Document View

We extend the literature on the impact of externalities using an approach based on a hybrid of hedonic and repeat-sales methods. The externality in question is groundwater contamination in Scottsdale, Arizona. The use of condominium sales allows us to assume that major physical characteristics remain unchanged, but location parameters may be altered by urban growth and development as well as contamination. We find an economically significant discount for properties located in the contaminated area. Interestingly, it does not appear until several years after the contamination becomes publicly known, and it seems to have disappeared before the end of the study period.

The last three decades have seen the emergence of a voluminous literature examining the interaction between environmental factors and real estate markets, much of it concerning the extent to which negative environmental spillovers are capitalized into real estate values. These interests have given rise to two broad categories of real estate literature reviewed in Jackson (2001). The first of these is dominated by the appraisal profession and has focused on valuation concepts and methods (see, e.g., Chalmers and Jackson 1996, Jackson 1997, Weber 1997, Kinnard and Worzala 1999). The second category reports the results of empirical studies for the effects of environmental spillovers on real estate prices, especially in residential markets. The forms of environmental spillovers considered relate to land, water and air (Segerson 2001) and have included proximity to landfill sites (McClelland, Shulze and Hurd 1990, Kohlhase 1991, Thayer, Albers and Rahmatian 1992) as well as the presence of overhead power lines (Colwell 1990), air pollution (Graves et al. 1988), water quality (Leggett and Boekstael 2000), noise pollution (Pennington, Topham and Ward 1990) and groundwater contamination (Kiel 1995, Kiel and McClain 1996).

There are three general scenarios to describe the response of real estate prices to an environmental spillover. In the most straightforward, the presence (or severity) of an environmental problem is associated simply with the constant presence (or severity) of a price discount on properties affected by that problem. For example, properties affected by environmental contamination could be priced at a discount (perhaps equal to a constant percentage of value) relative to unaffected properties, or the discount could be a linear function of the severity of the contamination affecting that property. As the discount is related only to the presence (or severity) of contamination, it remains constant until the contamination is mitigated (wholly or to some degree).

The presence or severity of contamination, though, can be considered a (negative) hedonic attribute of the property just like any other, and therefore the implicit market price associated with contamination may change over time just as it does for other hedonic attributes. In the second scenario, then, the presence (or severity) of the environmental occurrence is associated with a price discount on affected properties, but the magnitude of the discount evolves in the same way that implicit marginal prices on other hedonic attributes evolve. For example, if the implicit discount associated with environmental contamination is elastic with respect to wealth, then the magnitude of the discount can be expected to grow over time as wealth increases (controlling for any mitigation). Alternatively, if information about environmental contamination dissipates through time then the discount may itself dissipate even if the presence and severity of contamination do not.

Neither of these scenarios, however, takes into account the dynamic response of housing consumers to an occurrence such as environmental contamination. Consumers who attach a relatively high premium to environmental factors can be expected to bid (or ask) relatively high prices for properties with favorable environmental attributes, and to bid (or ask) relatively low prices for those with environmental problems. Therefore, these environmentally sensitive consumers will tend to outbid less sensitive consumers for properties with favorable environmental attributes, and will tend to be outbid by less sensitive consumers for properties with environmental problems. In response to the sudden occurrence (or recognition) of an environmental problem, the outcome of such a market adjustment or "sorting" process is likely to be this: the households that are most sensitive to the occurrence will offer their properties at the greatest discount in order to leave immediately; they will be replaced by households with the least sensitivity to the occurrence; this process will be repeated at successively smaller magnitudes for successively less sensitive sellers and more sensitive buyers; and thus the initial discount will decline steadily in magnitude, even in the absence of any mitigation of the environmental contamination. Of particular interest in this article is the application of house price modeling to assess whether transaction prices reveal the effects of such a sorting process, and if so how long the adjustment process seems to take.

In a recent review article, Boyle and Kiel (2001) examine more than 30 papers that use hedonic approaches to test the effects of environmental externalities on house prices. They note that a common problem is that the tests are often conducted over short periods and as a consequence are generally unable to effectively capture changes in price over time. The main problem that Boyle and Kiel identified is the absence of data from before and after an environmental event, which means that not even the immediate effect of the event on prices can be adequately identified. (A solution to this problem is found in Colwell, Dehring and Lash (2000).) Boyle and Kiel also, however, note a common inability to detect any change in the contamination effects after the initial contamination event-that is, the studies they examined tended to focus on the first and most straightforward scenario without considering any evolution of the price effect over time.

