Michal Grajek Christian Wey - PDF document

J Michal Grajek Christian Wey Berlin, November 2005 Königin-Luise-Str. 5 Tel. +49 (30) 897 89-0 Fax +49 (30) 897 89-200 requires the express written HypermarketcompetitionandthediffusionofretailcheckoutbarcodescanningJonathanBecka,b,MichalGrajeka,bChristianWeyc,d,eNovember2005AbstractThispaperpresentsasetofpaneldatatostudythediffusionofretailcheckoutbar-codescanningintenEuropeancountriesovertheperiod1981-1996.Estimatesfromastandarddiffusionmodelsuggestthatcountriesdiffermostinthelong-rundif-fusionlevelofbarcodescanningandlessintimingordiffusionspeed.Wepresentevidencethattheemergenceofhypermarketsraisescompetitiveintensityandusehypermarketdata,amongothervariables,inapooledestimation.Resultssuggestthathypermarketcompetitionreduceslong-runadoptionofinformationtechnology(IT)inretailing.Inparticular,theemergenceofhypermarketsseemstodeepenretailsegmentationbyinducingpotentialadopters(e.g.supermarkets)toexitthemarketand/orbydiscouragingadoptionbyotherretailformats.Consistentwithexpecta-tions,scaleandincomeeffectsspurITdiffusionandthereisaclassicsubstitutioneffect:whenwagesrise,diffusionofalabor-savingtechnologysuchasbarcodescan-ningismoreintense.Wedonotndasignicantimpactofemploymentprotectionlegislation.JELclassication:L5,L81,O33.Keywords:ITdiffusion;retailcompetition;hypermarkets.Afliations:aWissenschaftszentrumBerlin(WZB),bHumboldtUniversit¨atzuBerlin,cDeutschesInstitutf¨urWirtschaftsforschung(DIW),dTechnischeUniversit¨atBerlin,eCEPR,London. Correspondingauthor(beck@wz-berlin.de;postaladdress:WZB,Reichpietschufer50,D-10785Berlin).ThankstoJoeClougherty,LapoFilistrucchi,OzShy,IrinaSuleymanovaandparticipantsattheEARIEcon-ferenceinPorto,the5thZEWConferenceontheEconomicsofICTinMannheim,theBDPEMSworkshop,theInterValyearlymeetingandtheWZBseminarforcommentsandhelpfuldiscussions.AnnaK¨albererandKemalAzunprovidedableresearchassistance.FinancialsupportfromtheGermanFederalMinistryofEducationandResearch,projectInterVal–Internetandvaluechains(01AK702A),isgratefullyacknowledged.Theauthorsareresponsibleforallremainingerrors. 1IntroductionTheretailsectorhasrecentlygainedattentionfromtworatherindependentresearchstreams:aliteraturedevotedtoproductivityeffectsofinformationtechnology(IT)ontheonehand,andaliteratureconcerningtheindustrialorganizationandregulationofretailmarketsontheotherhand.First,cross-countryproductivitystudiesattributelargepost-1995productivitygainsintheUnitedStatestoincreasedITusagemainlyinthedistributionsector.Someevenidentifya`retailrevolution'(Arketal.,2005;Nakamura,1999).MostEuropeancountries,however,havenotexperiencedsuchamanifestdevelop-mentinretailproductivity(TimmerandInklaar,2005).Second,industrialorganizationscholars–alarmedbythegrowthofgloballyorganisedretailerssuchasWal-Mart,Car-refourandTesco–analysetheemergenceoflarge-scaleretailing,particularlyintheformofhypermarkets,andtheassociatedincreaseinmarketpower(DobsonandWaterson,1999;EuropeanCommission,1999;FTC,2001;CompetitionCommission,2000).Acloserlookrevealsthatbothresearchstreamsarestronglyrelated.Mostprominently,insufcientproductivityperformancesareoftenascribedtoanti-competitiveretailregu-lation(McKinseyGlobalInstitute,2002;Scarpettaetal.,2002).Yet,despitearichparallelliteratureonthelinkbetweenretailregulationandemployment(BertrandandKramarz,2002,forexample),empiricalstudiesoftherelationshipbetweenretailregulationorcom-petitionontheonehandandretailinnovationorproductivityontheotherhandarerare.Weareonlyawareofstudiesbasedonrm-leveldata,forexampleFosteretal.(2002)andLevinetal.(1992).Firm-leveldata,however,typicallylackvariationintheregulatoryen-vironmentandhenceprovidelittleopportunitytoexamineimportantpolicyissues.Inthispaper,wetakeacross-countryperspectivetoanalysetherelationshipbetweenadistinctformofregulatedretailcompetition(hypermarketcompetition)anddiffusionofadistinctretailIT(barcodescanning).Weabstractfromproductivityconcerns:giventhatITinvestmentsareproductive,whyistheirintensitysodifferentacrossindustri-alisedcountries?UsingdataonthediffusionofbarcodescanninginretailoutletsoftenEuropeancountries,werstobtaincountry-specicestimatesforastandarddiffusionmodel.Resultsindicatesubstantialcross-countrydifferencesinthelong-rundiffusionlevelofbarcodescanning,corroboratingthendingsintheproductivityliterature.Then,2 weadddataonanumberofexplanatoryfactorsandassesstheireffectonITdiffusioninapooledestimation.OurparticularfocusisontheroleofhypermarketsinretailcompetitionandITdiffusion.Whilewepresentevidencethattheemergenceofhypermarketsrepresentsanincreaseintheintensityofcompetitioninretailing,resultsfromthepooledestimationsuggestthatsuchcompetitionhasreducedthelong-runITdiffusionlevel.Weidentifytwo–poten-tiallyindependent–effectsthatmaydrivethisresult.First,hypermarketcompetitioninducesaselectioneffect:hypermarketentryseemstocauseexitofpotentialITadopters,namelysmaller-sizedsupermarkets.Second,thereisanencouragementeffect:hypermar-ketspresumablyadoptbarcodescanningearlyandtherebydis-orencouragesubsequentadoptionsofrivalretailformats.Wealsoconsiderotherexplanatoryfactors.Inparticular,wendevidenceforaclas-sicsubstitutioneffect:whenwagesrise,diffusionofalabor-savingtechnologysuchasbarcodescanningismoreintense.Scaleandincomeeffectsarealsoimportantdetermi-nantsofITdiffusion,resultswhichmayexplaintheU.S.-EuropedivideinITusage.Wedonotndasignicantimpactofemploymentprotectionlegislation.Whilerelevantintheirownright,theaboveresultsmayalsohelppredicttheupcoming“revolutionatthecheckoutcounter”(Brown,1997),whichwillinvolvethereplacementofbarcodescan-ningbyradiofrequencyidentication(RFID).Therestofthepaperisorganisedasfollows.Inthefollowingsection,wepresentthedataoncheckoutbarcodescanning,ouranalyticalapproachtoestimatingadiffusionfunction,andresultsfromcountry-wiseestimations.Insection3,weintroduceanddis-cussanumberofdeterminantsofITdiffusionandprovidesomeprecursoryevidenceonhypermarketcompetition.Insection4,weproposeaneconometricspecicationforpooledestimationandreportrespectiveresults,includinganumberofrobustnesschecks.Section5concludesthepaper.3 2CheckoutbarcodescanningacrossEuropeWebeginwitharstglanceatthedataonbarcodescanninginEurope.IncontrasttotheUnitedStates,wheretherstretailoutletwasequippedwithabarcodescanneralreadyin1974(Nelson,2001,andtable2),diffusionofbarcodescanninginEuropedidnottakeoffbeforethe1980s.Until1997,thenationalmemberorganizationsoftheEuropeanArticleNumberingAssociation(EAN)collecteddataonthenumberofretailoutletswithscannerinstallations.Thesedataarepublishedfortheyears1981to1996intheyearlyreportsoftheEAN.Wecombinethisinformationwithdataonthetotalnumberofretailoutletsinordertoobtainameasureoftherelativeintensityofbarcodescanningwithinacountry.1Accordingly,gure1indicateshowmanyretailoutletsusebarcodescanning(in%,scat-terpoints),separatelyforsixofthetencountrieswestudy.Figure2intheappendixplotsthedatafortheremainingcountries.Linesrepresentttedvaluesfromacountry-specicestimationofalogisticgrowthfunction(discussedbelow).Whileallseriesaccordwithasigmoid-shapedcurvecommontodiffusionprocesses,countriesdiffersignicantlyintheintensityofbarcodescanning.Ourobjectiveistoshedsomelightonthefactorsun-derlyingthesedifferencesbymeansofapooledestimation.Yet,absolutecross-countrydifferencesshouldnotbetakentooliterally.Countriesmaydifferinmeasurementofthetotalnumberofretailoutlets.Forexample,somecoun-triesmayincludemobileoutlets(`streettraders'),othersnot.Inourpooledestimationdescribedinsection4,weaccountforsuchpotentialdifferences.Nevertheless,coun-trydifferencespertainifwerelatethenumberofbarcodescanningstorestopopulationinsteadofoutletgures.Fromtoday'sperspective,theseguresalsoappeartoexhibitimplausiblylowadoptionrates.Afterall,ourdailygroceryshoppingexperiencesuggeststhatbarcodescannersareubiquitous.Yet,noticethatweconsidernotonlygroceryretailing,butthewholeretailsector;whichincludestypesofretailerswhosimplyarenotpotentialusersofbarcode 