The Effect of HVTLs on Property Values: An Event Study

Effect of HVTLs on Property Value 1 INTERNATIONAL REAL ESTATE REVI The Effect of HVTLs on Property Values: An Event Study Charles Thomas Black Swan Realty Advsors, 14 Monarch Bay Plaza, Sute 293, Monarch Beach, CA 92629. Gerd Welke * Department of Fnance, Real Estate and Law, Calforna State Polytechnc Unversty, Pomona, 3801 West Temple Avenue, Pomona, CA 91768. Emal: gmwelke@cpp.edu We present emprcal results for the eff ect on resdental property values from the constructon of hgh voltage transmsson towers on an exstng utlty rght of way. The event wndow conssts of a perod of two and a half years durng whch non-operatonal 500 kv towers were n place, after whch they were taken down. When comparng the proxmate transactons wthn the wndow to those outsde, relatve to the same dff erence for dstant transactons, we fnd a negatve eff ect on valuaton durng the presence of the towers, whch s most sgnfcant for encumbered and abuttng propertes, and vanshes quckly wth dstance from the rght of way or nearest tower. An event wndow that corresponds to wdespread publc knowledge of the pendng constructon leads to an nsgnfcant effect, whch suggests that the prce formaton process s possbly neffcent. Keywords Hedonc Prcng, Hgh Voltage Transmsson Lnes, Event Study, Market Effcency * Correspondng author

2 Thomas and Welke 1. Introducton Begnnng January 2011, sxteen 200 ft 500 kv non-operatonal double-crcut transmsson towers were constructed on an exstng 150 ft wde, 4 mle long utlty rght of way (ROW) through a resdental part of the Cty of Chno Hlls, Calforna. Before ths date, there were approxmately ten 70-100 ft 220 kv towers n place that had last been used 40 years earler. In September 2013, the new 500 kv towers were removed and the transmsson lnes were placed underground. Ths set of crcumstances allows us to revst the queston of the mpact of upgraded transmsson lnes on property values n a hedonc regresson analyss by usng the event wndow to control for non-hgh voltage transmsson lne (HVTL) local market condtons. Furthermore, snce court testmony makes t reasonably clear when resdents became wdely aware of the constructon plans, we use a second wndow to test for prce mpact before the new towers were n place and examne the effcency of the prce formaton process at an hedonc level. Extremely hgh voltage transmsson lnes are generally consdered to be a negatve amenty snce they are vsually not appealng, cause nose polluton (both aeolan and corona nose), and may be perceved as a health threat. The focus of ths study s on a partcularly onerous confguraton n that the new sxarm towers were very tall and the ROW s relatvely narrow. Studes suggest that a prce effect dmnshes wth tme, and the smaller preexstng towers let us reexamne ths fndng. The nstallaton of the new towers, on the other hand, should have maxmal mpact, although they, lke the old towers, were not operatonal. To our knowledge, ths s the frst regresson event study for transmsson lne mpact on resdental propertes. 1 Frst, we fnd no margnal effects from the extant, non-operatonal 220 kv towers. Second, double dfferences show a sgnfcant negatve mpact on prce when the 500 kv towers were present for propertes that abut or are encumbered by the ROW. The effect dmnshes rapdly wth dstance from the easement or nearest tower, whch ponts to the mportance of the nterplay between proxmty measures. Thrd, we fnd that prce and transacton volume are unaffected durng a precedng wndow that corresponds to concrete and wdely known plans for the new tower confguraton. Ths suggests that prce formaton s neffcent n our sample (Smons and Sagnor 2006). We revew the exstng lterature n the next secton, followed by a dscusson of the model and data. The results are presented n Secton 5, and Secton 6 concludes. 1 There have been some apprasal studes for new and upgraded transmsson lnes. See Kroll and Prestly (1991).

Effect of HVTLs on Property Value 3 2. Lterature Revew There s rch emprcal lterature on the mpact of negatve amentes on the value of resdental real estate. Smons and Sagnor (2006) revew 75 artcles and case studes of loss n value that stems from a wde range of ndvdual contamnants, ncludng power lnes, and combne these nto a meta-data set wth added covarates, such as study type, contamnaton source and dstance, geographc regon, market condtons, and rural settng. They fnd that regresson studes show systematcally lower losses n value compared to survey results, whch suggest that revealed preference s weaker than stated preference. Of partcular nterest to our study, they conclude that the announcement of a bad thng does not have a sgnfcant effect on value, whle nstead an announcement of a closng [of a negatve amenty] s sgnfcantly postve. Turnng to transmsson lnes n a resdental settng, several comprehensve revews (Knnard and Dckey 1995, Kroll and Prestly 1991, Jackson and Ptts 2010) break the emprcal lterature down nto three categores: surveys, regresson analyses, and a catch-all category that ncludes tradtonal apprasal technques, such as pared sales, case-based, and sale-resale studes. Ths latter group also ncludes studes of court rulngs (Bryant and Epley 1998, and Rkon 1996) and an analyss of perceved health rsks (Prestly and Evans 1990). Survey research generally ndcates that whle homeowners mght consder the power lnes a health rsk and an eyesore (Kung and Seagle 1992, Prestly and Evans 1996), they do not vew them as havng a sgnfcant mpact on property values (Knnard 1967, Solum 1985, Kung and Seagle 1992). The work of Prestly and Evans (1996) s of partcular nterest here snce t nvolved a survey of resdents after an upgrade to an exstng transmsson lne. Respondents that had lved n the neghborhood before the upgrade vewed t n the most negatve lght, consstent wth the results that ndcate prce mpact dmnshes over tme (Ignelz and Prestly 1991). Chalmers and Voorvaart (2009) summarze the survey lterature as beng consstent and pont out a puzzle from the general fndngs: most buyers knew of the power lnes at the tme of purchase, and a majorty of the owners had negatve feelngs about them, but most of these n turn stated that the transmsson lnes dd not factor nto ther decson to buy the property or the prce that they pad. In contrast, a survey of apprasers by Delaney and Tmmons (1992) revealed an average estmate of the negatve mpact on value to be 10%, wth only 10% and 6% of the respondents belevng there to be no effect or a premum, respectvely. The second strand of the emprcal lterature deals wth regressons that predct prce based on hedonc characterstcs, ncludng terms for the level of negatve amenty, such as encumbrance, adjacency, or dstance to an easement or a tower. Generally, these studes fnd a small effect, wth the loss n value varyng from no effect, to 2 10% (Kroll and Prestly 1991, Jackson and Ptts 2010). Colwell and Foley (1979), and Hamlton and Schwann (1995) fnd a sgnfcant