Although interest in the persistence of contamination effects and their longerterm impact on property values is not new (Riechert 1999), these effects have been difficult to determine in hedonic studies, so previous studies provide conflicting evidence of the temporal effects of contamination and remediation (see Kiel 1995 versus Kohlhase 1991 and Kiel and McClain 1996). Some researchers (Mieszkowski and Saper 1978, Hite et al. 2001) have tried to use control areas to infer what would have happened had the externality not occurred. The success of this approach is, of course, dependent on the researcher's ability to limit the differences in inter alia market conditions and neighborhood quality between the control and study areas or to accommodate the differences by including sufficient spatial variables. Dale etal. (1999) attempted a variant of the control area approach by examining various neighborhoods and found that after a cleanup house prices recover more slowly for dwellings in closest proximity to the negative externality.

In contexts other than contamination, real estate analysts have used repeat-sales regression analysis to examine temporal change in property values. Hedonic analysis is superior to repeat-sale analysis when there are few repeat sales, but repeat-sales analysis may be superior to hedonic analysis for this purpose when there are sufficient repeat sales, when data on some of the key property attributes are unavailable and when the analyst can be confident that those attributes will not have changed over time. Changing values for property attributes or model parameters can be modeled within a hedonic price model, but early applications of the repeat-sales model used simple versions that lacked this capability. A notable exception was Palmquist (1982), which incorporated changing values of an indicator of environmental quality (noise pollution), but subsequent studies failed to build on this observation. Shiller (1993), however, showed how changes in measured attributes or in parameter values could be incorporated more generally into the repeat-sales model, thereby removing a key advantage of the hedonic price model.

In this article, we employ the hybrid repeat-sale/hedonic approach suggested by Palmquist (1982) and Shiller (1993) to investigate the effect of environmental contamination in a particularly rewarding situation. In this application the timing of the first perception of contamination is well known while changes in unmeasured attributes are likely to be minimal, which means that observed changes in prices can be reliably decomposed into those attributable to contamination and those attributable to changing marginal prices in the market. Moreover, our data collection period is quite long, especially after the contamination was recognized, which enables us to investigate whether the price effect of the initial contamination persists, grows or dissipates over time.

The remainder of the article is divided into four major sections. The next section introduces the hybridization of hedonic and repeat-sale models. The third section discusses the data and the specific models to be estimated, while the fourth section discusses the empirical results. Our findings are summarized in the concluding section.

Hedonic and Repeat-Sale Models

Hedonic house price studies have been employed for nearly four decades to assess the impact of negative externalities such as air pollution (Ridker and Henning 1968). The hedonic method has subsequently yielded a vast applied literature, the basic premise of which is that by estimating the implicit price of each of the physical and locational attributes associated with a property it is possible to isolate the impact of environmental events on the price surface. In this context, estimates of the price surface from the period before and after the event are essential. For example, if the producers of negative externalities select sites in less desirable neighborhoods to reduce costs, then without before-andafter data lower prices might be interpreted as evidence of negative externalities while they may reflect only preexisting market conditions. As we note above, however, often the paucity of data imposes a constraint on before-and-after analysis.

Analysts tend to use repeat-sale analysis to estimate price indices when they do not have sufficient attribute data to use hedonic analysis for this purpose, but there are several dimensions to the downside of using repeat-sale analysis. First, data on assets that sold only once during the study period are ignored. Second, several empirical studies (Mark and Goldberg 1984 and Case, Pollakowski and Wachter 1991) have concluded that repeat sales understate house price inflation by failing to account for aging and depreciation between sales (Clapp and Giacotto 1992), overrepresentation of starter homes among repeat transactions (Clapp, Giacotto and Tirtiroglu 1991) and overrepresentation of "lemons" (properties that are substandard in some way) (Clapp and Giacotto 1992). Curiously perhaps, others (Crone and Voith 1992, Gatzlaff and Ling 1994) have found evidence that repeat-sales indices can overstate, rather than understate, price appreciation by failing to control for fix-up costs and the high representation of short hold properties (Steele and Goy 1997, Clapp and Giacotto 1999, Costello 2002). Advances in repeat-sales modeling have begun to ameliorate some of these problems: Cannaday, Munneke and Yang (2002), for example, developed a means of disentangling age effects from temporal effects.

