Finding an Impact of Preservation Policies:
Price Effects of Historic Landmarks on Attached Homes in Chicago 1990-1999
Douglas Noonan
Assistant Professor
School of Public Policy
Georgia Institute of Technology
Doug.Noonan@pubpolicy.gatech.edu
Draft Copy. Please do not cite without permission.
Comments welcome.
ABSTRACT:
Since the 1960s, historical preservation policies have been used to protect landmarks and
neighborhood character, as well as to promote local development. Landmark designation
by local governments may have a variety of effects on the value of historical properties
and their neighbors. This paper offers new empirical evidence on the effect of landmark
designation on property values. Using a hedonic price approach for 70,000 attached
home (i.e., condos, townhomes) sales during the 1990s in Chicago, robust estimates
suggest that homes in landmark buildings (and landmark districts to a lesser extent) sell at
a premium (6 - 15%) over other similar properties. Yet conclusions that historic
preservation policies cause this price increase are premature. Using a repeat-sales
framework (the subsample of properties that sold multiple times during the 1990s), the
effect of landmark designation on appreciation rates can be identified. Given that
construction quality and other unobserved property characteristics are likely to be
different for landmarks, the repeat-sales approach properly controls for a property's predesignation value. The repeat-sales results suggest that landmark designation in Chicago
does not have a positive effect on property values, and may even lead to lower
appreciation rates. These results are consistent with the idea that preservation policies
have struck a rough balance between competing effects, both positive (e.g., prestige,
stability) and negative (e.g., restricted uses).
I. Introduction
Urban renewal and community development have close and complex relationship with
historic preservation. Historic buildings and neighborhoods often find themselves the
focus of redevelopment efforts – either as an obstacle or a catalyst. On the one hand,
redevelopment and land-use changes may draw the ire of preservationists by
transforming historical structures, landscapes, or some “character” of a neighborhood.
On the other hand, historic landmarks may be exploited as amenities to foster local
redevelopment. In any event, the historic character of the built environment and the
extent to which it is preserved are pressing issues in many urban policy debates. “Smart
growth” advocates, for instance, commonly recognize the importance of historic
preservation.
Numerous policy tools directly affect historic preservation efforts. At the federal level,
the National Register of Historic Places boasted 76,835 properties in 2003 and adds over
1,000 sites annually. This amounts to over one million historic sites listed in the U.S.
(see Swaim 2003, Schuster 2002 for further discussion of the National Register). Federal
listing is voluntary and carries no restrictions on private property, but it does make
rehabilitation of some (income-producing) properties eligible for a 20% investment tax
credit. While the federal listing is largely honorific, especially for owner-occupied
residences, state and local historic landmark designations may carry considerably more
weight. State and local rules governing historic properties can vary widely. Some
localities include financial assistance or restrictions on use for historic property owners,
sometimes using the National Register to identify historic properties.
In addition, the National Historic Preservation Act of 1966 made provisions for property
owners to receive tax deductions for preservation easements on historic properties. An
owner of a historic hotel could, for instance, give an easement over the hotel’s façade that
would prevent the owner (and all future owners in perpetuity) from modifying its historic
character in exchange for a tax deduction equal to the lost value of the property.
In the local example considered in this paper, the Commission on Chicago Landmarks
has been recommending landmark designation for properties to the City Council since
established by ordinance in 1968. In the past 35 years, 4,500 properties in 34 historic
districts have been so designated. The Commission in the Landmarks Division of the
city’s Department of Planning and Development must review and approve alterations or
construction that affects landmarks. They also oversee a variety of financial incentive
programs for landmark building owners. Owner-occupied residences are eligible for a
12-year freeze on property taxes and waivers of building permit fees. Other incentives
apply to different landmark property types.
Background
One rationale for historic preservation policies is that markets fail to optimally provide
for preservation. A similar argument is often proffered for environmental preservation.
If the traditional identity of a neighborhood or community is a public good, competitive
markets may underprovide it. More specifically, individual properties may contribute to
1
the historic character of the urban environment, and this historical externality may affect
the well-being of others in ways not captured by prices. The owner of the historic hotel
may add to the charm of downtown, and certainly to the quality of the view from
apartments across the street, yet the owner receives no compensation from those external
benefits. Thus, it is argued, policies are needed to preserve those historic characteristics.
In practice, though, historic preservation may represent a very hard case for market
failure-motivated policy. Typical spillovers cause harms or benefits that are easier for
involved parties to conceptualize. Unlike smoke or noise spillovers, where people are
generally able to imagine the worlds with and without the externality, historical character
may speak to deeper questions of individual and group identity. This might even be
construed as cultural or social capital. Someone whose identity or values have been
influenced by growing up in the historical area is likely to measure the historical spillover
quite differently than an outsider. The possibility that the amount of historical
preservation may affect one’s value of historical preservation makes the policy questions
of entitlement to unrestricted property use or to neighborhood stability.
Historic preservation policies can have several effects on property markets. They are
often evident in prices. Restrictions on property use should reduce property values,
reflected in lower sale prices. Eligibility for tax deductions and other financial
assistance, on the other hand, should increase property values and be reflected in higher
sale prices. Landmark designation that confers honorific status may also see that
symbolic value captured by higher prices. Preservation policies also provide stability to a
neighborhood by limiting change, thereby reducing the investment for other property
owners. Frequently, observers cite intangible external benefits to historic designation
like signaling “public commitment” to an area (Schaeffer and Millerick 1991), solving
market failure in “providing a sense of unity with the past” (Asabere and Huffman
1994b), strengthening the “social fabric” of a community (New York Landmarks
Conservancy 1977), and “catalyzing” rehabilitation of nearby areas (Listokin et al. 1998,
Coulson and Leichenko 2001).
The Literature
Which of these effects dominates is an empirical question. The scholarly literature has
yielded mixed results on the price effects of historical preservation policies. Leichenko
et al. (2001) review empirical works in this area, showing how the “impact of designation
on property values” varies across studies and across empirical methods. Several earlier
studies (e.g., Scribner 1976, Gale 1991) use a difference-in-difference method to identify
price effect of historic designation. This method typically involves comparing sample
average property value growth rates in historic and non-historic districts. Many other
omitted factors that differ between areas may be relevant and better explain differential
growth rates. More recent scholarship has employed hedonic pricing method to assess
the implicit price of properties’ attributes, with historic designation being one of those
attributes. Examples of this approach (e.g., Schaeffer and Millerick 1991; Asabere and
Huffman 1994a, 1994b; Coulson and Leichenko 2001; Clark and Herrin 1997) control for
many other features of properties yet also find mixed results. Schaeffer and Millerick
(1991) claim that some of the difference in price effects is due to differences in landmark
2
regulation at local and national levels. Leichenko et al. (2001) control for different types
of historic designation and conclude that it does matter in some areas, and that price
effects generally differ across cities and sources of data.
With over 2,000 local historic district commissions and thousands of diverse properties
listed on the National Register, one might expect local studies to yield divergent results.
Leichenko et al. (2001) find historically designated properties in Texas to have 5 – 20%
higher appraised prices than other properties. Coulson and Leichenko (p.118, 2001) find
that local “historical designation adds about 17.6% percent to the value of a unit” in
Abilene, TX. In Philadelphia, owner-occupied properties located in historic districts
listed on the National Register sell at 26% higher prices than other properties sampled by
Asabere and Huffman (1994b). Philadelphia condominiums with historic easements,
however, sell for about 30% less than comparable properties, and that price is discounted
by 4.6% per year after the donation (Asabere and Huffman 1994a).
