The Journal of Real Estate Finance and Economics

, Volume 46, Issue 3, pp 543–563

Historic Preservation: Preserving Value?

Authors

    • School of BusinessClarkson University
  • Jason A. Altieri
    • School of BusinessClarkson University
Article

DOI: 10.1007/s11146-011-9338-8

Cite this article as:
Heintzelman, M.D. & Altieri, J.A. J Real Estate Finan Econ (2013) 46: 543. doi:10.1007/s11146-011-9338-8

Abstract

The creation of historic districts has become a common way to preserve historic buildings and neighborhoods. Advocates of historic districts assume that such districts augment, or at least, protect property values for homes within these districts. The existing economic literature supports this conclusion, but most studies seem to fall victim to an endogeneity bias since higher value homes are, all else equal, more likely to be included in districts. This study uses repeat-sales fixed effects (difference-in-differences) analysis to look at homes before and after the creation of districts in the Boston-Cambridge-Quincy MSA between 2000 and 2007, and thus control for this endogeneity bias. Secondarily, we re-examine the effects of a Massachusetts preservation policy, the Community Preservation Act (CPA) which, in part, supports historic preservation. We find evidence that the creation of a local historic district, on average, reduces home prices for homes in that district between 11.6 and 15.5%. This indicates that any restrictions implied by the creation of a district outweigh any benefits to homeowners within the district. If, instead, census block fixed effects are employed, the analysis shows a statistically insignificant impact, the sign and magnitude of which depends on the specification. Taken together with the repeat sales result, this confirms our intuition about the importance of controlling for omitted variables and endogeneity biases. Finally, we find evidence that the CPA also lowers property values, by less than 1%, and that being in a Historic District magnifies the negative effect of the CPA.

Keywords

Historic preservationHedonic analysisRepeat salesRegulation

Introduction

The creation of historic districts has become a common way to preserve our collective built heritage. According to the National Register of Historic Places, there are some 2,300 ‘local’ historic districts, created and administered by local governments and administrators, in the United States. The National Trust for Historic Preservation advertises such local measures as accomplishing five things:1
  1. 1.

    Provide a municipal policy for the protection of historic properties

     
  2. 2.

    Establish an objective and democratic process for designating historic properties

     
  3. 3.

    Protect the integrity of designated historic properties within a design review requirement

     
  4. 4.

    Authorize design guidelines for new development within historic districts to ensure that it is not destructive to the area’s historic character

     
  5. 5.

    Stabilize declining neighborhoods and protect and enhance property values.

     
This study looks to test the last of these advertised accomplishments of historic districts.

In general, historic designation and subsequent preservation does provide a public good—the protection of properties or buildings of historic significance or that have a unique character or architecture that the public wishes to see preserved.2 However, as with most public goods, this provision comes at a cost. In this case, owners of designated properties face restrictions on new development and on renovations that can be made to existing structures. Homes near districts, but not in districts, stand to benefit by receiving the augmented public good. Homes in regulated districts, however, may be affected negatively or positively by designation depending on the relative weight of the increased public good provision versus the implicit cost of restrictions on development and renovation.

This paper studies these effects using extensive home transaction data in the Boston-Cambridge-Quincy MSA from 2000–2007. Because of the quasi-panel nature of our dataset, we are able to overcome the endogeneity and omitted variables biases that are likely to affect many prior studies on the topic of historic preservation through the use of a repeat-sales fixed effects hedonic framework. We find that the creation of a local historic district, on average, reduces home prices for homes in that district, the internal effect, between 11.6 and 15.5%. This indicates that any restrictions implied by the creation of a district outweigh any benefits to homeowners in the district. We do not find any significant impacts of historic districts on nearby homes not included in districts. If, instead, census block fixed effects are employed, which does not properly control for endogeneity bias, the analysis provides evidence of a positive internal impact and some positive external effects for homes within 0.25 miles of a district.

As a secondary issue, we also examine the effects of a Massachusetts policy, the Community Preservation Act (CPA) which, in part, raises funds for local historic preservation. The CPA is a statewide policy that is implemented locally (or not implemented at all). Towns who opt-in assess a surcharge on property tax bills of up to 3%, and the funds raised are matched by the state and then must be spent on historic preservation, open-space preservation, or affordable housing. We find evidence that the CPA lowers property values, by less than 1%, and that being in a Historic District magnifies the negative effect of the CPA and vice versa. This may imply that the complementary nature of these policies, both of which restrict development, also complement each other in reducing property values.

Section “Background” provides background on the effects of government regulation on real estate markets, and discusses the existing literature. Section “Historic Districts in Massachusetts and the Community Preservation Act” provides information about the specific policies in Massachusetts. Section “Data and Methodology” discusses the details of our empirical approach and describes our dataset. Section “Results” discusses the results of our analysis and concludes.

Background

Theoretically, the market value of any property is determined by its current use as well as the value it would have in its privately optimal use, presuming that, for some cost, it could be converted to that use. That is, the value of a parcel will be maximized when the parcel is in its most valuable use. To the extent that regulation prevents this optimal use from being achieved, this will reduce parcel values. In a completely unregulated market, and in the absence of externalities, we would expect this optimal value to be achieved. However, given that there are externalities, there will be a divergence between private and public benefits. Possible externalities include benefits to others from a home remaining in a historic character, or costs from a a parcel being developed into a ‘locally undesirable land use’ (LULU). Regulation would tend to shrink this discrepancy by preventing negative externalities and promoting the provision of positive externalities and so reducing the value of regulated properties while increasing the value of nearby unregulated properties.

