regulation and capitalization of environmental amenities

REGULATION AND CAPITALIZATION OF ENVIRONMENTAL AMENITIES:EVIDENCE EROM THE TOXIC RELEASE INVENTORY

IN MASSACHUSETTS

Linda T. M. Bui and Christopher J. Mayer*

AhsiriKi—HinironniciUai regulatiun in the Unilcd Stales has undergone aslow evolution from command and conlrol strategies towsirds markct-hased regiilalions. One such innoviilion is the Toxics Release Inventory(TRI), ;i regulittion thai requires polluting fums lo publicly disclo.seinformation aboui their toxic emissions. The basic tenet of this regulationis tliai it conecls (or infomiational asymmetries between polluters andbouseholds. iilluwing eomniunities to pressure polluters to decrease theiremissions. Policy-makers have jtidged the TRI a tremendous success, asnational releases declined by 4y/c between 198H and 1999. Yet many ofthe fundamental problems which arc known to lead to the classic failureof the Coase theorem (.stich as high transaction costs and difficulties inorgatiizing) cast doubt on the effectiveness of disclosure rules, alone, tolead to an efiicient outcome in the ease of pollution. We use an event studymethodology with high-quality data on house prices and other localattributes to assess tlie extent to which the public values changes in toxicreleases and thus the success of TRI. Our major findings include: (I)declines in toxic releases appear unrelated to any political economyvariables thai might lead to public activism; (2) initial informationreleased under "fRI liad no signilicant effect on the distribution of houseprices; and l?i} house prices shov, no significant impact of declines inreported toxic releases overtime. Standard errors are small enough that wecan reject the hypothesis that large declines in toxic releases lead to morethan a {).>"/<- increase in house prices. These results also hold when wecontrol for differences in the availahility of information on TRI and thepossible effect of expectations. Our findings cast doubt on the ability ofthe public to process complex information on bazardous emissions andsupport the Coase theorem in that right-io-knt>w taws such as TRI may notbe the most effective form of environmental regulation.

1. Introduction

ENVIRONMENTAL regulation in the United Slates hasundergone a .slow evolution from command and control

strategies to more market-based approaches. In part, thistransition is in response to lhe overwhelming growth in thedirect cost oi regulation and the price of meeting thoseregulations. In the United States, pollution abatement con-trol expenditures are on the order of 1.5%"2.5%- of GDP peryear. As the trend towards stricter, more pervasive environ-mental regulation continues, both the cost and the effective-nes.s of regulation have become hot topics on the publicagenda.

Otie such market-based innovation adopted by the U.S.Environmental Protection Agency (EPA) is to require pol-luting iirtns to publicly disclose private information abouttheir emissions. The basic tenet of this regulation is that it

Received tor publication October 27. 2000. Revision accepted forpublication December 4. 2002.

'•* Boston University and The Whartt)n School, University of Pennsyl-vania, respectively.

The authors uisli lo thank seminar participants at the WEA meetings, theJFK School, Columbia University. Boston University, University of To-ronto, and The Wharton School, along with Katherine Carmen. LisaGeorge, Wayne Gray. Joe Gyoiirko. Charlie Mimmeiberg, Bob Inman, JoelWaldtbgel. Marty Weitztnan. and two anonymous referees, for helpfulcomments, Anv errors are our own.

corrects for informational asymmetries thai may exist.' Oneof the most pervasive examples of this is the EmergencyPlanning. Community Right to Know Act (EPCRA). Theappeal of this type of regulation is evidenced by the prolif-eration of community right-to-know laws at the state level.These laws lake various forms—from requiring state agen-cies to provide environmental data through the internet, toincreasing the reporting requirements for polluting plants.

Using disclosure of private information as an informalregulatory tool is attractive because it is relatively low-cost.Theoretically, both the cash-starved regulatory agency andthe regulated planl could face lower pollution abatementcontrol expenditures. Instead of directly regulating plantsand ensuring compliance, the enforcement agency onlywould be responsible for collecting and maintaining a pub-lic database, increasing community awareness, and penal-izing firms for inaccurate reporting. A polluting firm wouldbe free to choose how much to change its emissions and touse whatever abatement technology it wanted. Communitypolicing would pressure firms to reduce actual emissions.

In this paper, we focus on one example of this lype ofregulation, called the Toxics Relea.se Inventory (TRI). In-troduced by the EPA under the EPCRA in 19K6, TRIrequires manufacturing plants that emit more than a giventhreshold level of any listed toxic substance lo provideemissions data to the EPA for use in a publicly availabledatabase. Prior to the TRI, no record of toxic emissionsexisted.

Between 1988 and 1999, TRI reported emissions fell byapproximately 40%. Support for community right-to-knowlegislation and the TRI is very strong. In a presidentialmemo dated August 8, 1995, the administration wrote thatthe EPCRA of 1986 ". . . provides an innovative approach toprotecting public health and the environment by ensuringthat communities are informed about the toxic chemicalsbeing released . . ." and that ". . . Right-to-Know protectionsprovide a basic informational tool to encourage informedcommunity-based environmental decision making and pro-vide a strong incentive for businesses lo find their owti waysof preventing pollution." The apparent success of the TRI inreducing reported toxic emissions has made the TRI and

' Tbe disclosure of private information has been used as a regulatorymechanism in several instances. One example is the required labeling ofnutritional contents on food packages, and anotber is the OccupationalSafety and Health Act. which requires employers to inform employeesabout workplace hazards. The effectiveness of such regulation, to ourknowledge, has not been verified empirically.

hf AVt/cir iij'E, oinmiics ami Srali.\i!cs. August 2(H)3. 83(3): 69.̂ -7()K201).'̂ by ihe President and Ixllows of Harvard College and the Massachusetts Institute nf Technology

694 THB REVIEW OF ECONOMICS AND STATISTICS

community right-lo-know laws an altraclive tbrm of regu-lation that may become even more widespread.-

From an economic perspective, one might question theeffectiveness of disclosure-based regulation in the contextof environmental emissions. Pollution is, after all, the clas-sic example of an externality. Informing communities abouttoxic emissions in their neighborhood is quite different thangiving employees data about hazards in the workplace ordisclosing nutritional and fat content of prepackaged foods.In the latter two cases, consumers or employees can chooseto avoid a product or job if the price of the food, given itsnutritional and fat content, is too high, or the wage is toolow, given the workplace risks. Residents who live near aplant, however, face signihcant costs in leaving a neighbor-hood where a plant releases high levels of emissions; theirhomes may suffer from a decline in value. Whereas em-ployers or food manufacturers may care about a decline insales, a manufacturer may not care about the decline inproperty values in the surrounding area and may, in fact,benefit from such a decline if land becomes cheaper forfuture expansion. Under the Coase theorem, it is wellunderstood that local community pressure will only disci-pline a plants emissions level if: (I) most members of thecommunity care about the pollution, (2) costs of collectiveaction are low, and (3) communities can get around thefree-rider problem in gathering a coalition.

The TRl provides a unique opportunity to evaluate someimportant assumptions that underlie community right-to-know types of legislation. In particular, how do communi-ties and households react to information about environmen-tal amenities? To what extent is this information news, andhow do localities value changes in those environmentalamenities over time'? Have declines occurred in placeswhere political action is least costly or in communities thathave a strong aversion to pollution? If communities do notvalue the information, ov cannot use it efficiently, the effec-tiveness of community right-to-know laws as a standaloneregulatory instrument is drawn into question.

In this paper we explore the reaction to TRl reporting inMassachusetts between 1987 and 1992. Reported toxicreleases in Massachusetts arc larger on average than in therest of the nation, and percentage reductions in toxic re-leases have been close to the national average. A stateenvironmental regulation introduced in 1989, called theToxics Use Reduction Act. reinforced and supplementedTRl reporting requirements for Massachusetts manufactur-ing plants (as well asanumberof nonmanufacturing plants).

To understand the impact of TRl information, we use anevent study methodology that estimates Ihc capitalization ofenvironmental amenities on house prices. We consider sev-

- A presidenlial lucmo daled Oclnber 26, 1'I93 provides evidence thaithe administration believes in lln; cfliciicy i>f \.\K TRl, It stiittrs: •'Sharingvital TRl intormatioii with the piihlic has provided a siroiig incentive lorreduction in the generation, atid. ultimately, release into the enviroiimeniof toxic chemicals. Since the inception of the TRl program, reportedreleases to the environment under Tkl have decreased signihcantly."

era! events, including the introduetion of TRl reporting andsubsequent changes in reported releases over time. Thismethodology allows us to observe how house prices react tothe introduetion of new Information as well as how homeowners value changes in loxic releases. This approach has anumber of advantages over the cross-sectional hedonicmodels used in many earlier studies.' Our regressions userecently available, quality-controlled hou.se price indexesbased on repeat sales. These indexes, available at the ZIPcode level, are quite accurate and avoid the measurementerror problems of more commonly used median sales prices.(See Case & Shiller, 19K7, and Case & Mayer, 1996.) Also,the first-differences specification implicitly deals with theomitted variable problem that can lead to biased coefficientsiu a cross-sectional regression. Environmental variablesmay be correlated with unobserved town fixed effects (suchas proximity to tnanufacturing facilities or major polluters,the quality of housing, the amount of park space, or otheramenities). One can never be sure that the coefficient on anenvironmental variable is driven by a distaste for pollution,rather than the implicit value of some other amenity that iscorrelated with pollution. In the empirical work, below, weshow that a cross-sectional hedonic regression would obtaina quite different conclusion than our event study (fixed-effects) approach.

Another potential difficulty in using house prices toassess the impact of the TRl is that we do not observecommunity infonnation and expectations regarding toxicreleases. If communities know the level of releases in theirneighbt*rhood prior to the introduction of the TRl, houseprices should not change with the introduction of theTRl—no new information is bring provided. Similarly, ifcommunities anticipate the future path of toxic reductions,those reductions should be capitalized in advance of theactual reported reductions. A failure to find any capitaliza-tion of TRl information on house prices in either case couldbe misinterpreted as showing that communities do not careabout toxic emissions. We approach this problem in threeways. First, we use data on newspaper readership to controlfor differential local access to information about toxicreleases. Next, we attempt to model expectations directly,and explore whether house prices respond to deviationsfrom (modeled) expectations. Finally, recognizing the dif-ficulty in fully modeling expectations, we explore the extentto which changes in house prices lead annotincements ofchanges in toxic releases, as would be the case if commu-nities anticipated future changes in emissions.