4 Thomas and Welke prce mpact close to transmsson lnes, but t dsappears rapdly wth dstance. Colwell (1990) consders encumbered property and adds dstance to towers as an explanatory varable to fnd smlar results and that the mpact tends to dmnsh over tme (see also Ignelz and Prestly 1991). Des Rosers (2002) uses detaled mcro-spatal data to fnd a 10% drop n value f the property s abuttng or facng a tower, and a premum once t s 1 2 lots away from the easement. Chalmers and Voorvaart (2009) smlarly observe that once the level of vsblty of the tower s controlled for, only encumbrance has a sgnfcant but small effect on value whle mere proxmty does not. Bottemller and Wolverton (2013) fnd weak negatve effects for abuttng propertes, but that these are more sgnfcant for hgher-prced homes. They cauton that studes are not easly generalzable to other markets because of dfferences n terran, vegetaton, and local atttudes. Kroll and Prestly (1991) summarze these results: effects are generally lmted to 10% of the value and most evdent for encumbered property; proxmty effects vansh rapdly wth dstance; when a new lne s constructed, the negatve effects are ntally large, but attenuate over tme; the tower heght has lttle mpact on loss n value, but vsblty plays an mportant role; and proxmty may even lead to a premum, presumably because the ROW can provde recreatonal amentes and added prvacy (Rhodesde and Harwell 1988, Des Rosers 2002, but see Peser and Schwann 1993). Turnng to market effcency, there s a large body of lterature that nvestgates the centralzed and homogeneous securtes markets. 2 By effcent, one generally means the extent to whch relevant publc (weak-form) or prvate (strong-form effcency) currently avalable nformaton s mpounded n the prce of an asset, so that an nvestor cannot derve economcally sgnfcant excess returns. For real estate markets, there s rather less research, but effcency s tested n broadly three dfferent ways: research that examnes f the present value of future rents matches prce n the long run; whether excess returns exhbt seral correlaton and mght thus be forecasted; and tests of whether prces match ther underlyng determnants, such as ncome and employment. Meese and Wallace (1994) fnd that the present value relatonshp s volated n the short term, but holds over the long run. They also remark on the mportance of consderng transacton costs before concludng that the market s neffcent. Lnneman (1986) and Case and Shller (1989) fnd evdence of seral correlaton n house prces, whch mples that future prces are predctable. Guntermann and Smth (1987), however, conclude that the market s weak-form effcent once transacton costs are ncorporated. They examne the seral correlaton of excess total returns for 57 ndvdual real estate markets versus the market as a whole for the perod 1969 1982. At 1-3 year lags, they fnd no effect, and although seral correlaton s observed at 4-10 year ntervals, t s not economcally sgnfcant. They also nvestgate tradng strateges, and fnd that a mean- 2 A classc overvew s provded by Fama (1970); see also Dmson and Mussavan (1998).

Effect of HVTLs on Property Value 5 reverson strategy would be proftable, but only n the absence of tradng costs. Rosenthal (1994) fnds that shocks to prces relatve to constructon costs dsappear faster than constructon lags, whch mples that the market for mprovements s effcent, and that any remanng neffcency must be derved from the land market. 3. Background and Tme Lne The Tehachap Renewable Transmsson Project (TRTP) was a Southern Calforna Edson (SCE) program to construct and upgrade 173 mles of HVTLs to delver 4,500 MW of power from wnd farms near Tehachap n eastern Kern County, Calforna, to the Antelope Valley and Los Angeles Basn. The project contrbutes toward complance wth Calforna state law wth regard to the proporton of power consumpton that must be derved from renewable resources, and helps attan the greenhouse emsson caps n Calforna under the Global Warmng Solutons Act of 2006. SCE dvded the transmsson lne project nto segments. The last of these n the Los Angeles Basn s Segment 8, and runs from Monterey Park to the cty of Ontaro. The part of ths segment that traverses our study area n the cty of Chno Hlls was to nvolve the constructon of a double-crcut, 3 500 kv lne from the Mesa substaton n Monterey Park to a substaton n the Cty of Chno. 4 In Chno Hlls, the project would place roughly 200 ft tall tubular and lattce steel towers on the 150 ft-wde utlty ROW (see Aabo 2013). The ROW exsted well before our study perod wth approxmately ten 50 100 ft unused 220 kv towers placed on t pror to the upgrade. 5 SCE announced 6 ts ntenton to construct the TRTP n late 2004 when t appled to ts regulator, the Calforna Publc Utltes Commsson (CPUC), for permsson to construct Segments 1 3. Plans for Segment 8 n Chno Hlls were made publc n the second quarter of 2007 when SCE dstrbuted comprehensve nformaton packages, and advertsed open houses n local newspapers, on ts webste and maled nvtatons to property owners wthn 3 Wth a 3-phase system, sngle crcut means three cables, a double crcut, sx cables. 4 Chno Hlls s about 50 km east of downtown Los Angeles, n San Bernardno County. Monterey Park s west of Chno Hlls, and the Cty of Chno s the mmedate eastern neghbor of Chno Hlls, roughly dvded from t by the Chno Valley Freeway (CA 71). 5 Startng at the Chno Valley Freeway on the eastern border of Chno Hlls (33.986466 o /-117.713386 o ), the ROW runs drectly west for 1.75 km (to 33.986697 o /- 117.732786 o ), and then n a south-westerly drecton for about 4.5 km before enterng the relatvely unpopulated hlls that form the western border of the Chno Hlls. The western-most tower locaton that we consder s at 33.969762 o /-117.776489 o. 6 The followng tme lne s a summary of court documents, records and nformaton pamphlets from SCE, and artcles n the local press.