The empirical approach employed in this article can be thought of as a hybrid model incorporating hedonic characteristics in a repeat-sales framework. Although a hybrid approach has been used in other studies, the focus of these papers has tended to be on technical aspects of index construction, rather than the assessment of the impact of externalities (Palmquist 1982, case and Quigley 1991, case, Pollakowski and Wachter 1991, Shiller 1993, Quigley 1995, Meese and Wallace 1997, Hill, Knight and Sirmans 1997, England, Quigley and Redfearn 1999).

The Foundations of Repeat-Sale Analysis

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Equation (2) is the well-known repeat-sale analysis equation in which the dependent variable is the ratio of prices, the attributes and the implicit prices of the attributes are gone, and the time variables in parentheses take on the value - 1 if the first sale occurs during that period, +1 if the second sale occurs during that period and 0 if no sale occurs during that period. Thus, the dummy variable is no longer dichotomous. The equation is estimated by taking the natural logarithm of both sides and using ordinary least squares regression. Although very commonly used for the purpose of price index estimation, by itself this equation will not generate any but the most indirect evidence of a price effect for an environmental spillover.

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Given appropriate data, this formulation can be used to investigate (1) the discount associated with an environmental event (a discrete change in attributes), (2) the evolution of house prices in response to mitigation of the environmental problem (a continuous change in attributes) and (3) the evolution over time in the environmental discount even in the absence of mitigation (a parameter variation).

Unmeasured Physical Changes

Enlarge 200% Enlarge 400%

Data and Modeling

We employ a data set that is particularly appropriate for assessing changes over time in the discount associated with the sudden occurrence of an environmental problem. Our data set was developed as part of a case known as Baker vs. Motorola, in which a group of homeowners sued a manufacturer for damages sustained as a result of the discovery of long-term environmental contamination. Until the 1970s, industrial solvents containing volatile organic compounds (VOCs) were typically disposed of directly onto the ground or in dry wells. In Scottsdale and Tempe, Arizona, these practices resulted in contamination of groundwater over an area of approximately 13 square miles by VOCs including trichloroethylene (TCE), tetrachloroethylene (PCE) and chloroform.

The groundwater contamination was discovered only in late 1981 in several drinking water wells. Local water providers immediately stopped using those wells for drinking water, and in 1983 the entire site was placed on the U.S. Environmental Protection Agency's National Priorities (Superfund) List. Remediation efforts have since been ongoing; specifically, the contaminated groundwater is pumped to the surface where exposure to air removes the volatile compounds, after which the groundwater is pumped back underground. It is anticipated that this treatment will be required for the next 40 years, although the effectiveness of the treatment is uncertain. The length of time that mitigation is expected to require, as well as the uncertainty of ultimate success, suggest that any decline in the discount associated with groundwater contamination is likely to signal a decline in the market's assessment of the value of clean groundwater, rather than a decline in the physical severity of the contamination.

The defense in the lawsuit acquired data on transactions of properties in areas both affected and unaffected by the contamination, covering time periods both before and after the contamination was discovered. The data were accepted by the plaintiff as being reasonably accurate and were entered as evidence in the case, making it public information according to the plaintiffs' attorneys.

The area covered by the data set encompasses a large number of condominium properties, which makes it particularly useful for our purposes because structural characteristics of condominium properties are quite unlikely to have changed over time. Importantly, there was no public information regarding possible contamination prior to the announcement very late in 1981 that contamination had occurred. (If there was any leakage of information prior to the public announcement-which we do not believe occurred-then that fact would affect our estimate of the magnitude of the initial discount associated with environmental contamination, but not our analysis of its persistence after the public announcement.) The study period encompassed by the data set extends from the beginning of 1980 through 1998, which means that we have essentially two years of property transactions that were unaffected by contamination followed by 17 years of transactions after the event. (Of course it would have been preferable for our purposes to have more transactions prior to the event, but the defense selected the time period and the plaintiff acquiesced, each for interests that did not coincide with ours.)

Data

The initial data set describing condominium sales in Scottsdale and Tempe comprised 30,199 transactions. A small number (333) were duplicates, and a few (38) were eliminated because they were incomplete (lacking data on price and/or year) or because we judged them not to be valid repeat transactions. (For example, several of them appeared to be transactions not of property but only of partial interests, such as those resulting from a divorce; others appeared to involve an initial transaction of raw land along with a subsequent transaction of the developed property.)