Our primary hnding is that information about toxic re-leases had little impact on local house prices. This result isinconsistent with the hypothesis that community reaction tothe TRl has led to large declines in toxie releases. Inparticular, we find that (I) poorer, blue collar areas are more

' Some recent exceptions are Bui (2(H)2). Konar and Cohen (1997), andKamilto[! (1995). which use event studies based on stock market valua-tions. These are discussed further in section III,

REGULATION AND CAPITALIZATION OF ENVIRONMENTAL AMENITIES 695

likely to have loxic emissions and (2) larger reduelions intoxic emission.s occur in higher-value ZIP codes and ZIPcodes w\lh higher initial releases. However, (3) reductionsin releases are unrelated to any political economy factorsthat might relate to the ability or willingness of a commu-nity to organize against major polluters. When measured bythe initial level of toxie releases reported in 1987, (4) theintroduction of TRI reporting had virtually no effect onhousing prices, and—even more surprisingly—(5) subse-quent reductions in aggregate reported emissions between1987 and 1990 have no significant effect on house prices,either in the aggregate or when disaggregated into the mosthazardous types of chemicals or the most noxious airemissions. The standard errors are sufficiently small that wecan reject, with 95% eonfidenee, the hypothesis that de-clines in toxic releases of one standard deviation above themean are associated with more than a 0.5% increase inhouse prices. Finally, informational differences about TRIor expectations do not appear to explain these results. Weshow that (6) the above lindings hold even if we look at theunexpected information in TRI releases. (7) changes inhouse prices do not forecast changes in toxic emissions, asmight be expected if communities anticipated increases orreductions in toxic releases, and (8) these results persisteven in communities with high newspaper readership.

The paper is organized as follows. In section II wedescribe the TRI and provide background on reported TRfreleases over time. Section III provides an overview of theexisting literature that examines the use of property valuesto capture the value of different community amenities, andsection IV describes the data. Sections V and VI contain theempirical results, and the final section provides an interpre-tation, together with concluding comments and discussionof further avenues for research.

U. Background

Before 1986, when the Superfund law was revised as theSuperfund Amendments and Reauthorization Act (SARA),there was no systematic tracking of toxic releases in theUnited States. Under SARA, the Emergency Planning Com-munity Right to Know Act (EPCRA) was introduced.^included in the EPCRA is a provision known as the ToxieRelease Inventory. The TRI requires manufacturing tirms toreport their releases of listed toxic substances to the EPA forpublic disclosure.

The motivation behind EPCRA was twofold. In part.EPCRA was a response to a number of environmentaldisasters involving toxic substances, the best known ofwhich occurred in Love Canal, New York, and Bhopal,India. What became evident after those episodes was that,because residents were often unaware of what toxic sub-stances were being used by local plants, communities wereunprepared to deal with accidental releases. The EPCRA

•' The EPCRA Js also referred lo as Title Itl of .SARA.

requires that communities prepare emergency procedures todeal with the accidental release of any toxie substanceknown to be present in the area. This requires firefighters,hospitals, and the police to know both the storage locationof toxie substances in their community and how to handledifferent types of toxic substances in emergencies.

The second motivation of the EPCRA through the TRI,was to reduce toxic releases without formally regulatingpolluters. Up until that point, toxic releases had not beentouched directly by any existing environmental regulation.^Regulators hoped that by forcing private firms to disclosetheir toxic releases and by providing the public with thatinformation, firms" polluting behavior would be affected.

Under EPCRA section i 1023, a facility must report to theTRI if it is a manufacturing plant (SIC - 2000-3999) withmore than 10 full-time employees that either uses or man-ufactures more than a threshold level of any of the listedtoxic substances.'' Plants are required to file their reportswith their state EPA. This information is then collected bythe federal EPA and is made available to the public. Plantsmust file a separate form for each toxie substance for whichthey emit more than the specified threshold reporting level.Threshold reporting levels differ for manufacturers of atoxie substance and also differ across substances. Thresh-olds ean also change over time.

Between 1988 and 1999 EPA reports that toxic relea.sesfell by approximately 40%. In 1999, approximately 23.000different facilities submitted reports to the TRI. Nationwide,more than 2.3 billion pounds of toxic releases were re-ported.'' These releases include extremely toxic substancessuch as arsenic, mercury, and dioxin. as well as more benignsubstances (at least in small quantities) such as ammoniumsulfate. acetone, and sulfuric acid. It is important to note thatall of these substances are considered to be hazardous tohuman health at fairly low levels of concentration andexposure.

The TRI is meant to provide information to the public.The official release of TRI data is made by the EPA.Initially, hardcopy versions of the yearly releases eould beobtained directly from the EPA (at no cost) by individuals.Now. the data are made available on CD-ROM (or diskette)and on the Internet from the EPA or other environmentalsources, for example, the Right-To-Know Network (RTK-Net). The RTK-Net started operations in 1989 and is oper-ated by the OMB Watcfi and Unison institute (Washing-ton, DC). The TRI provides information on all reporting

^ There c\isl air polhilion stanJiird.s lor liazardous air pollulaius knownas the Naiioiial Eniissuins Standards for Hazardous Air Pollutants(NESHAP). but it is universally agreed upon that this panicular regulationhas been a)nipletcly unsuccessiul.

''During otir period oi study, [here were a stabie sei oC approxitnately300 listed loxic siibsianccs. This number has increased significantly since1995 to over 600 iisted substances. Several substances have also beendelistcd over the years, allhough not during our sample period.

' This includes emissions only from llic original induslries Ihai wererequired to report in 1987 and includes emissions from chemicals thatwere listed alter 1995.

696 THE REVIEW OF ECONOMICS AND STATISTICS

facilities as reported, including name and address of thefacility, type of chemical released, and amount released (inpounds). The hardcopy version of the data also includesummary statistics. After 1993, data were provided onwhether the substance was carcinogenic or had develop-mental or reproductive consequences.

The primary source of information to households, how-ever, was probably not through the raw data release itself,but from the media. Media accounts of TRI releases havebeen numerous since TRI data have become available. Asimple count of news reports in major newspapers between1988 and 1995 on LexisNexis was over 430. Some of the.sereports are very detailed in nature. The Los Angele.-i Times,for example, provided their readers with a list of the "Lead-ers in Toxic Releases,"^ which included a summary of whatsubstances were being released, tbe health consequences ofexposure to the substances, and tbe names and addresses ofthe facilities responsible for the emissions. In addition,community newspapers, which do not show up in Lexis-Nexis, are a well-known source of local information on TRI.An informal survey of local community newspapers InMassachusetts suggests that there were numerous stories onTRI during 1987-1992. However, these local newspapersdo not store their past stories in an electronic medium suchas LexisNexis, making it nearly impossible to obtain a moreprecise measure of their coverage during our sample period.

III. Previous Literature

The first work to incorporate environmental amenitiesinto the study of residential property values is Ridker andHenning (1967). Controlling for a number of property- andregion-specific characteristics, they regress mean propertyvalues by Census tract on air pollution measures. Ridker andHenning find that air pollution, as measured by sulfaleconcentration, has a significant effect on property v a l u e s -more so than measures of school quality and travel time tothe center of the city. Furthermore, they estimated that ifsulfation levels were reduced by at least 0.25 mg but notmore than 0.49 mg, households in tbe St. Louis metropolitanstatistical area would be willing to pay as much as$82,790,000 for tbat improvement in air quality.

Since 1967. there have been several other empiricalstudies tbat have found evidence that environmental amen-ities are capitalized in property values. Bednarz (1975)examines the relationship between individual selling pricesof land and aggregated values (to tbe census tract level) ofair pollution measures in Cook County (Chicago), Illinois.Bednarz finds ihat pollution decreases property values andthat the correlation between paniculate matter and theproportion of blacks in a community is 0.5, suggesting atleast tbe possibility of "environmental injustice." Greenbergand Hughes (1992) study the impact of hazardous waste

'̂ Hudson, Berkeley. Los Angele.s Tmies. December 3. 1991 (GlcndalcSection, part J: p. I. column 2).

Superfund sites on median housing prices in New Jersey.They focus their attention on 77 New Jersey communitiesthat have Superfund sites and 490 communities without anySuperfund sites between 1980 and 1988. Tbe authors findsome support for the hypothesis that Superfund hazardouswaste sites depress housing prices (relative to the controlcommunity with no Superfund sites), particularly if locatedin rural communities and in communities that had thehighest rate of price increases in the preceding five-yearperiod. Surprisingly, however, the level of risk associatedwith the hazardous waste site was not found to be associatedwith changes in sales prices. This finding is consistent withthe hypothesis that, although communities may be aware ofthe existence of a Superfund site in their locale, they do nothave the ability to evaluate the relative dangers associatedwith higher- or lower-risk sites. A recent paper by Gayer,Hamilton, and Viscusi (1998), however, finds that commu-nities are able to respond to differences in cancer risk, asmeasured by changes in house prices, when that informationis provided to them.

Both the methodology and the interpretation used byRidker and Henning (and, consequently, others) have beencontroversial, spawning a large number of studies thatsupport and that refute their results. [See, for example.Freeman (1971. 1974), Small (1975), Polinsky, Mitchell,and Sbavell (1975). and Harrison and Rubinfeld (1978).]Regardless of tbe controversy, hedonic studies on propertyvalues of the type that Ridker and Henning pioneered arenow commonly used to measure bow changes in environ-mental amenities are valued. Smith and Huang (1995)conducted a meta-analysis of 37 studies to evaluate therobustness of this methodology and find that there is aconsistent relationship between tbe marginal willingness topay for a reduction in air pollution and the level of pollutionin the local region. Kiel and Zabel (2000) provide morerecent estimates for the marginai willingness to pay for airquality in four major cities in the United States. Tbe authorsestimate that the benefits of achieving the National AmbientAir Quality Standards between 1974 and 1991 in those fourcities are between $171 million and $953 million.

