Shovel Ready Projects and Commercial Construction Activity's Long and Variable Lags

Finance and Economics Discussion Series Federal Reserve Board, Washington, D.C. ISSN 1936-2854 (Print) ISSN 2767-3898 (Online) Shovel Ready Projects and Commercial Construction Activity’s Long and Variable Lags David Glancy, Robert J. Kurtzman, and Lara Loewenstein 2024-016 Please cite this paper as: David Glancy, Robert J. Kurtzman, and Lara Loewenstein (2025). “Shovel Ready Projects and Commercial Construction Activity’s Long and Variable Lags,” Finance and Economics Discussion Series 2024-016r1. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2024.016r1. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Shovel Ready Projects and Commercial Construction Activity’s Long and Variable Lags∗ David Glancy† Robert Kurtzman‡ Lara Loewenstein§ FederalReserveBoard FederalReserveBoard FederalReserveBankofCleveland May 16, 2025 Abstract We use microdata on the phases of commercial construction projects to document three facts regarding the sector’s time-to-plan lags: (1) plan times are long and highly variable, (2) nearly half ofprojectsinplanningareabandoned, and(3)propertypriceappreciationreducesthelikelihoodof abandonment. Wewritedownatractablemodelofendogenousplanningstartsandabandonmentthatcan matchthesefacts. Themodelalsohasthetestableimplicationthatsupplyismoreelasticwhenthereare more“shovelready”projectsreadyforconstruction. Weuselocalprojectionstovalidatethisprediction inthecross-sectionforUScities. Keywords: buildingsupplyelasticities,commercialrealestate,construction,time-to-plan JELClassification: R33,E22,E32,L74 ∗Thankstotheparticipantsatthe2024NorthAmericanMeetingoftheUrbanEconomicsAssociation,2024Boca Finance and Real Estate Conference, Federal Reserve Board’s R&S workshop, and the Federal Reserve Bank of Clevelandbrownbagseriesfortheirhelpfulcommentsandsuggestions. Anearlierversionofthisworkwascirculated underthetitle“OnCommercialConstructionActivity’sLongandVariableLags.” Theviewsexpressedinthispaperare solelythoseoftheauthorsanddonotnecessarilyreflecttheopinionsoftheFederalReserveBoard,theFederalReserve BankofCleveland,ortheFederalReserveSystem. †PrincipalEconomist,DivisionofMonetaryAffairs,FederalReserveBoard,david.p.glancy@frb.gov ‡Principal Economist and Group Manager, Division of Research & Statistics, Federal Reserve Board, robert.j.kurtzman@frb.gov §ResearchEconomistII,FederalReserveBankofCleveland,lara.loewenstein@clev.frb.org 1

1. INTRODUCTION Constructionactivity—orthelackthereof—iscriticaltomacroeconomicoutcomes. Inthelongrun, constructionaddstotheproductivecapacityoftheeconomyandfacilitatesanefficientallocationof workersandcapitalacrossspace(Herkenhoffetal.,2018;HsiehandMoretti,2019;Babalievsky et al., 2023). In the short run, construction is resource intensive and highly cyclical, and thus an important driver of aggregate demand fluctuations (Leamer, 2008). Though there is a wealth of workonresidentialconstruction,commercialconstructionisrelativelyunderstudied—eventhough it accounts for around 20 percent of private domestic investment (Brandsaas et al., 2024) and commercial real estate (CRE) is one of the economy’s largest asset classes (Ghent et al., 2019). Consequently,CREmarketdevelopmentscanhavewide-rangingimplications,includingforlocal governmentfinancesandthehealthofthebankingsector(Guptaetal.,2022). One unique characteristic of commercial construction is its long planning horizons (Edge, 2000; Millar et al., 2016), which can cause investment to respond slowly to economic shocks (Edge, 2007).1 In this paper, we investigate two implications of these long planning horizons for CRE supplyelasticities. Thefirstimplicationisthathaving“shovelready”projectsthatcanimmediately begin construction is necessary for adding supply in the short run (since it would take years for new projects to complete the planning process). The second implication is that economic conditions can change drastically over the planning horizon and prompt developers to abandon projects before starting construction. Since abandonment decisions affect construction activity fasterthandecisionsaboutcommencingnewprojects,thisabandonmentmargincanbeakeydriver ofnear-termconstructionactivity. Existingwork onsucheffectsis limited,duein parttodifficulties measuringconstructionactivity that does not occur. In this paper, we take advantage of unique panel microdata on the phases ofUS commercialconstructionprojects—including planning,construction, andabandonmentor completion—to examine how planning lags affect construction dynamics. The key contribution of our work is to establish that the availability of ongoing projects in planning is an important determinantofcommercialbuildingsupplyelasticities. Inthe firstpartof thepaper,we presentthreestylized factsonplanning activityandabandonment 1FigureS1intheSupplementaryMaterialsshowsthatcommercialstructureinvestmentisslowertorespondto businesscyclefluctuationsthanresidentialconstructionortotalprivateinvestment. 2

fromplanning orconstruction. First, commercialconstruction projectshave longplanning horizons. The average time spent in planning for projects that make it to construction is about 1.5 years, roughlysimilartoaverageconstructiontimes. Second,asignificantnumberofprojectsinplanning (nearlyhalf)areabandonedbeforebeginningconstruction. Almostallabandonmentshappenduring the planning stage: of the projects that make it to the construction phase, over 99 percent are completed. Third, whetherprojectsadvancefromplanning toconstructionisdependent onthestate oftheeconomy. We then present a tractable time-to-plan model of building production that matches these facts. In the model, developers optimally choose how much to invest in planning starts and whether to proceed with construction when planning is completed. Projects in planning are options to engage in construction that developers choose to exercise based on prevailing property values and building costs at the conclusion of planning. We show analytically that the model not only rationalizesthethreestylizedfacts,butthatitalsohasanimportantimplicationforthesupplyof commercialbuildings: thenear-termresponseofconstructionactivitytopriceappreciationdepends on the availability of projects in planning. Price appreciation can affect construction activity by bothstimulatingplanningstartsandcausingmoreprojectsinplanningtoadvancetoconstruction. Because planningstarts areslow to translate intoconstruction activity, this secondchannel is the maindriverofshort-termsupplyelasticities. Inthefinalpartofthepaper,wetestthismodelimplicationbyempiricallyexaminingcross-sectional differencesintheresponseofconstructionactivitytopriceappreciation. Specifically,weuselocal projections to trace out the response of construction starts to commercial price appreciation for metropolitan statistical areas (MSAs) with different initial stocks of projects in planning. We demonstrate that construction starts are increasing in price growth, and this response depends importantlyonthestockofprojectsinplanning,aspredictedbythemodel. Asfurthervalidation,we findsimilarresultsforemploymentgrowthforthesectorsmostengagedincommercialconstruction activity. Thispaperisrelatedtoworkthatanalyzescross-sectionaldeterminantsofbuildingsupplyelasticities. Theworkonthistopichasmostlyfocusedonresidentialhousing,showingthatregulatory(Mayer and Somerville, 2000; Glaeser et al., 2006; Kok et al., 2014) and geographic (Saiz, 2010; Baum- SnowandHan,2024)constraintstodevelopmentaffecthousingsupplyelasticities. Forcommercial construction, we show the importance of the availability of ongoing projects in planning as a 3

determinantofsupplyelasticities. More narrowly, our paper lays out new facts regarding the development process for commercial buildings. Aswiththeliteratureonbuildingsupplyelasticities,existingworkontheconstruction processmostlystudiesresidentialhousingconstruction(see,forexample,Glaeseretal.2005,2008). Regarding commercial construction, our work builds upon Millar et al. (2016), who use similar data (but from 1997 to 2010) to examine time-to-plan lags for completed projects and how they differ across cities and over time. Our work documents novel facts about the abandonment of commercial constructionprojects and focuseson the broaderimplications of theseplanning lags andabandonmentdynamics. Finally, we contribute to the literature analyzing time-to-build dynamics (Kydland and Prescott, 1982; Christiano et al., 1996), in particular in real estate development (Del Boca et al., 2008). Recentworkinthisliterature buildsonMajdandPindyck(1987)toanalyze howtheoptionvalue ofdelayinginvestment(Oh andYoon,2020) ordiscountrateshocks (FernandesandRigato,2023) affect build times. Additionally, contemporaneous work by Oh et al. (2024) shows that long developmenttimelinesinresidentialhousingmaketheshort-runhousingsupplyinelasticandaffect thebusinesscyclepropertiesofhousing. Wedemonstratethatsupplycanbemoderatelyelasticin theshortrunifprojectsarealreadyinplanning,aspricechangesaffectnewconstructionbyaltering abandonmentdecisionsfasterthantheyaffectconstructionthroughtheinitiationofnewprojects. Theremainderofthepaperproceedsasfollows. InSection2,wedescribethedataandestablishthe factsoncommercialconstructionthatwewillusetodisciplinethemodel. InSection3,wepresenta simplemodelofthecommercialconstructionprocessthatcanmatchthesefacts,andweanalytically derive how theshort-termelasticity ofbuildingsupplydepends onthe stockofprojects inplanning. InSection4,wetestthispredictionanddemonstratethattheresponsivenessofconstructionactivity tochangesinpricesisindeedafunctionoftheplanningstock. InSection5,weconclude. 4

2. FACTSONCOMMERCIALBUILDINGCONSTRUCTION In this section, we first describe the data. We then provide an overview of typical planning and constructiontimelines,includinghowoftenandunderwhatcircumstancesprojectsareabandoned. 2.1. ConstructionPhaseDataandOtherDataDetails We use data on individual construction projects from 2003 to 2024 collected by Dodge Data & Analytics (Dodge), which is also an input to the Census calculations of monthly construction spending.2 ThedataincludemonthlyinformationonactiveconstructionprojectsfromwhenDodgefirstrecords theprojectasbeinginplanninguntileithertheprojectisabandonedorconstructioniscompleted. Eachmonth,Dodgerecordsthecurrentphaseoftheproject,wherethephasesincludepre-planning, planning, final planning, bidding, underway, completed, deferred, and abandoned. Pre-planning indicatesaprojectthatispurelyattheconceptstage,andthatanarchitecthasnotyetbeenhired. Movingfrompre-planningtoplanningmeansthattheprojecthasgenerallyalreadyhiredanarchitect whohasstartedtodrawupplans. Finalplanningimpliesthattheprojectisgettingfinalapprovals and should go out for construction bids (the bidding phase) within 4 months. The first month of theunder-constructionphase(thestart)occursafteracontracthasbeensignedbetweenageneral contractor and the developer and the project should break ground within the next six months. A projectcanbedeferredindefinitelyfromanyphaseinthedata. Thedatadonotincludeinformation onthereasonforthe deferral, butpossiblereasonsinclude goingoverbudget,marketconditions worsening,orfinancingbeingpulled. Afinalstateforaprojectiseithercompletedorabandoned. Forouranalysis,wegroupprojectsinpre-planning,planning,finalplanning,andbiddingtogether. Wetreatdeferredprojectsasaseparatecategoryunlesswestateotherwise. Alongwith thephases, thedata includeinformation onthe property’stype, itssquare footage,the totalcostofconstructionfortheproject(oranestimateforthatspendforprojectsinplanningor bidding),anddetailedgeographicinformation.3 2InFiguresS3andS4oftheSupplementaryMaterialsweprovidecomparisonsofdataseriesconstructedusingthe Dodgedatatoothersources. 3AdditionaldetailsonthedataareavailableinSupplementaryMaterialsSectionS.2. 5

