- Email: [email protected]
Housing submarkets and the impacts of foreclosures on property prices
Journal of Housing Economics 21 (2012) 235–245
Contents lists available at SciVerse ScienceDirect
Journal of Housing Economics journal homepage: www.elsevier.com/locate/jhec
Housing submarkets and the impacts of foreclosures on property prices Arnab Biswas Department of Marketing and Economics, University of West Florida, 11000 University Parkway, Pensacola, FL 32514, USA
a r t i c l e
i n f o
Article history: Received 17 October 2011 Available online 29 May 2012 JEL classification: R D Keywords: Foreclosure Property values Submarket Spillover Neighborhood
a b s t r a c t The dramatic rise in the number of foreclosed properties since 2006 has come to assume the proportions of a national crisis. It is widely acknowledged that foreclosures hurt neighborhoods by devaluing the nearby properties through various channels. This paper offers a new way of conceptualizing and then estimating the potential effects of foreclosures on property values. Housing stock heterogeneity in the central city old neighborhood allows for the possibility that the impacts of nearby foreclosures may differ across types of housing. This study uses a dataset that covers twenty years of housing values from the City of Worcester (MA), and finds evidence that foreclosures of multi-family houses in close proximity influence the sales price of surrounding single-family properties after controlling for impact from foreclosure of nearby single-family houses. The most preferred estimate suggests that each multi-family foreclosure that occurs between 660 and 1320 feet away from the sale lowers the predicted sales price by approximately 3%. Nearby multi-family spillover impacts also persist over time. In addition, a new approach advocating for an alternative definition of housing submarket suggests that a distant foreclosure within the same submarket also lower sales price of a single-family home by 0.1%. Ó 2012 Elsevier Inc. All rights reserved.
1. Introduction The dramatic rise in the number of foreclosures since 2006 has come to assume the proportions of a national crisis. Foreclosure allows a lender to claim legal rights to the amount owed on a defaulted loan by selling or taking ownership (repossession) of the property securing the loan. Foreclosed properties are likely to sell at discount, both because they may have been physically damaged during the foreclosure process, and because lenders have an incentive to sell them quickly to reduce their holding costs. However, there is a widespread concern that foreclosures may negatively affect neighborhoods by lowering the prices of nearby properties. In the event of foreclosure, properties may sit vacant, reducing the visual appeal and more likely to be attracting vandalism and crime (Immergluck and Smith, 2006; Schuetz et al., 2008; Rogers and Winter, 2009; Campbell et al., 2011). Even the crime-free well maintained E-mail address: [email protected] 1051-1377/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jhe.2012.05.002
vacant properties may depress nearby property values by adding to the local supply of available units. Foreclosures could also affect the price of ‘‘comparables’’ used to estimate neighboring property values (Lin et al., 2009). This study offers a new way of conceptualizing and then estimating the potential impacts of foreclosures on nonforeclosed property sales price. It examines how definitions of housing submarkets may influence the estimated impacts of foreclosures on non-foreclosed property sales prices. The traditional approach implicitly assumes homogeneous housing stocks, and narrowly defines submarkets spatially. It typically uses spatial proximity view to measure such spillover effects. In contrast, this study introduces a new approach, which accounts for housing stock heterogeneity (especially in the context of central city neighborhoods). This in turn firstly allows for the possibility that the impacts of nearby foreclosures may differ across types of housing. For example, multi-family foreclosures can affect nearby single-family properties. Secondly, it suggests that an alternative definition of a non-localized
236
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Table 1 Demographic comparisons across cities. City
Worcester (MA)
Springfield (MA)
Cincinnati (OH)
Providence (RI)
New Haven (CT)
Population density (per square mile) (%) White population (%) Black population Per capita income in 1999 dollars (%) Below poverty level Housing density (per square mile) (%) Occupied housing unit (%) Owner-occupied housing unit (%) Single-family unit (%) Multi-family unit Median block group level multi-family house (%) (%) Built before 1960 Median year structure was built Median gross rent as a % of household income in 1999 dollars Owner-occupied housing units: Median value
4592 77 7 18,614 18 1881 95 43 37 45 37 66 1946 25
4738 56 21 15,232 23 1906 93 50 49 35 24 67 1951 28
4239 53 43 19,962 22 2128 89 39 39 35 49 69 1948 25
9385 55 14 15,525 29 3671 92 35 26 59 51 70 1943 28
6541 44 37 16,393 24 2801 89 30 25 50 24 62 1951 30
118,400
86,500
93,200
101,700
104,300
Notes: Above information is based on the data from U.S. Census (2000). Multi-family units consider up to 19 units only. The median block group level measure is based on an adjustment made towards transforming housing units to number of houses. This measure represents the percentage of multi-family houses at the census block group level.