Tbe vast majority of these studies use a cross-sectionalhedonic methodology in which they regress house prices onvarious observed attributes plus an environmental variablewithin a given city. As Gyourko, Kahn, and Tracy (1999)point out. ". . . spatial sorting on unobservables" maypresent a problem. After all, "If suppliers build nicer homesin terms of unobservables in the nicer parts of the city, thenthe econometrieian will overestimate the value of the QOLIquality of lifel. Moreover, low environmental qualitywithin a neighborhood may proxy for low quality of hous-ing structure" (p. 1438).

A few studies look at changes over time in the capitali-zation of news about a specific amenity, such as highwaynoise in the Seattle suburbs (Palmquist, 1982), PCBs (Men-delsohn et al., 1992), a waste incinerator (Kiel and McCain,


1995). or a hazardous waste sile (Kohlhase, 1991), avoidingthe problem of unobservable quality. Palmquist (1982). inparticular, ha.s a niee discussion of the potential advantagesof such an approach. Al! o( these studies find that changesin news about the toxicity of the site are capitalized intohouse prices. One should use eaution in applying the resultsof Ihese studies more generally, however, as most thesestudies look at specific sites that usually received muchmedia attention.

Probably ihe most reliable estimates on the capitalizationof environmenUil amenities in house prices, however, comefrom a recenl study by Chay and Greenstone (2()()0), whoshow that a one-unit reduction in suspended particulatesresulted in a {).4^/r-().5% increase in house prices. Unlikeprevious papers. Chay and Greenstone use regulatorychanges during the 197()s as instruments to control forpotentially endogenous ehanges in emissions.

Below, we e.xamine the release of inrormation aboutactual releases under TRI. whieh allows us to examinethe eflecl of information about pollution on house values.Three recent studies use an event study approach loevaluate the stock market response to TRI information.Hamilton (1995) linds that firms with higher reportedreleases have larger abnt>rmal negative returns. Konarand Cohen (19971 model expectations and Iind that firmswith Liuexpeciedly high releases have lower stock marketreturns. Both studies conclude that (I) the TRI providednew infonnation to the market and that (2) the stockmarkei responded to that information. In eontrasi, how-ever, Bui (2002) elaiins that there may be an error in theestimation of the standard errors used by Hamilton andby Konar and Cohen, which casts doubts on the robust-ness o\ their findings. Bui linds that for petroleum com-panies small negative abnormal returns exist only for theinitial TRI announcement of 1987 emissions and thatthose returns are not significantly different from thereturns for nonreporting petroleum companies in anyperiod.

Finally, a recent study by Oberholzer-Gee and Milsunari(2002) looks at TRI-emitting plants in the Philadelphia areausing detailed location data and find that, if anything, theprices of houses located nearest TRI-emitting plants mayhave risen a little bit on the initial announcement of TRIemissions and the initial announeement had no impactbeyond a quarter of a mile. This study has the advantage oiusing detailed house locations, but has a much smallergeographic area and is limited to the iirsl year of TRIannouncements. "Fhe results of that paper are complemen-tary with whal we Iind below.

IV. The Data

We use data from several dilferent sources, including:plani-level data on toxic emissions reported under the TRIfrom 1987 to 1992: repeal-sales house price data from 1982to 1993: and \arious community characteristics from the

1980 and 1990 Censuses and the Commonwealth of Mas-sachusetts. The separate data sets are linked either by ZIPcode or by eommunity name.

Plant-level data on toxic emissions are taken from theTRI.'' Between 1987 and 1993, over 2000 different toxieemission reeords were filed annually in an average of 231different ZIP eodes in Massaehusetts. Over the sampleperiod, an average of 24,876,500 !b. of toxic substanceswere released into the environment every year. We aggre-gate the emissions data up to the ZIP code level and focuson the sum of environmental relea.ses by substance, underthe assumption that communities have a similar valuationfor toxic releases that occur through different environmentalmedia.'" However, we create separate categories for sub-stances that are carcinogenic or have adverse developmentalor reproductive effects on humans." Summary statistics areprovided in Table I. ZIP eodes that do not report any toxicreleases are given a value of zero.'-

A (Ota! of 144 different toxic substances were reported asTRI releases between 1987 and 1992 in Massachusetts. Ofthese, 18 (I2.5'7r) were known to be carcinogenic, 21(14.6%) were known to have adverse reproductive or de-velopmental consequences, and 12 (8.3%) were known tohave both carcinogenic and developmental or reproductiveconsequences."

Between 1987 and 1992. reported toxic reiea.ses in Mas-sachusetts fell by 70%. Similar reductions are found forcarcinogenic releases (a 60%.' deerease). Substances withknown adverse developmental or reproductive implieationsdecreased as well, but to a lesser extent (19%). The numberof reeords actually filed (separate releases) in the TRI wentfrom 2129 to 1654, a decline of 22%. The average reportedsingle toxie release in 1987 in Massaehusetts was 32,636 Ib.The amounts released ranged from a low of I Ib. to a highof 1.019,600 Ib. Both the mean release and the range felldramatically in 1992 to 9674 Ib. with a range from I Ib. to502,1.57 Ib.'

The house price indexes used in this paper are obtainedtVom Case. Shilier and Weiss, Ine. and are estimated using

'' h should be noied here that Ihe TRI provides daUi that is scll-ieportcdby Ibe poiluling tirins and is not verified |or easily veiitiable) by inde-pendent third parties. There may be reason Io believe Ihat tbe data maynol. in lact. be very aeeur;itc. Syslenialie o\'er- or undcrestimaiion ofactual releases, however, does not appear lo be a problem. For ourpurposes, whalever [noasurement errors may c.\ist in the reported data areol' liitle imporiance, because il is the reported TRI values that are providedto Ihe public, and presumably thai Is the inf\)r[iialion on which householdsbase llicir response.

'" The sum ol enviroiiinenlal releases equals the sum cil' air -I- water 4-land releases. In et)mparison wiili lhe so-called criieria air pollutants,toxic releases have a more localized etTect on their surroundings, sostudying Ibeir el'feeis at the ZIP eode level is not unreasonable.

'' Only substances Ihat are "known" by the liPA or California EPA to becarcinogenic or have developnienta! or reproduetive ba/ards were tagged.

'- In tbese / I P codes, either loxic releases fall below the thresholdreporting level (and may be equal to zero), ov plants fail to report releases,

' ' A list of all toxic substances covered under the TRI that were reportedas releases i[i Massachusetts between I9S7 and 1993 may be obtainedfrom the authors.


TABM. - S S : M M A H Y Of- T R I Rii.i;ASi;s' '

Year

198719881989199019911992

Change1987-1992

•' All releas,!s i r e meii'.urL-d

United States

AggregateReleases

I6.3l4..32t.8545.676.700.2694,446,143.0383.751,675.3943.499.710,9183.319,222.779

-80%

AggregateReleases

50.029.90031.791.50726.379.90922.210.87718.1,32.203I5.44,S.165

-7O'/r

CarcinogenicReleases

5.491.7834.712.8414.121.1112,9.36,7522.411.3212,207.644

-60^.

Massachusciis

Developmental orReproductive Releases

738.390442.136472.059445,278466.084600,019

-l9'"/(

NoxiousReleases

2.614.8772.274.4111.313.4921.574.979

565.7671,060.847

-59%

a variation on the weighted repeat sales methodology firstpresented in Case and Shilier (1987).''* Because the indexesinvolve repeat sales of the same property, they are notaffected by the mix of properties sold in a given time periodor differences in average housing quality across communi-ties. In Massachusetts, indexes were estimated for 247 ZIPcodes with a sufficient number of transactions to obtainreliable estimates after 1982. All price changes are mea-sured from the second quarter of each year to correspond tothe timing of the release of TRI data. Nominal houses pricesin Massachusetts towns increased by an average of \16%from 1982 to 1989, and then declined by 10% in the nextfive years. However, the price increases and decreases werequite unevenly distributed across communities during thisperiod, ranging from 132% to 330% in the earlier timeperiod and between a 38% decline and a 4% increase in lhelater time period. Earlier work (Bradbury, Mayer, & Case,2001) shows that shifts in economic variables such as em-ployment, aggregate school enrollments, demographics, andchanges in fiscal factors such as Proposition 2 - are economi-cally important and statistically significant in explaining thecross-sectional variation in changes in house prices.

We also use data on newspaper readership lo explore theextent to which capitalization depends on information. Weobtain average newspaper readership from 1995 to 1998from the Audit Bureau of Circulations.'-^ Although the datesof the data do not exactly correspond to the dates of oursample, we would expect that newspaper readership willremain stable over time, or at least the rank order oflocalities will not change much.

Demographic data are laken primarily from the 1980 or1990 decennial Censuses. In many cases, data aie onlyavailable at the town level, and are attributed to ZIP codeson the basis of town. These data include median income and

'•• The method uses arithtnetie weighting described by Shilier (i 991) andis based on reeurded sales prices of all properties that pass throtigh themarket more than onee during the period. The Massachusetts file containsover 135.000 pairs of sales^drawn between 1982 and 1994. First, anaggregate index was calculated based on all recorded sale pans. Next,indexes were calculated for individual Juri.sdietions.

'̂ Source: Circulalion Duht Rank. New York: Audil Bureau of Circida-lions (1999).

housing values, and age variables. School test scores andtown unemployment rates, spending on health and welfare.and manufacturing employment are obtained from the Com-monwealth of Massachusetts, land available for new con-struction from the University of Massachusetts, and data onnew construction frotn the U.S. Department of Commerce.

V. Where Are the Toxic Emissions, and Where do theDeclines Occur?

We begin the analysis by documenting the characteristicsof neighborhoods that surround plants that produce toxicemissions. (See Table 2, left panel.) Not surprisingly, plantsihat emit TRl-M.sted substances tend to be located in com-munities with lower median incomes, house values, andschool assessment test scores than communities with noreported toxic emissions. Communities with toxic emissionsare smaller and have a higher percentage of workers inrnanufacturing, but are quite similar in age distribution andpercentage of minority residents. Thus children are no moreor less likely to live iti neighborhoods with higher toxicemissions. In addition, communities with higher toxic re-leases have residents that are less likely to bo college-educated, and more likely to be registered as Democrats, butdo not differ in their newspaper readership. More of theplants are located not in the Boston metropolitan area, butfurther out from the city, suggesting locations in manufac-turing subcenters such as Lowell, Lawrence, and NewBedford, rather than in ruial areas or downtown Boston. Interms of housing appreciation, the groups in the left andright columns appear to be quite similar, both before andafter TR! reports began in 1987.