1TheearliestEANreportavailableisthe1983report,whichalsogivesguresfor1981and1982formostcountries(orindicatesthattherewerenoscanningstoresbefore1983inaparticularcountry).TheEANreportsincludedataformorethanthetencountrieswestudy.Oursampleresultsfromotherdatalimitations.Section3.1presentsthedataanditssourcesinmoredetail.4 Figure1:Numberofbarcodescanningstores(in%,bycountry) scanning–forexampleowershops,repairshopsorbakeries.Furthermore,theEANdataconcernonlyxedscannersincheckoutcounters.Manysmallerretailersnowworkwithhand-heldormobilebarcodescanners.ThewaytheEANdataispresentedstronglysuggeststhatitdoesnotincludesuchhand-heldscanners:the1994reportattemptstodistinguishbetweenxedandhand-heldscanners,butmostcountries(includingthoseinoursample)onlyreportthetotalnumberofscanningstoresorrelativelylowguresforstoreswithhand-heldscanners.Apparently,theEANstoppeddatacollectionin1997notonlybecausebarcodescanningwasconsideredastandardtechnologybythen,but5 alsobecausetheincreasingnumberofscannertypeswashardtomanage.2Wearguebelowthat–sinceitisrestrictedtocheckoutbarcodescanning–formostcountriesinthesampleourdatasetissufcientforaneconometricstudy.2.1AnalyticalframeworkInlinewithmostempiricalstudiesofaggregatedataondiffusion,weemploythelogisticfunctionasanalyticaltool.3Weuseitinareduced-formmanner,althoughanumberoftheoreticalmodelsofnewtechnologyadoptiongeneratediffusionpatternsrepresentedbythelogisticorrelatedfunctions(Geroski,2000;Stoneman,2002).Thelogisticfunctioncapturesthetypicalsigmoidshapethroughthreeinterpretableparameters:St=S 1+exp(�b(t�t)),whereS=gNt.(1)Stindicatesthenumberofadopters(outletswithacheckoutbarcodescanner)attimet.Sisthepotentialnumberofadopters:the`ceiling'orsaturationleveltowhichStconverges.ItisafractiongofthetotalnumberofoutletsNt.Sincethelogisticfunctionissymmetric,Stequalshalfofitssaturationlevelatthecurve'sinectionpoint:thedatetatwhichthegrowthrateofthenumberofadoptersisnolongerincreasing.tindicatesthisinectionpointandishenceameasureforthetimingofadoption–itshiftstheS-curveforwardsorbackwardsonthetimeline.Tooseethis,considertk,themomentintimewhereasharekofthesaturationlevelisreached:S 1+exp(�b(tk�t))=kSortk=t�b�1log(k�1�1).Atk=.5,tk=t.Differentiatingequation1withrespecttotimeshowsthatcoefcientbisameasureforthespeedofadoption.ItgivesthegrowthrateofSt,relativetoitsdistance 2InatelephoneconversationwithGermanEANrepresentatives,weweretoldthatcollectionofthesedatabecameincreasinglydifcultduringthemid-1990s.3Withmicrodata,discretechoiceandhazardratemodelsarecommonlyused,forexampleseeKarshenasandStoneman(1993), Astebro(2004),andthereferencestherein.6 tothesaturationlevel:dSt dt1 St=bS�St S.ThegrowthrateofStattainsit'smaximum,b 2,attheinectionpointt=t.Ourapproachdiffersfromearlierdiffusionstudiesintworespects.First,thesestudiesoftenfollowtheseminalworkbyGriliches(1957)anduseanotherversionofequation1,whereSt=S 1+exp(�a�bt).Whereastheadvantageofthatapproachisthatitlendsitselfeasilytolog-linearisation,itsdisadvantageisthataiserroneouslyinterpretedasapuretimingindicator.Instead,equation1showsthata=btandhence`timing'estimatesforaresultingfromthetraditionalapproacharestronglycorrelatedwithrespective`speed'estimatesforb.Second,mostotherstudiesrelatethespeedortimingcoefcientsofequation1toindependentvariables(GruberandVerboven,2001,forexample),whereaswefocusonthesaturationlevelginourpooledestimation(section4).Actually,inalatecommentonhisearlierwork,ZviGrilichesproposedtodoexactlythat:“Addingparameterstothecurveitselforddlingwiththefunctionalformisnotanattractivealternative,inmyopinion.Whatonegainsintonelosesininterpretability.Instead,Iwouldnowrespecifythemodelsothattheceilingisitselfafunctionofeconomicvariablesthatchangeovertime.”(Griliches,1980,p.1463)2.2Country-wiseestimationWeobtainthettedvaluesshowningure1fromcountry-specicnonlinearleastsquares(NLS)estimationsofequation1withanadditivei.i.d.errorterm.Nt–thenumberofout-lets–iscountedinhundredssuchthatgindicatesthesaturationlevelasthepercentageofbarcodescanningstores.Table1providesmoredetailedresultsontheseestimations.Inlinewiththeproductivitystudiescitedbefore,cross-countrydifferencesseemtobemostpronouncedwithrespecttothesaturationlevelofITadoptionasmeasuredbyˆgi.Forexample,Austriaisestimatedtohaveabout24%ofoutletswithbarcodescanninginthelongrunbutItalyonly1%.Again,thesedifferencesarisenotonlyfromdifferentadoptionpatterns(St)butalsofromdifferentunderlyingretailmarketstructures(Nt).TheItalianretailmarket,forexample,isstillhighlysegmented,withmanysmallbutspecializedretailerssellinggoodsthatinothercountriesaresoldjointlybylargerretail-7 Table1:Estimatesfromcountry-wiseregressions CountryˆgiˆbiˆtiObservationsR2 Austria24.2a.50a1994.014.999Belgium16.0a.391994.112.999Denmark10.7.421992.115.992France10.7.411994.413.996Germany5.2a.411992.7a15.999Ireland1.3a.481992.7a16.998Italy1.1a.451992.015.986Netherlands7.6.31a1994.814.997Spain3.9a.391995.116.978UnitedKingdom15.4.411995.816.995 Cross-countryaverage9.6.421993.810 Parameterestimatesfromcountry-wiseNLSestimationofequation1.aCoefcientdifferssignicantlyfromcross-countryaverage(95%condencelevel,F-testbasedonasymptoticstandarderrors). ers.Cross-countrydifferenceswithrespecttotimingandspeedofdiffusionseemlesspronounced.Onlyintwocasesdoestimatesforbiandtidiffersignicantlyfromthecross-countryaverage(whichimplies20%growthinthenumberofbarcodescanningstoresaroundyear1994).Wethereforefocusonexplainingdifferencesingiwithajointregressionanalysisofthepanel.TheestimatedsaturationlevelforIrelandalsodeservesanote.IncontrasttotheItaliancase,wearerathersurprisedbythelowvalue,sinceIreland'sretailstructureismorecomparabletotheUK's(cf.table7intheappendix).AsIrelandhasdevelopedstronglythroughoutthe1990s,wepresumethatourdatacoveronlytheverybeginningofacorre-spondingdiffusionprocess.Inotherwords,ourseriesforIrelandmaylackitsinectionpoint,whichwouldleadtounreliableestimates(DebeckerandModis,1994).Wegetbacktothispointinsection4.Regardingallothercountriesinoursample,acomparisonoftheseestimatestoU.S.g-uressuggeststhatourdatashouldcoverasufcientpartofthediffusionprocess.U.S.trademagazinesstoppedreportingdetailedadoptionguresalreadyin1985.Table2,whichcompilesU.S.datafromvarioussources,showsthatthenumberofbarcodescan-ningstoresstartedtoriseslowlyinthemid-1970sandquicklywentupinthelate1970s8 andearly1980s.4Butatleastwithinthegroupoffoodretailers–themainusersofxedcheckoutscanners–U.S.growthappearstohavecometoahaltbytheendofthe1980s.WithEurope'stimelagbeingroughlyadecade,weshouldexpectdiffusionofxedscan-nerinstallationsinEuropetoslowdownbytheendofthe1990s,withtheseries'inectionpointsintheearly1990s,asreectedbythet-estimatesintheabovetable.Table2:Percentageofbarcodescanningstores:U.S.data,1974-1984 YearScanningstoresaOutletswithScanningstoresScanningfood(number)payrollb(%)stores(%)c 19746726940.00001197697744780.0000119802483738100.0000319825902784700.0000819849278831300.00011198859.7198957.7 Sources:aEuromonitor(1986),whichcitestradepublicationChainStoreAge.bU.S.BureauoftheCensus(1978,andlaterissues).cFoodMarketingInstitute(1989,1990),basedonasurveyofapprox.10,000foodretailers. 