Henley Busness School School of Real Estate & Plannng Workng Papers n Real Estate & Plannng 07/09 The copyrght of each Workng Paper remans wth the author. If you wsh to quote from or cte any Paper please contact the approprate author. In some cases a more recent verson of the paper may have been publshed elsewhere.
New Evdence on the Green Buldng Rent and Prce Premum by Franz Fuerst and Patrck McAllster* Paper presented at the Annual Meetng of the Amercan Real Estate Socety, Monterey, CA, Aprl 3, 2009. Acknowledgement The authors wsh to thank the CoStar Group for provdng the large dataset needed to perform ths analyss. All errors reman our own. *The Unversty of Readng, Henley Busness School, School of Real Estate and Plannng, PO Box 219, Whteknghts, Readng, RG6 6AW, UK, Tel : + 44 (0) 118 931 6657, Fax : + 44 (0) 118 931 8172 e.mal: p.m.mcallster@rdg.ac.uk or f.fuerst@rdg.ac.uk
Abstract Ths paper nvestgates the effect of voluntary eco-certfcaton on the rental and sale prces of US commercal offce propertes. Hedonc and logstc regressons are used to test whether there are rental and sale prce premums for LEED and Energy Star certfed buldngs. The results of the hedonc analyss suggest that there s a rental premum of approxmately 6% for LEED and Energy Star certfcaton. A sale prce premum of approxmately 35% was found for 127 prce observatons nvolvng LEED rated buldngs and 31% for 662 buldngs nvolvng Energy Star rated buldngs. When compared to samples of smlar buldngs dentfed by a bnomal logstc regresson for LEED-certfed buldngs, the exstence of a rent and sales prce premum s confrmed albet wth dfferences regardng the magntude of the premum. Overall, the results of ths study confrm that LEED and Energy Star buldngs exhbt hgher rental rates and sales prces per square foot controllng for a large number of locaton- and property-specfc factors. 2
Introducton The supply of envronmentally responsble goods and servces has experenced consderable growth rates n recent years for a dverse range of products ncludng agrcultural produce, clothng, consumer electroncs and real estate. What the markets for these products have n common s an ncreasng wllngness of customers to pay a premum whch s potentally based on a heghtened awareness of the envronmental mpact of producton and consumpton patterns. Ths trend has been asssted by ndependent thrd party (albet often government sponsored) certfcaton and labelng programs and codes. Indeed, n the real estate sector, eco-labelng has been a central element of a blend of governmental polces and voluntary market change that s attemptng to produce reductons n carbon emssons from the commercal real estate sector. Whle government nterventon can be n the form of new regulatons and standards, fnancal ncentves, leadershp or example and educaton and nformaton, ths paper focuses on the potental role of the voluntary eco-labellng of commercal offces to affect market prces and, consequently, to produce favourable envronmental outcomes. In essence, ths study attempts to measure the revealed preferences of market partcpants who lease and nvest n eco-labeled commercal offces. The framework for ths research s that occuper and nvestor preferences towards exstng stock may change n the lght of growng awareness of envronmental sustanablty and ths change n behavour should manfest tself n the form of prce sgnals (rental and captal values). A key expectaton s that prce sgnals wll emerge through the operaton of markets and result n changes n the relatve allocaton of resources wthn the commercal real estate sector. Hgher relatve rsk-adjusted returns from envronmentally responsble buldngs should provde an ncentve to market partcpants to allocate captal nvestment to ther producton. Buldng on prevous emprcal work, ths paper provdes an emprcal nvestgaton of the rental and sale prce dfferentals between LEED and Energy Star certfed buldngs and noncertfed commercal buldngs n the US. In the analyss, certfed buldngs are compared to a sample of non-certfed buldngs whch were selected to nclude propertes n the same submarket areas as the certfed sample. Two technques are used to test whether there are sgnfcant prce dfferentals. Frstly, rents and prces are related to a set of hedonc characterstcs of the buldngs such as age, locaton, number of stores nter ala. Essentally, our hedonc model measures prce dfferences between certfed buldngs and randomly selected non-certfed buldngs n the same submarkets area controllng for dfferences n lease contract, age, heght, qualty, sub-market etc. We frst estmate rental regressons for a 3
sample of approxmately 200 LEED and 800 Energy Star (the precse number vares slghtly wth model specfcaton) as well as approxmately 10,000 buldngs n the control group. The results of hedonc regresson suggest that there s a rental premum of approxmately 6% for LEED and Energy certfcaton. In terms of LEED level, although the coeffcents have the expected sgns, only the Certfed and Platnum level have a sgnfcant premum. Based on a sample of transacton prces for 662 Energy Star and 127 LEED-certfed buldngs, we fnd prce premums of 31% and 35% respectvely. In tryng to mtgate a potental omtted varable bas n the hedonc framework, we also apply a probablstc logstc regresson approach. Ths provdes a method of selectng buldngs n the control sample that are suffcently smlar to the eco-certfed group. Based on the ndvdual probablty scores obtaned from the logstc regresson, we then defne buldngs for ncluson n a peer group sample. When compared to samples of smlar buldngs dentfed by a bnomal logstc regresson for LEED-certfed buldngs, the exstence of a rent and sales prce premum s confrmed albet wth dfferences regardng the magntude of the premum. Ths paper s organzed as follows. The frst secton provdes background dscusson to the topc focusng on the growth n envronmental certfcaton, the nature of eco-labeled buldngs and prevous research on ther costs and benefts. The man emprcal secton outlnes the data and methods appled n the study followed by a dscusson of the results and consderatons of the valdty of the results as well as possble paths for future research. Background and Context Eco-labelng n Commercal Real Estate Markets Many certfcaton and labelng codes are vewed as contrbutng to a prce-based soluton to promote ncreased provson of envronmental publc goods (Kotchen, 2006). A commonly accepted advantage of voluntary eco-labels s that the prces of sustanable attrbutes can be revealed through the operaton of markets and potental neffcences assocated wth government nterventon and compulsory standards are avoded. In common wth many ecocertfcaton schemes, the major objectve s to provde nformaton to market partcpants about the envronmental effects of the producton and consumpton of products and/or servces. The ams are broadly twofold. The frst s, by provdng nformaton, to encourage a shft towards more envronmentally responsble consumpton. The second s to stmulate producers and other market partcpants to mprove the envronmental standards of products 4