Next, we divide communities that reported positive TRIemissions in 1987 into iwo groups based on their subse-quent change in emissions. The right panel in Table 2summarizes the characteristics of communities whosechange in cmissitms between 1987 and 1992 was belowaverage for all (Massachusetts) TRI-reporting communities.Notice that reported toxic emissions actually ro.se in thetypical community within this group, indicating that de-clines in emissions were not uniform across locations. This


TABLK 2.—MASSACHUSETTS TOXIC EMISSIONS BY ZIP CODE FOR SELECT VAUIABLES

Emissions in 1987 by ZIP CodeChanges in Emissions 1987-1992

by ZIP Code

Statistic

NonzeruEmi

Mean

218.213

18.947.9

49,4610.04

0.290.160.430.270.35

25880.36

20.31.580.08

-0.110.34

3.7

ssion.s

•Std. Dev.

407.521

4.813.5

75,8580,06

0.120.090.140.030.031790.48

16.50,170.060.070.10

16.6

ZeroEmissions''

Mean

21.756.5

52,9190.04

0.280.2.30.350.280.35

26790.52

22.81.550.07

-0.090.29

7.2

Std. Dev.

6.822.7118,580

0.07

O.ll0.130.0150.040.04185

0,5017,40.230.060.080.10

27.9

Small Reductionsor IncreasedEmissions

Mean

41.16524,059

18.547.2

52,0130,05

0.270.160.430,270.35

25750.41

17.61,580.08

-0.110.33

-0.184.0

0.31

Std. Dev.

89,77065,896

4.712.9

77,6310.07

0.990.090,140.030.031890.49

15,70.190.070.07O.IO

0.1517.0

0.78

Large ReductionsIn Emissions

Mean

397.863-287,408

19.248.6

46.8710.03

0.300.160.430.280.35

26000.32

23.01.580.08

-0.100.36

-0.143.2

0.26

Std. Dev.

513,701404.328

4.914.074.5030.05

0.140.100.0140.030.03169

0.4717.10.150.060.070.11

0.1716.5

0.51

Average reported TRl {lb.)Average change in TR! in 1987-1992 (lb.)Median household income in 1980 {(K)Os)Median house price in 1980 (OOO.s)Town population in 1980Percentage of minority residents in 1980Avg. daily newspaper circulation per capita, 1995-

1998Percentage of residents with a college degree in 1980Percentage of regislered voters who are DemocratsPercentage of residents aged 35-60 in 1980Percentage of rosidenis under age 2! in 1980Average math and reading assessment test scoreDummy variable if in the Bosion metro areaDistance from Boston if in the Boston metro areaPercentage change in house prices, 1982-1987Percentage change in house prices, 1987-1989Percentage change in house prices, 1989-1994Percentage of residents in manufacturing secior. 1980Percentage change in manufacturing employment

between !987 and 1992Health and welfare spending in 1987 ($ millions)Percentage change in health and welfare spending

between 1987 and 1992

N 137 10 fi9 68' ZIP i:(x)es th;il do nol lepmt j ny TR l releases bul may have actual enijssji>n': grealer ihan U,

finding casts doubt on the hypothesis that declines in TRlemissions are strictly due to firms intentionally overreport-ing emissions in 1987—over one-third of the ZIP codes inthis sample experienced an increase in reported emissions.

If community activism were an important factor in con-tributing to declines in emissions, we might expect greaterdeclines in toxic emissions in communities with morecollege-educated or higher income residents, in communi-ties with more children, or in places with a higher percent-age of registered voters or higher newspaper readership. Yet,the two groups differ little in their average income, votingpatterns, education, age distribution, or any other factorsthat might be related to political or community activism.

Communities that had the largest absolute reductions inreleases tended to have slightly higher house prices andlower minority populations, were more likely to be locatedin the Boston area, and had a work force that was moreheavily concentrated in manufacturing. The data also showvery few differences between the communities in theirpopulation growth or housing appreciation during any of theperiods in question.

Because the correlations above are based on simple com-parisons of means, we estimate regressions to identify thecharacteristics of communities where TRI-emitting plants arelocated, and where emissions fell furthest. These regressionsshould be interpreted in a strictly descriptive manner, not assuggesting any form of causality. The results, reported in Table

3, are mostly consistent with the findings above. Holding otherfactors constant, TRI-emitting plants are more likely to belocated in poorer, Democratic-vofing communities, with moremiddle-aged residents, and further from Boston. In otherwords, these are solid blue collar communities.

Yet few of the above factors are particularly characteristicof communities with the largest declines in emissions. (Seecolumn 2.) In fact, only the median house price is correlatedwith changes in toxic releases—greater declines occurred inhigher-priced communities. Political economy variables orfactors that might relate to the ease or probability of polit-ical activism do not appear to matter. The third columnconfirms a strong trend towards reduced emissions in areaswith the highest initially reported releases. When control-ling for initial releases, the coefficient on median houseprice is no longer statistically different from 0.

VI. Estimation and Results

We now turn to house prices for direct evidence as to howhome buyers value the change in emissions that occurredafter the beginning of TRl reporting. To the extent thatpublic pressure is related to the subsequent fail in emissions,house prices should have increased in communities whoselocal plants successfully cut toxic releases. Even absentpublic pressure, evidence of capitalization will help showhow much the public values changes in reported emissions.


TABLL 3.—As.s(x.'iATioNS or TRi EMISSIONS

(1)

TRI Emissions. 1987

Tobit

Change in TRIEmissions.1987-1992

Ch;ingc in TRiEmissions,i 987-1992

OLS OLS

Dependent Variable Coef. Std. Err. Coef. Sid. Err, Coel. Sid, Err,

Median hoitscliuki incnnie in 1980 l()fX)s)Meiiiun house price in I'M) ((KH)slTown piipuhiliun in I9S()Pci'cenlage ol' minority residents in 1980Pcrccnlajic ol 1980 residents working in manutacliiringPereenlage of ivgisicred voters who are DemoeratsPercentage of adults who arc registered to voteAverage daily newspaper circulation per capita.

1993-1998 (dUOs)Pereenlage oi rcsidcius with a college degree in 1980Percentage oi residents under age 21 in 1980Pereenlage of rcsnienis aged 35-60 in 1980.Average math and reading assessment test scoreDummy variable if in the Boslon meiro areaDistance Ironi Boslon if in the Boston metro area(Uistanee from Boston il in Ihe Boston metro area)-TRI emissions in 1987 (000s)Constant (X lO" 'i

Log likelih(iod

Nimibcr of observations

-493.9

-0.00043

-394

763801

-756

356259780

4.618

0.3- 117

22.80,?l

-1.813

-2033.

247

25.8

6.10.00067

983465370601

308942

1.496

2,036

0-36

878.50.18

1.2X1

19

30.9

-13,1

-0.00027

286425•247

390

-178

449-1250

-2.669

• 0.39

126-28

0.017

1.940

0.13

137

26.56,9

0.00064

S74426331542

2669081483

1.981

0.34

847,80.17

1.171

-2.67

0,68

-O.O{R}I5

-181

-53.8

35.2

48.1

-227

74,2

499-574

-0.13

9.7- 1.57

0,009

-0.76

393

0-93137

7.87

2 08

O,O<K}I9

25712597,9

159.9

786267.5

439586

n.io24.9

2.29

0.049

0.02

347

A. Capitalization of Toxic Releases in House Prices

In this context, the equilibrium price of housing in com-munity / and base year T can be represented as'''

tt(i '^ U|(environmema! amenitiesj

^• ttj.(fixed amenities),^

+ a-i( housing stock),-^

+ Ujicconomic factors);,̂ + ^<i-

(1)

In equation (I), a i represents the value of environmentalamenities; in this case, toxic releases that are reported underTRI. The coefficietit a^ is the value to the marginal homebuyer of fixed amenities such as location and communitycharacteristics. The coefficient on the size of the housingstock («,) is expected to be negative as long as the supplyof new units is nt)t perfectly elastic. Zoning ordinances thatset minimum lot sizes, limit redevelopment at higher den-sities, and place restrictions on setting up new communitiesensure an upward-sloping supply curve within a metropol-

"> Epplc (1987) atid liartick (1987) provide critiques of Che basic hedonicequation in Rosen (1974). specitied above in equation (1). based on thelikelihood Ihat households simultaneously choose quantities and prices ofhotismg and land characteristics. Both papers suggest inslrutnents fromsuppliers and households that are not available in most appheations,incltiding ours. Epplc also suggests that identilicacion of hedonic modelscan be reached under a set of strong, but not completely unreasonable,assLimptiuns, 1iie subsequcni literature has generally accepted repeat-salesmodels as being a good approach to estimate capitalization in manycontexts, including environmental atnenities and school quality.

itan area, even in the long run.''' Finally, local economicfactors will also influence house prices, although they mayalso be determined simultaneously with house prices.

Our estimation procedure relies on an event study meth-odology, a variant of the process developed by Case andMayer (1996). In this context, we look for evidence as tohow changes in toxic releases are capitalized into houseprices. A simple first-differencing of equation (I) betweenthe base year T and a subsequent year / suggests thatchanges in house prices should be a function o'i changes infixed amenities, changes in environmental amenities,changes in the supply of units, and changes in local eco-nomic factors:

, — Pi, + 3i(Aenvironmcntal amenities),

(2)stock),

p4(Aeconomic factors), + |x,.

where A represents the change in the value between yearand the base year x.

'•̂ See Fischel (1990) for a summary of the literature on zoning. Ham-ilton (1975) shows that under a series of restrictive assumptions, includingperfectly elastic supply and zoning ordinances that control the exactquantity of housing eorisumption, there is no capitalization ot localamenities and thus « , . n^. and a , are equal lo zero. As noied by Fischel.these assumptions are not satisfted empirically.