Restrictions on land-use development take many forms. Zoning restrictions help to solve externalities between neighbors, as well as to restrict the density of development. Other municipal codes dictate anything from the placement of electrical outlets in a room to the length of one’s grass to the color of one’s home. The effects of these regulations are not obvious as they often provide the promise of a more attractive neighborhood at the cost of the rights of homeowners. In fact, the effects of regulation may follow the model of Brueckner (1982) for public goods provision in that property values would at first increase with more regulation and then decrease as the negative aspects dominate the positive so that prices are maximized at the optimal level of regulation.

An extensive empirical literature exists on the effects of these sorts of regulation on homeowners and developers. For instance, considering building restrictions generally, Glaeser and Gyourko (2003) argue that regulation is a major reason for the gap between home prices and construction costs in those areas where such a gap exists. Obviously such a gap is good for existing homeowners and reflects the implicit restriction in supply caused by building restrictions. It is also bad for those looking to buy into an affected market. Glaeser and Ward (2009) draw similar conclusions looking specifically at the real estate market of Greater Boston. In a study of Florida cities, Ihlanfeldt (2007), finds the opposite result that regulation, while increasing the price of homes, decreases the price of vacant land. This suggests that regulation leads to increased scarcity of developed and developable parcels, and may be driving new development out of regulated areas (perhaps by design).

Historic districting is another type of regulation that provides a public good at the cost of restricting the activities of homeowners. In general, designation of homes as part of a historic district restricts the types of changes that can be made to those homes but often allows for subsidies for maintenance of those homes (Noonan, forthcoming).3 As a consequence of these restrictions, subsidies, and public good aspects, there are a large number of possible categories of price impacts. We can broadly classify these impacts into internal impacts on homes within districts and external impacts on homes nearby but not within districts.

Internal Effects

The price of a home is a function of its physical characteristics, including its effective age and condition, characteristics of its neighborhood, and any zoning or other applicable regulations that impact the options available to the homeowner. The designation of a home as part of a historic district has the potential to impact all of these categories. First, the designation of the home and other neighboring homes provides stability; knowledge that the neighborhood is unlikely to change in the future. In this way, the ‘character’ of the neighborhood is preserved, which would be a positive impact. In addition, designation may bestow a cachet on included homes that separates them from other similarly aged homes that remain undesignated. On the other hand, the price of a home also reflects options to the homeowner to re-sell or re-develop the property to maximize its value. Designation would restrict these options by preventing the homeowner from making changes to the property that would alter its historical qualities, including substantial additions and, of course, demolition.

In older neighborhoods, and particularly in affluent and growing areas, the explicit restriction against demolition of a home may be very important. Estimates of the rate at which homes are torn down for replacement by newer homes are hard to come by, but there is a popular perception that this rate is increasing. Dye and McMillen (2007) provide data for the Chicago area where between 1999–2003 as much as 9.4% of the housing stock was torn down in some neighborhoods. Data from the National Trust for Historical Preservation indicates that in 2008, some 500 communities in 40 states across the country were ‘affected’ by teardowns.4 This activity seems to be concentrated in urban and suburban areas in major metropolitan regions (McMillen 2006). Dye and McMillen (2007) and McMillen (2006) also suggest that parcels that are ripe for redevelopment, as we would expect, would be worth more without the existing home which, in turn, suggests that the inability to teardown the home would reduce those parcels’ values.

The conclusion then about internal price effects must be ambiguous. The net effect will clearly depend on the magnitudes of the various effects mentioned here, and the effect may vary depending, in particular, on the age of each included home.5 A newer home within a district may be more likely to receive a benefit from designation if it is less likely to be impacted by the restriction on redevelopment, but will still receive the benefits from the preservation of the neighborhood’s character and may benefit more from subsidies that allow the owners to update the interior of the home (Coulson and Lahr 2005).

External Effects

The external effects of historic districts are also ambiguous. To the extent that the districts provide an amenity that is enjoyed by local residents, we would expect prices of local homes to increase. This effect is also likely to be strongest for those homes closest to historic districts. This is a demand effect—the creation of the district increases demand for homes near the district (and perhaps decreases demand and prices for homes further away from districts). There are also supply effects. By fixing the level of development within the historic district, the creation of the district fixes the supply of existing housing units and presumably shifts demand for redevelopment to nearby neighborhoods that do not carry a historic designation. These effects will tend to increase prices. However, if the creation of the district shifts undesirable development out of the district this could also depress nearby property values. Also, if demand is increased for homes in a district, if overall housing demand in the area does not also increase, this may result in a decrease in demand and value in the area surrounding a district. Many of these effects would tend to occur with some lag—preventing future changes in the real estate market. Assuming some amount of foresight, however, implies that these future impacts would be capitalized into values, to some extent, immediately.

Existing Literature

The existing literature on the effects of historic designation and historic districts is mixed but mostly finds positive impacts, both internal and external, as summarized in Table 1.6 Most studies of this issue use hedonic analysis of property sales transactions to estimate the effects of historic districts on values, but two notable exceptions are Coulson and Leichenko (2010), Leichenko et al. (2001) and Coulson and Lahr (2005) which use appraisal data instead. Appraisal data, which is often more readily available than sales data, while allowing for the inclusion of all properties in the study area, is somewhat unreliable because it involves the subjective judgements of appraisers rather than the actual, observed, market value. In other words, there is the possibility of significant measurement error when using appraisal data (Taylor 2003). In addition, appraisal procedures vary from community to community and so it is difficult to compare appraisals in a study that spans multiple administrative units.
Table 1

Selected previous studies on historic districts

Study

Location

Study type

Internal effects (% change)

External effects (% change)

Asabere et al. (1989)

Massahusetts

Sales prices

Insignificant

Asabere et al. (1994)

Philadelphia

Sales prices

−24

Coffin (1989)

Chicago Suburbs

Sales prices

6–7

Clark and Herrin (1997)

Sacramento

Sales prices

17.3

−20 for being near a historic district

Coulson and Leichenko (2010)

Abilene, Texas

Appraisal data

15–17.6

0.14 per nearby designated building

Leichenko et al. (2001)

