The Green Corporate Bond Issuance Premium

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1346 June 2022 The Green Corporate Bond Issuance Premium John Caramichael and Andreas Rapp Please cite this paper as: Caramichael, John and Andreas Rapp (2022). “The Green Corporate Bond Issuance Premium,” International Finance Discussion Papers 1346. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2022.1346. NOTE: International Finance Discussion Papers (IFDPs) 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 International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

The Green Corporate Bond Issuance Premium JohnCaramichael* AndreasC.Rapp† May2022 Abstract Westudyaglobalpanelofgreenandconventionalbondstoassesstheborrowing cost advantage at issuance for green bond issuers. We find that, on average, green bonds have a yield spread that is 8 basis points lower relative to conventional bonds. This borrowing cost advantage, or greenium, emerges asof2019andcoincideswiththegrowthofthesustainableassetmanagement industry following EU regulation. Within this context, we find that the greeniumislinkedtotwoproxiesofdemandpressure,bondoversubscriptionand bondindexinclusion. Moreover,whilegreenbondgovernanceappearstomatter for the greenium, the credibility of the underlying projects does not have a significant impact. Instead, the greenium is unevenly distributed to large, investment-grade issuers, primarily within the banking sector and developed economies. These findings have implications for the role of green bonds in incentivizingmeaningfulgreeninvestmentsthroughouttheglobaleconomy. Keywords: Green bonds, corporate bonds, green finance, sustainable finance, climatefinance,greenbondpremium,bondissuance JELClassifications: C33,G15,G18,G23,G28,Q54,Q56 *BoardofGovernorsoftheFederalReserveSystem.Email:john.caramichael@frb.gov. †BoardofGovernorsoftheFederalReserveSystem.Email:andreas.c.rapp@frb.org. ForhelpfuldiscussionsandvaluablecommentswethankRicardoCorrea,MicheleDathan,AnilJain,Yuriy Kitsul,GordonLiao,AndrewMcCallum,ViktorsStebunovs,EmreYoldas,TonyZhang,andparticipantsatthe DivisionofInternationalFinancePolicyWorkshopandtheGeorgetown-IDBGreenBondMethodsSeminar. Theviewsexpressedinthispaperaresolelytheresponsibilityoftheauthorsandshouldnotbeinterpretedas reflectingtheviewsoftheBoardofGovernorsoftheFederalReserveSystemoranyotherpersonassociated withtheFederalReserveSystem.Firstdraft:October2021.Finaldraft:May2022.

1 Introduction Accordingtointergovernmentalorganizationsandresearchinstitutesworldwide,toslash greenhousegasemissionstonetzeroby2050,itwillbenecessarytoinvestupto$275trillioninphysicalassetsoverthenextthirtyyears.1 Asubstantialpartofthisinvestmentwill likely come from the private sector. One prominent instrument for financing this private sectorinvestmentisthegreencorporatebond,along-termfixedincomedebtsecuritythat israpidlygainingpopularity. Greencorporatebondsaresimilartoconventionalcorporate bonds,buttheycontainprovisionsthatdirectthefundingraisedfromthebond’sissuance towardsenvironmental(green)projects. The issuance of green corporate bonds has grown rapidly in recent years, totaling almost $400 billion in 2021 (see Figure 1a). As of 2021, green corporate bonds account for nearly six percent of global corporate bonds outstanding, up from less than one percent in 2014 (seeFigure1b). Duringthisperiodofrapidgrowth, greenbondshavebeencitedasapotentialdriveroflarge-scale,rapidclimateinvestment. Theyhavebeencriticized,however, for their lack of standardization, high cost of issuance, and the potential for ”greenwashing”, or misusing a green label for a bond that does not finance eligible green projects. Furthermore, it remains unclear if green bonds actually incentivize green investment, or if they are an instrument that merely identifies green investments that otherwise would havebeenmadeandfinancedwithaconventionalbond. Figure1: Growthingreencorporatebonds (a)Annualissuancevolume (b)Shareofcorporatebondsoutstanding Billion ($) Share (%) 400 6 350 5 300 4 250 200 3 150 2 100 1 50 0 0 2014 2015 2016 2017 2018 2019 2020 2021 2014 2015 2016 2017 2018 2019 2020 2021 Panel (a) is the annual issuance volume of the global green corporate bond market. Panel (b) is thenotionalshareofgreencorporatebondsrelativetothenotionalsizeofthetotalcorporatebond marketatyear-end. Source: BloombergFinanceLP(2021). 1See,forexample,McKinsey(2022)aswellastheGlobalFinancialMarketsAssociationandBostonConsultingGroup(2020). Fortheenergysectorspecifically,whichaccountsforalargeportionofemissions,estimatesincludethosefromtheEnergyTransitionsCommission(2021),theInternationalEnergyAgency(2021), andBloombergNEF(2021)at$50trillion,$100trillion,and$92to$173trillionrespectively. 2

In this paper, we investigate whether green bonds offer a direct incentive to corporations that wish to invest in green projects. The potential direct incentive is in the form of a borrowingcostadvantage(loweryieldspreadatissuance)forgreenbondsversusconventional bonds, also called the ”greenium”. Our analysis focuses on three main questions: First, is there empirical evidence for an average greenium at issuance that is statistically significant? And, if so, is this average greenium economically meaningful so as to constitute a direct economic incentive for green investment? Second, does the greenium vary over time, and if so, how does this time variation connect with the development of the green bond market? Third, how is the greenium distributed across bond-level and issuer characteristics? In addressing these questions, we leverage a comprehensive global sample of the primarybondmarket,containing1,169greenand129,043conventionalcorporatebondsover a sample period from 2014 to 2021. We analyze our sample with a detailed and wellspecified fixed effects regression specification that accounts for potential nonlinearities, issuer- and bond-specific time-variation, and the pricing dynamics of the global corporate bond market. Our analysis relies on traditional financial securities data, such as the price,rating,andmaturityofeachbond,aswellasmetricsspecifictogreenbonds,suchas whether a green bond experienced strong investor demand at issuance or whether it adheres to voluntary green bond standards. We contribute to the literature by using a wellspecifiedempiricalapproachtounderstandtheprimarymarketpricingofgreencorporate bondsinaglobalcontext,withaneyetowardsthedifferentiatedpricingofthequalityand credibilityofgreencorporatebondsandthedifferencesbetweenvariousmarketsegments. Our main findings are as follows: On average, US dollar- and euro-denominated green corporate bonds offer a small borrowing cost advantage of about 8 basis points over our sampleperiod. Thisadvantageisdistinctfromagreenbondissuereffect,or”greenhalo”, whichisahypothesizedbenefitfromissuingagreenbondthatlowersanissuer’soverall borrowingcostsacrossallbonds,bothgreenandconventional(seeSection2). Asignificant greeniumemergesonlyin2019,whencorporatebondinvestorsandEuropeanUnionofficialsbegantoembracethegreenbondmarket. Wefindthatthegreeniumislinkedtotwo proxiesofdemandpressure, oversubscriptionandgreenbondindexinclusion, highlighting mechanisms through which the greenium can be allocated as demand for the bonds outpaces supply. While US dollar- and euro-denominated green bonds capture comparable greeniums, the greenium is, on average, allocated primarily to local euro and foreign USdollarissuers. Lastly,thegreeniumisunevenlydistributedtolarge,investment-grade issuers,primarilywithinthebankingsectoranddevelopedeconomies. With these findings, we argue that the issuance premium of green corporate bonds likely playsalimitedroleinincentivizingrapid,large-scalegreeninvestment. Instead,greencorporatebondsmayindirectlyincentivizegreeninvestmentbysignalingtheenvironmental credentials of the issuer. However, the signaling effect from green corporate bonds may be relatively suboptimal compared to alternative solutions, such as certain regulatory re- 3

quirementsorfinancinginstrumentsbasedongreenperformancetargetsratherthangreen projects. Situated within the critical need for investment to avert the worst potential outcomesofclimatechange,overrelianceongreencorporatebondsasanincentivemaylead toanundersupplyofgreeninvestment. The remainder of the paper proceeds as follows: In Section 2, we discuss the growth of thegreenbondmarket,thepotentialmotivationsforissuingandinvestingingreenbonds, and the existing body of literature on green bonds. In Section 3, we describe our dataset, regression methodology, and the limitations of our empirical approach. In Section 4, we presentandanalyzeourempiricalresults. InSection5,wediscussthebroadersignificance ofourresultsandconcludethepaper. 2 Background: Green bonds and the greenium Before we begin our analysis of green corporate bonds, we first review key concepts and developments within the green bond market at large. Green bonds have been one of the most recognized instruments for the financing of green projects. While the stock of green loans is also growing, most green debt is in the form of green bonds. Like conventional bonds,greenbondsarelong-termfixedincomedebtinstruments. Theygenerallyhavethe sameseniority,recourse,andratingasanissuer’sconventionalbonds. Themaindifference betweengreenandconventionalbondsisthattheproceedsofagreenbondareearmarked for climate- and environment-friendly projects.2 This typically includes investments into cleanenergy,energyefficiency,greenbuildings,orelectrifiedtransportation.3 Greenbonds relyonthird-partycertificationorthetrustofinvestorstoassurethatthebonds’proceeds arechanneledtowardtheintendedgreeninvestments. Inshort,greenbondsareessentially aconventionalbondpackagedwitha”greenpromise”.4 Greenbondscurrentlylackauniversalglobalregulatoryframework. Instead,greenbonds are customarily structured to align with the Green Bond Principles published by the International Capital Markets Association (ICMA).5 The principles provide voluntary bestpractice guidelines for the selection, management, evaluation, and disclosure of green projectswhenissuingagreenbond. Theprinciplesrecommend(butdonotrequire)thirdparty certification for green bonds. As alignment with the principles is voluntary, the proceedsofsomegreenbondsmaynotbeinvestedinmeaningfulgreenprojects,aprocess thatreflectsbroadertrendsofgreenwashing,ormisusingthegreenlabel,inthesustainable financeindustry.6 2Conventional bonds can still be used to finance green projects. Green bonds are distinguished by the specificationoftheuseofproceedsinthebondprospectus. 3Whilethispaperfocusesonthereductionofgreenhousegasemissionstomitigateclimatechange,green bonds may pledge investment to achieve a variety of aims not only limited to reducing emissions, such as increasingwaterefficiencyorsupportingbiodiversity. 4SeeLevine(2019). 5SeeInternationalCapitalMarketsAssociation(2021). 6For further discussion of greenwashing in green bonds, see Wirz (2022). For greenwashing in ESG investmentfunds,seeFletcherandOliver(2022). Forgeneralgreenwashingconcerns,seeMundy(2022),The Economist(2021),andFancy(2021). 4

Theoverallgreenbondmarkethasgrownrapidlyinrecentyears, withissuancereaching $556billionin2021. Thisisasubstantialincreasefromonly$31billionin2014,reflectinga compoundannualgrowthrateofabout50%fortheentiremarket. Greenbondsarepoised forcontinuedrapidgrowth,withsomemarketparticipantsforecastingsignificantgrowth in green bond issuance in 2022.7 While green bonds have been issued in many countries, Europeanissuershavebeenthemostactive,followedbyUSissuers,Chineseissuers,and supranationalissuers. Aboutthree-quartersofannualgreenbondissuanceisdenominated intheeuroortheUSdollar. 2.1 Greencorporatebonds Greenbondswerefirstissuedinthelate2000sbysupranationalorganizationssuchasthe European Investment Bank and the World Bank. Supranationals (and governments) still issue green bonds, but corporations now account for about two-thirds of global issuance. Private-sector issuers of green bonds tend to be large, mature firms or firms with strong access to debt capital markets. This is especially true of firms issuing green bonds in a foreign currency to tap international green capital markets. Issuers also tend to be from relativelylesscarbon-intensiveindustriesandgreenerthantheirindustrypeers.8 About half of green corporate issuance is concentrated in financial firms. The majority of this financial firm issuance is from banks, which use green bond proceeds to extend loanstofirmsthatneedfinancingforgreenprojects. Alternatively,banksmayfirstextend green loans, and then securitize the loans into a green bond.9 The remaining portion of financial firm issuance is mostly from real estate financing vehicles such as real estate investmenttrusts(REITs), whichtypicallyfinancethedevelopmentofgreenbuildings. The electricutilitysectoraccountsforanadditionalquarterofgreencorporatebondissuance. Theremainingquarterofgreencorporatebondissuanceisdistributedamongavarietyof sectorswithgrowingissuance,suchasalternateenergy,automobiles,andheavyindustry. Notably,issuancefromfossilfuelcompaniesisnegligible. During the recent period of remarkable growth in green bond issuance, policymakers, market participants, and financial journalists have acknowledged the potential for green bonds to fund green investments. However, while certain individual green bonds may have a discernible positive impact, the broader green bond market has several shortcomings. Keyissuesincludegreenwashing,thelackofauniversalgovernanceandcertification framework,andthesignificantcompliancecostsassociatedwithcertification,issuanceand reporting.10 Critics further question if the green bond market actually influences firms’ 7See,forexample,KuchtyakandBruce(2022). 8SeeFlammer(2021). 9Ineithercase,thisfinancialengineeringcanleadtothebankingsectoroverstatingitsgreeninvestments relativetoactualcapitalexpenditure. Thedouble-countingoccurswhenthebondholderandthelenderboth claimtobefundinggreeninvestments(Ritchie&Rocha,2021). Wefurtherdiscussthisrefinancingissueina latersection. 10Forfurtherdiscussionontheshortcomingsofgreenbonds,seeRitchie,Ward,Kishan,andGledhill(2021) andStubbingtonandNauman(2020).Fordiscussionongreenwashinginthesustainablefinanceindustry,see 5

