ContentslistsavailableatScienceDirect
Journal of Urban Economics
journalhomepage:www.elsevier.com/locate/jue
The amplifying effect of capitalization rates on housing supply ✩
Simon Büchler
a,b, Maximilian v. Ehrlich
b, Olivier Schöni
b,c,∗aMIT Center for Real Estate
bCenter for Regional Economic Development (CRED), Department of Economics, University of Bern
cDepartment of Finance, Insurance and Real Estate, Laval University
a r t i c le i n f o
JEL classification:
R 1 R 3 R 5 Keywords:
Housing supply Capitalization rate Land use regulation Geographic constraints
a b s t r a ct
Weprovideempiricalevidencethatincreasesinhousingrentalincomeleadtoalargersupplyresponsethanprice increasesofthesamepercentagevalue.Werationalizethisdifferentialinsupplyresponsivenesswithanampli- ficationmechanismarisingfromadownwardrevisionofcapitalizationratesfollowingarentalincomeincrease.
Wedocumentthattheamplificationofthehousingsupplypriceelasticityislesspronouncedingeographically constrainedandtightlyregulatedneighborhoodsandareashavingmoresophisticatedinvestors.Ourfindings holdvaluablelessonsforpublicpoliciesaffectingthehousingrentalincome,suchasrentcontrolandhousing subsidies.
1. Introduction
Existingresearchemphasizestheimportanceofhousingsupplyprice elasticityfora varietyof economicoutcomes.Theresponsivenessof housingsupplyaffects,amongotherthings,housingcycles,theallo- cationoflaboracrossspace,andthedegreeofcapitalizationofpublic policiessuchasplace-basedsubsidies.1However,todate,weknowrel- ativelylittleabouttheresponsivenessof housingsupplywithrespect tochangesinrents.Thisissurprising,asurbaneconomictheorytypi- callyfocusesonperiodichousingcosts,andmanypublicpoliciessuch asrentcontrolandhousingsubsidiesdirectlyactontherentalincome generatedbyrealestateproperties.Inthispaper,weinvestigateunder whichcircumstancesthehousingsupplyresponsivenesstochangesin rentsdiffersfromthesupplyresponsivenesstopricechangesandwhy theratioofrentandpriceelasticitiesvariesacrossregions.
Westartbydevelopingapartialequilibriumframeworkfeaturing housingsupplyanddemandaswellasrealestateinvestors.Thethe-
✩ WearegratefultoBrentAmbrose,PaulAnglin,NathanielBaum-Snow,AymoBrunetti,GuillaumeChapelle,DavidGeltner,LuHan,ChristianHilber,Albert Saiz,AlexvandeMinne,AnnieKinsellaThompson,BillWheaton,PaulWillen,Jeff Zabelandtwoanonymousrefereesforveryhelpfuladviceandsuggestions.We benefitedfromnumerouscommentsbytheparticipantsoftheBostonFEDUrbanandRealEstateSeminar,theUniversityofSt.GallenResearchSeminar,2019UEA inPhiladelphia,2018UEAinNewYork,2018SSES,MITCREResearchSeminar,SRERCZurich2017,CREDResearchSeminar,aswellasfromthemonitoringgroup atSECOandBWO.WethankMeta-Sys,Comparis,SwissFederalStatisticalOffice,SwisscantonalofficesandComparisforgenerouslysharingtheirdata.Ehrlichand SchöniacknowledgefundingfromtheSwissNationalScienceFoundation(grantnumber162589).
∗Correspondingauthorat:DepartmentofFinance,InsuranceandRealEstate,LavalUniversity,2325RuedelaTerrasse,G1V0A6,Québec.
E-mailaddresses:buechler@mit.edu(S.Büchler),maximilian.vonehrlich@vwi.unibe.ch(M.v.Ehrlich),olivier.schoni@fsa.ulaval.ca(O.Schöni).
1 Forinstance,Glaeseretal.(2008)investigatetheroleofhousingsupplyelasticitiesforpricedynamics,Diamond(2017)linksthedegreetowhichlocalgovern- mentscanextractrentstohousingsupplyelasticities,KlineandMoretti(2014)emphasizetheimportanceofhousingsupplyelasticityforthedistributionaleffects ofplace-basedpolicies,andHsiehandMoretti(2019)focusontheimplicationsofhousingsupplyconstraintsforthespatialmisallocationoflabor.SeeGlaeserand Gyourko(2018)andHilber(2017)forasynthesis.
oreticalframeworkservestwopurposes.First,itguidesourempirical analysisand, inparticular,motivates theidentificationstrategyused toestimatehousingsupplyelasticities.Second,itallowsustorational- izedifferencesintheestimatedsupplyresponsivenesstorentandprice dynamics.Weshowthatlocalchangesininvestors’expectationsabout rentalincomegrowthandriskpremiaaredecisivetoexplainsupplydif- ferences.Specifically,thehousingsupplyresponsetochangesinprices andrentsisidenticalifcapitalizationratesdonotadjusttochangesin rents.Ifcapitalizationratesdoadjust,housingsupplypriceelasticities areeitheramplifiedordampenedbythisadjustment.
Next,webringthetheoreticalframeworktothedataandestimate theaverageresponsivenessofhousingsupplywithrespecttopriceand rentchangesattheneighborhoodlevel.Todoso,weusedetailedgeoref- erenceddataonadvertisedresidentialpropertiesandbuildingstockfor Switzerlandcoveringtheperiod2005–2015.Wefindthatanincrease in rentsleadstoanaboutthriceaslargesupplyincreasethananin- creaseofpricesofthesamepercentagevalue:Thesupplyresponsefol-
https://doi.org/10.1016/j.jue.2021.103370
Received29February2020;Receivedinrevisedform14June2021 Availableonline3July2021
0094-1190/© 2021TheAuthors.PublishedbyElsevierInc.ThisisanopenaccessarticleundertheCCBYlicense(http://creativecommons.org/licenses/by/4.0/)
lowingatenpercentincreaseinsquaremeterrents(prices)isapprox- imately14percent(fourpercent).Accordingtoourframework,these resultssuggestthat,onaverage,realestateinvestorsrevisecapitaliza- tionratesdownwardfollowingapositivedemandshock.Thisrevision, inturn,amplifiesthesupplyresponsetopricechanges.Wedocument thatthesupplyamplification,andthecorrespondingadjustmentofex- pectations,areheterogeneousacrossspace.Geographicallyconstrained andtightlyregulatedmarkets– i.e.,majorurbanareasandalpinetourist areas– andneighborhoodshavingmoresophisticatedbuyersasproxied bylandlords,institutionalinvestors,orsecond-homeinvestorsdisplay loweramplificationvalues.
Switzerlandisanexcellentlaboratorytoinvestigatehousingsupply duetosubstantialheterogeneityinthelocalfactorsinfluencingit.The decentralizedformofgovernmentgrantslow-tier politicalunits(mu- nicipalities)large autonomyin landuse planningandfiscalpolicies.
Geographicfeaturesofthelandscape,suchaselevation,slope,andter- rainruggedness,alsovaryconsiderablyacrossspace.Thesecharacter- isticsofthecountrymakethereactionofhousingsupplycontingenton localizedfactors.Importantly,theowner-occupiedandrentalmarkets areapproximateofequalsizeinSwitzerland,whichfacilitatesthees- timationofrentalandpricesupplyelasticitiesthroughoutthecountry, therebyallowingustostudytheroleofcapitalizationratesforhousing supply.2Finally,theexistenceofdetailedinformationonproperty-level housingcharacteristicsallowsustoruleoutthatdifferentattributes– i.e.,aqualitygap– betweenpropertiesdrivethedifferencesinsupply elasticities.