Texas

Appraisal data

5–20

Coulson and Lahr (2005)

Memphis

Appraisal data

14–23

Noonan (2007)

Chicago

Sales prices

3–11

2 per nearby landmark

Ahlfeldt and Maennig (2010)

Berlin

Sales prices

−3–5

0–2.8 per nearby landmark

Noonan and Krupka (2011)

Chicago

Sales prices

−47 to +35

−2 per nearby landmark

Clark and Herrin (1997) and Coffin (1989) looked at the effects of historic preservation measures on property value in Sacramento and the Chicago suburbs respectively. Using various controls for neighborhood effects, these studies found, with varying degrees of significance, that preservation measures had a positive impact on property values of between 6 and 40%. Similarly, Leichenko et al. (2001) investigated the impact of a variety of different historic preservation measures on property values in nine Texas cities. They found a positive impact on property values of between 5 and 20%, and that the less restrictive the designation, the greater positive impact on property values. Early studies by Asabere et al. (1989) and Asabere et al. (1994) concentrating on small geographic areas found different results—that historic designation had insignificant or negative impacts on property values.

Focusing on attached homes such as townhouses or condominiums in the city of Chicago, and using a repeat sales fixed effects analysis to control for omitted variables and endogeneity biases, Noonan (2007) found a positive relationship, ranging from 3 to 11%, between historic preservation and home prices. More recently, Noonan and Krupka (2011), focusing again on attached homes in Chicago, come to a different conclusion. They run a simple OLS regression and find a premium for homes in historic districts, consistent with other studies. However, to control for potential endogeneity, they then employ an instrumental variables approach, and find a range of mostly negative price impacts depending on the specification on homes in designated historic districts. They also find small negative external effects. Ahlfeldt and Maennig (2010) use a cross-sectional spatial autoregressive approach to analyze the effects of historic landmarks and districts in Berlin. They concentrate on the external effects of these historic attributes and find significant positive impacts using a number of different measures of proximity, density, and size of designated historic landmarks and districts, as well as also finding small negative internal price effects.

Historic Districts in Massachusetts and the Community Preservation Act

Local historic districts, in Massachusetts, are allowed for by Massachusetts General Laws: Chapter40C, and can be established by towns or cities through a two-thirds majority vote at either a city council or town meeting. Prior to such a vote, a study committee must issue a “report on the historical and architectural significance of the buildings, structures or sites to be included in the proposed historic district.” This committee, which must include experts on historic preservation and is nominated by a town’s mayor and confirmed by the town meeting (or city council), in effect, draws the borders of the district to be then considered by the town meeting or city council. There is no set timeline for this process except that a public hearing must be held within 60 days or receipt of this report by the town meeting or city council. Chapter 40C sets out guidelines for what restrictions may be established when the district is created. In Section “Discussion”, the law states that “no building or structure within an historic district shall be constructed or altered in any way that affects exterior architectural features” unless an exception is issued by the local historical commission. Obviously, this will generally prevent the demolition of homes or structures within historic districts. Examples of alterations typically exempt from review, as provided by Section 8 of the law, however, are interior features, paint color, air conditioning units, storm doors and windows, and temporary buildings. Once a district has been established, a Local Historical Commission is appointed by the town or city to oversee the management of the historic districts. This committee reviews any proposed changes that are governed by the district ordinance and determines if the changes are acceptable within the historic framework of the neighborhood.7 Subsidies from the Massachusetts Historical Commission are available to aid homeowners in maintaining their historic properties. There are approximately 200 such local historic districts in the state of Massachusetts according to the Massachusetts Historical Commission.

The Massachusetts Community Preservation Act went into effect in 2001. This policy, which is entirely separate from the historic designation process described above, allows towns to opt-in to the program through a referendum, and if it passes, enact property tax surcharges of up to 3%. These funds are matched, at varying levels over time, by the state and then must be spent on historic preservation, open-space preservation, and affordable housing. Participating communities must spend at least 10% of all money raised on each of these areas, and may then allocate the rest as they see fit, but still on these priorities. As of December 2010, 147 communities have adopted the CPA out of the 351 towns and cities in Massachusetts, and these adoptions happen at different times in our study period. This heterogeneity in timing allows us to implement, in effect, a difference-in-differences estimation strategy taking advantage of both before/after and treatment/control variation in identifying the effect of the CPA. According to the Community Preservation Coalition, an organization that advocates for the CPA, “more than 1,300 appropriations for historic preservation projects have been approved under the program.”8 On average, towns have spent about 29.8% of available funds on Historic Preservation (Heintzelman 2010b). This spending takes many forms including installing signs marking historic district boundaries, purchasing historic properties, restoring historic buildings, and refurbishing historical monuments.9

As with historic districts, it is ambiguous what effect the CPA should have on property values because the policy implies both increased public goods provision and higher property taxes, and the model of Brueckner (1982) again comes into play. However, as Heintzelman (2010b) points out, while the tax-cost of the CPA is transparent and takes effect immediately, the expenditure side of the policy is highly uncertain so that consumers may see the additional taxes without seeing the benefits. There are also supply-side issues at play with the CPA since its preservation goals presumably restrict development and the supply of housing, while at the same time any additional affordable housing provisions will tend to increase supply of housing. Previous studies of the CPA have confirmed this ambiguity, and shown that the effects tend to differ from county to county (Heintzelman 2010a, b). On average though, this work has shown the CPA to have a depressive effect on property values of about 1.5%. Our current analysis uses a subset of the data used in Heintzelman (2010b), but adds substantial detail on historic districts that was not available at the time of that earlier study. It is possible that homes designated as historic may be impacted differently by the CPA than undesignated homes, although this effect is also theoretically ambiguous. On the one hand, the CPA may provide additional funds for historic preservation which may be spent in a designated district. However, the CPA may also provide resources for additional districts, and this additional land regulation could erode value for existing historic homes. Any differential impacts like these will be picked up through a term interacting designation with whether or not the community has implemented the CPA.