submarket may also be appropriate in terms of estimating the impacts of foreclosures. It allows for the possibility that houses that are not close by could be viewed as substitutes and those spatially separated foreclosed properties may exert downward pressure on the prices of non-foreclosed properties by adding to the supply of already available dwelling units in the entire submarket. This study is based on the housing market in the City of Worcester, Massachusetts, a mid-sized New England city. Worcester can be used as an example of a large number of middle-sized cities that have a spatial mix of housing rather than homogenous neighborhoods of one type or the other.1 Table 1 indicates that Worcester is indeed similar to other older cities based on a set of demographic characteristics. Especially, the median block group level measure reveals that half of the census block groups in each of these cities contain at least 25% of multi-family houses. This in turn gives an overall idea of housing stock heterogeneity in these older neighborhoods. The city had also experienced a faster rise in home prices followed by a more rapid fall in recent years and has been hit the hardest by foreclosure fallout, which is evident from the Fig. 2. It also shows that the incidence of multi-family foreclosure is a serious issue, and generally higher than that of the singlefamily foreclosures. In an effort to detect spillover effects of foreclosures, this study distinguishes between foreclosures on singleand multi-family dwellings that take place within 660 feet (one-eighth of a mile) and within 1320 feet (one-quarter of a mile) of each single-family transaction in the dataset. The coverage of the dataset is the period from 1991 to February 2008. There is evidence that properties in close proximity to foreclosures sell at a discount. The most preferred estimate suggests that within a 660 foot radius of the subject property, each single-family foreclosure reduces the value
1 Fig. 1 shows the basic facts about the spatial pattern of different types of housing in the City of Worcester.
of nearby single-family properties by approximately 1%. That result is consistent with the current literature. Immergluck and Smith (2006) find that each additional single-family foreclosure within one-eighth of a mile is associated with roughly a 1% decline in single-family property value in Chicago. Leonard and Murdoch (2009) report a similar price discount from every additional foreclosure within 250 feet of a sale of a single-family home in the Dallas County. Recently, an estimate by Campbell et al. (2011) indicates an average price discount of 1% due to a marginal foreclosure occurring within 0.05 mile of the subject property in the state of Massachusetts. In contrast, Lin et al. (2009) report a significant negative impact of up to 8.7% on neighborhood property values up to 0.9 km from the foreclosure, and up to 5 years after the foreclosure in Chicago PMSA. More crucially, this study finds evidence that foreclosures of multi-family houses in close proximity influence the sales price of surrounding single-family properties after controlling for impact from foreclosure of nearby single-family houses.2 Multi-family foreclosures have an impact up to one-quarter mile away from the subject property. The most preferred estimate suggests that each multi-family foreclosure that occurs between 660–1320 feet away from a subject single-family dwelling lowers the predicted sales price by approximately 3%. Such nearby multi-family spillover effects are also persistent over time. In addition to the local spillover effects, a distant foreclosure within the entire submarket for a single-family house has a substantial negative impact of about 0.1% on the property sales prices. The organization of the paper is as follows. The following section discusses the research hypotheses. Data and research methodology make up section three. Results are discussed in section four. Section five concludes the paper.
2
Foreclosure sales deed indicates the end of the foreclosure process.
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
237
Fig. 1. The spatial distribution of single- and multi-family housing in Worcester, Massachusetts. Source: GIS map for the city of Worcester and Warren data.
2. Research hypotheses Most estimates of the impact of foreclosures are based on geographic (spatial) proximity, and most importantly are confined to only one kind of housing stock (e.g. either single-family or multi-family or condominium). However, within a heterogeneous housing market setup, foreclosure of other types of housing may also exert negative spatial spillover effects on the neighboring property sales price. For example, multi-family houses (which are mostly renter occupied) often reflect lower level of maintenance, and perhaps are more prone to criminal activity and vandalism thereby yielding to differential impact of foreclosure on the nearby single-family property sales price. It is not clear whether the price impact of nearby single-family foreclosures is due to the deterioration in the quality of the neighborhood, or through supply channel, or comparable mechanism, or any combination of the above. In contrast, spillovers from multi-family foreclosures on single-family
property sales prices reflect pure neighborhood externalities effect. The possibility of such neighborhood interaction effects from multi-family dwellings leads to the following ‘‘spatial proximity’’ hypothesis. Hypothesis: Properties in close proximity to foreclosures sell at a discount, and those impacts of nearby foreclosures may differ across types of housing units. Another notable difference in this study is related to the heterogeneity of the housing stock by location and other attributes, which is consistent with a housing market that is not fully integrated. Instead, a housing market can be portrayed as a set of distinct but interrelated submarkets that incorporate dwellings differentiated by one or several alternative features. Studies however indicate that there is no single definition of a housing submarket.3 Some define submarkets as consisting of all dwellings within a specific 3 For detail see Bourassa et al. (1999, 2003) and Goodman and Thibodeau (1998), etc.
