Document View

State voluntary cleanup programs (VCPs) were established in the 1990s to encourage cleanup and redevelopment of contaminated properties. I ask three questions: First, what properties are attracted to VCPs? Second, is there an interaction between VCP incentives and enterprise or brownfield zone incentives? Third, does participation in VCPs affect property values? Data from Colorado's VCP suggest that (a) the main determinants of participation are the size of the parcel and the surrounding land use, (b) other incentives have little effect, (c) properties with confirmed contamination sell at a 43%-56% discount, and (d) participation does tend to raise the property price. [PUBLICATION ABSTRACT]

(ProQuest-CSA LLC: ... denotes formulae omitted.)

I. INTRODUCTION

Environmental regulation and enforcement-based environmental programs sometimes result in unintended consequences that defeat the purpose of the programs themselves. Many observers argue that the U.S. Superfund program is one such program. Over the last 25 yr, Superfund has identified contaminated sites needing environmental remediation, tracked down the responsible parties, and forced them to pay for the cleanup (or reimburse the U.S. Environmental Protection Agency [U.S. EPA] for the cleanups it initiated). Liability for the cost of cleanup is retroactive, strict, and joint and several, with potentially responsible parties to be sought among the owners and operators of the site and transporters of the wastes.1

In theory, these features should deter firms from handling hazardous waste carelessly. In practice, since liability has in some cases been construed to apply to property owners and lenders who foreclose on contaminated properties (Fogleman, 1992), they have also been blamed for discouraging the purchase and reuse of contaminated or potentially contaminated sites, which have remained idle or underutilized.

Recent state programs and federal legislation have attempted to reverse these disincentives. For example, the federal Small Business Liability Relief and Brownfield Revitalization Act of 2002 provides conditional relief from environmental liability for property owners and purchasers of land. In addition, starting in the 1990s, several states began establishing voluntary cleanup programs (VCPs) offering liability relief, other economic inducements such as tax credits or low-cost loans, oversight and expedited approval of cleanup plans, and simplified cleanup standards in exchange for site cleanup (Bartsch and Dorfman, 2000; Meyer and VanLandingham, 2000).

Despite much interest in policy circles and the attractiveness of relying on economic incentives rather than enforcement-based approaches, the effectiveness of these programs in promoting environmental remediation has not been studied to date in the economics literature. Little is known about the responsiveness of cleanup and redevelopment activity to these inducements, and while several studies have examined the effects of contamination on the value of commercial and industrial properties, none has focused on the appreciation potential (if any) of parcels participating in VCPs. Yet, these are important environmental policy questions, especially if we consider that the emphasis of these VCPs has gradually shifted away from environmental remediation goals toward economic development goals and that participation in the VCP is required in some states-such as Pennsylvania-for transferring contaminated properties (Meyer, 2000).

In this paper, we ask three related questions: First, what are the characteristics of parcels that make them attractive candidates for voluntary cleanup? second, do other local and federal economic development programs (whether or not specifically targeted for contaminated sites, such as enterprise zone or brownfield zone designations) make voluntary cleanup more or less likely? Third, what are the effects of participation in VCPs on property values?

Given the wide variety in program features across states, and the dearth of data documenting program enrollment and the characteristics of participating and nonparticipating properties (Meyer, 2000), we do not attempt a national-level analysis of VCPs. Instead, we focus on one state, Colorado, which established its VCP in 1994 and use information about individual parcels to estimate (a) a probit model of participation in the Colorado VCP and (b) a hedonic pricing model that relates property value to characteristics of the parcel and the neighborhood, and on whether the parcel was signed up for the program.

Three important lessons emerge from our analysis. First, the program has not "absorbed" the existing supply of sites on EPA registries of contaminated sites but has rather created a new "crop" of sites. second, because the majority of these properties apply directly for a No Further Action determination, the program does not seem to have encouraged much environmental remediation. Third, the participating properties are probably those with the highest redevelopment potential.

The results of hedonic pricing regressions where participation is allowed to be endogenous with property values show that sites with confirmed-not merely suspected-contamination sell at a 43%-56% discount relative to comparable uncontaminated properties. Participation seems to result in a partial to complete price recovery.

The estimate of the contamination discount is robust to the way we construct our sample of properties-whether we focus on sites on EPA registries of contaminated sites (augmented with the sites that actually participated in the VCP) or form an alternate treatment-control sample where participating sites are pooled with nonparticipating parcels selected at random from the universe of properties in the same areas slated for similar uses. The exact extent of the rebound in property values does vary across these two samples, and price recovery appears to be complete only for the alternate sample.

Taken together with the evidence about participation, these findings raise questions as to whether participation in VCPs maydepending on program features-be sought primarily as a way to improve the market attractiveness of the parcels with minimal or no environmental remediation. Were this possibility confirmed by similar findings for other states' programs, this would cast doubts about the incentives created by VCP and about these programs' effectiveness in encouraging cleanups by reversing some of Superfund's unintended consequences.