Another interesting question, which the empirical literature has addressed more sparingly
to date, is one of externalities from historic preservation. It is hypothesized that historic
buildings may have positive (or negative) effects on neighboring properties, and that
preserved properties may also have positive external effects beyond the effect that unit’s
own price. Coulson and Leichenko (2001) seek to estimate the external impact of
historic designation, using a hedonic price method that includes the number of
historically designated properties in a unit’s Census tract as an attribute of that unit. They
find that each additional historic house in a tract is associated with sales price that is
0.14% higher. Thus, in Abilene, TX where an average tract has 13 historically
designated properties, houses sell for 1.8% higher than tracts with no historically
designated properties.
Other benefits may accrue from historic preservation policies that are not captured by
prices. There may be public goods benefits from preservation beyond the impact on
properties and nearby properties. Several authors have sought to measure those public
goods benefits of historic preservation using stated preference techniques such as
contingent valuation (see, e.g., Kling et al. 2004, Chambers et al. 1998).
There is a temptation to conclude that higher or lower prices associated with historic
landmarks are the consequence of their designation. Without a careful research design,
however, this conclusion may be unwarranted. Authors rarely directly confront this
issue.1 Historic designation is likely correlated with other (unobserved) characteristics of
the property. Higher quality properties, those maintained better, or those in premium
locations may be more likely to become designated.2 Moreover, properties in areas ripe
for revitalization or in “hot” areas may attract the attention of other landowners and local
officials who support landmark status, especially in the case of landmark districts. Thus,
buildings in neighborhoods characterized by high or rising prices may also receive
landmark designation (possibly against the desires of the owner). If we think that historic
designation is assigned deliberately depending on site characteristics and expectations
1
2
Exceptions include Gale (1991) and Schaeffer and Millerick (1991).
In theory, designation may “follow the market” (Schaeffer and Millerick 1991).
3
about the future (Coulson and Leichenko 2004), then a selection bias may limit our
ability to interpret the results. Does designating an Abilene home as historic cause its
price to rise by 18%, or do high priced (and high quality) homes become designated?
Caution is called for in interpreting the observed price differential associated with historic
buildings.
This paper offers more empirical evidence on the relationship between prices and historic
designation. The empirical methodology employed here allows for more robust
interpretation of the “effect” or “impact” of historical designation. Moreover, unlike
previous research, the methods used here account for spatial dependence in the data and
improve inferences about the statistical significance of marginal prices of landmark
characteristics. This spatial econometric approach informs our understanding of the
neighborhood effects historic landmarks.
II. Data description
The data for this analysis come from several sources. Landmark information comes from
the City of Chicago’s Landmarks Division of its Department of Planning and
Development (City of Chicago 2004). For the 193 landmarks and 37 landmark districts
in the city (including 4,500 properties), information on their date of designation,
architect, and year of construction is available. Numerous sources contribute to the
geographic data. The U.S. Census Bureau’s TIGER files provide most of the geographic
maps used. These were complemented by a map of Chicago’s community areas
(Siciliano 2004). The property and sales data come from the Multiple Listing Service
(MLS) records of sales of all “attached” residential property sales in the city of Chicago
from 1990 – 1999. These attached properties typically include condominiums and
townhomes (in contrast with detached single-family housing, or multi-family housing).
There are 73,106 attached residential property sales using the MLS in Chicago during the
1990s, which accounts for a large share of all residential property sales in the City. MLS
records include information on many property attributes (e.g., address, number of rooms,
parking availability, number of bedrooms and baths). The MLS records were sufficient
to map 71,893 of these observations, although many of these records are missing valid
information. This is especially true for the “year built” and “square footage” variables,
unfortunately two of the more important variables. As expected, many of the properties
are concentrated downtown, near Lake Michigan, and in other pockets scattered around
the city. Perhaps not surprisingly, most of the City’s landmarks are concentrated in
similar areas. (See Map 1. Blue indicates a landmark; green indicates the city border;
black indicates a property sale.)
Missing information for certain key variables in the MLS dataset has been dealt with in
the following way. First, the analysis is performed using the subsample of those
observations for which no information is missing. Second, the missing values are
replaced using the predicted values from an auxiliary regression.3 Third, the analysis is
For the “square footage” variable, which is missing 20,210 observations, several auxiliary regressions
were run sequentially to construct and estimated square footage variable. The first auxiliary regression
included covariates: year built, # units in building, # of rooms, # bedrooms, # baths, master bath dummy, #
3
4
performed without the “year built” variable. Insofar as the results (for the variables of
interest) differ substantially across these three different methods of handling missing
values, the differences will be so indicated.
Even though the MLS data is not of perfect quality, it does have several desirable
features. First, it captures actual sales and perceived attribute values of those involved in
the transaction. Actual market data are superior to appraisal data in this regard, although
sales data reveal prices only for properties actually sold rather than the universe of
properties in a city. A selection bias is thus possible.4 Second, the MLS data records
have information about a wide range of attributes of the property. This include
information about the list and sale price and date, the agent making the sale, the
dimensions of various rooms, tax payments, etc. Finally, because the sample covers
nearly all sales of attached properties during a 10-year span, many properties were sold
multiple times during that period. This allows for a repeat-sale approach.5
fireplaces, basement dummy, garage dummy, parking spot dummy, waterfront dummy, area of living room,
area of dining room, area of master bedroom, area of kitchen, year of sale, latitude, longitude, distance from
Lake, distance from CBD, distance from nearest park, distance from nearest water body, distance from
CTA rail line, estimated block group demographics (population density, income, housing value, percent
nonwhite, and year of construction), and some interaction terms. The R 2 for this regression is 0.54. Still,
for those observations lacking both “year built” and “square footage,” another auxiliary regression was run
to predict “square footage” using the same covariates as the previous regression, except that “year built”
was omitted. The R2 for this regression was 0.35. For those 713 observations also missing “# bedrooms,”
another auxiliary regression predicted square footage as before, except that “# bedrooms” was omitted. Its
R2 = 0.36. Likewise, a fourth auxiliary regression omitted “# units in building” to estimate square footage
for 2,775 observations. Its R2 = 0.36. For the “year built” variable, which is missing 39,081 observations,
the auxiliary regressions included covariates: square footage, # units in building, # rooms, # bedrooms, #
baths, master bath dummy, # fireplaces, basement dummy, garage dummy parking spot dummy, waterfront
dummy, area of living room, area of dining room, area of master bedroom, area of kitchen, year of sale,
latitude, longitude, distance from Lake, distance from CBD, distance from nearest park, distance from
nearest water body, distance from CTA rail line, estimated block group demographics (population density,
income, housing value, percent nonwhite, and year of construction), and landmark status. Its R 2 = 0.01,
obviously a cause for concern.
4
It is interesting to note that becoming a landmark during the 1990s appears unrelated to the number of
times a property has been sold in the 1990s. See the Appendix for the results of the first-stage of the twostage Heckman model to control for the selectivity bias in the repeat-sales sample. The probit in the first
stage indicates that many attributes of a property predict whether it will have multiple sales, yet properties
that became landmarks during the 1990s appear no more or less likely to have repeat sales. Moreover,
properties that were or will be in landmarks, designated at any time, are no more or less likely to have
repeat sales. Interestingly, proximity to the nearest landmark is negatively related to being a repeat-sale
property, suggesting that landmarks may make nearby neighbors less likely to sell. This may be consistent
with the hypothesized stabilizing effect of preservation.
