresultsunderassumptionD(nontradables,localgoods,anddistributioncosts).Here,thevariationsinarerelativelysmall.TheyarealwayslowerthanpercentinBrazilandMexicoandlowerthan20percentinArgentinaandKorea.Toputthesenumbersinperspective,supposethatthepre-devaluationgrossmarkupis.ThisassumptionimpliesthatinBrazilandMexiconeverfallsbelow.InKoreaneverfallsbelow1.02,butintheArgentiniancasetheimpliedmarkuprisessomewhat.Next,weconsiderourresultsunderassumptionC(nontradablesanddistributioncosts,butnolocalgoods).Movementsinaresmallerthan20percentforallcountries.Wenotethatinthiscase,themarkupinArgentinanolongerrisesbutdeclinesbyamodestamount.WhenweexamineourresultsunderassumptionB(nontradablesbutnodistributioncostsorlocalgoods)weseethatmovementsinaremuchlarger.Forexample,themaximaldeviationexceeds60percentforArgentinaandisover40percentfortheotherthreecountries.So,ifthepre-devaluationmarkupis1.25,thenthegrossmarkupfallsbelow0.69inArgentinaand0.84fortheothercountries.Thisbigfallinmeansthatretailerswouldhavebeensellingproductsatalossforlongperiodsoftime.Toavoidthisimplicationwhilemaintaining outlets.WegroupEIUitemsintofourcategories:clothing,durables,food,andothermiscella-neousgoods.Wethencomputeapriceindexforeachcategoryineachcountry.Theweightwegivetoanindividualitem(e.g.,yogurt,natural,150grams)inacategory(e.g.,food)isequaltotheweighttheitemreceivesintheU.S.CPI.Tocalculateanoverallrateofation,weassigntoeachcategorytheweightthatitreceivesintheU.S.CPIbasket.WealsousetheseweightstocomputeapartialCPIinationrateasanaverageoftheoationratesforthefourcategoriesusedincomputingtheEIUinationrate.Wethencomparethetwoinationratestoassesstheimportanceoftheightfromquality.AlthoughtheEIUinationrateisanimperfectmeasure,itprovidessomeusefulinformationaboutthemagnitudeofmeasurementerrorassociatedwiththeightfromquality.Foreachcountry,Table5reportsthepartialCPIinationrateandtwoEIUinationrates.WecomputethesetwoEIUinationmeasuresusingdatacollectedinlow-andhigh-priceoutlets,respectively.ForMexicowehaveinstitutionalinformationthattherewereveryfewunplannedsubstitutionsofitemsintheCPIbasket.ThisinformationsuggeststhattheightfromqualitydoesnothaveabigimpactontheMexicanCPI.Interestingly,theEIUinationratetrackstheocialCPIinationquitecloselyforMexico.ThisfactgivesussomecondencethattheEIUinationrates,aswecomputethem,areuseful.Forallothercountriesinoursample,thepartialCPIinationrateislowerthanbothEIUinationrates.ThedierencesareparticularlynotableforBrazil,Korea,andThailand.Ourresultssuggestthatightfromqualityprobablyinducedasystematicdownwardbiasininationratesafterlargedevaluations.Butitisdiculttobepreciseaboutthequantitativeimportanceofthisbias.4.EngelDecompositionsInthissectionweuseanapproachproposedbyEngel(1999)tostudythesourceofRERmovements.Weassumethatequation(3.1)denesthedomesticCPIandthattheforeign