Ourpaperbridgestwostrandsoftheliterature.Thefirststrandfo- cusesontheestimationoflocalhousingsupplypriceelasticities.Despite theimportanceofhousingsupplyelasticity,papersquantifyingitremain scarce.GyourkoandMolloy(2015)provideacomprehensivereviewof theliteratureinvestigatingtheestimationanddeterminantsofhousing supply.Inhisseminalarticle,Saiz(2010)estimateshousingsupplyelas- ticitiesacrossU.S.MetropolitanStatisticalAreas(MSAs)asafunctionof geographicandregulatoryconstraints.UsingaVectorErrorCorrection Model,Wheatonetal.(2014)alsoestimatehousingsupplyelasticities forU.S.MSA’s,obtainingestimatesinlinewiththoseofSaiz (2010). Baum-Snow andHan, 2021adopt a structural approachto quantify cross-andwithin-cityhousingsupplyelasticitiesforU.S.metropolitan areas,showingthathousingsupplyelasticitiesincreasemonotonically withthedistancetocitycenters.WhilerelyingonSaiz(2010)empirical specificationforestimatinginversehousingsupplyelasticities,wefol- lowasimilaridentificationstrategyasinBaum-SnowandHan,2021. Specifically,weidentifyhousingsupplyelasticitiesusinglocaldemand shockstriggeredbythehistoricspatialdistributionofsectoralemploy- mentsharesandtheconnectivityofaneighborhoodwithlocallabor markets.Asacomplementaryshift-shareinstrument,weusethehistor- icaldistributionof languagesharesacross Swissneighborhoodscom- binedwiththegrowthoflanguagegroupsataggregatelevels.
Thesecond strandof theliteraturefocuses onhowinvestors’ ex- pectations impact market dynamics. Time variation in expectations – as captured by price to rent ratios dynamics – have been used byCase andShiller(2003)toidentifybubble-like behavior.Bycon- structing a user cost model incorporating economic fundamentals, Himmelberget al.(2005) show that expectedhouse price apprecia- tionplaysanimportantrolein explaininglocal U.S.pricedynamics.
FocusingontheU.S.housingboomintheearly2000s,Ben-Davidetal., 2019showthatagentsinvest invacanthomesiftheyexpectfurther priceincreases.Kaplanetal.(2020)highlighttheroleofexpectations forhousepriceandrentmovementsaroundtheGreatRecession.
Theimportanceof expectationsformarketdynamics hassparked interest in the way individuals form expectations. Mayer and Sinai (2007) investigate theeffect of backward-looking expectations
2 Similarhomeownershipratesareobservede.g.,inAustria,Germany,and SouthKorea.
inhousepricebooms.Sivitanidesetal.(2010)findthatcapitalization rates behavesimilarlytoprice/earningsratios,witheconomicagents formingpricegrowthexpectationsbasedonpastdynamics.Glaeserand Nathanson,2017constructamodelwherebuyersarenotentirelyratio- nalinpredictingfuturepricedynamics,whichexplainsobservedprice correlationovertime.KuchlerandZafar(2019)showthatindividuals formexpectationsabouthousepricedynamicsfromrecentpersonalex- periences.Thefactthatinvestorsaremyopic,orbackward-looking,isin linewithourfindings,whichsuggestthatinvestorsexpectfurtherrent growthinlocationsthathaveexperienceddemandincreasesinthere- centpast.Usingspatialvariation,weshowthatpartofthelocalhousing supplyresponsetodemandshocksisaffectedbychangesininvestors’
expectations.
Wealsorelatetotheliteratureonlocalgeographicandregulatory constraints(e.g.,AuraandDavidoff,2008,HilberandVermeulen,2016, andLutzandSand,2019)andshowthatpricesupplyelasticitiesarede- terminednotonlybyregulationandgeographicconstraintsbutalsoby adjustingcapitalizationrates.3HilberandMense(2021)showthatla- bordemandshocksinconjunctionwithsupplyconstraintsexplainmost oftheincreaseintheprice-to-rentratioinGreaterLondonoverthelast twodecades.
Ourcontributiontotheliteratureisthreefold.First,weempirically establishalinkbetween housingsupplyresponsivenessandinvestors expectations.Thislinkisessential,aspublicpoliciesaffectingrentalin- comemightleadtounanticipatedconsequencesinthesupplyofhousing duetochangesinexpectations.Second,ourempiricalanalysisquantifies thespatialdynamicsoflocalexpectations.Specifically,weprovidenovel evidencethattheadaptationofinvestorsexpectationsoccursatthelocal levelandthatsuchadjustmentisconsistentwithapath-dependentview ofspatialdevelopment.Investorsexpectthatplacesthathavegainedin attractiveness(i.e.,experiencedapositivedemandshock)willcontinue todoso,asreflectedbyadecreaseincapitalizationrates,leadingtoad- ditionalhousingdevelopment.Third,weshowthatthehousingsupply elasticityvariesconsiderablywithinandacrossurbanareasduetothe fine-scaleimpactofgeographicandregulatoryconstraints.Thisvaria- tionleadstoaspatiallyheterogeneouscapitalizationofglobaldemand shocksthatcannotbeobservedwhenestimatingthehousingsupplyelas- ticityattheurbanarealevel,asdonebypreviousresearch.
Theremainderof thepaperisstructuredasfollows.Section2in- troduces the conceptual framework motivating our empirical analysis. Section 3 explains our empirical identification strategy.
Section4presents thedataandprovidesdescriptivestatistics forthe Swisshousingmarket.WediscusstheresultsinSection5andprovide severalrobustnesschecksinSection6.Section7concludes.
2. Conceptualframework
Thefollowingpartialequilibriumframeworkallowsustoformal- izetheidentificationassumptionsunderlyingtheempiricalanalysisand rationalizecorrespondingfindings.Wespecifythesupplyanddemand sideofthehousingmarketandoutlinetheroleplayedbyrealestate investors.4
2.1. Housingdevelopers
AsinGlaeser(2008),ineachneighborhood𝑛,housingdevelopers choosetheamountofhousingspacetodevelop.Tobuildhousing,de- velopersmustpaytheprice 𝑃𝑛𝑙𝑎𝑛𝑑 toacquireland,andthelocalcon- structioncost𝑐𝑛 topurchasebuildingmaterialsandremuneratelabor.
3Relatedly,Solé-Ollé andViladecans-Marsal(2012)analyzetheroleofpo- liticalcompetitionforresidentialdevelopment.LinandWachter,2020docu- mentspillovereffectsoflocalregulatoryconstraintsonneighboringlocalities, Cosmanetal.(2018)analyzehousingappreciationandmarginallandsupplyin adynamicframework.
4OnlineAppendixApresentsamoredetailedderivationofthemodel.
Withoutlossofgenerality,wecapturebothlocalproductivityofthecon- structionsector,aswellastheunitcostofinputs,in𝑐𝑛.Thedevelopers profitoptimizationproblemisgivenby
maxℎ (𝑃𝑛ℎ𝑙−𝑐𝑛ℎ𝛿𝑖𝑛𝑡𝑛+1𝑙−𝑃𝑛𝑙𝑎𝑛𝑑𝑙), (1)
where𝑃𝑛isthelocalpriceofhousingperunitoflivingspace,𝑙denotes theamountoflandandℎthebuildingheight.Weassumethatthecost component𝑐𝑛ℎ𝛿𝑛𝑖𝑛𝑡+1𝑙isconvexwithrespecttobuildingheight(𝛿𝑛𝑖𝑛𝑡>0), describingthefactthattheconstructionoftallerbuildingbecomespro- gressivelycostlierduetogeographicandregulatoryconstraintslimiting residentialdevelopmentontheintensivemargin.5
Developerschoosetheoptimalintensityofdevelopmentforeachho- mogeneousunitoflandintheneighborhood.Thezero-profitcondition ismetandgovernstheequilibriumpriceofland𝑃𝑛𝑙𝑎𝑛𝑑.Totalhousing supplyis givenbytheproductofoptimalbuildingheightℎ∗𝑛 andthe amountofdevelopableland𝐿𝑛availableintheneighborhood.Weas- sumethatthequantityoflandavailableforresidentialdevelopmentin theneighborhoodrespondsendogenouslytohousingpricesaccording to𝐿𝑛= ̄𝐿𝑛𝑃
1 𝛿𝑒𝑥𝑡𝑛 + 1
𝛿𝑖𝑛𝑡𝑛 𝛿𝑒𝑥𝑡𝑛
𝑛 ,where ̄𝐿𝑛 capturescharacteristicsoflocations shiftinglandsupply,andtheparameter𝛿𝑒𝑥𝑡𝑛 governsthe(inverse)re- sponsivenessofresidentiallandavailability(seeOnlineAppendixA).