6 Thomas and Welke 300 ft of the ROW. Later testmony by representatves of a local ctzens group ( Hope for the Hlls ) before a CPUC admnstratve law judge ndcates that ths s when the resdents of Chno Hlls frst became wdely aware of the plans of SCE (Goodwn 2013, Gens 2013). A formal project applcaton for constructon was submtted n June 2007, and the Cty of Chno Hlls mmedately fled a complant to the San Bernardno County Superor Court that the planned transmsson lne would overburden the easement over cty property (Fleager 2013). On December 17, 2009, the CPUC approved the SCE plan, and four months later, the San Bernardno County Superor Court ruled that the CPUC had exclusve jursdcton wth regard to ROW ssues. Tower constructon began on January 17, 2011. 7 Durng ths tme-frame (md-2007 through 2011), poltcal pressure mounted aganst the project as proposed, and the Cty and Hope for the Hlls campagned for alternate routng and desgn of the transmsson lne. On October 19, 2011, after 16 towers were completed, and no cables had yet been strung, the CPUC ordered a stay of constructon and asked SCE to provde fve alternatve proposals for routng and desgn. Among the alternatves that SCE submtted n January 2012 was a plan to place the cables underground, but t contnued to advocate for the tower approach. By March 2012, medaton between the Cty and SCE had ended n falure and proceedngs were moved before a CPUC admnstratve law judge n Aprl 2013. 8 By then, however, the CPUC had sded wth the Cty and n July voted 3-to-2 n favor of placng sngle crcut cables underground. The towers were removed begnnng September 24, 2013, and cvl constructon for underground transmsson lnes on the ROW commenced soon thereafter. 4. The Model and Data The tme lne descrbed above leads us to consder two event wndows. The frst s a smple hard wndow, from January 17, 2011 to September 24, 2013, durng whch the new 200 ft towers would have been vsble and possbly nfluenced even unnformed buyers. 9 We shall call ths perod 1 and desgnate pror transactons (January 2001 December 2010) as untreated subjects, durng whch the much shorter, unused 220 kv towers were n place. 1 covers only a part of the tme durng whch publc nformaton about the 7 In September 2011, the Court of Appeal affrmed the Superor Court rulng that the CPUC had exclusve jursdcton n the matter. A subsequent petton to the Calforna Supreme Court was dened n December 2011. 8 See Fleager (2013), Lombardo (2013), Aabo (2013), Goodwn (2013), and Gens (2013). 9 Sellers wthn a 300 ft dstance from the easement would have been notfed by SCE around the second quarter of 2007. In prncple, brokers would have dsclosed project plans to buyers as part of ther fducary duty.

Effect of HVTLs on Property Value 7 pendng transmsson lnes was avalable. 10 We choose not to nclude the perod after September 2013 when constructon for underground cables began snce t s not drectly comparable to the pror two perods. We cannot yet consder the (potentally postve) mpact from underground cables as constructon s ongong and there are nsuffcent transactons proxmate to the ROW. 11 As a second test, we nvestgate f there was any announcement eff ect. We choose to start ths wndow, 2, when SCE began ts notfcaton campagn n March 2007 because ths s when resdents frst became aware of the plans of SCE (Goodwn 2013 and Gens 2013). The wndow ends at the commencement of tower constructon n January 2011. If the market ncorporated publcly avalable nformaton to any degree, one would expect some negatve effect on property values and/or sales volume proxmate to the ROW durng that tme. The transactons data were obtaned from PropertyRadar (2015), a frm that specalzes n data for real estate professonals. To reduce heterogenety whlst keepng a suffcently large control group at dstances where the potental negatve prce effects from the towers and transmsson lnes are neglgble, we restrct transactons to wthn 1 km of the ROW, between January 2001 and November 2013. 12 Wthn ths band and tme frame, there were 2,735 transactons. To the hedoncs n the data set, we add the lattude and longtude of each parcel as obtaned from county records, along wth tower locatons from SCE. We use four dstance ndcators as a proxy for the level of negatve amenty: (1) dstance levels to the center of the ROW, d TL; 13 (2) dstance to the nearest tower, d TW; and, from an nspecton of the proxmate parcel maps, whether (3) the property abutted the ROW; and (4) whether t was encumbered. 10 December 2009 (when the CPUC approved tower constructon), or March 2007 (when SCE publczed the project), though July 2013, (when the CPUC voted to underground the transmsson lne). 11 See McNar and Abelson (2010) for an emprcal study of the economc mpact of placng utlty cables underground. 12 We nclude a one month lag between the date of the sales contract and transfer of ownershp. 13 Closest dstance to a path was computed brute force by placng a lattude/longtude marker every 10 m or so on ts centerlne, and then fndng the dstance of the property to the closest pont on the path.

8 Thomas and Welke The model that we use s motvated by a standard double dfference setup, and restrcted by the avalable characterstcs, sgnfcance, and collnearty tests: 14 ln PRICE = LIVINGAREA LA LOTSIZE LS AGE A BEDS BATHS D BD BT UNITS U D D D STORIES S SFR SFR POOL P ln HPI t HPI DCLOSEQTR CQ DZIP Z D D ELEMENTARY E SALETYPE ST [ D D D ] ABUTS A A A, [ D D D ] ABUTSONLY AO AO AO, [ D D D ] ENCUMBERED EN EN EN, [ D D D ] TL TL TL TL, [ D D D ], (1) TW TW TW TW, where D ndcates a dummy varable. Terms of prmary nterest are enclosed n brackets [...], and we refer to all other terms as the base regresson model. ENCUMBERED s an ndcator for whether the lot s encumbered by the ROW. Ths porton s taxed, owners must mantan t, and may not mprove t wthout permsson from SCE. Owners can, however, use encumbered land for coverage restrctons on mprovements. ABUTS here wll mean that the lot shares a border wth (or s also encumbered by) the ROW. In an alternate defnton, we use ABUTSONLY to exclude propertes wth an easement. TL and TW are dstance ndcators to the center of the ROW and nearest tower, respectvely. The nteracton between these four proxmty measures and allows us to control for local market condtons unrelated to transmsson lnes: the effect on prce for proxmate transactons durng compared to the perod when the new negatve amenty (or threat of one) was absent, relatve to the same dfference for dstant propertes. The house prce ndex (HPI) (see below) serves as a further control for the wder market. Sales wth mssng lvng area, year bult, bed/bathroom count, or lot sze were dropped. A small number of (non-abuttng) propertes have a bed and bath count that s ncongruous wth the lvng area and were removed. We assume that any property constructon was completed mdyear, and dropped any observaton for whch age at tme of sale (AGE) was less than negatve 0.5 years. 15 Four non-abuttng propertes wth ages that exceed 50 years sgnfcantly contrbute to non-lnearty n AGE, and were also removed. 16 14 We return to the ssue of a log-lnear specfcaton later. 15 There are 22 such excluded observatons, none of whch abut the ROW. Our fndngs do not change f we keep these observatons nstead. Some of them could, of course, correspond to pre-sales. 16 AGE-squared was not consdered as the VIF tests show t to be hghly collnear wth AGE; dwellngs n Chno Hlls are relatvely new, wth only few exceptons.

