VU Research Portal
Shopping externalities and retail concentration
Koster, Hans R.A.; Pasidis, Ilias; van Ommeren, Jos
published in
Journal of Urban Economics
2019
DOI (link to publisher)
10.1016/j.jue.2019.103194
document version
Publisher's PDF, also known as Version of record
document license
Article 25fa Dutch Copyright Act
Link to publication in VU Research Portal
citation for published version (APA)
Koster, H. R. A., Pasidis, I., & van Ommeren, J. (2019). Shopping externalities and retail concentration:
Evidence from dutch shopping streets. Journal of Urban Economics, 114, 1-29. [103194].
https://doi.org/10.1016/j.jue.2019.103194
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal ?
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
E-mail address:
vuresearchportal.ub@vu.nl
ContentslistsavailableatScienceDirect
Journal
of
Urban
Economics
journalhomepage:www.elsevier.com/locate/jue
Shopping
externalities
and
retail
concentration:
Evidence
from
dutch
shopping
streets
☆
Hans
R.A.
Koster
a,1,∗,
Ilias
Pasidis
b,
Jos
van
Ommeren
a,2a Department of Spatial Economics, Vrije Universiteit Amsterdam, De Boelelaan 1105, HV 1081 Amsterdam, the Netherlands b Barcelona Institute of Economics, John Maynard Keynes, 1-11, 08034, Barcelona, Spain
a
r
t
i
c
l
e
i
n
f
o
JEL classification: R30 R33 L81 Keywords: Retail Shopping externalities Rents Vacancies Agglomeration economies Footfalla
b
s
t
r
a
c
t
Whydoshopsclusterinshoppingstreets?Wearguethatretailfirmsbenefitfromshoppingexternalities.We identifytheseexternalitiesforthemainDutchshoppingstreetsbyestimatingtheeffectoffootfall– thenumber ofpedestriansthatpassby– andthenumberofshopsinthevicinityonstoreowners’rentalincome.Weaddress endogeneityissuesbyexploitingspatialvariationwithinshoppingstreetscombinedwithhistoriclong-lagged instruments.Ourestimatesimplyanelasticityofrentalincomewithrespecttofootfallaswellasnumberofshops inthevicinityof(atleast)0.25.Weshowthattheseshoppingexternalitiesareunlikelytobeinternalised.Itfollows thatsubstantialsubsidiestoshopownersarewelfareimproving,seeminglyjustifyingcurrentpolicies.Finally, wefindlimitedevidenceforheterogeneitybetweenretailfirmslocatedinshoppingstreetsintheirwillingness topayforshoppingexternalities.
1. Introduction
Oneofthemainreasonsthatpeoplechoosetoliveinthecityisthe presenceofarichvarietyofconsumergoodsandservicesofferedbythe retailsector(Glaeseretal.,2001).Shopstendtobeconcentratedeither inpedestrianshoppingstreetsandshoppingdistricts,oftenlocatedin citycentres,orinshoppingmallsinthesuburbs.InEurope,shopsare mostlyconcentratedinpedestrianstreets.Walkinginthosestreetsisso importantthatthemajorityofallpedestrianmovements occurwhile shopping.3
☆WethankLocatus andHISGIS forprovidingdata.TheeditorStuartRosenthal,
twoanonymousreferees,SergejGubin,JordiJofre-Monseny,AlbertSollé-Ollé, Jacques-FrançoisThisseandElisabetViladecans-Marsalarethankedforuseful commentsonearlierversionsofthispaper,aswellasparticipantsofseminars attheBalticInternationalCentreforEconomicPolicyStudiesinRiga,the Uni-versityofBarcelona,theAutonomousUniversityofBarcelona,IDE-JETROin Chiba(Japan),theLondonSchoolofEconomicsandparticipantsoftheUrban EconomicsAssociationconferencesinCopenhagenandVienna.Thispaperhas beenpreparedwithintheframeworkoftheHSEUniversityBasicResearch Pro-gramandfundedbytheRussianAcademicExcellenceProject‘5-100’.
∗Correspondingauthor.
E-mail addresses: h.koster@vu.nl(H.R.A.Koster),ipasidis@gmail.com(I. Pa-sidis),jos.van.ommeren@vu.nl(J.vanOmmeren).
1 HansisalsoresearchfellowattheHSEUniversityinTheRussianFederation,
theTinbergenInstituteandtheCentreforEconomicPolicyResearch.
2 JosisalsoresearchfellowattheTinbergenInstitute.
3 ThisisbasedondatafromStatisticsNetherlands.Weexcludehikingand
recreationalwalkingactivities.
Arguably,themostimportantreasonforretailfirmstoclusteristhe presenceofpositiveshoppingexternalities,whicharegeneratedby con-sumers’‘trip-chaining’behaviour.Shoppingexternalitieshaveasimple logic.Inretailmarketswherecustomershavetovisitstores, transporta-tioncostsarepaidbycustomersandincurredonashoppingtripbasis (Claycombe,1991).Consumerswhovisitseveralshopsbenefitfrom re-ductionsintransportandsearchcosts.Inthecontextofshoppingstreets, aretailfirm’sproductivityfunctiondependsonlocalfootfall– the num-berofpedestriansthatpassbyashop.Footfalltendstobehigherin ar-easwithmoreshops,sincepedestriansareattractedtoareaswithmany shops.Hence,theassociatedreductionsincostsforconsumersimplya positiveshoppingexternalityforretailfirms,whichisenhancedwhen multipleretailfirmsarelocatedincloseproximity(EatonandLipsey, 1982;Claycombe,1991;SchulzandStahl,1996).Similartoother ag-glomeration advantages,these shoppingexternalitiesareexpectedto capitaliseintostoreowners’rentalincome.4
Intheempiricalliterature,however,littleattentionhasbeengivento theimportanceofshoppingexternalities.Thefewstudiesthatmeasure shoppingexternalitiesfocusonU.S.shoppingmalls(seePashigianand Gould,1998).However,retailactivityinEuropeancitiesismainly
con-4Inretailmarkets,agglomerationeconomiesoccurlocally,socapitalisation
intoemployees’wagesishardlyrelevant.Incontrast,innon-retailmarkets, ag-glomerationadvantagesaremoredispersedandmainlycapitaliseintowages (seee.g. ArzaghiandHenderson,2008).
https://doi.org/10.1016/j.jue.2019.103194
Received20February2018;Receivedinrevisedform5August2019 Availableonline28September2019
centratedinshoppingstreetswhileshoppingmallsarefarlesscommon.5 Tothebestofourknowledge,thisisthefirstpaperthatquantifies shop-pingexternalitiesinshoppingstreets.
Shoppingstreetsarecharacterisedbyadifferentformofretail or-ganisationthanshoppingmalls.Incontrasttoshoppingmalls,wewill showthatpropertyownershipinshoppingstreetsishighlyfragmented andthateachshopisonlyaminorplayerlocally.Asaconsequence, internalisationofshoppingexternalitiesisunlikelytooccurinshopping streets.6 Thus,publicpoliciesthatfosterretailconcentrationby provid-ingsubsidiesarepotentiallywelfareimproving.7
Wefocusonthefullpopulationofmainshoppingstreetsinone coun-try:theNetherlands.Animportantfeatureofshoppingstreetsisthat theyaredominatedbyclothingstores(30%),cafés/restaurants(16%) andfoodstores(10%).Themainstrategyfollowedbytheretailfirmsin thesesectorsistodifferentiatethemselvesbysupplyingheterogeneous products.Thisis insharpcontrasttootherretailsectorsthatare ex-aminedintheeconomicliterature,whichofferhomogeneousproducts andwherespatialdifferentiationisthemainstrategy(e.g.gasstations, movie theatres,or videoretailers, seeNetzandTaylor,2002;Davis, 2006;Seim,2006,respectively).
Wecontributetotheliteratureinthreeways.First,weemploytwo measuresofshoppingexternalities.Ourfirstmeasure,footfall,isnovel intheagglomerationliterature.8 Wewillarguethatfootfallcaptures shoppingexternalitiesratherprecisely,alsobecausemeasuresoffootfall predominantlyincludeshopperswhovisit severalshops.Asasecond measureofshoppingexternalitiesweuse thenumberofshopsinthe vicinity.Thelatter is amorestandardproxy forexternalities, which isinlinewithalarge literatureonagglomerationeconomies(seee.g Combesetal.,2008;Meloetal.,2009).
Asecondcontributionofthepaperliesinanexplicittreatmentof het-erogeneity.Onemaysuspectthatthereisheterogeneityinthebenefits ofshoppingexternalities.Forexample,shopsthatarepartofachainare morelikelytolocateinstreetswithhighlevelsoffootfall.Using semi-parametricregressiontechniquesweshowthatheterogeneitybetween firmsintermsofmarginalwillingnesstopayforfootfallislimited.For example,shopsbelongtochainsandothershopshaveroughlythesame marginalwillingnesstopayforfootfall.Notethatthissuggeststhatstore ownerscannot,ordonot,discriminatebetweendifferenttypesofretail firmsinsettingtherent.
Thethird,andmain,contributionofthecurrentpaperisthe identifi-cationofshoppingexternalitiesbyestimatingthecausaleffectoffootfall ontherentalincomeofstoreowners,whichwederivebyestimating thecausaleffectsoffootfallontherentpaidbytenantsaswellason theprobabilitythataretailpropertyliesempty.Wewilladdressseveral sourcesofendogeneitybyexploitingspatialvariationwithinshopping streets,combinedwiththeuseofarangeofcontrolvariables(e.g.,size ofproperty,shoppingchainfixedeffects)andaninstrumentalvariable approach.