Totalhousingsupplyinaneighborhoodisthengivenby 𝑄𝑠𝑛=ℎ∗𝑛𝐿𝑛=
( 𝑃𝑛 (𝛿𝑖𝑛𝑡𝑛 +1)𝑐𝑛
) 1
𝛿𝑖𝑛𝑡𝑛 ̄𝐿𝑛𝑃
1 𝛿𝑒𝑥𝑡𝑛 + 1
𝛿𝑖𝑛𝑡𝑛 𝛿𝑒𝑥𝑡𝑛
𝑛 =𝑆𝑛𝑃𝑛𝜖𝑛𝑄,𝑃, (2)
where 𝑆𝑛=𝑆𝑛(̄𝐿𝑛,𝑐𝑛,𝛿𝑖𝑛𝑡𝑛 ) summarizes exogenous housing supply shifters.Thestructuralparameter𝜖𝑄,𝑃𝑛 = 1
𝛿𝑛𝑖𝑛𝑡+ 1
𝛿𝑛𝑒𝑥𝑡+ 1
𝛿𝑛𝑖𝑛𝑡𝛿𝑒𝑥𝑡𝑛 ≥0corre- spondstothelocalhousingsupplyprice-elasticity,whichdependson thelocalresponsivenessofresidentialdevelopmentontheintensive(𝛿𝑖𝑛𝑡) andextensive(𝛿𝑒𝑥𝑡)margin.6
2.2. Realestateinvestors
We build on the framework proposed by DiPasquale and Wheaton(1992)andassumethatinvestorsarewillingtopayasquare meterprice𝑃𝑛forapropertygeneratingaperiodicrentalincome𝑅𝑛in neighborhood𝑛.Investorsthusmediatebetweentheproperty market -inwhichhouseholdsconsumehousingservices-andhousingdevel- opers.7Weassumethattheelasticityofbuildingtenurewithregardto capitalizationratedifferencesbetweenowner-occupiedandrentalprop- ertiesisinfinite.Putdifferently,investorsoptimallychoosewhetherto sellor rentoutaproperty,whichimpliesthatthecapitalizationrate ofrentalandsellingpropertiesisthesame.Ifitwerenotso,arbitrage opportunitieswouldarise,leadinginvestorstoshifttheirdemandfrom onerealestateassettotheother.8
Wedepartfromtheliteraturebyassumingthatinvestorsformex- pectationsendogenouslyaboutlocalrisk-adjustedreturns𝑟𝑛 andrent growth 𝑔𝑛 according to observed contemporaneous rents and prices,
5 Tallandhigh-risebuildingstypicallyrequirespecificbuildingmaterialsand specializedworkers,suchasarchitectsandengineers,thatensurethestabilityof itsstructure.Additionally,geographicandregulatoryconstraintsbecomemore binding,astheyaremorelikelytohinderverticaldevelopment.
6 Itiscommonintheliteraturetorepresenthousingsupplyelasticitywith asinglestructuralparameter𝜖enteringahousingsupplyfunctionoftheform 𝑄𝑠=𝑆𝑃𝜖,seee.g.,HsiehandMoretti(2019),Baum-SnowandHan,2021,and LinandWachter,2020.
7 Inthecaseofowner-occupancy,theinvestorrentsouttherealestateasset toherself.
8 Weassumethatrentalandsellingunitsare,onaverage,identicalwithina givenneighborhood,suchthatinvestors’expectationsarethesameduetothe no-arbitragecondition.InSection3.1,wethuspartialoutpotentialqualitydif- ferencesfromrentandpricedynamics.Theno-arbitrageassumptionisatthe coreofthestandardusercostapproachemployedbyHendershottandSlem- rod(1983),Poterba(1984),andMayerandSinai(2007).
i.e., 𝑟𝑛=𝑟𝑛(𝑅𝑛,𝑃𝑛) and𝑔𝑛=𝑔𝑛(𝑅𝑛,𝑃𝑛). Across periods, investors up- datetheirexpectationsbasedon thecapitalinvestmentthey haveto makeandthecorrespondingrentalincometheycouldpotentiallyearn at that time, thus leadingto heterogeneousexpectation adjustments across neighborhoods.Thisleads totheformula𝑃𝑛= 𝑅𝑛
𝑖𝑛(𝑅𝑛,𝑃𝑛), where 𝑖𝑛=𝑟𝑛(𝑅𝑛,𝑃𝑛)−𝑔𝑛(𝑅𝑛,𝑃𝑛)isthelocalcapitalizationrate.
Using thissimple framework,we can analyzethe propagationof rentalincomechangestothesupplyof housing.Therelativerespon- sivenessofhousingsupplytorentchangesisgivenby
𝜖𝑛𝑄,𝑅= 𝑅𝑛
𝑄𝑠𝑛 𝑑𝑄𝑠𝑛 𝑑𝑅𝑛 = 𝑃𝑛
𝑄𝑠𝑛 𝑑𝑄𝑠𝑛 𝑑𝑃𝑛
𝑅𝑛
𝑃𝑛 𝑑𝑃𝑛
𝑑𝑅𝑛 =𝜖𝑛𝑄,𝑃𝜖𝑃 ,𝑅𝑛 , (3) where𝜖𝑄,𝑃𝑛 isthestandardhousingsupplypriceelasticityinEq.(2),and 𝜖𝑛𝑃 ,𝑅isanamplificationcoefficientthatcorrespondstothepriceelasticity withrespecttorentchanges.Thislatterisdeterminedbytheresponsive- nessofthelocalcapitalizationratetorentchanges,i.e.,𝜖𝑛𝑃 ,𝑅=1−𝜖𝑖,𝑅𝑛 (seeOnlineAppendixA).Eq.(3)tellsusthathousingsupplyresponses torent andprice changesdifferwhentheelasticityofpricestorent changesisnotunitary.Ifthevaluationoflocalrealestateassetsisvery sensitivetolocalrentchanges,i.e.,𝜖𝑛𝑃 ,𝑅>1,thehousingsupplywill respondmorestronglytorentchangesthantopricechanges.Investors adjustmentofgrowthexpectationsandlocalriskcapturedbythecap- italizationratedeterminetheelasticityoflocalpricestolocalrents.If these factorswereindependentof rentdynamics, weshould observe identicalsupplyresponsestorentandpricechanges,whichisacentral hypothesiswetestempirically.9
Aparametrizationoflocalcapitalizationratesisinstructivetopro- videanintuitionaboutthewaywedifferfromtheliteratureandtoun- derstandtheidentificationassumptionsexposedinthenextsection.Let usassumethat𝑖𝑛=𝑖0𝑅𝛾𝑛𝑛𝑅𝑃𝑛𝛾𝑃𝑛,where𝑖0isthe“standard” capitalization rate,whichtheliteratureusuallyassumestobeexogenouslydetermined bycapitalmarkets.10Theparameters𝛾𝑛𝑅and𝛾𝑛𝑃representthelocalelas- ticityofcapitalizationrateswithrespecttorentandpriceshocks.Itis easytoshowthattheamplificationcoefficientispinneddownbythese twoparametersviatheequation𝜖𝑛𝑃 ,𝑅=1−𝛾𝑛𝑅
1+𝛾𝑛𝑃.Thisparametrizationpar- simoniouslyendogenizescapitalizationrateswhileallowingforspatial differencesintheinvestors’discountratewhenevaluatingrealestate assets.Ifweset𝛾𝑛𝑅=𝛾𝑛𝑃 =0,weobtainthestandardGordongrowth model.Empirically,wetestwhetherthisspatialgeneralizationof the Gordongrowthmodelismeaningful.11
2.3. Residents
The economyis endowed witha continuous measure of 𝑁 indi- vidualsdistributedacrossneighborhoods.Buildingonrecentworkby Monteetal.(2018),eachindividualworkinginindustry𝑘decidesin whichneighborhood𝑛toliveandinwhicharea𝑖towork.Theidiosyn- craticindirectutility𝑈𝑛𝑖𝑘 ofindividual𝜔isgivenby
𝑈𝑛𝑖𝑘(𝜔)=𝑏𝑘𝑛𝑖(𝜔)𝑊̃𝑛𝑖𝑘
𝑅𝛼𝑛 , (4)
9NotethatEq.(3)isvalidforanysupplyfunctionwhosepriceelasticityis describedbyasingleparameter.ThestructureimposedbyEq.(2)onlyservesthe purposeofillustratingtheidentificationassumptionsunderlyingtheempirical estimation.