The remainder of the paper is organized as follows. In section II, we present an overview of the effects of liability at contaminated site on real estate development and of recent VCPs. In section III, we review the relevant previous literature. In section IV, we describe the Colorado VCP and our study plan. section V describes the data and section VI the estimation results. section VII offers our concluding remarks.

II. BACKGROUND

In the United States, there is a large supply of properties where prior industrial uses have resulted in contamination of soil, surface water, and/or groundwater with pollutants that are noxious to human health and ecological systems. The U.S. General Accounting Office (U.S. GAO, 1995) estimates that there are 130,000-450,000 contaminated commercial and industrial sites in the United States.

It is widely felt that site contamination, or the mere suspicion that a site might be contaminated, seriously hampers its reuse. The U.S. EPA defines as hrownflelds real property "the expansion, redevelopment or reuse of which may be complicated by the presence or potential presence of a hazardous substance, pollutant, or contaminant,"" and some observers have argued that Superfund has created the brownfield problem to the point that they equate the supply of brownfields with EPA's registries of (potentially) contaminated sites. They further claim that removal from such registries-delisting-automatically removes contamination stigma (Bartsch, Collaton, and Pepper, 1996)/

Another consequence of the Superfund law is that fear of liability has driven developers to turn to pristine properties in suburban areas, contributing to urban sprawl and congestion and to the loss of open space and agricultural land. Policies that encourage cleanup and reuse of (potentially) contaminated sites are, therefore, attractive to communities and policymakers because they reduce health risks to residents and workers, mitigate the adverse effects of pollution on ecological systems, avoid development patterns that result in sprawl and congestion, and may stimulate economic growth in inner cities.

Starting in the 1990s, the states, realizing that their enforcement-based programs did not have sufficient funding to address the large number of contaminated sites needing attention, began developing VCPs, which rely on a different approach to the problem of remediation of contaminated sites (U.S. GAO, 1997). By 2000, over 90% of the states had a VCP in place (Meyer, 2000).

Programs offering and requirements vary widely across states (Meyer, 2000; U.S. EPA, 2005; U.S. GAO, 1997). Many statelevel VCPs grant liability relief in exchange for voluntary cleanup, provided that the latter is approved by the state agency in the form of a letter of no further action, a certificate of completion, or a covenant not to sue.4

VCPs often spell out simplified or variable cleanup standards linked to land use and hence to residents and workers' likely exposure to contaminants. Some states allow for engineering controls, such as caps, fences, or other physical means of preventing contact with pollution, in lieu of a more permanent cleanup, and/or offer institutional controls, such as permanent land use restrictions at the site or monitoring of the contamination plume, in place of (more stringent) cleanups.5

In addition, the state often offers fast-track oversight of cleanup plans. This helps reduce the time it takes before remediation is undertaken, as well as the uncertainty associated with stringency of cleanup standards (Meyer, 2000). At many locales, completion of voluntary cleanups at eligible sites can be combined with local, state, and federal "brownfields" programs that offer subsidies in the form of tax credits or low-cost loans. State VCP managers believe that the programs have resulted in the reporting of contaminated sites that were previously unknown to the state agency and have truly encouraged cleanups as long as the program requirements are not too burdensome to the applicants.6

Although thousands of voluntary cleanups have been undertaken throughout the nation (U.S. EPA, 2005) and despite numerous case studies regarding specific properties (Rafson and Rafson, 1999; Simons, 1998), relatively little is known about the type of sites that tend to enroll in VCPs and the reasons why property and business owners undertake voluntary cleanups. Presumably, they do because the benefits exceed the costs, but what exactly are the benefits and the costs and how do they vary across parcels and locales? Does participation in VCPs raise the value of the parcels? Answering these questions is important and should provide useful information for designing better programs and economic inducements in the future.

III. PREVIOUS LITERATURE

A. The Effects of Liability and Liability Relief

VCPs are based on the premise that protection from liability is desirable and should increase the attractiveness of a property, thus encouraging market transactions. But what does theory say about how liability affects land use and transactions? Boyd, Harrington, and Macaulcy (1996) identify a role for asymmetric information about the contamination of a parcel and the cleanup costs, which may result in a "market for lemons," and the role of risk aversion, which is concluded to deter transactions only if buyers as a group are more risk averse than sellers. They also consider "imperfect detection," that is, the situation where the degree of contamination is unknown to the government and the owner. In this situation, the owner may forgo otherwise desirable transactions to avoid scrutiny.