5 ShoppingmallfloorspaceperpersonismorethantenfoldintheU.S.
com-paredtoEurope(2150m2per1000peopleintheU.S.comparedto182m2per
1000peopleinEuropein2011,seeCushman&Cushman,2011).
6 Inshoppingmalls,storeownerssettherentbasedonretailfirms’revenues
andthus,shoppingexternalitiesareinternalised.Morespecifically,shopping mallrealestatemanagerschargelowerrentstofootfall-generatingshops(or ‘anchorstores’),whichcouldberegardedasafirst-bestsubsidy.Hence,thereis noroomforpublicpolicytoimprovewelfare(Brueckner,1993;Pashigianand Gould,1998;KonishiandSandfort,2003).
7 Manyexamplesofsuchsubsidiescanbegiven,althoughthesesubsidiestend
tobeimplicit.Forexample,inmanyEuropeancountries,inparticularGermany, policieshavecreatedpedestrianareas.Thelatterimpliesimplicitsubsidiesto localstoreowners,astheadvantagesarelocal,whereasthedisadvantagesof prohibitingcaruseintheseareasfallonotheragents.Subsidiestoparkingclose toshoppingclusters,andsubsidiestopark-and-ridefacilities,includingfree pub-lictransporttowardscitycentres,areotherrelevantexamples.
8 Intheretailindustry,footfallisastandardmeasuretoexplainthe
attrac-tivenessofashoppinglocation.
Tobemorespecific,asemphasisedintheagglomerationliterature, bothourmeasuresofshoppingexternalities,i.e.footfallandnumberof shopsinthevicinity,areessentiallymeasuresofspatialconcentration thatmaybeendogenousbecausetheytendtobepositivelycorrelated tounobservedattractivelocationcharacteristics.Weaddressthisissue byfocusingonshopsthatarelocatedwithinthesameshoppingstreet andarethereforeclosetoeachother.9 Thisstrategylargelyaddresses theissuethatfootfallandnumberofshopsinthevicinityaregenerated bylocalamenitiesandaccessibility.Wefurthercontrolforawiderange ofshopandstreetcharacteristics,aswellaslocalamenitiesthatmay generate footfallthatisnot relatedtoshopping,suchasthenumber ofschools,publicbuildingsandreligiousbuildings.Furthermore,using web-scrapeddatafrom FlickR,we controlforthenumberofpictures takenbytouristsandresidentsintheshoppingstreet,whichproxyfor difficult-to-capturedifferencesinamenitiesandattractivenesswithina street(Gaigné etal.,2018).
Nevertheless,itisplausiblethatendogeneityissuesarestillpresent for both measuresof shopping externalitiesdue tomeasurement er-rorandreversecausation issues,whichwouldyieldanunderestimate of shopping externalities. We believethere are three reasons which maycausereversecausation.First,attractive,andthereforeexpensive, streetsdisproportionallyattractshopswithhigh-endbrands(e.g.Louis Vitton, GiorgioArmani). These shopsgenerate little footfall,so that there isadownward biasoftheestimatedfootfallcoefficient.Inthe Dutchcontext,themainexampleistheP.C.HooftstraatinAmsterdam, butoneobservesthesamephenomenoninwell-knownshoppingstreets in othercountries(NewBondStreet,London; Champs-Élysées,Paris; ViaMonteNapoleone,Milan;Bahnhofstrasse,Zurich).Second,aretail firmthatsignsacontracttopayhighfixedrentsmaybemorelikelyto ‘workhard’togeneratefootfall(and,therefore,sales) inorderto en-surethatitcanaffordtheserents.Third,onemayarguethatthereisa simultaneityproblem(whichusuallydoesnotoccurinahedonicprice setting),becauseweconsidertheaggregatenumberofshops(and foot-fall)inalocation,whichwouldfallwithrents,asinastandardtextbook supply-demandsettingwithahomogeneousgood.10
Endogeneityissuesalsoplayarolewhenfocusingontheeffectson vacancies.Forexample,ahigherlevelofvacanciesinastreetwilllikely inducelessfoottrafficand,mechanically,reducethesmallnumberof non-vacantstoresinthevicinity.
Toaddresstheseendogeneityissues,wewillemployan instrumen-talvariablesapproachusingtheexactlocationofcinemasin1930. Be-causethereisahighautocorrelationofretaillocations,itappearsthat thenumberofcinemasin1930hasastrongeffectonfootfallandon numberofshopsinthevicinity.Thisisnottoosurprising:forexample, themaimshoppingstreetinAmsterdamnowadays,deKalverstraat,has beenoneofthemainshoppingstreetsforalmost300years.Notethat weidentifytheseeffectswithinshoppingstreetsandcontrolforthe cur-rentlocationofcinemas,henceourinstrumentplausiblyaddressesthe variousendogeneityconcerns.
Wealsomakesurethatourresultsdonotdependonthis particu-larchoiceofinstrumentsbygoingbackevenfurtherintime.Wewill usethenumberofshopsin1832asaninstrument.Thisinstrumentis particularlyconvincingbecauseweshow,usingdataonlandvaluesin 1832,thatpastattractivelocationsdonotcommandhigherretailrents nowadays.Weareabletoshowthatthelocationofshopsin 1832is astronginstrument(forfootfallandnumberofshops),evenwhenwe controlforthenumberofbuildingsin1832.Wediscusstheassumptions underlyingouridentificationstrategyatlengthinthepaper.
Ourresults showthatfootfallandnumberofshopshaveastrong positiveeffectonrentalincomewithanelasticityrangingfrom0.25–
9Shoppingstreetstendtobeshort;inourdata,themedianlengthisonly
182m.
10Wewouldliketothankananonymousrefereeformentioningthelattertwo
0.50.Wheninstrumenting,theestimatesforbothfootfallandnumberof shopsaresomewhathigher.Thus,therearesubstantialexternalbenefits fromfosteringfootfallandretailconcentration.
Basedontheseestimates,theoptimallevelofsubsidytostoreowners appearstobesubstantial,whichpotentiallyjustifiesarangeofcurrent policiesthatareespeciallybeneficialtoshoppingstreets.Current poli-cieseithersubsidiseshopowners,e.g.throughpedestrianisationof pop-ularshoppingstreets,orsubsidiseshopperseitherthroughtheprovision of(subsidised,andsometimesevenfree)publictransporttoshopping districtsorparking.Zoningpolicieswhichconcentratesshoppinginto shoppingstreets(e.g.attheexpenseofresidentialhousing),mayalsobe beneficial.
Relatedliterature.Ourpaperlinksandcontributestothreestrands
of literature.First, itrelatesto theliterature on spatialcompetition andproduct differentiation (D’Aspremontet al., 1979; Osborne and Pitchik,1987).Davis(2006)focusesonmovietheatres,andevaluates consumers’transportcosts,theeffectofgeographicdifferentiation,and theextentofmarketpower.Seim(2006)showsthattherearesignificant returnstoproduct(orspatial)differentiationandillustratesthatmarkets withmorescopefordifferentiationsupportgreaterentry.Jia(2008)and Arcidiaconoetal.(2016)studytheimpactofWal-Martontheretail marketonincumbent(discount)supermarketsandsmallgrocerystores. Clappetal.(2016) focusonopenings andclosingsof multi-line de-partmentstoresandfindevidenceforstrongnegativecompetitive ef-fects withinthe ownbranch.Zhou (2014)shows that multi-product search,whichisimportantwhenconsumersbuymultipleproductsin oneshoppingtrip,cansignificantlyinfluenceretailfirms’pricing de-cisions.JohansenandNilssen(2016)investigatetheconditionsunder whichone-stopshoppingcausestheformationofbigstores.However, thesepapersignorethatmanyshopsbenefitfromeachotherwhen lo-catedclosetoeachother.
Ourpaperalsorelatestoasecondliteraturethatexplicitlyfocuses onthebenefitsofagglomerationforfirms.Thereisampleevidencethat firmsthatlocateclosetogetherbenefitthroughinput-andoutput shar-ing,labourmarketpoolingandknowledgespillovers(Marshall,1890). Currentevidencesuggeststhatthe elasticityof productivitywith re-specttodensityisaround0.05(seee.g.themeta-analysisbyMeloetal., 2009).This literature typicallyfocuses on the manufacturing indus-try. Compared to that literature, we find substantial agglomeration elasticities, which are(at least) 5times larger. This is in line with Kosteretal.(2014),whoshowthatagglomerationeconomiesaremuch moreimportantforretailfirms.Teulingsetal.(2017)usea monocen-tricmodelwherecustomersarriveinacentrallocationandhavetowalk toshopsaroundthislocation.Locationsthatarefurtherawayfromthe centrearethereforeexpectedtobelessprofitable.Theirempirical evi-denceindicatesthatrentsinshoppingdistrictsareindeedhigherclose tothecentre.11
Our findings also contribute toa more policy-oriented literature studyingtheeffectivenessofretailpolicies,inparticulartowardsthe effectsoftheopeningoflarge‘big-box’retailersneartheurbanfringe (Sanchez-Vidal,2016).Somestudiesdemonstratethatthewelfare ef-fectsofcurrentplanningpoliciesthathinderentryinretailmarkets,and particularlyoflargeretailers,arenegative.Severalstudieshaveshown thatregulationpoliciesreduceretailproductivityandjobgrowthand in-creasemarketpowerofincumbentstores(BertrandandKramarz,2002; SchivardiandViviano,2011;HaskelandSadun,2012;Cheshireetal., 2015).Bycontrast,ourstudyshowsthatcurrentpoliciesthat implic-itlysubsidisestoreownersinshoppingstreetsarepotentiallywelfare improving,astheymakeitmoreattractivetoopenadditionalstoresin denseareas.Thisconclusionisconsistentwiththeshoppingmall litera-ture,whichshowsthattheprovisionofsubstantialrentdiscountsto ‘an-chor’storesinUSshoppingmallstointernaliseexternalitiesiscommon
11 Tocapturethisphenomenon,wecontrolforwalkingtimetothecentreofa
shoppingdistrict(i.e. thelocationwiththehighestfootfall).
practice(PashigianandGould,1998;Gouldetal.,2005).Shopping ex-ternalitiesmayjustifypublicpoliciesthatpedestrianiseshoppingstreets inordertoeffectivelyclustershopsandinternaliseexternalbenefits.