5
Such an approach might be thought of as similar to the difference-in-difference approach previously
employed in comparing changes in average sales prices across neighborhoods, except now the we can
compare changes in actual sales prices for the same properties.
5
Map 1
6
The variables employed in this analysis are:
Table 1: Variables Used
Variable
Definition
log-price
ln (real sales price, adjusted to 1 January 2000 $ using Chicago’s
housing CPI deflator)
log-area
ln (area of unit in feet2)
floor
estimated floor number of unit (e.g., 1st floor, 4th floor)
year built-e
estimated year unit built (see footnote 3)
year built
year unit built
unitbldg
number of units in the building
rooms
number of rooms in unit
bedrooms
number of bedrooms
baths
number of baths
master bath
master bathroom dummy
fireplaces
number of fireplaces
garage
garage dummy
parking spot
parking spot dummy
parking
garage or parking spot dummy
waterfront
waterfront dummy
distance to CBD
distance to State and Monroe downtown (km)
distance to Lake
distance to Lake Michigan (km)
distance to water
distance to nearest river or lake (km)
distance to CTA
distance to nearest CTA rail line (km)
distance to park
distance to nearest park (km)
latitude
decimal degrees north
longitude
decimal degrees east
northside
northern half of the city dummy
BG-income
median household income (in $1000s), block-group, estimated*
BG-value
median house value (in $1000s), block-group, estimated*
BG-density
population density (1000s/km2), block-group, estimated*
BG-nonwhite
percent non-white, block-group, estimated*
BG-year built
median year built for residences, block-group, estimated*
BG-landmarks
number of landmarks, block-group, estimated*
landmark
designated a landmark by 2004 (includes properties in districts)
district
inside a landmark district designated by 2004
CL-year built
year built of closest landmark
CL-date designated date (in days) of designation of closest landmark
CL-distance
distance to closest landmark (km)
year
year of sale
appreciation
appreciation rate from first to last sale in 1990s
price1
real price (in $1000s) of first sale in 1990s
designation
designated a landmark between first and last sale in 1990s
desig90s
designated a landmark any time during the 1990s
* These block-group characteristics were estimated for the sale year using a linear interpolation of the 1990
and 2000 Census estimates for each block group.
7
The following table shows the descriptive statistics for the variables used.
Table 2: Descriptive Statistics
Variable
N
real price
71885
log-price
71884
log-area
66841
floor
63730
year built-e
59889
year built
32809
unitbldg
68941
rooms
71294
bedrooms
67890
baths
71866
master bath
71893
fireplaces
71893
garage
71893
parking spot
71893
parking
71893
waterfront
71893
distance to CBD
71893
distance to Lake
71893
distance to water
71893
distance to CTA
71893
distance to park
71893
latitude
71893
longitude
71893
northside
71893
BG-income
71891
BG-value
71891
BG-density
71889
BG-nonwhite
71891
BG-year built
71891
BG-landmarks
71891
landmark
71893
district
71893
CL-year built
71893
CL-date designated
71771
CL-distance
71893
year
71891
appreciation
6415
price1
6415
designation
6415
desig90s
71893
mean
$178,377.90
11.85
7.05
7.30
1932.39
1941.83
157.35
4.68
1.91
1.53
0.46
0.29
0.35
0.17
0.88
0.07
5.80
1.84
0.79
0.64
0.33
41.93
-87.66
0.91
$47,974.70
$244,206.00
12.27
0.24
1924.52
0.85
0.03
0.03
1967.88
10360.18
0.53
1995.35
0.09
$167,424.80
0.003
0.01
std. dev.
157227.50
0.69
0.46
10.79
160.50
214.72
247.76
1.81
0.80
0.66
0.50
0.50
0.48
0.37
0.32
0.25
4.04
2.81
0.74
0.67
0.28
0.05
0.04
0.29
21103.45
116753.50
8.43
0.19
235.24
1.22
0.17
0.16
376.42
4222.40
0.83
2.82
1.02
137560.30
0.05
0.09
8
III. Empirical Method
Hedonic Price Method
These data are analyzed using hedonic price models. Hedonic price models are based on
the theory that houses are bundled goods – goods with many different attributes – and
that the marginal prices for the attributes can be identified by assessing how sale prices
vary with bundles’ attributes. This technique is common in urban and environmental
economics, where researchers using hedonic analyses to identify the marginal price of
changes in location, environmental quality, and other neighborhood characteristics. See
Rosen (1974), Freeman (2003), and Cropper et al. (1988) for further discussion of the
method.
The first-stage of a hedonic analysis is usually all that is estimated. The first-stage
hedonic regression estimates the following general model:
Pricei = f(Attributesi) + εi
(1)
where Pricei is the sale price, Attributesi is a vector of attribute of a house, and εi is an
error term, all for the ith house. Equation (1) is typically estimated using a regression
framework, with specification often following a Box-Cox procedure. Cropper et al.
(1988) recommend a semi-log specification as most robust to omitted house attributes.
The estimated coefficients for each attribute can be interpreted as a marginal price for
that attribute. In the semi-log specification, the coefficient measures the percent change
in sale price associated with a marginal change in the attribute.
Repeat Sales
The Chicago MLS data offer an opportunity to use a repeat-sales framework for the
hedonic analysis. This approach has an advantage in that it controls for all time-invariant
unobserved or omitted attributes. If sale prices vary between the two different sales of
the same property, then attributes that do not vary over time offer no explanation.
Consider a semi-log specification of the hedonic price function for a sale of the ith house
in period t:
ln(Priceit) = βAttributesit + γInvarianti + δLandmarkit + εit
(2).
The coefficient β indicates the marginal price (in percent terms) of the attribute. Also, γ
is the marginal price of the Invarianti variable, representing time-invariant attributes of
house i. The Landmarkit variable is another attribute of house i in period t, indicating
whether the house is designated as a landmark (=1) or not (=0). Thus, δ is the marginal
price of landmark status. Estimating (2) would be limited in its ability to identify the
“effect” of landmark status on price if some attributes (time variant or invariant) were
correlated with Landmark and omitted. Taking the difference of ln(Priceit) and
ln(Priceis), for the sale of the same house in periods t and s, yields:
ln(Priceit) – ln(Priceis) = βAttributesit + γInvarianti + δLandmarkit + εit –
βAttributesis + γInvarianti + δLandmarkis + εis
ln(Priceit/Priceis) = β(ΔAttributesi) + δ(ΔLandmarki) + εit – εis
(3).
9
As the time-invariant attributes drop out, the change in ln(Price) is a function of the
change in attributes and the change in landmark status. This construction assumes that
the marginal attribute price is constant over time. Equation (3) can be easily adjusted to
allow for β to vary over time, however.
In equation (3), the estimated δ coefficient still represents the marginal price associated
with a change in landmark status. By estimating (3) instead of (2), however, δ is no
longer subject to bias from omitting Invarianti variables. This applies to observed or
unobserved time-invariant variables. This can be an especially important consideration in
for historic landmark properties, where a variable of intangible or unobserved property
characteristics may explain its different price rather than its mere formal designation. For
example, if landmark properties have higher quality construction, have owners who
actively maintain them, have special or unique “historic” design features, or have extra
prestige associated with them, then a hedonic analysis that omits these difficult-toobserve variables (which are constant over time) will not bias its δ estimate. In other
words, historic designation will not be proxying for all of the underlying features that got
the property designated in the first. It will better capture the before-and-after price effect
of designation.