2.2. SummaryStatistics Figure1showshowplanningandconstructiontimes(toppanel)andabandonmentrates(bottom panel) differ across projects.4 These figures demonstrate that planning times arelong and outcomes arehighlyvariable—intermsofbothtimelinesandwhetherprojectseverreachconstruction. Regardingprojecttimelines,medianplantimesare roughlycomparabletoconstructiontimes,but theyaremuchmorevariable. ThiscanbeseeninFigure1a,whichdepictsthedistributionofplan and construction times across property types (left panel) and quintile of construction cost (right panel). Multifamilybuildingshavesomeofthehighestplanandconstructiontimes,andthegreatest variability in plan times. However, all property types exhibit wider distributions in plan times than in construction times. Both planning and construction tend to take longer for more expensive projects.5 One potential reason that planning times could have such a long right tail is due to developers’ options to defer or abandon projects. If economic conditions deteriorate such that the economic viabilityoftheprojectcomesintoquestion,developersmaywaittoseeifconditionsimprove,and then abandon a project if they do not. Evidence of this is shown in Figure 1b, which presents a binscatterofanindicatorforwhetheraprojectinplanningultimatelyreachesconstructionagainst thecommercialpropertypriceappreciationintheyearaftertheplanstart. Thefigureshowsthata highshare(ontheorderof50percent)ofprojectsinplanningneverstartconstruction,butthatstrong growth in property values in a local market increases the probability that a project successfully advancestoconstruction. More information summarizing the planning and construction process is in the Supplementary Materials. First,TableS3providesatransitionmatrixbetweendifferentprojectphases. Itshows thatnearlyallabandonmentanddeferraloccursduringtheplanningphasesofaproject,whereas nearlyallprojectsthatadvancetoconstructionareeventuallycompleted. Second,TableS4provides additionalregressionspredictingabandonment,includingvariablesrelatedtoprojectsizeandfactors that are potentially related to local supply elasticities (to be discussed more in Section 4). The starkest finding is that larger projects are more likely to be abandoned and are more sensitive to changesinlocalpropertyvalues. Finally,FigureS5presentsmapsofabandonmentratesandplan 4MoredetailedsummarystatisticscanbefoundinSupplementaryMaterialsTablesS1andS2. 5FigureS2intheappendixalsoprovidesplanandconstructiontimesseparatelyforprivateandpublicprojects. 6

rates—theratioofthenumberofprojectsinplanningtothenumberofcommercialpropertiesinan MSA—across CBSAs, and Table S5 provides statistics for average planning times, construction times, abandonment rates, and plan rates for the top 50 largest CBSAs in our sample. There is significantgeographicheterogeneity,afewexamplesofwhichareparticularlynotable. Plantimes andabandonmentratestendtobehigherinCaliforniaandnortheasterncities. Mostlargecitiesin Californiahaveaverageplanning times ofaroundoneyear,aboutdoublethatofthetypicallarge cityinTexas. Californiacitiesalsotendtohavehigherabandonmentrates;abandonmentratesare above 50 percent for every large city in California and below 50 percent for every large city in Texas. In short, these statistics show that commercial construction projects have long planning periods (about1.5 yearsona projectvalueweightedbasis), andthat changesineconomic conditionsaffect whetherconstructionoccurs. Thesefactssuggestthatnewconstructionrequires(1)thepresenceof projectsinplanningthatcanadvancetoconstructioninthenearterm(theplanningstock),and(2) economicconditions thatmotivatedevelopersto advanceto construction. To betterunderstand this first fact, Figure 2 decomposes changes in the aggregate number of projects in planning over time intochangesfromnewplanstarts(whichincreasetheplanningstock)andthosefromabandonment or construction starts (which decrease the planning stock). The figure shows that the number of projects in planning fell notably during the global financial crisis of 2007–09 (GFC) due to a contractioninplanningstarts(whichfellbyoverhalf)andariseinabandonments(whichexceeded construction starts in 2009 and 2010). The planning stock then started to rise in 2013, and it experiencedpositivegrowthuntil2024whenabandonmentsincreased. We provide further information on cross-sectional differences in planning behavior in the SupplementaryMaterials. FigureS6summarizeshowthedistributionofplanratesdiffersacrosscitiesand overtime. Theplanraterosefromunder0.5percentintheaftermathoftheGFCtoover1.5percent in2022. IntheaftermathoftheGFC,fewMSAshadaplanrateabove1,butthedistributionshifted significantlytotherightby2019afteralongbusinesscycleexpansion. 7

3. PLANNINGMODELWITHABANDONMENTS OurgoalistobuildamodelconsistentwiththefactsoutlinedinSection2. Inthemodel,timeisdiscrete,labeledast =0,1,2,.... Thereisarepresentativebuildingproducer who optimallydecides how muchto invest inplanning andconstruction starts. The buildertakes as givenaparticularsequenceofinterestrates(r ),rentalratesofbuildings(rb),andcostsofplanning t t starts(ι ). IntheSupplementaryMaterials,weembedthemodelintoageneralequilibriumbusiness t cyclemodelwherethesevariablesareendogenouslydetermined. Sincethesemicrofoundationsare notneededforthemainfindings,wefocusontheproblemofdevelopershere. Theproductionofbuildingsissubjecttotwofrictions. First,thereisastochastictimelagbefore building construction occurs. Specifically, firms can invest in planning projects, but only a share λ of these projects can advance to construction in a given period. When the planning horizon is completed, firms draw a cost κ ∼F and can choose whether or not to pay κ to produce a unit of building. Firmschoosethemaximumamounttheyarewillingtopayforaprojectκ∗,resultingin t theconstructionofλP F(κ∗) buildings,whereP istheplanningstockchosenintheprevious t−1 t t−1 period. Projectswithcostsabovethisthresholdareabandoned. Second, firms face adjustment costs in starting projects. The cost of initiating a planning start at p timet,denoted ι ,isincreasingin theamountofplanninginvestment,denotedI . Weassumethese t t adjustment costs are external to the firm (reflecting factors such as the supply of permits rather than internal capacity constraints) and are thus reflected in the cost of planning starts, which are taken as exogenous to the developer. This assumption simplifies the first order conditions, but thequantitativeestimatesaresimilarwithinternaladjustmentcosts(seeSupplementaryMaterials SectionS.3). Consequently,theproblemofthedeveloperisasfollows:   (cid:32) s 1 (cid:33)  κ (cid:90) t ∗ +s   max E ∑ ∏ rb B −ι I p −λP κdF(κ)), {I t p +s ,κ t ∗ +s }∞ s=0 t s i=0 1+r t+i   R (cid:124) t e + nt s a (cid:123) lI (cid:122) t n + c s o − m (cid:125) 1 e t+s t+s t+s−1 0    (cid:124) (cid:123)(cid:122) (cid:125) Planning&ConstructionExpenditure 8

subjecttothelawsofmotionfortheplanningandbuildingstock(P andB ): t t p P =(1−δ −λ)P +I t+s p t+s−1 t+s B =(1−δ )B +λP F(κ ∗ ), (1) t+s b t+s−1 t+s−1 t+s (cid:124) (cid:123)(cid:122) (cid:125) Ib t+s whereδ andδ arethe depreciationratesfor theplanning andbuilding stock,respectively. This p b problemhasthesolution: κ ∗ =qb t t 1 (cid:16) (cid:17) qb =E rb +(1−δ )qb t t 1+r t+1 b t+1 t+1 p p ι (I )=q (2) t t t   κ∗ t+1 1 (cid:90) q p =E λ (qb −κ)dF(κ)+q p (1−δ −λ), t t 1+r  t+1 t+1 p  t+1 0 whereqp andqb aretheLagrangemultipliersontheplanningandbuildingaccumulationconstraints, reflectingthevaluesofaunitoftheplanningandbuildingstock.6 The first line shows that developers proceed with construction when the cost of construction is lessthanqb,whichisdefinedinthesecondlinetobethepresentdiscountedvalueoffuturerental t income (i.e., the value of a unit of B ). The third line says that developers will invest in planning t p p startsuntilthevalueofaunitoftheplanningstock,q ,isequaltothecostofastart. q isdefinedin t t thelastrowasthepresentdiscountedvalueofthesurplus(buildingvaluenetofconstructioncosts) expected to be received when the planning stage ends. Takentogether, the last two lines imply that there will be more plan starts when property values are expected to be high relative to construction costs. The model rationalizes the main facts listed in Section 2. First, there are planning lags for commercialconstruction: theaveragetime-to-planis 1. Second,thereisstate-dependentabandonment λ outofplanning: afractionλ(1−F(qb)) projectsinplanningareabandonedperperiod,meaning t commercialpriceappreciation(higherqb)reducesabandonment. t 6SeetheSupplementaryMaterialsSectionS.3forfurtherdetailsonthederivation. 9

The model also produces an additional implication: the short-run elasticity of supply depends on the planning stock. From (1) and (2), we have that growth in the building stock next period is Bˆ ≡ Bt−B t−1 = −δ +λ P t−1F(qb). Differentiating with respect to building values shows that t B b B t t−1 t−1 ∂Bˆ t =λ P t−1 f(qb). Inwords,theresponseofbuildingsupplytopriceappreciationintheshortrun q t b B t−1 t P dependsontheratiooftheplanningstocktothebuildingstock( t−1)andtheshareofprojectsin B t−1 planningthatareonthemarginofadvancingtoconstruction(λ f(qb)). t Overalongerhorizon—ascanbeseeninthelasttwolinesof(2)—priceappreciationalsoaffects buildingsupplybypromptingtheinitiationofnewplanningstarts. InSupplementary Materials SectionS.4, weembedthisbuilding sector modelintoa DSGEmodel andcalibrateittomatchthefindingsofSection2. Consistentwiththetestableimplicationdescribed in this section, we show that construction responds later to TFP-driven property price shocks in markets with a lower initial planning stock. An alternative model with exogenous abandonment overstatesgestationlagsbyshuttingdownamechanismbywhichconstructioncanrespondquickly todemandshocks. Thesimplestructureofthismodelisusefulasitallowsustoderiveeasily-interpretedexpressions thatlinkmodeloutcomestopatternsobservedinthedata. Furthermore,thestructurefitseasilyinto astandardDSGEmodel,allowingustoquantitativelyanalyzedynamicsinageneralequilibrium setting. However,thissimplicityrequiresustoabstractfromotherpotentiallyimportantmechanisms. For example, uncertainty over future property values could also affect whether projects advance toconstructionbyraisingthevalueoftheoptiontodelayconstruction(MajdandPindyck,1987). Giventhatuncertaintyisgenerallyhigherduringperiodsofstress(Juradoetal.,2015;Adrianetal., 2019),sucheffectsmightexacerbatetheresponseofconstructiontodownwardpricemovements; weindeedfindsomesuggestiveevidenceofthisinthenextsection. 4. MODELVALIDATIONUSINGLOCALPROJECTIONS Thissectionteststhepropositionthattheresponseofconstructioninvestmenttopriceappreciation depends on the initial stock of projects in planning. We first outline the methodology and then presenttheresults. 10

4.1. Methodology The goalis totest whether theresponse ofconstruction activity to commercial price appreciation dependsontheavailabilityofprojectsinplanning,aswasimpliedbythemodelinSection3. The planning stock is measured by the number of projects in planning for a given property type p, in MSAi,asofquartert,normalizedbythenumberofsuchbuildingsinthemarketatthetime(Plan Rate ). Commercial property price appreciation is measured by the year-over-year change in i,p,t CoStar’scommercialpropertypriceindexforagivenMSA-propertytype-quarter.7 Becauseconstructionoccursslowlyovertime,welookatthecumulativeeffectsonconstruction startsusinglocalprojections. Themaindependentvariableiscumulativeconstructionstartssincet (asafractionoftheinitialbuildingstock),thoughwealsoanalyzeeffectsonMSA-levelcommercial constructionemploymentgrowth. Specifically,weestimatetheequation: ConstructionStarts i,p,t,t+h =β h∆ln(Comrcl. PriceIndex ) i,p,t BuildingStock i,p,t +δ h∆ln(Comrcl. PriceIndex )×Plan. Rate (3) i,p,t i,p,t +γ hX +η h +γ +ε h , i,p,t i,t p i,p,t whereh={2,4,...30}indexesthehorizononwhichwemeasuretheconstructionresponse,{βh} tracesouttheestimatedcumulativeconstructionresponsetopriceappreciationwhennoprojectsare inplanningattimet,and{δh}tracesouttheextenttowhichhigherinitialplanningstockmarkets are more responsive. ηh and γ are MSA-quarter and property type fixed effects. The vector of i,t p controls X includes Plan. Rate and measures of recent construction and planning activity.8 i,p,t i,p,t Regressions are weighted by the number of properties in a market (based on CoStar data), and standarderrorsaretwo-wayclusteredbyMSAandquarter. Oursamplerunsfrom2005to2016;we startin2005toensurethatanystartsatthebeginningofoursamplearedeterminedafteraplanning phasethatwecanobserve,andweendin2016sothatthethree-yearresponse(thetimehorizonwe 7Weexcludehotelsfromthisanalysisduetothelackofconsistently-reportedpricedata. 8Wecontrolforconstructionstartsoverthepastyear(asashareofpropertystockasofayearbefore)andfour lagsoftheplanningstockandtheunder-constructionstock(thenumberofprojectsinaninMSAinplanningorunder construction,respectively,asashareofthepropertystockatthetime). 11

focus most on) excludesCOVID-relateddisruptions. That said, in unreportedresults, we find that theestimatesarerobusttoallowingthesampletorunlater. Theassumptionunderlyingthismethodologyisthatcross-sectionaldifferencesincommercialprice appreciationreflectchangesindemandratherthansupply. Estimateswillbedownwardbiasedto the extent that supply-related factors drive price appreciation (e.g., impediments to construction would increase commercial prices but reduce construction activity). We take two approaches to mitigate this downward bias. First, we deploy MSA-quarter fixed effects to control for various factors such as labor availability, local regulatory stringency, and costs of construction materials that may affect local supply. Since inputs into construction are fairly similar across property types, within-MSAvariationinpricesismorelikelytoreflectdemandfactorsthansupplyfactors. Second, we instrument for ∆ln(Comrcl. PriceIndex )×Plan. Rate with the interaction of national i,p,t i,p,t priceappreciationwithPlan. Rate . ThisstrategyissimilarinspirittoMianandSufi(2011)and i,p,t Chaney et al. (2012) in that we exploit heterogeneous effects of national shocks across markets with different supply elasticities. While these estimates could still have a downward bias to the extent that national price trends are driven by supply shocks, this is likely less of a concern than localimpedimentstoconstructionincreasingbothabandonmentandpropertyprices. Indeed,Table S6intheSupplementaryMaterialsshowsthattherelationshipbetweenconstructionactivityand changesinnationalconstructioncostsismoreconsistentwithdemand-factorsdrivingconstruction— constructionactivityishigher followingcostgrowth—thoughthereissomeevidencesuggesting thatsupplyshocksmattermoreduringthepandemic(afterthesampleends). A related threat to identification comes from the endogeneity of Plan. Rate . A high plan rate i,p,t couldbe correlatedwithother factorsthataffect theelasticityof supplyin amarket. The direction ofsuchbiasisambiguous: impedimentstoconstructionmayincreasetheplanrate(e.g.,ifhurdles cause it to take longer to get through the planning stage), or decrease it (e.g., if developers do not initiate a plan due to a lack of available land). To address this concern, we add to X the i,p,t interaction of commercial price appreciation with other variables such as land availability and zoning regulations that may affect supply elasticities (Saiz, 2010; Baum-Snow and Han, 2024; Bartiketal.,2023). 12