Theremainderofthispaperisstructuredasfollows.InSection2we discuss the theoretical framework that guides the empirical results. Section3introducesthedataandreportsdescriptivestatistics,followed byadiscussionoftheeconometricframeworkinSection4.InSection5, wepresentanddiscussourresults,includingtheestimatesofthe op-timalsubsidy.Section6discussesanextensionwhereweallowfora heterogeneousversionofourmodel.Alsothemainsensitivityanalyses arediscussed.ThelatteraredescribedinmoredetailinAppendixD.We drawconclusionsinSection7.
2. Theoreticalframework
2.1. Rentalincome,rentsandvacancies
Weaimtomeasurethepresenceofshoppingexternalitiesby esti-matingtheeffectoffootfallornumberofshopson(expected)rental incomeofshopi,denotedbyIi.12 Footfallisdefinedasthenumberof
pedestriansthatpassashop(perunitoftime).Weassumethat pedestri-answalkthroughoneshoppingstreetandpassallshopsinthisstreet.13 Let𝑠={𝑓,𝑁}denoteshoppingexternalities,wherefreferstofootfall andNtothenumberofshopsintheshoppingstreet.14 Inwhatfollows, wemakeadistinctionbetweenstoreownersthatownshopsandretail firmsthatrentshops.Shopscanbeeitheroccupiedbyretailfirmsor vacant.Storeownersofvacantshopsneedadvertisingservicestofind anewtenant,whichiscostly.Givenrentpiandvacancyratev,rental
incomeofashopisgivenby:
𝐼𝑖=𝑝𝑖(1−𝑣)−𝑐𝑖𝑣, (1)
where𝑝𝑖(1−𝑣)isrentalincomewhenthepropertyislettoaretailfirm andcivistheadvertisingcosts.Itseemsreasonabletoassumethat,at leastinthelongrun,theadvertisingcostsci areproportionaltopi,so
𝑐𝑖=𝜅𝑝𝑖,where𝜅 >0.Becausevacancyratestendtobesmall(usually smallerthan10%inourdata),log(1−(1+𝜅)𝑣)≈ −(1+𝜅)𝑣.Hence,the logofrentalincomeisthen(approximately)equaltolog𝑝𝑖−(1+𝜅)𝑣.
Ifshoppingexternalitieshaveaneffectontherentandvacancyrate, itfollowsthattheeffectoflogshoppingexternalitiesonlogrental in-comecanbewrittenasthesumofthemarginaleffectoflogshopping externalitiesonthelogarithmofrentandthemarginaleffectoflog shop-pingexternalitiesonthelevelofthevacancyrate:
𝜕log𝐼𝑖 𝜕log𝑠 ≈ 𝜕log𝑝𝑖 𝜕log𝑠 −(1+𝜅) 𝜕𝑣 𝜕log𝑠. (2)
Note that a standard hedonic model cannot be used to predict 𝜕logIi/𝜕logsbecausevacanciesarenotincorporatedinsucha
frame-work. InAppendixA.1 wetherefore setup asearch andbargaining framework,wherestoreownershavetodecideonthelevelof advertis-ingwhichisnecessarytofindretailfirmssearchingforretailspace. Con-sequently,inthissetup,vacancyratesareendogenouslydetermined. Furthermore,itisassumedthatownerswithvacantpropertiesandretail firmssearchingforretailspacebargainabouttherentlevelwhenthey makecontactwitheachother,whilebothareuncertainhowmuchtime ittakestomakeanothercontact.Wethenshowthat,inequilibrium,
12Asanalternative,onemayestimatetheeffectontransactionpricesofshops.
Therearetworeasonsweprefertofocusonrentalincome.First,transaction pricesreflectexpectationsaboutfuturerentsandthereforefuturelevelsof shop-pingexternalities.Second,salestransactionsarerarerelativetorent transac-tions.Inourdata,only10%oftheobservationsrefertosalestransactions.
13Thisassumptionismadeforconvenienceonly,asitisplausiblethatshoppers
usuallywalkthroughseveralshoppingstreets.
14Inourempiricalapplication,wetakeintoaccountthatshoppersusually
𝜕logpi/𝜕logs>0and𝜕v/𝜕logs<0.15 Hence,in equilibrium,shopping
externalitiesnotonlyincreaserents,butalsoincreasevacancyrates.If weassumethat𝜅 =0,wegetthelowerboundestimateoftheeffectof footfallonlogrentalincome.Wewillmakethatassumptionbutalso calculatetheeffectoffootfallonrentalincomefordifferentvaluesof𝜅 (basedontheadvertisingcostsforDutchretailproperties).
2.2. Welfareandretailpolicies
Letusnowfocusonwelfareeffectsofretailpolicies.Intuition sug-geststhatretailpolicycanbewelfareimproving,becauseretailfirms generatefootfall,andpoliciesmayinfluencethespatialconcentration ofretailfilms.Toinvestigatethis,wewillassumeherethattheeffect ofshoppingexternalitiesisentirelydrivenbytheeffectoffootfallon rentalincome.Theeffectofnumberofshopsisonlyindirect,asit af-fectsfootfall,butnotrentalincomedirectly.Hence,moreformally,we assumethatforashopitherentalincomeis:
𝐼𝑖(𝑓,𝑁)=𝐼𝑖(𝑓(𝑁))=𝑝𝑖(𝑓(𝑁))(1−𝑣(𝑓(𝑁)))−𝑐𝑖𝑣(𝑓(𝑁)). (3)
Storeownersmaydecidetoincreasethesupplyofstoresinastreet (ex-tensivemargin)or/andreduceexistingvacancies(intensivemargin).16 Hence,thenumberofshops,aswellasthenumberofvacantshopsina shoppingstreet,isendogenous.Bothdecisionsimplyanexternality be-causefootfalldependspositivelyonthenumberofshopsandnegatively onthevacancyrate.Hereweinvestigatethewelfareeffectofincreasing thenumberofoccupiedshops.17 Combinedwiththeassumptionthatwe focusonacompetitivemarket,thisimpliesthatitisnotnecessaryto distinguishbetweenshopownersandretailfirms.
Thenumberofoccupiedshops,N,inashoppingstreetis endoge-nouslydetermined.Fornowletusassumethatshopsarehomogeneous. Theper-periodmarginalconstructionandmaintenancecostsofNshops areequaltoC(N),whichisincreasinginN.Footfallinastreet,f,isan increasingfunctionofthenumberofshopsintheshoppingstreet,so 𝜕f/𝜕N>0,aswellasthefootfallgeneratedperindividualshop,denoted by𝑓̃𝑖,so𝜕𝑓∕𝜕𝑓̃𝑖>0.Theshop’scostsofgeneratingfootfallareequalto
𝑞𝑖(𝑓̃𝑖),whichisanincreasingandconvexfunctionof𝑓̃𝑖.Forexample,
shopsmaygeneratefootfallbyadvertising,whichiscostly.Welfarein theshoppingstreetisnowdefinedby:
∫ 𝑁 0 𝐼𝑖(𝑓(𝑁,𝑓̃𝑖))d𝑖−𝐶(𝑁)−∫ 𝑁 0 𝑞𝑖(𝑓̃𝑖)d𝑖. (4)
Thewelfare-optimalnumberofshopsisthendeterminedbythe follow-ingequation:
𝐼𝑖+𝑁𝜕𝑁𝜕𝐼𝑖 =𝐶′(𝑁)+𝑞𝑖(𝑓̃𝑖), (5) 15 Anotherissueisthatfootfalldependsonthenumberofvacancies,asvacant
shopswillattractfewercustomers.WeaddressthisissueinAppendixA.2where weendogenisefootfallsothatitisdependentonvacancyrates.Weshowthat themainconsequenceofignoringthisendogenousrelationshipforourempirical investigationisthatoneunderestimatestheeffectoffootfallonrentalincome. WecomebacktothisissueinSection4.
16 Thesupplyofshopscanbeincreasedbybuildingnewpropertiesforretail
activityorbyconvertingexistingspacetoretailuse.Existingvacanciesmaybe reducede.g. byincreasingthesearcheffort.