and servces. In partcular, as a form of eco-label, Energy Star allows comparson of the energy effcency of buldngs. Ths allows prospectve buyers and occupers to see nformaton on the energy effcency so they can consder these ssues as part of ther decson on nvestment or occupaton. It can be nferred that polcy makers expect market partcpants to respond to the greater transparency provded by certfcaton. Envsaged (nterlnked) outcomes may nvolve changes n the supply and demand curves for envronmentally responsble buldngs that alter nvestment decsons by owners and occupers and lead to dfferental rental prcng and rental growth (or deprecaton) of new and exstng buldngs lnked to the level of ther carbon emssons. Hence, t s expected that the nteracton of mproved nformaton, market transparency and the prce mechansm wll produce postve envronmental outcomes. Although arguably stll n nascent form, n many real estate markets a blend of voluntary and mandatory eco-labels have emerged n a number of commercal real estate markets. Voluntary envronmental certfcaton systems for buldngs nclude schemes such as Green Star (Australa), LEED (USA), Energy Star (USA), Green Globes (USA), and BREEAM (UK). These schemes supplement mandatory approaches that can specfy mnmum envronmental standards and the provson of nformaton on envronmental effects. Compulsory certfcaton of energy effcency was ntroduced n the European Unon n 2008 followng the EU Energy Performance of Buldngs Drectve and takes the form of Energy Performance Certfcates and Dsplay Energy Certfcates. As noted above, ths paper focuses on LEED and Energy Star certfcaton. The LEED Green Buldng Ratng System, developed by the U.S. Green Buldng Councl, conssts of set of standards for the assessment of envronmentally sustanable constructon. The rates of growth n numbers of 'green' buldngs have been rapd wth numbers doublng nearly every two years. As of Aprl 1, 2009 the CoStar database ndcates that there are 402 LEED rated offce buldngs and 1925 Energy Star rated offce buldngs. In common wth the major regonal certfcaton such as Green Star and BREEAM, the ratng system focuses on sx broad categores related to: sustanablty of locaton, water effcency, energy and atmosphere, materals and resources, ndoor envronmental qualty and nnovaton and desgn process. There are dfferent levels of LEED accredtaton based upon a scorng founded upon the sx major categores lsted above. In LEED v2.2 for new constructon and major renovatons for commercal premses, buldngs may qualfy for four levels of certfcaton. 5
Certfed: 26-32 ponts Slver: 33-38 ponts Gold: 39-51 ponts Platnum: 52-69 ponts For exstng buldngs, the Energy Star scheme provdes a more wdespread scheme for ecocertfcaton as Energy Star-certfed buldngs greatly outnumber those certfed under the USGBC s LEED-EB program for exstng buldngs. The Energy Star scheme nvolves an assessment of buldngs energy performance. Buldngs are awarded a score out of 100. Only buldngs that are n the top quartle are elgble for Energy Star accredtaton. Offce propertes tend to domnate both the LEED and Energy Star n terms of space and numbers (Nelson, 2007). Based on anecdotal evdence, LEED certfcaton may be more prestgous than Energy Star as t provdes a more comprehensve evaluaton of a buldng s envronmental mpact. It s notable that there have been reports of some real estate developers makng fraudulent clams about havng obtaned LEED certfcaton n the early stages of constructon (see Burr, 2009). Ths underlnes the perceved attractveness of the LEED certfcaton scheme. Furthermore, LEED certfcaton s more costly to obtan n terms of fees, encompasses a broader range of sustanable attrbutes and s comparable to other real estate eco-certfcaton schemes n the UK, Germany and Australa. There s an expectaton that premums should vary between Energy Star and LEED certfed buldngs and also wthn the dfferent levels of LEED buldngs. A common mstake made n the dscusson of green buldngs s that the costs and benefts of eco-labelng and the costs and benefts of ncorporatng envronmentally responsble features and practces n buldngs are lumped together, although they should be consdered separately. Regardless of the source (label or actual performance), there are a range of benefts for owners and occupers that have been reported n the majorty of emprcal studes. Moreover, these buldngs may be elgble for the growng array of subsdes and tax benefts 1 that have appeared for eco-labeled buldngs. Occupers can have reduced costs of operatng buldngs (manly assocated wth energy and other utlty savngs), mproved productvty (due to reduced staff turnover, absenteesm nter ala) and other compettve advantages lnked to marketng and mage benefts. As well as potental rental uplfts, nvestors may beneft from reduced holdng costs (due to lower vacancy rates and hgher tenant retenton), reduced operatonal costs (due to energy and 1 A number of US states have ntroduced dfferent types of ncentves to encourage greater supply of certfed buldngs. 6