238
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
geographical area defined by a census tract, a zip code, a school attendance zone, or political boundaries (see Bourassa et al., 1999; Palm, 1978; Watkins, 2001). In general, submarkets are defined to the extent that the housing stock is homogeneous within such geographic limits. Most studies estimating the impact of foreclosures implicitly define housing submarkets in such a manner where submarkets are based on geographic (spatial) proximity. Such an approach implicitly assumes homogeneous housing stocks (even within each housing type, e.g. single-family), and narrowly defines submarket spatially. Grigsby et al. (1963, 1987) offers an alternative view of housing submarkets that are defined regardless of location. Instead, a submarket is defined as a collection of dwellings that are close substitutes for one another, but relatively poor substitutes for dwellings in other submarkets in terms of offering perceived packages of housing services. This view suggests that spatially separated houses and neighborhoods may be considered close substitutes by market participants. This is because households purchase a bundle of housing attributes and neighborhood characteristics and they do not restrict their search necessarily to one neighborhood. Thus, submarkets are often comprised of non-contiguous pieces of real estate. For example, even two adjacent single-family houses may not be perfect substitutes for each other and hence may belong to different quality submarkets. Such an expanded view of a submarket suggests that as the foreclosed homes come to the market by adding to the supply of existing inventory, distant foreclosures may also negatively affect the price within the same submarket.4 This leads to the following ‘‘crowding-in’’ hypothesis. Hypothesis: Foreclosures may lead to excess supply and hence reduce value within an entire housing submarket whether or not properties are immediately adjacent to each other. This study will test both the hypotheses within the context of Worcester housing market.
Table 2 Descriptive statistics of the variables. Variable
Mean
Standard deviation
Median
Price Bathrooms Lot Size ( 1000) Interior Sqft (1000) Total Rooms Fireplaces Age ( 10) Assault Rate. Dist. to Railroad ( 100)ft. Number of observations
165896 1.594 10.327 1.434 6.296 0.473 5.389 0.018 44.676
92778.35 0.619 16.343 0.549 1.395 0.499 3.366 0.012 40.863
148500 1.5 8 1.3 6 0 5.45 0.014 30.784
18270
18270
18270
Notes: The property characteristic data on the city of Worcester are provided by the Warren Group. Neighborhood characteristics are taken from GIS of Worcester and the Worcester Police Department.
The property sales data for the City of Worcester are provided by the Warren Group, which provides information on real estate transactions for New England. For each transaction, the Warren Group data record basic characteristics of houses including the number of rooms, the lot size, the interior surface area, the style, and the year of construction. It also provides information about whether a property is a single-family house or one of several other kinds of residential housing (such as a two-family, threefamily or condominium). In this study, properties are defined as multifamily if the properties in the dataset are classified either as a two-family, a three-family, a 4–8 unit apartment building or a building with nine or more apartments. The Warren data also provide information on the
sales price of each house, the date of each transaction, and the names of the corresponding buyers and sellers. The data are cleaned by removing transactions that appear to be intra-family transfers of ownership (Campbell et al., 2011).5 Duplicate transactions that reflect intermediation or corrections of public records are also dropped. Finally, transactions with a sales price lower than first percentile ($10,000) are excluded to deal with the lower bound extreme values. The Warren data also record different types of deeds, including a notation that the deed indicates the end of the foreclosure process (foreclosure deed). Since it is assumed that foreclosure effects are not contemporaneous with the date of the sale, the sales data are restricted to 1993–2008 (18,270 single-family transactions), the years for which data on foreclosures during the prior two years are available. To identify the likely effects of foreclosures on each sale, information on only single- and multi-family foreclosures between 1991 and 2008 is considered. Condominium foreclosure cases are excluded. The sales dataset is merged with the GIS for the city of Worcester and with a dataset on crime rates. Data on distance of a property from railroads are also calculated in ArcGIS. Table 2 provides descriptive statistics for the variables used in the analysis. The median singe family house has 1300 square feet of finished area, six rooms, 1.5 bathrooms, no fireplace and is located on an 8000 square foot lot. The median age of the house is 55 years which is higher than the mean age. The median house sells for a nominal price of $148500. The median assault rate is 1.4%.6 The median distance to a railroad is 3078 feet (approximately 0.6 mile or 1 kilometer). Distance from the railroad is a rough measure of other neighborhood disamenities. Since industrial development followed railroads, neighborhoods near railroads are more likely to include abandoned or underused factory buildings, brown field sites and other neighborhood disamenities associated with the pattern of development of an older industrial city such as Worcester.
4 Empirically, it is hard to separate out supply effect from the neighborhood externalities effect of foreclosure locally. Consequently, the negative price effects of those spatially separated foreclosures can be better explained by supply channel, and not through any undesirable neighborhood activities.
5 Although in the absence of any additional information it is very difficult to establish any family relationship between two individuals with the same surname. 6 The assault rate is calculated as the number of assaults per thousand residents in the police statistical area.
3. Data and research methodology
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
239
Fig. 2. Incidence of foreclosure over time. Source: House price index data for Worcester is taken from Office of Federal Housing Enterprise Oversight (OFHEO).