17 Thewelfareeffectoffillinganexistingvacancyandincreasingthesupplyof
shopsisequivalentgiventwoconditions.Thefirstisthatfillinganvacancyand increasingthesupplyofshopshasanidenticaleffectonfootfall,whichseems reasonable.Thesecondconditionisthat,intheabsenceofafootfallexternality, thedecisiontofillavacancyissociallyoptimal,implyingthattherearenoother externalities.Notethatstoreownerswithvacanciesmayincreasetheir advertis-ingexpenditure,whichincreasestheprobabilitytofindatenant.That,inturn, mayincreasefootfallforothershops.Itisingeneralnotclearwhetherstore ownerswithvacancieschoosethesociallyoptimaladvertisingexpenditure be-causeofnegativecongestionandpositivesearchexternalities.See,forexample, Hosios(1990)forananalysisofoptimaladvertisingexpenditureinthelabour market.Weleavethisissueforfurtherresearch.
where𝑖=𝑁.Theaboveequationstatesthatthemarginalexpected in-comeofNisequaltothesumofthemarginalcostofopeningan addi-tionalstoreandthecostofgeneratingfootfall.Notethatthemarginal benefitofopeningashop– i.e.theleft-handsideoftheaboveequation – isanincreasingfunctionofN.Thisisintuitive,becauseeachshopthat opensupintheshoppingstreetbenefitsfromthefootfallgeneratedby nearbyretailfirms.Thisimpliesthatthecostofopeningashop(C(N)) mustbestronglyconvex,sothatthesecond-orderconditionholds.18
Tocalculatetheoptimalsubsidytostoreownersisstraightforward. ThemarginalstoreownerwillignorethetermN(𝜕Ii/𝜕N)when
consid-eringtoincrease(ordecrease)thesupplyofshops.Hence,themarginal externalbenefitofopeningupanewshopisequalto:
𝑁𝜕𝑁𝜕𝐼𝑖 =𝐼𝑖⋅ 𝜀𝐼,𝑁(𝑖)>0, (6) where𝜀I,N(i)denotestheelasticityofrentalincomeofshopiwithrespect
tothenumberofshops.Inourempiricalapplicationwewillestimate theelasticity𝜀I,N(i).
Importantly, wewill assume thatthis shoppingexternalityis not internalised bystoreowners.Thisseemsareasonableassumptionfor shoppingstreetsduetothefragmentedownershipofshops,whichis, aswewillshow,supportedbyourdata.19 Giventhisassumption,the Pigouviansubsidytothemarginalstoreownerinthefirst-bestoptimum is𝜀I,N(i)timestherentalincomeofashop.Inourempiricalanalysis,we
areabletocalculatethissubsidyaswewillobserverentalincomeand estimate𝜀I,N(i).
We willnow determinethewelfare-optimallevel of footfallin a street,whichistheresultoftheself-chosenfootfalllevelbyanindividual firm.Thewelfare-optimallevelatthestreetlevel,f,canthenbederived bymaximisingwelfarewithrespecttotheindividualleveloffootfall,
̃
𝑓𝑖.Itcanbederivedbyanequation,whichstatesthatthemarginal
ex-pectedincomeof individualfootfall, 𝑓̃𝑖,isequaltoitsmarginalcost:
𝜕𝐼𝑖 𝜕𝑓 𝜕𝑓 𝜕𝑓̃𝑖 =𝑞′ 𝑖(𝑓̃𝑖). (7)
Wewillnowassumethatshopsarehomogeneous.Thisimpliesthat𝑓̃𝑖=
̃
𝑓,𝐼𝑖=𝐼,𝑞𝑖(𝑓̃𝑖)=𝑞(𝑓̃)and𝜀𝐼,𝑁(𝑖)=𝜀𝐼,𝑁.Itfollowsthat𝑓=𝑁𝑓̃and
Eq.(7)canbewrittenas: 𝜕𝐼
𝜕𝑓𝑁=𝑞′(𝑓̃). (8)
Theleft-handsideofthisequationdepictsthemarginalbenefitof foot-falltosocietygeneratedbyanindividualshop.Themarginalbenefitof footfalltoashopisequalto𝜕I/𝜕f(astheshopignoresthatothershops inthesamestreetbenefitfromthefootfall).Giventhatthenumberof shopsisusuallylargeinastreet(andfaraboveone),theexternalbenefit offootfall,andhencethewelfare-optimalsubsidy,isequalto:20
𝜕𝐼
𝜕𝑓𝑁=𝑁𝑓⋅ 𝐼𝜀𝐼,𝑓>0, (9)
18Thelatterislikelytruebecauseinshoppingstreets,shopsalmostalways
occupythegroundfloorofabuilding(seeLiuetal.,2016),whichimpliesthat thenumberofshopsisrestrictedbythelengthofthestreet.
19Incontrast,inshoppingmalls,developerswillinternalisetheseexternalities
bydeterminingtheoptimalnumberofstoresandbycharginglowerrentsto footfall-generatingshops(or‘anchorstores’)(Brueckner,1993;Pashigianand Gould,1998;KonishiandSandfort,2003)Therefore,inashoppingmall,the developerisabletoprovidefirst-bestsubsidies,basedontheamountoffootfall generatedbyeachstore,andmaximizeashoppingmall’swelfare.
20Toevaluatetheexactgeneral-equilibriumwelfareimprovementsassociated
where𝜀I,fdenotestheelasticityofrentalincomewithrespecttofootfall.
Inourempiricalapplicationwewillalsoestimatetheelasticity𝜀I,f.We
willshowthatinourdata𝜀𝐼,𝑓=𝜀𝐼,𝑁(whichisinlinewithour
assump-tionthatfootfallisproportionaltothenumberofshopsinthestreet). Wepreviouslyassumedthatretailfirmsarehomogeneous.Itisclear howeverthatretailfirmsmaydifferintheextentthattheybenefitfrom thepresenceofothershops,i.e.footfall.Hence,wewillallow𝜀I,Nand
𝜀I,f toberetailfirm-specific.Wewillsee howeverthatheterogeneity
ofretailfirmswithrespecttothemarginalwillingnesstopayfor foot-fallgeneratedbyothershopsisquitelimitedinshoppingstreets,which suggeststhatmodernshoppingstreetsattractrelativelysimilarshops. InSection6wewillinvestigatethisfurther.
3. Dataanddescriptives
3.1. Data
Webaseourempiricalanalysisonvariousdatasets.Thefirstoneis obtainedfromStrabo,aconsultancyfirmthatgatherscommercial prop-ertydata.Itcomprisestransactionsofcommercialpropertiesprovided byrealestateagentsfrom2003to2015. Thedatasetcontains infor-mationaboutannualrents(reportedatthemomentthatthecontract hasbeensigned)andrentalpropertyattributes,suchasaddress,size (grossfloorareainm2)andwhetherthebuildingisnewlyconstructed orrenovated.Inourdata,allrentsareindependentofretailrevenues inlinewithcommonpracticefortheNetherlands.21 FromtheStrabo dataset,weexcludeobservationsforwhichnorentisreported(27.8% ofallshopsinthedataset).22 Therentaltransactionsarethenmatchedto datafromtheBuildingRegistry(BAG),whichprovidestheexactlocation andconstructionyearforallbuildingsintheNetherlands.Usinga25m distancethreshold,wematchedalmostallof theStraboshops.Based ontheListedBuildingRegister,wehaveaddedinformationonwhether therentalpropertyisinanareathatisassignedasahistoricdistrict. Thelatterisrelevantsincehistoricdistrictsmayattracttouriststhatare notinterestedinshopping(CarlinoandSaiz,2008).Thedatasetisalso mergedwithdetailedlandusedatafromStatisticsNetherlands.Thelatter dataenablethecalculationofdistancetothenearestrailwaystation. Todetermineamenitiesandaccessibility oftheshop,wefirstgather datafromDuoonthelocationsofkindergartens,primaryandsecondary schoolsinthestreet.FromImergisweusethelocationofpublic build-ings(townhalls,policeandfirestations)andreligiousbuildings.Using OpenStreetMapwealsodeterminethenumberofbusstopsina shop-pingstreet.Itishardtocontrolforallrelevantamenities.Wetherefore gatherdatafromEricFisher’sGeotagger’sWorldAtlas,whichcontainall thegeocodedpicturesonthewebsiteFlickr.Thesepicturesshouldbea reasonableproxyforalocation’sattractiveness(seeGaigné etal.,2018). Theothermaindataset isa retaildatasetobtained from Locatus, whichcontainstheentirepopulationofshops.Foreachestablishment, weknowwhetherashopisvacantoroccupiedanditsretailsector(when occupied),andwhetheraretailfirmoccupyingashopispartofachain. Furthermore,weknowthe8-digitretailsector,whichprovidesvery de-tailedinformation.Forexample,wedonotonlyknowwhetheraretail firmsellsapparel,butalsowhetherittargetedatkids,womenormen.
writtenas 𝐼 1⋅ 𝜀1
𝐼,𝑁−𝐼 2⋅ 𝜀2𝐼,𝑁.Nowsupposethatstreet1containsmoreshops
thanstreet2.Thisequationshowsthatlargershoppingstreetsmustreceive largersubsidiesaslongastheelasticityofrentalincomewithrespecttonumber ofshopsispositive(implyingI 1> I 0)andnon-decreasinginthenumberofshops
(so,𝜀 1
𝐼,𝑁≥𝜀 2𝐼,𝑁).Wewillseethatbothconditionsarefulfilled.Thisimpliesthat
subsidiestostoreownersinlargeshoppingstreetsmustexceedthoseinsmall shoppingstreets.
21 Moreover,ouridentificationstrategywilladdresstheissueof‘percentrents’. 22 Wedonotfindevidencefordifferencesbetweenthetransactionswithand
withoutinformationonrents,althoughthelatterseemtorefertosomewhat largerrentalproperties.