4.2. LocalProjectionResults Howdoestheavailabilityofprojectsinplanningaffectconstructionactivity? Ifprojectsinplanning mechanically advance to construction and completion, projects in planning will measure future additions to supply. To the extent that these projects are options, projects in planning affect the elasticity ofbuildingsupply, astheywilladd tothebuildingstockifcommercialpropertyprices warrantconstructionproceeding. LocalPlanningStockandthe3-yearConstructionResponse Table1presentsestimatesofthe cumulative response of construction starts to commercial property price appreciation at the three yearhorizon,and Figure3plotsthe localprojectionestimatesofthe responseovertime. Westart by discussing the findings in the table, as they are useful for demonstrating the robustness of the resultstodifferentestimationstrategies. Thefirstthreecolumnsestimatetheeffectsofplanningactivityandpriceappreciationonconstruction starts omitting the interaction between the main explanatory variables. Column 1 presents OLSestimates usingonlyMSA andproperty type fixed effects, Column 2presentsIV estimates fromthesamespecificationinstrumentingforpriceappreciationwithnationalappreciationforthe givenpropertytypeandColumn3presentsOLSestimateswhenaddingMSA-quarterfixedeffects. TheestimatesinColumn1indicatethatonepercentagepointhigherpriceappreciationincreases constructionstartsbyabout1.2bps(asashareofthebuildingstock). Asprojectsunderconstruction areessentiallyalwayscompleted,thisshouldtranslatetoabouta1.2bpincreaseinthebuildingstock onceconstruction iscompleted. That is,the coefficientestimatecan bethought ofasa measureof theshort-termelasticityofcommercialbuildingsupply. The estimates are nearly identical in the IV specification (Column 2), but fall by half when the MSA-quarterfixed effectsareincluded (Column3). One potentialexplanationfor thisresult isthat thereissubstantialnoiseinmeasuringlocalpriceappreciationduetothelownumberofcommercial propertytransactionsincertainquarters,andthismeasurementerroraccountsforagreaterportion ofthevariationwhenmoregranularfixedeffectsareemployed. Columns 4–6 present the primary findings, where specifications include the interaction of price appreciation with Plan. Rate . The IV specification in column 6 also includes the interaction i,p,t 13

of Plan. Rate with national price appreciation as an instrument. The main object of interest i,p,t is the coefficient on the interaction term, which estimates how much the availability of shovel ready projects increases short run supply elasticities. The estimates range from 0.9 in the OLS specificationwithMSA-quarterfixedeffectsto1.3intheIVspecification. Theseresultssuggest thatincreasingtheplanratebyonestandarddeviation(1.6percentagepoints)increasesthesupply elasticitybybetween1.4and2.1bpsacrossthethreeestimates. Giventhattheaverageresponseto priceappreciation wasonlyabout 1.2bps,this constitutesasignificant proportionalchangein local supplyelasticities. Thelast twocolumns showthat theIV estimatesare robust tothe inclusionof moregranular fixed effects. Column (7) adds MSA-quarter fixed effects to control for changes in local conditions andColumn(8)addspropertytypespecificslopes(theinteractionofpropertytypedummieswith Plan. Rate andlocalpriceappreciation)toaccountfordifferencesacrosspropertytypesinthe i,p,t responsetopriceappreciationortherateatwhichprojectsinplanningadvancetoconstruction. The estimatedinteractioneffectsaregenerallyinthesamerangeasintheotherspecifications. Overall, theseresultsdemonstratethatsupplyismoreelasticintheshortrunwhenthereareprojectsalready inplanningthatareavailabletogotoconstruction. Controls for Other Factors Relating to Supply Elasticities To increase confidence that it is the projects themselves that matter (as opposed to market conditions), we add controls for the interaction of various other market characteristics with price growth in Table 2. To account for growthpotentiallybeingmoreachievableinsmallermarkets(giventhesmallerbase),wecontrol of the logarithm of the initial property stock. To account for land availability, we control for the commercial-property-stock-weighted housing supply elasticity from Baum-Snow and Han (2024).9 To account for the regulatory environment, we add the first two principal components from the LLM-generated data on housing regulations from Bartik et al. (2023).10 The last two columns addcontrols forpriceappreciationinteracted with MSAdummies,thus accountingforunobserved MSA-levelfactorsaffectingsupplyelasticities. Theestimatesfallmodestlywhenweincludethe 9Baum-SnowandHan(2024)estimatehousingsupplyelasticitiesatthetractlevelbasedontheshareofatractthat isdeveloped,thedistancetotheCBD,andtopographicalcharacteristicsaroundthetract. WeaggregatetotheMSA level,weightingtractsbytheircommercialbuildingcountfromCoreLogic. 10Theauthorsdemonstratethatthefirstcomponentisassociatedwithfactorssuchasaffordabilitymandates,which aremeanttoextractsurplusinhighdemandareas,whilethesecondcomponentisassociatedwithexclusionaryzoning, suchasrestrictionsonmultifamilydevelopment. 14

MSA controls and rise modestly with MSA-specific slopes, but do not differ meaningfully from themainestimatesinTable1. Inshort,theincreasedelasticityinmarketswithmoreshovel-ready projectsappearstoreflecttheavailabilityofprojectsthatarereadyforconstructionratherthanother factorsthoughttomatterforlocalsupply. Effects Over Time How does the supply response change over time? If new projects quickly enterplanninginresponsetopriceappreciation,areaswithalowplanratemightcatchupovertime as these new projects start to reach construction. If this process is more frictional, the gap could widenovertimeasmoreexistingprojectsinhighplanareashavetimetocompleteplanningand enterconstruction. ThelocalprojectionestimatesinFigure3aremoreconsistentwiththelatterstory. Thetoppanel plots OLS estimates along the lines of Column (6) of Table 1, while the bottom panel plots IV estimatesalongthelinesofColumn(5). Thegreenlineshowsthepredictedresponseofcommercial constructionforamarketwithanaverageplanrate,whilethetheredonesshowtheresponsewhen theplanrateisonestandarddeviationabovethemean. InboththeOLSandIVestimates,thegapin theresponse betweenhigh-and average-planmarketsincreases aboutlinearlyover time. In theIV estimates,thesupplyelasticitygrowstoabout3bpsafter7.5yearsinamarketwithanaverageplan rate and almost 9bp for a market with a high plan rate. As we saw in Table 1, the elasticities are lowerwiththeOLSestimates,butthedifferencebetweenhigh-andaverage-planmarketsissimilar. Given that we have only a short time horizon in the data, it is not feasible to estimate longer-run responses to price appreciation. While it is possible that low plan markets are able to add more stockeventually,these results suggestthat theavailabilityof projectsalready inplanningis critical fortheabilityofamarkettoaddsupplyintheshortrun. Heterogeneityby LocalMarketConditions Theresultssofartreattheeffectsofpricegrowth anddeclinesassymmetric. Whilethismaybereasonableforsmallfluctuationsaffectingwhether marginal projects come to fruition, larger shocks may have asymmetric effects. For example, negativeeconomicshocksmighthavelargereffectsiftheyareassociatedwitheithertightlending conditionsorhighuncertainty(raisingthevalueoftheoptiontodelay). Toinvestigatesucheffects, FigureS7intheSupplementalMaterialsestimatesEquation(3),butreplacesPriceGrowth with i,p,t Price Growth + ≡ max{PriceGrowth ,0} and Price Growth− ≡ min{PriceGrowth ,0}, i,p,t i,p,t i,p,t i,p,t 15

thus allowing for a differential response to price appreciation when prices are rising and falling. Whilethedifferenceineffectswhenpricesarerisingorfallingarenotstatisticallydifferent,wedo seethattheplanningstockmattersmoreforsupplyelasticitieswhenpricesarefalling,particularlyat shorterhorizons. Inotherwords,havingahighplanningrateseemstoamplifythenegativeeffectof pricedeclinesmorethanitamplifiesthepositiveeffectofpriceincreases. However,theconfidence intervals are wide, so the extent to which supply elasticities vary by local market conditions is uncertain. EffectsOnConstructionEmployment FigureS8intheSupplementaryMaterialssimilarlyplots the response of MSA-level construction employment to commercial property price appreciation (aggregatingover propertytypes sincetheemployment datais attheMSA level).11 Theleft panel presents OLSestimates of theeffect ofprice appreciation ina specification withMSA and quarter fixedeffects,whiletherightpresentsestimatesinstrumentingwithnationalpriceappreciation(and omittingthequarterfixedeffects). Eachfigureshowsthatcommercialconstructionemployment responds more to appreciation in markets with a high plan rate. After 2 years, construction employmentincreasesby3–4bps(relativetototalemployment)inhigh-plancities,and2–3bpson average.12 Thedifferenceinemploymentresponsebyinitialplanratethendeclinestonearzeroin theOLSspecificationandreversesintheIVspecification. 5. CONCLUSION The planning phase for commercial construction is long. Using microdata on the phases of development for CRE construction projects, we show that abandonment during the planning phase iscommonandissensitivetochangesinlocaleconomicconditions. Weconstructatime-to-plan model consistent with these observations and predict that the response of construction activity to price appreciation depends on the stock of projects in planning. Using a local projections methodology,wefindthatthisrelationshipindeedholdsinthecross-sectionofU.S.cities. 11EmploymentdatacomefromtheQuarterlyCensusofEmploymentandWages. Constructionemploymentisthe sumofemploymentinNAICScodes2362(commercialconstruction),236116(multifamilyconstruction),alongwith all6-digitcodespertainingtocommercialconstructioncontractors(238112,238122,...,238912,238992). 12Commercialconstructionemploymentaveragesabout2percentoftotalemployment,sogrowthincommercial constructionemploymentmovesmorethan1-for-1withpriceappreciation. 16

Thesefindinghaveseveralpolicyimplications. First,thefindingsspeaktothepotentialfortargeting infrastructureinvestmentforeconomicstimulus. Stimulusshouldbe“timely”sothatexpenditure occurswhenthereisstillslackintheeconomy(Summers,2008),meaningthattime-to-plandelays may reduce infrastructure investment’s efficacy as stimulus (Leeper et al., 2010; Ramey, 2021). Accounting forabandonment resultsin a morenuanced prediction: policies targeting commercial construction can work quickly, but only when there are “shovel-ready” projects available. This highlightstheimportanceofearlyinterventionsincetheabandonmentofprojectsoverthecourseof adownturncouldreducethestockofavailableprojects,asoccurredaftertheGFC. Second, the findingshave implicationsfor policies witha longer-term aimof encouraging development, such as many place-based policies. Since planning and construction can take a few years, policiesneedtoestablishincentivesfordevelopmentyearsinthefuturetostimulatenewprojects. ThismeansthatprogramssuchasOpportunityZones—wherethebenefitstoparticipationphase outquicklyovertime(CorinthandFeldman,2024)—wouldpredominantlyaffectalready-planned projects. Stimulatingnewinvestmentintheareaswherelittleinvestmentwaspreviouslyplanned mayrequirelonger-lastingincentives. 17