Enlarge 200% Enlarge 400%

The remaining 29,827 transactions are summarized in Table 1. As the table shows, 7,347 transactions described properties which sold only once during the study period. Although case and Quigley (1991) and others have shown how data on once-transacting properties could be incorporated into a hybrid repeatsale/hedonic model, for this application we focus only on repeat-transacting properties and therefore removed the once-transacting observations from the analysis.

It is common practice to eliminate any pair of consecutive transactions that occurred during the available time interval (the calendar year for this data set) because the observations on the time dummies in a repeat-sales model would all equal zero. As the Appendix points out, however, this approach is at best inefficient and in practice may introduce bias. The use of consecutive (rather than nonconsecutive) transaction pairs is simply a matter of convenience and is not required. Thus, the alternative approach suggested in the Appendix is to replace pairs of consecutive transactions occurring in the same time interval with pairs of nonconsecutive transactions occurring in different time intervals to preserve the information contained in all independent price-relatives. A property that transacted only in one time interval, however, cannot be incorporated into a repeat-sales analysis no matter how many transactions were involved in that time interval; in our data set this meant that 191 properties that sold more than once but never in more than one year had to be deleted.

Enlarge 200% Enlarge 400%

Table 2 presents basic descriptive statistics for the 13,612 transaction-pairs that are available for use in this analysis. As noted, the properties in our data set are all condominium units, which implies that changes in structural characteristics were rare if indeed they ever occurred-an important advantage when a repeatsales modeling approach is employed. The average condominium was 12.5 years old at the time of its second transaction and sold for $64,428. Most importantly, 44% of the transaction-pairs involved properties that were located within the contaminated area.

Modeling

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Case and Shiller (1989) suggested that the drift of individual property values through time would likely be reflected in heteroskedastic disturbance terms in any repeat-sales model, and that this source of heteroskedasticity could be eliminated by regressing the squared residuals from a first-stage repeat-sales regression on the time interval between sales. We employ this correction (using a quadratic functional form) in all of our empirical estimates.

In addition to this now-standard correction of the basic repeat-sales model, Palmquist (1982) pointed out that the error covariance matrix of a repeat-sales model is not diagonal if any of the properties transacted more than twice during the study period. The exact structure of the covariance matrix is known, however, so it is straightforward to apply the estimator suggested by Aitken (1935), as we describe in the Appendix.1

Results

The regression results are very much what we anticipated. The results suggest that the shape of the value surface has undergone both discrete and continuous changes over time, indicating the value of including some hedonic characteristics in the empirical estimation. Most importantly, we discover a contamination discount that varied over the post-contamination period.

As a baseline we estimated the standard repeat-sales formulation commonly used in constructing price indices. (As in all the models, 1980 was the excluded dummy year.) The adjusted R2 is 29.0%, and F tests indicate that we can reject the null hypotheses that all annual price changes are zero (fig, 13593 = 295) and that all annual price changes are identical (Fn, 13593 = 312). The results are shown in Table 3. The most obvious feature of the results is that after a general rise in condo prices from 1980 through 1983, prices declined steadily through 1991, reaching a trough more than 20% below the 1980 base year (1 - e^sup -0.22520^ - 20.16%) before rising steadily through 1998 until they stood almost 17% above their base level (e^sup 0.15621^ - 1 = 16.91%). Another clear result is that the intercept is positive and significant, indicating nontemporal improvements on the order of approximately 3.6%.

Enlarge 200% Enlarge 400%

The next step is to introduce a constant contamination discount through the variable CNTM^sub i^([varphi]^sub 82,i^ - [varphi]^sub 82,i^). As shown in Table 4, this change improves the fit of the model only slightly, increasing the adjusted R^sup 2^ to 29.1 %. F tests again indicate that we can reject the hypotheses that all annual price changes are zero (F^sub 18,13592^ = 291) and that all annual price changes are identical (F^sub 17,13592^ = 308). The dip in condo prices is largely unaltered by the inclusion of the contamination variable. Perhaps the only notable change to the annual price indices is that prices in the 1998 were more than 19% above the 1980 prices as compared to slightly less than 17% in the baseline model.

The environmental story is that the coefficient on the contamination variable is negative and significant, indicating that condos located in the contaminated area suffered a discount of 4.65% relative to condos in noncontaminated areas. This is interesting from two perspectives. First, it strongly suggests that the effects of groundwater contamination were capitalized into lower property price: unless there is an excluded variable that is correlated with the contamination variable and that offers an alternative explanation, it is reasonable to conclude that groundwater contamination led to price discounts. Second, the results provide our first glimpse at the usefulness of including hedonic characteristics in a repeat-sales model.