To operationalize equation (3) for repeat-sales that occur at different intervals (t and s),
the dependent variable is normalized by the years between sales. This repeat-sales model
is then straightforward to estimate:
ln(Priceit/Priceis)/(yeart – years) = β(ΔAttributesi) + δ(ΔLandmarki) + θi
(4).
The dependent variable can now be interpreted as the annualized appreciation rate for
property i. The θi error term is estimated using Huber-White robust errors.6
Spatial dependence
Before proceeding to estimate the hedonic models, the matter of spatial dependence in the
data needs to be addressed. It is common, especially in cities like Chicago, to observe
very strong spatial clustering by attributes. Just as Pricei and Attributei are not randomly
distributed geographically around the city, neither are the error terms. This can lead to
spatial autocorrelation – where the model’s error terms are spatially clustered. As a
result, the standard errors of the marginal price estimates in the hedonic models may be
biased (typically downward in the case of positive spatial autocorrelation) and inferences
may be incorrect. The coefficients remain unbiased, however. This is just one way in
which spatial dependence may exist in the data, and it is the one most commonly
addressed in recent hedonic research.7
Note, however, that the error structure in equation (3) contained εit – εis , which may not be zero in
expectation if equation (2) is based on the full sample of all sales (includes repeat and single sales). To
account for this, θi is replaced with an estimate of the inverse Mills ratio (from a probit of whether a sale
was a repeat sale or not). Controlling for this selectivity bias does not appear to have substantial effects on
the results of interest to this paper. Nonetheless, these results are presented in Table 3 and Table 4.
7
Another form of spatial dependence, often called a “spatial lag”, posits that the dependent variable is
endogenous or that sale prices of neighbors affect each other. This sort of contagion effect has been
observed in residential property sales (e.g., Ioannides 2003) and can also be modeled. But this approach is
beyond the scope of this paper.
6
10
A spatial autoregressive (SAR) model can account for spatial structure of the error term.
(See Anselin 2001, 2003 for further discussion.) In this regression model, an N × N
weights matrix W is introduced. W describes the “neighborliness” of observations to each
other. A common W uses contiguity to define neighbors, assigning a positive value to all
observations adjacent to the observation in question and zero to all other observations.
Inverse distance matrices are also a common weights matrix, where each element of W is
1/dij and dij is the distance between observations i and j. Given the nature of the sales
data (i.e., each observation is a point in space, and many of them are “stacked”), a spatial
weight matrix based on distance bands is used. Here, all observations that are within a
specified distance to the observation in question are treated as neighbors, and the
remainder are not.
Defining W based on distance bounds, the hedonic model changes to incorporate the
spatial structure in the error. The vector of errors, θ, in model (4) is specified as:
θ = Wθ + µ ,
(5)
where µ is an independent and identically distributed vector of error terms, and λ is the
nuisance parameter. λ corrects for the spatial correlation in the error rather than any
interdependence among observed variables. Estimating (4) with the spatial
autoregressive error via OLS now involves a non-spherical error and will leave the
coefficients unbiased but the standard errors both biased and inefficient. Combining
equation (4) in vector notation and equation (5) reveals the following spatial
autoregressive model:
y = Wy + (ΔAttributes)β + (ΔLandmark)δ – W(ΔAttributes)β +
W (ΔLandmark)δ + µ
(6).
Equation (6) differs substantially from equation (4). Accounting for error structure
described in (5) avoids the biases that result from estimating (4).
Different ways to estimate this model include maximum likelihood (ML) and generalized
method of moments (GMM). This analysis uses the SpaceStat (Anselin 1995) software
package to estimate the spatial models. In practice, ML estimation requires inverting an
N x N matrix that is computationally demanding and, in the case of large datasets,
intractable. For that reason, this study (N > 70,000) opts for the GMM approach. The
moment conditions for the model can be solved to obtain consistent estimates of β and δ
via feasible generalized least squares. GMM does not provide a way to test for the
significance of the nuisance parameter λ in the model. The estimation technique uses λ to
find consistent estimates of β and δ but cannot make inferences about the presence of
spatial dependence. For this analysis, W is defined as a contiguity matrix where
neighbors are those within ½ mile of the repeat-sale property, or within ¼ mile of the full
sample of all sales.8
8
Alternative constructions of W were tested here, including 5-, 10-, and 20-nearest neighbors.
Constructions of W based on defining neighbors as those within ¼ mile or ½ mile (corresponding to 2 and 4
city blocks in Chicago, respectively) were ultimately selected due to computational limits. While the
structure of spatial dependence in this data appears more expansive than this small range allows, even a
sparse W matrix with contiguity defined as within a ½-mile band is nearly 900MB large for 60,000
observations. Future research will address alternative weights matrices.
11
Diagnostic tests for the presence of spatial dependence in the data are also conducted.
Two prominent tests are reported here. First is the Moran’s I test. Moran’s I, one of the
oldest and best-performing tests, is a two-dimensional variant of time series correlation
based on Moran (1950). The Robust LM (Error) statistic tests for spatial error robust to
the presence of another form of spatial dependence (Anselin et al. 1996).
IV. Empirical Results
The first set of results are presented for the hedonic regression using all sales of attached
housing in Chicago during the 1990s. This hedonic model estimates equation (2) for a
large set of attributes of the properties and neighborhoods. As is common practice, a
Box-Cox transformation was conducted first. The results of the Box-Cox suggested a
semi-log form for the appropriate specification – consistent with recommendations
elsewhere in the hedonics literature (Cropper et al. 1988).9 The coefficients estimated
using the semi-log model in Table 3 should be interpreted as percent changes in real sales
price for the property, on the margin. Thus, another room or a waterfront location is
associated with properties that sell for 2% and 3% more, respectively. Nearly 60,000
observations had complete (or estimated) information for all variables. Table 3 shows
results for 3 different models – two OLS models (restricted and unrestricted) and a
restricted model that accounts for spatially autocorrelated errors. All test statistics are
reported using robust errors.
The restricted model results indicate a good fit to the data. The R2 = 0.79 for the
restricted model, and 0.81 for the unrestricted model. The estimated coefficients for this
large dataset are typically significant at the 0.05% level, well beyond typical standards
for statistical significance.10 Each of the coefficients for property attributes and
neighborhood characteristics have the expected sign. Bigger units, with more rooms and
parking (but not a garage or an assigned spot), sell for higher prices. Prices were also
higher for sales downtown and near parks, not too close to the river or CTA lines, and in
areas with high property values, low density, fewer nonwhites, lower incomes, and new
buildings. A possible surprise is the positive coefficient for the “distance to Lake”
variable, which indicates that each mile away from Lake Michigan is associated with a
higher sales price. Strictly interpreted, this coefficient indicates that Lake proximity is a
disamenity, after controlling for being on the waterfront (an amenity), being near
lakefront parks (an amenity), being near downtown (an amenity), and being near any
body of water (an amenity). In light of these controls, the omission of variables directly
measuring access to transportation, and the “attached homes” nature of the sample, this
result may not be too surprising.