10Ourresultsgeneralizetoaparametrizationthatallowsforlocalexogenous capitalizationrates𝑖0𝑛.InSection6,wechecktherobustnessofourresultswhen 𝑖0𝑛includeslocalmeasuresofliquidityrisksanduncertaintyintherevenuegen- eratedbytheproperty.
11Theexistingliteratureoncapitalizationratesempiricallydocumentsastrong heterogeneityincapitalizationratesacrossspace,withurbanandhigh-amenity areastypicallydisplayinglowercapitalizationrates.However,theexistingur- banliteraturelargelyneglectssuchdifferences.Ourparametrizationaccommo- datessuchfeatures.
wherewe setthepriceof thetradablenuméraireequaltounityand assumethatindividualsspendashare𝛼onhousing.Thevariable𝑊̃𝑛𝑖𝑘 denotestheindustry-specificwageofworkerslivingin𝑛andcommuting to𝑖.
Theutilitycomponent𝑏𝑘𝑛𝑖capturesidiosyncraticpreferencesthatdo notdependonmarketfundamentalsbut,rather,ontheexogenoustastes ofworkersforagivenplaceofresidence/placeofworkcombination.We assumesuchpreferencestobei.i.d.realizationsofaFréchet-distributed randomvariablewithscaleparameter𝐵𝑛𝑖𝑘 andshapeparameter𝜀𝑘>1. Thegreaterthevalueof𝜀𝑘,thelessheterogeneousarelocationalpref- erencesofworkersinagivenindustry,thusimplyinggreatermobility acrossspace.
Wemodel𝑊̃𝑛𝑖𝑘aswagepereffectiveunitsoflabor𝑊𝑛𝑖𝑘dividedby commutingcosts𝑚𝑛𝑖,implyingthatworkersreducelaborsupplywhen commutingfromdistantlocations.Ourfocusbeingonhousingmarkets, wedonotexplicitlymodelthedemandsideoflabormarkets,andcon- siderwages𝑊𝑛𝑖𝑘asanexogenousvariable.12
Givenhouseholdshomotheticpreferences,totalhousingdemand𝑄𝑑𝑛 inneighborhood𝑛isgivenby
𝑄𝑑𝑛 =𝛼𝑊̄𝑛
𝑅𝑛𝑁𝑛=𝛼 1 𝑅𝛼𝑛(1+𝜖𝑘)
𝑁 Φ
∑
𝑖,𝑘𝐵𝑘𝑛𝑖 (𝑊𝑛𝑖𝑘
𝑚𝑛𝑖 )𝜖𝑘+1
, (5)
where𝑊̄𝑛= 𝑁1
𝑛
∑𝑖,𝑘𝑊𝑛𝑖𝑘
𝑚𝑛𝑖𝑁𝑛𝑖𝑘 istheweightedaverageincomeearnedin neighborhood𝑛,𝑁𝑛=∑
𝑖,𝑘𝑁𝑛𝑖𝑘 isthetotalnumberofhouseholdsliving in𝑛,andΦisacompositetermreflectingtheattractivenessofallother possiblepairsofresidence𝑟andemployment𝑠.13Eq.(5)providestwo insightsthatproveusefulwhenconstructinghousingdemandshifters.
Specifically,weshouldexpecthigherhousingdemandinneighborhoods thatarei)betterconnectedtoproductiveareas,andii)attractivealong oneofthedimensionscapturedbyidiosyncratictastes𝐵𝑘𝑛𝑖.InSection3, wederivetwoinstrumentsthatcaptureshiftsinhousingdemandtrig- geredbythesetwodimensionswhileremainingexogenouswithrespect tohousingsupplychanges.
2.4. Demandshocks,expectations,andhousingsupply
Inthissection,weoutlinecomparativestaticresultsonhowanex- ogenousdemandshockinformsusaboutchangesinhousingsupplyand adjustmentsofexpectations.Ininterestofparsimony,letusdenoteby 𝜃anylocalshockexclusivelyaffectingthedemandsideofthehousing marketasdescribedbyEq.(5),andassume thattheperiodiccostof housingservices– equilibriumrents– increasefollowingthisdemand shock,i.e.𝜕𝑅𝜕𝜃𝑛
𝑛 >0.
UsingEq.(2),thehousingsupplyresponsetoademandshock is givenby
𝜕𝑄𝑠𝑛
𝜕𝜃𝑛 = 𝑆𝑛𝜖𝑄,𝑃 𝑖𝑛 𝑃𝑛𝜖𝑄,𝑃−1
(𝜕𝑅𝑛
𝜕𝜃𝑛 −𝑃𝑛𝜕𝑖𝑛
𝜕𝜃𝑛
)
. (6)
Theterm𝑃𝑛𝜕𝜃𝜕𝑖𝑛
𝑛isresponsiblefortheamplificationofhousingsupply viachangesincapitalizationratesfollowingapositivedemandshock.14 Wearguethattheresponseoflocalcapitalizationrates toapositive demandshockisunequivocallynegative,i.e.𝜕𝑖𝑛
𝜕𝜃𝑛 <0,thusincreasingthe supplyofhousingwithrespecttothestandardGordonGrowthmodelin whichcapitalizationratesareexogenouslyfixedandonlythefirstterm inparenthesesremains.
12 Indeed,theidentificationstrategyexposedinSection3reliesonexogenous changesinlocallabordemand.Forthisreason,werefraintomodellaborde- mandendogenously.
13 SeeOnlineAppendixAforadetailedderivationofEq.(5).
14 Tokeepthenotationassimpleandgeneralaspossible,herewerefrainfrom explicitlyformalizingtheendogenousrelationshipbetweencapitalizationrates 𝑖𝑛andrents𝑅𝑛– whichareaffectedbythedemandshock– andsimplywrite
𝜕𝑖𝑛
𝜕𝜃𝑛.
This negativerelationshiparisesbecause,followingapositive de- mandshock,i)theperceptionoflocalrisklikelydecreases(𝜕𝑟𝑛
𝜕𝜃𝑛 <0), and/orii)rentgrowthexpectationsincrease(𝜕𝑔𝑛
𝜕𝜃𝑛>0).15Forexample, anunexpecteddemandshock𝜃capturingimprovementsoflocalameni- ties,orarelocationofasufficientlylargecompany,leadstochangesin localhousingdemandandthusalterinvestors’expectations.
Expectations arelikely to be formedaccording topastdynamics (see e.g.DiPasqualeandWheaton,1995). Our resultsare consistent with myopic expectation as documented for real estate markets by KuchlerandZafar(2019).Begleyetal.(2019)documentthathetero- geneousexpectationsaboutlocalpopulationgrowthleadtodifferences inlocalprice-rentratiosintheUS.Aslongasthelocaldemandincrease hassome unexpectedelement, myopicinvestors willadjusttheir ex- pectationsaboutthegrowthratesofrentsupwardsandthusdecrease capitalizationrates.
Theequilibriumfeatureofrentgrowthratesmakesitextremelydif- ficulttopredictthem,evenwheninvestorshavestrongpredictivecapa- bilitiesregardingfuturedemandchanges.First,housingsupplyreacts endogenouslyatthelocalleveltothedemandshock.Second,evenif ademandshockisinitiallylocalizedtoagivenneighborhood,itwill spreadthroughoutthecountryduetothespatialequilibriumcondition.
Thismakesitverydifficulttocorrectlypredictthe“final” effectofade- mandchangebasedonpastdynamics,anditrationalizestheadjustment ofexpectationsaftertherealizationofademandchange.16
Wenotethattheextenttowhichcapitalizationratesadjustfollow- ingapositivedemandshockarguablydependsontheinvestors’idiosyn- craticcharacteristicaffectingtheirabilitytobuildexpectations.Wein- vestigatethismatterinmoredetailinSection5.3.
3. Empiricalframework
Basedontheaboveframework,wederiveempiricalspecifications toestimatehousingsupplyelasticities𝜖𝑄,𝑃𝑛 and𝜖𝑄,𝑅𝑛 ,anddiscussthe corresponding identificationassumptions.Westartbyimposingaver- agesupplyelasticities𝜖𝑄,𝑃 and𝜖𝑄,𝑅commontoallneighborhoods.In thenextstep,weprovideaparametrizationallowingustoestimatehet- erogeneoushousingsupplyresponsivenessattheneighborhoodlevel.