Enlarge 200% Enlarge 400%

Although Model 2 suggests the existence of a contamination discount, we are more interested in whether the contamination discount grows or dissipates through time, and to this end Table 5 shows the results of Model 2a permitting the contamination effect to follow a linear trend. The contamination discount for the year 1982is the negative of the coefficient on the CNTM^sub i^([varphi]^sub 82,i^ - [varphi]^sub 82,i^) variable, that is, the immediate effect of the contamination announcement was to lower prices in the area of contamination by approximately 2.43% (slightly less than in Model 2). The negative of the coefficient on the CNTM^sub i^ ([varphi]^sub 82,i ^- [varphi]^sub 82,i^)(Δ^sub 82,i^) variable is the constant annual increase in the contamination discount. If the contamination discount dissipated, the coefficient on this variable would be positive; in contrast, we estimate it to be negative and significantly different from zero at almost the 95% level of confidence. The results provide no support for the hypothesis that the contamination discount dissipated through time-at least along a steady trend-and only weak evidence that it grew.

Enlarge 200% Enlarge 400%

While Model 2a restricts the contamination discount to follow a linear time trend, in Model 2b we use the formulation suggested by Shiller (1993) to generalize this time-varying effect by estimating a separate price index for the effect of contamination. The empirical estimates are shown in Table 6. F tests indicate that we can reject the hypotheses that all annual price changes are zero and that all annual price changes are identical for both the overall price index (F ^sub 18, 13577^ = 148 and F ^sub 17, 13577^ = 157) and the contamination price index (F ^sub 16, 3577^ = 12.9 and F ^sub 15, 13577^ = 11.8).

The pattern of the contamination coefficients, however, tells a very different story than was suggested by the more restrictive models: they suggest that the effect of contamination was not capitalized into the prices of properties in the contaminated area until about 1986, well after the contamination became public knowledge. During 1986-1996, however, the prices of properties in the contaminated area dropped below those in unaffected areas by up to 13.55% (in 1991) before recovering in the last two years of the study period. It is interesting to speculate why the market response to the contamination event should apparently have been delayed by several years. One possibility is that information about the contamination may have disseminated relatively slowly to buyers-or perhaps even to sellers, although the news coverage of this event makes it difficult to credit that explanation.

Enlarge 200% Enlarge 400%

Another possibility is that the relatively large transaction costs associated with moving made it difficult for owners to respond quickly. If transaction costs are large, then the earliest transactors following the announcement of the contamination may not have been those owners who were most sensitive to environmental problems, but rather those owners who were already close to overcoming the transaction costs before the announcement was made; the maximum discount, then, would occur only at a later time when the most environmentally sensitive owners overcame their transaction costs to become environmentally sensitive sellers.

It is also interesting that the most severe discount (in percentage terms) on properties in contaminated areas occurred in 1991, coinciding with the trough of the overall condominium market in the Scottsdale/Tempe area. As our data set covers only one market cycle we cannot use it to assess whether environmental contamination generally depresses the value of affected properties most severely during market downturns, but it is a possibility worth exploring in other contexts.

The third substantial step in our estimation experiment is to introduce varying location-price gradients as well as the contamination discount. The results are reported in Table 7. The adjusted R^sup 2^ is 30.2%, only slightly higher than the traditional repeat-sale and the simpler hybrid model, and F tests again indicate that we can reject the hypotheses that all annual price changes are zero (F ^sub 18, 13590^ = 144) and that all annual price changes are identical (F ^sub 17,13590^ = 151). Once again we observe the profound dip in condo prices. The contamination discount is estimated at 4.88%, slightly higher than in Model 2 and significant at a high level of confidence.

The results for the time trend in both of the location-price gradients are positive and significant, suggesting that relative prices are rising in the far north and the far south of the area from which the data are drawn, with the gradient for locations north of the Salt River growing more rapidly than the one for locations south of the Salt River.

The value of including hedonic variables is even more dramatic in Model 3 than in Model 2: all three hedonic variables are significant. As suggested by Case and Quigley (1991), we should expect that the coefficients on location variables are likely not to be stationary, so it may generally be important to include hedonic characteristics with a functional form enabling implicit prices as well as property attributes to vary.