other utlty savngs), reduced deprecaton (lnked to the use of latest technologes) and reduced regulatory rsks. In turn, surveys of wllngness-to-pay have found that occupers have stated that they are prepared to compensate owners for the addtonal costs of ecocertfed buldngs through hgher rents (see GVA Grmley, 2007 and McGraw Hll Constructon, 2006 for examples). However, t s mportant to dstngush between what occupers and nvestors state that they are ready to pay from what they really pay. Prce formaton n commercal real estate markets A potentally sgnfcant factor nfluencng the exstence and/or the level of a premum for envronmentally responsble buldngs s the prce settng process. In the marketng lterature, there are three broad strateges to prce settng, Estmaton of reservaton prces of occupers through market research technques. Lnk to compettors prcng levels Cost-plus prcng (based on average cost) Whle t may be the case that for many companes, average cost s the ntal, paramount, or only determnant n prce settng (Shpley and Jobber, 2001, 307), ths s unlkely to be the case n the commercal real estate rental market. It s more dffcult to lnk prcng to cost and anecdotal evdence suggests that competton-lnked prcng domnates. In the absence of sutable certfed comparables and clear prcng sgnals, owners may anchor on the prces of non-certfed buldngs. Prcng strategy may also be nfluenced by whether developers tend to adopt prce skmmng or penetraton strateges. Apart from a small number of studes on the emprcs of search theory and the mcrofoundatons of the natural vacancy rate theory, research on the prce settng strateges n commercal real estate markets s relatvely scant. Conventonally, marketng of commercal real estate assets for occupaton s based upon the postng of askng rents that are then subject to negotaton about the detaled terms of the lease contract and rent. Although developers of certfed buldngs may be confdent that potental occupers wll obtan an addtonal consumer surplus relatve to non-certfed buldngs, nformaton about potental occupers wllngness-to-pay may be costly to obtan. In contrast, n many consumer markets, buyers reservaton prces for envronmentally responsble products can be estmated from actual tradng prces. Snce the heterogenety of commercal real estate premses 7
reduces the avalablty of clear market prcng sgnals, owners of certfed buldngs are faced wth dffculty n assessng the reservaton prces of occupers for a relatvely novel product. Sales prces tend to be determned by a smlar process. Typcally, the vendor lsts the property wth brokers and sets an askng prce. Ths askng prce may be lnked to prcng levels observed n recent transactons. Prospectve buyers can then send Letters of Intent (LOIs) that descrbe the general terms (ncludng sale prce) under whch they would buy the property. A Cost Premum? Dependng on the relatonshp between prce and producton cost, the exstence and sze of a cost premum to construct certfed buldngs may be relevant to prce premums. There have been a number of studes of the constructon cost premum assocated wth achevng certfcaton (see, for example, Kats, 2003; Berry, 2007; Morrson Hershfeld, 2005). These studes suggest small constructon cost premums of around 2% on average. To nvestgate the cost premum n more depth, Matthesen and Morrs (2007) analyzed 83 buldng projects wth a prmary goal of LEED certfcaton and 138 smlar buldng projects wthout the goal of sustanable desgn. Confrmng the fndngs of earler studes, they found no sgnfcant dfference n average costs for buldng projects wth a prmary goal of LEED certfcaton as compared to non-certfed buldngs. It s worth notng that the methodologcal dffcultes n measurng the cost premum rase smlar ssues to measurng the prce premum. In partcular, there are dffcultes n dentfyng an approprate benchmark for comparson between the two types of buldng (conventonal and envronmentally responsble). In addton, there s expected to be substantal cross-sectonal varaton between buldng projects. However, the concluson from ths evdence of a largely nsgnfcant cost premum s that, f rental askng prces are formed on a cost plus bass, owners may not be drven to demand rental premums due to cost pressures. Lease Contracts The type of lease contract may also be a factor n the rent determnaton process. It s wellestablshed the allocaton of cost lablty for property taxes, nsurance, mantenance, repars and utlty costs wll have an mpact on the rental level. In research on the prcng of 8
varatons n lease terms, the standard assumpton of lease prcng models s that real estate nvestors wll extract the same value from the property regardless of leases structure (see Grenader, 1995, Booth and Walsh, 2001, Ambrose, Hendershott and Klosek, 2002). In short, nvestors are assumed to be fully compensated by rental adjustments for the costs of provdng benefts to tenants. However, n practce, nsttutonal features of the rent determnaton process may prevent the transmsson of expected prce effects to actual prces. For nstance, researchers have been unable to dentfy emprcally an expected term structure of rents (see Bond, Lozou and McAllster, 2008, Englund, Gunneln, Hoesl and Söderberg, 2003). In ths context, there s an mportant dstncton between lease structures n whch the tenant s responsble for the costs of energy and lease structures where the owner s lable. For the latter, t s expected that tenants wll pay a hgher rent for a leasng on a gross or full servce bass. However, snce an owner of energy effcent buldngs wll ncur relatvely lower energy costs, ceters parbus, assumng prce competton they may be prepared to accept a dscounted (relatve to full servce or gross rental contracts on poor energy effcent buldngs) rental level f leasng on gross or full servce terms. In addton, owners may also prefer to lease on gross or full servce lease terms n order to capture the savngs generated by the lower energy costs. In turn, for tenants on gross or full servce leases, there s no reduced operatng cost ncentve to offer a hgher rent for energy effcent buldngs. However, snce there may be mage benefts assocated wth eco-certfcaton or perceved busness benefts assocated wth the use of less artfcal heatng and coolng technologes, a rental premum may stll be dentfed gross and full servce leases n eco-certfed buldngs. In contrast, for tenants on net leases, snce they are drectly lable for energy costs, there are clear operatonal savngs assocated wth an energy effcent buldng. As a result, t s expected that tenants wll be prepared to make an ncreased rental bd relatve to comparable poor energy effcent buldngs. Assumng that the rent determnaton process s affected by prospectve energy costs, for eco-certfed buldngs wth net lease structures, there should be a hgher rental premum than for eco-certfed buldngs wth full servce or gross leases. As a result, whle we expect to confrm that net leases have a lower rent than gross full servce leases, we expect ths dfferental to narrow n eco-certfed buldngs. Measurng the Prce Dfferental When attemptng to measure a prce dfferental between a certfed and non-certfed product, there are two key methodologcal and data ssues. The frst problem s to dentfy approprate 9