3.1. Partiallingoutqualitydifferences
TheconceptualframeworkexposedinSection2implicitlyassumes thatthequalityofhousinggoodsishomogeneouswithinthesameneigh- borhoodandacrossrentalandsellingproperties.Therelativelysmall neighborhoodsinourprincipalempiricalanalysisjustifythisassump- tiontoacertainextent,aspropertiessharingsimilarhousingcharacter- isticstendtoclustertogether.
Yet,withinaneighborhood,differencesinthequalityandtypeof housinggoodsmayremain.Topreventpotentialqualitybias,inwhat follows,weremoveallpriceandrentvariationacrosslocationsthatorig- inatefromdifferencesinobservablehousingcharacteristics.Tothisend, weconstructlocal(log)priceandrentindicesfromhedonicregressions.
Specifically,ineachperiodweseparatelyestimate
ln𝜏𝑗𝑛𝑡=𝛾𝑛𝑡𝜏 +𝛽𝑡𝜏𝐀𝑗𝑛𝑡+𝜖𝜏𝑗𝑛𝑡, 𝜏=𝑅,𝑃 (7) where𝜏𝑗𝑛denoteseitherthepriceorrentofproperty𝑗inneighborhood 𝑛attime𝑡.Thevector𝐀𝑗𝑛𝑡includesacomprehensivesetofattributes suchashousingsurface,theaveragenumberofrooms,age,agesquared, andanindicatorforsingle-familyvs.multi-familyhouses.𝜖𝜏𝑗𝑛𝑡denotes theerrorterm.
15Giventhatrentexpectations entercapitalizationratesnegatively,anin- creaseinexpectedrentalgrowthdecreaseslocalcapitalizationrates.
16Fortheinstrumentsusedbelow,wefindthatpastdynamicsexplainonlyvery littleofthevariationsuchthatdemandchangescannotbeperfectlyanticipated basedonpastdynamics.
Weusetheestimatedneighborhood-timefixed-effects𝛾𝑛𝑡𝜏 asquality- adjustedlog-prices(log-rents)ln𝑃𝑛𝑡(ln𝑅𝑛𝑡).17Notethatthecoefficients 𝛽𝑡𝜏aretime-variant,suchthatthevaluationofhousingcharacteristics canflexiblychangefromperiodtoperiod.
3.2. Model-informedidentificationofaveragesupplyelasticities
Log-linearizing supplyEq. (2), expressing prices asa functionof quantities,andfirstdifferencing,in equilibrium18 weobtainthefol- lowingempiricalspecification
Δ ln𝑃𝑛=𝛼𝑃+ 1
𝜖𝑄,𝑃Δ ln𝑄𝑛+Δ ln𝑆𝑛𝑃, (8) where𝜖𝑄,𝑃1 istheaverageinversehousingsupplyelasticitycommonto allneighborhoodsandΔdenotesatimedifferencebetween2005and 2015.Theterm𝛼𝑃 denotestheaveragevalueofchangesinobserved supplyshifterscommontoallneighborhoodsand,inaslightabuseof notation,Δ ln𝑆𝑛𝑃representsthecorrespondingmean-centeredvariable.
Time-invariantcomponentsarepartialledoutfromΔ ln𝑆𝑛𝑃 byfirstdif- ferencing,suchthatonlydynamicsupplyshiftersenterEq.(8).
EstimatingEq.(8)byOLSlikelyleadstobiasedestimatesofthepa- rameterofinterestduetotheendogeneityofΔ ln𝑄𝑛 viathedemand side.This isapparentfromEq.(5)whenwriting changesinhousing demandas
Δ ln𝑄𝑛=Δ ln𝑊̄𝑛+Δ ln𝑁𝑛−Δ ln𝑅𝑛. (9) Therefore,estimatesof 1
𝜖𝑄,𝑃 inEq.(8)arepotentiallybiasedduetoi) acorrelationofthecomponentsof housingdemand changessuchas changesinaveragewagesandnumberofresidentswithchangesofun- observedsupplyshiftersinΔ ln𝑆𝑛𝑃 (omittedvariablebias),andii)the impactofhousingpricesviachangesinrentsonhousingdemand(re- versecausality).
AccordingtoEq.(2)(seeOnlineAppendixA),Δ ln𝑆𝑛𝑃 includes,in additiontoexogenoussupplyshifters,changesinlocalconstructioncost.
Changesinthepriceofconstructionmaterialsandthecosttomakeland availableforresidentialdevelopmentareunlikelytodifferacrossspace, mostlyduetothesmallcountrysize.Onthecontrary,shiftsinhousing demandmightaffectconstructioncostsviawages.
Topartiallyaddressdifferencesinconstructioncostsacrossneigh- borhoods,wecontrol,inafirststep,forseveralsupplyshiftersinEq.(8). Let𝐗𝑛 denotethevectorcontainingsuchcontrols.19Inparticular,𝐗𝑛 includesconstructioncostindicesi.e.,changes inlabor andmaterial costsdefinedforthemainnationalconstructionmarkets,theintensity ofhistoricaldevelopmentasmeasuredbydevelopmentdensityin1980 andterrainruggedness.AccordingtoHilberandRobert-Nicoud(2013), thehistoriclevelofhousingdevelopmentproxiesforthefactthathigh- amenityareasdevelopfirstandtendtoadoptmorestringentland-use regulationsovertime.Controllingforterrainruggednesstakesintoac- countthatplotsoflandfeaturinggeographiccharacteristicsfavorable todevelopmentsuchasflatandnon-rockysurfacesarelikelydeveloped beforethoseshowingadversegeographiccharacteristics.Therefore,we expectunfavorablegeographicfeaturestoincreaserentsandpricesover time,asdevelopersfacehigherconstructioncostsforprovidingaddi- tionalhousingunitsontheextensivemarginofexistingdevelopment.
Similarly,wecontrolfortheelevationoftheland.Wefurthercontrolfor thedistancetothenearestcentralbusinessdistrictstocapturepotential
17 Notethatallourresultsarerobusttoallowingforregion-specificvaluations ofhousingattributes,i.e.,theinclusionof𝛽𝑟𝑡 whereregions𝑟aredefinedas cantonsorcommutingzonescontainingasufficientnumberoflocations𝑛.
18 Inequilibrium𝑄𝑑𝑛=𝑄𝑠𝑛=𝑄𝑛holds,suchthatweomitsupplyanddemand superscriptsinwhatfollows.
19 Notethatsuchcontrolsmightalsopartialoutqualitydifferencesnotfully capturedbyhedonicindices.
timetrenddifferentialsinthelaborsupply(anddemand)acrossspace.
Thesevariablesmightalsocapturechangesintransportationcosts.20 Despite controlling for several supply shifters, omitted variables and reverse causality may still bias the estimation of Eq. (8). To solve this issue, we proposean instrumental variableapproach that aimstoexclusivelyshifthousingdemandwhileleavinghousingsup- plyunchanged.Specifically,werequireaninstrument,denotedΔ ln𝑍𝑛, whichisrelevantforhousingdemandchangeswhileremainingexoge- noustosupplychangesconditionalonthesetof controlsin 𝐗𝑛,i.e., 𝐸(Δ ln𝑍𝑛Δ ln𝑆𝑛𝑃|𝐗𝑛)=0.