2.3FunctionalformandautocorrelationThelogisticisprobablythesimplestfunctionalformavailabletostudysigmoid-shapeddiffusionpatterns,however,itmaynotbethemostappropriate.Inparticular,erro-neouslyassumedsymmetryaroundtheinectionpointmaybiasestimates.Wethere-forere-estimatedcountry-specicsaturationlevelsusingtheGompertzfunction,whichissimilartothelogisticbutasymmetricaroundtheinectionpoint(Franses,2002,forexample).Forvecountries,theseestimatesandtherespectivetstatisticsdonotdiffermuchfromthosebasedonthelogisticfunction.Fortheothervecountries,however,theestimatedinectionpointliesbeyondtheyear2003,tstatisticsarepoorerthanorcomparabletothoseforthelogisticfunction,andtheestimatedsaturationlevelisverylarge–inthreecasesevenlargerthan100%(resultsavailableuponrequest).Weconcludethatthelogisticfunctionisthemoreappropriatefunctionalformforourdata. 4DasandMulligan(2004)arguethatfrequentvintagechangesofxedscannersbetween1975and1985affectU.S.diffusionestimates.Aspost-1980vintagesweresoldforrelativelylongtimeperiods,wedonotconsiderthisasignicantproblemforourdataset.9 Anotherissueinestimatinggrowthcurvesispotentialautocorrelation.FollowingtheprocedureproposedbyFranses(2002),inunreportedtestregressionswerejectthenullhypothesisofnoautocorrelationagainstthealternativeofAR(1)errorsonlyfortwocoun-tries.Yet,re-estimatingalogisticfunctionwithanAR(1)errortermforthesecountriesleadstoautocorrelationcoefcientswhicharenotsignicantlydifferentfromzero.WethereforeretaintheassumptionofanAR(0)errortermthroughouttherestofthepaper.3ExplainingcountrydifferencesInthissection,wepresentoursetofindependentvariablesandrelatethemtotheoreticalexplanationsfordifferencesinthediffusionofbarcodescanning.Inparticular,weassesstechnology-specicfactors(section3.2),employmentprotectionlegislation(section3.3),aswellashypermarketcompetitionandproductmarketregulation(section3.4).Relatedliteratureisdiscussedalongtheway.Wealsopresentanddiscusssomeprecursoryev-idence.Insection4,wepresentoureconometricspecicationforthepooledestimationandthecorrespondingresults.3.1RetailsectordataPubliclyavailableinformationontheretailsectorisscarce,evenonacountry-yearba-sis.Althoughwecompiledatafromvarioussources,variouslimitationsmakeusrestrictattentiontothe10countrieslistedintable1.Table3givesadescriptionofthemaininde-pendentvariablesusedinsection4.Formoredetailedcross-countrysummarystatistics,seetable7intheappendix.SourceofGDPandpopulationguresistheWorldBank(2003).Dataonthenumberofhypermarketsandthetotalnumberofretailoutletsarefromvariousissuesof”Retailtradeinternational”,apublicationbymarketresearchrmEuromonitor.ThemostrecentissueisEuromonitor(2002).Asameasurefortheseverityoflabormarketrestrictions,weuseversion1oftherevisedOECDindicatorofthestrictnessofemploymentprotectionlegislation(OECD,2004).Theindicatorofretailsalesvolume(VOL)isalsofromtheOECD.5TheretailWAGEindex 5ForItalyandSpain,thisindicatordoesnotcoverthewholesampleperiod.Forthesetwocountries,wethereforeconstructedacomparableindicatorbasedonEuromonitorandGGDCdata(seeappendix).10 isconstructedusingdatafromthe60-IndustryDatabaseoftheGroningenGrowthandDevelopmentCentre(GGDC).Pre-1990valuesforuniedGermanyforthevariablesVOLandWAGEwereconstructedbyapplyingpre-1990trendsforWesternGermanyto1990valuesforuniedGermany.Wealsohadtoreplacesomemissingvalueswithunivariateprocedures.AppendixAprovidesadetailedlistofalldatamanipulations.Table3:Summaryofindependentvariables LabelDescriptionSourceCross-countrymean1981/1996 OUTNo.ofretailoutletsEuromonitor,9361.8/7952.4(permn.inhabitants)WorldBankHYPNo.ofhypermarketsEuromonitor,6.8/13.3(permn.inhabitants)WorldBankEPLOECDindicatorofstrictnessofOECD2.5/2.2employmentprotectionlegislationWAGERetailhourlyrealwageGGDC,74.2/101.1(index1995=100)WorldBankGDPPercapitarealGDPWorldBank74.8/102.1(index1995=100)VOLRetailsalesvolumeOECD,85.7/101.3(index1995=100)Euromonitor Wewereunabletoobtaininformationonanumberoffactorsthatmayalsobeimportantinouranalysis,suchaspricesforscanningequipment,openinghours,theimportanceofmultinationalrms,oraveragestoresize.Aslongastheseomittedfactorsarerelativelytime-invariantorequalforthecountriesinoursample,ourresultsshouldnotbeaffectedsignicantly.Letusnowturntothetheoreticalpredictionsregardingtheeffectsoftheincludedfactorsonthediffusionofcheckoutbarcodescanning.3.2Technology-speciceffectsWhendecidingabouttheadoptionofanewtechnology,armtypicallycomparescostsandbenetsofadoptionatagivenpointintime(HallandKhan,2003).Forexample,het-erogeneityacrosspotentialadoptersregardingthesecostsorbenetsmaybeonereasonwhydiffusionofnewtechnologyisrarelyinstantaneous.Inourcase,theinstallationofabarcodescannerrepresentsamajorcapitalinvestmentthatbasicallyenablesaretailerto11 checkoutmoreretailitemsinlesslabortime.6FollowingthediscussionbyLevinetal.(1987,1992),anumberoffactorscanmakebarcodescanningmoreorlessvaluableindifferentcountries.7First,thenancialreturnstosuchacapitalinvestmentdependonfuturemarketcon-ditions.Sincereturn-on-investmentisquickeringrowingmarkets,retailerstherewilladoptmoreintenselythanretailersinstagnatingorcontractingmarkets.Inaddition,bar-codescanningmayintroduceorincreaseeconomiesofscaleinretailing.Inbothcases,weexpectadoptionintensitytoincreasewithmarketvolume(VOL).Second,barcodescanningislikelytoreducecustomerwaitingtimeatthecheckout.Customersinhigh-incomecountrieshaveahigheropportunitycostofwaiting.UsingpercapitaGDPasincomemeasure,weexpectdiffusionofbarcodescanningtoincreasewithGDP.No-ticethatinthisinterpretation,barcodescanningisaproduct-enhancinginnovation:itincreasesthequalityofretailingforthecostumer.Another,ratherclassicalinterpretationregardsbarcodescanningasaprocess-enhancinginnovationthatreducesthecostsofretailing.Mostprominently,barcodescanningmaybealabor-savingtechnologythatreducestotallabordemand.Inadditiontothisclassiccapital-laborsubstitutioneffect,barcodescanningmayallowretailerstosubstituteun-skilledforcostlyskilledlabor.Clerksatscannercheckoutsneedneitherknowpricesnorbeabletotypequickly.Inbothcasesofsubstitution,wethereforeexpectcountrieswithrisingretailwages–asmeasuredbyvariableWAGE–toinvestmoreinalabor-savingtechnologysuchasbarcodescanning.3.3EmploymentprotectionlegislationArelatedquestioniswhetherlabormarketrestraintshinderITdiffusion.Forexample,strictemploymentprotectionlegislation(EPL)mayprohibitretailersfromsubstitutingbarcodescannersforlaborasextensivelyasthetechnologymightallow.Accordingly,aconventionalwisdomhasbeenthatlessexiblelabormarkets(withstricterEPL)im- 6Clearly,barcodescanningalsofacilitatesotherpotentiallyproductivity-enhancingpractices,e.g.sophis-ticatedlogisticssystems(`efcientconsumerresponse',`categorymanagement');butthesesystemsdidnotdevelopbeforethemid-1990sandstillrepresent“untappedpotential”(Haberman,2001).7Levinetal.(1987,1992)studytheadoptionofbarcodescanninginU.S.retailing.Theyanalyserm-specicdatarelatingtotheearlyyearsofthetechnology(1974-1985).12 pedeITadoption(IMF,2001,forexample);withcorrespondingpolicyrecommendations.Yet,theliteratureontherelationshipbetweenlabormarketregulationsuchasEPLandinnovationprovidesmixedresults(BassaniniandErnst,2002,forareview).Insupportoftheconventionalview,GustandMarquez(2004)analyseapanelofcross-countrydataandndthatITinvestmentsarelowerincountrieswithhigherEPL.Incontrast,Koeniger(2005)ndsapositiveeffectofEPLoninnovativeactivity–atleastintheshort-andmedium-term–forapanelofOECDcountries.Healsoshowstheoreti-cally,thatEPLintheformofcollectivedismissalcostsmayincreaseinnovativeactivities.Accordingly,Agell(1999)arguesthatlabormarketregulations,inparticularEPL,neednotreduceinvestmentincentivesandproductiveefciency,astheyprovideinsuranceagainstadverseeconomicshocksorstructuralshiftsinlabordemand.HaucapandWey(2004)showthatlabormarketrigiditiescanincreasermsinvestmentincentiveswhentheytendtoenforceegalitarianwagestructures.3.4ProductmarketregulationandhypermarketcompetitionIntheindustrialorganizationliterature,retailmarketshavetypicallybeenregardedasmoreorlessperfectlycompetitive.Thisperceptionhasledscholarstoabstractfromtheretaillevelandconcentrateonthemanufacturers'side.Yet,fragmentedretailstructuresaremostoftenthedirectresultofentryrestrictions.Ingeneral,theserestrictionstendtofavorsmallretailingindowntownareasagainstlargescaleretailformatsasexempliedbyWal-Mart.Mostprominently,planningandconstructionrestrictionshavebeenusedinallEuropeancountriestobanlargeretailingformation;e.g.,bynotgrantingconstruc-tionpermissionsorbylimitingstoresize.SeeFainietal.