benchmarks to compare certfed and non-certfed products. The second s to dentfy suffcent market prce nformaton from transacton actvty to measure prce dfferentals. Wth regard to the frst ssue, commercal real estate markets present some problems. In some product markets, apart from the certfcaton label, eco-frendly goods may be ndstngushable from conventonal goods e.g. some tmber or food commodtes. As a result, t s often straghtforward to dentfy a sutable benchmark aganst whch to measure a prce dfferental. Addtonally, assumng actve tradng, t s also lkely that there wll be adequate prce nformaton. In contrast, n markets where products are bespoke (such as commercal real estate), the constructon and desgn requrements of obtanng certfcaton may add to nherent product heterogenety. Further, thn tradng and low market transparency may reduce the amount and qualty of avalable prce nformaton. The result s that measurng the prce dfferental for eco-certfed buldngs s hndered by the combnaton of the lack of an approprate benchmark and lmted prce nformaton due to thn market effects. Snce markets are dynamc, t s also expected that levels of prce dfferental wll tend to change over tme. In a statc equlbrum framework, the prce dfferental at a gven pont n tme can be analysed as the product of dfferent demand and supply curves for certfed and non-certfed buldngs. However, snce supply and demand curves tend to be dynamc (e.g. due to the effects of changes n market penetraton and producton costs) prces adjust. Even n the short term, supply and demand elastctes wll not be statc for certfed and noncertfed buldngs and, as a result, prce dfferentals should vary over tme. Furthermore, lnked to the nomnaton problem nherent to real estate development, developer responses to a prce dfferental wll themselves shft the supply curve and affect the level of prce dfferental. There are also strong grounds to expect levels of prce dfferental to vary cross-sectonally. Certfed buldngs are obvously not homogenous. For nstance, as noted above, there are dfferent levels of certfcaton. As a result, there are lkely to be varatons between certfed buldngs n the levels of the potental benefts (reduced costs of occupancy, mage and busness performance) that may be obtaned by occupers. In turn, there may be varatons between buldngs n the prce effects of certfcaton. A further aspect of the prcng ssue s that certfcaton standards themselves tend to be dynamc usually (albet not exclusvely) n an upwardly drecton. For nstance, t s commonly suggested that voluntary government sponsored eco-labellng standards provde a 10
sgnal to market partcpants of the polcy drecton for future mandatory standards. In addton, where a certfcate s awarded for relatve envronmental performance compared to the exstng buldng stock e.g. Energy Star, over tme ths performance may deterorate n relatve terms as standards mprove as exstng stock s upgraded and new stock s developed. In smple terms, a buldng that was n the top 25% by energy performance n 1999 may not be elgble for an Energy Star ratng n 2009. We are not aware that there s any process for wthdrawng certfcaton as the buldng populaton changes. Related emprcal studes There have been few studes have attempted to measure the prce effects of green buldng ratng. Studes that have dentfed hgher rents and mproved returns based on the vews and experences of expert professons stll requre emprcal verfcaton. Whle recognzng the centralty of prcng to adopton, recent revews of the lterature have found lttle convncng research that dentfed a certfcaton premum (see Berry, 2007). Nelson (2007) examned the performance dfferences between certfed and non-certfed buldngs usng a number of crtera. Drawng upon the CoStar database, the study compared LEED rated buldngs and Energy Star buldngs wth a vastly larger sample of non-certfed buldngs n the CoStar database. Whle acknowledgng the sgnfcant dfferences between the sample and the wder populaton, t found that certfed buldngs tended to be newer, owner-occuped or sngle tenanted, concentrated geographcally and sectorally (n the offce sector). Recognzng that t dd not control for these dfferences, the study dentfed lower vacancy rates and hgher rents n LEED-rated buldngs. There have been a group of studes that draw upon the CoStar database of US propertes to dentfy the effect of envronmental certfcaton on sale prces and rents respectvely. The most wdely quoted among these was conducted by Mller et al (2008). To control for dfferences between ther sample of certfed buldngs (927 buldngs) and a much larger sample of non-certfed buldngs, the authors nclude a number of control varables such as sze, locaton and age n ther hedonc regresson framework. They fnd that dummy varables for Energy Star and LEED ratngs show rent prema of 6% and 10% respectvely but these results are not sgnfcant at the 5 percent level. 11
Wley, Benefeld and Johnson (2008) focused on the effect on rent, occupancy rate and sale prce of eco-certfcaton for Class A buldngs n 46 offce markets across the USA 2. Usng an hedonc prcng approach, they found rental premums rangng from approxmately 15-18% for LEED certfed buldngs and 7-9% for Energy Star certfed buldngs dependng on the model specfcaton. In terms of sales transactons, they estmated premums of $130 per sq ft for LEED certfed buldngs and $30 for Energy Star. However, although plausble, these results need to be treated wth some cauton. A lmtaton of ther hedonc model s that t lacks controls for mcro-locatonal effects. In essence, they dentfy rental and sale premums for certfed buldngs relatve to non-certfed buldngs n the same metropoltan area. However, f certfed buldngs tend to be more lkely to be found n better qualty locatons wthn a metropoltan area, observed premums may nclude a locaton as well as a certfcaton premum. In a recent