Enlarge 200% Enlarge 400%

Model 3a incorporates a straight-line trend into the hybrid model to investigate whether the effect of contamination grows or diminishes over time. F tests again indicate that we can reject the hypotheses that all annual price changes are zero (F ^sub 18, 13589^= 136) and that all annual price changes are identical (F ^sub 17, 3589^ = 143). Table 8 shows that the coefficient on the contamination term is somewhat smaller in Model 3a (2.05%) than in Model 2a (2.43%) and not significantly different from zero, but the coefficient on the contamination/time interaction term is also negative and significant at the 98% level of confidence. This can be interpreted as evidence that there was very little contamination discount immediately after the contamination became public knowledge, but that the discount grew over time and amounted to about 6.5% by 1998.

Finally, in Model 3b we estimate a price index for the effect of contamination in the manner suggested by Shiller (1993), relaxing the constraint that the effect of contamination grow or diminish at a constant rate.2 The results are shown in Table 9. F tests indicate that we can reject the hypotheses that all annual price changes are zero and that all annual price changes are identical for all four price indices: the overall price index (F ^sub 18, 13541^ = 48.8 and F ^sub 17,13541^ = 49.8), the contamination price index (F ^sub 10,13541^ = 4.63 and F ^sub 15,13541^ = 4.94), the gradient for distance north of the Salt River (F ^sub 18, 13541^ = 22.9 and F ^sub 17, 13541^ = 21.2), and the gradient for distance south of the Salt River (Fi8-B54] = 18.8 and F ^sub 17, 13541^ = 15.9).

Enlarge 200% Enlarge 400%

As we saw in Model 2b, the results for Model 3b suggest that the effect of contamination was not capitalized into property values immediately after the information became public, but rather several years later in 1990. The contamination discount peaked in 1991 at 6.66%, and by 1995 it had disappeared. (Indeed, the results suggest that in 1998 properties located in the contaminated area were actually selling at a premium relative to properties located in otherwise identical but unaffected areas.)

Enlarge 200% Enlarge 400%

Figure 1 shows the price indices estimated for contaminated and unaffected areas using versions of each of the three models, while Figure 2 shows the estimated difference in prices between contaminated and unaffected areas. The general pattern of the results is quite striking: the results suggest that properties in the contaminated area suffered more severely than did properties in unaffected areas during the trough of the market, from 1989 to 1994. According to Model 3b, for example, there was essentially no difference in prices between contaminated and unaffected areas during 1982 and 1983, the first years after the contamination became public knowledge. Aside from an odd uptick estimated for 1984 in contaminated areas relative to unaffected areas, over the next eight years the prices of properties in the unaffected areas declined by slightly less than 30% while those in the contaminated area sank by more than 34%, meaning that properties in the contaminated area lost about 16% more of their value than did properties in unaffected areas.

Enlarge 200% Enlarge 400%

Both Models 2b and 3b suggest that the effect of contamination was not capitalized into market values immediately after the contamination became public knowledge; rather, a contamination discount does not appear in transaction price data until 1985 (Model 2b) or even 1987 (Model 3b). Moreover, both models suggest that the contamination discount had been eliminated by 1998 (Model 2b) if not earlier (Model 3b). We believe this suggests the value of using a flexible specification of the type proposed by Shiller (1993) rather than a restrictive specification of the types embodied in Models 2, 2a, 3 and 3a. It also raises the question why there should be such a delay in realizing the price discount associated with such a widely recognized disamenity as groundwater contamination. We speculate that the answer to that question lies in the protracted spread of information to buyers (if not sellers as well), the large transaction costs associated with moving residence, or both.

Conclusions

Every empirical model that we estimated suggested that condominiums in the contaminated area transacted at a discount that was both statistically and economically significant during at least part of the study period. When the contamination discount was restricted to remain constant over the study period (Model 2) it was estimated at 4.65%. When the contamination discount was allowed to vary over time it grew steadily from 2.43% to 5.91% (Model 2a) or varied as high as 13.55% in 1991 (Model 2b). When the model was expanded to take into account changes in the price gradient associated with locations further north or south of the Salt River (Models 3, 3a and 3b), the contamination discount declined slightly in magnitude but remained significantly different from zero.

Perhaps the most interesting finding is that the effect of groundwater contamination does not appear to have been capitalized into condominium transaction prices until several years after the contamination became public knowledge. It is also interesting that the contamination discount was estimated to be growing rather than dissipating through the study period, according to the restrictive straight-line specification of Models 2a and 3a. When this was generalized to permit the estimation of a separate price index for the contamination discount (Models 2b and 3b), however, the results suggested that the contamination discount operated primarily during the trough of the property market cycle (1990-1993) and had disappeared by the end of the study period.