Weusetheseweightsbecauseofdatalimitations.Seetheappendixfordetails.BancodeMexicoupdatedthelistofgoodsusedtocomputetheMexicanCPIinFebruary1995.Thisupdatehadbeenplannedformonthsandwasunrelatedtothedevaluation.UsingseveralissuesoftheDiariocialdelaFederacionfor1995,weconcludedthatbetweenFebruaryandJulyonly233goodsoutof5494wereaddedtotheCPIbasketforMexicoCity.Onlyafractionofthe233newgoodswereintroducedtoectchangesinconsumptionpatterns. Also,log( RERisthedierenceacrosscountriesinthepriceofnontradables,relativetotradedpricesmeasuredatthedock:log( RERlog(log(Table6andFigure4reportchangesinlog( RERintheaftermathofourvelargedeval-uationepisodes.Wemeasureusinganequallyweightedgeometricaverageofimportandexportpriceindices.Wealsoreportresultsforthecaseinwhichwemeasureusingjusttheimportpriceindex.Wenotethatmovementsinlog( RERaccountforamuchsmallerfractionofthechangeslog(RERthanmovementsinlog(RER.WeconcludethatwhenRERsareconstructedusingat-the-dockpricesofpure-tradedgoodstheydonotfallbyasubstantialamountafterlargedevaluations.Onceweallowfortime-varyingmarkups,equation(3.8)impliesthattheEngeldecom-positiontakestheform:log(RER)=log( RER)+(1)log()+log( RERWenotethatthisequationimpliesthatourpreviouscalculationsabouttheimportanceof RERintheEngeldecompositionareunaectedbythepresenceoftime-varyingmarkups.Viewedasawhole,andconsistentwithourprice-accountingexercises,theresultsinthissectionprovidestrongevidenceinfavoroftheviewthat,inourlargedevaluationepisodes,movementsintherelativepriceofpure-tradedgoodsacrosscountriesisnotanimportantsourceofRERmovementsafterlargedevaluations.5.ACloserLookattheData:Argentina’s2001DevaluationWecomplementtheevidencepresentedabovewithanin-depthlookatanepisodeforwhichwehavemoredetaileddata,Argentina’s2001devaluation.Ourinformationcomesfromfourdierentdatasets:disaggregatedCPIdatafromINDEC(InstitutoNacionaldeEs-tadísticayCensos,theArgentinianNationalStatisticalAgency);datafromourownsurvey Forexpositionalpurposesonlyweassumethat(asdenedinequation(3.7))isthesameacrosscountries.Forsimplicityweabstractfrommarkupsinforeigncountries.ArgentinadevalueditsocialexchangerateinJanuary2002,butbankswereclosedinthelastfewdaysofDecember2001. percentofthetotalmarketshareforeachofthe21productcategories.ThesurveytookplaceinOctober2002inseveralBuenosAiressupermarkets.Productlabelsofthegoodsthatwesurveyedindicatewhetherthegoodwasimported,exported,orneither.Usingthisinformation,weclassiedeachproductasbeingimported,exportable,orlocal.UsingCCRdataonmarketsharesofindividualproducts,wecomputedthesharesofimportedandexportablesforeachproductcategory(e.g.beer,bread,etc.).WeobtainedtherateofationforeachproductcategoryfromINDECfortheperiodDecember2001toJune2002.Table8reportsthesharesofimportedandexportablesforeachproductcategoryandthecorrespondingrateofination.Ourkeyndingisthatthereisastrongpositivecorrelation)betweentherateofinationofaproductcategoryandthemarketshareofimportedandexportablegoodsinthatproductcategory.5.2.OurSurveyofPricesinBuenosAiresWenowturntoadatasetthatallowsustoassesshowoftentheretailpricesoftradableandnontradablegoodsareadjusted.ThisdatasetisbasedonasurveyofpricesinBuenosAiresthatweconductedbetweenMarch27andDecember24,2002.Ourdataappendixcontainsacompletedescriptionofthegoodsandservicesincludeinthesurvey.Wecollectedpricesfor58goodsineightsupermarketsatweeklyfrequency.Foreachgoodwegatheredinformationondierentitems.Anitemisatypeofbrandandsizeofagoodcollectedinagivensupermarket(e.g.cookingoil,Cocinerobrand,1.5literbottle,inDiscosupermarket).Wehavepriceson518items.Wealsocollecteddataontenservicesinoneortwodierentoutlets.Intotal,wehavepriceson17serviceitems.Allofthegoodsthatweconsiderwouldbeclassiedastradableunderastandardclassication.Theservicesinoursurveyareallclearlynontradableinnature.Wecomputethefrequencyofpricechangesforeachindividualitem.Frequencyisdenedasthenumberofperiodsinwhichapricechangeoccursrelativetothepreviousperiod,asafractionofthetotalnumberofperiodsinwhichtheitemisavailable.Forexample,supposethatwehaveweeklydataonthepriceofanitemforaneight-monthperiod.Ifthepriceoftheitemchangesinonlytwoweeksduringthatperiod,thenthefrequencyofpricechangeis2/(8x4)=1/16.Iftheitemchangespriceeveryweek,thenthefrequencyofpriceadjustmentisone. Inpractice,foragivengoodorservice,wecomputethefrequencyofpricechangesasanaverageofthefrequencyofpricechangesindierentsupermarketsorlocations. percentforgoodsandzeropercentforservices.OurndingthatservicepriceschangelessfrequentlythangoodspricesisconsistentwithBilsandKlenow’sresults(2004).OurresultsindicatethatgoodspricesadjustedmuchmorerapidlyinArgentinathanintheU.S.andIsrael.Incontrast,servicepricesadjustedmuchmoreslowlyinArgentinathanintheU.S.5.3.FlightfromQualityinArgentinaWeuseourscannerdatatoprovideevidenceoftheightfromqualityinArgentina.WecalculatethefractionofindividualproductsattheSKUlevelthatwere“destroyed,”i.e.,disappearedfromsupermarketshelves.Foreachof24productcategories,thecolumnsofTable10reportthefractionofSKUsthatweredestroyedinthesixmonthspriortoJune2000,December2000,June2001,December2001,andJune2002.ThemedianfractionofSKUsacrossallproductcategoriesthatdisappearedrosefrompercentinJune2001topercentinJune2002.Next,wecalculatetheweightedaveragepriceofSKUswhosemarketssharefallbymorethan2percentinanygivenperiod.WealsocomputetheaveragepriceofSKUswhosemarketsharerisesbymorethan2percent.ThelastthreecolumnsofTable10reporttheratioofthesetwoaveragepricesforeachofthe24goodscategoriesinoursample.ThemedianofthisratioacrossproductcategoriesrosefromintheyearprecedingJune2000intheyearprecedingJune2002.Totheextentthatpriceiscorrelatedwithqualitythischangeinthemedianisclearevidenceofaightfromquality:theproductsthatlosemarketsharehavesignicantlyhigherpricesthandothoseproductsthatgainmarketshare.AdditionalevidenceonightfromqualityinArgentinacomesfromCCR,whichclassibrandsintodierentcategoriesorderedbydecreasingquality:premium,rstbrands,newrstbrands,supermarketbrands,andlow-pricebrands.Table11reportsthepriceofgoodsineachcategoryrelativetothepriceofgoodsinallcategoriesfortheyear2001.Thetablealsocontainsthemarketsharesofeachcategoryfor2001,2002,and2003.Twokeyfactsareworthnoting.First,theaveragepriceofpremiumandrstbrandsisroughlypercenthigherthanistheaveragepriceoflower-qualitybrands.Thisfactisconsistentwiththenotionthatpriceiscorrelatedwithquality.Second,afterthedevaluationthereisacleardeclineinthemarketshareofpremiumandrstbrands,from71percentin2001to63percentin2003.Thisdeclineisconsistentwithanongoingightfromquality.UsingconsumptiondatacollectedbyLatinPanel,whichincludes3,000Argentinehouseholds, devaluedbyroughlypercentperyear.DuringthisperiodtherewasacumulativeRERappreciationofpercent.InMexico,therealexchangeratepeakedinJanuary1994.RealGDPgrewatanaverageannualrateof3.8percentbetween1988and1994.Table12summarizesthebehaviorofpricesandexchangeratesinourtwolargeap-preciationepisodes.TheRERsaremeasuredrelativetotheUSdollar.WenotethattheCPI-basedRERappreciatedby28percentinArgentinaand46percentinMexico.Thisappreciationwasaccompaniedbyariseinthedollarretailpriceoftradables(31and76per-centinArgentinaandMexico,respectively).Thesefactsimplythat,whenwemeasurethepriceoftradablegoodsusingretailprices,mostofthemovementintheRERisaccountedforbychangesinthepricesoftradablegoods(seedecomposition(4.1)).Inparticular,thepercentagesofmovementsintheRERaccountedforbychangesinthepricesoftradablegoodsare89and58percentinArgentinaandMexico,respectively.Incontrast,movementsinpure-tradedgoods,whichwemeasureusinganequallyweightedgeometricaverageofimportandexportpriceindices,accountforarelativelysmallfractionofmovementsintheRER:percentinArgentinaand-percentinMexico.Thisresultectsthefactthatconsistentwithourpreviousevidence,relativePPPisareasonabledescriptionofthebehaviorofimportandexportprices.Thechangeinthedollarpriceofpure-tradedgoodswasonly-2percentinMexicoand6percentinArgentina.7.Medium-SizeDevaluationsandSmallExchange-RateFluctua-tionsHere,weexaminethebehavioroftheRERinmedium-sizedevaluationsandsmallexchange-uctuations.Unfortunately,fordataavailabilityreasons,wecannotholdconstantthesetofcountriesthatwestudy.Thesetofmedium-sizedevaluationepisodesforwhichwehavereliabledataarethe1992devaluationsinFinland,Italy,Sweden,andtheUK.averagerateofdevaluationinthersthalfyearaftertheonsetoftheseepisodesis20percent.Thecorrespondinggureforourlargedevaluationepisodesis62percent.ConsistentwithEngel(1999),wealsostudyexchange-rateuctuationsbyusingquarterlydatafromelevenOECDcountries:Australia,Canada,Denmark,Finland,Germany,Italy, TheseresultsareconsistentwithMendoza(2000),whondsthatmovementsintherelativepriceofnontradablegoods,especiallyhousing,accountedforalargefractionofmovementsintheCPI-basedRERduringtheMexicanexchange-rate-basedstabilizationepisode.During1993,realGDPfellby-1.2percent,-2percent,and0.9percentinFinlandandSwedenandItaly,respectively.TheUKexperiencedamildexpansion,withrealGDPgrowingby2.3percentin1993. HP-ltereddata,weconstructtwostatistics:thecorrelationbetweenbothlog(RERlog( RERwiththenominalexchangerate;andtheratioofthestandarddeviationofourRERmeasurestothestandarddeviationofthenominalexchangerate.Table15inthedataappendixreportsresultsforbothtrade-weightedexchangeratesandbilateralexchangerateswiththeU.S.dollar.ConsistentwithresultsinMussa(1986),wendthereisaveryhighcorrelationbetweenthelogarithmofthenominalexchangerate,log(1,andlog(RER:theaveragecorrelationbetweenthesetwoseriesacrossourelevencountriesispercent.Thevolatilityofthesetwoseriesisroughlythesame.AuthorssuchasChari,Kehoe,andMcGrattan(2002)arguethattheycanobtainsimilarresultsiftheylog(RERwithlog(RER.TheseobservationslieattheheartoftheviewthatstickytradablegoodspricesareanimportantsourceofRERmovements.Next,weexamineourresultswhenweworkwithlog( RER.Wendthatlog( RERismuchlessvolatilethanlog(RER.However,thereissubstantialheterogeneityacrosscountriesintheratioofthestandarddeviationoflog( RERtothestandarddeviationoflog(RER.TorelatethisstatistictotheEngeldecomposition(4.2),wecomputethevarianceonbothsidesofthatequationbyabstractingfromthecovariancebetweenlog( RERlog( RERndthatmovementsinlog( RERaccountonaverageforpercentofthestandarddeviationinlog(RER.Wealsondthatlog( RERismuchlesscorrelatedwiththelogarithmofthenominalexchangerate,log(1,thanlog(RER.Theaveragecorrelationwiththelogarithmofthenominalexchangeispercentforlog(RERpercentforlog( RERWeconcludethatchangesinthepriceofnontradablegoodsrelativetopure-tradedgoodsaccountforasignicantfractionofmovementsintheRER.Butuctuationsinthepriceofpure-tradedgoodsacrosscountriesaccountforthemajorityofsmallmovementsinthe8.ConclusionInthispaperwearguethattheprimaryforcebehindthelargefallinrealexchangeratesthatoccursafterlargedevaluationsistheslowadjustmentinthepriceofnontradablegoodsandservices.ItisnotthefailureofrelativePPPforgoodsthatareactuallytraded. Thedataappendixisavailableuponrequest.Wecannotcomputethiscovariancetermbecause,formostofthecountriesinoursample,thetimeseriesforthepriceofnontradablegoodsistooshort. [11]Crucini,Mario,ChrisTelmer,andMariosZachariadis“UnderstandingEuropeanRealExchangeRates,”forthcoming,A.E.R.,2005.[12]Engel,Charles,“AccountingforU.S.RealExchange-rateChanges,”,107:507-538,1999.[13]Frankel,Jerey,DavidParsley,andShang-JinWei“SlowPassthroughAroundtheWorld:ANewImportforDevelopingCountries?,”mimeo,I.M.F.,2004.[14]Goldfajn,IlanandRodrigoO.Valdes“TheAftermathofAppreciations,”Q.J.E.,Vol.114,No.1.(Feb.,1999),pp.229-262.[15]Guimarães,Eduardo,ArmandoPinheiro,CarmenFalcão,HenryPourchet,andRicardMarkwald“ÍndicedePreçoeQuantumdasExportaçõesBrazileiras,”FundaçãoCentrodeEstudosdoComércioExterior,1997.[16]Klenow,Pete“MeasuringConsumptionGrowth:TheImpactofNewandBetterProd-ucts,”Fed.ReserveBankofMinneapolisQ.Rev.,10-23,Winter2003.[17]Lach,Saul,andDanielTsiddon,“TheBehaviorofPricesandInation:AnEmpiricalAnalysisofDisaggregatedPriceData.”J.P.E.,100:349-389,1992.[18]Maitland-Smith,Fenella“ProducerPriceIndices,”mimeo,OECD,2000.[19]Markwald,Ricard,ArmandoPinheiro,CarmemFalcão,HenryPourchet,“ÍndicedePreçoeQuantumdasImportaçõesBrazileiras,”FundaçãoCentrodeEstudosdoComér-cioExterior,1998.[20]McKenzie,DavidandErnestoSchargrodsky“BuyingLess,ButShoppingMoreandShoppingDown:ChangesinConsumptionPatternsDuringaCrisis,”mimeo,Univer-sidadTorcuatoDiTella,2004.[21]Meade,James“ThePriceAdjustmentandtheAustralianBalanceofPayments,”Econ.Record,1956.[22]Mendoza,Enrique,“OntheInstabilityofVarianceDecompositionsoftheRealExchangeRateacrossExchange-Rate-Regimes:EvidencefromMexicoandtheUnitedStates,”N.B.E.R.WP7768,2000. Table 1Prices and Exchange Rates in Large DevaluationsCumulative Logarithmic Change A rgentina - December 2001Brazil - December 19983 months6 months12 months24 months3 months6 months12 months24 monthsUS$ nominal exchange rate86.9128.2123.5107.445.338.242.448.7Trade-weighted nominal exchange rate86.6124.7110.6108.343.234.639.641.0Import prices (at the dock)68.1111.4111.399.344.336.743.149.5Export prices (at the dock)n.a.n.a.n.a.n.a.37.426.732.743.3Consumer price