workng paper, Echholtz, Kok and Qugley (2009) also used an hedonc framework to test for the effect of certfcaton on the contract rents of 694 offce buldngs. Usng GIS technques, they control for locaton effects by dentfyng other offce buldngs n the CoStar database wthn a radus of 0.25 mles of each certfed buldng. They dentfy a statstcally sgnfcant rent premum on rents per square foot of 3.3% for Energy Star certfed buldngs. Surprsngly, they fnd no sgnfcant rent premum for LEED-certfed buldngs. However, when they used effectve rents to reflect dfferent vacancy rates n certfed buldngs, the premum ncreased to around 10% for Energy Star certfed buldngs and 9% for LEED-certfed buldngs 3. Smlar results were found for transacton prces. Although not dscussed n the paper, they found a substantal 19% sale prce premum for Energy Star certfed buldngs but no statstcally sgnfcant premum for LEED-certfed buldngs. If they are confrmed, these fndngs have substantal mplcatons for developers who are consderng LEED certfcaton and t s mportant that they are corroborated by other studes. In ths paper, we apply a smlar hedonc methodology to Echholtz et al to a smlar data set. However, nstead of usng the askng rent multpled by the occupancy rate (termed effectve rent by the authors), we use the rental rates reported by CoStar to solate the effect of certfcaton on rent only. A further advantage of dong so s that we do not have to address the endogenety and complex nteracton of rents and vacancy rates n a buldng. More substantvely, we control for locaton effects usng actual submarkets (as defned by CoStar) rather than possbly arbtrary submarkets. 2 Sales data were avalable for 26 offce markets. 3 Echholtz et al also fnd that there s a hgher relatve premum for cheaper locatons. However, ths s lkely to be due to the fact that smlar absolute premums due, for example, to lower energy costs wll nvarably result n hgher relatve premums n less expensve locatons. 12
Whle there are clearly plausble a pror reasons to expect prce dfferences between certfed and non-certfed buldngs, ths s not necessarly certan. As noted below, prevous research has shown that not all varatons n asset attrbutes are necessarly reflected n asset prces (see, for example, Wheaton, 1984). Conversely, the lack of a sgnfcant premum would not necessarly preclude the exstence of consderable benefts of certfed buldngs as descrbed n the prevous secton. Models and data In the emprcal test of rent and sales prce prema of certfed buldngs, we apply a two-stage approach. Frstly, we adopt a standard hedonc framework. In the second stage, we select a matched peer group based on logstc regresson and compare the results obtaned from both types of estmates. Hedonc analyss Hedonc regresson modelng s the standard methodology for examnng prce determnants n real estate research. Ths method s used here prmarly to measure the prce effect of LEED and Energy Star certfcaton. Rosen (1974) frst generalzed that the hedonc prce functon coverng any good or servce conssted of a varety of utlty-bearng characterstcs. In offce rent determnaton lterature, hedonc modelng typcally specfes that a range of physcal, locatonal and lease characterstcs be used as the ndependent varables determnng prce. As descrbed n the lterature revew secton of ths paper, a crtcal ssue n measurng the prce effect of eco-certfcaton s to control for the fact that certfed buldngs may be newer, hgher or located n more attractve locatons or markets. The standard log-lnear hedonc rent model takes the followng form: ln R x Z (1) where R s the natural log of average rent per square foot n a gven buldng, x s a vector of the natural log of several explanatory locatonal and physcal characterstcs, β and ϕ are the respectve vectors of parameters to be estmated. Z s a vector of tme-related varables and s a random error and stochastc dsturbance term that s expected to take the form of a normal dstrbuton wth a mean of zero and a varance of σ 2. The hedonc weghts assgned to each 13
varable are equvalent to ths characterstc s overall contrbuton to the rental prce (Rosen 1974). For the purpose of ths study, we specfy two types of hedonc models. The frst type explans rents and the second explans prce per square foot n sales transactons. Hedonc Rent Model ln R 9 GR 0 1 ln A 2 ln S 3 ln L 4 lnt 5 ln G 6 N 7 BC 8 SU (2) In ths model, A represents the age of the property, measured from the year of constructon or the year of a major refurbshment (whchever occurred more recently), S s the number of stores of the property, L represents the lot sze, T and G are the lattude and longtude geographc coordnates of the property whch capture any large-scale effects of the spatal dstrbuton of propertes across the country, N s a dummy varable ndcatng a net lease (takng the value of zero for a gross or full-servce lease), BC are controls for buldng class (standard categores A,B,C and F) and SU are controls for submarkets (853 n total) and ε s the error term whch s assumed to be ndependent across observatons and normally dstrbuted wth constant varance and a mean of zero. A rent premum for LEED and/or Energy Star rated buldngs s captured by the GR term, a dchotomous varable that takes the value of 1 for certfed buldngs and a value of 0 otherwse. In alternatve model specfcatons, the GR dummy varable s replaced by separate terms for LEED and Energy Star certfcaton (Model 2) and level of LEED certfcaton (Model 3). Hedonc Sales Transacton Prce Model: Smlarly, the regresson for estmatng prce per square foot n sales transactons s estmated n the followng way: ln R 9 GR 0 1 ln A 2 ln S 3 ln L 4 lnt 5 ln G 6 E 6 MC 7 BC 8 SU (3) 14