AsapparentfromEq.(9),wecannotuseobservedchangesin the componentsofthehousingdemand,asinequilibriumtheyareaffected bytensionforcesbetweenhousingdemandandsupply.Followingthere- centworkbyBaum-SnowandHan,2021,weisolateexogenouschanges inlabordemandgiventhespatiallinkages(commutingcost)ofaneigh- borhoodwithemploymentcenters.Wedefine
Δ ln𝑍𝑛=∑
𝑖,𝑘
𝟏(𝑚𝑛𝑖𝑡0<1hour)
𝑚𝑛𝑖𝑡0 𝑓𝑛𝑡𝑘0Δ ln𝐹𝐶𝑘(𝑛), (10) wheretheindicatorfunction𝟏(𝑚𝑛𝑖𝑡0<1hour)equalsoneforneighbor- hoods𝑖thatarelocatedatmostat60minutestraveltimefromneigh- borhood𝑛,andzerootherwise.21Thequantity𝑓𝑛𝑡𝑘0representstheshare ofemploymentbelongingtosector𝑘inneighborhood𝑛attime𝑡0.The termΔ ln𝐹𝐶(𝑘𝑛)isthecorrespondingaggregategrowthrateofemploy- mentinindustry𝑘over[𝑡,𝑡0]inregion𝐶inwhichneighborhood𝑛is located.22
The intuition behind Eq. (10) is straightforward.We computea weightedemploymentgrowthofthepredeterminedsectoralcomposi- tionintheproximityofagivenneighborhoodbyimposingacommon industrygrowthequaltotheonethatoccurredintheregion𝐶inwhich theneighborhoodislocated.Takingintoaccounttheproximityinterms ofcommutingtimetootherneighborhoodsisofparticularrelevancein thecaseofsmallspatialunits,e.g.,neighborhoods,asmostindividuals likelydonotworkinthesameneighborhoodwheretheylive.Weuse aone-hourradiusforthemaximumcommutingdistanceinthebench- markanalysis,asthisleadstothestrongestpredictivepowerofthein- strument.
Two features in thewaywe computeΔ ln𝑍𝑛 support exogeneity claimswithrespecttounobservedsupplydynamics.First,inlinewith Bartik(1991),weexcludeneighborhood𝑛itselffromthecomputation ofΔ ln𝐹𝐶𝑘(𝑛).Second,weexcludeallsectorsrelatedtoconstructionand realestatefrom𝑓𝑛𝑡𝑘0andΔ ln𝐹𝐶𝑘(𝑛).Therefore,theinstrumentcaptures weightedchangesinlabordemandthatarenotrelatedtotheconstruc- tionsector.
TheinstrumentdefinedbyEq.(10)isashift-shareinstrumentinline withBartik(1991).23Recently,Adãoetal.,2019,Borusyaketal.,2021, andGoldsmith-Pinkhametal., 2020investigatedtheeconometricas- sumptionsnecessaryforthevalidityofshift-shareinstruments.These instrumentsarevalidifeitherinitialsharesareindependentandran- domlyassignedacrossobservationsorgrowthshocksoccurrandomly acrossregions.Inoursetting,wearguethatinitialsectoralsharesare exogenouswithrespecttochangesinunobservedsupplyshiftersandlo- calrentandpricedynamics.Thesectoraldistributionofemploymentis highlypersistentandlargelydeterminedbynaturalamenitiesandmar- ketaccess,suchthatthehistoricsectoraldistributionisunlikelytobe
20Ofcourse,thesecontrolvariablesalsoaffecthousingdemand.Inthiscase, includingtheminthesupplyfunctionisevenmoreimportantastheyreduce endogeneityissuesarisingfromchangesinhousingdemandaccordingtoEq.(9).
21Wesettraveltimewithinthesamemunicipalityto1minute.
22Inourbaselineresults,weuseCantonsasaggregateregion𝐶,whichrepre- sent26upper-tieradministrativeunits.Ourresultsarerobustusinglowertier regions,suchasdistricts(Bezirke).
23GrahamandMakridis,2021introducearelatedidentificationstrategycom- biningvariationintheinitialdistributionofhousingcharacteristicstogether withanationaldemandshiftforthesecharacteristics.
correlatedwithrecentchangesoflocalhousingsupplyshiftersΔ ln𝑆𝑃. FollowingGoldsmith-Pinkhametal.,2020,inSection6,weassestheva- lidityoftheidentifyingvariationbycomputingtheRotembergweights foreachsector.
WefurthersupportexogeneityclaimsregardingΔ ln𝑍𝑛bycompar- ingourresultswiththoseobtainedusinganalternativeshift-sharein- strumentbasedonthedistributionoflanguageshares.24Specifically,for eachneighborhoodwecomputethehistoricalshareoflanguageshares andinteractthemwiththecantonalgrowthoftherespectivelanguage group.25Assumingthattheidiosyncraticutilityshiftercanbedecom- posedas𝐵𝑘𝑛𝑖=𝐵𝑛𝐵𝑖𝑘inEq.(5),wecaninterpretthisinstrumentaspre- dictingdemandchangesviaashiftintheidiosyncraticpreferences𝐵𝑛 toliveinaneighborhood𝑛.
Theno-arbitrageconditionforinvestorspurchasingrealestateassets preventsthatthedemandshockscapturedbyourtwoinstrumentsare tenurespecific(seeSection2.2).Evenifoneofthedemandshockswere morerelevanttoaspecificpartofthepopulationthatismoreinclined torentorown,itwouldpropagateacrosstenuresduetochangesinat- tractivenessofinvestments.InthecontextofSwitzerland,thecloselink betweentherentalandowner-occupiedmarketsisreflectedinahousing marketthatisapproximatelyequallysplitbetweenrenters(56%)and owner-occupiers(44%).Thecloselinkbetweendifferenthousingmar- ketsegmentsissupportedbyrecentworkofMense(2020)forGermany, acountrydisplayingsimilarhousingmarketconditionsasSwitzerland.
Specifically,theauthorshowsthatthelocalconstructionofhigh-quality housingunitslowersrentsthroughouttherentdistributionshortlyafter thenewunitsarecompleted.
Theestimationofthesupplyelasticitywithrespecttorentchanges 𝜖𝑄,𝑅𝑛 followsthesamelogicasbefore.Bysubstitutingtherelationship betweenpricesandrentsarisingfromtheparametrizationofcapitaliza- tionratesintoEq.(8),andisolating𝑅𝑛,weobtain
Δ ln𝑅𝑛=𝛼𝑅+ 1
𝜖𝑄,𝑅Δ ln𝑄𝑛+Δ ln𝑆𝑛𝑅 (11) where𝜖𝑄,𝑅1 =𝜖𝑄,𝑃1𝜖𝑃 ,𝑅,andweassumethatthevector𝐗𝑛capturingthe observablesupplyshiftersdiscussedaboveiscontrolledfor.Thenew errortermisΔ ln𝑆𝑛𝑅= 1
𝜖𝑃 ,𝑅Δ ln𝑆𝑛𝑃.Becausethisnewerrortermequals theoneofEq.(8)timesaconstantstructuralparameter,theprevious discussionoftheidentificationassumptionsstillholds.26
Weconcludethissectionbynotingthataveragehousingsupplyprice andrentelasticityestimatescanunambiguouslyberecoveredfromin- versesupplyequations(8)and(11)simplybytakingtheinverseofthe estimatedvaluefor𝜖𝑄,𝑃1 and𝜖𝑄,𝑅1 ,respectively.27
Housingsupplyelasticityestimatesallowus,inturn,toestimateelas- ticityparameters𝜖𝑃 ,𝑅and𝜖𝑖,𝑅.28
Estimatingtheinversesupplyfunction,ratherthanthedirectone,is acommonapproachintheliterature(e.g.Saiz,2010)thatoffersseveral advantages.First,usingasdependentvariablesquality-adjustedprices andrentsresultingfromahedonicregressiondoesnotrequireadditional
24 Saiz(2007)usesasimilarapproachbycomputingashift-shareinstrument basedonthenumberofimmigrantsmovingintoU.S.cities.
25 SeeOnlineAppendixCforaformaldefinitionoftheinstrument.
26 Theexogenouscomponentofthecapitalizationrate𝑖0iscapturedbythe rent-specificconstant𝛼𝑅inEq.(11).Ifweassumetheparametrizationoflocal capitalizationratesexposedinSection2andallowforlocation-specificcapital- izationrates𝑖0𝑛,theerrorterminEq.(11)becomesΔ ln𝑆𝑛𝑅+𝜖𝑃 ,𝑅1 1+1𝛾𝑃
𝑛Δ ln𝑖0𝑛. Thus,theexogeneityoftheinstrumentmustalsoholdwithrespecttolocation- specificexogenouscapitalizationrates.InSection6,weshowthatincluding severalcontrolsproxyingforΔ ln𝑖0𝑛doesnotaffectourresults.
27 Thisrelationshipbetweeninversesupplyequationsandhousingsupplyelas- ticityparameterestimatesispurelymechanicalanddoesnothingeonadditional identificationassumptions.Weobtainidenticalpointestimateswhenestimating non-invertedsupplyequations.