In Segerson (1993), the effect of liability of land transactions depends crucially on whether the parties are solvent. Imposing liability on the seller or buyer does not matter, and efficient outcomes are reached when parties are solvent. Results are ambiguous when liability is shifted to one or the other parties. Sigman (2005) examined whether land development rates and prices are affected by the type of liability imposed by the state mini-Superfund programs, finding that the presence of joint and several reduces development rates and prices. Her study, however, cannot be used to predict the effects of voluntary programs that aim at removing or reducing liability.

We are aware of only few studies about the effects of a document of no further action from the state. Using data from the State of Ohio for 1989-1992, Sementelli and Simons (1997) found that receiving a letter of "no further action" from the state does not improve transaction rates for sites with leaking underground storage tanks (LUST), which remain much lower than those for nontank commercial properties. Lange and McNeil (2004a, 2004b) surveyed over 100 EPA brownfield grant recipients and other stakeholders. They found that community support, consistency with local plans, cost minimization, financial incentives, and minimizing the time it takes to put the site back into productive use are the most often-cited variables that influence brownfield development success.

B. The Effects of Contamination and Cleanup on Property Prices

In principle, it may be possible to infer the attractiveness of voluntary remediation from the difference in values across properties that are and are not contaminated. However, there is mixed evidence about the effects of contamination on the value of commercial and industrial properties and on the market transactions involving them. In one analysis of previously contaminated industrial properties in southern California, Jackson (2001 )concluded that prices are not statistically different from those of comparable but uncontaminated properties. In another analysis that covers 140 industrial property sales in southern California in the period 1995-1999, Jackson (2002) found that industrial contaminated properties sell at prices approximately 30'Xi less than unimpaired levels but that prices recover after cleanup to the same levels as those of comparable uncontaminated parcels.

McGrath (2000) found that contamination risk-that is, the probability that a previously used site is contaminated based on the previous use-does affect urban industrial redevelopment in Chicago both directly and indirectly via the differential in price before and after redevelopment. The price differential-about $1 million per acre-is high relative to typical cleanup costs, suggesting that the costs are fully or even overcapitalized.

Howland (2000, 2004) found that parcels in two industrial areas in Baltimore are significantly lower when the property is contaminated but that turnover is not affected by contamination. Based on interviews with real estate agents, Howland (2004) suggested that incompatible land uses, inadequate infrastructure, and obsolete buildings are more important barriers than contamination to the revitalization of brownfields in Baltimore. Focusing on a third industrial area in Baltimore, Schoenbaum (2002) found no significant difference in assessed land values, vacancy rates, property turnover, and redevelopment rates across brownfield and nonbrownfield properties over 1963-1999. Evidence that contamination in one area affects the prices of nearby commercial and industrial properties is also mixed and locale specific (Ihlanfeldt and Taylor, 2004; Longo and Alberini, 2006).

C. Economic Incentives

Economic inducements have been advocated as potentially effective for stimulating cleanup and redevelopment of brownfields (Bartsch, Collaton, and Pepper, 1996; DeSousa, 2004; Howland, 2000,2004; Yount and Meyer, 1999).

On their part, real estate developersclaim that they are responsive to a broad range of inducements. In surveys in Europe (Alberini et al., 2005) and in the United States (Wernstedt, Meyer, and Alberini, 2006), choice experiments reveal that developers can be attracted to contaminated sites by offering them subsidies, liability relief, and less stringent regulation. Prior experience with projects at contaminated sites matters, in the sense that these incentives do not appeal to the same extent to all developers. Meyer and Lyons (2000) suggested that low property prices have played a larger role than subsidies in stimulating entrepreneurial redevelopment activity on contaminated sites and that obtaining subsidies may entail significant transaction costs that offset their value.

Since it is sometimes argued that contaminated properties are located in economically disadvantaged areas (Simons, 1998) and that incentives are needed to encourage their reuse, in empirical work it is important to distinguish for incentives that are explicitly linked with cleanup and incentives to economic development and location decisions that are not linked with cleanup, such as industrial development bonds, tax credits for job creation or business location, property tax abatement, tax increment financing, and downtown development authorities.7

IV. STUDY DESIGN AND ECONOMETRIC MODELS

We focus on the Colorado voluntary program, which was established in 1994. We first relate participation to the characteristics of candidate parcels, restricting attention to properties on the Front Range and controlling for other economic development policies that offer economic incentives to businesses. We then examine whether participation in the VCP influences the market value of a parcel. To answer this question, we estimate a regression equation where we compare participating and nonparticipating sites, as well as participating sites before and after enrollment in the VCP. We allow participation and property value to be potentially endogenous with one another.

A. The Colorado VCP

The Colorado VCP was passed in 1994. The statute clearly emphasizes that cleanup and program participation are purely on a voluntary basis. There are two possible modes of participation: first, the applicant can request a No Action determination (NAD), provided that he can show that the site is virtually clean or that the likelihood of exposure to the contamination is low. Applicants have also successfully requested a NAD when they were able to show, in the case of contaminated groundwater, that the pollutant had migrated to the site from elsewhere.