Modeling changes rather than levels has the advantage ofremoving a considerable degree of fixed-effects differencesamong communities. By definition, changes in many fixedamenities, such as location and community characteristics.equal 0 and thus drop out of equation (2). This is important,as the maintained assumption is that the error term IJL, isuncorrelated with the included independent variables, in-cluding environmental amenities. For example, communi-ties with higher levels of toxic releases may have lower-qualily houses, fewer parks, or worse schools. To the extentthat these fixed factors are not fully observed, the estimatedcoefficient on environmental amenities in a hedonic equa-tion will be biased. However, this bias does not affectestimated coefficients in the difference equation (2) as longas the fixed amenities remain fixed over time.

If the coefficients in the levels equations (ao-a-;) alsoremain unchanged over time, then a, — fi,. Although many(fixed) amenities do not change noticeably over a two- tofour-year time span, which is the period for most of ouranalysis, these independent variables may stiil influencespending changes if their effects on spending levels (theircoefficients in the level equations) change during the period.For example. Case and Mayer {1996) show that demand forcertain local characteristics can change due to such aggre-gate shocks as an aging population or increasing schoolenrollmenls. Though their study covered a longer timeperiod than our analysis, we nonetheless include initiallevels of some town characteristics in the house priceregression to allow for possible changes over time in thecoefficients on these fixed attributes.

For example, the aging of the baby boom and the asso-ciated echo baby boom has led to an increase in publicschool enrollments in Massachusetts since 1990. The result-ing growth in the number of households with children inpublic schools has increased the demand for houses intowns wilh good schools. Similarly, a town's initial age mixmay be an indicator of amenities that are attractive tospecific age groups. During this period, most baby boomersentered the 35-60-year-old age group, raising demand foramenities typically valued at those ages, such as town-sponsored day care and after-school programs, and betterparks and playgrounds. Thus, communities with high initialconcentrations of middle-aged households would be ex-pected to experience a relative increase in housing demandand. hence, house prices.'^

Thus we estimate the equation below, which allows forthe value of some fixed amenities to change over time:

'** The iihiliiy ot jurisdictious lo replicate desired anieniiies and housingtypes could counienict ;my impact of the aging of the baby boomgeneration on cjoss-seciional house prices. Since the baby boom's move-ment into middle age cuuld be easily torecast, an eflicient housing marketwould ciiusc prices in towns that appeal to baby-boomers to bavc risen inprevious periods in anticipation of baby-boomers entering the housingmarket. This is purticulariy true of iowns with a large number of trade-uphomes, whicb are very expensive to replicaie given the limited supply ofundeveloped iand in many metropolitan areas.

-j-ii -I- 7i(AenvironmentaI amenities),

+ 72(fixed amenities),

+ 73(Ahousing stock),-

+ 74(Aeconomic factors), + |JL,.

(3)

Environmental emissions are measured either in a singleyear to pick up iui initial announcement effect, or betweentwo years to capture the impact of changes in emissionsover lime. Fixed amenities, including school test scores, thepercentage of middle-aged households, and a locationproxy, are assumed to be constant, so we use the value fromthe base year (T), or in some cases an earlier year when thesevalues were taken from the U.S. Census.'*^ We use thesupply of new housing permits as a close proxy for Ahous-ing stock, but we instrument for new permits with theamount of vacant land available in 1984 and lagged permitsto control for potential endogeneity problems.-*^

We also face potential endogeneity problems when con-sidering changes in economic factors, such as (town-level)manufacturing employment, the unemployment rate, andspending on health and welfare. These variables help correctfor a potential bias, because changes in toxic emissionsmight be negatively correlated with employment changes.That is, firms that reduce output will lay off workers andalso reduce emissions, or vice versa. The beneficial impactof reduced emissions on house prices might be offset byreduced demand for housing from laid-off workers. Sincethe primary purpose of this paper is to explore the impact ofTRI emissions on house prices, we are not interested instructural estimates of the coefficients of the economicvariables. Yet the economic variables might be simulta-neously determined with house prices, for example throughemployment changes in the construction sector, and thustheir inclusion can bias the coefficients on the environmen-tal variables. Again, such a bias is unlikely given the lags innew construction and the short time period of our study, butwe take it seriously nonetheless.

Below, we use three alternative approaches to deal withlocal economic changes. First, we include the employmentvariables directly in the equation, recognizing that the co-efficient estimates on these variables may be biased, butexpecting that such simultaneity biases might not affect theenvironmental variables, especially inasmuch as we have

'•̂ Our meastire of test scores cannot be compared across years, so we usean initial value rather than the last year to control for the (slim) possibilitythat changes in house prices might have affected the quality of studentsliving in a community, although such an effect is quiie unlikely over theshort period of time used in our regressious. Also, the rank order ofcommunity test scores changes little over the sample period, which isconsistent with the view that school quality is fixed over short periods oftime. Due to the infrequent occurrence of the census, we also do nol havemultiple observations on Ihe percentage of middle-aged households,

-" Although we are not interested in the coefficients on change inhousing stock atid change in economic factors for this analysis, we correctfor potential endogeneity so as to mitigate any correlations between theremaining error term and environmental amenities.

702 THE REVIEW OE ECONOMICS AND STATISTICS

already included instruments for local supply. Second, weinclude a proxy for employment that i.s the number ofworkers in the manufacturing sector in 1980. but drop thepotentially endogenous changes in economic factors. Thisproxy would be appropriate if ail communities had the sameproportional drop in manufacturing employment. Our thirdapproach is to instrument for the changes in economicfactors with lagged changes in economic factors. Althoughlagged values are not the ideal instruments, they are the onlyinstruments that we can find at the local level. Since serialcorrelation might be a problem that biases instrumentalvariables estimation with lagged values of the endogenousvariables as instruments, we also use longer lags. We reportestimates using the first approach, but all of the resultsbelow with regard to the TRI variables hold just as stronglywhen we use any of the other instrumental variables as inIhe third approach.-'

To begin, Table AI in the appendix presents estimates ofthe house price regressions without considering environ-mental amenities. The basic regression coeflicients are ofthe expected signs, are almost all statistically significant,and are consistent with the coefficients of Case and Mayer.These results provide support for the hypothesis that aggre-gate trends change the capitalized value of school qualityand location over time. In particular, houses in towns withgood schools (as measured by 1988 test scores) actuallyincreased in value from 1989 to 1992, as aggregate schoolenrollments rose following national demographic factors.House prices also grew faster in communities with betteraccess to downtown Boston and its suburbs, where jobgrowth was strong compared to other parts of the Common-wealth over this period. Local construction is negativelyrelated to changes in house prices, as anticipated. Einally,the coeflicient on change in the unemployment is negativeand signilicant, as expected, and the change in health andwelfare spending has a positive relationship with houseprices, as might be anticipated if state transfers raise localincome, although only significant at the 90"' percentile.Controlling for the overall unemployment rate, the coeffi-cient on change in manufacturing employment is highlyinsignificant.

B. Cross-Scciumal Relationship helween TRI Releasesam! House Prices

Before beginning the anafysis using our differencesmodel described above, we expfore the refationship betweenhouse prices and TRI releases using the cross-sectional datathat have been utilized by many earlier studies. In particular,we take median house prices from the 1990 Census andcompare those prices to amount of toxic releases reportedby TRI in i 990. As expected, in the raw data, aggregate TRIreleases have a -0.12 correlation with median house prices

•' Tilt; previous version o! our paper reporled estimates Ironi lhe secondmelliod. All esiimales yre available Irom the aiitliors upon request.

(/7-value 0.05). Communities with higher house prices havelower TRI releases. In a regression with log of medianhouse prices in 1990 as the dependent variable, the coeffi-cient on the log of TRI releases is negative and highlysignificant (/?-vaiue 0.001), suggesting that a 10% increasein TRI emissions leads to a 0.14% decrease in house values.Though small in size, that coefficient remains statisticallysignificant even when controlling for the percentage ofmanufacturing workers in a community (/)-value 0.005).Thus we can confirm previous findings that toxic releasesare negatively correlated with house prices in a cross-sectional hedonic regression. As we will see below, how-ever, this result will not hold up in the difference estimationand is likely a function of unobservable variables.

C Efi'ect of Initial TR] Releases

We begin by supplementing the basic (differenced) houseprice regressions from Table AI with a variable measuringthe initial TRI releases in 1987. The interpretation for thecoefficient on this variable is similar to that of other vari-ables in an event study; it measures the effect of theannouncement of TRI releases on house prices relative toexpectations. Thus, an insignificant coefficient might reflecteither that the public does not place a large negative valueon toxic releases or that the announcement was in line withprior expectations.

Table 4 presents the regression coefficients on TRI vari-ables using a number of different specifications. All equa-tions also include the other variables listed in Table Al,which are not reported here for reasons of space. Regressioncoefficients on these other variables change little from thosereported in Table Al. The dependent variable measureshouse price changes over a one-year period, reflecting thepossibility that house prices adjust slowly to new informa-tion. The results are quite similar if we use a two- orthree-year window, although these longer windows mayincorporate information from subsequent TRI releases andthus do not provide as clean a test.