The variables of interest in Table 3, the landmark variables, tell an interesting story.
Units in properties that are designated landmarks (districts or individual buildings) sell
The Box-Cox transformation yielded a θ = 0.048.
Admittedly, given such a large N, the models presented here lack numerous possible quadratic and
interaction terms, as well as other variables that may be relevant. Numerous other specifications have been
estimated, with little substantive effect on the variables of interest. The parsimonious models are presented
here.
9
10
12
for a substantial premium over comparable properties. First, each additional landmark in
a block group is associated with 1 – 2% lower sales prices for attached homes in that
block group. This contrasts with Coulson and Leichenko’s (2001) findings. Moreover,
older buildings in a block group also appear associated with lower prices. More directly,
however, the landmark status of the property itself is significantly related to its sale price.
Units in properties that were designated landmarks (by 2004) sold for 15.5% higher
prices. Controlling for community areas, that premium falls to 13.5%. (Controlling for
year of construction of the particular property using this dataset does not affect the
coefficient substantially. See Table A1 in the Appendix for results including “year builte”.) Those properties in landmarks, where the landmark is a district, have only a 6 – 8%
premium – suggesting that landmark buildings are greater amenities than landmark
districts. So far, these results are mostly consistent with previous literature and
conventional wisdom. The hedonic equations in Table 3 also account for the properties’
proximity to other landmarks. In the restricted model, the price effect of distance to the
nearest landmark is not significantly different from zero, but it becomes significant after
controlling for community areas. It appears that properties that are farther from
landmarks sell at a premium, and that premium is greater if the nearest landmark was
constructed more recently and designated longer ago. These effects are consistent with
attached homes prices being higher in newer buildings and neighborhoods, and may
reflect that recent landmarks are increasingly “marginal.” Yet the SAR results suggest
the effect of CL-distance may not be significant. While there may be a price effect of
proximity to landmarks, this effect is highly sensitive different modeling assumptions
about space and neighborhoods.
As mentioned before, care should be taken before interpreting the results shown in Table
3 to demonstrate that landmark designation has an “effect” or “impact” on prices.
Landmarks do sell for higher prices, 6 – 15% higher on average. Yet this may be
attributable to unobservable characteristics of the property that are correlated with
designation, rather than the designation itself. Estimating equation (4), and equation (6),
can help us discern the price effect of landmark designation as a tool for historical
preservation.
An important limitation of this research is in the construction of some of the landmark
variables. These variables (landmark, district, CL-year built, CL-date designated, CLdistance) are based on properties and districts that were landmarks as of 2004. They do
not account for the fact that some of these properties and districts became landmarks after
the sale. As new landmarks were designated during the 1990s, this may change a
property’s closest landmark. These variables are best interpreted as representing the
marginal price of current or future landmarks.
13
Table 3: Hedonic regressions for all attached home sales, Chicago, 1990-1999
Variables
constant
log-area
unitbldg
unitbldg2
rooms
bedrooms
baths
master bath
fireplaces
garage
parking
parking spot
waterfront
distance to CBD
distance to CBD2
distance to lake
distance to lake2
distance to water
distance to water2
distance to CTA
distance to CTA2
distance to park
distance to park2
northside
latitude
northside× latitude
longitude
BG-income
BG-value
BG-density
BG-nonwhite
BG-year built
BG-landmarks
district
landmark
CL-year built
CL-date designated
CL-distance
year
year2
community areas
N=
R2 =
a
OLS restricted
Coeff.
t-stat
107.26*
1.84
0.492***
17.34
-8.9E-05***
-7.60
1.8E-08***
5.38
0.020***
4.76
0.088***
6.25
0.172***
18.74
0.059***
15.85
0.062***
15.44
-0.021***
-5.57
0.101***
18.79
-0.065***
-15.8
0.039***
7.27
-0.141***
-28.18
0.003***
28.05
0.066***
6.31
-0.0001
-0.64
0.037***
6.67
-0.011***
-7.12
0.065***
11.76
-0.009***
-6.47
-0.039***
-2.78
0.095***
8.32
-126.6***
-4.36
-0.219
-0.49
3.027***
4.36
1.962**
2.52
-2.7E-04***
-2.91
5.1E-04***
28.35
-0.003***
-13.96
-0.455***
-38.16
-0.0001***
-13.76
-0.021***
-16.13
-0.095***
-3.94
0.155***
6.74
4.5E-05***
10.82
-3.9E-06***
-9.73
0.004
1.24
0.041***
59.77
59982
0.794
OLS unrestricted
Coeff.
t-stat
-2365.50***
-2.60
0.484***
17.18
-1.2E-04***
-9.77
2.0E-08***
6.29
0.020***
4.77
0.091***
6.22
0.174***
18.68
0.058***
15.96
0.056***
14.15
-0.037***
-8.81
0.121***
21.67
-0.084***
-17.64
0.035***
6.29
-0.077***
-7.26
0.008***
13.77
0.241***
12.02
-0.006***
-7.45
0.033***
3.89
-0.007*
-1.88
0.043***
4.94
-0.012***
-3.08
-0.020
-1.26
0.073***
5.87
541.5***
6.11
11.582***
7.98
-12.928***
-6.11
16.127***
10.81
-6.4E-05
-0.65
4.2E-04***
22.41
-0.003***
-13.17
-0.267***
-19.88
-0.0001***
-12.78
-0.007***
-4.78
-0.059**
-2.29
0.135***
5.44
1.1E-05**
2.29
-2.0E-06***
-4.46
0.033***
5.48
3.268***
3.64
-0.001***
-3.6
Included
59982
0.808
SAR restricteda
Coeff.
z-stat
-53.883
-0.56
0.456***
96.36
-6.5E-05***
-7.22
1.1E-08***
5.01
0.020***
20.62
0.099***
42.00
0.180***
57.72
0.050***
16.27
0.053***
18.28
-0.027***
-7.37
0.113**
24.88
-0.077
19.97
0.026***
5.16
-0.160***
23.18
0.003***
12.38
0.043**
2.41
0.001***
2.75
0.035***
3.67
-0.015***
-7.74
0.092***
9.87
-0.006***
-4.09
-0.054**
-2.51
0.072***
3.51
-248.1***
-6.58
-2.283***
-3.15
5.928***
6.58
-0.924
-0.68
0.001***
6.99
2.6E-04***
13.82
-2.0E-04
-0.84
-0.192
12.59
-0.0001***
-10.19
-0.012***
-6.09
-0.045**
-2.19
0.127***
6.38
1.1E-05
1.53
-2.5E-06***
-4.87
-0.007
-1.21
0.039***
60.29
59982
0.762
Contiguity defined as all observations within ¼ mile, or approx. 2 Chicago city blocks.  = 0.650.
* significant at 10%, ** significant at 5%, *** significant at 1% level throughout the paper.