Effect of HVTLs on Property Value 9 We note that Hamlton and Schwann (1995) advocate the use of logarthms of ndependent varables to capture nonlnearty. Gven that there are only very few propertes that cause the nonlnearty n our sample, none of whch are proxmate to the ROW, we choose nstead to drop these observatons and forego the added complcatons of a full translog specfcaton. There were sx propertes that transacted a total of seven tmes wth a lvng area of more than 4,500 sq ft. None abut the ROW and they are more than 400 m from any tower. Most belong to a non-contguous cluster of very large luxury homes about 500 to 1,000 m south-west of the western-most tower. Local regresson ndcates that they would be responsble for nducng large nonlnearty n LIVINGAREA, whch we do not cure wth hgher order terms because the varance nflaton factor (VIF) tests show strong collnearty effects. Gven that these homes are qualtatvely dfferent from the bulk of the sample and none abut the ROW, we dropped all propertes wth a lvng area greater than 4,500 sq ft. Smlarly, we cut LOTSIZE at one acre. SALETYPE s a PropertyRadar ndcator for the type of transacton, dentfed through the use of offcal records from the offce of the county recorder, and matched wth assessor records, geographc nformaton systems (GIS) data, and trustee sale results. The levels of SALETYPE n our data set are: Market, for whch none of the attrbutes lsted next apply; 3rd Resale, purchased by a 3rd party at a trustee sale and resold; Market Flp, sold n a market transacton f the pror purchase was market less than 180 days ago; REO Resale, purchased from a bank as a real estate owned (REO) property; REO Flp, the sale of a property f t was purchased REO less than 180 days pror; Short Sale, sold at a prce less than the outstandng mortgage (and requrng bank approval unless the seller made up the dfference n cash); Short Sale Flp, a sale that was preceded by a short sale wthn 180 days; and Short Sale Foreclosure, whch corresponds to a short sale durng an actve foreclosure. Non-Market transactons 17 make up about 25% of the total sample and we choose to keep them n the estmaton, and return to ths ssue below. Addtonally, we kept only one of any ``double transactons : f all attrbutes, ncludng prce, transfer date, and transfer type were dentcal, one observaton was dropped. For same-day sales wth dfferng transfer types, we removed the non-market transacton, or kept the hgher-prced sale. There are a total of sx same-day transacton pars, none of whch abut the ROW. ln HPI s the (log) quarterly Zllow all-property prce ndex level for Chno Hlls (zp code 91709), and accounts for general market condtons. We have also consdered our own quarterly date-of-sale dummes, and the resultng 17 In ths dscusson, we label all transactons wth specal qualfers as non market and those wthout as market. A more usual ndustry practce would be to refer only to transfers such as nter-spousal deeds, transfers nto a trust, or upon the death of a jont tenant, etc. as non-market. There are no such transactons n our data set.

10 Thomas and Welke ndex well reproduces that of Zllow, albet wth added nose. To account for any seasonal varaton and the fact that about 2% of the raw sample (5% of the fnal sample) has a Chno postal address (zp-code 91710), we nclude dummes for quarter of transacton (CLOSEQTR) and zp-code (ZIP), respectvely. Fnally, the other factor varables n Eq. (1) are: detached or duplex (UNITS), one or two stores (STORIES), sngle famly or condomnum (SFR), pool (POOL), and elementary school dstrct (ELEMENTARY). 18 An estmaton of the base model Eq. (1) reveals no mult-collnearty ssues va VIF tests, and local regresson analyss ndcates lnearty n all contnuous varables. A Breusch-Pagan test shows that the data are strongly heteroscedastc and scatter plots ndcate 36 observatons wth studentzed resduals less than - 2.5 (none greater than +2.5), mostly at md-range ftted log-prce. One of the outlers abuts the ROW, but outsde ether event wndow. A further 55 observatons exhbt a Cook s D greater than the cut-off value of 4/N, of whch sx abut the ROW and one abuts nsde 2 (none n 1). 19 These outlers and nfluental observatons are dropped they do not sgnfcantly affect our results, presumably snce they are not proxmate transactons durng ether event wndow (see below). The resduals reman heteroscedastc, so we shall use Huber-Whte standard errors n what follows. The summary statstcs are presented n Table 1. For 1, there are 2,569 transactons wth 97 that abut the ROW, 21 of whch are n 1. For 2, sales after January 2011 were dropped and the counts are 2,115, 76, and 17, respectvely. Fnally, we note that our results are robust to renstatng the outlers and nfluental observatons that we dropped above. Relaxng the cuts and removng nstead only the 15 observatons wth r st < 4 (keepng all Cook s D otherwse) leaves 2,631 observatons, of whch 20 have r st < 2.5 (none of whch abut) and a further 75 exhbt Cook s D > 4/N (one of whch abuts, but outsde both 1 and 2). The base regresson adjusted R 2 drops margnally to 93.7%, and we fnd that the double dfference log prce effects are qualtatvely dentcal to the results that we report below, albet wth a slghtly lowered sgnfcance. Smlar results are found f we drop all non market sales. 18 The ROW runs roughly east-west, approxmately perpendcular to Freeway 71. Snce Chno Hlls property values tend to rse toward the west, an alternate specfcaton to school dstrct that we consdered uses levels of dstance to the freeway, dhw. dhw s hghly sgnfcant when ELEMENTARY s excluded, but becomes nsgnfcant when t s ncluded. In partcular, even mmedate proxmty to the freeway was no longer sgnfcant. We chose the more customary school dstrct co-varate, and dspensed wth dhw altogether. 19 Roughly one-half of these outlers and nfluental observatons are market transactons, compared to 75% n populaton.