The second mode of participation is a voluntary cleanup (VCUP), whereby the applicant proposes to undertake remediation at the site. Once the application is approved and the cleanup has been completed and approved by the state, the applicant submits a separate application to receive a NAD. When the NAD is granted, the applicant is given a letter of no further action, relieving him of further liability over the site. Once cleanup is completed and the letter of no further action issued, VCUP properties that satisfy certain requirements qualify for tax credits through the state (or federal) brownfield program.

To participate in the Colorado VCP, parcels must meet certain eligibility criteria.8 An application of $2,000 is due at the time of the application, and decisions must be notified to the applicants by the Colorado Department of Public Health and Environment within 45 d.

B. Determinants of Enrollment in flic VCP

One goal of this paper is to establish which characteristics of a parcel (size, structures, distance from transportation nodes, availability of specified incentives, etc.) and its neighborhood make it more likely to participate in the Colorado VCP. Consider a set of "candidate" parcels (we discuss how this set is defined in section IV.F below). We assume that a parcel is enrolled in the VCP if net benefits of participation are positive. Let VCP* denote the net benefits of participation in the program and assume that:

(1) VCP*^sub i^ = x^sub i^β + η^sub i^,

where x is a vector of site characteristics, β is a vector of unknown coefficients, and n^sub i^ is an i.i.d. standard normal error term.

We cannot observe the net benefits of participation, but we assume that properties are signed up (i.e., VCP = 1 ) when the net benefits of participation are positive. This allows us to estimate a probit equation:

E(VCP^sub i^ = 1) = Pr(VCP*^sub i^ ≥ 0) = Φ(x^sub i^β), (2)

where Φ(.) is the standard normal c.d.f. Because a site can only participate in the program once, we specify the log likelihood function as:

(3) ...

where i denotes the site, t denotes the year of the program, and 5 is the set "at risk" (i.e., the set of candidate sites that have not participated yet). Equation (3) describes a discrete-time duration model, which can be used to understand which characteristics of the property will lead to earlier or later enrollment in the program.

C. Property Prices

Turning to the equation describing property values, let r denote the log market value of a property and z denote a vector of physical characteristics of the parcel, structural characteristics of the building, and neighborhood characteristics. Formally,

(4) y^sub ij^ = z^sub ij^γ + VCP^sub ij^δ + ε^sub ij^,

where i denotes the parcel and j the sale (j = 1, ..., J^sub i^), and ε is an econometric error term. Coefficient δ captures the effect on participation on property values.

To estimate Equation (4), it is necessary to gather sale information for the properties in our sample that do and do not sign up for the program and, for those which do, for sales before and after participation. Estimation is

complicated by the fact that participation and value may be influenced by common unobservable parcel characteristics, which make ε and η correlated within a parcel, y and VCP endogenous, and the ordinary least squares (OLS) estimates of δ in Equation (4) biased.

Several approaches are available to circumvent this problem and obtain consistent estimates of δ. The first is based on the so-called propensity score. This approach relies on the assumption that the parcel's propensity to participate in the program, when added in the right-hand side of regression equation (4), controls for possible self-selection bias and contains all the relevant information in the covariates that is relevant for disentangling the effect of participation (which is interpreted as a "treatment" on the parcel) (Rosenbaum and Rubin, 1983; Wooldridge, 2002, pp. 612, 615-620). In practice, estimation is carried out in two steps. The first step fits a probit model of participation and uses the estimated coefficients to form the predicted probability of participation. This predicted probability of participation is then included in the right-hand side of Equation (4), and an OLS regression is run.

Alternatively, if one is prepared to assume that ε and η are jointly normally distributed, the expectation of í conditionally on observing participation or nonparticipation is

(5) ...

where α = Cov(ε, η;) and λ^sub i^ is equal to φ(z^sub i^β)/ Φ(z^sub i^β) if VCP^sub i^ = 1 and to -φ(z^sub i^β)/[1 - Φ(z^sub i^β)] if VCP^sub i^ = 0. On appending an error term, Equation (5) becomes an econometric equation, which can be estimated using a two-step procedure. The first step is analogous to that of the propensity score approach. The second step simply replaces λ^sub i^ in Equation (4)-the inverse Mills' ratio term-with its prediction, A/, obtained by plugging the estimated probit coefficients into the expression for λ^sub i^ and then runs OLS.