14
Table 4: Repeat-Sales Hedonic Models
Variables
constant
price1
price12
log-area
unitbldg
unitbldg2
rooms
bedrooms
baths
master bath
fireplaces
garage
parking
parking spot
waterfront
BG-landmarks
designation
designation×years since
design.
district
landmark
BG-landmarks difference
CL-year built
CL-date designated
CL-distance
year
years between sales
years between sales2
distancea, lat/longb, BGc,
differencesd controls
inverse Mills ratio
OLS restricted
Coeff.
t-stat
-1001.2
-1.65
-0.001***
-3.32
3.3E-07***
2.66
0.026
-0.099*
0.112
-0.069
0.013
0.017
-0.298*
0.76
-1.81
1.51
-0.87
0.42
0.83
-1.77
0.196
-0.133
0.004
1.47
-1.01
0.07
-0.046**
0.037**
-2.21
2.39
OLS unrestricted
Coeff.
t-stat
-929.38
-1.31
-0.001***
-5.33
3.6E-07***
3.64
0.014
0.34
0.0002
1.19
-3.2E-08
-0.90
-0.009
-0.60
0.034*
1.95
0.020
0.64
0.019
0.47
0.035
0.99
-0.053
-0.89
0.046
0.66
-0.025
-0.30
0.006
0.25
0.017
0.75
-0.275
-1.49
0.021
0.69
0.189
-0.133
-0.046
6.9E-06
-1.9E-06
-0.043
0.028**
-0.061
0.007
(yes)
1.27
-0.87
-0.93
0.31
-0.54
-1.60
2.14
-1.18
1.27
Heckman restricted
Coeff.
z-stat
54.724
0.55
0.001*** -3.22
2.9E-07**
2.27
0.051
0.184
-0.208
0.251
0.047
0.015
-0.287
1.30
0.97
-0.96
1.09
0.75
1.05
-1.13
0.116
-0.090
-0.016
0.44
-0.36
-0.09
-0.013
-0.045
-0.45
-0.80
(yes)
R2=0.01
N=5462
R2=0.01
0.153
-0.119
-0.046
-1.2E-05
-3.0E-07
-0.022
-0.019
-0.059**
0.007**
(yes)
0.463
N=6011
Heckman unrestricted
Coeff.
z-stat
-0.001***
-3.14
3.6E-07***
2.51
0.008
0.13
0.0001
1.19
-3.0E-08
-0.48
-0.014
-0.77
0.025
0.67
0.022
0.53
0.030
0.73
0.061*
1.69
0.102
1.24
-0.133
-1.35
0.167
1.64
0.020
0.34
0.015
1.05
-0.285
-0.40
0.020
0.06
N=5479
1.47
Wald=
2377
0.60
-0.49
-0.25
-0.29
-0.08
-0.64
-0.92
-2.43
2.43
SAR restrictede
Coeff.
z-stat
-76.01*
-1.84
-0.001***
-2.81
2.4E-07*
1.93
0.011
-0.072
0.087
-0.080
0.018
0.016
-0.196
0.36
-1.47
1.52
-1.55
0.30
1.06
-0.77
0.121
-0.070
0.48
-0.29
-0.034
0.036***
-1.08
3.85
(yes)
0.275
N=5462
2.01
Wald=
2453
(yes)
N=5467
R2=0.01
a
Array of distance variables (to CBD, to lake, to water, to CTA, to park).
Array of geographic variables (northside, northside×latitude×distance to lake, latitude, longitude). Interaction term omitted for restricted model. Longitude omitted in
Heckman models due to collinearity.
c Array of block-group variables (income, value, density nonwhite, year built).
d Array of difference variables. Unrestricted: changes in BG variables, parking, fireplaces, all room types, and units. Restricted: changes in bath, bedrooms.
e Contiguity defined as all observations within ½ mile, or approximately 4 Chicago city blocks.  = 0.207. Moran’s I = 0.0105***. Robust LM(error) = 0.336. (p-value = 0.6)
b
15
Table 4 shows the results of estimating equation (4), where the dependent variable in the
repeat-sales hedonic framework is now the (annualized) appreciation rate. For attached
homes in Chicago, the MLS data has complete (or estimated) information for nearly
5,500 properties that were sold at least twice during the 1990s. As is common in repeatsales hedonic regressions, the explanatory power of the models wanes considerably
relative to the sort of hedonics in Table 3. This is due to the fact that all of the time
invariant attributes of a property (construction quality, neighborhood quality, etc.) drop
out of equation (4) and provide little explanation for differential appreciation rates.
Nonetheless, Table 4 depicts the results of the several repeat sales models: two using
OLS, two controlling for the selectivity bias, and one that controls for spatial
autocorrelation. As is common in repeat-sales hedonics (see, e.g., Kiel and McClain
1995), because changes in supply of or demand for (time-invariant) characteristics may
affect appreciation rates, these controls are included in the specifications in Table 4.
Results for many of these control variables are omitted from Table 4, but Table A2 in the
Appendix reprints the full results for the Heckman unrestricted model.
The repeat-sales framework offers limited evidence of the impact of Chicago’s landmark
designation program on the value of attached homes in 1990s. Unfortunately, even in a
city as large as Chicago with an extensive and vibrant historic landmarks program, only
16 attached homes were sold before and after a landmark designation during the 1990s.
The OLS models indicate that properties that became designated landmarks between
sales saw their property values appreciate at slower rates than other properties. The
restricted OLS model indicates a 29.8% lower appreciation rate for properties that
became landmarks, significant at the 8% level. This estimated effect is less precise in the
unrestricted model, which controls for more property attributes. Similarly, controlling for
the selectivity bias (e.g., homes that sold multiple times may be somehow different from
other homes), diminishes this negative landmark effect somewhat. No models estimate a
positive effect of landmark designation on appreciation rates.
A few other results evident in Table 4 are worth noting. Properties that were or would
become landmarks appreciated more slowly (although not statistically significant from
zero). This effect was reversed for those in landmark districts (again, also not statistically
significant). Appreciation rates for properties that became landmarks do not appear to
differ substantially over time. Finally, there does not appear to be a statistically
significant effect on appreciation rates of increasing the number of landmarks in a block
group.
V. Summary and Conclusion
This paper has added to the empirical literature on the price effects of historical
preservation (landmark) programs. It has used an extensive MLS dataset of attached
home sales in the city of Chicago during the 1990s, combined with other geographic and
demographic data. The empirical method employed, a hedonic price method, estimates
the implicit price of housing attributes, including historic landmark status. Unfortunately,
landmark designation effects cannot be differentiated from correlated, unobserved
housing traits. Consistent with previous results, the hedonic regressions show landmarks
16
having substantially higher prices than comparable properties. This premium is smaller if
the landmark is district rather than a building. In addition, while older buildings and
neighborhoods receive a discount, proximity to a landmark that is newer (in construction
date) and less recent (in designation date) confers a premium. Yet, these effects may be
capturing unobserved neighborhood characteristics as well. A repeat-sales framework
was introduced to address this identification issue. While few properties in the sample
became a landmark in between sales, those that did suggest that the price effect of
landmark designation is not positive. If anything, local landmark designation appears to
harm property values.
This result is consistent with another study of price effects of landmark designation in
Chicago (Schaeffer and Millerick 1991). The results here apply only to local, City of
Chicago landmark status. Future work will control for National Register status.
Additionally, the Chicago Historical Resources Survey (for more information, see City of
Chicago 2004) can be incorporated to improve the controls for historical qualities of
properties that have not yet been designated landmarks. This 1996 survey identified over
17,000 properties in the city with historical or architectural significance. It collected
information about building style and type, architect, and amount of alteration that has
taken place. Using this data source will offer better controls on historical qualities
independent of official designation.