3.1. Econometric specification A variation of a hedonic regression model is chosen to test the negative impact of foreclosures on non-foreclosed housing prices: LnðProperty Valueijt Þ ¼ b0 þ Xij b1 þ Fijt b2 þ Tij b3 þ C ijt
ð1Þ The model’s dependent variable is the natural log per unit sales prices of single-family non-foreclosed property i in Police Statistical Area (PSA) j in year t.7 The log transformation is commonly used in hedonic studies to reflect the non-linearity of bundling characteristics. Xij is a vector of structural characteristics of the subject property i, including the number of rooms, lot size, living area, age etc. The variable Fijt is a vector of the count of foreclosures within a given time and distance interval of the subject property i; Tij is a set of PSA-year fixed effects, which allow for house price variation over time at the PSA level. PSAs are a good proxy for local neighborhoods as they control for time invariant amenities and characteristics of the same. Finally, the regression includes quarterly dummies to account for the seasonality of housing prices. C ijt is an error term. 3.2. Spatial proximity This study specifies the foreclosure measures (Fijt) in two ways. The spatial proximity measure of foreclosure takes count of all nearby foreclosures in different categories (single- and multi-family units) for the twelve months prior to the property sale. Two distance buffers are used to identify those nearby foreclosures around each subject property, consistent with the literature on the effects of proximate phenomena on property values. Fig. 3 provides 7
PSA subdivides all of Worcester into 58 small areas.
a schematic representation of the general spatial model of nearby foreclosures where two buffer areas are drawn, one with a radius of 660 feet, and the other with a radius of 1320 feet. Table 3 shows that the average count of foreclosures near each property sale varies noticeably across different time–distance intervals and types of foreclosure. Around each non-foreclosed single-family house, the concentration of multi-family foreclosed houses on average is less than the single-family foreclosed ones. The average number of multi-family foreclosures also increases with distance from the subject property. Table 4 presents the distribution of property sales by percentage of foreclosure. Almost 90% of the property transactions have no singlefamily foreclosures within 660 feet throughout the year before sale while 96% of the property sales have no multi-family foreclosures.
3.3. Expanded submarket view This study empirically derives a classification of housing submarkets, where the submarket is defined on the basis of substitutability across all characteristics, regardless of the location. It uses a statistical clustering technique to identify housing submarkets. The clustering approach is based on the principle of minimizing the dissimilarities in dwelling characteristics within a cluster, and maximizing the across cluster heterogeneity. Variables of structural and neighborhood characteristics are used in the clustering method.8 The assault rate and distance from railroad are the two variables that constitute the neighborhood variables. While the former controls for the crime rate in the locality, the latter one is important in this context as it captures the basic facts about the spatial pattern of different types of housing in the City of Worcester. Fig. 1 clearly shows that 8
For detail, see Chen et al. (2009).
240
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Fig. 3. A general spatial model of the impact of a foreclosure on property value.
Table 3 Average number of foreclosures.
Table 4 Distribution of property sales by percentage of foreclosures.
0–660 ft Buffer
660–1320 ft Buffer
Period
Mean
Mean
Std. dev.
Count
Single-family housing 1 Year before 1 Year after 1–2 Years before 1–2 Years after
foreclosures 0.124 0.379 0.130 0.391 0.115 0.364 0.120 0.376
0.256 0.267 0.247 0.246
0.554 0.573 0.551 0.556
Multi-family housing 1 Year before 1 Year after 1–2 Years before 1–2 Years after
foreclosures 0.057 0.314 0.057 0.303 0.055 0.288 0.058 0.317
0.139 0.140 0.139 0.137
0.544 0.542 0.532 0.558
Std. dev.
Notes: The distance buffers are drawn with the help of ArcGIS software. Foreclosure data on the city of Worcester are from the Warren Group and Worcester Registry of Deeds.
the multi-family houses are mostly located along the railroads, whereas single-family houses on average are away from the same. In particular, this study uses the two-step clustering method that results in three definable submarkets: high, medium, and low. Table 5 provides the distribution of the single-family house and neighborhood characteristics as well as the foreclosure measures by submarket- which in turn somewhat helps determine the ranking of those submarkets on a set of quality parameters. These submarkets however show a good deal of geographic overlapping, which is evident from Fig. 4. On average, the properties in the high submarket are located at the greatest distance from railroads.
0–660 ft Buffer
660–1320 ft Buffer
1 year before
1–2 Years before
1 Year before
1–2 Years before
Single-family 0 89.1 1 9.57 2 1.17 3 0.13 4 0.02 5 _
89.86 9 1.02 0.09 0.02 0.01
79.22 16.86 3.21 0.63 0.07 0.01
79.95 16.32 3.11 0.49 0.08 0.06
Multi-family 0 95.76 1 3.24 2 0.69 3 0.18 4 0.1 5 0.03
95.67 3.46 0.63 0.18 0.05 0.01
90.8 6.63 1.51 0.6 0.19 0.28
90.82 6.47 1.65 0.56 0.31 0.2
Notes: The distance buffers are drawn with the help of ArcGIS software. Foreclosure data on the city of Worcester are from the Warren Group and Worcester Registry of Deeds.