index9.226.634.337.92.83.98.614.4Nontradable prices2.88.313.018.31.42.25.19.7Retail price of tradables15.241.651.553.84.45.711.416.8Public goods prices0.84.54.77.52.22.98.1n.a.CPI-based real exchange rate (trade weighted)-78.5-100.2-82.2-81.7-40.5-31.5-32.7-30.7Korea - September 1997Mexico - November 199 4 3 months6 months12 months24 months3 months6 months12 months24 monthsUS$ nominal exchange rate49.049.341.227.650.254.980.083.3Trade-weighted nominal exchange rate46.346.637.330.850.256.180.482.9Import prices (at the dock)28.738.821.519.752.559.184.086.9Export prices (at the dock)32.444.822.511.051.657.979.687.1Consumer price index2.66.56.67.48.726.239.564.0Nontradable prices2.15.05.14.86.721.631.653.6Retail price of tradables3.08.08.210.210.029.645.672.1Public goods prices1.07.17.5n.a.3.620.729.253.1CPI-based real exchange rate (trade weighted)-43.3-39.4-30.4-24.6-42.0-31.5-42.7-24.33 months6 months12 months24 monthsUS$ nominal exchange rate34.256.349.735.8Trade-weighted nominal exchange rate30.546.235.928.6Import prices (at the dock)30.150.340.420.4Export prices (at the dock)31.447.532.317.6Consumer price index3.75.310.18.9Nontradable pricesn.a.n.a.n.a.n.a.Retail price of tradablesn.a.n.a.n.a.n.a.Public goods pricesn.a.n.an.a.n.a.CPI-based real exchange rate (trade weighted)-27.4-41.1-26.2-20.7 ArgentinaBrazil200120022003199819992000US$ nominal exchange rate0.0112.0106.50.047.047.9Trade-weighted nominal exchange raten.a.n.a.n.a.n.a.n.a.n.a.Consumer price index0.023.035.60.08.014.0Imports0.0107.4104.30.048.851.0Exports0.0103.6102.50.034.741.2 Korea Mexico1996199719981999199419951996US$ nominal exchange rate0.016.855.539.10.064.381.2Trade-weighted nominal exchange raten.a.n.a.n.a.n.a.n.a.n.a.n.a.Consumer price index0.04.311.612.40.030.059.6Imports0.011.035.416.20.066.986.1Exports0.04.925.63.10.058.679.1 Table 3Share of Tradables in CPI, and Import Content of Consumption A rgentinaBrazilKoreaMexicoThailand53.059.348.053.543.3 Direct import content4.3 2.3 *6.64.77.8 * Total import content10.5 8.9 **20.610.920.7 *FinlandItalySwedenUKUS58.765.848.069.042.9 Direct import content13.16.613.612.04.7 Total import content24.016.225.320.99.1 Table 6 How Much of the Decline in the RER Is Due to the Decline in the Price of Tradable Goods ? Percent as a Fraction of Change in CPI-based-RER A rgentina - December 2001Brazil - December 199 8 3 months6 months12 months24 months3 months6 months12 months24 monthsRetail prices92.485.079.180.596.194.491.492.0Simple geometric average of import and export prices22.419.720.826.55.210.412.7-1.2Import prices21.819.122.526.5-3.4-3.72.5-1.4 7 Mexico - November 199 4 3 months6 months12 months24 months3 months6 months12 months24 monthsRetail prices99.096.194.888.796.989.285.766.8Simple geometric average of import and export prices38.47.140.835.8-0.91.73.6-3.1Import prices42.511.336.918.8-3.2-1.2-3.8-0.7 7 3 months6 months12 months24 monthsRetail pricesn.a.n.a.n.a.n.a.Simple geometric average of import and export prices2.83.26.526.6Import prices3.7-1.5-16.912.9 Table 8 Inflation and Market Share of Imports and Exportables in Argentin a Product Categor y InflationMarket