(2004)forarecentaccountofretailrestrictionsintheUK,ItalyandGermany.BertrandandKramarz(2002)provideempiricalevidenceforFrance.TheserestrictionshavebeeneasedrstintheU.K.bytheThatchergovernmentandlaterinotherEuropeancountriesaswell.8Withthesedevelopments,hypermarketshavebecomeanintegralelementofEuropeanretailmarkets.Accordingtoawidelyuseddenition,hypermarketshaveaminimumsizeof2,500squaremeters,andsellbothfoodandnon-fooditems.Hypermarketsoften 8In1996,henceattheveryendofoursampleperiod,U.K.retailregulationturnedtowardsamorere-strictiveapproachfavoringcitycentres(HaskelandSadun,2005).13 locateinperipheralareaswhichareeasilyaccessiblebycar.InmostEuropeancountries,thehypermarketretailformatemergedinthe1970sand1980s,paralleltoanincreaseinmotorization.9Weclaimthatthenumberofhypermarketspercapita(HYP)isaninverseindicatorofen-tryrestrictions.Anincreasingnumberofhypermarketsisaresultfromlessrestrictiveen-tryregulations,andhenceaproxyforincreasingcompetitiveintensityduetoregulatorychange.Moreover,hypermarketsmayreectcompetitiveintensityonothergrounds.Theycanberegardedaslow-costcompetitorswhoexploitthecostbenetsofout-of-townlocations,sophisticatedlogistics,andeconomiesofscale(Basker,2004).Onemayalsoviewretailcompetitionascompetitionofretailchannelsorformats(Michael,1994;SmithandHay,2005).Inthatsense,theemergenceandgrowthofanewformat,likethehypermarket,intensiesretailcompetitionassuch.Table4presentssomeevidenceinsupportofourclaim.Sinceretailcompetitionessen-tiallyworksthroughentryandexitofrms(Fosteretal.,2002),theappearanceofcom-petitivehypermarketsshouldhaveledtoincreasedexitrates.Wethereforeregressthenumberofnon-hypermarketoutletsonthenumberofhypermarkets:(OUT-HYP)onHYP(allinpercapitaterms).Twocountriesinoursample–GermanyandDenmark–applyaslightlybroaderhypermarketdenitionwhichincludessuperstores(supermar-ketswithaoorspacebetween1,500and2,500squaremeters).Accordingly,weallowforadifferenteffectforthesetwocountries,thedifferencemeasuredbythecoefcientforD*HYP.Asexpected,anincreaseinthenumberofhypermarketsisestimatedtoleadtoade-creaseinthenumberofotherretailoutlets.Therstcolumnoftable4providesresultsundertheassumptionthattherearecountryxedeffectsbutnotimetrendsinthenum-berofoutletstimeseries.Thecoefcientforthestandardhypermarketdenition(HYP)impliesthatanadditionalhypermarketoutletpermillioninhabitantsisestimatedtoin-ducealmost197otherretailoutletspermillioninhabitantstoexitthemarket.Yet,theestimatedcoefcientforthebroaderdenitionemployedbyGermanyandDenmarkissignicantlypositive(-196.9+298.3).WeinterpretthisresultasanindicationthatGer- 9TheFrenchretailgroupCarrefourclaimstohaveinventedtheconcept.Itopeneditsrsthypermarketin1963nearParis,“withaoorspaceof2,500squaremeters,12checkoutsand400parkingspaces”(seewww.carrefour.com/english/groupecarrefour/annees60.jsp).14 Table4:Regressionresultsonhypermarketcompetition Dependentvariable:(OUT-HYP)(OUT-HYP)(OUT-HYP)(OUT-SUPHYP)IndependentCoefcientCoefcientCoefcientCoefcientvariable HYP-196.921a-196.921a-31.204(25.898)(26.881)(48.340)D*HYP298.300a103.574-68.124(60.212)(122.356)(121.573)SUPHYP-.972(1.052)Countryexcluded:GermanyGermanyCountryxedeffects:yesyesyesyesCountrytimetrend:nonoyesyesTimespan(max.)1980-20011980-20011980-20011980-2001 Observations215193215182R2.959.958.991.991 OLSestimates(countryxedeffectsandtimetrendsomitted).Standarderrorsinparentheses(aindicatessignicancewith95%condence).DisadummyvariableequaltooneforGermanyandDenmark,whouseadifferenthypermarketdenitionthantheothercountries.SUPHYPisthenumberofsuper-andhypermarketspermn.inhabitants(Source:Euromonitor;12obs.missing,seedataappendix). many'sseriesforoutletsandhypermarketsaresomewhatspecial.Theyseemheavilyaffectedbytworatherindependentpost-1990developmentsfollowingre-unication:(i)anoverallcatch-upinthenumberofoutletsinformerEastGermanyand(ii)theconstruc-tionoflargeretailsites–namelysuperstoresandhypermarkets–outsideofformerEastGermancities.Indeed,whenweexcludeGermanyfromthesample(secondcolumn),theestimateddifferencebetweentheDanish-Germanandthestandardhypermarketden-ition(D*HYP)isnolongersignicantlydifferentfromzero,suggestingthattheoverallhypermarketeffectisnegativeforDenmarkaswell.Theestimateforallothercountries(HYP)bydenitionremainsunchanged.Theinclusionofcountry-specictimetrendssomewhatimprovest,asmeasuredbyR2(thirdcolumn).Theaveragehypermarketeffectisthensmaller,butnegativeforbothhypermarketdenitions.Anadditionalhypermarketpermillioninhabitants(HYP)isestimatedtoinduceexitof31otherretailers(99forDenmark),althoughtheeffectisnotsignicant.Also,theDenmark-specicinsignicanceincolumntwosuggeststhathy-permarketsindeedimplymorecompetitivethreatthansmallermodernretailformatslikesuperstoresorsupermarkets,whicharepartiallyincludedinDenmark'shypermar-15 ketgures.Inordertoassessthishypothesisinmoredetail,welookedatresultswiththejointnumberofhyper-andsupermarkets(SUPHYP)asanalternativeregressor(fourthcolumn).10TheestimatedaverageeffectofSUPHYPonthenumberofotheroutletsisclosetozero.Weconcludethatthenumberofhypermarketsisavalidproxyfortheintensityofretailcompetitionandabetterindicatorthanthenumberofsuperstoresorsupermarkets.11Havingestablishedthatemerginghypermarketsrepresentmoreintensecompetition,whatshouldweexpectregardingtheireffectonthediffusionofbarcodescanning?Wearenotawareoftheoreticalorempiricalworkthatrelatesparticularlytoretailderegula-tionorhypermarketretailingandITdiffusion.Buttherelationshipbetweencompetitionandtechnologydiffusionhasbeenstudiedonamoregenerallevel(forareviewsee,inparticular,Stoneman,2002).G¨otz(1999)studiesthediffusionofnewtechnologyinamo-nopolisticallycompetitiveindustry.Hendsthatincreasedcompetitionoftenpromotesdiffusion.Incontrast,Boucekkineetal.(2004)studyadifferentiated-productsCournotduopolyandndaninverselyU-shapedrelationshipbetweencompetitionanddiffu-sion.Intheirmodel,anincreaseincompetition(adecreaseinproductdifferentiation)stimulatesdiffusiononlywhenproductsaresufcientlysubstitutable.Thecloselyrelatedliteratureontherelationshipbetweenmarketstructureandinnova-tionincentiveshasalsodeliveredcontradictoryresults.WhiletheSchumpeterian(1942)ideahasbeenthatthereisapositiverelationshipbetweeninnovationincentivesandconcentrationorlargerms,othershaveemphasisedthenegativeeffectsofmonopolypoweroninnovation.Borrowingfromtheparallelliteratureonmarketstructureandproductquality,onemayalsoclaimthattheinuenceofmarketstructureoninnovationisneutral,oringeneralambiguous(Swan,1970;Spence,1975).Empiricalresultsmirrorthistheoreticalambiguity(Geroski,2000;KarshenasandStone-man,1995,forreviews).Forexample,Levinetal.