3.4. Issues with error term Errors are no longer i.i.d. due to two reasons. First, the errors are not identically distributed in the presence of age related heteroskedasticity. Second, the errors may be correlated over the houses within a neighborhood (cluster). In the presence of such intra-cluster correlation of disturbances as well as heteroskedasticity, this study adopts a robust approach that corrects the variance of
241
A. Biswas / Journal of Housing Economics 21 (2012) 235–245 Table 5 Descriptive statistics of the variables for each housing submarket. High
Medium
Low
Variable
Mean
Std. dev
Median
Mean
Std. dev
Median
Mean
Std. dev
Median
Price Bathrooms Lot size (1000) Interior sqft(1000) Total rooms Fireplaces Age (10) Assault rate Dist. to railroad (100) ft. 1 Year before foreclosure Number of observations
261142 2.484 19.187 2.429 8.392 0.968 5.269 0.013 56.279 3.899 2241
139486 0.732 34.994 0.668 1.450 0.175 3.873 0.010 34.422 3.366
245000 2.5 12 2.061 8 1 5.7 0.014 49.70 3
168450 1.539 9.983 1.384 6.183 1 4.999 0.016 53.606 12.654 6475
74055 0.487 15.922 0.327 0.973 0 2.638 0.012 46.289 7.125
154000 1.5 8.393 1.355 6 1 5.3 0.014 36.929 11
141825 1.423 8.482 1.233 5.881 0.0001 5.682 0.020 35.903 32.399 9554
73965 0.479 5.475 0.354 1.173 0.010 3.638 0.013 36.066 20.132
127000 1.5 7.328 1.159 6 0 5.7 0.017 25.670 32
Notes: The property characteristic data on the city of Worcester are provided by the Warren Group. Neighborhood characteristics are taken from GIS of Worcester and the Worcester Police Department.
the linear regression estimator. The standard errors are cluster-corrected at the PSA-year level. There can be another problem in interpreting the estimated nearby impacts since both house prices and foreclosures may be influenced by a common local economic shock. This may result in the correlation of the error term with the measures of nearby foreclosures. For example, if a plant that provides employment to a neighborhood shuts down, it could prompt localized unemployment and foreclosures. Moreover, foreclosures are endogenous to house prices to some extent. With a fall in the house price, homeowners are more prone to default as they face lower or even negative equity. Preferably, separating out these two effects warrants an instrumental variable that would be correlated with foreclosures, but orthogonal to property prices. However, this study is unable to find such an instrument. One way to correct the problem is to use the equivalent of the difference-in-difference approach described in Campbell et al. (2011). This approach includes foreclosures occurring after the sale as a control for a localized higher probability of a foreclosure. The future foreclosures are considered to have at least as much explanatory power for property sale prices as prior foreclosures. In such a framework, the estimated impact of a foreclosure on the sales prices of surrounding properties should be interpreted as the difference between the prior and future foreclosure coefficient estimates. The following equation illustrates the difference-in-difference approach:
LnðProperty Valueijt Þ ¼ b0 þ Xijt b1 þ Fijt-k b2 þ Fijtþk b3 þ Tijt b4 þ C ijt
ð2Þ
The variable Fijt k is a vector of foreclosure count occurring within a given distance interval and within k years prior to the sale of the subject property i. The variable Fijt + k is a vector of foreclosure counts that occur within the same given distance interval and within k years after the sale of the subject property i. The coefficient difference b2–b3 captures the net marginal (difference-in-difference) impact of foreclosures on the sales prices of nearby properties.
Spatial autocorrelation is another common source of imperfection in house price modeling. It can take two forms namely, spatial error dependence and spatial lag dependence. A weight matrix approach is commonly used to model the spatial pattern in error term due to omitted variables. However, there is little agreement on the choice of a proper weight matrix. Several forms are commonly used depending upon the separation distance (such as distance decay and sharing common borders). In the presence of spatially clustered foreclosures, this study uses a strategy that reflects a more direct form of spatial interaction than spatial auto models, since it does not impose any restrictions on the pattern of spatial autocorrelation. In contrast, spatial lag dependence refers to the assumption of correlated errors as they occur between the dependent variables. In this case, the price history in the immediate vicinity of a given property plays a significant role in terms of determining its market value. The consequences of ignoring spatial lag dependence are more severe than the consequences of ignoring spatial error dependence as the former is related to theoretical considerations while the latter is related to a statistical one (LeSage, 1997; Patton and McErleans, 2003). Within this context, the regression model includes spatially weighted average price of non-foreclosed property sales within a radius of 1320 feet and within a year before the sale of the property. The weight is defined as 1320 less the distance between the subject property and a particular non-foreclosed house, divided by 1320. This assigns a maximum weight of 1 if the house is next to the subject property, and a minimum weight of zero to a house 1320 feet away (i.e. on the boundary). A dummy is also introduced to control for the cases where no transaction takes place within 1320 feet during the previous year. 4. Results of estimation This section presents the results of the tests of the spatial proximity and the expanded submarket hypotheses. Table 6 reports the base hedonic regression results without the foreclosure variables. The coefficients on all of the structural variables have the expected signs and
242
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Fig. 4. Single-family housing submarkets. Source: GIS map for the city of Worcester and Warren data.