Effect of HVTLs on Property Value 11 Table 1 Summary Statstcs for the Transacton Data 1 2 Mean Mn Mean Mn SD. Max SD. Max Observatons 2569 2115 Transfer Date (Y.Q) 2006.66 2001.25 2005.35 2001.25 3.85 2014.00 2.88 2011.25 Prce per sq ft 232 95 231 95 62 442 66 442 Lvng Area (sq ft) 1951 974 1933 974 635 4500 632 4500 Lot Sze (acres) 0.154 0.022 0.152 0.022 0.087 0.849 0.086 0.849 Age (yrs) 19.2-0.3 18.1-0.3 10.1 49 9.6 47 Beds 3.57 2 3.54 2 0.79 6 0.78 6 Baths 2.52 1.5 2.50 1.5 0.52 5 0.52 5 Unts (1 or 2) 1.0012 1.0009 Stores (1 or 2) 1.701 1.707 SFR [Condo] 0.854 0.845 Pool 0.171 0.165 Chno Zp 0.048 0.042 Market 0.764 0.810 3rd Resale 0.020 0.010 Market Flp 0.007 0.008 REO Flp 0.004 0.001 REO Resale 0.106 0.098 Short Sale 0.068 0.054 Short Sale Flp 0.002 0.001 Short Sale Forecl. 0.030 0.018 Elem. Sch. Cntry Sprngs 0.102 0.111 Hdden Trals 0.324 0.326 Ltel 0.337 0.317 Glenmeade 0.171 0.179 Oak Rdge 0.049 0.049 Chaparral 0.018 0.018 Abuts 0.0378 0.0359 Abuts Only 0.0230 0.0213 Encumbered 0.0148 0.0147 Dst. ROW d TL (m) 492 16 494 16 277 999 275 999 Dst. Tower d TW (m) 519 18 521 18 278 1342 277 1342 0.177 0.252 Abuts 0.00817 0.00804 Abuts Only 0.00545 0.00378 Encumbered 0.00273 0.00426

12 Thomas and Welke 5. Results We begn wth the results for 1, as dscussed n the prevous secton. Table 2 shows the coeffcents from an estmaton of the base model, Eq. (1), wthout the terms of fnal nterest (those n square brackets). We fnd a small but sgnfcant seasonal effect, wth prces n the thrd quarter exceedng the frst quarter prces by 2.2%. The Chno postal addresses n our sample are 9% lowerprced when compared to Chno Hlls propertes. Lvng area s hghly sgnfcant, and once ths s controlled for, bedroom and bathroom counts add very lttle to prce snce they are correlated wth lvng area. Lot sze s hghly sgnfcant at about 40% of prce per acre. Condos and duplexes are 20 and 16% cheaper than sngle famly resdences, all else equal, whle pools add 2% to the value. Transfer types vary n sgnfcance, but Wald tests ndcate that they are hghly sgnfcant as a group. Propertes purchased as REO and n short sales transfer are 3.8% and 2.9% less than market transactons, respectvely. The correspondng flps are sgnfcantly postve, 4.1% and 2.3%, respectvely, thus ndcatng that any upgrades are reflected n prce, relatve to market sales. 20 Thrd party resales and market flps are nsgnfcantly dfferent from market sales, whle short sales durng an actve foreclosure sell at a 10% dscount. Elementary school catchment areas are ndvdually sgnfcant; the Chaparral dstrct straddles the freeway, whch mght account for ts large dscount. The adjusted R 2 s hgh at 95.05%, whch s not surprsng snce the study area s small (roughly 13 km 2 ), relatvely homogeneous, and we have removed outlers. 21 We now add the event wndow dummy 1, a proxmty measure, and ts nteracton wth 1. The levels for the dstance to the center of the ROW (d TL) are 0 100 m, 100 200 m, and 200-1000 m, and the same for the dstance to the nearest tower (d TW). To examne the prce mpact from the transmsson lnes, frst consder the more usual average margnal effect: Δ m,jn(log Prce) = log Prce(P =, 1=m) log Prce(P =j, 1 =n), (2) where log Prce s the predcted prce, (, j) ndcate the level of proxmty P, and (m, n) are 0 or 1 for outsde and nsde the event wndow, respectvely. Δ 11,01 and Δ 10,00 are thus between-subjects estmates the predcted prce effect from movng a transacton from far (P = 0) to near (P = 1), wthn the event wndow and outsde of t, respectvely. Smlarly, Δ 11,10 and Δ 01,00 are wthnsubject estmates of the effect of movng near and far transactons, respectvely, from outsde to nsde the event wndow. 20 Note that the results do not reflect excess returns for nsders, as we have not controlled for purchase prce, extngushed opton value, and costs. However, the parng of sales wth flps suggests a 7 8% captal gan (excludng costs). 21 The adj. R 2 s 88% when all transactons are utlzed.