28 Assumingthattheidentificationassumptionsunderlyingthe2SLSestima- tionof𝜖𝑄,𝑃 and𝜖𝑄,𝑅hold,accordingto(3)wecanestimate ̂𝜖𝑃 ,𝑅= ̂𝜖̂𝜖𝑄,𝑅𝑄,𝑃 and
̂𝜖𝑖,𝑅=1−̂𝜖𝑃 ,𝑅.
statisticaltreatment.Thisisnotthecaseifquality-adjustedprices(and rents)areusedasregressors,astheirstandarderrorsarenotvalidand needtobebootstrapped.Second,instrumentalvariablestendtobemore relevantforquantitychangesthanforpriceones,therebyimprovingthe precisionof theestimates.Third, estimatinginversesupplyfunctions allowsustoinstrumentthesamevariable(i.e.,Δ ln𝑄𝑛)torecoverthe responsivenessofhousingsupplytopriceandrentchanges.This,inturn, facilitatesthecomparisonofthecoefficientsacrossspecifications.
3.3. Localsupplyresponsivenessandtheroleofgeographicandregulatory constraints
Eqs.(8)and(11)assumethatinversesupplyelasticitieswithrespect topricesandrentsareconstantacrosslocations.However,ourconcep- tualframeworksuggeststhathousingsupplyelasticityvariesatthelocal level,bothattheintensiveandextensivemarginofresidentialdevelop- ment.Ontheintensivemargin,theempiricalliteraturepointsoutthat supplyelasticitymightvary considerablyaccordingtoregulatoryre- strictions– suchasheightrestriction,floortoarearatios,etc.– adopted bylocalgovernments.AccordingtoHilberandRobert-Nicoud(2013), attractiveplacesarehistoricallymoredevelopedand,asanoutcomeof thepoliticalgamebetweenlanddevelopersandownersofdeveloped land,moreregulated.Ontheextensivemargin,Saiz(2010)pointsout thatsuchconstraintsareempiricallyrelevantonlywhenthereisenough development tomakethembinding.29 Toimplementsuchconsidera- tionsempirically,wethusfollowSaiz(2010)andusethefollowingap- proximation
1
𝜖𝑛𝑄,𝜏 ≈𝛽𝑎𝑣𝑔,𝜏+𝛽ℎ𝑖𝑠𝑡,𝜏×𝑄1980𝑛 +𝛽𝑐𝑜𝑛𝑠𝑡𝑟,𝜏× Λ𝑛×𝑄1980𝑛 , 𝜏=𝑅,𝑃, (12) wheretheobservedhistoricstockin1980𝑄1980𝑛 proxiesforregulation constraintson theintensivemargin,accordingtoHilberandRobert- Nicoud(2013).ThevariableΛ𝑛isameasuresummarizingthemostim- portantgeographicandregulatoryconstraintsontheextensivemargin.
InsertingsuchapproximationinEqs.(8)and(11),wethenestimatethe followingequation
Δln(𝜏𝑛)=𝛼𝜏+𝛽𝑎𝑣𝑔,𝜏Δln(𝑄𝑛)+𝛽ℎ𝑖𝑠𝑡,𝜏Δln(𝑄𝑛)×𝑄1980𝑛
+𝛽𝑐𝑜𝑛𝑠𝑡𝑟,𝜏Δln(𝑄𝑛)× Λ𝑛×𝑄1980𝑛 +Δ ln𝑆𝑛𝜏, 𝜏=𝑅,𝑃, (13) wherewecontrolagainfor𝐗𝑛 (whichincludeshistoricdevelopment 𝑄1980𝑛 )aswellasforthemaineffectofextensivemarginconstraintsΛ𝑛. Tworemarksareworthnoting.First,Λ𝑛 isinteractedwiththehistoric stock level 𝑄1980𝑛 , thusallowing theimpact ofregulatory constraints tobecomemorebindinginmore-developedplaces.Havingestimated the parameters 𝛽𝑎𝑣𝑔,𝜏,𝛽ℎ𝑖𝑠𝑡,𝜏,and𝛽𝑐𝑜𝑛𝑠𝑡𝑟,𝜏, we thenrecover 𝜖𝑄,𝜏𝑛 using Eq.(12).Second,theidentificationassumptionsofthepriceandrent equationinEq.(13)arethesameasinEqs.(8)and(11)provided𝑄1980𝑛 andΛ𝑛areexogenouswithrespecttoΔ ln𝑆𝜏.
4. Dataanddescriptivestatistics
Theempiricalanalysisreliesonseveraldatasources.Furtherinfor- mationisavailableinOnlineAppendixB.
4.1. Datasources
HousingdataWeusegeo-referenceddataonadvertisedresidential propertiesprovidedbyMeta-Sys.Thedatasetcontainsapproximately 2.1millionpostingsofrentalpropertiesandabout0.8millionpostings ofsellingresidencesforthewholeofSwitzerlandfrom2004to2016.
Inadditiontoaskingrentsandprices,thedatasetincludescompre- hensive informationon housingcharacteristics.The FederalRegister
29Forexample,protectedforestlikelyhinderresidentialdevelopmentonlyif noothertypeofdevelopablelandisavailableintheneighborhood.
ofBuildingsandHabitationspublishedbytheSwissFederalStatistical Office(FSO)providesacensusoftheresidentialhousingstockofthe country.Changesinthehousingstockaremeasuredeveryfiveyears, providingthreetimeperiods2005,2010,and2015thatoverlapwith ouradvertisementdata.Upto2015,theregistercontainsapproximately 4.8millionhousingunitsforthewholeofSwitzerland,11.5percentof whichwerebuiltbetween2005and2015.The2000BuildingCensus (publishedbytheFSO)providesinformationonwhetheradwellingis aprimaryorsecondaryresidenceinthatyear.
Socio-demographicandeconomicdataWeusetheFederalPopu- lationCensusof2000(publishedbytheFSO)aswellasthePopulation andHouseholdsSurveyfrom2010to2015(publishedbytheFSO)to infergeo-referencedhomeownershipratesandtoobtaininformationon predeterminedlevelsandchangesinthelocalsocio-demographiccom- positioni.e.,nationalityandlanguageofresidentslivinginagivenarea.
The2000FederalPopulationCensusprovidesinformationonthetype of buildingowner,suchaslandlords andinstitutional investors.The FSOpublishesaconstructionindextrackingthecostevolutionofmate- rialandlaborintheconstructionsectorforsevenstatisticalareas.We obtaineddetailedinformationaboutthespatialdistributionofemploy- mentandfirmsbysectorfromtheStructuralBusinessStatistics(STA- TENT).WeusetheNOGA1sectorclassification,whichcomprises16 differentsectorsinthecaseofSwitzerland.30 Wecombinethis infor- mationwithdataaboutroadtraveltimeprovidedbywww.search.chto constructalocalmeasurelabormarketaccess.
RegulatoryandgeographicdataTheLandUseStatisticsofSwitzer- land(publishedbytheFSO)providessatellite-basedlandcoverdata, allowingustoidentifygeographicconstraints,suchaslakes,rocks,and glaciers,andareassubjecttoparticularregulations.Informationabout regulationsontheextensivemarginandprotectedareasinparticular isobtainedfromCantonalofficesof spatialplanningandtheFederal OfficefortheEnvironment(FOEN).
OtherdataWecomplementtheabovedatawithavarietyofdata onSwissadministrativeunitsandmetropolitanareas(publishedbythe FSO)andelevation(EuropeanEnvironmentAgency).Weidentifythe agglomerationsofthe15maincitiesinSwitzerland,asdefinedbythe FSO,andcomputethedistanceofeachneighborhoodtotheclosestcity center.
4.2. Datastructureanddescriptivestatistics
Westructurethedatabypartitioningthewholeterritoryofthecoun- tryintosmall squareneighborhoodsof2x2km.Weaggregate hous- ingstock,socio-demographicandeconomicdata,andgeographicand regulatoryconstraintswithinthese neighborhoods.31 Weassigneach neighborhoodtooneof2324municipalities,whichrepresentthelow- estgovernmentaltierinSwitzerlandandhaveaninfluenceonlanduse regulation.