TheLocatusdatasetalsoprovides3936annualcountsoffootfallin allmainshoppingstreetsoftheNetherlandsfrom2003to2015,sothat wehavemorethan50thousandfootfallcountsoverthewholestudy pe-riod.Themeasurementpointsarenottheshopsthemselves,butselected pointsinpedestrianstreetsoratsidewalkswithanaveragedistanceof 45mbetweenthem.Themainshoppingstreetscontainabout13.4% ofallshopsintheNetherlands.Theannualfootfalldata,providedby Locatus, isbasedonfootfallmeasurementscollected onfour‘regular’ Saturdays(twoinSpringandtwoinAutumn)atfourdifferenthours oftheday.Usingthese16measurements,Locatuscalculatesthe aver-agefootfallperday.Thefootfalldataarematchedtoallshopsinthe previously-defined shoppingstreets.Withineach shoppingstreet,the averagedistancebetweenfootfallmeasuresisapproximately45m.
Formostobservations(about90%)in theStrabo datasetwehave detailedinformationonthenameoftheretailfirmrentingthestore. WehavematchedeachrenttransactionintheStrabodatasettoashop intheLocatusdatasetinordertorecoverthe8-digitsectorandwhether thefirmispartofachain.
Wehavedefined ashoppingstreetasacontinuousstraightstreet (orslightlycurvedstreet)basedonmanuallycreatedGISpolylinesfor allstreetsforwhichthereisatleastonelocationwherefootfalldatais available.Usingthisdefinition,basedontheabove-discussedBuilding Registrydataset,wedefine1253uniqueshoppingstreets.Themedian lengthofthesestreetsis182m,sostreetsareshort.Themedianwidth ofastreetis12m.Thisunderlinesthatwefocusonshoppingstreets withinEuropeancitieswithshortandnarrowstreets,ratherthan,for example,U.S.cities,whichusuallyhavelongerandwiderstreets.
WeillustratethedatainFig.1basedonasampleofourdatafor thecitycentreof Amsterdam,whichcontains twoshoppingdistricts includingthecitycentre.Thisareaisthebusiestshoppingdistrictofthe Netherlandswiththehighestlevelofdailyfootfall.Itshowsthatthere is substantialspatialvariationin theannualaverageof footfallboth withinshoppingstreetsandbetweenshoppingdistricts.Moreover,rent transactions(thestarsinthemap)arenumerousandcoveralmostthe wholeshoppingdistrict.Thecentreofshoppingdistrictsisdetermined bytakingthelocationwiththehighestfootfallwithinthedistrict.
Forouridentificationstrategy,explainedlateron,wewillalsorely onhistoricdatagoingbackto1930oreven1832.Wegathered infor-mationon theexactlocationofcinemasin1930,aswellasin 2010 (in1930therewere315cinemas,whilein2010thishasbeenreduced to180)fromSpinLab.Formapsanddescriptivestatistics,wereferthe readertoAppendixB.1.
We furtheruse datafrom HISGIS, whichprovides geocodeddata of the first Dutchcensusof 1830. Itis the oldest nationwide regis-trationsystemofpropertyandlandownership.Wehaveinformation on thefootprintsizeofeach building.Moreover, foreach parcelwe knowtheownerandtheowner’soccupation.Basedontheoccupation andwhethertherewasabuildinglocatedontheparcel,wedetermine whetherabuildingwasusedasashop.Weexplainthisin more de-tailinAppendixB.1.Foreachparcelwealsohaveinformationonthe so-calledCadastralIncome,whichisaproxyforthelandvalue,asland taxeswerebasedonthisvalue.Again,wereportdescriptivesandmaps inAppendixB.1.
3.2. Descriptives
Inthissection,wepresentthedescriptivestatisticsforthemain vari-ablesthatweincludeinouranalysis.Ourmaindependentvariableis theannualrentalprice.Table 1summarisesthedescriptivestatistics fortheStrabodataset.Wehave4738rentaltransactionslocatedin131 shoppingdistricts.Themeanrentalpriceisequalto€ 51,449.Footfall exhibitssubstantialvariationrangingfrom100to71,000pedestrians passingbyeachday.Meandailyfootfallis13,328withastandard de-viationof8935.
Fig.1. SamplemapfortheAmsterdamcitycentre.
Table1
DescriptivestatisticsfortheStrabodataset.
(1) (2) (3) (4)
mean sd min max
Rent (in €) 51,234 83,227 3400 2,700,000
Footfall (number of shoppers per day) 13,328 8935 100 71,000
Number of shops, < 200 m 132.2 68.89 0 406
In pedestrianised street 0.862 0.345 0 1
Size of property (in m 2 ) 201.2 268.2 19 7200
Building – new 0.0112 0.105 0 1
Building – renovated 0.00708 0.0839 0 1
Construction year 1933 85.30 1325 2016
In historic district 0.474 0.499 0 1
Number of photos in street, < 200 m 283.4 451.3 0 5181 Religious buildings in street, < 200 m 0.714 1.058 0 6
Bus stops in street, < 200 m 0.861 1.478 0 10
Public buildings in street, < 200 m 0.0955 0.302 0 2 Schools, kindergartens in street, < 200 m 0.153 0.515 0 4
Railway stations, < 200 m 0.0247 0.155 0 1
Notes :Thenumberofobservationsis4378.
whowalkthroughseveralstreets.Therefore,wewillmeasurethe num-berofshopswithinaradiusof200mfromeachshoplocation(butwe willalsoexperimentwithotherthresholds).Theareadefinedbya200m radiusmaycontainseveralshoppingstreets.Thereareonaverage132 shopswithin200m,hencethenumberofshopswithinthevicinityis veryhigh.Moreover,thelargemajorityofshopsaresmall,withamean of201m2.Oneimplicationofthisisthatindividualshopsusually con-tributelittletooverallfootfall,whichmakesitveryunlikelythatshop ownerswithinshoppingstreetshavemarketpowerandsuggeststhat theretailrealestatemarketishighlycompetitive.
Thecorrelationbetweenfootfallandnumberofshopsinthevicinity ismoderate(𝜌 =0.375).About1%ofshopsareeitherneworrenovated
whentherentaltransactiontookplace.About6%oftheshopsisbuilt before1832,whileabouthalfoftheshopsarebuiltbeforetheSecond WorldWar.Therefore,almosthalfoftheobservationsisinhistoric dis-tricts.InTable1wealsoprovideinformationaboutpedestrianstreets. Wedonotcontrolforpedestrianstreetsintheanalysis (pedestrianisa-tionmightbeendogenous),butnotethatthelargemajorityofshops, about85%,arelocatedinstreetsthatarepedestrianised.
Inourdata,wewilldistinguishbetween153shoppingdistricts(a shoppingdistrictcontainsabout260shops,onaverage).Asubstantial proportionof shoppingdistricts(about45%)arenotwithin 5kmof thecentreof acity.Hence,intermsofshoppingdistricts,wehavea goodrepresentationofshoppingdistrictsthatareoutofthecitycentre. However,theproportionofshopsnotwithin5kmofthecentreofacity is muchsmallerandonlyabout25%(assuburbanshopping districts tendtobesmaller).
InTable2,wereportdescriptivestatisticsforshopsintheLocatus dataset.Wehave410,544observationsofshopsin133shopping dis-trictsand1243shoppingstreets.About6%oftheshopsarevacant.23 Thedescriptivestatisticsofthelocationvariablesarecomparabletothe descriptivestatisticsfortheLocatusdata.
Wefocusonshopsinshoppingstreetsthatpresumablyaimtobenefit fromfootfall,henceoursampleofshopsisfarfromarandomsample ofshops.Thisisparticularlythecaseforclothing.Clothingstoresare themostcommonbranch(30%ofshops)– almost4timeshigherthan thenationalaverage.However,thisisnotthecaseforrestaurantsand
23The Strabo datasetcontainsaconsiderablyhighershareofshopsinolder
Table2
DescriptivestatisticsfortheLocatusdataset.
(1) (2) (3) (4)
mean sd min max
Shop is vacant 0.0610 0.239 0 1
Footfall (number of shoppers per day) 12,379 10,732 100 102,600
Number of shops, < 200 m 129.0 62.77 0 439
In pedestrianised street 0.843 0.364 0 1
Building size (in m 2 ) 179.9 1089 25 27,694
Construction year 1967 35.83 1445 2012
In historic district 0.403 0.490 0 1
Number of photos in street, < 200 m 280.5 560.7 0 5493 Religious buildings in street, < 200 m 0.755 1.096 0 7
Bus stops in street, < 200 m 0.705 1.338 0 10
Public buildings in street, < 200 m 0.0951 0.301 0 2 Schools, kindergartens in street, < 200 m 0.185 0.552 0 4
Railway stations, < 200 m 0.0193 0.138 0 1
Notes :Thenumberofobservationsis410,544.
cafés:theshareofrestaurantsandcafés(16%)isexactlyequaltothe nationalaverage.Inbothbranchesthedegreeofproductdifferentiation isextremelyhigh.
InEurope,shoppingdistrictsusuallyexhibitapatternofmixedland uses.ThisisalsothecaseintheNetherlands.Usinginformationfromthe BuildingRegistryonbuildingswithin25mofashoppingstreet,itappears thatasmallminority,onlyaboutonequarter,ofthepropertiesisused forshopping,whereasalmosthalfofthepropertiesisresidential,and theotherquarterisusedforotherpurposes(e.g.offices,publicservices). Theobservationthatshopsaretypicallysmallsuggeststhatthe re-tailrealestatemarketishighlycompetitive,butthisisonlytruewhen storeownership(andthereforelandownership)withinshoppingstreets ishighlyfragmented.Thisappearstobethecase.Thisclaimisbased ontheStrabodatasetforwhich storeowner typeandregularlyeven owner’snameisreported(weknowthestoreownertypeforabouttwo thirdsoftheobservationsandstoreowner’snameforaboutonethird oftheobservations).Usinginformationaboutnames,itappearsthaton averageonly28%oftheshopswithinashoppingstreetbelongtoastore ownerwhoownsmorethanoneshopinthesameshoppingstreet.24 On average,storeownerspossess1.32shopsperstreet,whichisalow num-bergiventhatthereareonaverage55shopsperstreet.Thisindicates thatitishighlyunlikelythattheshoppingexternalitythatwemeasure isinternalised.