(1987,1992)ndthatretailersadoptearlierandthatintra-rmdiffusionofbarcodescanningisfasterinmarketsthatarelessconcentrated,buttheseeffectspartiallybaresignicance.Moreimportantly,aswear- 10Inthiscase,wedonothavetodistinguishbetweendenitions,sinceitdoesnotmatterhowstoresatthemarginbetweensuper-andhypermarketsareclassied.11Duetoanumberofproblemsassociatedwiththesupermarketdata(seeappendix),weregardthesesupermarket-specicresultsascomplementary,butrefrainfromusingrespectivedatainmoredetail.16 guedabove,concentrationmeasuresarenotnecessarilygoodproxiesforcompetitiveintensityinretailmarkets.Inourparticularcase,hypermarketcompetitionmayhavetwo–potentiallyindependent–effectsonITadoptionbyotherretailers.Ontheonehand,wendthathypermarketentryinducesexitofotherretailers.Iftheexitingretailerspredominantlybelongtothegroupof(potential)ITadopters,thisselectioneffectreducestheshareofadoptersinthegroupofremainingretailers.Ontheotherhand,hypermarketcompetitioncanhaveanencouragementeffectonthegroupofremainingretailers,forexamplewhenhypermarketentryleadsformernon-adopterstobecome(potential)adopters.WithaggregatedataonITdiffusion,wecanonlyidentifythejointimpactofthesetwoeffects,whichcanbepositiveornegative.Considerasimplenumericalexampleasillustration.Imagineacountrywith100retail-ers,50ofwhicharepotentialadoptersofbarcodescanning.Therearenohypermarketsyet.Whilebarcodescanningdiffuses,oneoftheretailersdecidestotransformintoahy-permarket,whichdrives10otherretailersoutofbusiness.Dependingonwhetherthese10quittingretailerswerepotentialadoptersornot,theselectioneffectofthehypermarketonlong-rundiffusionofbarcodescanningcanbenegativeorpositive.Incaseallquit-terswerenon-adopters,thelong-rundiffusionlevelofbarcodescanningincreasesfrom.5to.56(50outof90).Incasetheyhadbeen(potential)adopters,itreducesfrom.5to.44(40outof90).Moreover,anencouragementeffectofincreasedcompetitioncouldbethatsomeofthepreviousnon-adoptersbecomepotentialadoptersofbarcodescanning,whichraisesitslong-rundiffusionlevel.3.5BivariatecorrelationsForarstideaonhowthediscussedfactorsmightrelatetocross-countrydifferencesinbarcodescanning,ausefulapproachistheonepioneeredbyZviGriliches(1957).Herelatesgroup-specicparameterestimatestoindependentstatistics.Inthisvein,weas-sesshowthecountries'separatelyestimatedsaturationlevelscorrelatewithtrendsoftheproposedvariablesintherespectivetimeperiod.Table5liststhecorrelationcoefcients.17 Table5:Bivariatecorrelationsbetweenˆgiandindependentvariables Correlationbetweentrendcoefcientforlog(HYP)log(EPL)log(WAGE)log(GDP)log(VOL) andˆgi-.526.512.501.015.581 Basedonnineobservations(onepercountry,excludingIreland):ˆgiandtrendcoefcientfromcountry-wiseregressionoflog(independentvariable)ontime.Trendcoefcientsaresignicantwith95%condenceforallcountriesandvariablesexceptfortwocountrieswithvariableVOL. Allbivariatecorrelationcoefcientsareinlinewiththeabovetheoreticaldiscussions.Es-timatedsaturationlevelsarehigherincountrieswithlargergrowthofGDP,retailsalesvolume,retailwagesandemploymentprotection,andlowerincountrieswithlargerhypermarketgrowth.12Anegativehypermarketeffectissurprisinglyclearinthedata:between1981and1996,5outof10countrieshaveanaverageyearlygrowthintheper-capitanumberofhypermarketsbelow3%–asproxiedbyatrendcoefcientinaregres-sionoflog(HYP).Averageestimatedsaturationlevelis12.7%forthesecountries(Austria,Belgium,Denmark,Germany,Netherlands),butonly6.5%fortheothervecountriesthathadstrongerhypermarketgrowth.Yet,thesebivariatecorrelationsneitheraccountforcountry-specicxedeffectsingi,whichmayarisesolelyfromdifferencesincount-ingretailoutlets,norforyear-to-yearandmultivariatecorrelations.4PooledestimationInordertoassesstheeffectsoftheproposedvariablesandatthesametimeaccountfortime-invariantcountry-speciceffectsonthesaturationlevel,wepoolcountriestoestimateajointdiffusionfunction,inwhichweparameterisegasfollows:g=gi+Xitgx,(2) 12WhenweincludeIrelandincalculatingthesecorrelationcoefcients,onlythecoefcientforGDPchangesqualitatively,resultingfromIreland'scombinationofstrongGDPgrowthwithalowg-estimate.18 whereXitcontainsthevariablesHYP,EPL,VOL,GDP,WAGEandD*HYP.AsbeforeDisadummyvariableequaltooneforGermanyandDenmarktoaccountforthedifferenthypermarketdenition,andthenumberofoutlets(Nt)isgiveninhundreds.Subscripti=1,...,10indicatescountriesandt=1981,...,1996indicatesperiods.Thecoefcientsgiaccountfortime-invariantcountry-speciceffects,aswellasfortime-invariantcross-countrydifferencesinmeasurementoftheindependentvariables.Hence,gxestimatestheaveragemarginaleffectofvariablexonthecountry-specicsaturationlevel.Thespeedandtimingcoefcientsofequation1areallowedtovaryacrosscoun-tries,hencewespecifyb=biandt=ti.Inotherwords,weretainthefullexibilityofthecountry-wiseregressionsandusethespecicationofequation2toaskwhethertheindependentvariablescontainadditionalinformationregardingthediffusionofbar-codescanning.Afterinsertingequation2andanadditivei.i.d.errorterm,weestimateequation1byNLS.13UnreportedregressionsbasedonthefullsampleexhibitedconvergenceproblemsandledtolargeandunstableestimatesforIreland'scountry-specicestimates(gi,bi,ti).Weactu-allyndthisresultreafrmingintworespects.First,thisseemstoconrmthesuspicionthatthedataforIrelanddonotcoverasufcientlylargeportionofitsdiffusionofbar-codescanning.Second,itsuggeststhattheindependentvariablesdocontainadditionalinformation,sinceIreland-specicestimateswithoutthesevariablesspuriouslyappearedstable.Inwhatfollows,wethereforepresentestimationresultsexcludingIreland.Theindependentvariables'coefcientsarevirtuallyunaffected,comparedtoestimatesin-cludingIreland,butconvergenceissmootherandallcountry-specicestimatesarenowstable.4.1ResultsTherstcolumnofTable6presentstheresultsforourbaselinespecication(I).Inaddi-tion,wepresentresultsforthreealternativespecications:inspecicationsII,IIIandIVweexcludeGermany;inspecicationIII,wealsoexcludethevariablesEPLandD*HYP; 13Weusetheestimatesfromthecountry-wiseregressionsasinitialvaluesforcountry-speciceffects.Fortheindependentvariables'coefcients,wesetinitialvaluesequalto0.19 inspecicationIV,weexcludeD*HYPandaddaninteractiontermforthevariablesEPLandHYP.Table6:NLSestimationresults Dependentvariable:NumberofbarcodescanningstoresSpecication(I)(II)(III)(IV) HYP-1.852a-1.744a-1.756a-2.153a(.426)(.408)(.425)(.596)D*HYP6.668a1.837(2.505)(2.154)EPL-1.333-1.154-1.807(2.287)(2.108)(2.311)HYP*EPL.203(.179)WAGE.119a.123a.116a.126a(.032)(.031)(.028)(.033)GDP.394a.385a.408a.368a(.069)(.069)(.061)(.070)VOL.087b.072.062.101a(.048)(.047)(.047)(.048) Countryexcluded:IrelandIrelandIrelandIrelandGermanyGermanyGermanyTimespan(max.)1981-19961981-19961981-19961981-1996Observations130115115115Adj.R2.994.994.994.994RootMSE494.6504.2499.4499.8 Estimatesforgi,biandtiomitted.Asymptoticstandarderrorsinparentheses.Signicancelevels:a95%,b90% TheeffectsforvariablesWAGE,GDPandVOLvarylittleacrossspecicationsandareforthemostpartsignicantlyestimated.A10-pointincreaseintheretailwageindexises-timatedtoraisethesaturationpercentageofbarcodescanningstoresbyabout1.2pointsonaverage.A10-pointincreaseinrealGDPpercapitaandretailsalesvolumeraisesthesaturationpercentagebyabout4and1points,respectively.Allthreeresultsconrminitialexpectations.First,investmentinlabor-savingretailITcanbeinterpretedasareac-tiontochangesinlaborcosts.Second,income,scaleandreturns-to-investmenteffectsareimportant.AlthoughtheincomeeffectmeasuredbyGDPseemsmoreimportantthanthescaleeffectmeasuredbyVOL,botheffectsarehardtodistinguishempiricallysincethetwovariablesarehighlycorrelatedbydenition.Nevertheless,theseestimatedeffectsalreadycanexplainwhytheU.S.isaheadofmostEuropeancountrieswhenitcomesto20 ITdiffusionintheretailsectorandtheresultingproductivitygainsthroughoutthe1990s:strongoveralleconomicgrowthdrivenbyasurgeinconsumerspending.14Consistentwiththebivariatecorrelationfoundbefore,anincreaseinthenumberofhy-permarketsbyonepermillioninhabitantsisestimatedtodecreasethesaturationper-centageofbarcodescanningstoresbyabout2points.Inaggregateterms,hypermarketcompetitionthereforeseemstoreducelong-runITusageintheretailsector.Thequestionwhetherthishypermarketeffectworksbydiscouragingexistingretailersfromadoption(encouragementeffect)orratherbydrivingpotentialadoptersoutofthemarket(selectioneffect)isonewecannotaddresswiththedataathand.Wesuspectthatbotheffectsareatplay.Theimpactoftheselectioneffectmaybemoreimportant,however,sincehy-permarketsmainlycompetewithsupermarkets–themaingroupofpotentialadopters–andlesswithother,smallerretailers.Yet,thenegativeresultseemstoholdonlyforthestandardhypermarketdenition.InourbaselinespecicationI,theestimatefortheDanish/Germandenitionispositive(-1.9+6.7).Aswithourprecursoryresultsonhypermarketcompetition,thiseffectpre-sumablyarisesfromGermanre-unicationefforts,wherealargenumberofnewlybuiltretailoutletsinEastGermanyhavebeenequippedwithbarcodescannersfromthestart.Wethereforere-estimatedthemodelexcludingGermanyandndthatGermanyindeedseemstobeaspecialcase.Theestimateddifferencebetweenthebroadhypermarketdenitionandthestandardone,nowaDenmark-speciceffect,ismuchlowerandnotsignicantlydifferentfromzero.Accordingly,anestimationwhichignoresdifferenthy-permarketdenitionsbyexcludingtheinteractiontermD*HYP(specicationIII)leadstoanessentiallyunchangedhypermarketeffect,aslongasGermanyremainsexcluded.InspecicationIIIwealsoexcludetheEPLindicator,whoseeffecthasthecommonlyex-pectednegativesignbutisinsignicantinallestimations.Otherestimatesremainlargelythesame. 