Data and Methodology

Our approach to measuring the effect of historic districting on property values is a repeat sales fixed-effects hedonic analysis. That is, we regress the observed transaction prices on a set of explanatory variables including property characteristics such as the size of the home and number of rooms, locational variables such as the distance to various amenities, the type of ‘zone’ in which the property is located and a set of variables measuring the presence of historic districts. In addition, we include a set of year and month dummies that capture dataset-wide, and seasonal price trends respectively. Sales prices are also normalized, or deflated, using the House Price Index from the Federal Housing Finance Agency. This normalization is undertaken at the MSA Metropolitan Division level, of which there are three within our study area (the Boston-Cambridge-Quincy MSA).

We use two measures of the effect of historic districts on home prices: the distance between each transacted property and the nearest historic district at the time of the sale and a dummy variable indicating whether or not each property is included within a historic district at the time of the sale. However, because of the expected discontinuity of price effects at the borders of districts, rather than including distance as a linear or otherwise continuous measure, we use a set of distance denominated dummy variables representing the distance to the nearest historic district.10 Because new districts were established during the study period, these measures have temporal variation that we can take advantage of in our analysis. We include dummy variables representing whether the nearest district is up to ten miles away.11

To establish these measures, we identified 118 local historic districts in the study area using the Massachusetts Cultural Resource Information System. This also told us when each district had been created. Unfortunately, the Massachusetts Historic Commission (MHC) does not maintain an accurate GIS database of historic districts so we mapped the districts using a combination of digital maps from local officials and physical maps from the MHC. These maps were then drawn “by hand” in ArcGIS. Using these methods we were able to locate 113 of the 118 identified districts. We then used STATA and ArcGIS to link each transaction to the districts that existed at the time of the sale. To be clear, the dummy variable, InDistrict, is 0 for all transactions of homes that are not in a local historic district and for tranasctions on homes that will, within our sample period, be part of a district but are not at the time of sale. It is 1 for all transactions of homes that are, at the time of sale, in a local historic district. In this sense, we are performing a difference-in-differences analysis. It is possible that new districts are anticipated by local residents ahead of their final approval date. If this is true, this effect would tend to bias our estimates of the effects of districts towards zero, which means that our estimates are conservative in the sense of underestimating the true effects of historic districts.12

In general, three empirical issues must concern analysts in performing hedonic analyses: omitted variables bias, endogeneity, and spatial issues (dependence, autocorrelation). Greenstone and Gayer (2009) establish that hedonic analyses are often severely impacted by omitted variables bias. Intuitively, home prices are very complex and subject to almost innumerable factors, many of which are unobservable to researchers. When one of these unobserved variables is correlated with both the dependent and at least one independent variable, then estimates of the coefficients on those independent variables will be biased. In our case, suppose that being in a historic district is correlated with an unobserved characteristic such as building quality that we would expect to impact home price positively. Then the estimated coefficient on being in a historic district will be biased upwards since we are capturing effects of building quality in that estimate.

Endogeneity bias occurs when the dependent variable is co-determined with one or more independent variables. In our case, the endogeneity is in the form of a selection bias because of the non-random selection of homes into historic districts. As discussed above, the borders of a proposed district are determined by a committee made up, in part, of experts on historic preservation and architecture, and it is unlikely that they explicitly consider value in drawing district boundaries. However, older homes that survive to be considered for historic preservation are likely to have been well-maintained over the years and will have a higher market value, and be more likely to be deemed historic than an older home whose character, perhaps, has been eroded through less maintenance or care. This correlation between ‘quality’ and being designated for preservation is the genesis of the endogeneity problem. Noonan and Krupka (2010, 2011) provide empirical evidence that higher value homes are indeed more likely to be included in historic districts. If this selection effect is not controlled for, we would be likely to find a positive correlation between sales price and being in a district even if there is no causative effect of designation on price.

Greenstone and Gayer (2009) recommend three approaches for overcoming omitted variables and endogeneity in hedonic analyses: instrumental variables, regression discontinuity approaches, and fixed effects analysis. We employ the last of these.13 Fixed effects analysis controls for omitted variables and endogeneity by including a large set of dummy variables for small groups of observations, allowing for the constant term in the regression to be group-specific. These dummy variables control for any factors or characteristics of properties within a group that help determine price and are constant over time and thus allow for unbiased estimates of any remaining variables that either vary amongst homes within the chosen scope of the fixed effects or over time. In this paper we will use two scales: the census block and for the sub-sample of homes that sell more than once during the sample period, individual homes.

Unfortunately, census block fixed effects alone will be unlikely to mitigate the endogeneity problem if, for instance, it is more likely that higher value homes would be picked to be a part of a historic district even within census blocks. Since historic districts need not conform to simple boundaries, it is possible that individual properties could be carved out of otherwise contiguous districts. This implies that one either needs to be able to predict which properties will be included based on exogenous factors through an instrumental variables approach (Noonan and Krupka 2011), or must control for this selection effect through property-specific fixed effects. Because of the size and quasi-panel nature of our dataset, we observe a large number of individual properties that sell more than once during the study period. This allows us to employ property-level fixed effects.