investments, or if green bonds instead fund green projects that would have otherwise receivedfundingthroughconventionalchannels. Theunansweredquestionis: Isthegreen bondmarketasawholeincentivizingmeaningfulgreeninvestment? 2.2 Motivationforissuingandinvestingingreenbonds 2.2.1 Thegreenium Firms that issue a green bond may receive several direct and indirect benefits that potentially incentivize green investment. Issuing a green bond may directly lower the interest rate paid on the bond relative to conventional bonds. If a firm chooses to issue a green bond,itmayattractnewinvestorsinterestedinsustainableinvestment,therebyincreasing demandforthebond. Shouldtheaddeddemandpushthegreenbond’syieldlowerthan that of a comparable or hypothetical conventional bond, this yield difference is called the greenpremium,or”greenium”.11 Thatsaid,evenifagreenbondissueoffersagreenium, itmaystillbecostliertoissueagreenbondcomparedtoaconventionalbond. Theprocess ofcertifying,issuing,monitoring,andreportingoverthelifetimeofthebondishigh,especially for complex green projects and small or first-time issuers. For some issuers, it may takeseveralgreenbondissuesoraverylargegreeniumfortheborrowingcostadvantage ofthegreeniumtobreakevenagainstthesignificantcomplianceandissuancecosts. From the investor’s perspective, there are several explanation for why one would accept a lower yield on a green bond relative to a comparable conventional bond. The most common explanation for the greenium is that investors are willing to sacrifice immediate financial returns in exchange for an environmental benefit.12 According to this framework, because a green bond is effectively comprised of a conventional bond and a green promise, investors assign a positive value to the green promise and are willing to pay a higher price for a bond at issuance, which means they accept a lower yield. Under this working model, the marginal green bond investor is understood to be concessionary (as opposedtofull-return),meaningtheyarewillingtoacceptlowerinvestmentreturnstofinancehigh-qualitygreenprojects.13 Byoptimizingthetrade-offbetweenreturnsandenvironmentalimpact,theseinvestorsaresupposedtoexertademandpressurethatproduces Quinson(2021a),Quinson(2021b),TheEconomist(2021),Fancy(2021),andTemple-WestandPalma(2022). ForacriticalassessmentofthechangesincarbonemissionsfromgreenbondissuersseeEhlers,Mojon,and Packer(2020). 11Alternatively,theissuermayrespondtotheincreaseddemandbyissuingmoredebtatathesameinterest rate, rather than lowering the interest rate. In practice, this is constrained by the limited number of green projectsthatareavailabletotheissuer. 12SeeSectionA1intheAppendixfordiscussionofthetwomainalternativeexplanationsforthegreenium: higherrisk-adjustedreturnsandthe”greencrisispremium”. 13ArecentdraftmethodologyreleasedbythePartnershipforCarbonAccountingproposesthatgreenbond investorsshouldbeabletoclaimareductiontotheirScope3Category15(investment-related)greenhouse gas emissions in accordance with their investment in a given green bond. This accounting methodology couldpotentiallyincreasetheincentivetopayahigherpriceforagivengreenbond(concedelowerexpected returns)byeffectivelyallowinggreenbondinvestorstopurchaseanemissionsreductionbybuyingagreen bond(PartnershipforCarbonAccountingFinancials,2021). Somemarketparticipantsarereportedlyalready usingthismethodology(Edwardsetal.,2022). 6

a ”smart greenium”, rewarding credible, high-impact green bonds with an incrementally lowerinterestrate. Our research is not the first to focus on the greenium at issuance. Previous studies have deliveredmixedempiricalresults.14 Severalpapers,usingdifferentempiricalmethodologies and green bond samples, have quantified greeniums in the primary market that are rangingfromastrict0toupto19basispoints. See,forexample,EhlersandPacker(2017), Gianfrate and Peri (2019), Partridge and Medda (2020), Larcker and Watts (2020), Baker, Bergstresser, Serafeim, and Wurgler (2022), and Kapraun, Latino, Scheins, and Schlag (2021). However, it can be difficult to parse and compare the results across the papers because they differ widely in their sample period, green bond market segment, focus on primary and/or secondary markets, and empirical methodologies. In addition, some papersimplementamatchingapproachthatconfinessamplestoasmallsetofissuerswitha largenumberofbondsoutstandinginordertoestablishmeaningfulmatches,whileother papersusefixed-effectregressionapproachesthatrequirelarge,globalpanelsandahostof controlvariablestomeaningfullyaccountforissuer-andbond-leveldifferences. Therefore, theexistenceandrobustnessofapotentialcorporategreeniumatissuanceisinconclusive. In this paper, we aim to be careful in selecting the most appropriate empirical approach, given the global corporate green bond sample at our hands, in an attempt to gauge the greeniumforaspecificsetofgreenbonds. 2.2.2 Thegreenhalo Besides the direct benefit of the greenium, green bonds may provide additional indirect benefits for the issuer. For instance, by highlighting the environmental credentials of an issuer,greenbondsofferamarketingbenefit,potentiallyloweringthefirm’scostofcapital by attracting new investors, or potentially improving business performance by attracting newcustomers. Thishypothesizedindirecteffectiscalledthegreenhalo. Empirical research on the green halo has only emerged recently. The existing literature highlights short-term increases in issuers’ stock prices or decreases in issuers’ secondary market bond yields upon announcement of the first green bond issue. See, for example, Flammer (2021), Tang and Zhang (2020), NatWest Markets (2019), Baulkaran (2019), and Forfot and Fosse (2021). These announcement returns are captured by stockholders and bondholders, but they do not necessarily impact the firm’s actual cost of capital beyond the short-term window. The green halo is not a focus of our paper, but we do look to understand whether the benefits of issuing a green bond are distinct from any potential long-termimprovementstoafirm’scostofcapital. 14SeeLau,Sze,Wan,andWong(2022)forasummaryoftheempiricalevidenceonthegreenium,covering bothprimaryandsecondarymarketresults. 7

The argument for issuing green bonds to capture a green halo can be situated within the theoretical context of signaling problems.15 Firms possess asymmetric information about their environmental credentials, such as future plans to reduce emissions. If this information is not or cannot be communicated effectively to investors with a preference for sustainability, firms may suffer from suboptimal costs of capital. Issuing a green bond may serve as (potentially costly) solution to address the signaling problem and achieve a moreoptimalcapitalcost. 3 Data and empirical approach For the period from 2014 to 2021, we compile a global panel of 129,043 conventional corporate bonds and 1,169 green corporate bonds. These bonds have been issued by 12,736 corporations. Our primary source of data to construct this sample is bond-level data from Bloomberg Back Office. For the bonds in our sample, Bloomberg Back Office data provides detailed information on the characteristics of each bond, such as the issue price, issue and maturitydate,thehistoryoftheparamountoutstanding,creditratings,thebond’scurrency,as well as a variable to identify green bonds Bloomberg Finance LP (2021). The Bloomberg green bond identifier strictly requires referenced bonds to be aligned with the first principle of the Green Bond Principles, which states that the green bond’s use of proceeds forenvironmentalprojectsshouldbe”appropriatelydescribedinthelegaldocumentation of the security”. We supplement the Bloomberg data with additional data from Refinitiv Workspace(Refinitiv,2021). To construct our sample, we begin by selecting conventional and green corporate bonds andmedium-termnotes(MTNs)frombothgreenandconventionalissuers. Wefurtherrequireabondtobeafixed-orzero-couponbondwithanotionalamountofatleast$500,000 issuance. We only consider bonds issued by private-sector corporations but allow stateownedenterprisesinoursample.16 Wedonotconsiderbondsdirectlyissuedbysupranationalentitiesormunicipalities. We require green bonds to be issued in either the euro or US dollar because these green bond markets are the deepest and most liquid. We drop green bonds if their ultimate parent does not have at least one conventional bond in our sample.17 For conventional 15SeeDaubanes,Mitali,andRochet(2021),whichdevelopsatheoreticalmodelofsignalingbygreenbond issuers. See Maltais and Nykvist (2020) for further discussion of potential indirect, non-financial benefits toissuinggreenbonds, suchasalleviatinginstitutionalpressures, seekingsociallegitimacy(the”licenseto operate”),andattractingorretainingvaluedemployees. 16Manylargeemergingmarketissuerswithconsistentaccesstointernationalbondmarketshavesomeform ofgovernmentbacking;thisincludesstate-ownedenterprises(SOEs).WeretainSOEsbecauseexcludingthem wouldreduceourglobalsamplebyroughly20%.Thisdoesnotqualitativelyaffectourresults.Moreover,we add regression controls on the issuer- and issuer-times-country level to account for potential differences in SOEsbonds. 17We obtain data on issuers’ ownership structure from Bloomberg Capital Structure (CAST) (Bloomberg FinanceLP,2021). 8

bonds, we include only bonds denominated in currencies where issuers of the selected greenbondshaveissuedatleastoneadditionalbondandwherethereisatleast$10billion intotalissuanceinthecurrencyoverthesampleperiod. Ourgoalistocompareeuroand US dollar green bonds issued by firms outside the US and euro area to the conventional bonds in the issuers’ domestic markets, provided the market has sufficient issuance volumeforarobust,meaningfulcomparison. Theendresultisthatweincludeconventional bonds from 23 currencies. A breakdown of our sample by currency is provided in the AppendixinTableA1. After constructing this initial sample of bonds, we calculate the exact yield to maturity of each bond at issuance using each bond’s issue price, coupon rate, coupon frequency, and day-count convention. We focus on the primary market yield (the yield at issuance), insteadofthesecondarymarketyield,becauseitdeterminestheactualinterestratepaidby the issuer to borrow funds.18 Additionally, because the global corporate bond market is relatively illiquid and has substantial transaction costs, many investors have a preference for purchasing bonds in the primary market (Flanagan, Kedia, & Zhou, 2021). When investorsareconsideringwheretoallocatetheirbondportfolio,theyareoftenmorelikelyto selectoneofseveralcomparablebondofferingsintheprimarymarketratherthanchoose between a given issuer’s new primary market issue and their outstanding bonds trading in the secondary market. This means it is often more appropriate to compare an issuer’s primary market yields to the primary market yields of other issuers in a given period, ratherthancompareanissuer’sprimarymarketyieldstotheirsecondarymarketyields. Wethencalculateeachbond’syieldspreadatissuancebytakingthedifferencebetweenthe yield to maturity and the linearly interpolated maturity-matched government bond yield curveforthegivenbond’scurrencyonitsdateofissuance. Lastly,wetakeseveralstepsto eliminateoutliersanddataerrors. Wedropbondswithyieldspreadsgreaterthantenpercent, as these bonds are generally considered distressed. We also drop bonds with prices at issuance of less than 90 and greater than 250, as these are likely distressed, data errors, or very large outliers. By focusing on yield spreads, we are able to control for the differencesininterestrateenvironmentsacrosscurrencies. Bycomparingyieldspreadsbetween bonds denominated in different currencies without accounting for currency hedging, we are assuming that exchange rates follow a random walk. Summary statistics for yield spread,yieldtomaturity,andamountissuedacrossthebondsinoursampleareprovided intheAppendixinTableA2. 3.1 Variableconstruction For bonds included in our sample, we construct composite credit ratings by taking the meanofthebond’screditrating, whereavailable, fromMoody’s, Standard&Poor’s, and 18Secondarymarketyieldsmaynotaccuratelyreflectborrowingcostsintheprimarymarketforcorporate issuerswithoutabroad,liquidsetofoutstandingbondsthatcanbeusedtoconstructayieldcurve.Liquidity conditionsmayalsovarysignificantlyacrossinternationalmarketsandinterestrateenvironments,whichcan furtherdistortsecondarymarketyields. 9