14ComparableOECDdatafortheretailvolumeindicatorVOLindicatethat,between1990and2000,U.S.retailvolumeincreasedbyabout67%,whereasitincreasedbyabout30%intheU.K.andbyabout7%inFrance.InGermany,retailvolumedecreasedbyabout1%between1990and2000.21 4.2RobustnessAllndingsremainqualitativelyunchangedinanumberofrobustnesschecks.First,complementaritiesbetweenlaborandproductmarketconditionsmayaffectourresults.WethereforelookedatresultsincludinganinteractiontermHYP*EPL(specicationIVintable6)orEPL*WAGE(resultsomitted).Inbothcases,theinteractiontermcoefcientsareratherclosetoandnotsignicantlydifferentfromzero,whilesomeotherestimatesslightlychangeinmagnitudeandprecisionbutareotherwiseunaffected.Apotentialsourceofendogeneitybiasisthepresumptionthateverynewhypermarketbuiltfromthemid-1980sincreasesthenumberofscanningoutletsbyone.Althoughnotnecessarily,sincehypermarketsoperatedlongbeforetheintroductionofbarcodescan-ningandhencethetechnologymaynotbeascrucialforthemasitmightappearfromtoday'sperspective.Inanycase,thenegativeestimatesintable6alreadysuggestthatthisendogeneitybiascannotbeveryinuential.Bydeductingthenumberofhyper-marketsfromboththenumberofbarcodescanningstoresandthenumberofoutlets,itisneverthelesspossibletofocusontheeffectofhypermarketcompetitionontheadop-tionofbarcodescanningbyallotherretailers.ThecorrespondingunreportedresultsforspecicationsItoIVarevirtuallyidenticaltothoseintable6.Onemayalsosuspectthatthereareeffectsdrivingreversecausality,namelythatbarcodescanningleadstoanincreaseinaveragestoresizeandeventuallytomore“superstores”orhypermarkets(Holmes,2001).Yet,thefactsthat(i)hypermarketsexistedlongbeforebarcodescanningwasintroducedand(ii)Holmes(2001)modelpredictsapositivecorre-lationwhilewendanegativeoneleadustobelievethatreversecausalityisnotasevereissueinourcase.Anotherpotentialsourceoferroraretheimplicitassumptionsinourmethodtoconstructtimeseriesforthetotalnumberofretailoutlets(seedataappendix).Wethereforeesti-matedspecicationsItoIVwithacountry'spopulation(inmillions)replacingthenum-berofoutletsinequation1.Table8intheappendixpresentsthecorrespondingresults.ForspecicationsIItoIV,allvariablesyieldestimateswiththesamequalitativeeffectsonthelong-runnumberofbarcodescanningstorespercapita;exceptEPL,whosecoef-cientschangesignbutareagaininsignicant.OnlyforspecicationI,whichincludes22 thespecialcaseofGermany,someresultsdiffer.Weinferthatourresultsarenotcruciallyaffectedbythedatamanipulationsthatwerenecessarytoobtainaworkabletimeseriesforthenumberofretailoutlets.Finally,ourconclusionsregardingtheeffectofEPLmaybepremature.Givensubstan-tialmanipulationsnecessarytoobtainacompletetimeseries(seeappendix),andothermeasurementproblemsassociatedwiththeOECDEPLindex(BlanchardandWolfers,2000),therearereasonstodoubtthevalidityoftheindicatorused.Inordertocross-checkresults,wereplacedtheEPLindicatorwithvariablesconstructedusingdatafromtheSocialReformsDatabaseoftheFondazioneRodolfoDeBenedetti.Amongstotherin-formation,thisdatabaseprovidesalistofEPLreformsforallcountriesinoursample,andclassiesthemasexibility-increasingor-decreasing.15Fromthisinformation,weconstructedtwotimeseriesonthecumulativenumberofEPLreformsforeachcountry.WhenreplacedforourinitialEPLindicatorinspecicationII,thesevariablesalsoyieldinsignicantresults(availableuponrequest).5ConclusionBarcodescanning,acriticalinformationtechnologyintheretailsector,hasdiffusedtodifferentsaturationlevelsacrossEuropeancountries.Econometricresultsbasedondatafromvarioussourcessuggestthat,asexpected,thisretailtechnologydiffusesmorein-tenselyincountrieswithlargeandgrowingretailsectorsandeconomies.ItisthereforenotsurprisingthattheUnitedStatesisaheadofmostEuropeancountrieswhenitcomestoITdiffusionintheretailsectorandtheresultingproductivitygainsinthe1990s.Withrespecttoanupcoming`retailrevolution'thatreliesonRFIDtechnology,ourresultsleadustoexpectstrongerRFIDdiffusionincountriesthatallowretailerstoexploitscaleef-fects.Inlinewithclassictheory,wealsondthatraisinglaborcostsinduceretailerstosubstitutebarcodescannersforlabor.Incontrast,wedonotndemploymentprotectionlegislationtosignicantlyimpactretailITdiffusion. 15Thereisalsoaclassicationintomarginalandstructuralreforms,butasmostlistedreformsaremarginalwedidnotusethisdistinction.Seehttp://www.frdb.org/documentazione/scheda.php?id=55&doc_pk=9027formoredetail.23 Ourresultsconcerningtheimpactofcompetitiveintensityseemtodifferwithconven-tionalwisdom.Wendthattheemergenceofhypermarketsrepresentsincreasedretailcompetitionandthatsuchcompetitionreduceslong-runretailITdiffusion.Thiseffect,whichisrobustinavarietyofspecications,hastwopotentialexplanations.First,hyper-marketcompetitionmayverywellcauseexitofpotentialITadopters,namelysmaller-sizedsupermarkets.Second,hypermarkets–whicharemostlikelytoadoptbarcodescanningearly–discouragesubsequentadoptionsofrivalretailformats.Overall,theseresultssuggestthatliberalisationofretailmarketentryandtheassociatedemergenceofhypermarketsdeepensretailsegmentationsuchthathypermarketsontheonehandandsmalldowntownretailing(includingshoppingmallretailing)prevail.Incontrast,inter-mediateretailformats–inparticularmedium-sizedsupermarkets–arelikelytosufferfrommarketliberalization.Oneshouldnotice,however,thattheproductivityimplicationsofthesendingsarenotevident:dependingonhowmuchretailvolumegoesthroughbarcodescanningretailers,ITproductivitymayincreaseeventhoughaggregateITintensitydecreases.Also,ourdataarenotdirectlycomparabletomeasuresofretailITinvestments,sincetheycountthenumberofbarcodescanningstores,notthenumberofscannerinstallations.Inourdata,asmallersupermarketwith,say,onescannercheckouthasthesameweightasalargeronewithmultiplescannercheckouts.FurtherresearchmayincludemeasuresofforeigndirectinvestmentsinordertoassesstheroleoflargemultinationalretailrmsinITdiffusion.Givendataontheemergenceofone-stop-shopping(e.g.,motorizationanddemographics),itmayalsobepossibletoaddressthepotentialendogeneityofhypermarketdevelopmentmorerigorously.Finally,thepresentresultsarebasedonarathersmallnumberofobservations.Itshouldbeinterestingtoincludemorecountriesandexplanatoryvariables.Reviewingoureffortstoputtogetherthepresentdataset,wehoweverfearsuchataskismoredemandingthanitseemsatrstsight.24 ADataappendixFigure2:Numberofbarcodescanningstores(in%,bycountry) Scanningoutlets.DatasourcearethestatisticalappendicesoftheEuropeanArticleNumberingAssociation's(EAN)yearlyreportsfor1983through1997(availableatwww.ean-int.org).Theygivethenumberofbarcodescanningoutletspercountryfortheyears1981through1996,althoughthisperiodisnotentirelycoveredforallcountries.DataforBelgiumincludeLuxembourg.InthecasesofAustria,Denmark,IrelandandSpain,itwasclearfromthetextinthecountrysectionsofthereportsthatthenumberofscanningoutletswaszerobefore1984,althoughitisreportedasmissinginthere-spectiveappendixtable.MissingobservationsintheseriesforItaly(1982)andIreland(1989)werereplacedbylinearinterpolationusingadjacentobservations.Dataforthelastyearsofobservation,1995and/or1996,seemedinconsistentwithdataforprecedingyearsinthecasesofAustria,ItalyandtheNetherlands.Theyindicatedeitheradecreaseofthethenumberofscanningoutlets(Netherlands,1995;Italy,1996)oranoverlystrongincrease(Austria,1995and1996).16Inatelephoneinterview,weweretoldbyGermanEANrepresentativesthatcollectionofthesedatabecameincreasinglydifcultduringthemid-1990s,asbarcodescannersbecamestandardtechnology,differenttypesofscanner 