The last class of empirical issues we need to address are spatial in nature (LeSage and Pace 2009) Spatial dependence is when the values of nearby or neighboring homes are co-determined; the value of my home depends on the characteristics of my home and the value of my neighbor’s. Spatial autocorrelation, as the name implies, is when the error terms for different observations, rather than being independently identically distributed are, in fact, correlated for spatially close observations. The most complete way to solve these issues is to employ a fully-general spatial econometric approach using a spatial weights matrix to dictate the strength of relationship between any two observations in the dataset. Unfortunately, for a dataset of our size, this is not feasible. However, we can perform a simplified spatial approach using fixed effects and error-clustering at various geographic levels. These tools essentially employ a simplified version of the spatial weighting matrix where we allow for spatial dependence at the scale of the fixed effects; observations within the same block, or property, are assumed to be spatially dependent, but independence is assumed for any observations not in the same respective area. Similarly, we allow error terms to be correlated for observations within blocks, or properties, but assume zero correlation of the error terms for observations in different areas. This clustering also implicitly adjusts the standard error estimates for heteroskedasticity.

Our first, census block fixed effects, regression then takes the form:
$$ \ln{p_{ijt}}=\lambda_t+\alpha_j+z_{jt}\beta+x_{ijt}\delta_{jt}+\eta_{jt}+\varepsilon_{ijt} $$
(1)
where pijt represents the price of property i in group j at time t; λt represents the set of time dummy variables; αj represents the group fixed effects; zjt represents the treatment variables—whether or not a home is in an historic district at the time of sale and whether or not the town in which a home is located has enacted the CPA at the time of sale; xijt represents the set of other explanatory variables; and ηjt and εijt represent group and individual-level error terms respectively. This specification is basically the same as that employed in Heintzelman (2010a, b) for a different purpose and follows from Bertrand et al. (2004) and Parmeter and Pope (2011).
The repeat sales regression is similar and takes the form:
$$ \ln{p_{it}}=\lambda_t+\alpha_i+z_{it}\beta+\varepsilon_{it} $$
(2)
where λt represents year dummies, αi represents parcel fixed effects, zit represents time varying parcel level characteristics including the age of the home and the measures of proximity to historic districts, and εit is the error term. This approach is similar to a first-difference estimation in that, for each parcel that sells more than once during the sample period we are basing our estimates on changes in price as they relate to changes in these time varying characteristics.14

Data

Our sample consists of all single-family home sales from 2000–2007 in the Boston-Cambridge-Quincy Metropolitan Statistical Area, as defined by the U.S. Census Bureau, except for the small portion of the MSA located in New Hampshire. This amounts to a total of 311,635 observations. In Massachusetts, the Boston-Cambridge-Quincy MSA includes Suffolk, Middlesex, Essex, Norfolk, and Plymouth counties. In addition to sales price and date of sale, our dataset includes home characteristics such as number of bedrooms, number of bathrooms, lot size, and building area.15 We measure other geographic characteristics such as distance to the nearest highway, zoning categories, and others using ArcGIS. Table 2 contains summary statistics for our dataset by whether or not homes are within historic districts at the time of sale.
Table 2

Selected summary statistics

Variable

Never in a district

Always in a district

Status changes

Mean

Std. Dev.

Mean

Std. Dev.

Mean

Std. Dev.

Sale price ($)

392,102.40

832,556.60

585,173.00

586,483.90

664,871.30

929,974.80

Lot size (sq. ft.)

17,859.12

84,405.86

2,552.06

17,508.75

3,858.55

9,336.79

Building living area (sq. ft.)

1,638.03

2,674.19

1,207.47

1,802.02

1,555.00

1,219.55

Bedrooms

2.72

1.19

1.83

1.09

2.40

1.77

Bathrooms

1.54

0.75

1.39

0.71

1.60

0.92

Half bathrooms

0.49

0.54

0.26

0.48

0.51

0.66

Distance highway exit (miles)

2.45

1.76

0.96

1.07

0.96

0.51

Distance highway (miles)

0.26

0.52

0.08

0.42

0.05

0.08

Distance train station (miles)

2.24

1.99

0.85

0.97

1.16

0.95

Distance train line (miles)

0.82

1.02

0.21

0.23

0.36

0.49

Age (years)

50.12

39.75

93.17

42.60

72.11

38.56

Distance historic district (miles)

3.27

2.88

1.62

1.00

The summary statistics show that homes within historic districts are generally older, smaller, in more dense neighborhoods, and more expensive. In addition they are generally further away from exits, train stations, and train lines. These numbers make sense considering that homes within districts are designated for historic qualities, suggesting that they will, in general, be older homes with the characteristics of older neighborhoods. The fact that homes in Districts are older and more expensive could result from factors related to the historic nature of the housing, or it could be a result of what Noonan and Krupka (2011) describe as “picking winners” where regulators choose superior housing to include in the districts, while leaving less valuable properties out. Note that those properties which have more recently changed status are younger than other designated properties and, in general, more similar to undesignated homes.

It is worth noting that historic districts are relatively sparse geographically in our dataset. There are 75,780 census blocks in our dataset and 113 historic districts. Of these census blocks, only 1,804, or 2.3%, contain or are a part of one or more historic districts. Only 83 blocks contain or are part of 2 or more districts. The average historic district in our dataset contains all or part of 16.75 census blocks although this number is somewhat skewed by two large districts in Boston itself which contain 241 and 161 census blocks respectively.16

Results

We begin by using the full sample of homes and employing census block level fixed effects analysis as described by Eq. 1 above.17 Table 3 reports the results of these regressions. At this scale, the effect of being in a historic district is positive and significant when the interaction term between age and InDistrict is left out. This effect switches sign and becomes insignificant when this interaction term is added. The interaction term, however, is positive and significant which is still indicative of a positive internal effect of historic districts.18 The estimate of the external district effects, as measured by the series of distance-related dummy variables indicates positive but insignificant impacts, except at the quarter-mile level where the effect is mostly significant. We experimented quite a bit with the specification of the proximity variables. Using a simple linear specification was suggestive of a positive proximity effect. Using natural log, quadratic, or square root specifications suggested a negative proximity effect. However, these specifications, since they included zero values for homes in districts, would be biased downwards if, as we suspect, the internal effects of districts are, in fact, negative. We also experimented with including the size of the nearest district, and the distance of each home to downtown Boston. These variables were not significant, and so have been omitted.
Table 3