Wealso useinformationabout owner typethat distinguishes be-tweenprivate-storeowners, realestateagencies,pension funds, con-structioncompaniesetc.Only34%ofshopsareownedbyrealestate investors.Thus,thelarge majorityofshopsareownedbyindividual privateinvestors,whichfurthersupportsourclaimthatownershipof shopswithinshoppingstreetsishighlyfragmented.Wealsogathered in-formationfromMenger(2014)onlocalisedshopassociations,inwhich externalitiesmaybeinternalised (butwhich maynotengagein anti-competitiveactivities).Theseshopassociationsarearea-basedand or-ganisee.g.collectivemarketingandprovidepublicgoodssuchas Christ-maslightning.Onaverage14%oftheretailfirmsarepartofsuch asso-ciations.
4. Econometricframework
4.1. Rents
Wefirstfocusontheestimationoftheeffectofshoppingexternalities onrentsofretailestablishments.Letpijtbetherentpaidbyretailfirmi
inshoppingstreetjinyeart.Weusetwoproxiesforshopping
external-24 Giventhatthisdatasetonlycontainsrentaltransactions,itislikelythatthe
shareofmulti-storeownersisoverrepresentedinoursample,becausethisshare islikelylowerforownedshops.
ities𝑠𝑖𝑗𝑡={𝑓𝑖𝑗𝑡,𝑁𝑖𝑗𝑡}:footfallandthenumberofshopsinthevicinity. Furthermore,letxijtbeothershopandlocationcharacteristics(e.g.shop size,constructionyear,historicdistrictetc.).Thebasicequationtobe estimatedyields:
log𝑝𝑖𝑗𝑡=𝛼 log𝑠𝑖𝑗𝑡+𝛽𝑥𝑖𝑗𝑡+𝜃𝑡+𝜖𝑖𝑗𝑡, (10) where𝛼 and𝛽 areparameterstobeestimated,𝜃tareyearfixedeffects
and𝜖ijtisarandomerrorterm.
Thereareanumberofconcernswhenusingfootfallasaproxyfor shoppingexternalities.Onemaybeworriedthatnon-shoppersand one-stopshoppersmaybeincludedinfootfall.Arguably,thisisaminorissue becauseourmeasureoffootfallpredominantlyincludesshopperswho visitseveralshops.25
Anotherconcernismeasurementerror.Measurementerrorislikely presentforarangeofreasons.First,forfootfall,weonlyhavemeasures onspecificSaturdays,whichmaynotberepresentativefortheaverage Saturday,orforotherdaysoftheweek.Webelievethattheextentof measurementerrorhereisnottooserious.Second,thenumberofshops inthevicinitymayhavesubstantialmeasurementerror,becauseshops differsubstantiallyinsize.Inspecificationswithstreetfixedeffects,the biascausedbymeasurementerrorwithinthesamestreetmaybecome morepronounced.Mostlikely,wewillgetunderestimatesbecauseof measurementerror.Totakemeasurementerrorintoaccount,wewill useanIVapproach,asexplainedlateron.
Amoreseriousconcernispotentialsimultaneity.Rentsofshopsmay determinethetypeofshop,whereasshopsvaryintheextentoffootfall theygenerate;hencethereispotentiallyaneffectofrentsonfootfall. Furthermore,thelevelofrentmaydeterminethesizeofshopsand there-fore,thenumberofshopsinvicinity.Hence,thenumberofshopsmay alsobeendogenous.Moreover,theremaybeunobservedstoreand loca-tioncharacteristicsthatarecorrelatedwithfootfallandthenumberof shopsinthevicinity.Forexample,astore’sbuildingqualitymaybe im-portantforprofits.Whenbuildingqualityisnon-randomlydistributed overspace(e.g.nicerbuildingsinareaswithmorefootfallandshops) andcustomersvaluebuildingquality,anaïvehedonicregressionwould sufferfromomittedvariablebias.
Tomaketheidentificationofacausaleffectofshoppingexternalities moreplausible,wetakeanumberofsteps.First,weincludeshopping districtfixedeffects,implyingthatweidentifythedifferencesin foot-fallwithintheshoppingdistrict.Thisapproachmitigatestheproblem ofunobservedendowments,butmaynotsolvetheproblementirely be-causeshoppingdistrictsmaybequitelarge.Wethereforealsopropose anotheridentificationstrategyusingspatialvariationinlocalfootfall withinshoppingstreets.Wewillalsocontrolforweb-scrapeddataon picturestakenbyresidentsandtouristsandcountthenumberofpictures ineachshoppingstreet.Theideaisthatlocationswithanabundant sup-plyofaestheticamenitiesmayhaveahighpicturedensityandahigh numberofnon-shoppersorone-stopshoppers.26
Second,tomitigatetheissuethatspecificretailfirmsthatgenerate alotoffootfallmaynegotiatelowerrents,weuseinformationonthe 8-digitretailsectorandwhethertheretailfirmispartofachain.𝜓i∈kis
thenafixedeffectforeachretailsectorbychaincombination,denoted byk.Wethenestimate:
log𝑝𝑖𝑗𝑡=𝛼 log𝑠𝑖𝑗𝑡+𝛽𝑥𝑖𝑗𝑡+𝜑𝑗+𝜓𝑖∈𝑘+𝜃𝑡+𝜖𝑖𝑗𝑡, (11)
25Sincetheprobabilitythatashopperisincludedinourmeasureoffootfall
is(approximately)proportionaltothenumberofshopsvisited,theshareof non-shoppersorone-stopshoppersshouldbemuchsmallerthantheshareof multi-stopshoppers.Forexample,if25%offootfallwereone-stopshoppers, andtheother75%visitfourshops,thentheproportionofone-stopshoppers wouldbeonly7.8%.
26Ahlfeldt(2016)andGaigné etal.(2018)showthatthereisastrongpositive
where𝜑jareshoppingstreetfixedeffects.
Weimproveonthisspecificationbyestimatingspecificationswhere (i)weincluderetailfirmfixedeffectstofurthercontrolforunobserved retailfirm characteristics,(ii) onlyfocusonsmall shops,whereitis unlikelythattheretailfirmwillcontributesubstantiallytofootfall,and (iii)focusonfirmsthatarenotpartofalargerretailchain,whichare morelikelytoadvertiseandthereforetogeneratefootfall.Wealsoshow that(iv)whenflexiblycontrollingforthewalkingtimetothecitycentre, theresultsareessentiallyunaffected.
Still,onemaynotbefullyconvincedthatEq.(11)fullyaddressesthe issueofunobservedstoreandlocationalcharacteristics.Furthermore,it isplausiblethatreversecausationstillplaysarole.
Inordertoaddresssuchconcerns,wewill alsopursuean instru-mentalvariablesstrategy.Wewilluselocallylong-laggedinstruments, whichdonotaffectdirectlytoday’srents,butdeterminecontemporary concentrationofshopsand,therefore,footfall.Morespecifically, condi-tionalonshoppingstreetfixedeffects,wewillfirstusethenumberof cinemasin1930inthevicinityasaninstrumentforthecurrentnumber ofshopsinthevicinityandfootfall.
Weusecinemasin1930asaproxyforhistoricshops,becausewedo nothavehistoricdataonthelocationofshops.Historicshopswouldbe apreferredinstrument,becauseonemayarguethatchanginghistorical buildingsfordifferentuse(e.g.fromretailintoresidentialortheother wayaround)isacostlyprocessandrequireschangingzoningplans.Asa result,currentretailoutletstendtobelocatedinbuildingswhereretail was1930evenwithinthesamestreet.
Historically,mostcinemasweresmall, withonescreen only,and werelocatedinshoppingstreets(thosewhohavesurvivedtendtobe relativelylarge).Thebuildingscontainingsuchcinemasarenot very differentfromthesurroundingbuildingstypicallyusedforshops.One maybe concernedthatcinemasthemselvescreatefootfall,while this doesnotnecessarilyimplyshoppingexternalities(aspeoplemayonly visitthecinemaandnotvisitothershops).Hence,wecontrolforthe numberofcinemasinthevicinityin2010.27 Thebuildingsoftheclosed cinemasnowadaysarefrequentlyusedasshops,butalsoattractother businesses.
Themainidentifyingassumptionwhenrelyingonlong-lagged in-strumentsisthatpastunobservablecharacteristicsofeitherstoresor locationsareuncorrelatedtocurrentunobservables (whichaffect de-mand).Thisassumptionisoftencriticisedbecausethelocationsofcities aredeterminedlongago,e.g.duetonaturaladvantages.However,we identifytheeffectwithinshoppingstreets,making theidentifying as-sumptionmorecredible.28 Themostobviousthreattoidentificationis thatpastbuildingandlocationqualityarecorrelatedtoshopping exter-nalities.Wethereforecontrolforconstructionperioddummiesincluding adummywhethertheshopshasbeenbuiltbefore1930.Wealsoinclude adummyindicatingwhetherthepropertyisinahistoricdistrictto con-trolforthequalityof(historic)buildingsinthesurroundings.