16Accordingtotheoriginalgures,thenumberofscanningoutletsinAustriarosefrom4,670to13,827(henceby300%)between1994and1995.InrelationtothetotalnumberofretailoutletsinAustria,whichEuromonitorInternationalestimatesat38,546for1995,thiswouldimplyanincreaseinpenetrationfrom12to36%inoneyear.Webelievethatthepost-1994guresrefertothenumberofscannercheckoutsratherthanthenumberofscanningoutlets.25 Table7:Detailedsummaryofvariables VariableOUTHYPEPLWAGEGDPVOLSUP CountryAustria4762.529.92.287.994.095.2696.1356.215.6.245.337.631.8664.5Belgium4653.97.52.887.893.994.8166.2677.11.7.937.236.642.9115.6Germany4652.722.72.991.993.994.2102.91937.811.6.937.031.527.251.9Denmark6841.316.01.890.394.796.5216.3896.88.0.749.034.723.2102.7Spain20730.44.03.493.792.8100.0192.28573.06.6.843.146.027.3182.7France7159.515.32.893.695.598.486.72380.611.7.330.133.317.152.6Ireland9177.15.3.992.791.1100.035.1707.913.1.168.399.074.524.4Italy16139.83.93.694.793.499.0104.36221.59.11.917.033.725.0139.0Nether-5469.42.42.595.894.3101.5276.2lands652.01.5.624.040.225.076.9United6740.13.4.591.093.193.6101.3Kingdom2564.84.6.240.844.861.525.9 SUPisthenumberofsupermarketspermn.inhabitants(Source:Euromonitor).Seetable3forafulldescriptionoftheothervariables.Country-specicmeansintherstline,inthesecondlinethedifferencebetweenthemaximumandtheminimumvalueobservedintherespectiveseries(range). wereintroducedandsmallrmswereunwillingtoanswerquestionnaires.Apparentlyforthesereasons,theEANstoppedcollectingthesedataafter1997.Weinterpretedtheinconsistentpost-1995dataforAustria,ItalyandtheNetherlandsasarstsignofthesedifcultiesandthereforeexcludedthemfromoursample.Retailoutlets.Dataonthenumberofretailoutletsweretakenfromvariousissuesof”Retailtradeinternational”,apublicationbymarketresearcherEuromonitorInternational.Everyissueprovidescountry-specicdataontheretailsector,mostlycollectedfromof-cialandindustrysources(suchastrademagazines)forveconsecutiveyears.ThelatestavailableissueisEuromonitor(2002),whichcoverstheyears1997-2001.However,earlierissuescoveringthelate1970sandthe1980sonlyprovideguresforfewsingleyears.Wethereforehadtoreplacemissingvaluesbyinterpolationforthefollowingobservations:Austria,1981,1982,1984-1987;Belgium,1981,1983,1985,1986,1988,1989;Denmark,26 1982-1984,1986;France,1982,1983,1985-1987;Germany,1981-1983,1987,1989,1991;Ire-land,1981-1987,1989-1991;Italy,1982-194;Netherlands,1981,1983,1985,1986;Spain,1981-1984,1988;UnitedKingdom,1981,1983,1985,1989,1991.Foreverycountrycov-ered,notalltimeseriespublishedinthevariousEuromonitorissueswereconsistentinoverlappingyearsofcoverage.Mostprobably,thisisduetovarying(non-)inclusionofgasstations,automobiledealersandmobileretailoutlets.Wethereforeusedthemostrecentavailableseries(Euromonitor,2002)forabsolutevaluesandprojectedthisseriesbackto1981usingthetrendsfromprecedingseries.17Whenevertwoissuesgavein-consistentguresforthesameyear,weusedthegurefromthemorerecentpublica-tion.Thisapproachentailstheimplicitassumptionthattheoutletshareofwhatevertypeofretailformatincluded(notincluded)intheEuromonitor(2002)guresbutnotincluded(included)intheearliergureshasremainedconstantovertime.Then,ourconstructedtimeseriesreectchangesinthenumberofretailoutletsaccurately,anddif-ferencesacrosscountriesregardingtheinclusionofacertainretailformatinthetimeseriesareaccountedforinestimationbythecountry-speciccoefcients.Hypermarkets.DataonthenumberofhypermarketswerealsotakenfromtheEuromon-itorpublicationscitedabove.Thefollowingmissingvaluesforsingleyearshavebeenreplacedbyinterpolation:Belgium,1982,1983;Denmark,1984;Ireland,1991;Italy,1985;UnitedKingdom,1983.MissingvaluesforItaly,1987and1988,andtheUnitedKing-dom,1981,werereplacedbydatafromtheEuropeanCommission(1997,p.21-17,ta-ble9),whichareconsistentwiththeEuromonitordataforsubsequentyears.InthecasesofAustria,Belgium,Denmark,theNetherlandsandtheUnitedKingdom,thetimeseriespublishedinthevariousEuromonitorissueswerenotalwaysconsistentinoverlappingyearsofcoverage.Thismaybeduetochangesinoriginalindustrysources.Inthesecases,theseriesfromEuromonitor(2002)wasprojectedback,inasimilarwaythande-scribedfortheoutletsseries,usingtrendsfromprecedingseries.InthecasesofDenmarkandGermany,theguresbaseonadifferenthypermarketdenition,whichconsidersashypermarketsfoodretailerswhoalsosellnon-fooditems(asinthestandarddenition)andhavemorethan1,500squaremetersofretailspace(asopposedto2,500squaremetersinthestandarddenition). 17InthecasesofAustriaandFrance,theseriescoveringthelate1980sdidnotoverlapwiththesubsequentseries.Wethereforeextrapolatedtheearlierseries,usinginformationfor1985-1988,toobtainavaluefor1989whichwecouldcomparewiththe1989valueofthefollowingseries.27 Supermarkets.TheEuromonitorpublicationsalsoincludedataonthenumberofsuper-markets,butwithmanymissingvalues.Moreover,supermarketdenitionsarenotascomparableacrosscountriesashypermarketdenitions.Forexample,Austriadenesassupermarketsstoreswitharetailspacebetween400and1000squaremeters,whereasFranceandSpaindenesassupermarketsstoreswitharetailspacebetween400and2,500squaremetersbutSpainonlycountssuchstoresassupermarketsthatadditionallyhaveatleast5checkouts(Euromonitor,1989).Wethereforeusedrespectivedataonlyinanauxiliaryregression(table4).Before,wereplacedthefollowingmissingvaluesforsingleyearsbyinterpolation:Austria,1984;Belgium,1981-1982,1987-1988,1990-1991;Denmark,1984,1990-1991;Ireland,1980-1987,1989,1991-1992;Netherlands,1988;Spain,1989-1990.DatafortheNetherlands,1980-1986,andtheUnitedKingdom,1980-1984,re-mainmissing.InallcasesexceptDenmarkandGermany,thetimeseriespublishedinthevariousEuromonitorissueswerenotalwaysconsistentinoverlappingyearsofcoverage.Inthesecases,theseriesfromEuromonitor(2002)wasprojectedback,inasimilarwaythandescribedfortheoutletsseries,usingtrendsfromprecedingseries.Employmentprotectionlegislation.TherevisedOECDindicatorforemploymentpro-tectionlegislation(EPL)ispublishedbytheOECD(2004)forthreemomentsintime:the`late1980s'(1989),the`late1990s'(1998),and2003.WefollowedBlanchardandWolfers(2000)inordertoconstructatimeseriesfromthesedata:fortheyears1990-1997,were-placedmissingvaluesbylinearinterpolationandweassumedthatEPLhasnotchangedsignicantlythroughoutthe1980s.ThefactthattheSocialReformsDatabaseoftheFon-dazioneRodolfoDeBenedettilistsonlythreemarginalEPLreformspriorto1989–twoforFrance,1986,andoneforItaly,1987–reconrmsthisassertion.Salesvolume.TheOECDindicatorofthevolumeofretailsalesisnotavailableforSpain,1981-1990andforItaly,1981-1985.WeconstructedacomparableindicatorusingEu-romonitordataonretailsalesanddatafromtheGGDC60-industrydatabaseonretailvalueaddeddeators.ForItalyandSpain,weusedthisindicatorinsteadoftheOECDindicatorforthewholesampleperiod.Wagesandhoursworked.TheGGDCdatabasecontainsinformationonthenumberofpersonsemployed,annualhoursworkedandlaborcompensationperemployee,andavaluedeatorfortheretailsector.Unfortunately,thenumberofretailemployees–which28 excludesself-employedpersonsorfamilymembers–isnotavailableforallcountries.Thetotalnumberofhoursworkedaswellasourindexofthedeatedaveragehourlywagearethereforebasedonthenumberofpersonsengaged.BResultsforrobustnesschecksTable8:Estimationresultsusingpopulationasdenominator