Fitch, and then rounding down. We aggregate the composite ratings into rating buckets. WeprovideabreakdownofthebondsamplebyratingbucketinTableA4. Sector categories are based on the Global Industry Classical Standard (GICS) codes and used to analyze green bonds by industry. We tailor our sector categories to the major segmentsofthegreenbondmarket. Ourgreenbondsectorcategoriesare: alternateenergy; banks; electric utilities and fossil fuels; industry and materials; non-bank financials; real estate,transportation,andother.19 Weprovideabreakdownofthegreenbondsampleby industryinTableA5. Weconstructarefinancingvariable,whichisequaltooneifsomeportionofthebondwas used to refinance an existing liability. Otherwise, this variable has a value of zero. We are able to construct this variable for 706 green bonds and 40,794 conventional bonds in our sample. Within this subset, 200 (28%) of the green bonds are used for refinancing, comparedto4,028(10%)ofconventionalbonds. 3.1.1 Greenbondvariables Where data is available, we calculate each green bond’s oversubscription from a textual variablethatcontainsadditionalinformationonthebondissue. Abondisoversubscribed if the investment bank underwriting a bond issuance receives excess orders relative to the actual amount of debt being issued. High oversubscription rates can indicate strong demand for a bond issuance, relative to supply. Relative to the typical demand seen for conventional bonds, green bonds have been well-received by investors, with some issues being several times oversubscribed, a fact that is often noted in the financial press. For a given bond, we calculate oversubscription as the ratio of the notional amount of orders to the actual amount of debt that is issued. Data on investment bank order books for the corporatebondmarketisverylimited,butweareabletoconstructthisvariablefor474of our1,169greenbonds. Wealsoconstructavariablereflectingeachbond’sadherencetotheGreenBondPrinciples. Ataminimum,greenbondsmustadheretothefirstcomponentoftheGreenBondPrinciples,UseofProceeds,tobeidentifiedasagreenbondbyBloomberg. Theremainingthree componentsoftheGreenBondPrinciplesareoptionaltobeconsideredagreenbondand reflectthequalityofthegreenbond’sgovernance.20 Foreachgreenbond,ourGreenBond PrinciplesvariableisequaltozeroifagreenbondisnotfullyalignedwiththeGreenBond Principles and one if the bond is fully aligned. We construct a variable reflecting if the 19Transportationfirmsincludefirmsinthetransportationandautomanufacturingindustrygroups. Firms intheoilandgassectorsissueveryfewgreenbonds,sotheyareincludedwithelectricutilities. 20The remaining three components of the Green Bond Principles are: (1) Process for Evaluation and Selection, whichindicatesiftheissuerhasclearlycommunicatedtheeligibility, objectives, andpotentialrisks ofthebond’sgreenprojects; (2)ManagementofProceeds, whichindicatesifthegreenbond’sproceedsare transferredtoasub-accountorsub-portfolioorotherwisemanagedinan”appropriatemanner”;and(3)Reporting,whichindicatesiftheissuerproducesanannualreportuntilthebond’sproceedsarefullyallocated. SeeInternationalCapitalMarketsAssociation(2021). 10

green bond was subject to a pre-issuance external review by a third party. While external reviewisnotacomponentoftheGreenBondPrinciples,itisconsidered”recommended”. Finally, we construct a triple index inclusion indicator variable that is equal to one if a green bond was a constituent of three major green bond indices in our sample: the ICE GreenBondIndex(542bondsinthesample),theSolactiveGreenBondIndex(505)andthe JP Morgan JESG Green Bond Index (436).21 The triple index inclusion variable is equal to onefor365bondsinoursample. 3.2 Empiricalapproach TheregressionsinthispaperareinspiritofBakeretal.(2022)inthatweestimateafixedeffects regression across an unbalanced panel of corporate bonds, with a given bond’s yield spread at issuance as the dependent variable. In doing so, we allow our regression specification to account for potential nonlinearities as well as issuer- and bond-specific timevariation. We chose a fixed effects regression approach over a matching approach for two reasons. First, the matching approach generally requires a triplet of bonds from the same issuer: one green bond and two comparable conventional bonds that are used to interpolate a localized conventional yield curve. This requirement drastically reduces the number of green bonds that can be investigated and biases the sample to issuers with strong capital marketaccessandtheabilitytofrequentlyissuecomparablebonds. Issuersthatcannotfrequentlyissuecomparablebonds,suchassmall-andmedium-sizedenterprises(SMEs)and issuersinemergingmarkets,willbeunderrepresented. Second,thematchingapproachreducesthesampletoonlygreenbondissuers. Thisbiasesthecontrolgroupofconventional bonds to those issued by green bond issuers, which is problematic because green bond issuers have been shown to be different from grey issuers (Flammer, 2021). In contrast, theregressionapproachallowsustoestimatethegreeniumbyregressingyieldspreadson bond, issuer, and macro characteristics of bonds issued by both green and conventionalonlybondissuers. In our regression approach, we compare the borrowing costs of green and conventional bondsusinganindicatorvariablethatflagsgreenbonds,whileholdingotherfactorsconstant. Ourempiricalbaselinemodelisasfollows: Yieldspread = α Green +β Controls T +µT +ϵ (1) i,f i i,r,t i,r,m,f i,f for bond i of ultimate parent company f issued in currency region r on date t in yearmonth m. In Equation (1), the key variable of interest is the indicator variable Green , i which takes the value of one if bond i is a green bond. The coefficient on this indicator 21TheJPMorganindexwaslaunchedinNovember2020. BeforeNovember2020,werelyonJPMorgan’s simulatedindexconstituents,whichreflectswhattheindexconstituentswouldhavebeeniftheindexmethodologywasretroactivelyappliedtohistoricaldata. 11

variable, α, captures the average difference in primary market yield spreads for green and conventional bonds, holding other factors constant. A negative α indicates a lower yieldspreadforgreenbonds,andthereforeaborrowingcostadvantageoverconventional T bonds (or a positive greenium). The vector Controls contains bond-level numeric and i,t,r macro-level controls, observed for bond i in currency region r on issue date t, and their interactionsi×randi×t. ThevectorµT containsbond-level,firm-level,andtime-level i,m,f fixed effects for bond i, ultimate parent f, and year-month m, as well as an interaction of year-monthandcurrencyregion,m×r. Standarderrorsareclusteredontheissuerultimate parentandyear-monthlevelsandarethusrobusttobothcross-sectionaldependenceand serialcorrelation. Specifically, our vector of controls includes the following variables: We have two bondlevel numeric variables, the log years to maturity and log notional amount issued. We further control for multiple variables that capture the general condition of credit markets in different currencies: The level, slope, and curvature of the sovereign yield curve for a given bond’s currency region, calculated between 1 and 10 years to maturity using a principal component analysis; and the 30-day realized volatility of the 10-year sovereign bond for a given bond’s currency region. To add a dimension for aggregate credit risk, wecontrolfortheoption-adjustedspread(OAS)oftheICEBofAglobalinvestmentgrade corporatebondindex(G0BC)andthedifferenceintheOASbetweentheICEBofAglobal high-yieldcorporatebond(HW00)andtheinvestmentgradecorporatebondindexes. In addition to the numeric controls, we include several fixed effects to capture potential nonlinearities: the issuer ultimate parent, the rating bucket, seniority, whether a bond is callable, whether a bond is putable, whether a bond is sinkable, the year-month of issuance, and the currency of issuance interacted with year-month, as well as country of issuanceinteractedwithanindicatorvariablemarkingstate-ownedenterprises. Lastly, to control for potential bond-specific time variation in the maturity risk, the credit risk,andthepotentialcall-optionfeature,weallowforinteractionsofthenumericcontrols and the fixed effect. That is, we include interactions between the slope of the yield curve andlogmaturity;theratingbucketandthehighyieldminusinvestmentgradeindex;and whether a bond is callable and the level, slope, curvature, and realized volatility of the sovereignyieldcurve.22 Initscurrentversion, thepotentialshortcomingsofourempiricalanalysismayberooted in our lack of firm-level metrics for ESG outcomes and balance sheet data. Nevertheless, our baseline results remain robust to the inclusion of time-varying issuer fixed effects, such as issuer ultimate parent interacted with year-quarter. Balancing the granularity of the time-varying issuer fixed effects and the associate reduction in sample size due to the elimination of singleton observations (Correia, 2015), we ultimately adopt a regression specification that interacts issuer and annual fixed effects. There is possibly some addi- 22See Gilchrist and Zakrajsˇek (2012) for the importance of controlling for interactions between the yield curveandwhetherabondiscallable 12

tional bias from our lack of data on expected liquidity. We further discuss this potential shortcomingintheAppendixinSectionA2. 4 Empirical results In this section, we present our empirical results. We start with the baseline estimate of the greenium for our sample. We then continue to account for the time-variation in the greenium over our sample period, relating it to the rapid growth of the sustainable asset management industry. Within the context of a growing demand from the asset management industry, we connect the greenium to two proxies for demand pressure, namely oversubscriptionandbondindexinclusion,anddocumentarelationparticularlyforeurodenominated green bonds. The latter motivates us to differentiate the greenium by currency and whether or not corporations are local or foreign currency issuers. We conclude by relating the greenium to the ”greenness” of the bonds as well as to specific bond- and firm-levelcharacteristics. 4.1 Baselineresults: 8bpgreenium,notfromgreenhalo WepresentourbaselineresultsonthegreeniumandthegreenhaloinTable1. InColumns (1) to (3), we find negative and highly significant coefficients on the green bond indicator variable from Equation (1). The negative coefficient indicates that, on average, there is a greeniumholdingotherfactorsconstant. InColumn(1),weestimateEquation(1)without time-varyingissuerfixedeffects,findingagreeniumofabout11basispoints. InColumns (2) and (3), we control for time-varying issuer fixed effects at the annual and quarterly frequency,respectively,findingagreeniumofabout8and9basispoints,respectively. We select the regression specification of Column (2), which uses annual time-varying issuer fixed effects, as our baseline estimate of the greenium. This is motivated by the followingconsideration: Itisimportantthatourregressionsaccountforissuer-specifictimevariation,butthereisatrade-offbetweenthegranularityofthefixedeffectsandthesample sizeduetooureliminationofsingletonobservations.23 AsshowninColumn(3), increasingthegranularityofthetime-varyingissuerfixedeffectstothequarterlyfrequencydoes not meaningfully impact the regression estimate, but it causes a sharp drop in our sample size by 12,755 bonds. This reduction in the sample most impacts the bonds issued by smaller issuers, which are more likely to issue a single bond in a given quarter, which wouldbecountedasasingleton. Taking Column (2) as our baseline regression, the green bonds in our sample receive an averageyieldspreadatissuancethatisabout8basispointslowerrelativetoconventional bonds. Thisindicatesthat,onaverage,corporatebondissuerspayalowerinterestrateon their green bonds, holding other factors constant. Relative to the average yield spread of our sample, this reflects a roughly 5% decrease in the borrowing cost to the issuer. This 23SeeCorreia(2015)fordiscussionofsingletonobservations. 13

borrowing cost advantage is economically meaningful, but at the same time it is reduced by green bonds’ compliance costs (certification, monitoring, reporting) and a longer issuance process, particularly for complex green projects and small or first-time issuers.24 However, anecdotal evidence suggests the costs of issuance may decrease as firms issue multiple green bonds, which is when an average 5% borrowing cost advantage may start toaccumulate. [INSERTTABLE1] In Columns (4) and (5), we test whether green bond issuers receive a borrowing cost advantageacrossalloftheirbonds, bothgreenandconventional. AsdescribedinSection2, thispotentialbenefittoafirm’soverallcostofcapitaliscalledthegreenhalo. Weconstruct anindicatorvariablethatisequaltooneforagivenbondifitsissuerhaspreviouslyissued a green bond (or if the given bond is the issuer’s first green bond). The coefficient on the greenissuervariableisneithereconomicallymeaningfulnorstatisticallysignificant. That is,relativetoconventionalissuers,greenissuersdonotcaptureanoverallborrowingcost advantageacrossalloftheirbonds. Next,inColumns(6)and(7),wetestiftheborrowing costadvantageofthegreeniumisdistinctfromanypotentialgreenhalobenefitsforgreen bond issuers. We include both the green bond and green bond issuer indicator variables. While the greenium coefficient remains significant at about 8 basis points, the coefficient onthegreenissuerindicatorvariableisnotstatisticallysignificant. In combination the results in Columns (4) to (7) suggest that the greenium is a separate phenomenon from any potential green halo benefit to the bonds of green bond issuers. Admittedly, this test does not constitute an exhaustive invalidation of the green halo, as there are other possible channels by which issuing green bonds could lower a firm’s cost ofcapital(forinstance,throughhigherstockpricesasdocumentedbyFlammer(2021)and TangandZhang(2020). 4.2 Accountingfortime-variation: Greeniumemergesin2019 In Figure 2, we use a similar regression specification to Equation (1) to construct a time series of the greenium at issuance at an annual frequency. We interact the green bond indicator variable with a yearly time fixed effect; the regression estimates are detailed in TableA6intheAppendix. InFigure2,blackdotsreflectthegreeniumcoefficientforeach year, and the gray ribbon reflects a 95% confidence interval. We find that a statistically significantaveragegreeniuminoursampleonlyemergesin2019atabout14basispoints, andthentightenstoabout9and8basispointsin2020and2021,respectively. The emergence of an average greenium in 2019 follows a discrete jump in the maturation ofthegreenbondmarketovertheperiodfrom2018to2019. Duringthistime,greenbonds 24Inunreported resultsthatgo beyondthescope ofthis analysis, wefind that, on average, underwriters chargelowerfeesforunderwritinggreenbondsrelativetoconventionalbonds.Thismightreflectunderwriters’incentivestoestablishafootholdinthenascentmarketforissuinggreenbonds. 14