The coefficient estimates show that TRI releases had verylittle impact on house prices over this time period. TRIreleases are made public with approximately an 18-monthlag. In this case. TRI information from 1987 was madepublic in the summer of 1989. Thus the dependent variablemeasures house price changes from the second quarter of19H9 through the second quarter of 1990, a full year. Thetop panel explores changes in house prices following theinitial announcement and finds no statistically significantimpact of the initial announcement on house prices. In fact,both the coefficient and standard errors suggest that theeconomic impact of TRI information is exceedingly low.The estimated coefficient indicates that a one-standard-deviation increase in announced aggregate TRI releases(408,521 Ib. among communities reporting positive emis-sions) leads to a 0.095% decline in house prices over twoyears. Even looking at a two-standard-deviation confidence


TABI.[-. 4,—2SLS RLGKiissn)NS ON TRI FMISSIONS IK I987''

iasc tiquaiionAlternative Eqiiaiion wifli

SiiiTOundirm Towns

Specification Coef, Std, tirror Cocf, Std. Hrror

•asc Equation: Communitieswith Highest NewspaperReadership (;V = 124)

Sid, Error

Announcement Effect of Initial TRi EmissionsDependenl Variable: Change in House Prices between I989Q2 and I990Q2

Aggrtr j ia ie T R I e m i s s i o n s m I 9 S 7 (X 10

C a r c i n o g e n i c e m i s s i o n in iy}<7 ( x ! 0 ^)

Adverse developmental or reproductiveemissions in 1987 (X 10 '*)

N o x i o u s e m i s s i o n s in I9H7 ( X I O

-2,32

-0,22

1,21

-2,19

4,10

1,74

5,45

2.26

0,20

0,41

0,70

3,57

2,99

1,,17

4.46

1,95

-0,033

0,.̂ 2

- 1,04

0,071

4,72

1.91

5,79

2,93

Contemporaneous Effect of Initial TRI EmissionsDependenl Variable: Change in House Prices between I987QI and I988QI

Aggregate TRI emissions in 1987 (XIO

C a r c i n o g e n i c e m i s s i o n in !9S7 ( X l O '^)

A d v e r s e d e v e l o p m e n t a l o r r e p r o d u c t i \ e

e m i s s i o n s in 1987 ( X I O ")

N o x i o u s e m i s s i o n s in 1987 ( X l l ) **)

-3,23

3.10

-2.07

-1,11

nihlcAl. y.hwh l̂ iliL- 1

6,64

2,76

8,74

3,63

use SpOLilkillilJE!

-3.64

1.77

4.24

- 1,09

5.11

2.55

7.46

2,93

[« cwflkiews fri,"!! hcpjralf KV

-2,80

0.21

- 2,68

2,99

rosMons. Nuiiihcr

7,.S9

3,07

9.52

4.75

lit i,"ihstrv;moiis ISA l l ^:l|l^y(h^>Il^

T f nc' lcd

L' Ihc iilliLT iii(k'poiuk-iil v.iniihks lis[i:il ill

inlervai around ihis estimate, we can reject that a 408,521lb. decrease in emissions leads to more than a 0.33% rise inhou.se prices.

To confirm these results, we investigate other possiblealternatives. Many of the toxics covered by TRI are rela-tively benign, but a few have a particularly strong effect onpublic health. In the lower part of the top panel, we onlyinclude measures of chemicals that have a strong link tocancer or developmental and reproductive problems, repre-senting some of the most hazardous chemicals covered byTRI. Again the coefficients are quite small and the standarderrors show that one-standard-deviation changes in reportedemissions of these severe toxics have a very small effect onlocai house prices. In both cases, we can reject (with 95%confidence) that a one-standard-deviation decrease in an-nounced toxic releases leads to more than a 0.16% increasein house prices over a year. The third set of regressions inthe top panel considers emissions of noxious chemicals, thatis. chemicals that emit distasteful odors and are thus mosteasily observable to the general public. The coefJicient isnegative, but the standard error is larger than the coeflieientestimate, and the tight standard errors allow us to reject anymeaningful effect of noxious emissions on house prices.

Next, we examine the possibility that TRI emissions havean impact beyond the immediate ZIP code where the plantis located. The middle column reports results in which weadd one-half of Ihe emissions from immediately surround-ing communities to the emissions in a given ZIP code. Inthese specitications. 218 of the 247 ZIP codes are nowmeasured as having some po.sitive exposure to emissionsfrom plants in them or in a surrounding ZIP code. This

specification may be reasonable it TRI emissions travel tosurrounding locales thru the air or ground water. Yet in threeof the four rows, the coefficients drop in size, and in alleases the standard errors fail. None of these regressioncoefficients is close to conventional significance levels. Thedecreased standard errors suggest that TRI would have hadan even smaller effect on house prices than in the firstcolumn. We have also tried other weights on emissions insurrounding communities, but have always found the sameresults, suggesting that the lack of impact of TRI announee-ments on house prices does not appear to be due to diffi-culties in defining the relevant affected area.

Another possibility is that some communities may bebetter informed than others as to the extent of the toxicreleases. After all. TRI covers all types of releases—ground.water and air—and many of the most hazardous chemicalsare invisible and odorless. To control for this possibihty. welook at communities whose newspaper readership is in thetop one-half of the sample. These regressions are reported inthe right-hand column of Table 4. Even communities withhigh newspaper readership do not have statistically differentcapitalization behavior. In regressions not reported here, weexamine alternative functional forms, including an interac-tion between newspaper readership and toxic releases. Innone of these specifications was the coefficient on a toxicreleases variable ever close to statistical significance.

We also ran regressions, not reported here, in which weremoved the communities with the largest emissions, pos-sibly supposing these communities already knew about theemissions in their area. Again, the results indicate thatreleases had no statistically significant impact on house


TABLE 5.—2SLS RECRESSiONS OF CHANGE IN HOUSE PRICES ON CHANGL IN TRI EMISSIONS ULIWLLN I9K7 AMI

Specification

Dependent

Aggregate TRI emissions in 1987 (X]0"**)

Carcinogenic emission in 1987 (X 10" **)

Adverse developmental or reproductiveemissions in 1987 (XIO~^)

Noxious emissions in 1987 (XU) ^)

Dependent

Aggregate TRI emissions in 1987 (XlO""^)

Carcinogenic emission in 1987 (XlO **)

Adverse developmental or reproductiveemissions in 1987 (XlO ^|

Noxious emissions in 1987 (XlO ^)

Base Equation

Coef. Std. Error

Alternative Equation withSurrounding Towns

Coef.

Announcement Effect of Initial TRI EmissionsVariable: Change in House Prices between I989Q2

1.30

-4.52

3.81

4.12

1.30

10.6

8.47

2.80

Contemporaneous Effect olVariable; Change in House

-0.97

3.67

0.34

5.70

1.88

15.3

1.26

3.98

-0,61

-3,82

-0.14

2,65

•' Initial TRI EmissionsPrices between i987QI

0.64

3.24

(1.040

3,76

Std. Enoi

and 199IQ2

0.96

5.66

7.52

2.5!

and 198901

1.32

10,5

1.21

3.05

Hdsc [ 'quaiiiii i :

with HighcslReadership i

Cod,

0.68

0.016

7 32

0.19

0.63

-2.44

0,87

0.36

ConlnlUllillC^

;V - 124)

Sid. Error

l.()4

X.56

4.22

2.3S

15.6

1 J?

ii,02

• A l l cqvuui t in ' . j l s i i i rL-luJe ihf iiIhiT i n d c p e i i d e i i l v^i r iables l i s led in T a b l e A l , w h i f h i s Ihi-

i e r t nnre i l .

. E a c h seL-iu>n r e p o n s

prices.'2 An allemative hypothesis is thai the public onlycares about large releases. In other regressions, we foundsimilar results when looking at capitalization of the largest50% of releases.

Einally, we consider the possibility that the toxic releasesin the reported TRI data were actually observed contempo-raneously by local residents. For example, residents mightbe able to infer actual toxic releases from proxies such asaggregate production or employment at a plant. The bottompanel of Table 4 reports the results of the same regressionsin the top panel, except looking at contemporaneouschanges in house prices from 1987 to 1989. Once again, wefind no evidence of any capitalization of TRI releases intohouse prices. None of the coefficients are statistically dif-ferent from 0 at the 95'*' percentile. Tn these regressions,however, the standard errors rise from those found in the toppanel, so a one-standard-deviation decrease in aggregateTRI emissions can contemporaneously raise house prices byup to 0.5% at the 95"' percentile.

D. Changes in TRI Emissions. 1987-1990

Although the results in Table 4 show little effect of theinitial TRI reports, it is possible that house prices hadalready capitalized the level of toxic emissions. If true,house prices would be unlikely to react to initial TRI data,but should react to changes in releases over time. We find noevidence of any such reaction.

To begin, we explore reported changes in the first year ofTRI in Table 5, which repeats the basic specifications in

--This conclusion does not change if we remove the top 10"' or 20'"percentile of emissions.

Table 4, except that we k)ok at changes in reported emis-sions between 1987 and 1988 and examine changes in houseprices over a two-year period that corresponds to the dalesof reporting of TRl releases. Aggregate releases were re-ported to decline almost 40% in the first year of the kuv. Asshown above, these releases were not completely random:larger releases were reported for plants in high-housing-value communities and in plants with higher initial reportedreleases. Yet these reported releases had little impact onlocal house prices. In fact, three of the lour coefficients onthe change in reported TRI releases in the lirst column arepositive, although none are even close lo being statisticallydifferent from 0 at conventional levels. Two standard dc\i-ations around the estimate using total TRI releases in thefirst row and column suggests that we can reject with about95% confidence the hypothesis thai house prices will rise byless than 0.17% as a result of a rather large hypotheticaldecrease in emissions of 133,000 lb. (a one-standard-deviation decline in emissions). These basic conclusions donot change on looking at different specifications, typos oftoxics, or contemporaneous changes in house prices. Wealso obtain similar results with all of the alternative speci-fications.

Some observers of TRl have argued that early TRI reportswere quite inaccurate because many firms did iu)t havesufficient expertise to estimate releases accurately. Also,some firms might have overreported their initial releases.even if accurate, to avoid the possible negative publicity oflater having to admit that they underreported loxic releases.Also, firms might have expected to receive positive public-ity associated with reporting strong initial declines in emis-sions. To the extent that firms pursued such strategics, and


h 6,—2SL.S RHGRESSIONS OF CHANGB IN HOU^F PRICKS ON CHANGF IN TRI EMISSIONS HFTWFEN 1988 AND 1989'

Base EquationAlternative Equation

wilh Surrounding Towns

Base Equation: Communitieswith Highest Newspaper

Readership (N = 124)

Specification Coef, Std, Error Coef, Std. Error Coei Std, Error

Announcement Effect of initial TRI EmissionsDependent Variable: Change in House Prices between 1990Q2 and 1991Q;

Aggregate TRI emissions in 1987

Carcinogenic emission in 1987

Adverse developmental or reprodueiiveemissions in 1987 (XIO ̂ )

Noxious emissions in 1987

- 8 . 3 0

-0.7.5

- 2 , 1 9

3.03

22..'̂

1,09

3,45

4,74

- 8 . 7 1

- 0 , 5 3

- 1 . 3 6

- 0 , 7 3

12,0

0.85

0.94

2,58

7,36

0,42

-1 .21

23,6

22.7

1.68

4,4.1

20.8

' A l l i iiu-liide Ilie olhci indcpoi id i :n l v ; i [ iaMcs l i s ted in T a h l f A l , whic l i I,H Ihc b a s e vpfi;i l ic;i l ioii . sct-l ion i v p u r t s c o i : t l i c i c m s Ir

the public knew this was happening, house prices might nolhave reacted to early announcements as much as to subse-quent reports.