Table 2 Effect of HVTLs on Property Value 13 Base Prce Regresson, Eq. (1), wthout the Terms Enclosed n [...] ln Prce Coef. t ln HPI 1.05317** 142.57 Closng Qtr: [1st] 2nd 0.02079** 4.23 3rd 0.02225** 4.55 4th 0.01397* 2.70 Chno [Chno Hlls] -0.09083** -10.41 Lvng Area (1,000 SqFt) 0.28717** 43.96 Lot Sze (acres) 0.39251** 11.20 Age (yrs) -0.00633** -20.43 Beds 0.00400 1.16 Baths 0.00711 0.96 Unts: 2 [1] -0.16363** -4.77 Stores: 2 [1] 0.00521 0.94 Type: Condo [SFR] -0.20183** -24.33 Pool 0.02101** 4.21 Transfer Type: [Market] 3rd Resale 0.00704 0.63 Market Flp 0.01554 1.15 REO Flp 0.04081* 2.89 REO Resale -0.03792** -6.03 Short Sale -0.02949** -4.11 Short Sale Flp 0.02306 1.47 Short Sale Forecl. -0.10267** -9.98 Elementary School [Oak Rdge] Country Sprngs -0.12214** -12.63 Hdden Trals -0.03932** -4.74 Ltel -0.05690** -6.82 Glenmeade -0.05527** -5.40 Chaparral -0.34503** -20.33 Intercept -1.18332** -12.52 Note: The data correspond to the 1 estmaton, wth 2,569 transactons. Base levels are shown n square brackets. The adjusted R-sq s 95.05%, and Huber-Whte t statstcs are reported. *p < 0.01, **p < 0.001 The results are summarzed n Tables 3 and 4. Abuttng propertes are predcted to suffer only a very small and nsgnfcant loss n value when compared to otherwse smlar non-abuttng propertes, provded that the sale took place outsde the event wndow: wth the old 220 kv towers n place, prces are not affected by the abuttng ROW, consstent wth fndngs by other authors that mpact dmnshes wth tme. For the new and taller towers, on the other hand, there s a loss of about 6.0% for abuttng propertes, and ths result s sgnfcant at the 99.9% level. An abuttng sale nsde the event wndow loses 4.6% n value

14 Thomas and Welke (wth 98% confdence) compared to the same sale outsde the wndow, whle non-abuttng propertes suffer no such loss. A smlar set of results s obtaned for encumbered sales: on average, an encumbered property sold for 9.6% less (at 99% CL) than ts unencumbered twn nsde 1, whle outsde of the wndow there s no dscernable effect. Market partcpants dd not dscount the loss of rghts wth the old towers on the easement, but sgnfcantly so once the new towers were constructed. The loss from movng the transacton from outsde the event wndow to nsde s 8.1% when encumbered, and nsgnfcant otherwse. Table 3 The Average Margnal Effect on Log Prce for Abuttng and Encumbered Varables, and 1, Relatve to The Indcted Base Value Ref. 1 = 0 1 = 1 Abuts [Not] -0.0104-0.0595*** (0.91) (3.45) Not Abuttng [ 1 = 0] 0.0029 (0.52) Abuts [ 1 = 0] -0.0463** (2.28) Encumbered [Not] -0.0128-0.0957*** (0.64) (2.59) Not Encumb. [ 1 = 0] 0.0021 (0.39) Encumbered [ 1 = 0] -0.0808* (1.95) Note: Huber-Whte t statstcs n parentheses below the coeffcent. * p < 0.1, ** p < 0.05, *** p < 0.01. Turnng to the dstance ndcators, d TL and d TW (Table 4), we fnd that up to 100 m nsde the event wndow, the loss n value s 3.4%, and 3.6% when compared to sales more than 200 m from the ROW or nearest tower, respectvely. Outsde 1, the small loss s nsgnfcant the old and shorter towers dd not matter. Correspondng results are obtaned f we swtch the most proxmate sale nto the event wndow. However, the effects already vansh at 100 200 m n dstance (Colwell and Foley 1979, Des Rosers 2002, Chalmers and Voorvaart 2009). Assumng that the unrelated local effects over tme are dentcal for both groups far and near, 22 the transmsson lne effect can be solated by removng the estmate of the between subject dfference outsde the event wndow (Δ 10,00) from the between subject dfference nsde the wndow (Δ 11,01) Δ 2 = Δ 11,01 Δ 10,00. (3) 22 After testng ths parallel trend assumpton, we fnd that the covarance of, P, and P wth the resduals n our regresson are zero wth p-values near 100%.

Effect of HVTLs on Property Value 15 In our setup, ths double dfference s smply the respectve coeffcent β P,1 n Eq. (1). The results are collected n Table 5. For 1 (Case A), beng encumbered nduces the largest loss of value, 8.3%. Abuttng propertes (whch nclude encumbered ones) lose a smaller, but sgnfcant 4.9% of value as a result of the presence of the new towers. As before, dstance to the ROW and nearest tower matters only wthn the frst 100 m (about 3% n each case), but not partcularly sgnfcantly so. Table 4 The Average Margnal Effect on Log Prce n 1 for Dstance Indcators d: 0=(0,100 m], 1=(100, 200 m], and 2=(200, 1000 m]. Relatve to the Indcted Base Value. Ref. Dst. ROW dtl Dst. Tower dtw 1 = 0 1 = 1 1 = 0 1 = 1 d = 0 [d = 2] -0.0024-0.0340** -0.0093-0.0363** (0.33) (2.46) (0.73) (2.10) d = 1 [d = 2] -0.0114* 0.0077-0.0033-0.0038 (1.86) (0.63) (0.57) (0.30) d = 0 [ 1 = 0] -0.0301** -0.0252 (2.04) (1.21) d = 1 [ 1 = 0] 0.0206* 0.0013 (1.66) (0.10) d = 2 [ 1 = 0] 0.0015 0.0018 (0.25) (0.30) Note: Huber-Whte t statstcs n parentheses below the coeffcent. * p < 0.1, **p < 0.05. The results for abuttng and encumbered sales mmedately rase the queston of whether abuttng propertes that were not encumbered suffered a sgnfcant loss n value durng 1. There are 62 non-encumbered transactons that abut, and 14 of these fall n 1. The double dfference n log prce wth ths redefned abuttng ndcator s added as Regresson B n Table 5. The prce effect declnes to a loss of 3.1% (vs. 4.9% before) and t s not qute sgnfcant at the 90% level (p = 12.7%). Abuttng-only does result n a loss n value, albet by a lower amount unless the property s encumbered. The results for d TL and d TW at 0-100 m become nsgnfcant when abuttng propertes are excluded. Ths could be nterpreted as meanng that the vsual effect from the towers was not very mportant, 23 but s also lkely the result of the low transacton count. 24 23 Ether because buyers dd not dscount t, or because one lot n the vsual mpact s hdden by landscapng and/or unmportant because of the orentaton of the house. 24 Conservatvely assumng average 6,500 sq ft (80 ft) 2 lots to be square, the ROW plus abuttng propertes (150 ft + 2 80 ft 310 ft 100 m) leaves lttle room for nonabuttng sales wthn a 200 m dameter of any tower. In fact, the number of sales wthn 100 m of any tower drops from 90 to 39 when abuttng propertes are removed and there are only 6 wthn 1.