Wealsoconstructanalternativeinstrumentbasedondataofthe ex-actlocationsofallbuildingsin1832,whichisavailableforabout50% ofourdata.Basedonoccupationalcensusoflandowners,wedetermine whetheracertainbuildingwasashop.Wethencalculatethenumber ofshopsin1832inthevicinity.Thecensusfrom1832alsoprovides informationontheso-calledCadastralIncome,whichwasataxbased ontheassessedvalueofeachparcelofthebuildingandthe
surround-27 Intheshoppingstreetswefocuson,therearecurrently180cinemas,but
in1930therewereabout60%more(315cinemas).Wecontrolforcurrent cin-emas,sovariationincinemascomesfromthepresenceofclosed,andlikely smaller,cinemas.
28 Gaigné etal.(2018)arguethatforDutchcities,theamenitydistribution
withincitieshasconsiderablychanged.Forexample,aroundthe1930s,open wateranddenselybuilt-upareaswerenotnecessarilyconsideredamenities.It wasalsobeforethetimewhencarsbecamethedominantmodeoftransport. Peopleusuallywalkedtotheirworkingplace,andthuscommutingdistances wereveryshort.
ingland.Wewillthereforecontrolforthevalueoflandin1832,which shouldaddresstheissuethatpastunobservablesarecorrelatedto cur-rentunobservables.Tofurthercontrolforthefactthatalocationisin ahistoricallydenserpartofthestreet(e.g.closertothecitycentre),we controlforthenumberofbuildingsin1832inthevicinity.
Hence,ouridentificationstrategyexploitsthatthereisareasonably strongautocorrelationofshoplocationsovertime.Thisautocorrelation ofretaillocationsisalsoinlinewithanecdotalevidence.Forexample, Amsterdam’scurrentmainshoppingstreet,theKalverstraat,hasbeenan importantshoppingstreetformorethan300years.29 Weemphasisehere thattheunobservedfactorswhichmakeshoplocationsattractivenow, arelikelyverydifferentfromthe(economic)forceswhichinducedshops tosettleat certainlocationshistorically:up toa hundredyearsago, shopperswalkedtothesestreets,andshopworkers,whowerealsoshop owners,frequentlylivedabovetheirshops.Nowadays,duetochanges in transporttechnology,shoppersdonot walktotheshoppingstreet (butonlywithin),whereasshopworkers,whoareemployeesandnot shopownersany more,come fromfarawayusingmoderntransport technologies(motorvehicles,publictransport,bicycle).
Thefirststageofourestimationprocedureentails:
log𝑠𝑖𝑗𝑡=̃𝛿 log𝑧𝑖𝑗+ ̃𝛽𝑥𝑖𝑗𝑡+̃𝛾ℎ𝑖𝑗+𝜑̃𝑗+ ̃𝜓𝑖∈𝑗+ ̃𝜃𝑡+𝜉𝑖𝑗𝑡, (12)
wherethe ∼ refer tofirst-stagecoefficients, zij iseither thenumber ofcinemasin1930orthenumberofshopsin1832inthevicinity.hij
arecontrolvariablesrelatedtothehistoricinstruments(e.g.numberof cinemasin2010,CadastralIncomein1832).Thesecond-stageisgiven by:
log𝑝𝑖𝑗𝑡=𝛼 loĝ𝑠𝑖𝑡+𝛽𝑥𝑖𝑗𝑡+𝛾ℎ𝑖𝑗+𝜑𝑗+𝜓𝑖∈𝑗+𝜃𝑡+𝜖𝑖𝑗𝑡, (13)
where loĝ𝑠𝑖𝑗𝑡={log𝑓̂𝑖𝑗𝑡,log𝑁̂𝑖𝑗𝑡} represents the predicted footfall or numberofshopsfromthefirststage.
4.2. Vacancies
Wewillalsoestimatetheeffectoflogfootfalaswellasthelog num-berofshopsinthevicinity,onwhethertheshopisvacant,indicatedby thedummyvariablevijt.Onceagain,wewilluseshoppingstreetfixed effectsandexploit thevariationin footfallandtheprobabilitytobe vacantwithinshoppingstreets.Forthesamereasonsthatfootfallmay beendogenouswithrespecttorent,itispossiblethatfootfallis endoge-nouswithrespectofvacancies.Furthermore,theremaybeadirectform ofreversecausation,becauseoneexpectsfootfalltodecreasewhen va-canciesincrease.Wewilluseagainlong-laggedhistoricinstrumentsto dealwithendogeneityissues,asitisimpossiblethatcurrentvacancies directlyimpactthenumberofcinemasin1930orthenumberofshops in1832.
5. Results
Theresultssectionisstructuredasfollows.Wefirstdiscusstheeffects offootfallandthenumberofshopsintheshoppingstreetonrents.In thesecondpart,wefocusontheeffectsoffootfallandthenumberof shopsontheprobabilityofashoptobevacant.Wetaketheseresults togethertoestimatetheeffectsoffootfallandnumberofshopsonrental income.Weclosethissectionbydiscussingthewelfareimplicationsand derivingthePigouviansubsidy.
5.1. Shoppingexternalitiesandrents
Table3reportstheresultsofourbaselineregressions.The specifica-tionincolumn(1)isanordinaryleastsquares(OLS)regressionofthe logrentalpriceonlogfootfall,logsizeoftheshop,building character-istics,andshoppingdistrictfixedeffects.Theelasticityoffootfallwith
Table3
Resultsforretailrents.
(Dependent variable: log of rent per m 2 )
Footfall Number of shops
(1) (2) (3) (4) (5) (6)
OLS OLS OLS OLS OLS OLS
Footfall (log) 0.3737 ∗∗∗ 0.3181 ∗∗∗ 0.2609 ∗∗∗ (0.0149) (0.0203) (0.0181)
Number of shops, < 200 m (log) 0.1786 ∗∗∗ 0.2066 ∗∗∗ 0.1513 ∗∗∗
(0.0214) (0.0359) (0.0324) Size of property (log) 0.6376 ∗∗∗ 0.6267 ∗∗∗ 0.5929 ∗∗∗ 0.6323 ∗∗∗ 0.6283 ∗∗∗ 0.5938 ∗∗∗ (0.0119) (0.0121) (0.0136) (0.0131) (0.0126) (0.0141)
Building – new 0.1341 0.0844 0.0877 0.0980 0.0684 0.0671
(0.0948) (0.0790) (0.0795) (0.0843) (0.0805) (0.0816) Building – renovated 0.4123 ∗∗∗ 0.3715 ∗∗∗ 0.3148 ∗∗∗ 0.3256 ∗∗∗ 0.3391 ∗∗∗ 0.2908 ∗∗∗ (0.1162) (0.0830) (0.0842) (0.0928) (0.0880) (0.0847)
Construction year dummies Yes Yes Yes Yes Yes Yes
Location characteristics No Yes Yes No Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes
Shopping district fixed effects Yes Yes Yes Yes Yes Yes
Shopping street fixed effects No Yes Yes No Yes Yes
Retail sector × chain fixed effects No No Yes No No Yes
Number of observations 4378 4378 4378 4378 4378 4378
R2 0.682 0.782 0.821 0.604 0.760 0.807
Notes :Footfallismeasuredasthenumberofshoppersperday.Theconstructionyeardummiesarecategorisedas follows:< 1832,1832–1930,1931–1950,1951–1960,1961–1970,1971–1980,1981–1990,1991–2000, > 2000. Locationcharacteristicsincludewhetherthepropertyisinahistoricdistrict,andthenumberofgeocodedpictures
< 200m,religiousbuildings< 200m,busstops < 200m,publicbuildings< 200m,schools,andrailwaystations
< 200m.Robuststandarderrorsareclusteredatthepostcodeandinparentheses.∗∗∗p < 0.01,∗∗p < 0.05,∗
p < 0.10
respecttorentalpriceis0.37.Thecontrolvariableshaveplausiblesigns, withlargerandrenovatedpropertiesbeingmoreexpensive.30
Thespecificationincolumn(1)mightsufferfromomittedvariable bias due to theomission of unobserved features of a shop location thatarecorrelatedwithfootfall.Forexample,someshoppingareasare moreattractivedue totheirproximitytoabusstop,schoolorother neighbourhood-specificamenities.Apartialsolutiontothisproblemis theinclusionofshoppingstreetfixedeffectsandlocationcharacteristics incolumn(2),whichmaymitigatethisendogeneityissue.Thishardly affectstheresults.
Onemaystillbeconcernedthatrentsarepartlydependentonthe amountoffootfallashopgenerates– sothatreversecausalityisanissue. Tomitigatethisissue,wewillincluderetailsector×chainfixedeffects. Thatis,weincludeadummyforwhetheraretailfirmispartofachain ineach retailsector,whichshould absorbmost oftheheterogeneity inexpectedsalesbetweenretailfirms.Thecoefficientincolumn(3)is somewhatsmallerandnowequals0.26.Thus,a10%increaseinfootfall isassociatedwitharentincreaseof2.6%.
Columns(4)–(6)inTable3focusesonanotherproxyforshopping externalities:thenumberofshopswithin200m.Whenusingshopping districtfixedeffects,theelasticityis0.179(seecolumn(4)).This elastic-ityisslightlyhigherwhenusingshoppingstreetfixedeffectsincolumn (5).Whencontrollingforretailsector×chainfixedeffects,the elastic-ityis0.151.Hence,a10%increaseinthenumberofshopsleadstoan increaseinrentsof1.5%.Thelatterelasticityisabout40%lowerthan theelasticitywithrespecttofootfall.However,thenumberofshopsin thevicinityasaproxyforshoppingexternalitiesmaybemeasuredwith error,e.g.becausewedonotknowtheexactrelevantspatialscaleof shoppingexternalities.Wewillthereforeinvestigatewhetherthis con-clusionstillholdswhenusinginstrumentalvariables.