Diminished prices for landmarks may not be worrisome on their own, especially if the
purpose of historical preservation is to maintain affordable housing. It does seem
plausible that historical preservation tools such as those used in Chicago – which makes it
costly or impossible to update older residential properties – will keep housing affordable
by lowering the relative quality of landmark residences. Effective landmark preservation
policies may constrain owners ability provide optimal mixes of housing service, and the
properties accordingly sell at a discount. Unfortunately, affordable housing goals may be
reached not through lowering housing costs, but by lowering the quantity or quality of
housing.
Another view of lower prices for landmarks, suggests a sort of “taxation by regulation”
approach to historical preservation. Landmark owners may lose a little property value, it
is argued, but preservation is serving the broader public interest. Restrictions on property
use are justified by the sizeable external benefits of preservation (Coulson and Leichenko
2001). These externalities accrue from neighborhood stabilization, adding prestige or
maintaining the “charm” of a neighborhood, and other alleged positive spillovers. Thus,
lower prices for designated landmark may be merely the cost of achieving the external
benefits from preservation. The results presented here give limited empirical support for
this view. The hedonic regression for all attached home sales in Chicago in the 1990s
indicates that block groups with more landmarks are disamenities. It also suggests that
properties nearer to landmarks, near to landmarks that were constructed long ago, and
near to landmarks that were newly designated sell at discounts. Although these negative
price effects may be an artifact of the sample, they provide little support for the view that
17
landmark designation confers substantial external benefits to other properties.11 The
hedonic regression for repeat sales actually provides some weak evidence that landmark
designation has negative spillovers for attached homes. Properties’ appreciation rates
may not differ significantly depending on their own landmark status, but they do appear
to be somewhat lower as the number of landmarks in the block group increases.
Space matters
Judging by the effects of various weights matrices (i.e., definitions of neighborliness), it
appears that the neighbor effects in the attached housing market in Chicago are strong for
price levels and weaker for appreciation rates. The neighbor effects on prices appear at
its strongest around one-half mile of a property, though appreciation rates seem to be
interdependent with much closer neighbors. Accounting for spatial interdependence in
the data has relatively limited influence on the estimated marginal prices associated
landmarks. By and large, little changed in the SAR estimates, except that some statistical
significance was lost. This is a common result when the data are positively spatially
correlated (as standard errors are biased down). Not surprisingly, landmark designation
in Chicago exhibits spatial dependence (see Map 1). Perhaps most interesting is the
different in the BG-landmarks coefficient between the OLS and SAR models in Table 3.
This suggests that spatial interdependence in the data may inflate the effects of an area’s
landmarks count in hedonic analyses.
Otherwise, controlling for spatial effects has minor effect on the estimated coefficients.
This weak effect of spatial interdependence is of particular interest for landmarks – which
are often touted as having powerful effects on neighborhood identity, character, and
social fabric. If the spatial interdependence was indeed especially strong concerning
landmarks, we might have expected that to be captured when controlling for neighbor
interactions. Future research that better accounts for the spatial structure of the data (i.e.,
better W matrices) may revise this conclusion.
Conclusive evidence of the price effects of historic preservation programs is elusive, even
for a single market. Properties in landmarks and landmark districts in Chicago clearly
sell for much higher prices than other properties. Yet, we cannot distinguish these effects
from other unobservable traits of the property that are correlated with designation status.
Very little attention has been paid to controlling for these unobserved quality
characteristics, a shortcoming addressed by the repeat-sales approach introduced in this
paper. Doing so demonstrates the difficulty in identifying causal effects of preservation
policies. Additional research is needed to explore the possible external effects of
preservation on other properties and on welfare more generally.
11
There may be sizable external benefits not captured by properties, e.g., public goods values. These were
found in studies like Kling et al. (2004) and Chambers et al. (1998).
18
References
Anselin, Luc. 1995. SpaceStat version 1.80 User’s Guide. Morgantown, WV: Regional
Research Institute, West Virginia University.
Anselin, Luc. 2001. “Spatial econometrics.” in B. Baltagi, ed. Companion to
Theoretical Econometrics. Oxford: Basil Blackwell, pp. 310 – 330.
Anselin, Luc. 2003. “Spatial Externalities.” International Regional Science Review. 26
(2): 147 – 152.
Anselin, Luc, A. K. Bera, Raymond Florax, and M. J. Yoon. 1996. “Simple Diagnostic
Tests for Spatial Dependence.” Regional Science and Urban Economics. 26: 77 –
104.
Asabere, Paul K. and Forrest E. Huffman. 1994a. “The Value Discounts Associated
with Historic Façade Easements.” The Appraisal Journal. 62 (2): 270 – 277.
Asabere, Paul K. and Forrest E. Huffman. 1994b. “Historic Designation and Residential
Market Values.” The Appraisal Journal. 62 (3): 396 – 401.
Chambers, Catherine M., Paul E. Chambers, and John C. Whitehead. 1998. “Contingent
Valuation of Quasi-Public Goods: Validity, Reliability, and Application to
Valuing an Historic Site.” Public Finance Review. 26(2): 137 –154.
City of Chicago. 2004. “Chicago Landmarks” website.
http://www.cityofchicago.org/Landmarks/ Last accessed 6 October 2004.
Clark, D. E., and W. E. Herrin. 1997. “Historical Preservation and Home Sale Prices:
Evidence from the Sacramento Housing Market.” Review of Regional Studies 27:
29 – 48.
Coulson, N. Edward and Robin M. Leichenko. 2001. “The Internal and External Impact
of Historical Designation on Property Values.” Journal of Real Estate Finance
and Economics. 23 (1): 113 – 124.
Cropper, Maureen L, Leland Deck, and Kenneth E. McConnell. 1988. “On the Choice of
Functional Forms for Hedonic Price Functions.” Review of Economics and
Statistics. 70(4): 668 – 675.
Freeman, A. Myrick. 2003. The Measurement of Environmental and Resource Values.
Washington, DC: Resources for the Future.
Gale, Dennis E. 1991. “The Impacts of Historic District Designation.” Journal of the
American Planning Association. 57 (3): 325 – 341.
19
Ioannides, Yannis M. 2003. “Interactive Property Valuations.” Journal of Urban
Economics. 145 – 170.
Kiel, Katherine A. and Katherine T. McClain. 1995. “The Effect of an Incinerator Siting
on Housing Appreciation Rates.” Journal of Urban Economics. 37: 311 – 323.
Kling, Robert, Charles Revier, and Karin Sable. 2004. “Estimating the Public Good
Value of Preserving a Local Historic Landmark: The Role of Non-Substitutability
and Citizen Information.” Urban Studies. 41 (10): 2025 – 2042.
Leichenko, Robin. M., N. Edward Coulson, and David Listokin. 2001. “Historic
Preservation and Residential Property Values: An Analysis of Texas Cities.”
Urban Studies. 38 (11): 1973 – 1987.
Listokin, David, Barbara Listokin, and Michael Lahr. 1998. “The Contributions of
Historic Preservation to Housing and Economic Development.” Housing Policy
Debate. 9 (3): 431 – 478.
Moran, P. A. P. 1950. “A Test for the Serial Dependence of Residuals.” Biometrika. 37:
178 – 181.
New York Landmarks Conservancy. 1977. The Impacts of Historic Designation—
Summary. Study conducted by Raymond, Parish, Pine and Weiner, Inc. new
York Landmarks Conservancy, New York.
Rosen, Sherwin. 1974. “Hedonic Prices and Implicity Markets: Product Differentiation
in Perfect Competition.” Journal of Political Economy. 82 (1): 34 – 55.