Figure2: Timeseriesofcorporatebondgreeniumatissuance Basis points 20 Green higher yield 10 spread at issuance 0 −10 −20 −30 Green lower yield spread at issuance −40 2014 2015 2016 2017 2018 2019 2020 2021 Regression estimate of the average annual yield spread of green corporate bonds versus conventional corporate bonds, holding other factors constant. Negative values indicate green corporate bonds have a lower yield spread than conventional corporate bonds, and therefore a borrowing costadvantage. Shadedribbonindicatesa95%confidenceinterval. Source: Regressionestimation fromTableA6. gainedacceptancefrominvestorsandregulatorsasamainstreaminvestmentvehicle,particularlyinEurope. Forregulators,thisperiodmarkedasubstantialdevelopmentwiththe release of the European Union Sustainable Finance Action Plan (EU SFAP) by the EuropeanCommissioninMarch2018. Theplanreflectedabroadcommitmenttoreorientcapitalflowstowardsamoresustainableeconomy,withspecificcommitmentstostandardize and regulate green securities through regulations such as the EU Green Bond Standard. The plan included a regulation to mandate and standardize the sustainability disclosures of financial market participants (or SFDR) in an effort to increase transparency. Following the release of the action plan by the European Commission, the SFDR was eventually adoptedbytheEuropeanParliamentonApril18,2019andsignedintolawonNovember 27,2019. This period of regulatory crystallization coincided with the uptake of green bonds by investorsandassetmanagers. Thisuptakeisreflectedinthegrowthinassetsundermanagement (AUM) by green bond funds beginning in 2018, as charted in Figure 3a using data from Morningstar (Morningstar, Inc., 2021). The growth is driven almost exclusively by green bond funds domiciled in Europe, which account for more than 90% of green bond fundAUM. It appears plausible that the increased appetite by investors, combined with a push for disclosuresandstandardization,contributedtotheemergenceoftheaveragegreeniumin 2019. We test this conjecture in Table 2a by creating an indicator variable SFDR , which t 15

Figure3: Greenbondfundassetsundermanagement(AUM) (a)GreenbondfundAUM (b)GreenshareoftotalfixedincomeAUM ($) Billion Share (%) 35 1.0 Fund domicile region Fund domicile region 30 Emerging markets European Union 0.8 Other developed markets Other developed markets 25 United States United States 20 European Union 0.6 Emerging markets 15 0.4 10 0.2 5 0 0.0 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 Panel (a) is the assets under management (AUM) of green bond funds by fund domicile region. Panel(b)istheshareofgreenbondfundAUMrelativetoeachdomicileregion’stotalfixedincome fund AUM. The data extends from 2014-Q1 through 2021-Q4 at a quarterly frequency. Source: Morningstar,Inc.(2021). is equal to zero before and one on and after April 18, 2019, the date the EU Parliament adopted the SFDR. In Column (1), we estimate the greenium in the pre-SFDR and post- SFDR periods. In line with our results from Figure 2, we find a significantly negative greenium of about 13 basis points in the period following the EU regulation. In Column (2), we show that this post-SFDR greenium is little changed by the inclusion of annual time-varyingissuerfixedeffects,remainingsignificantlynegativeatabout10basispoints. [INSERTTABLE2a] In Columns (3) and (4), we adopt a difference-in-differences specification, including the green bond indicator variable in combination with the post-SFDR interaction term. The coefficientontheinteractiontermshouldcapturetheresponseinthegreeniumfollowing the regulatory change.25 In Column (3), without controlling for time-varying issuer fixed effects, we still find a significantly negative coefficient on the interaction term. However, inColumn(4),controllingforannualtime-varyingissuerfixedeffects,theinteractionterm is no longer significant, suggesting that issuer-specific time-variation rather than just the regulatorychangecontributedtothesignificantgreeniumasof2019inFigure2. InColumns(5)and(6),wedifferentiatethegreeniumintothepre-andpost-SFDRperiods withfurthersegmentationbycurrency. ConsistentwithColumns(1)through(4),wefind significant negative greenium coefficients for euro and US dollar green bonds in Column (5), but these coefficients become insignificant in Column (6) with the addition of time- 25Thebaselineeffectforthepost-SFDRindicatorispartialledoutduetoouryear-monthfixedeffectsinour regressions. 16

varying issuer fixed effects. Segmentation by region, as captured in Table 2b Panel (b) Column(1),suggeststhatonlyissuersfromtheEUandtheUnitedStatescapturedsignificant greeniums following the introduction of the SFDR. Other developed market issuers as well as emerging market issuers appear not to capture a significant average greenium asof2019. Withtime-varyingissuerfixedeffects,Column(2),thiseffectonlyholdsweakly forUSissuers. [INSERTTABLE2b] FromthismixedempiricalevidencewecannotconcludethattheSFDRregulationcaused theemergenceofasignificantgreeniumasof2019. ThisisnottosaythattheEUSFAPand theSFDRhavebeensupportivetotheoveralldevelopmentofthegreenbondmarketand inparticularitsstrengthinEurope. 4.3 Oversubscriptionandindexinclusion Inthetwosubsectionsabove,wedocumentedasignificantaveragegreeniumanddifferentiatedthisgreeniumacrosstime. Weshowedthattheemergenceofasignificantgreenium coincides with the growth of the sustainable asset management industry. In this subsection,wefocusontheimpactofsupplyanddemandimbalancesforgreenbondsandhow excessdemandpressurerelativetoagreenbond’sfundamentalscoulddrivethegreenium. We examine two proxies for excess demand that could contribute to the allocation of the greenium: Oversubscriptionandbondindexinclusion. We link the greenium to a green bond’s excess demand at issuance using bond oversubscription rates in Table 3. Anecdotally, market participants and financial journalists have reported that green corporate bonds can be heavily oversubscribed, indicating excess demand for a bond in the primary market.26 Holding other factors constant, excessive demand from investors relative to the notional size of a green bond may allow an underwriter to lower the bond’s yield on behalf of the issuer. Our hypothesis is therefore that oversubscriptionratesarenegativelycorrelatedwithyieldspreads. We argue that oversubscription rates are a good proxy for excess demand in the case of green bonds. The notional size of a green bond is constrained by the opportunities for greeninvestmentavailabletoanissuer. Ifthereisverystrongdemandforagivenissuer’s green bond, it is difficult to increase the amount of debt issued without degrading the greennessofthebond. Itismorefeasiblefortheissuertocapturethehighdemandwitha lowerbondyieldratheralargersupplyofdebt. In Table 3 Column (1), we estimate the effect of green bonds’ log oversubscription rates on yield spreads. We find a statistically significant negative coefficient, suggesting that highergreenbondoversubscriptionratesareassociatedwithaborrowingcostadvantage. 26Fordiscussionofgreenbondoversubscriptionintheprimarymarket,seeClimateBondsInitiative(2021). 17

In our estimation, a 1% increase in oversubscription is associated with a 0.058 basis point increaseinthegreenium. Fortheaveragesamplelogoversubscriptionof1.48,thiswould indicate an average greenium of about 8 basis points, consistent with our baseline results in Table 1 Column (2). Admittedly, one could argue that underwriters of issuers with oversubscribed bonds are more likely to report oversubscription data. In Column (2), in an attempt to tackle such a selection issue, we assume that if underwriters did not reportanoversubscriptionrate,thegreenbondwasneitherover-norundersubscribed,and assign an oversubscription ratio of one. These imputed oversubscription rates serve as a lower bound, as markedly undersubscribed bond offerings are typically not brought to the market. The coefficient retains its sign and significance and increases by a little more than a basis point, suggesting that green bonds without reported oversubscription data and an imputed oversubscription rate of one still capture an average greenium. That is, Column (1) is not necessarily biased toward bonds that capture a greenium. In Column (3), we continue our analysis by controlling for the green bond indicator. Despite some collinearity between the two variables, we find that the oversubscription effect remains weakly significant and negative, while the coefficient on the green bond indicator turns insignificant. [INSERTTABLE3] InColumn(4),weexploreapotentialnonlinearrelationshipbetweenoversubscriptionand the greenium, breaking the green bond sample into terciles we form on oversubscription rates. We find that only highly oversubscribed green bonds receive an statistically significant average greenium at about 17 basis points. In Column (5), we confirm this effect even when controlling for the green bond indicator variable, finding statistically significant differences in the coefficients for low, medium, and high oversubscription rates. In Columns (6) and (7), we refine our result by segmenting the green bonds by currency. While controlling for the green bond indicator variable, we find that oversubscription in euro-denominated green bonds leads to a statistically significant greenium; for dollar greenbonds,theeffectisinsignificant. Another potential proxy for excess demand relative to a green bond’s fundamentals is whetherthegreenbondisincludedinagreenbondindex. Indexinclusioniseitherknown in advance or can be anticipated based on inclusion criteria and the inclusion or prior issues. Previousliteratureshowsthatdemandfrommutualfundsandexchange-tradefunds (ETFs) benchmarked to a given bond index can impact the pricing of bonds included in the index (Dannhauser, 2017; Holden & Nam, 2019; Ye, 2019). This phenomenon of additional demand for index constituents has been shown to encourage bond issuers to alter the characteristics of their bonds in order to meet index inclusion criteria (Dathan & Davydenko,2020). Indexinclusionmayplausiblydrivethegreeniumbyforcingdemand 18

for constituent green bonds from green bond funds that are benchmarked against the indexes.27 Our hypothesis is that green bond index inclusion is negatively related to yield spreads at issuance. In Table 4 Column (1), we confirm that green bonds capture a significant averagegreeniumofabout9basispointswhentheyareincludedinanyofthethreegreen bond indexes that we consider (ICE, Solactive, JP Morgan).28 In Column (2), we control for the green bond indicator to test if the greenium for index constituent green bonds is distinctfromtheadvantageforexcludedgreenbonds. Wefindthatindexinclusionisnot, onaverage,associatedwithastatisticallysignificantgreeniumrelativetoexcludedbonds. However, in Columns (3) and (4), we segment the index inclusion effect by currency. We findthateurogreenbondsincludedinthegreenbondindexesreceivealarge,statistically significantgreeniumofabout12basispoints,whileexcludedeurogreenbondsdonot. On theotherhand,includedUSdollargreenbondsdonotreceiveagreenium,whileexcluded dollar green bonds receive a sizeable, statistically significant greenium. Notably, most of the excluded US dollar green bonds have been issued by foreign dollar issuers, a green bondfeaturewestudyinmoredetailinthenextsubsection. [INSERTTABLE4] Incombination,theresultsinTables3and4suggestthatoversubscriptionandbondindex inclusion are relevant proxies for how the greenium is allocated. Our results point to a morepronouncedgreeniumthroughdemandpressuresineuro-denominatedgreenbonds and US dollar green bonds of foreign issuers. One possible explanation for this effect couldbethemorestringenttransparencyrequirementsthatEuropeanissuersandfinancial service providers are subjected too. Anecdotally, European issuers are more transparent about the governance and impact of their green projects, making it easier for investors to justifyanddisclosetheirgreenbondpurchases. Thismaycreateadditionaldemandforthe respectivegreenbonds. Atthispoint,however,wecannotbackupthisnotionempirically. 4.4 Eurogreeniummorerobust Next, we examine the effect of currency denomination and issuer geography on the greenium in Table 5. In Column (1), we show that euro-denominated green corporate bonds captureanaveragegreeniumofabout6basispoints, whileUSdollar-denominatedgreen bondscaptureanaveragegreeniumof12basispoints. Bothestimatesaresignificantatthe 5%significancelevel. InColumn(2),weshowthattheroughly7basispointdifferencein 27Institutionalinvestorswithbuy-and-holdstrategiesprefertopurchasebondswithaprimarymarketallocationtoavoidthesubstantialtransactioncostsassociatedwithbuyingthesamepositioninthesecondary bondmarket(Flanaganetal.,2021). Asgreenbondsareanecdotallyreportedtobelessliquidthanconventional bonds in the secondary market, this rationale should exacerbate the competition for green primary marketallocationsamongtheESGassetmanagementindustry. 28Wedonotreporttheresultsfromregressionsusingeachgreenbondindexseparately,buttheresultsare similar. 19