Table 6 Standard hedonic regression result. Variable
Coefficient
Robust standard error
Bathrooms Lot size (1000) Interior sqft (1000) Total rooms Fireplaces Age (10) Age squared Age cubed N R-square
0.060⁄⁄ 0.001⁄ 0.206⁄⁄ 0.025⁄⁄ 0.092⁄⁄ 0.010 0.004⁄⁄ 0.0002⁄⁄ 18,270 0.642
0.007 0.0003 0.012 0.004 0.007 0.008 0.001 0.0001
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value Notes: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Standard errors are clustered at PSA-year level.
reasonable magnitudes, and are consistent with the literature. A larger number of bathrooms, lot size, living area, and number of rooms all add to the value of the property. For example, each additional bathroom raises the house price by about 6%. A one thousand square foot increase in living area is associated with a 21% increase in the sales price. Coefficients on the squared and the cubed terms of the age variable turn out to be significant, which indicates the presence of non-linear relationship between dwelling price and age. 4.1. Tests of spatial proximity hypothesis Table 7 reports the results of estimating the hedonic regression that includes two alternative sets of specifications for the spatial spillover variables. The first set of specifications (columns 1 through 3) includes information on the combined total of nearby foreclosures of both single- and
243
A. Biswas / Journal of Housing Economics 21 (2012) 235–245 Table 7 Impacts of nearby foreclosures on the sales price of single-family homes: 1 year window. Variable Near before 0–660 ft Far before 660–1320 ft
(1)
(2)
0.033⁄⁄ (0.008) 0.012⁄ (0.005)
All 0.029⁄⁄ (0.008) 0.009+ (0.005)
(3) 0.027⁄⁄ (0.008) 0.009+ (0.005)
(4)
(5)
(6)
0.023⁄⁄ (0.008) 0.002 (0.006)
All 0.018⁄ (0.008) 0.002 (0.006)
0.017⁄ (0.008) 0.002 (0.006)
Controls for potential endogeneity 0.019⁄⁄ (0.006) 0.003 (0.004)
Near after 0–660 ft Far after 660–1320 ft
0.009 (0.007) 0.007 (0.005) 0.029+ (0.017) 0.031⁄⁄ (0.010)
Near before 0–660 ft Far before 660–1320 ft
Multi-family only 0.030+ (0.017) 0.030⁄⁄ (0.010)
0.027 (0.017) 0.029⁄⁄ (0.010)
Controls for potential endogeneity Near after 0–660 ft Far after 660–1320 ft
N R-square
0.029⁄ (0.014) 0.005 (0.010) No 18270 0.642
Yes 18270 0.645
Spatial dependence control Yes 18270 0.646
No 18270 0.643
Yes 18270 0.646
Yes 18270 0.646
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value. Note: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Foreclosure variables are the number of foreclosures that occurred within a year of the sale and within two buffer distances (0–660 ft and 660–1320 ft) of the subject property. Standard errors are clustered at PSA-year level.
multi-family houses that took place within a year before the sale of the subject property. In order to test for an additional effect of multi-family foreclosures, the final set of specifications (columns 4 through 6) includes only the number of multi-family foreclosures as separate explanatory variables. Additional controls for spatial dependence are included in specifications 2, 3, 5 and 6. Finally, the third and sixth specifications include counts of foreclosures that occur within a year after the sale of the subject property. The results in Table 7 indicate that recent neighborhood foreclosures have a significant negative impact on the price at which a house will sell. Each foreclosure within 660 feet of the subject property and that occurs within a year prior to a sale lowers the sales price of a single-family home by 3.3% (column 1). The impact falls to 2.9% (column 2) when the average level of recent unforced sales prices in the neighborhood is included as a regressor. In contrast, foreclosures of properties in the outer ring (660–1320 feet) result in a discount of 0.9–1.2%. Column 3 adds the ‘‘aftersale’’ foreclosures, and only the coefficient on the ‘‘aftersale’’ variable within 660 feet of the sale is negative and statistically significant, suggesting that the occurrence of future foreclosures is indeed correlated with current conditions and property values.9 The magnitude of the coefficient on the ‘‘after-sale’’ variables is smaller than the ‘‘before-sale’’ counterparts, and decreases as the distance 9 Although the reported p-values are based on two-tailed t-test, statistical significance of foreclosure coefficients should be read based on onetailed t-test critical value. Since it is assumed that foreclosures negatively affect the sales prices of nearby houses, the alternative hypothesis is H1: bForeclosure < 0.