Full sample results

 

Model 1

Model 2

Model 3

Model 4

coef

p-value

coef

p-value

coef

p-value

coef

p-value

InDistrict

0.187**

0.022

−0.026

0.636

−0.023

0.666

−0.011

0.858

Nearest district w/in 0.25 miles

0.067

0.114

0.070*

0.098

0.075*

0.058

0.076*

0.053

Nearest district w/in 0.5 miles

0.049

0.214

0.052

0.188

0.057

0.128

0.057

0.121

Nearest district w/in 1 mile

0.050

0.188

0.051

0.184

0.055

0.122

0.055

0.122

Nearest district w/in 2 miles

0.048

0.204

0.048

0.203

0.052

0.139

0.052

0.139

Nearest district w/in 3 miles

0.038

0.238

0.038

0.242

0.040

0.178

0.040

0.178

Nearest district w/in 4 miles

0.024

0.345

0.023

0.353

0.024

0.308

0.024

0.307

Nearest district w/in 5 miles

0.025

0.299

0.025

0.306

0.024

0.291

0.024

0.290

Nearest district w/in 10 miles

0.019

0.257

0.019

0.269

0.019

0.270

0.019

0.270

Ln(Age of home)

−0.035***

0.000

−0.037***

0.000

−0.038***

0.000

−0.038***

0.000

Ln(Age)*InDistrict

0.052***

0.000

0.052***

0.000

0.053***

0.000

CPA

0.060**

0.015

0.061**

0.014

CPA*InDistrict

−0.059

0.515

Lot size (sq. ft.)

0.000***

0.000

0.000***

0.000

0.000***

0.000

0.000***

0.000

Lot size squared

−0.000***

0.000

−0.000***

0.000

−0.000***

0.000

−0.000***

0.000

Ln(Building Living Area)

0.567***

0.000

0.567***

0.000

0.568***

0.000

0.568***

0.000

Bedrooms

0.006

0.297

0.006

0.287

0.005

0.326

0.005

0.328

Bathrooms

0.086***

0.000

0.085***

0.000

0.084***

0.000

0.084***

0.000

Half bathrooms

0.066***

0.000

0.065***

0.000

0.065***

0.000

0.065***

0.000

Distance highway exit (miles)

0.002

0.875

0.002

0.839

0.001

0.890

0.001

0.884

Distance highway (miles)

0.007***

0.000

0.007***

0.001

0.007***

0.001

0.007***

0.001

Distance train station (miles)

−0.020**

0.011

−0.021***

0.008

−0.020**

0.012

−0.019**

0.013

Distance train line (miles)

0.001

0.856

0.001

0.839

0.001

0.843

0.001

0.846

Condominium

−0.150***

0.000

−0.152***

0.000

−0.152***

0.000

−0.153***

0.000

CapeCod

−0.002

0.788

−0.002

0.769

−0.003

0.679

−0.003

0.676

Ranch

0.040***

0.000

0.040***

0.000

0.040***

0.000

0.040***

0.000

Townhouse

−0.009

0.712

−0.010

0.700

−0.014

0.596

−0.014

0.588

Colonial

0.001

0.937

−0.000

0.986

−0.001

0.932

−0.001

0.930

Contemporary

0.034**

0.023

0.032**

0.032

0.031**

0.036

0.031**

0.037

Apartment

−0.067**

0.013

−0.067**

0.013

−0.069***

0.009

−0.069***

0.009

Constant

7.841***

0.000

7.856***

0.000

7.855***

0.000

7.857***

0.000

Zoning dummies

Yes

Yes

Yes

Yes

Month dummies

Yes

Yes

Yes

Yes

Year dummies

Yes

Yes

Yes

Yes

Fixed effects level

Census block

Census block

Census block

Census block

Clustering level

Census block

Census block

Census block

Census block

Number of observations

309,680

309,680

309,680

309,680

Adjusted R-squared

0.679

0.679

0.681

0.681

***p < 0.01, **p < 0.05, *p < 0.1

We see that age negatively impacts a homes price, while additional living area, and the number of half and full baths, all positively impact sales price. The lot size has a positive but decreasing marginal impact on price. We use a quadratic here, rather than a natural log because of the number of condominiums in our sample which have a lot size of zero. A home’s distance to the nearest highway has a positive and significant effect on price, although there is no effect of being close to highway exits. Similarly, homeowners pay a premium to be closer to commuter rail stations, but prefer to be further away from active rail lines, although this last effect is not significant. The style of the home also matters.19

In much of the literature, in a semi-logarithmic specification, it is assumed that the coefficient on a dummy variable represents the percent change on the dependent variable from the effect represented by the dummy variable. However, as Halvorsen and Palmquist (1980) show, this is only approximately correct, and the bias of this assumption increases with the absolute value of the coefficient. They show that the relative effect of the dummy variable is, in fact, g = exp(c) − 1 where g is the percent change and c is the estimated coefficient. Applying this to the InDistrict variable in these regressions, we find that being in a district increases the price by 20.6%. Being within 0.25 miles increases price by between 7.25 and 7.9% in models 2–4. The CPA is estimated to have a positive effect of about 6.2% on average.