As long as the error terms are truly normally distributed, a consistent estimate of δ can also be obtained by simply regressing y on z and VCP^sub it^ = Φ(x^sub it^β), as shown by Heckman (1978). All the three approaches require using heteroskedasticity-robust standard errors because heteroskedasticity is introduced into the second-step equation when we use predictions in lieu of the true parameters to calculate the propensity or the inverse Mills' ratio term.9

Since the two approaches based on Equation (5) are sensitive to the departure from the normality assumption (Wooldridge, 2002, p. 564), in this paper attention is restricted to the propensity score approach.lo The first-stage probit model must use information from all parcels in all periods rather than "retiring" a parcel from the sample once it is enrolled as we did in the discrete-time duration model of Equation (3).11 In the absence of unobserved heterogeneity, the log likelihood function for the first-stage probit is thus: 2000 432

(6) ....

An amended version of Equation (6) allows for random effects ( see Greene, 2003).

As a final point, it should be noted that sale prices are observed only when a parcel is sold. It is possible that characteristics of a parcel unobserved to the investigator influence its likelihood to be sold as well as its price and that a form of self-selection takes place. Ideally, one would want to control for these unobserved characteristics by specifying an additional equation that describes propensity to sell and by implementing a "heckit" sample selection correction ( see Gatzlaff and Haurin, 1998). We attempted to implement the heckit procedure but in the end abandoned it because we were not able to identify significant predictors of a sale.12 In sum, we correct our price equation for self-selection into the VCP, but we are forced to ignore self-selection into market transactions.

D. The Benefits and Costs of Participation in the VCP

We assume that the owner of a property will enroll a parcel in the Colorado VCP if the net benefits of participation are positive. The benefits of participation should include the avoided expected liability costs plus any reduction in the environmental assessment of the site, preparation of cleanup plan, and cleanup costs afforded by participation in the program and due to agency oversight and simplified cleanup standards. These should be added to the tax credit received if the site meets certain requirements and the property owner, upon completing cleanup and receiving a "NAD" from the Colorado Department of Public Health and Environment, applies for such a credit through the state's brownfield program.

An additional component of the benefits of participation may be (the value of) the reduction in uncertainty surrounding remedial work (Urban Institute et al., 1997). Finally, whether or not any actual remediation work takes place, the final NAD may well serve as a clean bill of health from the state, dispel any possible contamination stigma, and raise the value of the site by more than the mere saving in cleanup and liability costs.

The costs of participation include, of course, the cost of cleanup (if any), the participation fee charged by the state ($2,000), plus the cost of dealing with the state agency during the various stages of participation in the program. All else the same, one would, therefore, expect the costs of participation to be low at those sites where it can be shown that cleanup is not necessary (or where the owner or operator can escape the responsibility for cleanup because pollution has migrated from an offsite source).

E. The Choice of Independent Variables

We do not observe the net benefits of participation or any of the individual categories of costs and benefits described in the previous section. We therefore capture them with characteristics of the parcel and the neighborhood.

These include the size of the parcel, a dummy for the presence of structures at the site (BUILDING), and capital intensity (CAPITAL), defined as the square footage of the building divided by the size of the site, and the age of the building. Earlier studies of residential or industrial real estate have consistently found these factors to be associated with the value of a parcel (see Dobson and Goddard, 1992; Sivitanidou, 1994; Sivitanidou and Sivitanides, 1995). They may also influence remediation costs and any demolition costs. For example, heavily built sites may differ from others in terms of demolition costs and cleanup costs because of toxic construction materials (e.g., asbestos, heavy metals).

Distance to the central business district (CBD) and access to various types of roads (e.g., interstates, state highways, and local roads) should influence the attractiveness of a site. Our regressors also include a dummy for whether the parcel lies in a brownfield zone (BZ), in which case, the owner (or developer) is entitled to tax credits after cleanup. We also include controls for economic incentives or burdens that are not linked to cleanup but are in effect when a business is established at the parcel's location or for owning property at the parcel's location. Specifically, we form a dummy indicator for whether the lot is in a Colorado or federal enterprise zone (which grants tax credits for establishing a business in a zone designated as such but is unrelated to contamination or cleanup) (EZ) and property tax rate that applies to the parcel (MILLLEVY).13

In some cases, we were unable to determine whether the site is in an enterprise or in a brownfield zone.14 We created two dummy variables (EZJJN KNOWN and BF_ UNKNOWN) to keep track of this, which we both entered in our regressions along with the EZ and BF dummies. The coefficients of the latter should then be interpreted as capturing the effect of being in such zones on participation in the program, conditional on the availability of information.

The benefit-cost calculus driving participation could also be influenced by the land use around the site. We have information on the percentage of land in the 500- and 1,500-ft buffer around the parcel slated for residential, industrial, and commercial use (from EPA's Multi-Resolution Land Characteristics Consortium15). We gathered information from the 1990 and 2000 censuses about median housing rents in the parcel's census tract and characteristics of the residents in the parcel's ZIP code and county (education, race, votes for the Democratic candidate in the most recent presidential elections, and sources of drinking water).