Table4furtherinvestigateswhetherthepreviousresultsmakesense. Wefirstfocusonfootfall.Incolumn(1)weincluderetailfirmfixed
ef-30 Notethatthedummyvariableindicatingwhetherthepropertyisnewis
conditionalontheconstructionyear,sopartoftheeffectofbeinginanewer buildingisabsorbedbytheconstructiondecadedummies.
fects,implyingthatweincludedummiesforeachandeveryretailchain andindependentretailfirm.Thisimpliesthatwecomparerentand foot-falldifferencesofidenticalfirmswithinthesameshoppingstreet.The resultsstillindicateasizeableeffectoffootfall:a10%increasein foot-fallisassociatedwitha1.8%increaseinrents.Notsurprisingly,because ofthelargenumberoffixedeffects,theestimateislesspreciseandonly statisticallysignificantatthe5%level.Giventhestandarderror,the es-timateisnotstatisticallysignificantlydifferentfromthepreferred spec-ificationincolumn(3),Table3.
Incolumn(2)ofTable4,wecontrolflexibly(usinga5th-order poly-nomial)forthewalkingtimetothecentreoftheshoppingdistrict,by includingafifth-orderpolynomialofwalkingtimetothecentre.When one-stopshoppersareimportant,theymayaccessshoppingdistrictson onelocation(e.g.thecentreortheedgeofacitycentre)andthenwalk tothe intendedshop in thecentre,while passing manyother shops (Teulingsetal.,2017).Ontheotherhand,controllingforwalkingtime tothecentremayalsopartly absorbshoppingexternalities. Reassur-ingly,thefootfallcoefficientishardlyaffected.
Onemaystillbeworriedthatfootfallisinfluencedbytherentthat retailfirmshavetopay(e.g.duetorentspaidasafunctionofexpected retailrevenuesorexpectedgeneratedfootfall).Incolumns(3)and(4) we thereforeonlyincluderetailfirmsthatareunlikelytocontribute (substantially)tolocalfootfall.Column(3)onlyincludesshopsthatare smallerthan90m2(the25thpercentile).Theestimateisvirtually iden-ticaltothepreferredspecification.Alsowhenweonlyfocusonretail firmsthatarenotpartofachain,theestimateisessentiallythesame. Thisseemstosuggestthattheissueofreversecausalityislimited. Impor-tantly,italsostronglysuggeststhattheelasticityofrentswithrespect tofootfalldoesnotvarybetweentypesoffirms,i.e.heterogeneityofthe estimatesisabsent,atleastwhenwedistinguishbetweendifferentsizes offirmsorwhetherretailfirmsarepartofachain.Wecomebackto thisissuelateron.
ap-Table4
Footfallandretailrents:extensions.
(Dependent variable: log of rent per m 2 )
Retailer Walking time Small No chain Historic instruments: Historic instruments:
fixed effects to the centre shops shops cinemas in 1930 shops in 1832
(1) (2) (3) (4) (5) (6) (7) (8) (9)
OLS OLS OLS OLS 2SLS 2SLS 2SLS 2SLS 2SLS
Footfall (log) 0.1820 ∗∗ 0.2032 ∗∗∗ 0.2556 ∗∗∗ 0.2435 ∗∗∗ 0.4563 ∗∗∗ 0.4348 ∗∗∗ 0.4034 ∗∗∗ 0.3740 ∗∗∗ 0.4799 ∗∗ (0.0724) (0.0189) (0.0553) (0.0228) (0.1061) (0.0982) (0.1131) (0.1105) (0.2203)
Number of cinemas in 2010, < 200 m (std) 0.0107
(0.0115)
Cadastral income (log) − 0.0177 − 0.0213
(0.0236) (0.0256)
Cadastral income is zero in 1832 − 0.1822 − 0.2700
(0.2659) (0.3240)
Number of buildings in 1832, < 200 m (std) − 0.0461
(0.0868)
Shop characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes
Location characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes
Walking time to centre, f (·) No Yes No No No No No No No
Shopping street fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Retail firm fixed effects Yes No No No No No No No No
Retail sector × chain fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes
Number of observations 4378 4378 987 2932 4378 4378 2496 2496 2496
R2 0.969 0.826 0.873 0.792
Kleibergen-Paap F -statistic 42.65 48.68 31.46 32.67 17.95
Endogeneity test 4.858 4.426 0.790 0.795 1.153
𝜉2 (2) p -value 0.0275 0.0354 0.374 0.373 0.283
Notes :Footfallismeasuredasthenumberofshoppersperday.Shopcharacteristicsincludepropertysize,whetherthebuildingisneworrenovatedandconstruction yeardummies,whicharecategorisedasfollows:< 1832,1832–1930,1931–1950,1951–1960,1961–1970,1971–1980,1981–1990,1991–2000,> 2000.Location characteristicsincludewhetherthepropertyisinahistoricdistrict,andthenumberofgeocodedpictures < 200m,religiousbuildings < 200m,busstops < 200m, publicbuildings< 200m,schools,andrailwaystations< 200m.Robuststandarderrorsareclusteredatthepostcodeandinparentheses.∗∗∗p < 0.01,∗∗p < 0.05,∗
p < 0.10
pearstobestrongandmeaningful:astandarddeviationincreaseinthe numberofcinemasin1930increasescurrentfootfallby12%.31 The ef-fectoffootfallisnowsomewhatstronger(0.456),whichisstatistically differentfromthebaselineestimateusingaHausman-test,reportedat thebottomofthetable.32 Onemaybeworriedthatcinemasmostly at-tractone-stopshoppers.Wethereforecontrolforthenumberofcinemas in2010incolumn(6).Itcanimmediatelybeseenthatthecoefficient offootfallishardlyaffected.Moreover,thecurrentlocationofcinemas donotseemtogeneratehigherrents.
Columns(7)–(9)useanalternativesetofinstruments,byrelyingon datafromthe1832census.Becausethesedataareonlyavailablefor about50%of theNetherlands, ournumberofobservationsis lower. Column(7)indicatesastrongfirststage:astandarddeviationincrease inthenumberofshopsin1832increasesfootfallby35%.Thefirststage isagainstrong,albeitlessstrongthanwhenusingcinemasin1930asan instrument.Forthisspecification,theIVestimateisnotstatistically sig-nificantlyhigherthanthebaselineOLSestimate(theendogeneitytest hasap-valueof0.374).Themainadvantageofthe1832dataisthat wehaveinformationontheCadastralIncome,whichisagoodproxyof thevalueofpropertyin1832.Ofcourse,thereisafairshare(20%)of missingvalues,eitherbecauselandwasnot(commercially)owned(e.g. becauseitwasnotyetreclaimedfromthesea),oritwasmissing.We thereforeincludeadummyindicatingwhethertheCadastralIncomeis missing.Inanycase,controllingfortheCadastralIncomedoesnotmake
31 Thefirst-stageresultsarereportedinAppendixD.1.
32 Wehavealsore-estimatedthemodelbyredefiningtheinstrumentintoa
dummyvariable(wethenuseabinaryvariablefortheproximitytoacinema). WiththedummyinstrumenttheIVestimatesaresomewhatlower,andbased onaHausmantest,notstatisticallysignificantfromtheOLSestimates. Conse-quently,followingargumentsdevelopedbySargan(1958),andmorerecently emphasisedbyYoung(2019),onemayprefertheOLSestimates,asthesehave beenmorepreciselyestimated.
anydifference,astheCadastralIncomedoesnotseemtoberelatedto currentrentsofretailproperties.33 Hence,theassumptionthat unob-servedendowmentswithinshoppingstreetsofthepastareuncorrelated tocurrentunobservedstoreandlocationcharacteristics,isevenmore plausible.Inthefinalspecificationwealsocontrolforthenumberof buildingswithin200m,tomakesurethatwedonotjustmeasurethat certainshopsareinlocations thataredenser(e.g.citycentres). Col-umn(9)showsthatitisindeedthelocationsshopsin1832thatmatter forcurrentfootfall,astheeffectofnumberofbuildingsisstatistically insignificant.Thecoefficientalsobecomessomewhatstronger,but be-causeofthewideconfidencebands,itisnotstatisticallysignificantly differentfromthebaselineOLSestimate.
InTable5werepeatthesamesetofspecifications,butnowusing theothermeasureforshoppingexternalities:thenumberof shopsin thevicinity.Whenusingretailfirmfixedeffects,orwhencontrolling flexiblyforthewalkingtimetothecentreoftheshoppingdistrict,the estimatesareabout50%lower(respectivelycolumns(1)and(2)). How-ever,whenfocusingonsmallstoresorwhenonlyincludingchainstores, theelasticitiesarecomparabletothebaselineestimates(respectively columns(3)and(4)).Again,theresultsstronglysuggestanabsenceof between-firmheterogeneityintheelasticityofrentwithrespectto shop-pingexternalities.
Incolumns(5)–(9)weusehistoricinstruments.Inallspecifications, theimpactofnumberofshopsismuchhigherwhenusingIVthanwhen using OLS.Inmostspecifications,endogeneitytestsindicate thatthe estimatesusinginstrumentsaresignificantlyhigher.Animportant ob-servationisthattheIVestimatesfortheelasticityofnumberofshopsare verycomparabletoIVestimatesfortheelasticityofrentwithrespectto
33Furthermore,theunconditionalcorrelationbetweenlogpriceandlog