Dependentvariable:NumberofbarcodescanningstoresSpecication(I)(II)(III)(IV) HYP-101.797b-98.111a-96.814a-82.775a(56.681)(34.648)(34.156)(38.829)D*HYP143.195a142.682(68.168)(146.656)EPL-321.143a50.255212.352(134.152)(216.056)(320.688)HYP*EPL-30.843(26.711)WAGE-3.847b4.886b5.011b4.800(2.196)(2.873)(2.766)(3.047)GDP-2.69214.088a13.117a14.299b(3.710)(6.456)(4.881)(7.280)VOL9.592a10.914b11.837a10.664(4.179)(6.166)(5.371)(6.956) Countryexcluded:IrelandIrelandIrelandIrelandGermanyGermanyGermanyTimespan(max.)1981-19961981-19961981-19961981-1996Observations130115115115Adj.R2.993.993.993.993RootMSE550.7545.6540.3544.1 NLSestimatesofequation1withequation2inserted,whereNtdenotespopulation(inmillions)insteadofthenumberofretailoutlets.Estimatesforgi,biandtiomitted.Asymptoticstandarderrorsinparentheses.Signicancelevels:a95%,b90% 29 ReferencesAgell,Jonas(1999),Onthebenetsfromrigidlabourmarkets:norms,marketfailures,andsocialinsurance,TheEconomicJournal,109(127),pp.F143–64.Ark,Bartvan,RobertH.McGuckinandMatthewSpiegelman(2005),Theretailrevolu-tion–canEuropematchU.S.productivityperformance?,Report1358,TheConferenceBoard. Astebro,Thomas(2004),Sunkcostsandthedepthandprobabilityoftechnologyadop-tion,JournalofIndustrialEconomics,52(3),pp.381–399.Basker,Emek(2004),Sellingacheapermousetrap:Entryandcompetitionintheretailsector,WorkingPapers0401,DepartmentofEconomics,UniversityofMissouri.Bassanini,AndreaandEkkehardErnst(2002),Labourmarketinstitutions,productmar-ketregulation,andinnovation:Cross-countryevidence,WorkingPaper316,OECDEconomicsDepartment.Bertrand,MarianneandFrancisKramarz(2002),Doesentryregulationhinderjobcre-ation?EvidencefromtheFrenchretailindustry,TheQuarterlyJournalofEconomics,117(4),pp.1369–1413.Blanchard,OlivierandJustinWolfers(2000),TheroleofshocksandinstitutionsintheriseofEuropeanunemployment:Theaggregateevidence,TheEconomicJournal,110(462),pp.C1–33.Boucekkine,Raouf,OmarLicandroandAntonioMinniti(2004),Adoptionanddiffusionofcostreducinginnovations:Cournotcompetitioninduopoly,mimeo.Brown,StephenA.(1997),Revolutionatthecheckoutcounter(HarvardUniv.Press,Cam-bridge,Mass.).CompetitionCommission(2000),Supermarkets:areportonthesupplyofgroceriesfrommultiplestoresintheUnitedKingdom,ReportCm-4842.Das,NilotpalandJamesG.Mulligan(2004),Vintageeffectsandthediffusionoftime-savingtechnologicalinnovations:TheadoptionofopticalscannersbyU.S.supermar-kets,WorkingPaper04-06,UniversityofDelaware,DepartmentofEconomics.30 Debecker,AlainandTheodoreModis(1994),DeterminationoftheuncertaintiesinS-curvelogisticts,TechnologicalForecasting&SocialChange,46,pp.153–173.Dobson,PaulandMichaelWaterson(1999),Retailerpower:howregulatorsshouldre-spondtogreaterconcentrationinretailing,EconomicPolicy,14(28),pp.133–164.Euromonitor(1989),RetailTradeInternational:1989Edition(EuromonitorInternationalPlc.,London).—(2002),RetailTradeInternational:2002Edition(EuromonitorInternationalPlc.,London).EuropeanCommission(1997),PanoramaofEUIndustry(EuropeanCommission).—(1999),BuyingpoweranditsimpactoncompetitioninthefoodretaildistributionsectoroftheEuropeanUnion,ReportproducedfortheEuropeanCommission,DGIV,Brussels.Faini,Ricardo,JonathanHaskel,GiorgioB.Navaretti,CarloScarpaandChristianWey(2004),ContrastingEuropesdecline:doproductmarketreformshelp?,mimeo.FoodMarketingInstitute(1989),Thefoodmarketingindustryspeaks,1989.Detailedtabula-tions(FoodMarketingInstitute).—(1990),Thefoodmarketingindustryspeaks,1990.Detailedtabulations(FoodMarketingInstitute).Foster,Lucia,JohnHaltiwangerandC.J.Krizan(2002),Thelinkbetweenaggregateandmicroproductivitygrowth:Evidencefromretailtrade,NBERWorkingPaperSeries,No.9120.Franses,PhilipHans(2002),Testingforresidualautocorrelationingrowthcurvemodels,TechnologicalForecasting&SocialChange,69,pp.195–204.FTC(2001),ReportontheFederalTradeCommissionworkshoponslottingallowancesandothermarketingpracticesinthegroceryindustry,ReportbytheFederalTradeCommissionstaff,WashingtonD.C.Geroski,PaulA.(2000),Modelsoftechnologydiffusion,ResearchPolicy,29,pp.603–625.31 G¨otz,Georg(1999),Monopolisticcompetitionandthediffusionofnewtechnology,RANDJournalofEconomics,30(4),pp.679–693.Griliches,Zvi(1957),Hybridcorn:anexplorationintheeconomicsoftechnologicalchange,Econometrica,25(4),pp.501–522.—(1980),Hybridcornrevisited:areply,Econometrica,48(6),pp.1463–1465.Gruber,HaraldandFrankVerboven(2001),ThediffusionofmobiletelecommunicationsintheEuropeanUnion,EuropeanEconomicReview,45,pp.577–588.Gust,ChristopherJ.andJamieMarquez(2004),Internationalcomparisonsofproductiv-itygrowth:Theroleofinformationtechnologyandregulatorypractices,LabourEco-nomics,11,pp.33–58.Haberman,AlanL.(2001),Appendix:17billionreasonstosaythanks,in:AlanL.Haber-man(ed.),Twenty-veyearsbehindbars,pp.113–151(HarvardUniversityPress),Pro-ceedingsoftheTwenty-fthanniversaryoftheU.P.C.attheSmithsonianInstitution,September30,1999.Hall,BronwynH.andBeethikaKhan(2003),Adoptionofnewtechnology,in:DerekC.Jones(ed.),NewEconomyHandbook(AcademicPress).Haskel,JonathanandRaffaellaSadun(2005),Entry,exitandlabourproductivityinU.K.retailing:evidencefrommicrodata,mimeo.Haucap,JustusandChristianWey(2004),Unionisationstructuresandinnovationincen-tives,TheEconomicJournal,114(127),pp.C149–C165.Holmes,ThomasJ(2001),Barcodesleadtofrequentdeliveriesandsuperstores,RANDJournalofEconomics,32(4),pp.708–25.IMF(2001),TheInformationTechnologyRevolution,chapterIII,pp.105–144(InternationalMonetaryFund).Karshenas,MassoudandPaulL.Stoneman(1993),Rank,stock,order,andepidemicef-fectsinthediffusionofnewprocesstechnologies:Anempiricalmodel,RANDJournalofEconomics,24(4),pp.503–528.32 —(1995),Technologicaldiffusion,in:PaulL.Stoneman(ed.),Handbookoftheeconomicsofinnovationandtechnologicalchange,pp.265–297(Blackwell).Koeniger,Winfried(2005),Dismissalcostsandinnovation,EconomicsLetters,88(1),pp.79–84.Levin,SharonG.,StanfordL.LevinandJohnB.Meisel(1987),Adynamicanalysisoftheadoptionofanewtechnology:thecaseofopticalscanners,TheReviewofEconomicsandStatistics,69,pp.12–17.—(1992),Marketstructure,uncertainty,andintrarmdiffusion:Thecaseofopticalscan-nersingrocerystores,TheReviewofEconomicsandStatistics,74(2),pp.345–50.McKinseyGlobalInstitute(2002),HowITEnablesProductivityGrowth,MGIReports(McK-insey&Company).Michael,StevenC.(1994),Competitioninorganizationalform:Mailorderversusretailstores,1910-1940,JournalofEconomicBehavior&Organization,23(3),pp.269–286.Nakamura,LeonardI.(1999),Themeasurementofretailoutputandtheretailrevolution,CanadianJournalofEconomics,32(2),pp.408–425.Nelson,JohnE.(2001),Scanning'ssilvercelebration,in:AlanL.Haberman(ed.),Twenty-veyearsbehindbars,pp.25–33(HarvardUniversityPress),ProceedingsoftheTwenty-fthanniversaryoftheU.P.C.attheSmithsonianInstitution,September30,1999.OECD(2004),Employmentoutlook(OrganisationforEconomicCo-operationandDevel-opment).Scarpetta,Stefano,PhilipHemmings,ThierryTresselandJaejoonWoo(2002),Theroleofpolicyandinstitutionsforproductivityandrmdynamics:evidencefrommicroandindustrydata,Technicalreport,OECDEconomicsDepartment.Schumpeter,JosephA.(1942),Capitalism,socialismanddemocracy(Routledge,LondonandNewYork),5thedition,reprinted1968.Smith,HowardandDonaldHay(2005),Streets,malls,andsupermarkets,JournalofEco-nomics&ManagementStrategy,14(1),pp.29–59.33 Spence,A.Michael(1975),Monopoly,quality,andregulation,BellJournalofEconomics,6(2),pp.417–429.Stoneman,Paul(2002),Theeconomicsoftechnologicaldiffusion(Blackwell,Oxford).Swan,PeterL..(1970),Marketstructureandtechnologicalprogress:Theinuenceofmonopolyonproductinnovation,TheQuarterlyJournalofEconomics,84(4),pp.627–38.Timmer,M.andR.Inklaar(2005),ProductivitydifferentialsintheU.S.andEUdistrib-utivetradesector:Statisticalmythorreality?,GGDCResearchMemorandumGD-76,GroningenGrowthandDevelopmentCentre,UniversityofGroningen.WorldBank(2003),WorlddevelopmentindicatorsCD-ROM(WorldBank).34