Table 6 reports the same coefficients as Table 5, exceptthat it explores changes in house prices in the second fullyear of TRI announcements. As suggested above, the resultsfor changes in aggregate emissions suggest that the publichad a stronger reaction to these second-year results. Coef-ficients in the first row of Table 6 are increased in size fromthe corresponding coefficients in the first row of Table 5,and ure almost uniformly negative in all specificationsexcept for noxious emissions. However, none of thesecoefficients exceeds its standard error, and the magnitude ofthe effects are quite small. We estimate that a one-standard-deviation decrease in reported emissions {17,700 lb.) in-creases house prices by 0.015%, and we can reject at the95"" percentile the hypothesis that it has more than a 0.08%effect on house prices. Our basic conclusions do not changeif we examine any of the other coefficients in this Table.

Finally, Table 7 presents evidence on the capitalizationof changes in reported releases from the first three yearsof the program. While the coefficients on overall emis-sions and developmental and reproductive releases areactually positive and insignificant, the coefficient esti-mates on carcinogenic releases are negative and above

conventional significance levels in the middle column. Inthis case, a one-standard-deviation decline in carcino-genic emissions (approximately 18,000 ib.) results in a0.20% increase in house prices over a four-year perutu.Tight standard errors still allow us to reject with 95Vfconfidence the hypothesis that this large decrease incarcinogenic releases leads to more than a 0,43% in-crease in house prices.

The results above are reasonably consistent. No matterwhat the time period, the housing market does not appear toreact very strongly to any of the data released by TRI.Although the coefficients on TRI announcements oichanges in toxic releases appear mostly negative, we canreject the hypothesis that a large (onc-standard-deviation)change in the announcement of any one of these variabie,shas affected house prices by more than 0.5% over this timeperiod. These results hold no matter what econometricspecification we use, no matter how we treat emissions insurrounding communities, and even if we limit the sampleto communities with greater access to information abcniitoxic releases through high newspaper readership. We a^j.n',note, however, that almost al! other variables, including newconstruction, change in economic conditions, and proxiesfor location and school quality, do have a statistically

7,—2SLS RL-GRKSSIONS OF CHANGH IN HOU-SB PRICES ON CHANGE IN TRI EMISSIONS 1987 .^ 1990='

Base EquationAlternative Bquaiiun

with Surrounding Towns

Base HquaHon: Communitieswith Highest NewspaperReadership (,V = 124)

Spec iti cation Coef, Std. Error Coef. Std, Error Coef. Sid. El

Announcement Effect of Change in TRI EmissionsDependent Variable: Change in House Prices between 1989Q2 and I993Q2

Aggregate TRI emissions in 1987 ( X I O ' ' ) 2,43 12,3 2.24 10.0 - 5.S8 13.4

Carcinogenic emission in 1987 (X10" ' ' j - 1 1 , 0 6,0 - 9 . 1 9 4.37 9,74 7,IK

Adverse developmental or reproductiveemissions in 1987 (XK)-**) 1.74 12.3 1,44

Noxious einissions in 1987 (XIO **) 4,62 4,03 3,93

• All e^]ilnll.^ll^ iilst> i n k l u d e iho . i ihcr m d q K E i d t i i t w i n a b l e s l i s ted in T a b l e A l , w h i c h is (he h. ise s p c t i t i c a t i o n h a t h ,M.VIJOII reporl'. coi

10.0

4,37

10,2

3,74

;s from scp.ir.ii

-•:),88

- 9,74

2,01)

1,71

[c il:L'rL•^^ll)ll^, NiimbLT ol" ob

706 THF REVIEW OE ECONOMICS AND STATISTICS

I- H. • RI^<;RI;SSK)N,S (!(• T R i K M I S S I O N S O N LACCini ('HAKni-..s IN Hoi.sr-: I '

DcpciKlcni Variuhle

Change in liou.se prices. I989'I99O

Change in house prices. 1988 1989

Change in lioiise prices, 1987-1988

Change in house prices, 1986-1987

Change in house prices. 1985 1986

Consuuil

ft-square

TRI Emission'-

Coel.

-472,406

4K6.676

568.035

-73,319

275.K96

O.(H)7

in 1987

Snl. Hnor

694.1 10

62,3,196

485.661

468,220

218.415

Change in TRI Emissions,1987 19SS

Co,i.

151 .K49

-182,730

128.5,56

-40. ) 92

-33.514

Ski, ['iTor

2S7,8.3K

258,431

201.397

194.164

90.573

0,0(.)3

Change1

Coef.

37.K90

219.567

• 1OK,757

132,668

• 56.987

in TRI Emissions.9KX-I989

Std. Rrroi-

311.791

198,579

I98.("i82

147.382

46.836

0.009

significant and econoniicaiiy important impact on localhouse pfices.

E. TRI Relea.ws and Expectations

Although these re.sults seem to suggest that home buyersdo not piaee a large value on toxic releases by plants locatedin Iheir ZIP code, other explanations are possible. Homebuyers might have had expectations that vt'ere consistentwith the exact path of reported TRI releases. According tothis view. TRi releases were in line with expectations andthus the TRI announeements should not be expected ioalTect local house prices. One way to test this is to allow forthe possibility that house prices anticipate announced TRIemissions. To do this we regress TRI emissions on laggedhouse prices. If the publie places a negative value on TRIreleases, bul is able to predict future changes, the coefficienton lagged ehanges in house prices should be negative. Theempirieal results in Table S seem to reject this hypothesis.Lagged house prices have little predictive power for futureTRI announcements, no matter whai the time period. The/^-square for all three equations (TRI announeements in19H7, ehange in TRI emissions between 1987 and 1988. andchange in TRI emissions between 1988 and 1989) is ex-ceedingly low, and F tests cannot reject the hypotheses thatthe coefficients on all of the lagged hou.se price terms areindividually and jointly equal to 0.

Alternatively, our results may simply reflect the faet thatTRI data might have been regarded as sufficiently low inquality that they were ignored by home btiyers. This wouldbe consistent with the stock market results found by Bui(2()()2) for petroleum companies.

VII. Concluding Remarks

In this paper, we explore lhe impaet of informationreleased under the Toxics Release Inventory. Our resultsshow little evidence to support the claim by many propo-nents of TRI that the acfs introduction has caused commu-nities to mobili/e against polluters and induce large reduc-lions hi toxic emissions. We find that (I) poorer, blue collar

areas are more likely to have toxic emissions and (2) largerreduclions in toxic emissions oecur in higher-value regionsand regions with higher initial releases. However, (3) re-ductions in releases are unrelated to any political economyfactors that might relate to the ability or willingness of acommunity to organize against major polluters. When mea-sured by the initial level of toxic releases reported in 1987,(4) introduction of the TRI reporting had virtually no effecton housing prices, and, even more surprisingly, (5) subse-quent reduetions in aggregate reporled emissions between1987 and 1990 have no significant effect on house prices.either in the aggregate or when disaggregated by type oftoxic release. Einally. expectations do not appear to explainihese results. We show that (6) changes in house prices donot forecast ehanges in toxic emissions, as might be ex-pected if communities anticipated increases or reductions intoxic releases, and (7) these results persist even in commu-nities with high newspaper readership.

It is important to emphasize that our findings do not showthat TRI has no impact on local house prices, but rather thatthe estimated effect of TRI is suflieientiy small as to beundetectable using our rather accurate house price series. Invirtually all equations the coefficients on TRI infonnationare not statistically different from zero. This finding is notdue lo large standard errors. We can reject with 95'Xconfidence the hypothesis that a one-standard-deviation in-erease in announced TRI releases (or announced changes inTRI releases over time) lowers lK)use prices by more than0.5%. It is possible that we might find a larger effeet if welooked at a mueh smaller neighborhood around plants thatrelease toxics. However, these results seem lo reject abroad-based community response to TRI.

Our findings that communities do not place a significantvalue on deelines in toxic releases is in sharp contrast to thebeliefs held by environmenial policy-makers and to theexisting literature on the publie valuation of another envi-ronmental amenity, ah* quality. How can we reeoncile theseresults with the evidence in the existing literature, alongwith our basie intuition ihat households value clean air andwater? One major difference between air quality and TRI

REGULATION AND CAPITAMZATION OF HNVIRONMBNTAL AMENITIES 707

releases is that the former i.s immediately identitiable to allhouseholds by bolh (heir visibility and their odor (thesebeing tiieasures oi the concentration of such pollutants assulfur dioxide, particulale matter, and ozone), whereas thechemical releases measured by TRl are difficult to observe{TR! emissions tend to be both colorless and odorless).However, even when we disaggregate the TRl releases intothose that are particularly noxious, we still find no statisticalrelationship between changes in toxic releases and changesin house prices. Our results may simply reflect an "out ofsight, out of muui" problem. A community's inability toassess the risk associated with toxic releases would beconsistent with the findings of Greenberg and Hughes(1992).

Another important difference between our re.sults andother studies is uiethodologicai. Previous research relies onhedonic regressions that can produce negatively biasedcoefficients if air quality is positively correlated with othercommunity factors that negatively affect house prices. FACU

in our data it is true that there is a strong and significantcorrelation between median house prices and aggregate TRIreleases in 1990. However, that correlation disappears whenwe look at differences over time, suggesting that the corre-lation is likely driven by negative unobservable faelors thatexist in hou^es and neighborhoods located near plants thaiemit TRI chemicals.

Finally, households may asMuiie that existing hPA regu-lation is strong enough to protect them against harmfullevels of chemical emissions. While firms were required torelease information about toxic emissions under TRl. thesefirms were still subject to the usual battery of EPA regula-tions. Thus it would be inappropriate to extrapolate theseresults to a regime in which the only type of regulation isdisclosure of public releases atid firms were free lo etnitwhatever they wanted.