Regarding contamination, we consulted dockets at the Colorado Department of the Environment and U.S. EPA. Contamination can be described using dummy variables for the presence (actual or suspected) of specified pollutants at the site (e.g., heavy metals, petroleum, solvents, etc.) and the contaminated environmental medium (e.g., groundwater or soil). The former or latter type of information is available for 73% of the sites. Alternatively, as Howland (2000. 2004), one can create a dummy denoting whether contamination was confirmed (or was simply suspected and eventually ruled out).

The variables that we have described thus far are good candidates for the vector ? of determinants of participation in the VCP. They are also good candidates for the vector z.16

F. The Sample

Observers of the Superfund program and the brownfield problem assert that sites previously or currently listed on EPA's registry of sites needing investigation or cleanup are the supply of brownfields (Bartsch, Collaton, and Pepper, 1996). Based on this view, the main sample ("Sample 1") for our analysis comprises sites in Colorado that are known to be contaminated or previously believed to be contaminated and meet the VCP eligibility requirements.

Specifically, we pooled (a) Colorado sites in the CERCLA Information System database (CERCLIS) (H = 222), (a) Colorado sites in the CERCLIS archives, which consist of sites previously placed in CERCLIS, but delisted in 1995 because they were not found to pose meaningful risks (or had been cleaned up) (;; = 456), and (c) sites that participated in the Colorado VCP as of August 2000 (n = 188).17 We then excluded sites that do not meet the eligibility requirements, such as sites on the National Priorities List, solid and hazardous waste sites, LUST sites, sites covered under the Uranium Mining and Tailing Recovery Act, military installations, and federal sites. This reduced the sample from 857 to 623. We further restricted attention to sites in the nine Front Range counties (Adams, Arapahoe, Boulder, Denver, El Paso, Jefferson, Larimer, Pueblo, and Weld),18 which results in a final sample size of 432 sites.

To examine the sensitivity of the results to this definition of candidate parcels, we repeated all analyses and models on an alternate sample that is the union of (c)-all properties that participated in the Colorado VCP as of August 2000-and (d) a random sample of properties in the same areas as in (c) and with similar land use restrictions (i.e., slated for commercial, industrial, and mixed use but not for single-family homes). Sites in Group (d) are not listed in any EPA or state registries of contaminated or LUST sites, and there is no particular reason to believe that they are contaminated. In what follows, we refer to this alternative sample as "Sample 2." There were a total of 156 sites in (c) after we restrict attention to the Front Range and 188 sites in (d), making the total sample size for Sample 2 equal to 344.

The difference between Sample 1 and Sample 2 should be clear: the former focuses on what many observers consider the obvious supply of brownfields (listed sites), while with the latter we attempt to create a study with a "control" and a "treatment" group ((d) and (c), respectively; see Wooldridge, 2002, for a discussion of quasi experiments). For good measure, with both samples we implement estimation procedures that allow for the treatment (participation in the VCP) to be endogenous with property values.

V. THEDATA

Our first order of business is to examine the composition of our samples in terms of site contamination and participation in the VCP. We begin with Sample 1. Of the 432 eligible sites in the Colorado Front Range, 159 (36.5%) applied for participation in the Colorado VCP. The majority of the participants ( 102 or 64.5%) entered the program by directly applying for a letter of NAD. The remaining 57 (35.5%) submitted a cleanup proposal for review by the Colorado Department of Public Health and Environment.

Only six sites in the CERCLIS registry and only three sites in the CERCLIS archives signed up with the Colorado VCP.'9 This suggests that the Colorado VCP has not absorbed contaminated sites or brownfields from these sources and has solicited participation from an entirely new crop of parcels. This is consistent with the notion that VCPs lead to the discovery of new contaminated sites (U.S. GAO. 1997) but casts doubts about the extent of the actual cleanup activity.

Descriptive statistics of our universe of parcels are reported in Table 1. We were able to obtain acreage and improvement information for 61% of parcels. The average parcel size is 1,421,033 sq. ft. Almost 51% of the parcels for which we have improvement information contain a building. The mean "capital intensity" is 0.36, and our parcels are at an average distance of 5,429 m from the CBD and at an average of 770 m from the nearest road.

Table 2 presents economic inducements and land use for Sample 1. Of the parcels for which we have this information, we know that 36.7% are in an enterprise zone and 39.11% in a brownfield zone. The property tax is on average $73.79 per $1,000 assessed property value, and there is quite a bit of variability in property taxes within and between counties. On average, residential property accounts for 16.31% of the 1,500-m buffer zone around our parcels.

The presence of contamination is confirmed at 31.5% of the sites. Groundwater and soil are contaminated at almost 44% and 31.6% of these sites, respectively. Solvents are the most common pollutant, and they are present at 45% of the contaminated sites, followed by petroleum and hydrocarbons (30% of the contaminated sites).