Schaeffer, Peter V. and Cecily Ahern Millerick. 1991. “The Impact of Historic District
Designation on Property Values: An Empirical Study.” Economic Development
Quarterly. 5 (4): 301 – 312.
Schuster, J. Mark. 2002. “Making a List and Checking it Twice: The List as a Tool of
Historic Preservation.” The Cultural Policy Center at the University of Chicago,
working paper #14.
http://culturalpolicy.uchicago.edu/workingpapers/Schuster14.pdf Last accessed
10 October 2004.
Scribner, D. 1976. “Historic Districts as an Economic Asset to Cities.” The Real Estate
Appraiser. May/June: 7 – 12.
Siciliano, Christopher. 2004. Community area map, downloaded from University of
Chicago library website: http://www.lib.uchicago.edu/e/su/maps/chicomm.zip
Last accessed 6 October 2004.
20
Swaim, Richard. 2003. “Politics and Policymaking: Tax Credits and Historic
Preservation.” Journal of Arts, Management, Law and Society. 33 (1): 32 – 39.
21
Appendix
Table A1: Hedonic model, with all sales, including “year-built” variable
Dep. Var.: log-price
Variables
constant
log-area
floor
year built-e
unitbldg
unitbldg2
rooms
bedrooms
baths
master bath
fireplaces
garage
parking
parking spot
waterfront
distance to CBD
distance to CBD2
distance to lake
distance to lake2
distance to water
distance to water2
distance to CTA
distance to CTA2
distance to park
distance to park2
northside
latitude
northside× latitude
northside× latitude× distance to lake
longitude
BG-income
BG-value
BG-density
BG-nonwhite
BG-year built
BG-landmarks
district
landmark
CL-year built
CL-date designated
CL-distance
year
year2
N=
R2 =
Coeff.
-3379.231***
0.5117***
0.0048***
-1.4E-05*
-0.0002***
3.3E-08***
0.0173***
0.0912***
0.1731***
0.0578***
0.0566***
-0.0535***
0.1288***
-0.0891***
0.0210***
-0.0110
0.0076***
0.2529***
-0.0084***
0.0183**
-0.0041
0.0470***
-0.0152***
-0.0094
0.0713***
932.06***
15.1148***
-22.25***
-0.0013***
14.6732***
0.0006***
0.0004***
-0.0025***
-0.1942***
-0.0001***
-0.0071***
-0.0356
0.1390***
1.1E-05**
-2.2E-06***
0.0182***
4.0059***
-0.0010***
48863
0.8157
t-stat
-3.64
15.16
25.98
-1.66
-16.05
8.11
4.02
4.92
15.04
14.88
12.83
-12.10
22.18
-17.78
3.79
-0.99
13.95
13.59
-10.20
2.21
-1.27
4.78
-3.13
-0.55
5.16
9.69
9.27
-9.69
-6.73
10.35
4.93
21.05
-11.18
-13.05
-14.10
-4.27
-1.37
5.61
2.07
-4.47
2.93
4.35
-4.30
p-value
0
0
0
0.097
0
0
0
0
0
0
0
0
0
0
0
0.321
0
0
0
0.027
0.206
0
0.002
0.583
0
0
0
0
0
0
0
0
0
0
0
0
0.17
0
0.039
0
0.003
0
0
22
Table A2: Full Heckman (ML) model for Repeat Sales
Appreciation Rate regression
Selection (repeat sales) probit
Variable
Coeff.
z-stat
p-value
Coeff.
z-stat
p-value
constant
217.49600**
1.98
0.047
350.2115***
21.28
0
price1
-0.00071***
-3.42
0.001
0.04341*a
1.74
0.082
price12
3.14E-07***
2.81
0.005
log-area
0.00779
0.17
0.864
-0.05032*
-1.84
0.066
unitbldg
0.00015
1.30
0.195
-0.00008
-1.63
0.102
unitbldg2
4.65E-08
-1.42
0.156
6.99E-09
0.91
0.363
rooms
-0.01980
-1.14
0.254
-0.01806**
-2.00
0.046
bedrooms
0.00162
0.07
0.947
-0.04589**
-2.48
0.013
baths
0.01572
0.46
0.643
-0.01066
-0.55
0.585
master bath
0.04412
1.00
0.316
0.05255***
2.83
0.005
fireplaces
0.10693**
2.32
0.021
0.10101***
5.81
0
garage
0.49731**
2.35
0.019
0.66481***
19.32
0
parking
-0.60945***
-2.64
0.008
-0.76222***
-18.9
0
parking spot
0.67186***
2.66
0.008
0.82285***
22.52
0
waterfront
0.05878*
1.80
0.071
0.04082
1.30
0.194
BG-income
0.00107
0.87
0.384
0.00017
0.32
0.748
BG-value
-0.00016
-1.05
0.294
-0.00011
-1.19
0.233
BG-density
0.00437**
2.38
0.017
0.00574***
5.74
0
BG-nonwhite
-0.41092*
-1.94
0.052
-0.19274***
-3.20
0.001
BG-year built
-0.00001
-0.45
0.653
BG-landmarks
0.00822
0.51
0.608
-0.00635
-0.76
0.444
designation
-0.16326
-1.37
0.170
designation×years since
design.
0.00102
0.05
0.961
year
-0.14065**
-2.39
0.017
-0.20791***
-32.71
0
distance to CBD
-0.02723*
-1.73
0.083
-0.03950***
-6.63
0
distance to lake
0.01143
0.44
0.663
0.02197***
3.40
0.001
distance to water
-0.02107
-0.89
0.375
0.00465
0.30
0.763
distance to CTA
0.00433
0.13
0.899
-0.04087**
-2.22
0.026
distance to park
0.24087
3.15
0.002
0.15089***
4.61
0
northside
-0.14228
-0.84
0.403
northside× latitude×
distance to lake
-0.00010
-0.36
0.721
latitude
1.48168*
1.66
0.098
1.51950***
4.62
0
district
-0.02964
-0.19
0.849
-0.15042
-1.09
0.276
landmark
0.02416
0.17
0.866
0.08265
0.63
0.526
CL-year built
-0.00004
-1.35
0.178
-0.00006**
-2.38
0.017
CL-date designated
1.56E-06
0.46
0.645
3.51E-06
1.63
0.102
CL-distance
0.00237
0.07
0.942
0.04026**
2.59
0.01
BG-nonwhite
-0.27664
-1.12
0.263
BG-density
-0.00104
-0.33
0.740
BG-value
0.00005
0.32
0.751
BG-income
-0.00057
-0.62
0.534
BG-landmarks
-0.06295
-1.43
0.152
parking spot
0.00833
0.30
0.763
parking
0.00675
0.20
0.841
diff’s
garage
-0.00448
-0.14
0.888
btwn.
fireplaces
0.07908
0.98
0.329
master bath
-0.00600
-0.23
0.819
baths
0.10043
2.77
0.006
bedrooms
0.03121*
1.75
0.080
rooms
-0.00042
-0.04
0.971
unitbldg
-1.36E-08
-0.98
0.327
unitbldg2
0.00010
0.90
0.366
years between sales
-0.00294
-0.12
0.903
years between sales2
0.00100
0.41
0.681
desig90s
-0.01542
-0.20
0.839
N = 59946
uncensored N = 5462
censored N = 54484
a log-price was used in the selection equation.
23
24