16 Thomas and Welke Fnally, we turn to the second event wndow, 2, whch starts when SCE began ts nformaton campagn and ends wth the commencement of the tower constructon. Results C n Table 5 show the dfference n dfferences: none of the negatve amenty varables are sgnfcant durng the tme that publc nformaton about the towers was avalable. Ths s n stark contrast to the prce effect when the new towers were n place. Table 5 The Double Dfference Log Prce, Δ 2 (log Prce), for Varous Measures of Proxmty to the Transmsson Lnes A B C Abuts -0.0492** -0.0310-0.0149 (2.38) (1.52) (0.49) Encumbered -0.0829** -0.0829** 0.0263 (1.99) (1.99) (0.67) d TL [200 m+] 0-100 m -0.0316** -0.0313-0.0091 (2.04) (1.57) (0.54) 100-200 m 0.0191 0.0189 0.0133 (1.42) (1.40) (1.05) d TW [200 m+] 0-100 m -0.0270-0.0282-0.0213 (1.26) (1.02) (0.70) 100-200 m -0.0004 0.0053 0.0125 (0.03) (0.36) (0.98) Note: Case A: event wndow 1. Case B: 1, abuts excludes encumbered propertes, and d excludes all abuttng propertes. Case C: 2, announcement wndow. Omtted levels n square brackets. Huber-Whte t-statstcs n parenthess below the coeffcent, *p < 0.1, **p < 0.05. Of course, the announcement effect could have manfested tself n quantty rather than prce, so t s nstructve to test for a structural break n abuttng (V A) versus non-abuttng (V NA) sales volume durng 2. Consder the smple tmeseres regresson (Hansen 2000): V A(t) = β V NA(t)+ θ V NA(t) 2 + e(t) (4) The errors are serally correlated but normal, so we use Huber-Whte standard errors and a standard Wald test. We fnd β =0.0385 ± 0.0044 and an nsgnfcant θ =0.0020 ± 0.0098, so that we cannot reject the null of no structural break n sales volume durng 2. A more senstve test mght be to examne tme on market, for whch we currently have no data.

Effect of HVTLs on Property Value 17 6. Dscusson We have used an event study to examne the revealed prce effect on resdental propertes from an upgrade to hgh voltage transmsson towers that were constructed on an exstng ROW. The transacton data cover the perods before and durng ths wndow, and allow us to obtan not only a between subjects comparson (proxmate versus not), whch s used n most studes n the lterature, but also a wthn subjects dfference (nsde versus outsde the event wndow) and therefore a double-dfference to control for non-transmsson lne local effects. From the tme lne for the project, we can also nvestgate a potental announcement effect: does publc nformaton about constructon plans affect value, as opposed to the physcal presence of the towers? We fnd a sgnfcant loss n value from the upgrade for encumbered (8.3%) and abuttng (4.9%) propertes, and nsgnfcant losses when the older towers were present, even for lots wth an easement. These results are consstent wth prevous studes that show that the prce mpact s ntally large, but dmnshes over tme. Propertes that abut the ROW, but not encumbered by t, suffer a smaller and less sgnfcant loss from the constructon of the new towers (3.1%). Sales wthn 100 m of the ROW or nearest tower experence a 3% drop n value whch vanshes beyond that dstance. However, the result for the most proxmate sales s weaker (1-2%) and nsgnfcant f one excludes abuttng sales. Ths llustrates the mportance of dstngushng between adjacency or proxmty, and a rghts takng. Broadly speakng, our results are consstent wth fndngs by other authors: () over tme, prce mpact s dmnshed; () effects vansh beyond about 100 m; () the proxmate sales results are largely drven by abuttng lots; (v) encumbered sales are sgnfcantly negatvely affected; and (v) abuttng propertes somewhat less so. We further fnd no evdence that publc nformaton pror to the constructon of the towers affected sales prces, even f the property abutted or was encumbered by the ROW: market partcpants dd not mpound avalable nformaton n prce over the 3-4 year announcement nterval. The nformaton-only wndow (March 2007 January 2011) s well defned by court testmony, and also squarely falls n the downturn when seller concessons would have presumably been more easly obtanable. Whle ths result s consstent wth the fndng of Smons and Sagnor (2006) that announcement effects have no sgnfcant prce mpact, t s nonetheless somewhat surprsng n lght of the large effect from encumbrance durng the presence of the new towers, and because many studes suggest that the real estate market s weakly effcent, at least n an economc sense. 25 Before concludng that our results pont to neffcency, more study s needed to determne f they could be explaned by a drop n sales volume durng the announcement wndow. Whle we dd not fnd any evdence of a structural 25 A subset of market partcpants do seem to be able to extract excess returns locally, as hnted at n an examnaton of non market sales.

18 Thomas and Welke break durng 2, our sample probably contans too few abuttng transactons to draw a defntve concluson. A better test may be to examne the dfference n tme on market compared to dstant propertes, for whch we currently have no data. The buryng of 500 kv cables requres perodc vaults for coolng and ventng whch are dsruptve to mmedately proxmate home owners, but presumably at a much lower level than towers. A potental future avenue of study would be therefore to examne the prce mpact of the postve announcement of tower removal (see Smons and Sagnor 2006 and McNar and Abelson 2010). Current ongong constructon of the underground transmsson lne complcates the study, but eventually, transacton volume n the post tower perod wll be suffcent to nvestgate the apparent asymmetry between bad and good announcements as noted by Smons and Sagnor (2006). Acknowledgement We thank PropertyRadar for the use of ther data, and Ashok Bardhan, Thomas Davdoff, Sean O Toole, and Oscar We for useful dscussons. Any errors are entrely our own. References Aabo, T. (2013). Testmony on Behalf of the Cty of Chno Hlls, Exhbt CH7. Publc Utltes Commsson of the State of Calforna, March 20: Applcaton No. 07-06-031, Fled June 29, 2007. Bottemller, S.C. and Wolverton, M.L. (2013). Prce Effects of HVTLs on Abuttng Homes, The Apprasal Journal, 45 61. Bryant, J.A., and Epley, D.R. (1998). Cancerphoba: Electromagnetc Felds and ther Impact n Resdental Loan Values, Journal of Real Estate Research, 15.1, 115 29. Case, K.E., and Shller, R.J. (1989). The Effcency of the Market for Sngle Famly Resdences, Amercan Economc Revew, 79, 125 137. Chalmers, J.A., and Voorvaart F.A. (2009). Hgh-Voltage Transmsson Lnes: Proxmty, Vsblty and Encumbrance Effects, The Apprasal Journal, 227 45.