References ADRIAN, T., N. BOYARCHENKO, AND D. GIANNONE (2019): “VulnerableGrowth,”American EconomicReview,109,1263–1289. BABALIEVSKY, F., K. F. HERKENHOFF, L. E. OHANIAN, AND E. C. PRESCOTT (2023): “The ImpactofCommercialRealEstateRegulationsonU.S.Output,”NBERWorkingPaperNo.31895. BARTIK, A., A. GUPTA, AND D. MILO (2023): “The Costs of Housing Regulation: Evidence FromGenerativeRegulatoryMeasurement,”WorkingPaper. BAUM-SNOW, N. AND L. HAN (2024): “The Microgeography of Housing Supply,” Journal of PoliticalEconomy,132,1897–1946. BRANDSAAS, E. E., R. J. KURTZMAN, AND J. NICHOLS (2024): “FromPlanstoStarts: Examining Recent Trends in Manufacturing Plant Construction,” FEDS Notes. Washington: Board of GovernorsoftheFederalReserveSystem. CATTANEO, M. D., R. K. CRUMP, M. H. FARRELL, AND Y. FENG (2024): “On Binscatter,” AmericanEconomicReview,114,1488–1514. CHANEY, T., D. SRAER, AND D. THESMAR (2012): “TheCollateralChannel: HowRealEstate ShocksaffectCorporateInvestment,”AmericanEconomicReview,102,2381–2409. CHETTY, R. (2006): “ANewMethodofEstimatingRiskAversion,”AmericanEconomicReview, 96,1821–1834. CHRISTIANO, L. J., R. M. TODD, ET AL. (1996): “Time to Plan and Aggregate Fluctuations,” FederalReserveBankofMinneapolisQuarterlyReview,20,14–27. CORINTH, K. AND N. FELDMAN (2024): “Are Opportunity Zones an Effective Place-Based Policy?” JournalofEconomicPerspectives,38,113–36. D’AMICO, L., E. L. GLAESER, J. GYOURKO, W. R. KERR, AND G. A. PONZETTO(2024): “Why Has ConstructionProductivity Stagnated? TheRole of Land-UseRegulation,” NBER Working PaperNo.33188. DEL BOCA, A., M. GALEOTTI, C. P. HIMMELBERG, AND P. ROTA (2008): “Investment and 18

TimetoPlanandBuild: AComparisonofStructuresvs.EquipmentinaPanelofItalianFirms,” JournaloftheEuropeanEconomicAssociation,6,864–889. EDGE, R. M. (2000): “The Effect of Monetary Policy on Residential and Structures Investment underDifferentialProjectPlanningandCompletionTimes,”WorkingPaper. ———(2007): “Time-to-Build,Time-to-Plan,Habit-Persistence,andtheLiquidityEffect,”Journal ofMonetaryEconomics,54,1644–1669. FERNANDES, A. AND R. RIGATO (2023): “KWasn’tBuiltinaDay: InvestmentwithEndogenous TimetoBuild,”WorkingPaper. GERTLER, M. AND P. KARADI (2013): “QE1vs.2vs.3...: AFrameworkforAnalyzingLarge- ScaleAssetPurchasesasaMonetaryPolicyTool,”InternationalJournalofCentralBanking,9, 5–53. GHENT, A. C., W. N. TOROUS, AND R. I. VALKANOV (2019): “CommercialRealEstateasan AssetClass,”AnnualReviewofFinancialEconomics,11,153–171. GLAESER, E. L., J. GYOURKO, AND A. SAIZ (2008): “HousingSupplyandHousingBubbles,” JournalofUrbanEconomics,64,198–217. GLAESER, E. L., J. GYOURKO, AND R. E. SAKS (2005): “WhyHaveHousingPricesGoneUp?” AmericanEconomicReview,95,329–333. ———(2006): “UrbangrowthandHousingSupply,”JournalofEconomicGeography,6,71–89. GUPTA, A., V. MITTAL, AND S. VAN NIEUWERBURGH (2022): “WorkfromHomeandtheOffice RealEstateApocalypse,”NBERWorkingPaperNo.30526. HERKENHOFF, K., L. OHANIAN, AND E. PRESCOTT (2018): “Tarnishing the Golden and Empire States: Land-Use Restrictions and the U.S. Economic Slowdown,” Journal of Monetary Economics,93,89–109. HSIEH, C.-T. AND E. MORETTI (2019): “HousingConstraints andSpatial Misallocation,” AmericanEconomicJournal: Macroeconomics,11,1–39. JURADO, K., S. C. LUDVIGSON, AND S. NG (2015): “MeasuringUncertainty,”AmericanEconomicReview,105,1177–1216. 19

KOK, N., P. MONKKONEN, AND J. M. QUIGLEY (2014): “Land UseRegulationsand theValue of LandandHousing: AnIntra-metropolitanAnalysis,”JournalofUrbanEconomics,81,136–148. KYDLAND, F. E. AND E. C. PRESCOTT (1982): “Time to Build and Aggregate Fluctuations,” Econometrica,50,1345–1370. LEAMER, E. E. (2008): “Housing IS the business cycle,” in Housing, Housing Finance and MonetaryPolicy,AsymposiumsponsoredbytheFederalReserveofKansasCity,149–233. LEEPER, E. M., T. B. WALKER, AND S.-C. S. YANG (2010): “GovernmentInvestmentandFiscal Stimulus,”JournalofMonetaryEconomics,57,1000–1012. MAJD, S. AND R. PINDYCK (1987): “Time to Build, Option Value, and Investment Decisions,” JournalofFinancialEconomics,18,7–27. MAYER, C. J. AND C. T. SOMERVILLE (2000): “Land Use Regulation and New Construction,” RegionalScienceandUrbanEconomics,30,639–662. MIAN, A. AND A. SUFI (2011): “House Prices, Home Equity–Based Borrowing, and the US HouseholdLeverageCrisis,”AmericanEconomicReview,101,2132–2156. MILLAR, J. N., S. D. OLINER, AND D. E. SICHEL (2016): “Time-to-PlanLagsforCommercial ConstructionProjects,”RegionalScienceandUrbanEconomics,59,75–89. OH, H., C. YANG, AND C. YOON (2024): “LandDevelopmentandFrictionstoHousingSupply overtheBusinessCycle,”FEDSWorkingPaperNo.2024-010. OH, H. AND C. YOON (2020): “Time to Build and the Real-Options Channel of Residential Investment,”JournalofFinancialEconomics,135,255–269. RAMEY, V. A. (2021): “The Macroeconomic Consequences of Infrastructure Investment,” in Economic Analysis and Infrastructure Investment, ed. by E. L. Glaeser and J. M. Poterba, Chicago: UniversityofChicagoPress,chap.4,219–276. SAIZ, A. (2010): “TheGeographicDeterminantsofHousingSupply,”TheQuarterlyJournalof Economics,125,1253–1296. SUMMERS, L. (2008): “FiscalStimulusIssues: TestimonybeforetheJointEconomicCommittee,” TestimonybeforetheJointEconomicCommittee,Washington,DC,January,16. 20

(a)PlanningandConstructionTimes ByPropertyType ByQuintileofConstructionCost shtnoM 05 04 03 02 01 0 Planning Time Construction Time MU LO OF RE WH shtnoM 05 04 03 02 01 0 Planning Time Construction Time 1 2 3 4 5 (b)ProbabilityofAdvancingtoConstruction erahS detelpmoC revE 55. 5. 54. 4. 53. -.2 -.1 0 .1 .2 Price appreciation Figure1: PLANNING AND CONSTRUCTION TIMES BY PROPERTY TYPE AND CONSTRUCTION COST AND ESTIMATES OF THE RELATIONSHIP BETWEEN A PROJECT ADVANCING TO CON- STRUCTION AND PRICE APPRECIATION EARLY IN THE PROJECT’S LIFE. Note: Figure1aplots thedistributionofplanning(green)andconstruction(orange)timesbypropertytype(leftpanel) andconstructioncost(rightpanel). Propertytypesincludemultifamily(MU),hotels(LO),office (OF), retail(RE) andwarehouses (WH).Figure 1bpresents semi-linearregression estimatesof the relationship between whether a project eventually advances to construction and the commercial propertypriceappreciationinthefirstyearafteraplanisinitiated. Priceappreciationismeasuredby CoStar’scommercialpropertypriceindexforthegivenpropertytypeandMSA.Thespecification controlsforthenaturallogarithmofrealprojectcostandbuildingarea,andincludesMSA,property type, and quarter-of-plan start fixed effects. Estimates use the Stata “binsreg” command (see Cattaneoetal.2024). Thedotsreflectbinscatterestimatesandthelinereflectsalinearregression estimate; price appreciation is significant at the 0.1% level when standard errors are two-way clusteredbyMSAandquarter-of-planstart. Source: Authors’calculationsusingdatafromDodge Data&Analytics,Inc. andCoStarSuite(US).

)s000,#( stcejorP fo rebmuN 04 02 0 02- D Planning Stock Planning Starts Construction Starts Abandonments 4 6 8 0 2 4 6 8 0 2 4 0 0 0 1 1 1 1 1 2 2 2 0 0 0 0 0 0 0 0 0 0 0 2 2 2 2 2 2 2 2 2 2 2 Figure2: DECOMPOSITION OF CHANGES TO THE PLANNING STOCK. Note: Thisfigureshows thecomponentsofthechangeinthestockofprojectsinplanningbyyear. Theblacklineplotsthe changeinprojectsinplanning,whichisequaltoinflowsminusoutflows. Theredbarsareinflows intotheplanningstock,whichcomefromplanningstarts. Thebluebarsareoutflowsintheform of construction starts. The green bars are outflows in terms of abandoned and deferred projects. Source: Authors’calculationsusingdatafromDodgeData&Analytics,Inc. 22

3-yearCommercialConstructionResponse (1) (2) (3) (4) (5) (6) (7) (8) PlanRate 0.67** 0.67** 0.68** 0.63** 0.63** 0.60** 0.59** i,p,t (0.06) (0.06) (0.06) (0.07) (0.07) (0.06) (0.06) PriceGrowth 1.19** 1.16** 0.58 0.11 0.10 -1.78** -2.19** i,p,t (0.16) (0.17) (0.69) (0.13) (0.12) (0.57) (0.63) ×PlanRate 0.97** 0.93** 1.26** 1.25** 0.95** i,p,t (0.18) (0.18) (0.16) (0.16) (0.16) IV? (cid:88) (cid:88) (cid:88) (cid:88) LaggedPlan/Constr. (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) PropertytypeFEs (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) MSAFEs (cid:88) (cid:88) (cid:88) (cid:88) MSA-QuarterFEs (cid:88) (cid:88) (cid:88) (cid:88) Propertytypespecificslopes (cid:88) Observations 54,228 54,228 54,228 54,228 54,228 54,228 54,228 54,228 Table 1: 3-YEAR CUMULATIVE RESPONSE OF CONSTRUCTION STARTS TO PRICE APPRECIA- TION. Note: Thistable showsthe cumulative3-year response of construction starts(normalized to the initial building stock) to commercial price appreciation for a given MSA, property type, and quarter. Thefirstthreecolumnspresentestimateswiththeplanrateandpriceappreciationentering independently,whiletheremainingspecificationsincludetheirinteraction. IVestimatesinstrument for price appreciation and its interaction with the plan rate with national price appreciation for the given property type and quarter and its interaction with the local plan rate. Fixed effects for eachspecification areindicated atthe bottomof thetable; “Propertytype specificslopes” indicates thepresenceof propertytypedummiesinteracted withthelocalplanrate andlocalpropertyprice appreciation. Eachspecificationcontrolsforcumulativeconstructionstartsoverthepastyear,and four lags of the plan and construction rates (Lagged Plan/Constr.). Standard errors are two-way clusteredbyMSAandyear-quarter. +,*,**indicatesignificanceatthe10percent,5percent,and1 percentlevels,respectively. Source: Authors’calculationsusingdatafromDodgeData&Analytics, Inc. andCoStarSuite(US). 23