Regarding sale price data, we were able to identify a total of 245 transactions that took place between 1974 and 2002. We restrict attention to the 119 arm's-length sales with positive price recoded by the tax assessor's office between 1980-the year the Superfund law was passed-and 2002. Of these, 43.7O1X, were for sites that at some point enrolled in the program. Ninety sales took place at nonparticipating parcels (67) or at participating parcels (23) before they enrolled in the program. Twenty-nine parcels occurred after a participating sale signed up for the VCP.

The sale price, expressed in 1988 constant dollar, is on average about $854,116, ranging from $5,326 to $6.5 million (from $7 to $112/sq. ft). Split-sample Mests suggest that the mean price (whether total price of the property or per square foot) does not differ across participating and nonparticipating parcels and that no difference exists among the mean prices of participating parcels before and after participation. It remains to be seen, however, if these results are confirmed or refuted when we run regressions that control for site characteristics.

Descriptive statistics for Sample 2 are displayed in Tables 3 and 4. We checked whether parcel and neighborhood characteristics were different across sites that did and did not participate in the VCP in this sample and, as expected (nonparticipating sites were selected at random from the tax assessor records within the same areas as participating properties), did not find significant differences. The only exception was distance to the CBD, which was slightly shorter among participating properties. We have a total of 205 arm's-length transactions. A /-test of 1.96 indicates that sale prices are marginally statistically different across the two groups of parcels, begging the question whether this difference remains or disappears after cleaning the data and controlling for parcel characteristics.

Vl. ESTIMATION RESULTS

A. Participation in the Colorado VCP

In this section, we discuss the results of probit model (Equation (3)) for the main sample. By participation, we mean a direct application for either a NAD or an actual cleanup proposal (VCLJP application).

The results of two specifications of probit model (Equation (3)) are reported in Table 5 (Panels A and B). The specification in Panel A is the more parsimonious and suggests that participation occurs sooner among larger candidate sites. Doubling the size of the parcel increases the likelihood of participating in any given year for the average parcel that has not been signed up yet from 0.04 to 0.07, a 73'Xi increase. Since lot size information was missing for a number of sites, we included a dummy (SQFTMISS) taking on a value of 1 if no information on the size of the parcel is available.20 The coefficient on this dummy is statistically insignificant, implying that sites for which we were not able to gather acreage information do not enroll in the program any sooner (or later) than those for which we do have this information.

The presence of a building tends to hinder participation. For properties without buildings, the likelihood of participating in any given year is over 61% greater than that for comparable properties with buildings. Capital intensity is not a significant determinant of participation. Sites located farther from the CBD are less likely to participate, but in this specification, the coefficient is not statistically significant at the conventional levels.21

We attempted various proxies for the site's contamination, and in the end, we settled for a dummy denoting whether contamination was confirmed, or simply suspected and then ruled out, at the site. The coefficient on this dummy is positive but not significant at the conventional levels. This confirms our earlier assessment that most of the sites participating in the program are not on official registries and wish to obtain a No Further Action determination without offering to conduct remediation at the site.

The variable most strongly associated with participation is the percentage of land slated for residential use within 1,500 m of the parcel, interacted with a dummy for whether the parcel is intended for commercial or industrial use. The coefficient on this variable is positive and significant at the 1% level, suggesting that property owners or developers may have had conversion to residential use and residential development in mind when they opted for participating in the VCP.22 The elasticity of participation with respect to the percent of residential area in the immediate vicinity is about 0.20.

The specification of Panel B adds controls for other economic development incentives or disincentives. In this specification, the physical characteristics of the parcel are no longer statistically significant, perhaps as a result of the collinearity between several of the regressors,23 but their coefficients retain the signs as in specification A.

The coefficients on the enterprise zone (EZ) and brownfield zone (BF) dummies are positive-a result consistent with our expectation that being in an enterprise incentive zone may increase the attractiveness of a site for investment purposes and hence the likelihood of participation in voluntary cleanup-but small and insignificant. Dummies denoting lack of information for these variables are likewise statistically insignificant, confirming that these sites are not different in their probability of participation from those for which we do have information. The property tax rate has the expected negative coefficient, but again the association with participation is very weak. We rerun the specification of Panel B as a random-effects probit, but we did not find any evidence of unobserved heterogeneity.24

B. Effects on Price

Estimation results for Equation (4) based on the propensity score technique are displayed in Table 6 for the main sample.25 Property prices increase significantly with lot size and building capital. Parcels for which we do not have size information are not different in value, on average, from those for which we do. The coefficients on building age and its square are weakly associated with prices, and their magnitude indicates that value tends to decline in an approximately exponential fashion with age rather than in a strictly quadratic fashion.

Contrary to expectations, distance to the CBD is not a significant determinant of value. The applicable property tax rate is negatively associated with value, an effect that is significant at the 10% level. Parcels in an enterprise zone transact at a 38'Xi discount with respect to comparable lots in other areas (an effect that, however, is not statistically significant), but being in a brownfield zone does not have a discernible effect on price.