Effect of HVTLs on Property Value 19 Colwell, P.F. (1990). Power Lnes and Land Value, Journal of Real Estate Research,5, 1, 117 27. Colwell, P.F., and Foley, K.W. (1979). Electrc Transmsson Lnes and the Sellng Prce of Resdental Property, The Apprasal Journal, 47, 4, 490 99. Delaney, C.J., and Tmmons D. (1992). Hgh Voltage Power Lnes: Do They Affect Resdental Property Value?, Journal of Real Estate Research, 7, 3, 315 29. Des Rosers, F. (2002). Vsual Encumbrance and House Values: A Mcrospatal Approach to Impact Measurement, Journal of Real Estate Research, 23, 3, 275 301. Dmson, E., and Mussavan, M. (1998). A Bref Hstory of Market Effcency, European Fnancal Management, 4, 1, 91 193. Fama, E. (1970). Effcent Captal Markets: A Revew of Theory and Emprcal Work, Journal of Fnance, 25, 2, 383 417. Fleager, M. (2013). Testmony on Behalf of the Cty of Chno Hlls, Exhbt CH-1. Publc Utltes Commsson of the State of Calforna, March 20: Applcaton No. 07-06031, Fled June 29, 2007. Gens, J. (2013). Testmony on Behalf of the Cty of Chno Hlls, Exhbt CH- 4. Publc Utltes Commsson of the State of Calforna, March 20: Applcaton No. 07-06-031, Fled June 29, 2007. Goodwn, R. (2013). Testmony on Behalf of the Cty of Chno Hlls, Exhbt CH-5. Publc Utltes Commsson of the State of Calforna, March 20: Applcaton No. 07-06031, Fled June 29, 2007. Guntermann, K.L., and Smth, R.L. (1987). Effcency of the Market for Resdental Real Estate. Land Economcs, 63, 1, 34 45. Hamlton, S.W., and Schwann, G.M. (1995). Do Hgh Voltage Electrc Transmsson Lnes Affect Property Value, Land Economcs, 71, 4, 436 44. Hansen, B.E. (2000). Testng for Structural Change n Condtonal Models, Journal of Econometrcs, 97, 1, 93 115. Ignelz, P., and Prestly, T. (1991). A Statstcal Analyss of Transmsson Lne Impacts on Resdental Property Values n Sx Neghborhoods. Albany, CA: Fnal Report Prepared for Pacfc Consultng Servces.

20 Thomas and Welke Jackson, T., and Ptts J. (2010). The Effects of Electrc Transmsson Lnes on Property Values: A Lterature Revew, Journal of Real Estate Lterature, 18, 2, 239 59. Knnard, W.N. (1967). Tower Lnes and Resdental Property Values, The Apprasal Journal, 269 84. Knnard, W.N., and Dckey, S.A. (1995). A Prmer on Proxmty Impact Research: Resdental Property Values Near Hgh-Voltage Transmsson Lnes. Real Estate Issues, 23 9. Kroll, C., and Prestly, T. (1991). The Effects of Overhead Transmsson Lnes on Property Values: A Revew and Analyss from the Lterature. Washngton, D.C.: Prepared for the Stng and Envronmental Plannng Task Force of the Edson Electrc Insttute. Kung, H., and Seagle, C.F. (1992). Impact of Power Transmsson Lnes on Property Values: A Case Study, The Apprasal Journal, 60, 3, 413 18. Lnneman, P. (1986). An Emprcal Test of the Effcency of the Housng Market, Journal of Urban Economcs, 20, 140 154. Lombardo, J. (2013). Testmony on Behalf of the Cty of Chno Hlls, Exhbt CH-2. Publc Utltes Commsson of the State of Calforna, March 20: Applcaton No. 07-06031, Fled June 29, 2007. McNar, B., and Abelson, P. (2010). Estmatng the Value of Undergroundng Electrcy and Telecommuncatons Networks, The Australan Economc Revew, 43, 4, 376 88. Meese, R., and Wallace, N. (1994). Testng the present value relaton for housng prces: Should I leave my house n San Francsco?, Journal of Urban Economcs, 35, 245 66. Peser, R., and Schwann, G. (1993). The Prvate Value of Publc Open Space wthn Subdvsons, Journal of Archtectural and Plannng Research, 10, 91 104. Prestly, T., and Evans, G.W. (1990). Perceptons of Transmsson Lnes n Resdental Neghborhoods: Results of a Case Study n Vallejo, Calforna. Study prepared for Southern Calforna Edson. Prestly, T., and Evans, G.W. (1996). Resdent Perceptons of a Nearby Electrc Transmsson Lne, Journal of Envronmental Psychology, 16, 65 74. PropertyRadar, Inc. (2015). Data downloaded Aprl 17, 2015. url: www.propertyradar.com.

Effect of HVTLs on Property Value 21 Rhodesde and Harwell, Inc. (1988). Perceptons of Power Lnes. Resdent s Atttudes. Rchmond, VA: Report prepared for the Vrgna Power Company. Rkon, M. (1996). Electromagnetc Radaton Feld Property Devaluaton, The Apprasal Journal, 87 90. Rosenthal, S. (1994). Resdental Buldng and the cost of Constructon: New Evdence on the effcency of the housng market, Revew of Economcs and Statstcs, 81, 2, 288 302. Smons, R., and Sagnor, J. (2006). A Meta-Analyss of the Effect of Envronmental Contamnaton and Postve Amentes on Resdental Real Estate Values, Journal of Real Estate Research, 28, 1, 71 104. Solum, C.L. (1985). Transmsson Lne Easement Effect on Rural Land n Northwest Wsconsn, Rght of Way, 14 8.