Share(log percent)Imported + ExportablesDec 01 - June 02Oct-02Beer32.412.7Bread33.452.7Cereals40.355.7Cleaning liquids50.486.2Coffee45.155.8Deodorant50.086.4Detergents for clothes67.166.0Diapers83.372.4Dish detergents50.132.1Female protection67.085.7Hamburgers17.90.7Insect killer53.377.7Juice50.011.5Mayonnaise64.595.4Milk41.90.2Paper towels52.450.8Shampoo47.371.4Soap46.070.1Soda31.53.9Toothpaste67.068.3Yogurt27.74.1 s GoodsService s Weekl y Number of products5810Median frequency of price adjustment (%)30.00.0Median time between price change5.3[39 , ) y Number of products5810Median frequency of price adjustment (%)66.50.0Median time between price change1.7[9 , ) Table 10: Prices, Market Shares, and Product Destruction in Argentin a Product destruction Average price of SKUs whose market share Percentage of SKUs destroyed in previous 6 months� fell by 2% between June t , t+1 / June 2000 June 2001 June 2002 June 2000Dec. 2000 June 2001Dec. 2001 June 2002 June 1999 June 2000 June 2001Beer19152617320.91 0.86 a Price of category/ Average priceMarket Shareacross categories, in 2001200120022003Premium brands1.513.710.710.4 Table 12 Real Exchange Rate Appreciations in Exchange-Rate-Based StabilizationsCumulative Logarithmic ChangeMexico , March 1988 - January 1994 A rgentina , April 1991 - January 199 5 US$ nominal exchange rate30.9US$ nominal exchange rate0.0Import prices (at the dock)43.0Import prices (at the dock)10.9Export prices (at the dock)32.8Export prices (at the dock)n.a.Consumer price index100.1Consumer price index38.1Retail price of tradables76.2Retail price of tradables31.1Nontradable prices excluding housing103.4Nontradable prices56.7Housing183.6CPI-based RER46.3CPI-based RER27.6CPI (excluding housing) - based RER32.8CPI (excluding housing) - based RERn.a.Retail price of tradables - based RER26.9Retail price of tradables - based RER24.5Simple geometric average of import and export price - based RER-1.7Simple geometric average of import and export price - based RER5.8 Table 1 4 How Much of the Decline in the RER Is Due to the Decline in the Price of Tradable Goods ? Percent as a Fraction of Change in CPI-based-RERFinland - August 199 2 Italy - September 199 2 3 months6 months12 months24 months3 months6 months12 months24 monthsRetail prices93.991.485.374.296.997.896.294.9Simple geometric average of import and export prices50.473.760.141.437.940.447.148.6Import prices38.464.847.338.125.034.448.645.5 2 UK - August 199 2 3 months6 months12 months15 months3 months6 months12 months15 monthsRetail prices93.092.792.392.599.698.099.2107.2Simple geometric average of import and export prices73.363.761.945.274.534.8-4.5-11.4Import prices69.155.049.036.159.331.4-7.9-23.4 2002 2002.5 2003 2003.5 0 20 40 60 80 100 120 Argentina 1995.5 1996 1996.5 0 20 40 60 80 Mexico 1997.5 1998 1998.5 1999 1997.5 0 10 20 30 40 50 60 Thailand 2003 2003.5 0 20 40 60 80 100 120 2002 2002.5 2003 2003.5 -60 -40 -20 0 20 1999 1999.5 2000 2000.5 -40 -30 -20 -10 0 1999 1999.5 -40 -30 -20 -10 0 -50 -40 -30 -20 -10 0 10 1995 1995.5 1996 1996.5 -50 -40 -30 -20 -10 0 1998 1998.5 1999 1999.5 -50 -40 -30 -20 -10 0