Investor response to TRI announeements seem in linewith our findings. There is only weak empirical evidencethat investors responded to the initial announeement of TRIreleases, and subsequent announcements appear to have nosignificant effects on stock market returns. The lack of asignificant response in the housing market seems to suggestthat communities either do not care about toxic releases ordo not internalize the differential risk associaied with beingin a relatively dirty versus a slightly cleaner community.Investors may have initially responded to the TRI informa-tion because large reported releases might be a sign that theHrm will faee future pressure from community groups orlegal liability at some future date. To date, however, there islittle evidence that TRI information has been successfullyused for litigation against polluting plants.-' The vast ma-jority of legal cases that have been resolved that are relatedto TRI are about communities suing individual plants fornonreporting. Taken together, uuv results suggest that broad-

-' Based on a ^ummaiy of" cases reported by LexisNeMs bclweenand 1996.

based eommunity pressure is the least likely of these alter-natives to explain the reported reduction in releases. This isconsistent with the Coase theorem, especially if one be-lieves that the costs of organizing community residents arehigh and the public does not place a high premium onhouses in communities with low or declining emissions.

So what was responsible for the decline in reported toxicreleases':' Factors that are unrelated to TRI regulation mayplay an important role in explaining ihe reduction. Forexample, between 1988 and 1992, Massachusetts lost al-most one-third of its manufacturing jobs. This suggests thatmuch of the reported TRI reductions may be due to de-creases in manufacturing output. Further research can ex-plore this proposition by looking ai the specific declines inreleases and compare them with the manufacturing indus-tries that have left the state or reduced output. Disaggre-gated TRl data make this possible. In addition, firms mayhave improved their manufacturing efficiency over thisperiod, ieading to a voluntary decline in emissions. Also,much of the repotted decline in emissions occurred in earlyyears, when firnis may have been working on methodolo-gies to accurately estimate their toxic releases. If firmsinitially overreported emissions due to uncertainty abouttheir actual values, this would lead to large initial declines.Our regression results are somewhat consistent with thishypothesis: ZIP codes with the highest initial releases hadthe greatest subsequent declines in releases. Yet almostone-third of plants reponed increases in emissions, so allplants cannot have followed this strategy.

Another possibility is that polluting plants simply substi-tuted away from the listed TRI substances to equally toxie.but unlisted, substances. In that case, if communities areaware oT this behavior, we would expect to see little or nocapitalization occurring with reported reductions in TRIemissions. This possibility would be worst of all outcomesin terms of the effectiveness of disclosure-based regulation.

The results here cast considerable doubt on the premise ofTR!, that public reaction can discipline industrial behavior.Before embarking on a major policy shift in favor ofinformation-based regulation, regulators should as.sess themechanism through which this regulation is expected towork and the areas where it will likely be most effective.

REHERENCHS

iiariiL'k. TiiTioihy .!,. "Tlie t^siima'.ion of Demand Parameters in HedonicPrice Models," .Iniinw! of Politica! iuoriomv 95:1 (1987), 81-88.

Bednar/.. Roberl, "Tbe Eflecl of An- Polknioii on Property Value inChicayo," Linivcrsity of Chicitco. LX-partmcnt o( (ieotiraphyU')7.S).

(iradbtiry. Katlieiinc L., Christopher J. Mayer, and Karl k Case, "PropertyTax Limits, Local Fiscal Behavior, and Properly Vaiiics: EvidenceIrom MassaehiisetT'; under Proposilioii 2-."" Journal of Publichcononiir-s HO (20(11 ), 2S7-312. '

Bni. Lintla T, M., -Plant Level Responses lo Public Di;sclosure Rules; TheStor> of i'etroteudi Refineries and [he Toxics Release tnventory,"Boston University Uepartnieni of Economies mimeograph (2002).

Case. Karl t . . and Chrisiophec J. Mayer, '-Housing Price Dynamics withinit Melrupolitan Area," Regional Scieiur aiul Urban Economics 2611996). .̂ iX7̂ O7


Case. Karl E.. and Roberl J. Shitler. "Prices of Single-Family Homes since1970: New Indexes for Four Cities," New England Economic-Review (September/Oclober 1987). 45-56.

Chity. Kenneth Y. aiid Michael Greenstone. "Does Air- Qualify Mattel-'.' Evidencefrom the Housing Market," UC Beiiceley niimeogmpli (20(X)).

Epple, Dennis. "Hedonic Prices and Implicit Markets: Hstimating Demandand Supply Functions for Differentiated Products," Journal nfPolitical Economy 95:1 (1987), 59-80.

Fischel. William. "Do Growth Controls Malter?" Lincoln Institute of LandPolicy working paper (1990).

Freeman, .\. Myrick, HI. "Air Pollution and Properly Values: A FurtherComment," this REVIEW 56 (!974). 554-556.

Freeman. A. Myrick, III. "Air Pollution and Property Values: A Method-ological Comment." this KEVIKW 53 (1971), 415-416,

Gayer. Ted. Hamilton, James T . and W. Kip Viscusi, "Market Evidence onLearning about Cancer Risks: A Repeat Sales Housing MarketAnalysis," Harvard Law School mimeograph (1998).

Greenberg, M.. and J. Hughes, "The Impac! of Hazardous Waste Super-fund Sites on the Value of Houses Sold in New Jersey," Annals ofRegional Science 26 (1992). 147-153.

Gyourko, Joseph, Matthew Kahn, and Joseph Tracy, '"Quality of Life andEnvironmental Comparisons" (pp. 1413-1454), in Paul Cheshireand Edwin Miils (Eds.). Hamlbook of Urban and Regional Eco-nomics: Applied Urban Economics, Vol. 3 (1999).

Hamilton, Bruce. "Zoning and Property Taxation in a System of LocalGovernments." Urban Studies 12 (June 1975), 205-211.

Hamilton, James T , "Pollution as News: Media and Stock Market Reac-tions to the Toxics Release Inventory Data," Journal of Environ-mental Economics- and Management 28 (1995), 98-i 13.

Harrison, David. Jr, and Daniel L. Rubinfeld, "The Air Pollution andProperty Value Debate: Some Empirical Evidence." Review ofEconomics and Statistics 60 (1978), 635-638.

KJel. Katherine. ;ind K. McCain. "House Prices during Siting Decision Stages:The Case of An incinerator from Rumor thin Oper.ition." Joumal ofEnvironnwntal Economics and Management 28 (1995), 241-255.

Kiel. Katherine, and Jeffrey Zabel. "Estimating the Demand for Clean Airin Four United States Cities," hind Economics 76 (2000). 174-194.

Kohlhase. J. H.. "The Impact of Toxic Waste Sites on Hou.'̂ ing Values."Journal of Urban Economics 30 (1991), 1-26.

Konar. Shaameek, iind Mark A. Cohen, "Infonnation as Regulation: I'he Eftect ofCommunity Right to Know LJIWS on Toxic Ejiiissions." Journal ofEnvimnmental Ecoiwmics ami Management 32 (1997), 109-124.

Mendelsohn, Robert, Daniel Hellerstein, Michael Hugeniii. Robert \.]n-sworth. and Richard Brazee, "Measuring Hazardous Waste Dam-ages with Panel Models." Journal of Environmenlal Economicsand Management 22 (1992), 259-271.

Oberhol/,er-Gee, Felix, and Miki Mitsunari. "Informalion Regulation: Dothe Victims of Externalities Pay Attention','" University oi' Penn-sylvania, Wharton School mimeograph (2(K)2).

Paimquist, Raymond B., "Measuring Environmental Rtfects of PropertyValues without Hedonic Regressions." Journal of Urban Ei-onom-ics 11 11982). 33,^347,

Polinsky. A. Mitchell, and Steven Shavell. "The Air Pollution and Prop-erty Value Debate." thi.s Rlivir-.w 57 (February 1975), 100-104,

Polinsky, A, Mitchell, and Steven Shavell, "Amenities and PropertyValues in a Model of an Urb-M Area." Journal of Pnhtic Economics5 (1976). 119-129.

Ridker, Ronald G.. and John A. Henning, "The Determinants of Residen-tial Property Values with Special Reference to Air Poliution." thisREVIEW 49 (May l%7). 246-257,

Rosen, Sherwin. "Hedonic Prices and Imphcit Markets: Pr()duct Differ-entiation in Pure Competition," Jourmil of Political Economy 82(January-February 1974), 34-55,

Shiller, Robert J.. "Arithmetic Repeat Sales Price Estimators," Journal ofHousing Economics 1:1 (March, 1991), 110-126.

Small, Kenneth A.. "Air Pollution and Property Values: Further Com-ment," this REVIEW 57 (February 1975), 105-107.

Smith. V. Kerry, and Ju-Chin Huang. "Can Markets Value Air Quality'? AMeta-analysis of Hedonic Propert> Value Models." Journal ofPolilical Economy 103:1 (1995). 209-227.

APPENDIX

TABLE; AL—2SLS RECiRE,ssioNS OP CHANGL IN HOUSK PRICHS

( ! ) • '

Change in House Price1989-1992

Dependent Variable Coef. Sui. Error

Change in House Price1989-1992

Coet; Sid. Krroc

New housing permits issued as a percentage of the total number of units

Percentage change in manufacturing employment in the town

Change in unemployment rate in the town

Percentage change in health and welfare spending in the town

Percentage of lesidents who workeJ in the manufacturing sector in 1980

Average math and reading assessment lest .score (OOOs)

Percentage of residents aged 35-60 In 1980

Dummy variable if in the Bosion metro area

Distance from Boston if in (he Boston metro area (00s)

(Distance from Boston if in the Boston metro area)- (O,O(X)s)

Constant

•0.66

-0.0039

-(I.,19

0,0033

0.058

-0.023

0.024

0,12

-0.30

-0.27

0,20

0,0061

0.20

0.0019

0,018

0.076

0.005

0,05

0.10

0,04

-0.57

-0,06(.l

0,062

(1015

0,02.̂ ,

O.iO

-0.27

-0.2^

0.21

0,024

0,016

0,075

0.005

0,04

O.Ul

0.03

-iii|iloymeiu

regulation and capitalization of environmental amenities

Documents