where E s a tme trend varable whch accounts for general prce nflaton and other unobserved trends over tme. Ths varable ncreases n sem-annual ncrements. Beyond ths control for the overall trend, we also ncluded E, whch ndcates market condtons at the tme of sale proxed by the average quarterly return of the NAREIT ndex. All other varables are the same as n the rent model. The type of specfcaton used n the rent and transacton prce models allows us to detect dfferences n the weght of parameter estmates across submarkets, buldng class categores and market condtons by estmatng separate ntercepts. Ths Least Squares Dummy Varable (LSDV) approach has the advantage of controllng for a number of omtted varables, for example small-scale spatal effects at the submarket level that we could not model explctly as the data necessary to do ths were not avalable to us. The LSDV approach allows ntercepts of the regresson to dffer across markets whle assumng constant varable coeffcents. Ths s mportant not only because of the dfference n prce levels across markets but also because t controls for tax and other ncentves that several states and ctes grant for buldngs that are certfed ncludng tax credts, reduced permttng fees and property tax abatements (Roberts, 2007). Bnomal logstc regresson Whle the hedonc regresson approach s the prncpal method for determnng rent and prce premum snce t enables the researcher to control for a host of relevant buldng characterstcs, t s subject to a potentally serous methodologcal problem. If buldngs n the treatment group (eco-certfed buldngs n our case) and the control group (non-ecocertfed buldngs) dffer systematcally wth respect to characterstcs that are sgnfcant factors n rent formaton, the hedonc model wll not attrbute the prce effect of ndvdual factors accurately and the model as such s subject to omtted varable bas. Ths problem may arse because of unmeasured common features of eco-certfed buldngs, for example certan mcro-locatonal characterstcs that are not entered as ndependent varables n the hedonc model. Therefore, we complement the hedonc analyss wth a logstc regresson framework whch serves as a bass for selectng buldngs n the control sample that are suffcently smlar to the eco-certfed group. Based on the ndvdual probablty scores obtaned from the logstc regresson, we can then defne a cutoff pont for ncluson n the peer group sample. Our logstc model assumes a dchotomous dependent varable whch measures the probablty π of beng an eco-certfed buldng as 15
exp( ) (4) 1 exp( ) Thus, we can determne a lkelhood functon lf for n observatons y 1,...,y n, wth probabltes π 1,..., π n and case weghts w 1,...,w n, can be expressed as n 1 In the logarthmc form used n our paper, the full model L s thus: L ln(lf) n 1 w y (1 y ) lf (1- ) (5) w y ln( ) w (1- y )ln(1- ) (6) Havng obtaned probablty values for each observaton n our database, we can then proceed to re-defne our sample of comparable buldngs by ncludng only those non-certfed buldngs that suffcently resemble certfed buldngs based on the features ncluded n the logstc regresson model. Data The taggng of LEED and Energy Star buldngs by CoStar enables researchers to dentfy numbers and types of eco-certfed buldngs n the database. Gven the dscusson above, a key ssue s the benchmark aganst whch the sample of certfed buldngs can be compared. Our benchmark sample conssts of approxmately 24,479 offce buldngs n 853 submarkets n 81 metropoltan areas spread throughout the Unted States. Ths means that our hedonc model s measurng prce dfferences between certfed buldngs and randomly selected noncertfed buldngs n the same metropoltan area controllng for dfferences n age, sze, heght, locaton, lease type, buldng class and submarket. In the frst step, we drew detals of approxmately 1,900 eco-certfed buldngs of whch 626 were LEED certfed and 1,282 were Energy Star. In the second step, buldngs were selected n the same metropoltan areas and submarket as the certfed sample. Sample selecton was based on the crtera a) same submarket or market as certfed buldngs and b) at least 10 comparable observatons for each certfed buldng n the database. Although the market weghtngs may be dfferent between the benchmark and the certfed samples, our regresson 16
model controls for market-specfc effects. Of the LEED buldngs, 31% (n=192) are certfcaton-level, 29% (n=180) are Slver, 32% (n=201) are Gold and 7% (n=45) are Platnum level. In total, we have used 9,806 observatons of transacton prces and 18,519 (askng) rent observatons. Results Descrptve Statstcs The descrptve statstcs are dsplayed n Exhbt 1. There are clearly some dfferences between eco-certfed and non-certfed buldngs. The former tend to be newer. In partcular, the medan age of LEED certfed buldngs s fve years. The comparable fgure for the benchmark sample s 23. Whle there s relatvely lttle dfference between buldngs wth Energy Star certfcaton and the benchmark sample n terms of age, the former tend to be domnated by tall buldngs suggestng that they are manly located n CBD locatons. Ths s supported by the fact that Energy Star buldngs tend to be on average nearly 20 tmes larger than non-certfed buldngs. Wthout controllng for the dfferences between the samples, certfed buldngs have hgher askng rents and lower vacancy rates than noncertfed buldngs. Medan askng rents are approxmately 35% hgher n LEED and Energy Star certfed buldngs. There are also some notable dfferences n terms of the proportons of each sample that are on trple let leases compared to gross or full servce leases. Energy Star buldngs have 12% and LEED buldngs have 10% on net leases. The comparable fgures for the control sample s 22%. More thorough nvestgaton s requred, however, to nfer a general prevalence of gross leases n certfed buldngs as the hgher share may smply be reflectve of dfferences n property types (partcularly mono- vs. mult-tenanted propertes) between the certfed and the non-certfed samples. 17
ln R 0 1 lny 2 lno 3 ln S 4 ln L 5 ln F GR t Exhbt 1: Descrptve statstcs of overall sample wth LEED and Energy Star sample Overall RENT $ psf PRICE $ psf % LEASED SIZE (sq ft) STORIES AGE Mean 19.50 141.19 63.82 52,771 3.32 28.37 Medan 18.00 113.81 79.80 10,800 2.00 23.00 Std. Dev. 9.16 112.50 38.87 145,147 5.75 27.48 Skewness 2.40 1.77-0.69 7.57 5.92 1.97 Kurtoss 14.47 8.77 1.88 92,807 50.21 8.42 Observatons 16,488 9,120 24,951 16,488 24,479 21,147 LEED RENT PRICE % LEASED SIZE (sq ft) STORIES AGE $ psf $ psf Mean 26.74 251.12 91.06 179,290 6.45 11.77 Medan 24.50 247.41 100.00 95,000 4.00 5.00 Std. Dev. 11.00 136.33 22.46 262,071 8.50 19..06 Skewness 1.79 0.37-2.91 4.68 3.13 3.30 Kurtoss 7.21 3.32 10.78 43.49 13.76 14.89 Observatons 210 127 667 667 622 504 Energy Star RENT PRICE % LEASED SIZE (sq ft) STORIES AGE $ psf $ psf Mean 27.76 254.93 91.43 315,052 13.40 19.43 Medan 25.04 230.88 95.76 217,082 9.00 20.00 Std. Dev. 11.37 138.20 