3-yearCommercialConstructionResponse (1) (2) (3) (4) (5) (6) PlanRate 0.67** 0.65** 0.66** 0.65** 0.59** 0.59** i,p,t (0.07) (0.06) (0.07) (0.06) (0.07) (0.08) PriceGrowth -1.86** 2.46 -2.23** 0.95 i,p,t (0.60) (1.91) (0.68) (1.74) ×PlanRate 1.31** 1.20** 1.30** 1.21** 1.46** 1.46** i,p,t (0.16) (0.16) (0.16) (0.17) (0.20) (0.20) ×ln(PropertyStock) -0.46+ -0.35 -0.40+ -0.21 i,p,t (0.27) (0.24) (0.23) (0.28) ×HousingElasticity (BH24) -1.46 -1.39 i (1.02) (1.09) ×RegulatoryIndex,1stPC (BGM24) 0.47+ 0.50* i (0.24) (0.21) ×RegulatoryIndex,2ndPC -0.64 -0.75+ i (0.46) (0.44) IV? (cid:88) (cid:88) (cid:88) LaggedPlan/Constr. (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) PropertytypeFEs (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) MSA-QuarterFEs (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) MSA×PriceGrowth (cid:88) MSA×NationalPriceGrowth (cid:88) Observations 38,127 38,127 38,127 38,127 38,127 38,127 Table 2: OTHER POTENTIAL DRIVERS OF SUPPLY ELASTICITIES. Note: This table shows the cumulative 3-year response of construction starts (normalized to the initial building stock) to commercial price appreciation for a given MSA, property type, and quarter, controlling for the interaction of price appreciation with other factors that potentially affect supply elasticities. Columns (1) and (3) repeat previous OLS and IV analysis for MSAs where additional factors affecting elasticities are available. Columns (2) and (4) add in interactions of price appreciation with the following variables to the specification: the logarithm of the initial property stock, the commercial-property-weightedhousingsupplyelasticityintheMSAfromBaum-SnowandHan (2024), and population-weighted-average of the first two principal components of the housing regulation variables collected by Bartik et al. (2023). Columns (5) and (6) repeat Columns (1) and(3), respectively, butadd ininteractionsof priceappreciation withthelogarithm ofthe initial property stock and interactions with market-level (Column (5)) or national (Column (6)) price appreciationwithMSAdummies. IVestimatesinstrumentforpriceappreciationanditsinteraction withthevariablesofinterestwithnationalpriceappreciationforthegivenpropertytypeandquarter and its interaction with the set of explanatory variables. Fixed effects for each specification are indicatedatthebottomofthetable. Eachspecificationcontrolsforcumulativeconstructionstarts over the past year, and four lags of the plan and construction rates. Standard errors are two-way clusteredbyMSAandyear-quarter. +,*,**indicatesignificanceatthe10percent,5percent,and1 percentlevels,respectively. Source: Authors’calculationsusingdatafromDodgeData&Analytics, Inc.,CoStarSuite(US),andpubliclyshareddatafromtheabove-listedpapers.

htworg ecirp pp1 fo spb ni tceffE 01 5 0 5- Commercial Construction Response 1sd Above Mean Mean Plan Rate 0 10 20 30 Quarters htworg ecirp pp1 fo spb ni tceffE 51 01 5 0 Commercial Construction Response 1sd Above Mean Mean Plan Rate 0 10 20 30 Quarters Figure3: LOCAL PROJECTIONS ESTIMATE OF COMMERCIAL PROPERTY SUPPLY ELASTICITIES. Note: This figure shows the cumulative response of construction starts (normalized to the initial building stock) to commercial property price appreciation in a market. The top panel plots the response of construction estimated by OLS, including CBSA-quarter fixed effects (following Column6ofTable1),whilethebottompanelpresentsestimatesinstrumentingforpriceappreciation with the national appreciation for the given property type (following Column 5 of Table 1). The y-axispresentsthecumulativenumberofconstructionstarts(inbasispoints)occurringinresponse to a1 percentagepoint increasein commercial price appreciation fora CBSA withan average plan rate(green)oraplanrate1standarddeviationabovethemean(red). Thex-axisindexesthenumber of quarters elapsed since the increase in price appreciation. The shaded regions give 95 percent confidence intervals. Standard errors are two-way clustered by MSA and year-quarter. Source: Authors’calculationsusingdatafromDodgeData&Analytics,Inc. andCoStarSuite(US).

SUPPLEMENTARYMATERIALS Thisdocumentcontainsthesupplementarymaterialsasreferencedinthemanuscript. S.1. SupplementaryFiguresandTablesReferencedinText 1

htworG Y-o-Y 57. 5. 52. 0 52.- 5.- 57.- Commercial Structure Investment Private Domestic Investment Residential Investment 1970 1980 1990 2000 2010 2020 Figure S1: COMMERCIAL VS. TOTAL AND RESIDENTIAL AND PRIVATE DOMESTIC INVEST- MENT. Note: The figure shows year-over-year changes in investment in nonresidential and multifamily structures (black—which we label as commercial structure investment), gross private domesticinvestment(red),andsingle-familyhousinginvestment(bluedashed–whichwelabelas residential investment). Commercial Structure investment is calculated from the sum of private investmentinnonresidentialandmultifamilystructures(FREDseriesB009RC1Q027SBEAand C292RC1Q027SBEA,respectively). TotalprivateinvestmentiscalculatedfromFREDseriesGPDI. Residentialinvestmentiscalculatedasprivateinvestmentinsingle-familyhousingstructures(FRED seriesA944RC1Q027SBEA).Notethatnonresidentialstructuresincludestructuretypesotherthan thoseinthemicrodatainthispaper,suchasmanufacturingandpower. Source: Authors’calculations usingdatafromtheBureauofEconomicAnalysis,retrievedfromFRED. 2

shtnoM 03 02 01 0 Planning Time Construction Time e c at bli Pri v P u FigureS2: PLANNING AND CONSTRUCTION TIMES BY PRIVATE OR PUBLIC OWNERSHIP. Note: Thisfigure plotsthe distribution ofplanning (green)and construction (orange)times bywhethera projectis privateor public. Source: Authors’calculations usingdata fromDodgeData &Analytics, Inc. 3

(a)MultifamilyStarts stinU fo sdnasuohT 06 04 02 0 (b)MultifamilyUnderConstruction Dodge Census/HUD Jan2005 Jan2007 Jan2009 Jan2011 Jan2013 Jan2015 Jan2017 Jan2019 Jan2021 Jan2023 Jan2025 stinU fo sdnasuohT 0001 008 006 004 002 0 Dodge Census/HUD Jan2005 Jan2007 Jan2009 Jan2011 Jan2013 Jan2015 Jan2017 Jan2019 Jan2021 Jan2023 Jan2025 (c)MultifamilyCompletions stinU fo sdnasuohT 08 06 04 02 0 Dodge Census/HUD Jan2005 Jan2007 Jan2009 Jan2011 Jan2013 Jan2015 Jan2017 Jan2019 Jan2021 Jan2023 Jan2025 FigureS3: MULTIFAMILY HOUSING STATISTICS. Note: Wecomparedatainimpliedunitstarts, units under construction, and unit completions from the Dodge data to reported statistics for buildingswithfiveormoreunitsfromtheU.S.CensusandtheDepartmentofHousingandUrban Development. Source: Authors’ calculations using data from Dodge Data & Analytics, Inc. and the U.S. Census and the Department of Housing and Urban Development, retrieved from Haver Analytics. 4

(a)OfficeCompletions sgnidliuB fo rebmuN 0052 0002 0051 0001 005 0 (b)RetailCompletions Dodge CoStar Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 2005: 2007: 2009: 2011: 2013: 2015: 2017: 2019: 2021: 2023: 2025: sgnidliuB fo rebmuN 0005 0004 0003 0002 0001 0 Dodge CoStar Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 2005: 2007: 2009: 2011: 2013: 2015: 2017: 2019: 2021: 2023: 2025: (c)WarehouseCompletions sgnidliuB fo rebmuN 0051 0001 005 0 Dodge CoStar Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 2005: 2007: 2009: 2011: 2013: 2015: 2017: 2019: 2021: 2023: 2025: FigureS4: COMPARISON OF COMPLETIONS IN DODGE REAL ESTATE ANALYZER TO COSTAR. Note: WarehousesarecomparedtologisticalindustrialpropertiesfromCoStar. Source: Authors calculationsusingdatafromDodgeData&Analytics,Inc. andCoStarSuite(US). 5

Unweighted Weighted Mean Std Median Mean Std Median N PlanningStarttoConstructionStart(months) 11.7 13.9 7 18.9 19.9 13 179,529 ConstructionStarttoCompletion(months) 9.1 6.8 7 18.3 12.2 16 174,433 PlanningStarttoAbandonment(months) 32.7 18.2 36 34.4 20.7 36 112,432 PlanningStarttoCompletion(months) 20.6 16.4 15 36.0 24.1 30 170,822 ProjectConstructionValue(millionsof2012USD) 12.7 61.2 3.0 386,648 BuildingArea(1000sofSq.Ft.) 104.4 731.2 30.1 386,648 AbandonmentShare 0.46 0.47 HotelShare 0.05 0.1 OfficeShare 0.19 0.2 RetailShare 0.31 0.11 WarehouseShare 0.14 0.12 MultifamilyShare 0.3 0.47 Table S1: SUMMARY STATISTICS FOR ALL PROJECTS. Note: Thistable showssummary statistics forthe projectsin oursampleon anunweighted andweighted(by realproject value) basis. Source: Authors’calculationsusingdatafromDodgeData&Analytics,Inc. 6

Hotel Unweighted Weighted Mean Std Median Mean Std Median N PlanningStarttoConstructionStart(months) 13.7 14.7 9 19.3 18.5 14 9,018 ConstructionStarttoCompletion(months) 13.4 7.9 12 21.0 11.7 19 8,590 PlanningStarttoAbandonment(months) 32.1 20.2 36 33.2 24.3 36 5,270 PlanningStarttoCompletion(months) 26.7 17.1 22 39.2 22.8 33 8,256 ProjectConstructionValue(millionsof2012USD) 22.3 106.3 7.8 20,284 BuildingArea(1000sofSq.Ft.) 143.8 497.4 68.8 20,284 AbandonmentShare 0.51 0.59 Office Unweighted Weighted Mean Std Median Mean Std Median N PlanningStarttoConstructionStart(months) 9.6 12.7 6 18.3 22.1 11 33,046 ConstructionStarttoCompletion(months) 8.6 5.7 8 21.1 13.4 19 32,635 PlanningStarttoAbandonment(months) 32.6 18.3 36 36.3 22.2 36 22,275 PlanningStarttoCompletion(months) 18.4 15.0 14 38.3 26.4 32 31,994 ProjectConstructionValue(millionsof2012USD) 14.9 91.4 2.0 71,214 BuildingArea(1000sofSq.Ft.) 72.2 268.9 17.1 71,214 AbandonmentShare 0.48 0.48 Retail Unweighted Weighted Mean Std Median Mean Std Median N PlanningStarttoConstructionStart(months) 8.3 10.8 5 11.8 14.9 7 57,193 ConstructionStarttoCompletion(months) 5.7 3.1 6 8.9 7.4 7 56,603 PlanningStarttoAbandonment(months) 33.2 18.8 36 32.9 18.9 36 35,611 PlanningStarttoCompletion(months) 14.2 11.8 10 20.6 18.2 14 56,089 ProjectConstructionValue(millionsof2012USD) 3.9 35.9 1.1 115,020 BuildingArea(1000sofSq.Ft.) 41.3 283.1 11.0 115,020 AbandonmentShare 0.44 0.55 Warehouse Unweighted Weighted Mean Std Median Mean Std Median N PlanningStarttoConstructionStart(months) 10.3 13.0 6 14.0 16.9 8 23,058 ConstructionStarttoCompletion(months) 5.8 3.9 5 9.6 6.3 8 22,331 PlanningStarttoAbandonment(months) 33.3 17.6 36 33.3 19.7 36 18,612 PlanningStarttoCompletion(months) 16.4 14.4 12 23.5 18.6 18 22,098 ProjectConstructionValue(millionsof2012USD) 10.7 27.8 3.1 57,508 BuildingArea(1000sofSq.Ft.) 147.5 365.9 45.2 57,508 AbandonmentShare 0.50 0.41 Multifamily Unweighted Weighted Mean Std Median Mean Std Median N PlanningStarttoConstructionStart(months) 16.6 15.9 12 21.5 20.0 16 57,214 ConstructionStarttoCompletion(months) 13.5 8.1 12 20.8 11.9 18 54,274 PlanningStarttoAbandonment(months) 32.1 17.4 36 34.6 19.6 36 30,664 PlanningStarttoCompletion(months) 29.6 17.7 25 40.8 23.2 35 52,385 ProjectConstructionValue(millionsof2012USD) 19.1 58.3 8.1 122,622 BuildingArea(1000sofSq.Ft.) 155.5 1207.0 75.0 122,622 AbandonmentShare 0.45 0.43 TableS2: SUMMARY STATISTICS FOR PROJECTS BY PROPERTY TYPE. Note: Thistableshows summarystatisticsfortheprojectsinoursampleonanunweightedandweighted(byrealproject value) basis, broken out by property type. Source: Authors’ calculations using data from Dodge Data&Analytics,Inc. 7

phase[t+1] phase[t] Pre-Planning Planning FinalPlanning Bidding Underconstruction Completed Deferred Abandoned Total Row% Row% Row% Row% Row% Row% Row% Row% Row% Pre-Planning 95.51 1.25 0.10 0.11 0.32 0.13 0.41 2.18 100.00 Planning 0.09 94.81 0.49 0.24 1.77 0.37 0.70 1.53 100.00 FinalPlanning 0.03 0.80 86.97 2.04 7.59 0.78 0.67 1.11 100.00 Bidding 0.06 0.44 0.41 84.63 11.45 1.09 0.42 1.51 100.00 Underconstruction 0.00 0.00 0.00 0.00 91.14 8.80 0.03 0.03 100.00 Deferred 0.06 0.12 0.12 0.07 0.38 0.15 96.48 2.62 100.00 Total 13.27 39.44 3.24 3.35 25.53 2.57 11.27 1.33 100.00 Table S3: TRANSITION MATRIX FOR PHASE DATA Note: This table shows a transition matrix for the sample of projects consideredinthispaper. Specifically,eachcell (i, j) givestheshareofprojectsthatstartinphaseiinmontht transitions tophase j inmontht+1. Source: Authors’calculationsusingdatafromDodgeData&Analytics,Inc. 8