from the sale increases. Within this framework, the difference-in-difference approach that takes the differential impact of a foreclosure yields a net impact of about 1% on the sales price for foreclosures occurring within one-eight mile of the subject property. The final three columns explicitly show that the impact of multi-family foreclosures is indeed different from that of single-family foreclosures, since the coefficients on the multi-family variables are mostly statistically significant. The results in specification (4) suggests that each singlefamily foreclosure within 660 feet of the subject property and within a year before the sale lowers the sales price of a single-family home by 2.3% in column 4, or 1.8% in column 5. In contrast, each multi-family foreclosure within 660 feet of the subject property lowers the sales price of a single-family home by 4.8–5.2% (2.3% + 2.9%). The multi-family foreclosure drives the results in the outer ring. For example, impact of single-family foreclosure in the outer ring is not statistically significant whereas the price discount from an additional multi-family foreclosure is about 2.9% in column 4, or 2.8% in column 5. In the presence of ‘‘after-sale’’ foreclosures, the difference-in-difference approach from column 6 suggests that each multi-family foreclosure within 660 feet of the subject property results in a 0.6% (4.4%–3.8%) decline in the sales price of a single-family home while the discount from a single-family foreclosure is 0.8% (1.7%–0.9%).10 Again,
10 The ‘‘before’’ coefficient on multi-family variable within 660 feet of the sale is still statistically significant at 10% level of significance based on onetailed t-test critical value.
244
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Table 8 Impacts of nearby foreclosures over time: Two year window.
1 Year Before
1–2 Year Before
1 Year Before
1–2 Year Before
Variable
(1)
(2)
Near before 0–660 ft Far before 660–1320 ft Near before 0–660 ft Far before 660–1320 ft
All 0.028⁄⁄ (0.008) 0.009+ (0.005) 0.021⁄⁄ (0.007) 0.005 (0.004)
All 0.017⁄ (0.008) 0.003 (0.006) 0.012 (0.008) 0.001 (0.006) Multi-family only 0.044⁄⁄ (0.017) 0.025⁄⁄ (0.008) 0.036⁄ (0.016) 0.009 (0.007)
Near before 0–660 ft Far before 660–1320 ft Near before 0–660 ft Far before 660–1320 ft
N R-square
Spatial dependence control Yes Yes 18270 18270 0.643 0.646
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value Notes: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Foreclosure variables are the number of foreclosures that occurred within two year windows before the sale and within two buffer distances (0–660 ft and 660–1320 ft) of the subject property. Standard errors are clustered at PSA-year level.
only multi-family foreclosures are driving the results in the outer ring with a significant negative price impact of about 3%. Two main points can be drawn from the last three specifications. First, multi-family houses seem to be more likely to be affected by a common spatial shock (as the ‘‘after’’ coefficient is statistically significant for the inner ring). Second, multi-family foreclosures have an impact up to 1320 feet away from the subject property whereas the impact from single-family foreclosure is restricted within a 660 foot radius of the subject property. The final assessment of the spatial proximity hypothesis addresses an issue that relates to the lasting impact of foreclosures. Results from Table 8 indicate that the marginal impacts of both types foreclosure from between one and two years prior to the sale decrease over time. While these estimates do not control for future foreclosures, a persistent effect is evident in the results for multi-family dwellings located within one-eight mile of a subject single-family home.
Table 9 Proximity versus within-submarket effects of foreclosures.
Variable
Semi-log (1)
Semi-log (2)
Elasticity (log–log) (3)
Near before 1 year 0–660 ft Far before 660–1320 ft
0.017⁄ (0.008) 0.002 (0.006)
All 0.015+ (0.008) 0.004 (0.006)
0.026⁄ (0.013) 0.004 (0.011)
Near after 1 year 0–660 ft Far after 660–1320 ft
Controls for potential endogeneity 0.009 0.009 0.015 (0.007) (0.007) (0.012) 0.007 0.006 0.011 (0.005) (0.005) (0.010)
Near before 1 year 0–660 ft Far before 660–1320 ft
0.027 (0.017) 0.029⁄⁄ (0.010)
Near after 1 year 0–660 ft Far after 660–1320 ft
Controls for potential endogeneity 0.029⁄ 0.029⁄ 0.055⁄ (0.014) (0.014) (0.023) 0.005 0.005 0.011 (0.010) (0.010) (0.016) Crowding-in effect 0.001⁄⁄ 0.020⁄ (0.0004) (0.009)
Submarket effect Before 1 year
N R-square
Multi-family only 0.027 0.040 (0.017) (0.027) 0.028⁄⁄ 0.046⁄ (0.010) (0.018)
Yes 18270 0.646
Spatial dependence control Yes Yes 18270 18270 0.646 0.646
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value. Notes: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Foreclosure variables are the number of foreclosures that occurred within a year of the sale and within two buffer distances (0–660 ft and 660–1320 ft) of the subject property. It also adds the number of foreclosures that occurred within 1 year before the sale at the submarket level. Standard errors are clustered at PSA-year level.