The spatial gradients appear to have been evolving during the study period. Ordinarily, a gradient is the proportionate rate of decline in price as distance increases away from the downtown. In this article, the gradients are the proportionate rate of increase in price as distance from the Salt River increases. The Salt River cuts from east to west across the approximate middle of the area from which data are drawn. The price gradient north of the Salt River grows by an annual amount that is slightly more than the growth of the gradient south of the Salt River.

The methodological results are nearly as interesting as the environmental or economic results. We have demonstrated the value of incorporating hedonic characteristics such as location into repeat-sale models. It may be generally important to do this because the parameters on location variables can change profoundly over time. We found that including gradient variables caused the path of the price index to change dramatically. The distinctions allowed by this approach reveal relative changes in condo prices within the region.

[Reference]

References

Aitken, A.C. 1935. On Least Squares and Linear Combinations of Observations. Proceedings of the Royal Society of Edinburgh 55: 42-48.

Bailey, M.J., R.F. Muth and H.O. Nourse. 1963. A Regression Method for Real Estate Price Index Construction. Journal of the American Statistical Association 58: 933942.

Boyle, M.A. and K.A. Kiel. 2001. A Survey of Hedonic Studies of the Impact of Environmental Externalities. Journal of Real Estate Literature 9: 117-144.

Cannaday, R.E., HJ. Munneke and T.T. Yang. 2002. A Repeat Sales Index for Durable Assets: Disentangling the Price Effects of Age and Time. Working Paper. University of Georgia: Athens, GA.

Case, B., H.O. Pollakowski and S.M. Wachter. 1991. On Choosing Among House Price Index Methodologies. Journal of the American Real Estate and Urban Economics Association 19: 286-307.

Case, B. and J.M. Quigley. 1991. The Dynamics of Real Estate Prices. Review of Economics and Statistics 83: 50-58.

Case, K. and R. Shiller. 1989. The Efficiency of the Market for Single Family Homes. American Economic Review 79: 45-56.

Chalmers, J.A. and T.O. Jackson. 1996. Risk Factors in the Appraisal of Contaminated Property. The Appraisal Journal 64: 44-59.

Clapp, J.M. and C. Giacotto. 1992. Estimating Price Indices for Residential Property: A Comparison of Repeat Sales and Assessed Value Methods. Journal of the American Statistical Association 87: 300-306.

______. 1999. Revisions in Repeat-Sales Price Indexes: Here Today, Gone Tomorrow? Real Estate Economics 27: 79-104.

Clapp, J.M., C. Giacotto and D. Tirtiroglu. 1991. Housing Price Indices Based on All Transactions Compared to Repeat Subsamples. Journal of the American Real Estate and Urban Economics Association 19: 270-285.

Colwell, RF. 1990. Power Lines and Land Values. Journal of Real Estate Research 5: 117-127.

Colwell, RR, C. Dehring and N.A. Lash. 2000. The Effect of Group Homes on Neighborhood Property Values. Land Economics 76: 615-636.

Costello, G. 2002. Pricing Size Effects in Housing Markets. Journal of Property Research 17: 203-220.

Crone, T.M. and R.P. Voith. 1992. Estimating House Price Appreciation: A Comparison of Methods. Journal of Housing Economics 2: 324-338.

Dale, L., J.C. Murdoch, M.A. Thayer and RA. Waddell. 1999. Do Property Values Rebound from Environmental Stigmas? Evidence from Dallas. Land Economics 75: 311-315.

Englund, P., J.M. Quigley and C.L. Redfearn. 1999. The Choice of Methodology for Computing Housing Price Indexes: Comparisons of Temporal Aggregation and Sample Definition. The Journal of Real Estate Finance and Economics 19: 91112.

Gao, A.H. and G.H.K. Wang. 2004. Multi Transactions Model for Constructing Housing Market Index. Paper from the 2005 Fannie Mae Association Annual Meeting. Available at http://www.fma.org/Chicago/Papers/q041230.000MultiTransactionsModelAll.Zpdf.

Gatzlaff, D.H. and D.C. Ling. 1994. Measuring Changes in Local House Prices: An Empirical Investigation of Alternative Methodologies. Journal of Urban Economics 35: 221-244.