From2005to2015,rentshaveincreasedbyapproximately14per- centwhileprices haveincreasedbyapproximately35percent atthe countrylevel.32Overthesameperiod,thehousingstockgrewbyap- proximately11percent.Despitethesegeneraltrends,stock,rent,and pricedynamicsareheterogeneousacrossspace,asillustratedinFig.1. Specifically,Fig.1shows stock,rent,andpriceindex growthin ma- jorcities(PanelA)andthecountryside(PanelB)from2005to2015, using2005asthebaseyear(=100).Overtheconsideredperiod,hous- ingstockgrewalmosttwiceasmuchinthecountrysideareasthanin cities.Thissuggeststhatincitiesalackofdevelopablelandinconjunc-
30 ThisclassificationcorrespondstothemajorSICindustrygroupsintheU.S.
31 Ourresultsarerobusttoalternativeneighborhoodsizes,asdiscussedin Section6.
32 FornewtenancyagreementsmarketrentsapplyinSwitzerland.Toprevent abusiveincreases,propertyownerscanadjustrentsforexistingtenancyagree- mentsonlyifsomeformalcriteriaaremet.However,severalexceptionsinthe regulationallowlandlordstoadjustrentstolocalmarketlevels.
Fig.1. Rent,price,andstockdynamicsNotes:Citiesinclude2x2kmneigh- borhoodslocatedinoneofthe15mainmunicipalitiesofthecorresponding biggesturbanareasaccordingto2015boundaries.Thetwopanelsshowindex growthfrom2005to2015oftheconsideredvariables,using2005asthebase year(=100).Stockismeasuredasthetotalnumberofdwellings,andrentsand pricesaremeasuredasadvertisedaveragerentsandpricespersquaremeter.
SeeOnlineAppendixB.2foradefinitionofmajorSwissagglomerations.
tionwithgeographicandregulatoryconstraintshindersfurtherdevel- opment.Giventhiscomparativelylowerresponsivenessofhousingde- velopmentincities,itisnotsurprisingthatrentsandpricesgrewmore intheseareasthaninthecountryside.Interestingly,from2007onward, rentsandpricedynamicshavestartedtodivergeconsiderably-with pricesrisingatafasterpace-whichimpliesthatcapitalizationrates havebeenreviseddownwardintheselocations.
InFig.2,weshow themostimportantgeographicandregulatory constraintsfor housingdevelopment inSwitzerland. Geographic fea- tures preventingany form of development areanimportantcompo- nent of the Swisslandscape. Similarlyto Saiz (2010) andLutz and Sand(2019),wedefineundevelopablelandaslandthatislocatedabove 2000 m andwhose landcover corresponds to unproductive vegeta- tion,vegetation-freeareas,orrocks,andglaciers.33 Waterbodiessig- nificantlyreduce theamountofdevelopablelandintheproximityof majoragglomerations,asvirtuallyallmajorCBDsareadjacenttoalake orriver.34
Inadditiontogeographicconstraints,therearesignificantregula- toryrestrictionsinplacethatpreventorhinderdevelopmentinspecific areas.We refertomeasuresthatprevent newconstructiononunde- velopedlandasregulationsontheextensivemargin.Regulationsonthe extensivemarginincludeforests35,UNESCOculturalornaturalheritage sites,parks,andotherhigh-valuenaturalamenityareas.Theserestric- tionsaccountforapproximately49.1percentoftheSwissterritory.Total restrictedareasobtainedbyoverlappinggeographicandregulatorycon- straintsattheextensivemarginamounttoapproximately67.3percent ofthecountry’ssurface.Theremaining32.7percentofthecountry’ssur- face(whiteareainFig.2)isavailablefordevelopmentunderdifferent
33Ourdataallowsustocompilethismeasureatevenfinerscalethanprevious literature.
34Thisismainlyduetothecompetitiveadvantageofareasintheproximityof waterbodiesduringtheIndustrialRevolutionandthesubsequenturbanization ofSwitzerland.
35Inresponsetothegrowingindustrializationofthecountry,in1876,Switzer- landpassedafederallawprohibitingfurtherdeforestation,defactofreezing forestareastothelevelobservedatthattime.Thelawhasremainedmainlyun- changedtothepresentday.Asaconsequence,forestareasinhighlypopulated regionshaveremainedpracticallyunchangedsince1876.
Fig.2.ConstraintstodevelopmentNotes:Wedefinegeographicconstraintsasundevelopableland,whichcorrespondstoaplotoflandlocatedabove2000m,or whoselanduseclassificationcorrespondstounproductivevegetation,vegetation-freeareas,orrocksandglaciers.Exceptforforests,redareassummarizeregulatory constraintsontheextensivemargin.Swissforestsareprotectedareassince1876.(Forinterpretationofthereferencestocolourinthisfigurelegend,thereaderis referredtothewebversionofthisarticle.)
Table1
Inversehousingsupplyelasticity– averageestimates.
(1) (2) (3) (4) (5) (6)
ΔLog P ΔLog R ΔLog P ΔLog R ΔLog P ΔLog R
ΔLog Q 2.367 ∗∗∗ 0.694 ∗∗ 2.004 ∗∗∗ 0.747 ∗∗∗ 2.084 ∗∗∗ 0.735 ∗∗∗
(0.442) (0.287) (0.342) (0.215) (0.316) (0.191)
Amplification 𝜖𝑃 ,𝑅 3.411 2.683 2.835
Instruments I I L L I & L I & L
Observations 2,498 2,498 2,498 2,498 2,498 2,498
Kleibergen-Paap F 15.61 15.61 71.39 71.39 42.76 42.76
Overidentification - - - - 0.40 0.87
Notes:Standarderrorsinparentheses∗𝑝<.10,∗∗𝑝<.05,∗∗∗𝑝<.01.Standarderrorsareclusteredatthemunicipalitylevel.Thesamplecoverstheperiod2005–2015.
TheAdãoetal.,2019standarderrorsforcolumns(1),(2),(3),and(4)are0.100,0.048,0.386,and0.198,respectively.Theunitsofobservationsareobtained bypartitioningSwitzerlandin2x2kmneighborhoods.Logrentsandlogpricesarequality-adjustedwithrespecttothelivingsurface,thenumberofrooms,age, agesquared,andbuildingtype.Allregressionscontrolforsupplyshifters,whichincludeelevation,elevationstandarddeviation,log-distancetothenearestCBD, log-housingstockin1980,totalrestrictedareas,andchange(2005–2015)inconstructioncosts.SeeOnlineAppendixEfordetailedestimationresultsandfirststages.
ChangesinhousingstockΔLogQ,areinstrumentedusingashift-shareinstrumentforindustriesIincolumns(1)and(2),ashift-shareinstrumentformainspoken languagesLincolumns(3)and(4),andboththeseinstrumentsI&Lincolumns(5)and(6).
regulatorymeasuresdeterminingtheintensityandtypeofresidential development.
5. Results
5.1. Supplyelasticityestimatesandamplificationmechanism
Table1summarizesaveragesupplyelasticityestimateswithrespect toprice(columns1,3,and5)andrent(columns2,4,and6)changes,re- spectively.Columns1and2reportestimatesbasedontheshift-sharein- strumentΔ ln𝑍𝑛derivedfromhistoricindustryshares(usedasabench- mark).Columns3and4reportthecorrespondingeffectsfortheshift-
shareinstrumentderivedfromhistoriclanguageshares.Columns5and 6showtheresultswhenusingthetwoinstrumentssimultaneously.Since ourmodelframeworkinSection2establisheslabormarketshocksasa source ofshiftsin housingdemand,we refertotheresults basedon industrysharesasourbaselineestimates.
Responsivenessestimatesbasedontheindustryinstrumentareequal to𝜖𝑄,𝑅= 1
0.694 =1.44forrentand𝜖𝑄,𝑃= 1
2.367 =0.42forpricechanges.
These results show that, on average,the housingsupplyin Switzer- landisrelativelyelastictorentchanges,butlesssotopricechanges.
Thecorrespondingamplificationeffectis𝜖𝑃 ,𝑅=𝜖𝑄,𝑅
𝜖𝑄,𝑃 =3.43.Thecoun- try’saverage responseoflocal capitalizationratestorent changesis thus𝜖𝑖,𝑅=1−𝜖𝑃 ,𝑅=−2.43,suggestingthatinvestorsreviselocalrent