Abandonment (1) (2) (3) (4) (5) (6) PriceGrowth -54.15** -43.92** -46.67** -8.74* -53.83** -25.15* i,p,t (3.45) (3.40) (3.59) (3.82) (3.95) (10.82) LogRealProjectCost -3.18** -2.73** -2.56** -2.11** (0.12) (0.12) (0.13) (0.12) HousingElasticity (BH24) -3.60** -3.69** i (0.97) (1.00) RegulatoryIndex,1stPC (BGM24) 0.36+ 0.49* i (0.21) (0.21) RegulatoryIndex,2ndPC -0.72* -0.67* i (0.28) (0.28) PriceGrowth ×LogRealProjectCost -22.27** -21.76** i,p,t (0.95) (0.99) ×HousingElasticity (BH24) 17.92 i (11.78) ×RegulatoryIndex,1stPC (BGM24) -7.54** i (2.09) ×RegulatoryIndex,2ndPC -4.73 i (3.08) R2 0.101 0.138 0.147 0.151 0.119 0.123 a StartYearFEs (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) MSAFEs (cid:88) (cid:88) (cid:88) QuarterFEs (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) PropertytypeFEs (cid:88) (cid:88) (cid:88) (cid:88) Observations 286,154 286,154 286,154 286,154 286,154 286,154 Table S4: DETERMINANTS OF ABANDONMENT. Note: The dependent variable is an indicator if a project is abandoned. Each observation is one project. Columns (5) and (6) exclude CBSA fixedeffects toincludeCBSA-level measuresof housingsupplyelasticities fromBaum-Snowand Han(2024)andBartiketal.(2023). +,*,**indicatesignificanceatthe10percent,5percent,and1 percentlevels,respectively. Source: Authors’calculationsusingdatafromDodgeData&Analytics, Inc.,CoStarSuite(US),andpubliclyshareddatafromtheabove-listedpapers. 9

Pl a n ni n g Ti m e ( M o nt hs) C o nstr u cti o n Ti m e ( M o nt hs) A b a n d o n m e nt R at e ( %) I n Pl a n ni n g R at e 2 0 1 1 ( %) I n Pl a n ni n g R at e 2 0 1 9 ( %) NewYork-Newark-JerseyCity,NY-NJ-PA 12 11 37 0.69 1.83 LosAngeles-LongBeach-Anaheim,CA 14 10 53 0.18 0.61 Chicago-Naperville-Elgin,IL-IN-WI 6 7 51 0.27 0.51 Dallas-FortWorth-Arlington,TX 6 7 45 0.62 1.34 Houston-TheWoodlands-SugarLand,TX 6 6 36 0.50 0.76 Washington-Arlington-Alexandria,DC-VA-MD-WV 13 11 63 1.29 4.15 Philadelphia-Camden-Wilmington,PA-NJ-DE-MD 11 8 55 0.46 1.19 Miami-FortLauderdale-WestPalmBeach,FL 15 8 53 0.62 1.78 Atlanta-SandySprings-Roswell,GA 8 7 60 0.75 1.43 Boston-Cambridge-Newton,MA-NH 13 10 47 1.32 2.28 Phoenix-Mesa-Scottsdale,AZ 7 7 48 0.88 1.19 SanFrancisco-Oakland-Hayward,CA 16 11 55 0.22 0.77 Riverside-SanBernardino-Ontario,CA 10 6 61 0.98 1.31 Detroit-Warren-Dearborn,MI 7 7 56 0.22 0.42 Seattle-Tacoma-Bellevue,WA 13 10 42 0.38 1.71 Minneapolis-St.Paul-Bloomington,MN-WI 8 8 45 0.40 1.42 SanDiego-Carlsbad,CA 11 10 56 0.31 0.61 Tampa-St.Petersburg-Clearwater,FL 7 7 39 0.39 0.80 Denver-Aurora-Lakewood,CO 9 8 49 0.62 1.91 Baltimore-Columbia-Towson,MD 11 8 67 1.06 2.67 St.Louis,MO-IL 5 7 44 0.32 0.52 Orlando-Kissimmee-Sanford,FL 9 6 44 0.92 1.88 Charlotte-Concord-Gastonia,NC-SC 7 8 44 0.43 1.25 SanAntonio-NewBraunfels,TX 5 6 23 0.56 0.65 Portland-Vancouver-Hillsboro,OR-WA 10 9 47 0.32 0.92 Sacramento–Roseville–Arden-Arcade,CA 10 7 64 0.48 1.48 Pittsburgh,PA 8 7 52 0.44 0.81 Austin-RoundRock,TX 7 7 25 0.40 1.15 LasVegas-Henderson-Paradise,NV 8 7 60 0.41 0.97 Cincinnati,OH-KY-IN 5 8 41 0.46 0.70 KansasCity,MO-KS 6 7 48 0.58 1.21 Columbus,OH 7 8 44 0.65 1.04 Indianapolis-Carmel-Anderson,IN 4 7 31 0.32 0.63 Cleveland-Elyria,OH 6 7 44 0.30 0.33 SanJose-Sunnyvale-SantaClara,CA 14 11 60 0.62 1.04 Nashville-Davidson–Murfreesboro–Franklin,TN 6 8 29 0.52 1.17 VirginiaBeach-Norfolk-NewportNews,VA-NC 8 7 54 0.49 1.01 Providence-Warwick,RI-MA 12 7 50 0.56 0.99 Jacksonville,FL 6 6 39 0.32 1.14 Milwaukee-Waukesha-WestAllis,WI 7 8 42 0.36 0.46 OklahomaCity,OK 5 6 47 0.83 0.15 Raleigh,NC 11 8 44 0.70 2.15 Memphis,TN-MS-AR 5 6 31 0.26 0.74 Richmond,VA 8 8 55 0.61 1.44 Louisville/JeffersonCounty,KY-IN 6 7 41 0.53 1.09 NewOrleans-Metairie,LA 5 6 37 0.30 0.33 SaltLakeCity,UT 6 8 45 0.86 0.48 Hartford-WestHartford-EastHartford,CT 9 7 49 0.75 1.10 Buffalo-Cheektowaga-NiagaraFalls,NY 9 7 52 0.87 1.37 Birmingham-Hoover,AL 5 6 56 0.42 0.52 TableS5: CBSA-LEVEL STATISTICS. Note: Thetabledisplaysstatisticsfromtheconstructiondata forthe 50mostpopulousCBSAs inourdata asof2020 (sortedbypopulation). Source: Authors’ calculationsusingdatafromDodgeData&Analytics,Inc. 10

(a)AbandonmentRate 50 - 77 40 - 50 30 - 40 20 - 30 0 - 20 No data (b)InPlanRate2011 (c)InPlanRate2019 .69 - 3.59 1.15 - 6.52 .42 - .69 .69 - 1.15 .25 - .42 .39 - .69 .01 - .25 0 - .39 No data No data FigureS5: GEOGRAPHIC VARIATION IN ABANDONMENT AND PLANNING RATES. Note: Figure S5aistheCBSA-levelaverageabandonmentrateacrossourentiresample. FiguresS5bandS5care themedian planningratein 2011and 2019,respectively. Source: Authors’calculations usingdata fromDodgeData&Analytics,Inc. 11

3-yearConst. Response 1-yearConstructionResponse 2005-2018 2005-2018 2020-on (1) (2) (3) (4) (5) (6) PlanRate 0.63** 0.45** 0.13** 0.07** 0.07** 0.06** i,p,t (0.06) (0.04) (0.01) (0.01) (0.01) (0.02) ×PriceGrowth 1.34** 0.23** 0.07 i,p,t (0.18) (0.05) (0.07) ×BuildingCostGrowth 1.77* 0.60** -0.01 t (0.66) (0.14) (0.15) PriceGrowth 1.30** 0.04 0.19** -0.02 -0.01 -0.10 i,p,t (0.15) (0.13) (0.06) (0.05) (0.06) (0.06) BuildingCostGrowth 0.59** -0.26 0.33** -0.03 -0.25+ -0.20 t (0.17) (0.34) (0.06) (0.08) (0.12) (0.14) LaggedPlan/Constr. (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) PropertytypeFEs (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) MSAFEs (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) (cid:88) Observations 56,977 56,977 69,177 69,177 23,633 23,633 Table S6: TESTS FOR SUPPLY SHOCKS. Note: This table demonstrates that for the sample period in the paper, construction increases when construction costs rise, consistent with fluctuations in costs reflecting construction demand rather than supply shocks. However, there is evidence of constructionsupplyshocksdominatingduringthepandemic. BuildingCostGrowth istheannual t growthintheTurnerNonresidentialBuildingCostIndex,whileothervariablesareasinTable1. Odd columnsadd interactionsof building cost growth withthe planningstock. Columns (1)and (2) studytheconstructioncostatthe3-yearhorizonforthemainsampleperiod,whileColumns(3)and (4)studythe1-yearresponseofconstructionbeforethepandemic,andColumns(5)and(6)study the1-yearresponseduringthepandemic. +,*,**indicatesignificanceatthe10percent,5percent, and 1 percent levels, respectively. Source: Authors’ calculations using data from Dodge Data & Analytics,Inc. andCoStarSuite(US). 12

)tnecrep( etaR gninnalP 2 5.1 1 5. 0 Percentile 75th 50th 25th Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 2006: 2008: 2010: 2012: 2014: 2016: 2018: 2020: 2022: Quarter ytisneD 5.1 1 5. 0 Year 2011 2019 0 1 2 3 4 5 Planning Rate (percent) FigureS6: THE DISTRIBUTION OF PLANNING RATES OVER TIME. Note: Intheleftpanel,this figure presents time series of various quantiles of planning rates. In the right panel, the figure presentskerneldensitiesofthedistributionin2011and2019. MSAsareweightedbynumberof commercial properties. Source: Authors’ calculations using data from Dodge Data & Analytics, Inc. 13

htworg ecirp pp1 fo spb ni tceffE 51 01 5 0 5- Supply Elasticity (to Price Appreciation) 1sd Above Mean Mean Plan Rate 0 10 20 30 Quarters htworg ecirp pp1 fo spb ni tceffE 51 01 5 0 5- Supply Elasticity (to Price Declines) 1sd Above Mean Mean Plan Rate 0 10 20 30 Quarters Figure S7: ASYMMETRIC EFFECTS OF PRICE APPRECIATION. Note: This table shows the cumulative response of constructionstarts(normalizedtotheinitialbuildingstock)tocommercialpriceappreciationforagivenMSA,property type,andquarter. AllvariablesandspecificationsarethesameasthetoppanelofFigure3,butPriceGrowth isreplaced i,p,t withPriceGrowth + ≡max{PriceGrowth ,0}andPriceGrowth− ≡min{PriceGrowth ,0},thusallowingfora i,p,t i,p,t i,p,t i,p,t differentialresponsetopriceappreciationwhenpricesarerisingandfalling. Thetopandbottompanelreportestimated supply elasticities in markets experiencing positive and negative price growth, respectively. Green and red lines show theestimated responsefor marketswitha planrateat themean andonestandard deviationabove themean, respectively. Shadedregionsshow95percentconfidenceintervals. Standarderrorsaretwo-wayclusteredbyMSAandyear-quarter. +,*,**indicatesignificanceatthe10percent,5percent,and1percentlevels,respectively. Source: Authors’calculations usingdatafromDodgeData&Analytics,Inc,andCoStarSuite(US).