As discussed above, however, the full sample results remain open to concerns of omitted variables and, in particular, endogeneity bias. To control for this, we now present results of an analysis where we focus on homes that sell more than once during our study period and use property-level fixed effects to control as completely as possible for these biases, as in Eq. 2. Table 4 presents these results.20
Table 4

Repeat sales results

 

Model 5

Model 6

Model 7

Model 8

coef

p-value

coef

p-value

coef

p-value

coef

p-value

InDistrict

−0.123*

0.091

−0.168*

0.061

−0.160*

0.074

−0.136

0.142

Nearest district w/in 0.25 miles

−0.038

0.178

−0.037

0.184

−0.037

0.183

−0.036

0.195

Nearest district w/in 0.5 miles

−0.019

0.511

−0.019

0.519

−0.018

0.530

−0.018

0.533

Nearest district w/in 1 mile

0.001

0.984

0.001

0.973

0.002

0.930

0.002

0.930

Nearest district w/in 2 miles

0.007

0.796

0.007

0.787

0.009

0.724

0.009

0.732

Nearest district w/in 3 miles

0.011

0.686

0.011

0.683

0.011

0.676

0.010

0.706

Nearest district w/in 4 miles

0.008

0.755

0.008

0.751

0.009

0.740

0.008

0.760

Nearest district w/in 5 miles

−0.001

0.978

−0.001

0.982

0.002

0.952

0.001

0.973

Nearest district w/in 10 miles

0.008

0.751

0.008

0.747

0.008

0.734

0.008

0.738

Ln(Age of home)

−0.049***

0.000

−0.050***

0.000

−0.049***

0.000

−0.050***

0.000

Ln(Age)*InDistrict

0.011

0.377

0.009

0.467

0.012

0.327

CPA

−0.009***

0.001

−0.008***

0.006

CPA*InDistrict

−0.068***

0.000

Constant

11.995***

0.000

11.996***

0.000

11.994***

0.000

11.993***

0.000

Month dummies

Yes

Yes

Yes

Yes

Year dummies

Yes

Yes

Yes

Yes

Clustering level

Property

Property

Property

Property

Number of observations

125,719

125,719

125,719

125,719

Adjusted R-squared (within)

0.023

0.023

0.023

0.023

***p < 0.01, **p < 0.05, *p < 0.1

In contrast to the full sample results, we now see that when a home becomes designated as part of a historic district its value declines by between 11.6 and 15.5%. In addition, there are now no significant external effects. Age still has a negative impact on value with a slightly higher magnitude as in the block-level fixed effects analysis, although the interaction term, when included, is not significant. The age variable is now measuring the effect of the number of years between sales rather than the absolute age of the home since it necessarily increases by the number of years between sales for each observed parcel. In these models, the CPA has a small negative impact on values of less than 1%, and the interaction between the CPA and InDistrict indicates that these effects magnify each other; homes in districts are more negatively impacted by the CPA than homes not in districts and vice versa.21 It is not surprising that the historic district and CPA coefficients, and those on the interaction terms, change sign in the implementation of the repeat sales regressions. The repeat sales regression does the best job of handling omitted variables bias which will isolate the impacts of factors which change over time, as these variables do. In addition, if the historic designation process is endogenous, these effects will now be mitigated. The particular direction of the change for the InDistrict estimate supports the proposition that the process is endogenous. If higher value or higher quality homes are more likely to be included in historic districts, this will bias naive estimates upwards. Controlling for this endogeneity, as we do with the repeat sales analysis will correct this bias. This is what appears to happen in our results.

Discussion

Our analysis has a number of important implications. First, in the property-level fixed effects model which most completely controls for endogeneity and omitted variables biases we find significant negative internal impacts of historic designation and no significant external effects. This is despite the fact that, comparing means, homes in districts sell at a premium of nearly 60% over homes not in historic districts despite being smaller and older. This is clearly suggestive that while homes with historic characteristics are attractive and sell for a premium, the restrictions that come with designation erode some of these benefits. The lack of any significant external effects is difficult to interpret, but suggests that spillover benefits of preservation may be small. It is also possible that, following the model of Brueckner (1982), the level of historic preservation in our study area is approximately optimal so that additional small changes to this level will have no effect on observed prices.

This work also confirms, using some of the same data, analyses that find negative impacts of the Massachusetts Community Preservation Act and that homes in historic districts appear to be harmed disproportionately by the CPA. Because the CPA has multiple objectives it is difficult to narrow in on any one explanation for why it would have a negative impact. In particular, the requirement to spend at least 10% of raised funds on affordable housing would seem to be a drag on the property value benefits of the program. However, that factor is unlikely to explain why property values in historic districts decline disproportionately.

On possible explanation has to do with the regulation of externalities discussed above. Regulation would be expected to reduce home values if it prevents optimal development from happening. Suppose, for instance, that a parcel’s highest value use is as a high-density residential development, but currently contains an historic estate. Making that parcel part of a historic district would clearly reduce its value since its use can no longer be changed. If the CPA is expected to allow for more historic districts to be created, or to support additional restrictions on land uses in the area, for instance by providing funds for the preservation of open space, this would likely compound the effects of the historic district. In this way, historic districts and the CPA, by complementing each other in outcome, may be having complementary negative impacts on property values.

This study does not imply, however, that, as public goods, historic districts are bad. The public benefits of districts are hard to quantify, and so while we do not see substantial evidence of external price appreciation, we cannot be confident that such external benefits do not exist. It may be simply that there are either non-use benefits that accrue to everyone or that benefits are not clearly increasing with proximity to districts. However, what is clear is that advocates, including the National Trust for Historic Preservation, may be mistaken in asserting that one benefit of local historic districts is that they ‘protect and enhance property values’ for residents of these districts. From a policy perspective, it may be necessary to include or increase tax breaks or other financial incentives for district residents in order to secure their support. Most importantly, homeowners considering attempts to have their neighborhoods designated as locally historic may be disappointed in the effects on their home values. Furthermore, since these results seem to go against prior studies using assessment data, it seems likely that tax assessors are over-valuing homes in historic districts, which has clear economic implications for the owners of these homes.

Finally, there are important methodological implications of this study. When we use census block level fixed effects, we see evidence of a positive internal impact on values, but this effect becomes negative and significant in the repeat sales analysis. This points to the necessity of controlling for selection effects which cause omitted variables and endogeneity biases in hedonic analyses. The qualitative differences between the census block and property-level fixed effects models are strong evidence that at the census block level we are failing to control for unobserved factors which are driving the selection of homes into historic districts. This also makes us second guess much of the prior literature on this topic which did not account for this bias.