12.44 301,264 12.89 12.77 Skewness 1.66 1.47-3.06 1.99 1.62 2.31 Kurtoss 7.21 6.57 17.78 7.59 5.55 13.66 Observatons 990 662 1,480 1,480 1,453 1,474 Hedonc regresson results and the rent premum To further nvestgate the hypothess of a rent and prce premum for certfed buldngs, we estmate hedonc regressons as outlned above. Two separate regressons are estmated to model rent and transacton prce separately. All contnuous numerc varables were transformed to log values to (1) reduce non-normalty found n ntal examnatons of the dataset, (2) to reduce heteroskedastcty and (3) to be able to nterpret the results as elastctes. The results are summarzed n Exhbts 2 and 3. When controllng for the most mportant rent determnants such as age, heght, sze and submarket locaton, we fnd a statstcally sgnfcant rent premum of 6% n eco-certfed buldngs compared to non-certfed buldngs n the same sub-market area. The control varables used n the regresson show the expected sgns. Ths regresson explans approxmately 60% of the cross-sectonal varaton n rents n the entre sample. 18
Model 2 shows the results of the regresson wth separate dchotomous varables for LEED and Energy Star certfcaton. Both types of certfcaton are found to exert a postve and sgnfcant mpact on rents. Whle the premum for LEED s hgher as expected, there s very lttle dfference between the premums for LEED and Energy Star buldngs. A further common assumpton that we set out to test s that the rent premum of LEED buldngs s ncreasng wth the level of certfcaton. Model 3 n Exhbt 2 reports the estmaton results wth a LEED level varable. In ths specfcaton, the dchotomous LEED varable s modfed to reflect the certfcaton standard,.e. Certfed, Slver, Gold and Platnum. Whle the coeffcents have the expected sgns, only the Gold level s sgnfcant. 19
Exhbt 2 Results from hedonc model estmaton of rental rates Model 1 Model 2 Model 3 Dependent varable Rent psf (log) Rent psf (log) Rent psf (log) Constant 3.61*** 3.81*** 3.80*** Eco-certfed 0.06*** LEED 0.06** Certfed 0.09** Slver 0.04 Gold 0.04 Platnum 0.16*** Energy Star 0.06*** 0.06*** Net Lease -0.11*** -0.11*** -0.11*** No. of stores (log) 0.06*** 0.06*** 0.06*** Sze square feet (log) 0.02*** -0.01*** -0.01*** Ste area (log) 0.01* 0.00 0.00 Age (log) 3-6 years -0.06*** -0.06*** -0.06*** 7-10 years -0.12*** -0.12*** -0.12*** 11-19 years -0.14*** -0.14*** -0.14*** 20-23 years -0.16*** -0.16*** -0.16*** 23-26 years -0.18*** -0.18*** -0.18*** 27-31 years -0.19*** -0.19*** -0.19*** 32-42 years -0.20*** -0.20*** -0.20*** 43-62 years -0.23*** -0.24*** -0.24*** >62 years -0.23*** -0.23*** -0.23*** Longtude (log) -0.01*** -0.01*** -0.01*** Lattude (log) -0.43*** 0.45** 0.45** Class A 0.20*** 0.20*** 0.20*** Class B 0.09*** 0.09*** 0.09*** Adjusted R-squared 0.61 0.61 0.61 F test 26.44*** 26.40*** 26.28*** Included observatons 10,970 10,970 10,969 *** - sgnfcant at 1% level ** - sgnfcant at 5% level * - sgnfcant at 10% level 20
Although t s not a central part of the study, t s nterestng to compare the results of the control factors wth the fndngs of other studes of offce rent determnants. Gven a varaton n data sources and model specfcatons, prevous studes do not always provde consstent fndngs on the relatonshp between varables such as age, sze and heght nter ala and offce rents/prces. As expected, we fnd that the coeffcent for the age varable s negatve. In addton, consstent wth prevous research (for example, see Bollnger, Ihlanfeldt and Bowes, 1998; Shlton and Zaccara, 1994), we fnd that there s a sgnfcantly postve relatonshp between heght and rent. We also fnd a negatve relatonshp between sze and rent. In common wth Laverne and Wnson-Gedeman (2004), we fnd a negatve relatonshp between trple net leases and rental level. Hedonc regresson results and the prce premum Exhbt 3 reports the results of the hedonc regressons wth sales prce per square foot as the dependent varable. Three separate models were estmated wth the smlar ndependent varables. All models dsplay smlar results and have smlar explanatory power. The explanatory power s lower relatve to the regressons for the sample of rents. For most of the ndependent varables, the coeffcents have the expected sgns. Of the addtonal four varables to control for rsng and fallng market condtons, droppng the varable for strong postve growth all exhbt the expected sgn and are sgnfcant (confrmng the lnk between drect and ndrect real estate returns). Model 1 suggests a sales premum of 36% for ecocertfed buldngs. In Model 2, we dstngush between LEED and Energy Star and fnd premums of 35% and 31% respectvely 4. When we break down the LEED sample nto ts varous categores, premums are sgnfcant for each ndvdual category. Whlst the premum for platnum-rated buldngs may seem unrealstc, t s based on the sales of eght buldngs. The results suggest a much hgher relatve sales prce premum compared to rental prce premums. There are a number of potental explanatons. A possble explanaton may le n the combned effects of nvestors expectatons of hgher rental ncome, lower operatng costs, hgher occupancy rates, mage benefts to nvestors and a lower rsk premum. 4 The larger average premum of 30% for eco-certfed buldngs compared to LEED and Energy Star prema s due to the exstence of a number of buldngs that hold both types of certfcaton. For these buldngs, the rental premum wll effectvely be splt between the LEED and Energy Star coeffcents resultng n a lower premum compared to the overall eco-certfed varable. 21
Exhbt 3 Results from hedonc model estmaton of sales prces Model 1 Model 2 Model 3 Dependent varable Sale prce psf (log) Sale prce psf (log) Sale prce psf (log) Constant -0.02 0.09 0.08 Eco-certfed 0.36*** LEED 0.35*** Certfed 0.20*** Slver 0.38*** Gold 0.36*** Platnum 1.00*** Energy Star 0.31*** 0.31*** No. of stores (log) 0.15*** 0.15*** 0.15*** Sze square feet (log) -0.23*** -0.23*** -0.23*** Ste area (log) 0.09*** 0.09*** 0.09*** Age (log) 3-6 years 0.15*** 0.15*** 0.16*** 7-10 years 0.50*** 0.51*** 0.51*** 11-19 years 0.45*** 0.50*** 0.46*** 20-23 years 0.40*** 0.45*** 0.41*** 23-26 years 0.38*** 0.40*** 0.38*** 27-31 years 0.37*** 0.37*** 0.38*** 32-42 years 0.27*** 0.37*** 0.28*** 43-62 years 0.27*** 0.27*** 0.28*** >62 years 0.28*** 0.27*** 0.30*** Longtude (log) -0.02*** -0.02*** -0.02*** Lattude (log) 1.03** 1.06** 1.04** Class A 0.42*** 0.43*** 0.43*** Class B 0.06*** 0.06*** 0.06*** Tme trend varable 0.03*** 0.03*** 0.03*** Moderate postve market -0.08*** -0.08*** -0.08*** Moderate negatve market -0.10*** -0.11*** -0.11*** Strong negatve market -0.10*** -0.10*** -0.10*** SUBMARKET CONTROLS Adjusted R-squared 0.42 0.42 0.42 F test 8.96*** 8.92*** 8.89*** Included observatons 6,158 6,158 6,157 *** - sgnfcant at 1% level ** - sgnfcant at 5% level * - sgnfcant at 10% level 22