thegreeniumforeuroandUSdollargreenbondsisnotstatisticallysignificantfromzero, meaning the bonds capture statistically comparable greeniums when the point estimates’ confidenceintervalsaretakenintoaccount. [INSERTTABLE5] We further segment the green bond sample by currency and whether a bond was issued byadomesticorforeignfirmrelativetothecurrencyregion. InColumns(3),ourfindings suggest that, on average, the greenium for US dollar green bonds is conferred to foreign issuers,notAmericanissuers. ThisisconsistentwithourresultsfromTable4Column(3), asmostofthedollargreenbondsexcludedfromgreenbondindexesareissuedbyforeign firms. InColumn(4),weshowthatthe15basispointdifferencebetweenlocalandforeign issuers in the US dollar green bond market is significant at the 5% significance level. On the other hand, the euro greenium appears to be conferred primarily to local issuers, see Column (3). However, in comparison to the US dollar green bond market, we cannot confirmastatisticallysignificantdifferenceinthegreeniumoflocalandforeigneurogreen bondissuers,seeColumn(4). InColumn(5), weexaminetheimpactofissuergeographyonthegreenium. Wesplitthe green bond market into five segments: the euro area, the United States, other developed markets, China, and other emerging markets.29 We find statistically weak evidence of a 6 basis point average greenium for euro area issuers, but weak to no statistical evidence, respectively,ofagreeniumforotherdevelopedmarketissuersandfromtheUnitedStates. AsshowninColumn(6),wecannotconfirmthatthesedifferencesinthegreeniumacross regionsarestatisticallysignificant. Overall, the results from Table 5 suggest that, on average, euro area firms issuing local currencybondsaswellasforeign-currencydollarborrowersappeartoreceiveasignificant average greenium, while there appears to be no average greenium in the domestic dollar market,andinthenon-euroareamarketmoregenerally. 4.5 Mixedresultsongovernance,externalreview,andcredibility InTable6, we analyzewhetherthe”greenness”ofagreen bondhasameasurableimpact onitsyield. Webeginbyassessingtheimpactofagreenbond’sgovernanceandalignment withvoluntaryinternationalstandards. InColumn(1),weseparatethegreenbondsample into those that are aligned and not aligned with the Green Bond Principles (GBP). In our sample, 82% of green bonds are aligned with the GBP (that is, 839 vs. 190 green bonds). Wefindthatonaverage,GBP-alignedgreenbondscapturearobuststatisticallysignificant greenium of 7 basis points, while non-GBP-aligned green bonds also show a weakly significantcoefficientofabout20basispointsatthe10%significancelevel. InColumn(2),we 29Theissuercountryisthecountryoftheissuer’sultimateparentasof2021. Issuercountrygroupingsare basedonclassificationsfromInternationalMonetaryFund(2021). 20

checkwhetherthereisastatisticallysignificantormeasurabledifferencebetweenthetwo groupsofbonds,whichwecannotconfirm. Thelackofastatisticallysignificantdifference between the two groups suggests that their pricing is not inherently different and green bonds with poor governance frameworks may still receive a comparable borrowing cost advantage. Next,weseparatethegreenbondsintothosethatdoanddonotreceiveanexternalreview (alsoknownasthird-partyverification). Inoursample,78%ofgreenbondshavereceived an external review (that is, 800 vs. 229 green bonds). In Column (3), we find that, on average, green bonds with an external review capture a statistically significant greenium of 8 basis points, while green bonds without an external review capture a statistically insignificant estimate of about 9 basis points. Again, as shown in Column (4), we do not findastatisticallysignificantdifferenceinthegreeniumofgreenbondswithandwithout externalreview. Thissuggeststhatgreenbondswithoutexternalreviewmaystillcapture aborrowingcostadvantage,eventhoughitis,onaverage,notsignificantlydifferentfrom zero. Lastly, in Columns (5) and (6), we separate green bonds by whether the bond has been used to refinance a firm’s existing debt. For green bonds used for refinancing, we estimate a weakly significant greenium of about 6 basis points; for green bonds not used for refinancing,thegreeniumisaweaklysignificantatabout13basispointsatthe5%significancelevel. WeshowinColumn(6)thatthis7basispointdifferencebetweenrefinancing and non-refinancing green bonds is not statistically significant, indicating no measurable impact of the refinancing variable on the greenium. This result is notable because it suggests that whether a green bond drives additional environmental impact from new green projects,asopposedtorefinancingexistingprojects,isnotpricedintheprimarymarket.30 In combination, the results from Table 6 suggest that the greenium is primarily allocated to green bonds that are sufficiently green beyond some minimum threshold. That is, on themargin,itappearsthatwhatdeterminesthegreeniumisnotthecredibilityofthegreen promise,butrathertherequirementsthatmakebonds“greenenough”(fullorpartialGBP alignmentandexternalreview)tobeincludedinaconventionalESGorgreenbondportfolio. 4.6 Bond-andfirm-levelcharacteristics Inthislastsubsection, weanalyzetheconnectionbetweenbondandfirm-levelcharacteristics and the greenium. Table 7 presents the impact on the greenium by bond size, firm capital market access, and bond rating. In Column (1), we do not find that the size of the greenbonditselfhasanimpactonthegreenium. However,inColumn(2),weestimatethe greeniumwhilecontrollingfortheaveragesizeofanissuer’sbondswithinthesameyear. 30Greeninvestmentsaresaidtohaveanadditionalenvironmentalimpact(inESGjargon,”additionality”) whentheyproducebeneficialoutcomesthatwouldnothaveoccurredwithouttheinvestment. Forareview ofenvironmentalandsocialimpactinvesting,seeBrestandBorn(2013). 21

We find that, on average, firms that are able to place larger bonds (a proxy for access to debt capital markets) capture a larger greenium. Specifically, the coefficient on the green bond indicator decreases in the average size of an issuer’s bonds. Taking the average of the issuer log average bond size, 18.8 (about $152 million), the greenium is about 7 basis points(58.4minus3.5times18.8). InColumn(3),weassesstheimpactofagreenbond’sratingonthegreenium,segmenting thesampleintoinvestmentgrade(AAAtoBBB),highyield(BBtoC),andnotratedbonds. Wefindthatonlyinvestmentgradegreenbondscaptureastatisticallysignificantaverage greenium of about 10 basis points. In contrast, the greenium coefficients for high yield and not rated green bonds are not statistically significant. In Column (4), we do not find significant differences in the greenium between rating classes, despite the differences in groupmeans. Inaone-sidedt-testbetweeninvestmentgradeandnotratedbondsweneed to reject the null hypothesis that the estimated difference of -8 basis points is positive (pvalue: 0.051),suggestingthat,onaverage,investmentgradebondshavealargergreenium relative to no rated bonds. This finding indicates that firms which cannot afford to have their bonds rated or firms not covered by ratings agencies (for instance, SMEs and EM issuers)maynotcaptureagreenium. The results from Table 7 are consistent with our findings in Table 4 in that they point to a greenium for firms that are able to meet typical index inclusion criteria, namely issuers withanabilitytoregularlyissuelarge,investment-graderatedbonds. [INSERTTABLE7] In Table 8a, we analyze how the greenium is allocated by issuer sector. In Column (1), wefindthatonlyissuersfromthebankingsectorreceiveastatisticallysignificantaverage greeniumofabout9basispoints. Apossibleexplanationforthebankinggreeniumcould bethatbanksreceiveacompensationforthecostsofcertifying,extending,andmonitoring green loans to their customers, or that they capture a potential borrowing cost advantage thatmaybepassedthroughtothebanks’greendebtors. What’smore,inanattempttoprepare for anticipated climate-related financial disclosure the banking sector has increased their climate disclosure efforts, which may increase the attractiveness of banks’ bonds to investorswhoaresubjecttosustainability-relatedreportingrequirements. Mostothersectorsreceiveastatisticallyinsignificantnegativecoefficientestimate. Thenotableexception is the high-emission industry and materials sector, for which we estimate a positive yet statisticallyinsignificantcoefficient. InColumn(2),weshowthatwedonotfindstatisticallysignificantdifferencesinthegreenium between banks and issuers in other sectors, suggesting that other sectors show statistically comparable greeniums when the estimates’ confidence intervals are taken into account. The notable exception is the difference in the greenium between banks and industryandmaterials. Thisdifferenceisabout23basispointsandweaklysignificant. 22

[INSERTTABLE8a] Finally, in Table 8b, we break down the greenium in the banking sector by region. We find that it is primarily banks from emerging markets (EM), in particular China, as well asbanksfromtheEuropeanUnionthatare,onaverage,abletocaptureasignificantgreenium.31 The significant greenium estimate for EU banks lines up with our overall results on a robust euro greenium for large, investment-grade issuers. And, it is in line with the ideathatincreasedclimatedisclosures,inparticularinEurope,increasetheattractiveness of the region’s bonds relative to other developed markets. For emerging markets, where informationasymmetriescanbelargeandaccesstoissuerscanbelimited,theexplanation that the estimated greenium could be compensating banks for their monitoring costs and accesstogreendebtorsapplieswell. Lastly,nearlyalloftheEMandChinesebanks’green bondshavefullGBPalignment,externalreview,andahighcreditrating,whichcanmake thebondsattractiveforportfoliodiversificationforasustainableassetmanager. [INSERTTABLE8b] 5 Conclusion According to intergovernmental organizations and research institutes worldwide, hundreds of trillions of dollars of investment is required to achieve global net zero emissions by 2050. About two-thirds of this investment is expected to come from private sector investments in climate- and environment-friendly projects. Currently, one of the most recognized instruments for financing this type of investment is the green corporate bond, a long-termfixedincomedebtinstrumentthathasitsproceedsearmarkedforgreenprojects. In this paper, we study a large global panel of corporate bonds to understand the potential for green corporate bonds to incentivize green investments. To do so, we investigatewhethergreenbondsofferlowerborrowingcostsforissuersrelativetoconventional bonds. We find that, on average, dollar- and euro-denominated green corporate bonds haveaprimarymarketcreditspreadthatiseightbasispointslowercomparedtoconventional corporate bonds, reflecting a 5% reduction relative to the average credit spread in our sample. This greenium is distinct from a green bond issuer effect, or so-called “green halo”. It emerges, on average, only as of 2019, coinciding with the growth of the sustainable asset management industry following EU regulation. While this EU regulation appears to have been supportive for the development of the green bond market and its strengthinEurope,ourresultsdonotestablishacausallinkbetweentheregulationandthe emergenceofthegreeniumin2019. However,theexcessdemandfromgreenbonds,possiblyfromthesustainableassetmanagementindustry,appearstobeanimportantdriverof 31Besidesthebondrating,ourregressionspecificationcontrolsforwhetheranissuerisanSOEthroughthe issuerfixedeffect.Italsoaccountsfordifferencesinstateownershipacrossjurisdictionswithacountry-times- SOEinteraction. Inunreportedresults,thiseffectremainsqualitativelysimilarwhenwedropallSOEsfrom thesample. 23

thegreenbondborrowingcostadvantage. Infact,weshowthatthegreeniumislinkedto twoproxiesforexcessdemandpressure,bondoversubscriptionandbondindexinclusion. WhileU.S.dollar-andeuro-denominatedgreenbondscapturecomparablegreeniums,we find that the greenium is allocated primarily to local euro and foreign U.S. dollar issuers. While green bond governance and external review appear to matter for the greenium, the credibility of the underlying projects does not have a significant impact. Instead, the greeniumisunevenlydistributedtolarge,investment-gradeissuers,primarilywithinthe bankingsectoranddevelopedeconomies. Ourfindingshaveimplicationsfortheroleofcorporategreenbondsinincentivizingmeaningful green investments throughout the global economy. The empirical evidence suggests that a greenium exists, but it primarily favors large, rated European firms, does not necessarily reward high-quality green projects, and is small or potentially negative when taken net of fees and compliance costs. Part of the funding advantage of green corporate bonds may be driven by potentially temporary supply and demand imbalances between green bond issuers and the growing sustainable asset management industry.32 Furthermore, questions of additionality−whether green bonds incentivize additional green investment,orsimplyrewardgreeninvestmentsthatwouldhaveoccurredanyway−remain unresolved. For these reasons, the borrowing cost advantage of green corporate bonds likely plays a limited role in incentivizing global, large-scale climate investment. Instead, the potential benefitofissuingagreenbondwouldlikelybeanindirectsignalingbenefitthatimproves theenvironmentalcredentialsoftheissuer. Thisraisesquestionsaboutthepotentialqualityofthesignalfromissuingagreenbond. Whileissuingagreenbondmayindicatethat a firm is more green than its peers, the potential for greenwashing is high and there is no guaranteethatthegreenprojectsofagreenbondarelinkedwithalong-term,meaningful commitmentforgreeninvestment. Mandated,standardizedreportingonenvironmentalmetricssuchasgreenhousegasemissions, as well as required disclosure of climate-related risks and future corporate development, may provide a more effective signaling solution. Financial instruments that are linked to firm-level environmental metrics, such as sustainability-linked bonds, rather than specific green projects (like green bonds), may be more effective for connecting sustainabilitymindedinvestorswithfirmsmakingcredible,systemicgreeninvestments. Situatedwithintheneedforsubstantial,rapidinvestmentstoreachnetzeroemissionsby2050, an over-reliance on the potentially limited incentives conferred from issuing green bonds could lead to an undersupply of green investment if green bonds are not complemented byothersourcesofprivatesectorfunding. 32Continuedgrowthingreencorporatebondissuancecouldthereforeleadtoadilutionofthegreenium, should supply catch up with demand. That is, an exhaustion at the margin of the direct financial benefit to green issuance. This potential dilution has already been noted anecdotally by market participants and financialjournalists(see,forexample,Bahceli(2021)andStubbington(2021)). 24