htworg ecirp pp1 fo spb ni tceffE 6 4 2 0 2- 4- Construction Employment Response 1sd Above Mean Mean Plan Rate 0 10 20 30 Quarters htworg ecirp pp1 fo spb ni tceffE 4 2 0 2- Construction Employment Response 1sd Above Mean Mean Plan Rate 0 10 20 30 Quarters Figure S8: LOCAL PROJECTIONS ESTIMATE OF RESPONSE OF COMMERCIAL CONSTRUCTION EMPLOYMENT TO PRICE APPRECIATION. Note: Thisfigureshowstheresponseofcommercialconstructionemployment(asapercentof initialtotalemployment)tocommercialpropertypriceappreciationinaCBSA.Thetoppanelplotstheresponseestimated by OLS, including CBSA and quarter fixed effects, while the bottom panel presents estimates instrumenting for price appreciation with the national appreciation (and omitting the quarter fixed effects). The y-axis presents the increase in constructionemploymentinresponsetoa1percentagepointincreaseincommercialpriceappreciationforaCBSAwith anaverageplanrate(green)oraplanrate1standarddeviationabovethemean(red). Thex-axisindexesthenumberof quarters elapsed since the increase in price appreciation. The shaded regions give 95 percent confidence intervals. All specificationcontrolforthenumberofconstructionstartsoverthepastyear,and4lagsoftheplanrate,theconstruction rate,andthelogarithmsofcommercialconstructionemploymentandtotalemployment. OLSestimatesincludeCBSAand quarterfixedeffects,whileIVestimatesonlyincludeCBSAfixedeffects. Standarderrorsaretwo-wayclusteredbyMSA and quarter. Source: Authors’ calculations using data from Dodge Data & Analytics, Inc., CoStar Suite (US), and the QuarterlyCensusofEmploymentandWages. 15

S.2. AdditionalDataDetails Exceptasotherwisestated,theadditionaldatacleaningdetailsapplytoourdata. Wekeepcommercialrealestateprojectsforthemainpropertytypes: multifamily(whichwegroup withapartments),hotel,office,retail,andwarehouse. We drop any observations marked as deleted by Dodge. These include projects Dodge has determined as duplicates or that do not meet their criteria for inclusion. We drop data where building areasquarefootageorprojectvalueismissing(allotherinformationisalwaysnon-missing). We drop any non-US projects. We drop any projects with only one observation. We also drop the small number of observations after the first instance that the project is indicated as completed or abandoned. Additionally, though Dodge typically marks projects as abandoned if they have not beenupdatedinthreeyearsintherecentdata,inthehistory,theydidnotalwaysdoso;inturn,we markanyprojectsthathavenotbeenupdatedinthreeyearsasabandoned. Some projectshave informationon phases forsub-projects (called “child”projects). We treat child projects as individual projects, but use the information for the planning phase from the master project. Notethattheconstructionspendisnotthevalueofthebuilding(ortheland)butrathertheestimated valuetobepaidtothegeneralcontractor(GC)beforeconstructioniscompleted,ortheactualspend ontheGCifconstructioniscompleted. Thespendvariableonlyincludesconstructioncostsandnot designfeesorothernon-constructioncosts. Weputtheconstructionspendin2012dollarsusingthe PCEpricedeflator.13 13WeusethePCEchain-typepriceindex,availableathttps://fred.stlouisfed.org/series/PCEPI.Wesetthedatetobe theminimumdateoftheprojectbeingcompletedifitreachesoverallcompletion,beingunderwayifitonlygetstothe underwayphase,biddingifitonlygetstothebiddingphase,andthelateststageofplanningifitgetstotheplanning stage. 16

S.3. AdditionalTheoryDetails Developer’sProblem TheLagrangianofthedeveloper’sproblemthatwesetupinSection3is: s (cid:34) (cid:16) κ (cid:90) t ∗ +s (cid:17) L =E ∑(∏(1+r ))−1 rb B − ι I p +λP κdF(κ) t t+i t+s t+s−1 t+s t+s t+s−1 s i=0 0 (cid:16) (cid:17) p p +q −P +(1−δ −λ)P +I t+s t+s p t+s−1 t+s (cid:35) (cid:16) (cid:17) +qb −B +(1−δ )B +λP F(κ ∗ ) , t+s t+s b t+s−1 t+s−1 t+s whereqp andqb arethecostatevariablesgivingtheshadowvalueofplanningstockandbuildings, respectively. ThishastheFOCs: ∂L 1 (cid:16) (cid:17) =−qb+E rb+(1−δ b)qb =0 ∂B t t 1+r t t+1 t t+1 ∂L p =−ι +q =0 p t t ∂I t ∂L =−κ ∗ λP f(κ ∗)+qb λP f(κ ∗)=0 ∂κ∗ t t−1 t t t−1 t t   κ∗ ∂L 1 (cid:90) t+1 =−q p +E λ (qb −κ)dF(κ)+q p (1−δ −λ)=0. ∂P t t 1+r  t+1 t+1 p  t t+1 0 Thefirsttwoexpressionsdefinetheoptimalinvestmentamountsasafunctionofbuildingvalues. p q =ι means that investors initiate planning starts until thevalueof a unit of planningequals the t t marginalcostofastart.14 Theexpressionκ∗ =qb meansthatdeveloperschoosetoproceedwith t t constructionprojectswheneverthecostislessthanthevalueofabuilding. The last two expressions define the values of projects in planning and of completed buildings. Combiningtheseconditionsovertimeshowsthatthesevaluesreflectthepresentdiscountedvalueof 14Thismarginalcostisfromtheperspectiveoftheindividualdeveloper. Sincethereareexternaladjustmentcosts, themarginalcostinaggregateishigher. 17

κ∗ t+1 payoutsfromplanningandconstruction. Forplanning,thispayoutisπ p ≡λ (cid:82) (qb −κ)dF(κ)— t t+1 0 thatis,theprobabilityoftheplanbeingcompletedmultipliedbythesurplusexpectedtobereceived from construction, or E (max(0,qb −κ)). For construction, the payout is the rent received on t t+1 thebuilding,rb. Thesediscountedvaluesare: t (cid:18) (cid:19)s 1−δ −λ q p =E ∑ π p t t 1+r t+s s t,t+s (cid:18) (cid:19)s 1−δ qb =E ∑ rb , t t 1+r t+s s t,t+s where1+r t,t+s ≡(∏ s i=0 (1+r t+i )) 1 s. Adjustment Costs We assume the following functional form for the costs to planning starts: ι =ι+ 1( Pt−P t−1). Thisspecificationimpliesthatcostsarequadraticinthenumberofstarts,with t φ P t−1 ι measuring the steady-state cost of a plan start and φ the elasticity of starts with respect to qp. Combiningthefirst-orderconditionthatq =ι withtheexpressionforP −P intheplanning p t t t−1 accumulationequation,wegetthat: I p =P (φ(qp−ι)+λ +δ ). t t−1 p Inthefirstorderconditionsabove,developerstakethecostofplanningstartstobeexogenous. This assumptionallowedforsimplerandmoreeasily-interpretedexpressionsinSection3,anditislikely economically reasonablegiventhat the construction sector is highly fragmented, leaving little room forindividualfirmstoaffectinputprices(D’Amicoetal.,2024). However,wefoundqualitatively similarresultswithinternaladjustmentcosts(i.e.,whendevelopersaccountedfortheeffectsplan startshadoncurrentandfuturecostsofplanstarts).15 15If we keep other parameter values the same, but make adjustment costs internal, we see a smaller difference betweentheresponseofhigh-andlow-planeconomies,becausedeviationsfromthesteadystateplanraterevertmore quickly. Ifwerecalibrateφ (inthewaywewilldiscussinSectionS.4.2),wefindsimilareffectswithinternaland externaladjustmentcosts. 18

S.4. QuantitativeModel Section S.4.1 presents the DSGE model, and Section S.4.2 discusses model calibration. Section S.4.3 presents how the planning stock affects supply elasticities in the model, and Section S.4.4 shows howresultsdifferfromanotherwiseequivalentmodelwithoutendogenousabandonment. Thetableandfiguresthatarereferencedinthissectionarepresentedattheendofthesection. S.4.1. DSGEModel The model has the following agents: households, capital producers, building producers (whose problem was defined in Section 3), final goods producers, and a government. Although most of theproblemoutsideofbuildingproductionisstandard,wereviewtheirproblemsinthatorderfor completeness. Households Attimet, arepresentativehousehold maximizeslifetime utility—whichis assumed tobeseparableandisoelastic—overconsumption(ofthefinalgood),C ,anditslaborsupplied,L : t t (cid:32) (cid:33) 1−γ C ω E ∑β s t+s − L1+ν , t 1−γ 1+ν t+s s whereω >0,ν >0,andγ >0. Thehouseholdmaximizesutilitysubjecttoabudgetconstraint: Dh +C =(1+r )Dh +w L +Π −T, (4) t+s t+s t+s t+s−1 t+s t+s t t where Dh is government debt held by households at time t; r is the one-period real return on t t governmentdebt;w istherealwagetheyarepaidfortheirlabor; Π areanynetprofitsreturnedby t t firms—developers, capital producers, and finalgoods producers—which households whollyown; andT arenettaxespaidtothegovernment. t Thesolutiontothehouseholdproblemthusimpliesstandardlabor-incomeandEulerequations: w −ωC γ Lν =0 t t t C −γ −βEC −γ (1+r )=0. t t t+1 t+1 19

Capital ProducerProblem Capitaldepreciatesat rateδ andisrented tofirmsatrental rate rk. k t Thereisthusarepresentativecapitalproducerthatsolvesthefollowingproblem: s 1 max E ∑(∏ )(rk K −Ik ), t 1+r t+s t+s−1 t+s s i=0 t+i subjecttothecapitalaccumulationequation: K =(1−δ )K +Ik . (5) t+s k t+s−1 t+s Giventherearenoadjustmentcoststocapitalinvestment, thefirst-ordercondition(FOC)fromthe capitalproducer’sproblemimpliesthestandardrentalrateofcapital: rk =r +δ . (6) t t k Final Good Sector A continuum of competitive firms produce output Y by hiring labor L at t t wagew andrentingcapitalandbuildings,K andB ,respectively,withtechnology:16 t t−1 t−1 Y =Z Kα B η L 1−α−η , (7) t t t−1 t−1 t where Z is firm productivity, α ∈ (0,1), and η ∈ (0,1−α). As in Section 3, buildings are t constructedwithaseparateinvestmentprocessfromcapital. Firmschoosetheamountoflabortouseinproductionandtheamountcapitalandbuildingstorent inordertomaximizeprofits(whicharezeroinequilibrium): s 1 E ∑(∏ )(Y −w L −rk K −rb B ). t 1+r t+s t+s t+s t+s t+s−1 t+s t+s−1 s i=0 t+i 16Wefollowtheconventionthatvariablesaredatedasofwhentheyaredetermined. Buildingsandcapitalusedat timet arechosenattimet−1. 20

WethusobtainthefollowingFOCs: w =(1−α−η)Z Kα B η L −α−η t t t−1 t−1 t rk =αZ Kα−1B η L 1−α−η (8) t t t−1 t−1 t rb =ηZ Kα B η−1 L 1−α−η . t t t−1 t−1 t Government,Clearing,andEquilibrium Thegovernmentcomesintotheperiodwithalevelof debtD ,whichisallheldbyhouseholds. Governmentspending,G ,isexogenouslyspecifiedandis t t financedwithtaxesandnewdebtissuance. Thegovernmentthusfacesbudgetconstraint:17 D (1+r )+G =D +T. (9) t t t t+1 t Governmentdebtissuanceisequaltohouseholdbondholdingssuchthat: D =Dh. (10) t t Givenasequenceofproductivitiesandgovernmentpolicies({Z ,G ,T } )andasetofinitial t+s t+s t+s s conditions(B ,P,K ,D ),acompetitiveequilibriumisasequenceofprices{r ,rk ,rb ,w } t t t t t+s t+s t+s t+s s and quantities {C ,L ,Y ,K ,B ,P ,Π ,D ,Dh } such that households and the t+s t+s t+s t+s t+s t+s t+s t+s t+s s producers of capital buildings and final goods all solve their respective maximization problems, households’laborsuppliedequalsfirmlabordemanded,capitalandbuildingssuppliedbycapital and building producers are equal to capital and buildings demanded, respectively, building and capital accumulation follow equations (1) and (5), and bond markets clear following equation (10).18 17Wemakethestandardassumptionofnon-explosivegovernmentdebt. 18Wewriteallbudgetconstraintsasbinding,butifthesewerewrittenasinequalities,theywouldalsoneedtoholdin equilibrium. 21