Logstc regresson results Our bnomal logstc regresson model ncludes as covarates the contnuous varables buldng age, land area, rentable buldng area, stores, occupancy rate as well as the categorcal varables buldng class and submarket. The Nagelkerke R squared for ths estmaton s 0.33. Further estmatons wth more varables dd not mprove the model results sgnfcantly and are therefore not reported. Havng obtaned probablty values for each buldng n our sample of approxmately 15,700 offce buldngs wth vald observatons, we select a smaller peer sample of matched buldngs wth a calculated probablty value of at least 18 percent of belongng n the eco-certfed group (wthout actually beng an eco-certfed). The cut-off value was defned to represent above-average probabltes, thus all buldngs wth probablty values above the arthmetc mean were ncluded. Whle a probablty value of 18 percent may appear low n overall terms, t s mportant to keep n mnd that the vast majorty of buldngs n our sample exhbt very low or zero probabltes. Put dfferently, a buldng wth a 18 percent probablty value s wthn the top 6% of all buldngs n terms of probablty (ncludng 'true' eco-certfed buldngs). It appears thus justfed to set a relatvely low probablty value. Exhbt 4 reports the mean and medan for key values for both the eco-certfed and the matched peer sample. Compared to the large dfferences between eco-certfed and the overall control sample shown n Exhbt 4, these two groups show a much greater degree of smlarty of buldng attrbutes. More mportantly, average rental rates are $25.47/sq.ft (medan value of $22.75/sq.ft) n the matched peer sample and $26.38/sq.ft. (medan of $24.50/sq.ft.) for the LEED sample. Ths corresponds to a 3.7% average rental premum (7.7% for medan values). Smlarly, sales prces for the matched peer sample are $206/sq.ft (medan value of $163/sq.ft) and $247/sq.ft (medan value of $240/sq.ft). Ths translates nto a 19.6% sales premum for LEED buldngs (47% based on medan values). Whle ths appears to corroborate the exstence of a rent and sales prce premum detected by the hedonc regresson, a caveat s n order regardng the nterpretaton of these results. Summary measures such as the arthmetc mean and the medan are very basc ndcators of the dfferences of two dstrbutons. More sophstcated tests and measures are requred to confrm the exstence of a premum. 23
Exhbt 4: Comparson of key varables based on bnomal logstc regresson Overall RENT $ psf PRICE $ psf % LEASED SIZE (sq ft) STORIES AGE Mean 19.50 141.19 63.82 52,771 3.32 28.37 Medan 18.00 113.81 79.80 10,800 2.00 23.00 Std. Dev. 9.16 112.50 38.87 145,147 5.75 27.48 Observatons 16,488 9,120 24,951 16,488 24,479 21,147 Peer sample RENT $ psf PRICE $ psf % LEASED SIZE (sq ft) STORIES AGE Mean 25.47 206.45 91.00 261,611 10.22 14.08 Medan 22.75 163.19 98.00 125,645 5.00 11.00 Std. Dev. 11.29 165.13 15.25 371,038 12.05 10.54 Observatons 447 387 710 710 710 710 LEED RENT PRICE % LEASED SIZE (sq ft) STORIES AGE $ psf $ psf Mean 26.39 247.07 90.89 176,080 6.39 12.14 Medan 24.50 240.00 100.00 94,945 4.00 5.00 Std. Dev. 10.34 137.85 22.95 25,882 8.22 19.46 Observatons 197 127 626 626 581 469 24
Conclusons In many product markets eco-labelng s a common method of sgnalng superor envronmental performance. The growth n compulsory and voluntary eco-labelng of commercal real estate assets s part of a number of regulatory and voluntary ntatves to reduce carbon emssons from the commercal buldng stock. The central am of eco-labelng s to alter the behavor of market partcpants through the provson of mproved nformaton about the envronmental effects of ther real estate decsons. It s antcpated that better nformed captal provders and occupers wll shft towards more envronmentally responsble producton, nvestment and use. A key expectaton s that the prce mechansm wll create prce dfferentals lnked to envronmental performance and result n the allocaton of more captal to envronmentally benefcal nvestment. There are strong a pror grounds to expect prce dfferentals between eco-labeled and noneco-labeled offces. It s generally accepted that there are benefts assocated wth envronmentally responsble buldngs. For occupers, there wll often be reductons n operatonal costs assocated wth occupyng the buldngs, mprovements n productvty of the occuper s busness and mage benefts for the occuper. For nvestors, there may be hgher Net Operatng Incomes due ncreased demand from occupers, lower vod rates, lower costs of ownershp and an element of protecton from future regulatory changes. The results of the emprcal analyss confrm these expectatons. The hedonc regressons suggest that there s a rental premum of approxmately 6% for LEED certfcaton and 5% for Energy Star. The type of lease contract may be a very mportant ssue here. It has been argued that, for propertes on gross or full servce leases, occupers have less ncentve to pay a rental premum snce they do not beneft from reduced operatng costs. It s notable that a much larger proporton of Energy Star certfed buldngs have gross or full servce leases. For sales prces, we fnd prce premums of 31% for Energy Star and 35% for LEED certfed buldngs. The latter fndngs, n partcular, are strkng and need to be confrmed. In order to control for the possblty of an omtted varable bas, we also used a logstc regresson. Ths ndcated that, compared to smlar buldngs, LEED buldngs had premums of 3.7% (7.7% based on medan values).and 19.6% on sales prces (medan premum of 47%). The larger medan premums ndcate that LEED buldngs have hgher rents and sales prces (ceters parbus) even n the lower-prced half of the market. There are a number of mportant ssues that suggest that emprcal studes of typcal prce dfferentals need to be nterpreted wth care. Frstly, as we have seen there are sgnfcant 25