References Bahceli,Y. (2021). Analysis: ’Greenium’shrinksasclimatebondsalesswelltorecord. Reuters. Baker, M., Bergstresser, D., Serafeim, G.,&Wurgler, J. (2022). Thepricingandownership ofUSgreenbonds. AnnualReviewofFinancialEconomics,forthcoming. Baulkaran, V. (2019). Stock market reaction to green bond issuance. Journal of Asset Management,20(5),331–340. BloombergFinanceLP. (2021). BloombergBackOfficeCreditRiskData. BloombergNEF. (2021). NewEnergyOutlook. Brest, P., & Born, K. (2013). Unpacking the impact in impact investing. Stanford Social InnovationReview,11. Climate Bonds Initiative. (2021). Green Bond Pricing in the Primary Market: January - June 2021. Correia,S. (2015). Singletons,cluster-robuststandarderrorsandfixedeffects: Abadmix. TechnicalNote,DukeUniversity,7. Dannhauser, C. D. (2017). The impact of innovation: Evidence from corporate bond exchange-tradedfunds(ETFs). JournalofFinancialEconomics,125(3),537–560. Dathan, M., & Davydenko, S. (2020). Debt issuance in the era of passive investment. AvailableatSSRN3152612. Daubanes,J.X.,Mitali,S.F.,&Rochet,J.-C. (2021). Whydofirmsissuegreenbonds? Swiss FinanceInstituteResearchPaper,21. Do¨ttling,R.,&Kim,S. (2020). Sustainabilitypreferencesunderstress: Mutualfundflows duringCOVID-19. VoxEU.org,19. Edwards, C., Harju, J., Aksu, C., Foux, M., & Herold, F. (2020). Green is but a colour. Barclays. Edwards, C., Rekrut, D., Boulanger, C., Elias, K., O’Neal, M., Boyipadu, P., ... Pobjoy, J. (2022). AcompilationofESGbondtradeideas. Barclays. Ehlers, T., Mojon, B., & Packer, F. (2020). Green bonds and carbon emissions: Exploring thecaseforaratingsystematthefirmlevel. BISQuarterlyReview,September. Ehlers,T.,&Packer,F. (2017). GreenBondFinanceandCertification. BISQuarterlyReview September2017. Energy Transitions Commission. (2021). Making Mission Possible: Delivering a Net-Zero Economy. Fancy, T. (2021). TariqFancyonthefailureofgreeninvestingandtheneedforstateaction. The Economist. Flammer,C. (2021). Corporategreenbonds. JournalofFinancialEconomics,142. Flanagan, T., Kedia, S., & Zhou, X. A. (2019). Secondary market liquidity and primary marketallocationsincorporatebonds. AvailableatSSRN3449431. Flanagan,T.,Kedia,S.,&Zhou,X.A. (2021). Assessinggainsfromprimarymarketallocationsincorporatebonds. AvailableatSSRN3859063. Fletcher,L.,&Oliver,J. (2022). Greeninvesting: Theriskofanewmis-sellingscandal. Financial Times. 25

Forfot, J. S., & Fosse, H. G. (2021). The green halo debt effect. Norwegian University of ScienceandTechnology: Master’sThesis. Gianfrate, G., & Peri, M. (2019). The green advantage: Exploring the convenience of issuinggreenbonds. JournalofCleanerProduction,forthcoming. Gilchrist,S.,&Zakrajsˇek,E. (2012). Creditspreadsandbusinesscyclefluctuations. AmericanEconomicReview,102. Global Financial Markets Association and Boston Consulting Group. (2020). Climate FinanceMarketsandtheRealEconomy. Holden,C.W.,&Nam,J. (2019). Marketaccessibility,corporatebondETFs,andliquidity. KelleySchoolofBusinessResearchPaper,3083257. InternationalCapitalMarketsAssociation. (2021). GreenBondPrinciples: VoluntaryProcess GuidelinesforIssuingGreenBonds. InternationalEnergyAgency. (2021). WorldEnergyOutlook2021. InternationalMonetaryFund. (2021). WorldEconomicOutlookOctober2021. Kapraun,J.,Latino,C.,Scheins,C.,&Schlag,C. (2021). (In)-crediblygreen: Whichbonds tradeatagreenbondpremium? InProceedingsofParisDecember2019FinanceMeeting EUROFIDAI-ESSEC. Kuchtyak, M., & Bruce, E. (2022). Sustainable bonds to hit record $1.35 trillion in 2022. Moody’s. Larcker, D. F., & Watts, E. M. (2020). Where’s the greenium? Journal of Accounting and Economics,69. Lau,P.,Sze,A.,Wan,W.,&Wong,A. (2022). Theeconomicsofthegreenium: Howmuch is the world willing to pay to save the earth? Environmental & Resource Economics, 81. Levine,M. (2019). GreenBondsWithouttheBonds. Bloomberg. Maltais, A., & Nykvist, B. (2020). Understanding the role of green bonds in advancing sustainability. JournalofSustainableFinance&Investment,1–20. Marsh, A. (2020). Green Bond Crisis Premium Theory Debunked by Covid, Manager Says. Bloomberg. McKinsey. (2022). Theeconomictransformation: Whatwouldchangeinthenet-zerotransition. Morningstar, Inc. (2021). Morningstar Direct. Retrieved from http://corporate .morningstar.com/US/asp/subject.aspx?xmlfile=40.xml Mundy,S.(2022).‘Greenwashing’warningsacceleratedriveforbusinesssustainabilitystandards. FinancialTimes. NatWestMarkets. (2019). Greenhalo2.0. PartnershipforCarbonAccountingFinancials. (2021). Draft: Newmethodsforpublicconsultation: Forfinancialinstitutionsmeasuringandreportingscope3category15emissions. Partridge, C., & Medda, F. R. (2020). Green bond pricing: The search for greenium. The JournalofAlternativeInvestments,23. Quinson,T. (2021a). HowWallStreetisGamingESGScores. Bloomberg. 26

Quinson, T. (2021b). Regulators Intensify ESG Scrutiny as Greenwashing Explodes. Bloomberg. Refinitiv. (2021). Workspace. Ritchie, G., & Rocha, P. A. (2021). A $180 Billion Green-Debt Boom Grows Faster Than Its Impact. Bloomberg. Ritchie,G.,Ward,J.,Kishan,S.,&Gledhill,A. (2021). BondInvestorRevoltBrewsOverBogus GreenDebtFloodingMarket. Bloomberg. Stubbington, T. (2021). Squeeze on ‘greenium’ as ESG bond investors demand more value. FinancialTimes. Stubbington, T., & Nauman, B. (2020). Investors probe ESG credentials of bond sellers on ’greenwashing’fears. FinancialTimes. Tang, D. Y., & Zhang, Y. (2020). Do shareholders benefit from green bonds? Journal of CorporateFinance,61. Temple-West,P.,&Palma,S. (2022).SECpreparestocrackdownonmisleadingESGinvestment claims. FinancialTimes. TheEconomist. (2021). Sustainablefinanceisrifewithgreenwash.Timeformoredisclosure. Wirz,M. (2022). BondInvestorsChallengeWallStreetGreenwashing. WallStreetJournal. Ye,S. (2019). HowdoETFsaffecttheliquidityoftheunderlyingcorporatebonds. Chinese UniversityofHongKongWorkingPaper. 27

Table1: Greeniumandgreenhalobaselineregressions Yieldspread (1) (2) (3) (4) (5) (6) (7) Green -11.35∗∗∗ -8.232∗∗∗ -9.020∗∗∗ -9.782∗∗∗ -8.350∗∗∗ (2.388) (2.527) (2.532) (2.696) (2.688) Greenissuer -4.781∗ -1.493 -3.876 0.521 (2.516) (3.976) (2.600) (4.157) ✓ ✓ ✓ ✓ ✓ ✓ ✓ Controls Firm×YearFE ✓ ✓ ✓ Firm×QuarterFE ✓ Observations 126,373 114,879 102,124 126,373 114,879 126,373 114,879 R2 0.783 0.842 0.872 0.783 0.842 0.783 0.842 AdjustedR2 0.762 0.809 0.833 0.762 0.809 0.762 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondandGreenissuerindicatorvariables. Thesample periodcoversgreenandconventionalbondissuesfrom2014to2021(Section3detailsthesampleconstruction). The dependentvariableYieldspreadreferstoabond’syieldspreadonitsdateofissuanceoverthematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. TheindependentvariableGreentakesthevalueofonefor greenbondsandzerootherwise.Greenissuerisequaltooneforagivenbondifitsissuerhaspreviouslyissuedagreen bond(orifthegivenbondistheissuer’sfirstgreenbond). Allothercontrolvariablesandfixedeffectsaredetailedin Section3. Standarderrors,reportedinparentheses,areclusteredontheissuerultimateparentandyear-monthlevels. ***,**,and*indicatestatisticalsignificanceatthe1%,5%,and10%level,respectively. 28

Table2: GreeniumandEUregulation (a)Baselineresultsandsegmentationbycurrency Yieldspread (1) (2) (3) (4) (5) (6) Green -2.660 -4.385 (3.511) (3.748) Green×Pre-SFDR -2.660 -4.385 (3.511) (3.748) Green×Post-SFDR -14.39∗∗∗ -9.687∗∗∗ -11.73∗∗∗ -5.302 (3.180) (3.136) (4.439) (4.525) Green×EUR -0.711 -2.357 (3.273) (3.017) Green×EUR×Post-SFDR -10.38∗∗ -4.406 (4.003) (3.740) Green×USD -4.855 -6.672 (6.406) (7.063) Green×USD×Post-SFDR -15.37∗ -8.666 (8.259) (8.597) ✓ ✓ ✓ ✓ ✓ ✓ Controls Firm×YearFE ✓ ✓ ✓ Observations 126,373 114,879 126,373 114,879 126,373 114,879 R2 0.783 0.842 0.783 0.842 0.783 0.842 AdjustedR2 0.762 0.809 0.762 0.809 0.762 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorvariableandEUregulationindicator variables. The sample period covers green and conventional bond issues from 2014 to 2021 (Section 3 details the sampleconstruction). ThedependentvariableYieldspreadreferstoabond’syieldspreadonitsdateofissuanceover thematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. TheindependentvariableGreen takesthevalueofoneforgreenbondsandzerootherwise. Pre-SFDRisequaltoonebeforeApril18, 2019andzero afterwards.Post-SFDRisequaltozerobeforeApril18,2019andoneafterwards.EURandUSDareindicatorvariables thatflageuroandUSdollargreenbonds.AllothercontrolvariablesandfixedeffectsaredetailedinSection3.Standard errors,reportedinparentheses,areclusteredontheissuerultimateparentandyear-monthlevels.***,**,and*indicate statisticalsignificanceatthe1%,5%,and10%level,respectively. 29

(b)Segmentationbyregion Yieldspread (1) (2) Green×China 8.355 7.717 (12.70) (10.95) Green×EU -3.002 -5.778∗ (3.612) (3.027) Green×USA 10.40∗ 9.311∗ (5.712) (5.293) Green×OtherDM -25.27∗∗ -24.18 (12.61) (16.93) Green×OtherEM 4.683 2.811 (17.02) (25.61) Green×China×Post-SFDR -17.11 -24.98 (15.37) (15.45) Green×EU×Post-SFDR -10.14∗∗ -1.025 (4.448) (4.321) Green×USA×Post-SFDR -29.64∗∗∗ -15.51 (10.54) (11.55) Green×OtherDM×Post-SFDR 11.45 6.712 (10.28) (12.90) Green×OtherEM×Post-SFDR -30.31 -23.10 (23.59) (28.83) ✓ ✓ Controls Firm×YearFE ✓ Observations 126,373 114,879 R2 0.783 0.842 AdjustedR2 0.762 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorvariableandEUregulationindicator variables. The sample period covers green and conventional bond issues from 2014 to 2021 (Section 3 details the sampleconstruction). ThedependentvariableYieldspreadreferstoabond’syieldspreadonitsdateofissuanceover thematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. TheindependentvariableGreen takesthevalueofoneforgreenbondsandzerootherwise. Post-SFDRisequaltozerobeforeApril18,2019andone afterwards. Euro area, USA, Other DM, China, and EM are indicator variables reflecting issuer geography. All other controlvariablesandfixedeffectsaredetailedinSection3. Standarderrors,reportedinparentheses,areclusteredon theissuerultimateparentandyear-monthlevels. ***,**,and*indicatestatisticalsignificanceatthe1%,5%,and10% level,respectively. 30