S.4.2. Calibration WepresentthecalibrationsforthemodelparametersinTableS7. Thetable’sparametersaregrouped firstbythestandardmacroparametersandthenbythenovelconstruction-relatedparameters. Wecalibratetherelativeutilityweightonlabor,ω,andproductivityinthesteadystate,Z,sothat aggregatelaborsupply,L,andaggregateoutput,Y,arenormalizedto1,leadingtovaluesof0.91and 0.49,respectively. Wesetgovernmentspending,taxes,andgovernmentdebttozero. Formostof theotherstandardmacroparameters,wefollowGertlerandKaradi(2013). Specifically,following theirwork,wecalibrateβ tohita2percentinterestrate,leadingtoavalueof0.995(quarterly),the inverseFrischelasticity,ν,to0.276(whichimpliesaFrischelasticityofabout3.6),andthecapital depreciationrate,δ ,to0.025. Wesettherelative riskaversion parameter, γ,to1followingChetty k (2006). Given that we introduce buildings as a second capital input into production, we set the sum of α andη sothatthecapitalshareofincome(inclusiveofbothK andB)isthestandardvalueof1/3. Wesettherelativeincomesharestomatchtheestimatefrom? thatrealestateisabout30percent ofafirm’sbookassets(i.e.,we setthesevariablestosatisfy qbB = 3);this conditiongivesusthat K 7 α =0.287andη =0.046.19 Theotherkeybuilding andplanningparametersarecalibratedasfollows. Wesetthehazard from planningtoconstruction,λ,to0.167tohaveasix-quarteraveragetime-to-plan. Wesetthebuilding depreciationrate,δ ,tobe0.0062tomatchtheannualdepreciationrateforofficebuildingsusedin b thenationalincomeandproductaccounts(NIPAs).20 δ isnotseparatelyidentifiedfromλ andthe p parameterspertainingtothedistributionofconstructioncosts,sowejustsetittothesamevalueas δ .21 k Regarding parameterspertaining to the costof planning starts, weset ι (reflecting thesteady-state costofplanningstarts)tonormalizeqb to1. φ,whichreflectshowelasticplanstartsare,doesnot affectthesteadystatebutaffectshowquicklytheplanningstockrespondstoshocks. Wesetthis 19SeeFigure3in?. 20The annual depreciation rate for office buildings is 0.0247, which is around the middle of the range of depreciationestimatesforprivatenonresidentialstructures. Seehttps://apps.bea.gov/national/pdf/BEA_ depreciation_rates.pdf. 21Ahigherdepreciationrateisequivalenttohavingacombinationofahigherλ,butalsoanincreaseintheprobability thatdrawsareunfavorable. 22

to 1.25 so that a 50 percent reduction in P would have a half-life of six years, which is roughly consistentwiththepost-GFCrecoveryshowninFigureS6. Wetake thedistributionforconstructioncoststobeParetodistributed: F(κ)=1−(s)a,wheres κ istheminimumpossiblecostofconstructionanda>1determineshowmuchmassisaroundthis minimum. This makes the probability of abandonment 1−F(q ) = ( s )a and the expected conb q b structionexpenditure(ifconstructiongoesahead)equaltosa a (q1−a−s1−a). Wecalibratesand 1−a aso that the probabilityof abandonmentis 47percent (matchingthe value-weightedabandonment shareinTableS2)andconstructioncostsare85percentofbuildingvalues.22 S.4.3. BuildingSupplyElasticitiesandthePlanningStock Wenowpresentthequantitativeresultsfromthemodel. InSection 4,weshowedthatconstruction responds more to commercial price appreciation in localities with a greater stock of projects in planning. Wenowdemonstrateequivalentdynamicsinthecalibratedmodel. Figure S9 plots the response of construction to a 1 percent TFP shock that decays at a rate of 20 percent per quarter in two economies differing only in their initial levels of planning stock. The orangelineshowstheeffectoftheshockinamarketstartingatthesteadystate,whereastheblue lineshows theresponse foran economystartingat onlyhalf ofthe steady-statelevelof P.23 The1 percentTFPshockinitiallyraisesbuildingproductionbyabitmorethanaquarterofapercentage pointinthesteadystate,butonlybyabouthalfthisamountforthelowplanningeconomy.24 The effect of the TFP shock on construction then grows over time as developers start to build up the planningstock,resultinginmoreconstructionstartsovertime. However,thisprocessisdelayedfor the economy with a low planning stock; construction activity does not peak until six years after the 22We assume that “soft costs” as a share of building value are similar to those in the NAIOP’s report on the economic impacts of commercial real estate investment. The NAIOP estimates that in 2018 developers incurred 31.71 billion dollars in soft costs relative to total expenditure of 207.77 billion dollars. See Table 2 here: https://www.naiop.org/globalassets/research-and-publications/ report/economic-impacts-of-commercial-real-estate-2019-edition/ researchreportnaiop-2019-fuller-report-online-version.pdf. 23SincetheeconomywithalowPwouldhavetransitionaldynamicsevenabsenttheshock,theeffectoftheshock is (cid:0) I t b +s ({Z(cid:48)};P t =.5P)−I t b +s ({Z};P t =.5P) (cid:1) /Ib,whereI t b +s ({Z(cid:48)};P t )andI t b +s ({Z};P t )areconstructionlevelsthat wouldoccurwithandwithouttheshock,andIbandParesteady-statebuildinginvestmentandplanlevels. 24Theincreaseintheprobabilitythataprojectadvancestoconstructionisroughlysimilarinbotheconomies,sothe differenceisthatthelowplanningeconomyhasonlyhalfasmanyplanningprojectsavailabletoadvance. 23

shock(comparedwithfouryearsunderthesteady-stateinitialconditions). SinceTFPshocksdrivechangesinbothbuildingvaluesandconstructionactivity,wecanalsopresent theseimpulsesalongthelinespresentedinFigureS10. InFigureS10,weplotthecumulativere- (cid:18) (cid:19) s sponseofbuildingconstructionasashareofthebuildingstock, ∑ (Ib ({Z(cid:48)})−Ib ({Z})) /B, t+i t+i i=0 normalized by the price appreciation caused by the shock, (qb−qb)/q . The left panel plots the t b cumulativeelasticityofconstructionstartswithrespecttopriceappreciationforthelow-plan-rate andsteady-stateeconomies,whiletherightpanelplotsthedifferenceinelasticities. The effects of the shocks are qualitatively similar to the local projection estimates displayed in Figure3. Thecumulativeeffectofthepriceshockrisessteadilyovertime,asinthedata,withthe effectlevelingoffafter aboutsixyears.25 Therise inconstructionis slowerfor theeconomywith thelowerinitialplanningstock,butthedifferencelevelsoffafteraboutfiveyears. Altogether,the calibratedmodelisbroadlyconsistentwiththepatternsinthelocalprojections,thoughthemodel estimatessupplyasmoreelastic.26 S.4.4. Endogenousvs. ExogenousAbandonments Forourfinalexercise,weanalyzetheroleendogenousabandonmentplaysinthemodel. InFigure S11, we present impulses of construction in response to a 1 percent TFP shock in the baseline calibrated model and an alternative model without endogenous abandonment. In this alternative model,thereisjustafixedprobabilitythatprojectscompletingplanningareabandonedandafixed costtoundertakingconstruction. Wesetthisabandonmentprobabilityandconstructioncostequal tothesteady-stateabandonmentprobabilityand(average)constructioncostsothatthesteadystates ofthetwomodelsareidentical.27 25Wecannotcomputesimilarestimatesoflongerhorizonsinthedata,aswehaveashorttimeseriesandthusquickly losedegreesoffreedom. 26Thereareacoupleoffactorsthatmightcontributetothisdifferenceinelasticitiesoutsideofmodelmisspecification. First,empiricalestimatesofthesupplyelasticitywouldbebiaseddownwardtotheextentthatpricechangesreflect supplyshocks(movementsalongthedemandcurve). Second,priceappreciationismeasuredwitherrorsincetheCRE marketisilliquidenoughthatthepriceindexisbasedonappraisalsratherthantransactions. 27Thisexogenousabandonmentmodelcanbethoughtofintermsofthebaselinemodelbutwithadiscretedistribution ofconstructioncosts: thecostisarbitrarilyhighwiththeprobabilityofabandonmentandequaltotheaveragecostof constructionotherwise. Theimportantdifferencebetweenthemodelsisthuswhetherornotthereareprojectsthatare atthemarginofbeingabandoned. 24

The main takeaway from this exercise is that endogenous abandonment speedsup the response of constructiontodemandshocks. Intheexogenousabandonmentmodel,thereisnomechanismto increaseconstructionimmediately: constructionisjustaconstantshare(thehazardofcompletion multipliedbytheexogenousprobabilityofconstruction)ofthepredeterminedplanningstock. This means that construction only rises because of the initiation of new projects in planning, which eventuallytranslate intonew construction. Incontrast, withendogenousabandonment,construction risesimmediatelybecauseofareductioninthenumberofprojectsinplanningbeingabandoned. 25

Parameters Value Description Target StandardMacroparameters ω 0.907 LaborDisutility L=1 Z 0.490 Productivity Y =1 β 0.995 HouseholdDiscountFactor r=2%(annual) γ 1.0 CoefficientofRelativeRiskAversion Chetty(2006) ν 0.276 InverseFrischelasticityoflaborsupply GertlerandKaradi(2013) δ 0.025 CapitalDepreciation GertlerandKaradi(2013) k α 0.287 Kincomeshare Capital(K+B)share= 1 3 ConstructionandPlanningParameters η 0.046 Bincomeshare qbB = 3 K 7 λ 0.167 HazardofCompletingPlanning 1.5-yearplantime δ 0.025 PlanningDepreciationRate Equatetoδ p k δ 0.0062 BuildingDepreciationRate NIPA b ι 0.067 CostofPlanningStart qb =1 φ 1.25 PlanningAdjustmentCosts Post-GFCPlanStockRecovery s 0.744 Min. ConstructionCost(paretodist.) 15%softcoststoconstruction a 2.556 Paretoshapeparameter 37%abandonmentfromplanning TableS7: CALIBRATION. Note: Thistablepresentsthecalibrationoftheparametersforthemodel. Fromlefttoright,thecolumnsprovidetheparameter,thecalibratedvalue,adescriptionofwhatthe parameterreflects,andthetarget. 26

1.2 1.0 0.8 0.6 0.4 0.2 0.0 0 10 20 30 40 50 Quarters )level SS fo %( tceffE kcohS PFT Low Planning Starting as SS FigureS9: CONSTRUCTION INVESTMENT RESPONSE TO A TFP SHOCK BY PLANNING STOCK. Note: This figure shows the impulse response of construction to a 1 percent positive TFP shock. The orange line plots the response in an economy starting at the steady state, while the blue line plots the effect of the shock in an economy with an initial planning stock of half the steady state level. Formally,thebluelineplotsthesequence (cid:0) Ib ({Z(cid:48)};P =.5P)−Ib ({Z};P =.5P) (cid:1) /Ib t+s t t+s t and the orange line (cid:0) Ib ({Z(cid:48)};P =P)−Ib(cid:1) /Ib, where arguments without time subscripts denote t+s t steady-statelevelsandZ(cid:48) =Z+.01×.8s. t+s 27

2.0 1.5 1.0 0.5 0.0 0 10 20 30 40 50 Quarters )level SS fo %( tceffE kcohS noitaulaV 0.4 Low Planning Starting as SS 0.3 0.2 0.1 0.0 0 10 20 30 40 50 Quarters )level SS fo %( tceffe kcohS Difference Figure S10: CUMULATIVE BUILDING RESPONSE TO PRICE APPRECIATION. Note: This figure shows the cumulative response of building construction (as a share of the steadystate building stock) to a 1 percentage point building value shock. The left figure plots (cid:18) (cid:19) s ∑ (Ib ({Z(cid:48)})−Ib ({Z})) /B, normalized by the price appreciation caused by the shock, t+i t+i i=0 (cid:0) qb({Z(cid:48)})−qb ({Z}) (cid:1) /qb ({Z}). ThesequencesIb ({Z(cid:48)}) andqb({Z(cid:48)}) arebuildinginvestt t−1 t−1 t+i t ment and building values i quarters after the shock to Z, and these functions with respect to {Z} givetheinvestmentandvaluesthatwouldoccurwithouttheshock(whichwouldcorrespondwith steady-state values for the economy starting there). These sequences are plotted for an economy startingatthesteadystate(orange)andonestartingwithaP ofhalfthislevel(blue). Theleftpanel t showsbothresponsesindividually,whiletherightpanelplotstheirdifference(baselineminuslow initialplanningstock). 28

1.2 1.0 0.8 0.6 0.4 0.2 0.0 0 10 20 30 40 50 Quarters ss morf noitaived % Endogenous Abandonment Exogenous Abandonment FigureS11: EFFECT OF ENDOGENOUS ABANDONMENT. Note: Thisfigureshowstheconstruction responsetoa1percentTFPshockinthemodelwithendogenousabandonment(blue)andamodel withexogenousabandonment(orange). Theexogenousabandonmentmodelhasfixedabandonment rates and construction costs equal to steady-state abandonment rates and average construction costs intheendogenousabandonmentmodel. 29