The confirmed presence of contamination also implies a discount relative to parcels where the contamination was ruled out. This discount is about 53%, which is comparable to that estimated by Jackson (2002) and Howland (2000, 2004). The coefficient at the heart of this analysis is, of course, that on VCP participation dummy, here dubbed PARTICIPANT.-6 The magnitude of this coefficient implies that if a parcel in our universe is sold after it signs up with the VCP, its value rebounds by 481Xi over its pre-participation price. Taken at face value, this would suggest that participation in VCP is beneficial to property values and that prices recover partially with participation but will remain below those of comparable uncontaminated sites.27 However, this effect is not statistically significant at the conventional levels.

We experimented with rerunning the regression after excluding several regressors with insignificant coefficients and obtained similar coefficient and p values for the coefficients on contamination and participation. The alternative estimation approach that relies on Equation (5) and uses the first-stage probit to compute inverse Mills' ratios yields similar results. The price discount due to the presence of confirmed contamination on the premises is 56%, which compares well with the 53% identified by the propensity score approach. The coefficients on virtually all other variables are extremely close to those of the model of Table 6, and participation is likewise found to raise property values albeit weakly in terms of statistical significance. However, the size of this effect-participation would, all else the same, more than double property values-is not credible. We conclude that the data hint at the possibility that participating parcels appreciate in value, but the evidence about this effect is statistically weak and it is difficult to nail down its magnitude.

C. Results for Sample 2

Since Sample 2 was formed by augmenting participating sites with sites selected at random in similar neighborhoods, we do not expect many variables to be significantly associated with the probability of participation. This expectation is borne out in the data: the only significant predictor of participation is the property tax rate (see Tables A2 and A3).

The results of the hedonic price model (corrected for endogeneity using the propensity score approach) are displayed in Table 7. These regressions broadly confirm the results of the analysis for the main sample. The price of a property increases, as expected, with property size and with the presence of the building on the property. The coefficients on these regressors are within approximately 10% of their counterparts for the main sample.28

We were forced to omit brownfield zone classification from this regression to limit collinearity problems, but a property's location within a county enterprise zone was found to be weakly associated with lower values. Property values with confirmed contamination problems sell at a 43% discount relative to sites with no known contamination, but participation in the VCP results in a virtually complete recovery of value. These effects broadly confirm those found for the main sample, but the price discount due to contamination is less severe and the recovery in value from participation is stronger than that seen for the main sample.

VII. CONCLUSIONS

This paper explores the determinants of participation in the Colorado VCP and the effect of participation on the property's value. We find that very few of the properties that participate in the program are on the EPA's registries of contaminated sites or of sites previously thought to be contaminated. This confirms the remark that VCPs may lead to the discovery of new sites. In Colorado, however, these sites are probably not very seriously contaminated and/or their owners tend to apply directly for a No Further Action determination rather than proposing to undertake environmental remediation.

The most important factors associated with participation are the size of the site and whether the site is surrounded by properties slated for residential use. Additional incentives to economic development and remediation, such as the availability of enterprise or brownfield zone tax credits, do not discernibly affect the likelihood of participation. These findings should be interpreted with caution -since the presence/absence of these incentives tends to be correlated with other site characteristics (see Bartik, 2004, for a discussion of this problem). We conclude that sites that are more likely to participate in the Colorado VCP are probably those with comparatively high redevelopment potential.

Does participation raise property value? When attention is restricted to our main sample-participating properties plus sites on EPA's lists-we find that contamination reduces substantially property values, all else the same, and that there is some appreciation associated with participation in the program. The recovery in value for contaminated properties is not complete. However, the results are statistically weak. This is probably due to the relatively small number of arm's-length sales we were able to identify over 1980-2002 for these parcels.

We look for more evidence using an alternate sample comprising the properties participating in the VCP plus others selected at random from the universe of parcels in the same areas slated for the same uses (mostly commercial, industrial, and mixed use). In sum, we find that the contamination discount and the effect of participation are robust, with the former in the range of 43%-56% and the latter suggesting complete recovery.

Taken together, these findings cast doubt on whether the VCP is truly attaining its original cleanup and environmental remediation goals and hint at the possibility that participation might be driven exclusively by the desire to rid the parcel of any stigma associated with current or previous use of the land (or to prevent such an effect with future buyers). Since the 1997 U.S. GAO study found that permanent remediation was undertaken for less than 50% of the voluntary cleanups, it would seem that the reliance on VCP may be insufficient to encourage true environmental remediation and/or may create its own set of perverse incentives rather than reversing those linked with liability- and enforcement-based programs. Clearly, more research is needed in this area, and data from programs of different states must be collected and analyzed to answer these questions.