Goetzmann, W.N. and M. Spiegel. 1995. Non-Temporal Components of Residential Real Estate Appreciation. Review of Economics and Statistics 77: 199-206.

______. 1997. A Spatial Model of Housing Returns and Neighbourhood Substitutability. The Journal of Real Estate Finance and Economics 14: 11-31.

Graves, P., J.C. Murdoch, M.A. Thayer and D. Waldman. 1988. The Robustness of Hedonic Price Estimation: Urban Air Quality. Land Economics 64: 220233.

Hill, R.C., J.R. Knight and C.F. Sirmans. 1997. Estimating Capital Asset Prices. Review of Economics and Statistics 79: 226-233.

Kite, D., W. Chern, F. Hitzhusen and A. Randall. 2001. Property Value Impacts of an Environmental Disamenity: The Case of Landfills. The Journal of Real Estate Finance and Economics 22: 185-202.

Jackson, T.O. 1997. Investing in Contaminated Real Estate. Real Estate Review 26: 38-43.

Jackson, T.O. 2001. The Effects of Environmental Contamination on Real Estate: A Literature Review. Journal of Real Estate Literature 9: 93-116.

Kiel, K.A. 1995. Measuring the Impact of the Discovery and Cleaning of Hazardous Waste Sites on House Values. Land Economics 71: 428-435.

Kiel, K. A. and K.T. McClain. 1996. House Price Recovery and Stigma After a Failed Siting. Applied Economics 28: 1351-1358.

Kinnard, W.N. and E.M. Worzala. 1999. How North American Appraisers Value Contaminated Property and Associated Stigma. Appraisal Journal 67: 269-279.

Kohlhase, J.E. 1991. The Impact of Toxic Waste Sites on Housing Values. Journal of Urban Economics 30: 1-26.

Leggett, CG. and N.B. Boekstael. 2000. Evidence of the Effects of Water Quality on Residential Land Prices. Journal of Environmental Economics and Management 39: 121-144.

Mark, J.H. and M.A. Goldberg. 1984. Alternative House Price Indices: An Evaluation. Journal of the American Real Estate and Urban Economics Association 12: 30-49.

McClelland, G.H., D.H. Shulze and B. Hurd. 1990. The Effects of Risk Beliefs on Property Values: A Case Study of a Hazardous Waste Site. Risk Analysis 10: 485-497.

Meese, R.A. and N.E. Wallace. 1997. The Construction of Residential Housing Price Indices: A Comparison of Repeat-Sales, Hedonic Regression and Hybrid Approaches. The Journal of Real Estate Finance and Economics 14: 51-73.

Mieszkowski, P. and A.M. Saper. 1978. An Estimate of the Effects of Airport Noise on Property Values. Journal of Urban Economics 5: 425-440.

Palmquist, R.B. 1982. Measuring Environmental Effects on Property Values Without Hedonic Regressions. Journal of Urban Economics 11: 333-347.

Pennington, G., N. Topham and R. Ward. 1990. Aircraft Noise and Residential Property Values Adjacent to Manchester Airport. Journal of Transport Economics and Policy 24: 49-59.

Quigley, J.M. 1995. A Simple Hybrid Model for Estimating Real-Estate Price Indexes. Journal of Housing Economics 4: 1-12.

Ridker, R.G. and J.A. Henning. 1968. The Determination of Residential Property Value with Special Reference to Air Pollution. Review of Economics and Statistics 49: 246-257.

Riechert, A. 1999. The Persistence of Contamination Effects: A Superfund Site Revisited. Appraisal Journal 67: 126-135.

Segerson, K. 2001. Real Estate and the Environment: An Introduction. The Journal of Real Estate Finance and Economics 22: 135-139.

Shiller, R.J. 1993. Measuring Asset Values for Cash Settlement in Derivative Markets: Hedonic Repeated Measures Indices and Perpetual Futures. The Journal of Finance 48: 911-931.

Steele, M. and R. Goy. 1997. Short Holds, the Distribution of First and Second Sales and Bias in the Repeat-Sales Price Index. The Journal of Real Estate Finance and Economics 14: 133-154.

Thayer, M., H. Albers and M. Rahmatian. 1992. The Benefits of Reducing Exposure to Waste Disposal Sites: A Hedonic Housing Value Approach. Journal of Real Estate Research 7: 265-282.

Weber, B.R. 1997. The Valuation of Contaminated Land. Journal of Real Estate Research 14: 379-398.

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%

Enlarge 200% Enlarge 400%