Table3: Greeniumandoversubscription Yieldspread (1) (2) (3) (4) (5) (6) (7) Green×Oversubscription -5.791∗∗∗ -10.76∗ (2.109) (6.386) Green×Imputedoversubscription -7.145∗∗∗ (2.067) Green 7.826 -17.81∗∗∗ 10.95 (8.450) (5.679) (8.713) Green×Lowoversubscriptionbucket -4.397 13.41∗∗ (3.036) (6.221) Green×Mediumoversubscriptionbucket -2.490 15.32∗∗ (4.905) (6.529) Green×Highoversubscriptionbucket -17.81∗∗∗ (5.679) Green×EUR×Oversubscription -8.597∗∗∗ -15.84∗∗ (2.933) (7.339) Green×USD×Oversubscription -2.454 -9.067 (3.129) (6.137) ✓ ✓ ✓ ✓ ✓ ✓ ✓ Controls Firm×YearFE ✓ ✓ ✓ ✓ ✓ ✓ ✓ Observations 114,168 114,879 114,168 114168 114,168 114,168 114,168 R2 0.842 0.842 0.842 0.842 0.842 0.842 0.842 AdjustedR2 0.808 0.809 0.808 0.808 0.808 0.808 0.808 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorandbondoversubscriptionvariables. Thesampleperiodcoversgreenandconventional bond issues from 2014 to 2021 (Section 3 details the sample construction). The dependent variable Yield spread refers to a bond’s yield spread on its date of issuance over the maturity-matched government bond yield for the given bond’s currency region. The independent variable Green takes the value of one for greenbondsandzerootherwise. Oversubscriptionistheratioofthenotionalamountoforderstotheactualamountofissueddebt. Low/medium/highoversubscription bucketrepresenttercilesformedonbonds’oversubscriptionratios. EURandUSDareindicatorvariablesthatflageuroandUSdollargreenbonds. Allothercontrol variablesandfixedeffectsaredetailedinSection3. Standarderrors,reportedinparentheses,areclusteredontheissuerultimateparentandyear-monthlevels. ***, **,and*indicatestatisticalsignificanceatthe1%,5%,and10%level,respectively. 31

Table4: Greeniumandbondindexinclusion Yieldspread (1) (2) (3) (4) Green×Tripleindexinclusion -10.04∗∗∗ -4.549 (2.869) (4.697) Green -5.610 (3.814) Green×EUR×TripleIndexInclusion -11.81∗∗∗ -12.52∗∗ (3.340) (4.880) Green×USD×TripleIndexInclusion -7.847 12.58 (5.065) (10.75) Green×EUR 0.696 (3.047) Green×USD -20.93∗∗ (10.34) ✓ ✓ ✓ ✓ Controls Firm×YearFE ✓ ✓ ✓ ✓ Observations 114879 114879 114879 114879 R2 0.842 0.842 0.842 0.842 AdjustedR2 0.809 0.809 0.809 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorandthebondindexinclusionindicator variables. The sample period covers green and conventional bond issues from 2014 to 2021 (Section 3 details the sampleconstruction). ThedependentvariableYieldspreadreferstoabond’syieldspreadonitsdateofissuanceover thematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. TheindependentvariableGreen takesthevalueofoneforgreenbondsandzerootherwise. Tripleindexinclusionisequaltooneifagreenbondwasa constituentofthreemajorgreenbondindices(ICE,Solactive,JPMorgan). EURandUSDareindicatorvariablesthat flageuroandUSdollargreenbonds. AllothercontrolvariablesandfixedeffectsaredetailedinSection3. Standard errors,reportedinparentheses,areclusteredontheissuerultimateparentandyear-monthlevels.***,**,and*indicate statisticalsignificanceatthe1%,5%,and10%level,respectively. 32

Table5: Greeniumandcurrencydenominationandregion Yieldspread (1) (2) (3) (4) (5) (6) Green×EUR -5.734∗∗∗ -5.404 (2.155) (5.357) Green×USD -12.36∗∗ -6.625 -14.94∗∗ (5.081) (5.451) (6.195) Green -5.734∗∗∗ -6.203∗∗ (2.155) (2.406) Green×LocalEUR -5.823∗∗∗ -0.419 (2.175) (5.428) Green×ForeignEUR -5.404 (5.357) Green×LocalUSD -4.218 10.72 (9.893) (12.12) Green×ForeignUSD -14.94∗∗ (6.195) Green×China -10.03 -3.824 (9.005) (8.974) Green×Euroarea -6.203∗∗ (2.406) Green×USA -0.846 5.357 (7.635) (8.089) Green×OtherDM -19.67∗∗ -13.47 (8.929) (9.155) Green×OtherEM -12.36 -6.155 (16.53) (17.00) ✓ ✓ ✓ ✓ ✓ ✓ Controls Firm×YearFE ✓ ✓ ✓ ✓ ✓ ✓ Observations 114879 114879 114879 114879 114879 114879 R2 0.842 0.842 0.842 0.842 0.842 0.842 AdjustedR2 0.809 0.809 0.809 0.809 0.809 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorandcurrencydenominationindicator variables. The sample period covers green and conventional bond issues from 2014 to 2021 (Section 3 details the sampleconstruction). ThedependentvariableYieldspreadreferstoabond’syieldspreadonitsdateofissuanceover thematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. TheindependentvariableGreen takesthevalueofoneforgreenbondsandzerootherwise.EURandUSDareindicatorvariablesthatflageuroandU.S. dollargreenbonds.LocalandForeigndifferentiatecurrencydenominationbylocalandforeigncurrencyissuers.China, Euroarea,USA,OtherDM,andOtherEMareindicatorvariablesreflectingissuergeography.Allothercontrolvariables andfixedeffectsaredetailedinSection3.Standarderrors,reportedinparentheses,areclusteredontheissuerultimate parentandyear-monthlevels.***,**,and*indicatestatisticalsignificanceatthe1%,5%,and10%level,respectively. 33

Table6: GreeniumandbondGovernance,externalreview,andproceedcredibility Yieldspread (1) (2) (3) (4) (5) (6) Green×GBPaligned -7.214∗∗∗ (2.363) Green×GBPNotaligned -20.25∗ -13.04 (11.22) (11.13) Green -7.214∗∗∗ -8.088∗∗∗ -6.453∗∗ (2.363) (2.585) (2.726) Green×Externalreview -8.088∗∗∗ (2.585) Green×Noexternalreview -9.317 -1.229 (8.113) (8.385) Green×Norefinancing -6.453∗∗ (2.726) Green×Refinancing -13.73∗∗ -7.281 (5.676) (5.651) ✓ ✓ ✓ ✓ ✓ ✓ Controls Firm×YearFE ✓ ✓ ✓ ✓ ✓ ✓ Observations 114,879 114,879 114,879 114,879 114,403 114,403 R2 0.842 0.842 0.842 0.842 0.842 0.842 AdjustedR2 0.809 0.809 0.809 0.809 0.809 0.809 Regressions of corporate yield spreads at issuance on the Green bond indicator and variables capturing green bond governance,externalreview,andcredibility. Thesampleperiodcoversgreenandconventionalbondissuesfrom2014 to2021(Section3detailsthesampleconstruction). ThedependentvariableYieldspreadreferstoabond’syieldspread onitsdateofissuanceoverthematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. The independentvariableGreentakesthevalueofoneforgreenbondsandzerootherwise.GBPalignedisequaltozeroifa greenbondisnotfullyalignedwiththeGreenBondPrinciplesandoneifitisfullyaligned. Externalreviewisequalto oneifagreenbondwassubjecttoapre-issuanceexternalreviewbyathirdparty.Norefinancingisequaltooneifsome portionofthebondwasusedtorefinanceanexistingliability,otherwiseitiszero.Allothercontrolvariablesandfixed effectsaredetailedinSection3. Standarderrors,reportedinparentheses,areclusteredontheissuerultimateparent andyear-monthlevels.***,**,and*indicatestatisticalsignificanceatthe1%,5%,and10%level,respectively. 34

Table7: Greeniumandbond-andfirm-levelcharacteristics Yieldspread (1) (2) (3) (4) Green 14.74 58.40∗∗ -9.793∗∗∗ (26.70) (28.18) (3.063) Green×Size -1.201 (1.413) Green×Averageissuerbondsize -3.473∗∗ (1.489) Green×Investmentgrade -9.793∗∗∗ (3.063) Green×Highyield -23.99 -14.20 (18.28) (18.79) Green×Notrated -1.669 8.124 (3.278) (4.904) ✓ ✓ ✓ ✓ Controls Firm×YearFE ✓ ✓ ✓ ✓ Observations 114,879 114,879 114,879 114,879 R2 0.842 0.842 0.842 0.842 AdjustedR2 0.809 0.809 0.809 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorvariableandvariablescapturingbond andissuercharacteristics. Thesampleperiodcoversgreenandconventionalbondissuesfrom2014to2021(Section 3detailsthesampleconstruction). ThedependentvariableYieldspreadreferstoabond’syieldspreadonitsdateof issuanceoverthematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. Theindependent variableGreentakesthevalueofoneforgreenbondsandzerootherwise. Sizeisthelogarithmofthebond’snotional amountissued.Averageissuerbondsizeistheaveragesizeofanissuer’sbondsissuedwithinthesameyearofthegreen bond. Investmentgrade,Highyield,andNoratedareindicatorvariablesthatflagtherespectivebondratings. Allother controlvariablesandfixedeffectsaredetailedinSection3. Standarderrors,reportedinparentheses,areclusteredon theissuerultimateparentandyear-monthlevels. ***,**,and*indicatestatisticalsignificanceatthe1%,5%,and10% level,respectively. 35

Table8: Greeniumandissuerindustry (a)Industry Yieldspread (1) (2) Green -8.682∗∗∗ (2.970) Green×Alternateenergy -26.40 -17.72 (25.70) (25.48) Green×Banks -8.682∗∗∗ (2.970) Green×Electricutilitiesandfossilfuels -14.88∗ -6.195 (7.826) (8.725) Green×Industryandmaterials 14.37 23.06∗ (11.33) (11.83) Green×Non-bankfinancials 0.209 8.891 (15.35) (15.37) Green×Realestate 0.0330 8.715 (9.577) (10.10) Green×Transportation -15.27∗ -6.591 (8.699) (9.618) Green×Other -8.064 0.618 (5.744) (6.400) ✓ ✓ Controls Firm×YearFE ✓ ✓ Observations 114879 114879 R2 0.842 0.842 AdjustedR2 0.809 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorandissuerindustryindicatorvariables. Thesampleperiodcoversgreenandconventionalbondissuesfrom2014to2021(Section3detailsthesampleconstruction). The dependent variable Yield spread refers to a bond’s yield spread on its date of issuance over the maturitymatchedgovernmentbondyieldforthegivenbond’scurrencyregion.TheindependentvariableGreentakesthevalue ofoneforgreenbondsandzerootherwise.Alternativeenergy,Banks,Electricutilitiesandfossilfuels,Industryandmaterials, Non-bankfinancials,Realestate,Transportation,andOtherareindicatorvariablesthatflagtherespectiveissuerindustries. AllothercontrolvariablesandfixedeffectsaredetailedinSection3.Standarderrors,reportedinparentheses,areclusteredontheissuerultimateparentandyear-monthlevels. ***,**,and*indicatestatisticalsignificanceatthe1%,5%, and10%level,respectively. 36