Footnotes
1

National Trust for Historic Preservation, http://www.nps.gov/hps/workingonthepast/does_doesnot.htm.

 
2

Historic preservation can be characterized this way because it exhibits both characteristics of a public good: it is both non-rivalrous and non-excludable. To the extent that historic preservation has non-use values, that is, people value knowing that historical properties exist, even if they do not plan to visit or otherwise exploit them, these traits are clear. Even in considering use values, however, in most ways, my enjoyment of a historic neighborhood, by visiting it, does not prevent anyone else from similarly enjoying it (non-rival) and we similarly cannot exclude people from enjoying a historic district once it is preserved.

 
3

What we refer to as subsidies are, in fact, any policy that provides a monetary benefit to homeowners whose homes are designated as part of a historic district. These can take the form of tax breaks, grants, or explicit subsidies to enable the protection of designated properties. See (Noonan, forthcoming) for more detail on these policies.

 
4

See the website of the National Trust for Historical Preservation for descriptions and data regarding teardowns in the United States. http://www.preservationnation.org/issues/teardowns/.

 
5

If the effect does vary with the age of the home, this would be made evident through the inclusion of a term interacting designation with age.

 
6

For a more thorough review of this literature, see (Noonan, forthcoming). For a survey and analysis of hedonic analyses more generally see Sirmans et al. (2005) and Sirmans et al. (2006).

 
7

Massachusetts General Laws: Chapter 40C.

 
8

“Summary of an Act to Sustain Community Preservation, SB 90”, available at http://www.communitypreservation.org/CPALegislation.cfm.

 
9

A database of CPA projects can be accessed at http://www.communitypreservation.org/CPAProjectsSearchStart.cfm.

 
10

It is not sufficient to include the District dummy variable and the linear distance since this would still force a monotonic relationship between distance and price because the distance for those inside of a district is zero. Such a monotonic relationship is unrealistic.

 
11

As discussed above, if external effects exist, we would expect homes closer to districts to be more impacted than those further away. We include such large dummies in the hopes of identifying a distance at which districts cease to have external effects, if they exist.

 
12

Unfortunately, detailed information on the approval process for the districts in our sample is not readily available, and there is likely considerable heterogeneity in this process from district to district, so we are unable to systematically control for any anticipatory bias. There are fewer than ten transactions in our dataset, however, that take place within a new district fewer than 12 months before that district was created, so any bias is presumably quite small.

 
13

We employ fixed effects because it is the most appropriate for our context. Instrumental variables requires us to be able to identify an observable variable that is correlated with historic designation but not property values. We were unable to identify such a variable. A regression discontinuity approach relies on using homes on either side of an exogenous border. This approach was considered, but evidence suggests that historic district boundaries are not exogenous, otherwise we would not have to worry about this at all. We were unable to identify another border, geographic or otherwise, which could provide this exogeneity.

 
14

There is one implicit assumption that we have to make for this fixed effects analysis to be valid. This is that there are no time varying unobservables that are correlated with the creation of historic districts and with property values at the very local level. This type of omitted variable could have similar effects as traditional omitted variables bias. Our normalization of the prices reduces concerns about any large-scale omitted variables of this type, but, while we have no reason to believe that any smaller scale effects of this type are at play, we cannot completely rule them out.

 
15

For homes that sell more than once, we only observe property characteristics at the time of the most recent sale. To the extent that there are renovations to the home, we are unable to observe this. We are also limited in terms of which characteristics can be included, again, because of what is available in the dataset.

 
16

Since our fixed effects analysis depends on changes in a parcel’s status as regards historic preservation over time, it is important to know how much variation we actually observe in these variables. In our dataset, there are 87 observations for 39 parcels in seven different districts whose InDistrict status changes during the sample period.

 
17

We also tried using fixed effects and clustering at the census block-group level. Results were broadly consistent with those presented here, and are available from the authors.

 
18

Importantly, the InDistrict and Age/District interaction terms are extremely highly collinear which helps explain why the inclusion of the interaction term makes the InDistrict coefficient become insignificant. This also helps to justify the interpretation above that the results are indicative of a positive internal effect of historic districts.

 
19

We did experiment with non-linear specifications of many of these variables, especially the distance variables. Because these effects are not the focus of this paper, these results are not reported, but are available from the authors.

 
20

By restricting the sample in this way to homes that sell more than once, we risk the introduction of a sample bias. However, an examination of summary statistics suggests no substantial differences in observed characteristics between those homes that do and do not sell more than once during the sample period. Homes selling more than once tend to be somewhat smaller and have a correspondingly lower sales price, but these differences are not large. More importantly, the fixed-effects analysis implicitly controls for anything, observed or unobserved, that makes these homes different, and so any differences should not be reflected in the estimated effect of historic districting. Due to space constraints, summary statistics broken out by repeat/non-repeat are not included in this manuscript, but are available from the authors.

 
21

Notice that the adjusted R2s in these models are very small. This is because, at this scale, we are only using within parcel variation, so that most of the overall variation is explained by the unestimated parcel level fixed effects parameters. The R2 statistic does not include that variation explained by these parameters, αi in Eq. 2. See Cameron and Trivedi (2010), Chapter 8.

 

Acknowledgements

Thanks are due to Todd Easton, Douglas S. Noonan, Stephen Sauer, and seminar participants at Binghamton University and the 2011 Eastern Economics Association Meetings for comments and suggestions which have improved this manuscript. Funding from the Clarkson University School of Business and the Fredric C. Menz Endowment Fund made this work possible. All errors are, of course, our own.

Copyright information

© Springer Science+Business Media, LLC 2011