(b)Banksbyregion Yieldspread (1) (2) Green -27.61∗∗∗ (5.060) Green×Banks×China -27.61∗∗∗ (5.060) Green×Banks×EU -6.170∗∗∗ 21.44∗∗∗ (2.320) (5.188) Green×Banks×USA 11.89 39.50∗∗∗ (9.002) (11.09) Green×Banks×OtherDM -2.144 25.46∗∗ (8.864) (10.99) Green×Banks×OtherEM -66.40∗ -38.79 (37.03) (37.06) Green×Non-bankfinancials -0.00835 27.60∗ (15.37) (16.11) Green×Non-financials -8.558∗ 19.05∗∗∗ (5.031) (6.450) ✓ ✓ Controls Firm×YearFE ✓ ✓ Observations 114,879 114,879 R2 0.842 0.842 AdjustedR2 0.809 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorandissuerindustryandregionvariables. Thesampleperiodcoversgreenandconventionalbondissuesfrom2014to2021(Section3detailsthesampleconstruction). The dependent variable Yield spread refers to a bond’s yield spread on its date of issuance over the maturitymatchedgovernmentbondyieldforthegivenbond’scurrencyregion.TheindependentvariableGreentakesthevalue ofoneforgreenbondsandzerootherwise. Banks, Non-bankfinancials, andNon-financialsareindicatorvariablesthat flagtherespectiveissuerindustries. China,EU,USA,OtherDM,andEMareindicatorvariablesreflectingissuergeography. AllothercontrolvariablesandfixedeffectsaredetailedinSection3. Standarderrors,reportedinparentheses, areclusteredontheissuerultimateparentandyear-monthlevels. ***,**,and*indicatestatisticalsignificanceatthe 1%,5%,and10%level,respectively. 37

Appendix A1 Alternategreeniumtheories Therearetwomainalternativeexplanationsforthegreenium. Thefirstisthatgreenbonds are more valuable because they are expected to achieve higher risk-adjusted returns relative to conventional bonds, as they are issued by firms that are reducing their exposure to climate- and other environment-related physical and transition risks. This explanation is problematic because the coupon and principal payments from green bonds are almost never paid out from cash flows generated by the green bond’s underlying green projects. Instead,thecreditriskofagreenbondshouldbeconsideredidenticaltothecreditriskof afirm’sconventionalbonds,particularlywhenthebondshavethesamerecourse. Evenif a firm’s issuance of a green bond reduces its overall credit risk by lowering its exposure tosustainabilityriskfactors,thiswouldnotnecessarilyimprovetherisk-adjustedreturns expected for green bonds. Accordingly, we show in Table 1 Column (7) that green bonds receive a greenium distinct from the green halo benefits to a firm’s overall cost of capital, suggestingthisexplanationisinsufficient. The second explanation for the greenium, called the crisis premium theory, is that green bonds offer better returns during crisis periods, because environmentally conscious investors are less likely to sell green bonds during market distress. For this explanation to hold,itwouldhavetobetruethatgreeninvestorsare,atthemargin,lessriskaversethan conventionalinvestors. Thisseemsplausibleforinstitutionalinvestors,whomaytradeoff sellingsustainableassetsagainsttheperceivedpenaltytotheircostofcapitalfromfailing to hit their sustainable investment goals, typically articulated in an ESG shareholder engagement. However, this seems less plausible for retail investors, who are typically more risk averse and may view green investments as a luxury preference that withers under financialoreconomicstress.33 Whilethecrisispremiumtheoryshouldbeconsidered,the limited available market participant commentary suggests this theory is not widely held. See,forexample,Edwards,Harju,Aksu,Foux,andHerold(2020)andMarsh(2020). A2 Thegreeniumandliquidity Ourregressionspecificationsusecontrols,suchasfirm,rating,andtimefixedeffects,that mayproxyfortheimpactofexpectedsecondarymarketliquidityonpricingintheprimary market. However, data availability limits our ability to construct measure of expected liquidity to directly control for it at issuance. As such, it is possible that we are not fully capturing the impact of expected liquidity on the greenium. In this section, we briefly 33Forexample,Do¨ttlingandKim(2020)findevidencethatthesustainabilitypreferencesofretailinvestors deterioratedrelativetoinstitutionalinvestorsduringthemarketcrashofMarch2020. 38

outline how the role of buy-and-hold investors in the green bond market may affect the greenium.34 A2.1 Expectedsecondarymarketliquidity Assumetheyieldforagivenbondcanbedecomposedintothreecomponents,suchthat y = y +y +y (A1) F D EL wherey isthecompensationforabond’sfundamentalcharacteristics;y isthecompen- F D sation for investor demand for the bond; and y is the compensation for a bond’s ex- EL pectedliquidityinthesecondarymarket. Thebond’sfundamentalcharacteristics,suchas itscreditratingandmaturity,determineitsriskprofileacrossseveraldimensions,includingcreditrisk,interestraterisk,andinflationrisk. Agivenbondwithriskierfundamentals willhavehighervaluesofy (lowerpricep ). Beyondthefundamentals, excessinvestor F F demand for a given bond will exert further demand pressure on the bond, lowering the value of y (raising price p ). Finally, lower expected liquidity in the secondary market D D shouldraisey (lowerthepricep ),becausebondsthataremoredifficulttosellorbuy EL EL inthesecondarymarketarelessvaluableortiedtohightransactioncosts. Whenmeasuringthegreeniuminourregressions, wecomparetheyieldonagreenbond tothatofacomparableconventionalbond(observedorhypothetical)withthesamefundamentals. Inotherwords,weassumethatforacomparablegreenbond,yGreen = yConventional. F F Therefore, the yield differential between a green bond and a comparable conventional bondisgivenby: (cid:16) (cid:17) (cid:16) (cid:17) yGreen−yConventional = yGreen−yConventional + yGreen−yConventional (A2) D D EL EL (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) Yielddifferential Greenium(-) Expectedliquiditydifferential(+) WearguethatyGreen < yConventional,becauseexcessdemandforthegreenbondlowersits D D yieldrelativetotheyieldfortheconventionalbond. Thereforethegreeniumtermisnegative. However, if a green bond has a lower expected liquidity than a comparable conventional bond, because green bond investors are more likely to be buy-and-hold investors, then yGreen > yConventional, and the expected liquidity term will be positive. Controlling EL EL for relevant bond characteristics, we find that the green bond yield differential, given by yGreen−yConventional,isnegative. Ifthereisanexpectedliquiditypenaltyforgreenbonds beyondtheirfundamentals,meaningtheexpectedliquiditytermispositive,thenwemay 34SeeFlanagan, Kedia,andZhou(2019)forempiricalevidencethatlinksprimarymarketallocationsand secondarymarketliquidityforasubsetofbuy-and-holdinvestors. 39

be underestimating the magnitude of the ”true” greenium by not controlling for differences in expected liquidity. However, even if this were the case, it would not impact the ultimate borrowing cost from the issuer’s perspective, measured by the yield differential yGreen−yConventional. A2.2 Near-termexpectedliquidity We can decompose the yield even further. First, note that y refers to investor compen- EL sation for long-term expected secondary market liquidity, referring to the expected transactioncostsofsellingthebondbeforeitmatures. Inthiscontext,lowersecondarymarket liquiditywouldreducetheoverallpricep ofthebond(raiseyieldy ). However,there EL EL isanadditionalfactorinplay–thenear-termexpectedliquidity. Ifinvestorsbelieveitwill be costly to buy the bond in the secondary market, they may, at the margin, be willing to buy the bond in the primary market, driving further excess demand for a bond beyond its fundamentals. Let us denote the yield compensation from this primary market excess demand y , where a stronger preference for a primary market allocation lowers y PM PM (raises y ). Let us also relabel the residual demand uncorrelated with expected bond PM liquidityasy . Thus,theyielddifferentialfromEquationA2canberewritten: RD (cid:16) (cid:17) (cid:16) (cid:17) yGreen−yConventional = yGreen−yConventional + yGreen−yConventional + D D EL EL (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) Yielddifferential Greenium(-) Expectedliquiditydifferential(+) (A3) (cid:16) (cid:17) yGreen−yConventional PM PM (cid:124) (cid:123)(cid:122) (cid:125) Primarymarketdemand(-) Ifagreenbondhasalowerexpectedliquidityinthesecondarymarketthanacomparable conventional bond, this should increase the preference to buy the green bond in the primary market. This means yGreen < yConventional, so the additional primary market term PM PM will still be negative. As in the previous subsection, note that while this may impact the estimationofthegreeniumterminEquationA3,itultimatelydoesnotimpacttheborrowingcostoftheissuertotheextentthattheadditionalprimarymarketdemandisdrivenby thegreenlabelitself. 40

FigureA1: Impactoftime-varyingfixedeffectsongreenbondsamplesize (a)Numberofgreenbonds Number of bonds 500 Sample Full sample 400 Annual FE Quarterly FE 300 200 100 0 2014 2015 2016 2017 2018 2019 2020 2021 (b)Notionalamountofgreenbondissuance Amount issued ($B) 200 Sample Full sample 150 Annual FE Quarterly FE 100 50 0 2014 2015 2016 2017 2018 2019 2020 2021 Panel (a) is the number of green bonds in our sample. Panel (b) is the notional amount of green bondissuanceinoursample. Thedataisreportedatanannualfrequency. Fullsampleisthesample when we do not include time-varying fixed effects. Annual FE and Quarterly FE are the samples whenweincludeannualorquarterlytime-varyingfixedeffects,respectively. 41

TableA1: Bondsamplebycurrency Currency Amt. issued($B) N AUD 243.0 2,109 BRL 13.5 361 CAD 605.4 2,386 CHF 397.3 1,763 CNY 4,161.0 21,572 DKK 24.7 313 EUR 6,708.6 22,094 GBP 542.1 1,222 HKD 67.7 1,226 IDR 54.3 1,572 ILS 40.7 304 INR 515.5 9,146 JPY 897.6 6,583 KRW 824.3 14,440 MXN 25.8 191 MYR 126.4 2,191 NOK 69.6 874 NZD 37.0 476 RUB 61.7 660 SEK 311.6 790 SGD 57.5 451 USD 17,768.9 39,239 ZAR 9.3 249 Total 33,563.6 130,212 42

TableA2: Bondsamplesummarystatistics Mean S.D. Min. 25th Median 75th Max. Greenbonds Yieldspread(bp) 149.8 128.3 -212.3 75.3 100.3 190.1 959.1 Coupon(%) 2.0 1.9 0.0 0.5 1.5 3.0 12.0 Yearstomaturity 7.8 4.1 1.0 5.0 7.0 10.0 30.0 Amountissued($M) 419.4 488.3 0.6 29.0 370.2 598.2 7,030.2 Observations 1,169 Conventionalbonds Creditspread(bp) 137.3 168.3 -499.5 43.0 97.8 198.4 997.5 Coupon(%) 3.5 2.8 0.0 1.4 3.0 5.0 20.8 Yearstomaturity 6.1 4.7 1.0 3.0 5.0 8.0 30.0 Amountissued($M) 256.3 623.9 0.5 16.9 76.0 300.0 111,639.5 Observations 129,043 TableA3: Greenbondsampleoversubscriptionsummarystatistics Mean S.D. Min. 25th Median 75th Max. Oversubscription 3.8 2.1 1.0 2.2 3.3 4.7 15.0 Observations 474 43

TableA4: Bondsamplebyrating Rating Amt. issued($B) N Greenbonds AAA 64.5 47 AA 36.8 63 A 154.5 256 BBB 157.4 260 BB 39.0 73 B 14.3 35 C 0.4 1 Notrated 23.5 434 Total 490.3 1,169 Conventionalbonds AAA 2,360.5 3,467 AA 2,485.0 6,109 A 6,955.2 14,557 BBB 7,059.1 12,634 BB 3,122.6 4,946 B 2,351.5 3,789 C 396.8 699 Notrated 8,342.6 82,842 Total 33,073.3 129,043 TableA5: Greenbondsamplebysector Sector Amt. issued($B) N Alternateenergy 19.7 170 Banks 162.4 450 Electricutilitiesandfossilfuels 121.3 188 Industryandmaterials 31.1 75 Non-bankfinancials 24.4 47 Realestate 74.9 156 Transportation 20.6 29 Other 35.8 54 Total 490.3 1,169 44

TableA6: Greeniumandtimevariation Yieldspread (1) Green×2014 -10.44 (11.98) Green×2015 -2.525 (6.049) Green×2016 -12.01 (8.898) Green×2017 -4.837 (8.448) Green×2018 3.036 (4.934) Green×2019 -14.40∗∗ (6.193) Green×2020 -9.312∗∗∗ (3.497) Green×2021 -8.302∗∗ (3.799) ✓ Controls Firm×YearFE ✓ Observations 114,879 R2 0.842 AdjustedR2 0.809 RegressionsofcorporateyieldspreadsatissuanceontheGreenbondindicatorvariableinteractedwithannual indicatorvariables.Thesampleperiodcoversgreenandconventionalbondissuesfrom2014to2021(Section 3 details the sample construction). The dependent variable Yield spread refers to a bond’s yield spread on itsdateofissuanceoverthematurity-matchedgovernmentbondyieldforthegivenbond’scurrencyregion. The independent variable Green takes the value of one for green bonds and zero otherwise. 2014 to 2021 areindicatorvariablesthatflagtherespectivecalendaryear. Allothercontrolvariablesandfixedeffectsare detailedinSection3.Standarderrors,reportedinparentheses,areclusteredontheissuerultimateparentand year-monthlevels.***,**,and*indicatestatisticalsignificanceatthe1%,5%,and10%level,respectively. 45