4.2. Test of the crowding-in hypothesis
column (6) from Table 7; columns (2) and (3) include the number of foreclosures occurring within the relevant submarket for the year prior to the sale of the subject property. In a specification that already includes controls for nearby foreclosures, the submarket variable captures the pure supply effect of (only) distant foreclosures within the same submarket. The final column reports the elasticity coefficient estimates for the purpose of comparing different types of impacts in a standardized form.11 The results of column (2) indicate that each distant foreclosure that occurred within a year before the sale of the subject property lowered the sales price of a single-family home within the same submarket by about 0.1%.12 Although the
In the context of a heterogeneous stock of housing, an alternative view of housing submarkets suggests that there will be additional effects of distant foreclosures on the sales price of homes as foreclosure leads to crowding-in of excess dwelling units within an entire submarket. Table 9 includes the key results of the test of the submarket hypothesis. Column (1) reproduces the results found in
11 The number of foreclosures at the submarket level is expected to be much higher, so the associated coefficient is going to be much lower. Elasticity coefficients are reported so that it will be easier to see which type of foreclosure is really driving the results. 12 This submarket impact is robust across different sets of submarket delineated using alternative neighborhood variables such as area unemployment rate, median household income, mean travel time to work, percentage of people having bachelor degree etc.
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
effect seems small, the elasticity measure indicates that foreclosures at the submarket level have substantial impact on the sales price of the subject property within the same submarket. Column 3 indicates that a one percent increase in submarket foreclosures leads to a 0.02% decrease in the predicted sales price.13 An interesting way of assessing the quantitative significance of these results is to evaluate the impact of the surge in foreclosures that began in 2006 and continued in 2007. Consider the change in the average foreclosure measure between two years: 2005, a year of low foreclosure activity at the height of the housing bubble, and 2007, which experienced the largest number of foreclosures most likely observed since the 1930s. The increase in foreclosures from 2005 to 2007 on average reduces the predicted sales price of a 2005 single-family house by 5.5% by 2007.14 For example, given the average sales price of $252,425 in 2005, a 5.5% reduction corresponds to the loss of almost $14,000 by 2007. This predicted impact of changes in the housing market is associated only with impacts of foreclosures on the sales price. 5. Conclusion This study offers a new way of conceptualizing and then estimating the potential spillover effects of foreclosures on the sales prices of non-foreclosed single-family houses. Using traditional time-distance intervals, the results confirm the occurrence of substantial additional spillover effects from multi-family foreclosures on the sales prices of neighboring single-family properties. While the impact of single-family foreclosure is relevant within 660 feet of a sale, the discount is substantial (about 3%) for multi-family foreclosures occurring in the outer ring (660–1320 feet), and these multi-family effects do persist over time. Such persistence suggests that foreclosures do not just temporarily depress the nearby property prices but they may have negative physical impact on the immediate surroundings which lasts for some time. When a more expansive definition of a housing submarket is used, results suggest that a distant foreclosure within the entire submarket also has a substantial negative impact on property sales prices. The evidence suggests that the total estimated impacts of all types of foreclosure are much higher than what is reported in the current literature. The results of this study may offer some useful policy implications against the backdrop of recent surge in foreclosure. Evidence calls for government intervention as foreclosure impact is not limited only to those who are directly involved but also the entire community faces the consequences. Especially in the context of old central city
13 This paper also estimates the foreclosure impacts by three quality submarkets. Impacts are found to be limited to high and low submarket. In the high submarket, the nearby foreclosures of both types of housing matter, but not the foreclosures at the submarket level. In contrast, the nearby multi-family foreclosures along with foreclosures at the submarket level have significant impact on the sales prices in the low submarket. Results are available from the author upon request. 14 Estimation uses coefficient measures from column (2) of Table 9.
245
neighborhood with a heterogeneous stock of housing, policymakers should consider the possibility of a heightened impact of multi-family foreclosures. For the policy makers, it will be also worth considering the possibility of additional impacts due to within-submarket effects of distant foreclosures in the context of city neighborhood like Worcester where the phases of housing development took place at different points in time unlike the relatively new suburbs. Acknowledgment The author would like to thank John Brown, Tom Mayock, Junfu Zhang, Wayne Gray, Rahul Rakshit, and Aniruddha Mitra for helpful comments. Special thanks to John Brown and Jacqueline Geoghegan for providing the dataset. References Bourassa, Steven.C., Hamelink, Foort., Hoesli, Martin., MacGregor, Bryan.D., 1999. Defining housing submarkets. Journal of Housing Economics. 8, 160–183. Bourassa, S., Hoesli, M., Peng, V.S., 2003. Do housing submarkets really matter? Journal of Housing Economics. 12, 12–18. Campbell, John.Y., Giglio, Stefano., Pathak, Parag., 2011. Forced sales and housing prices. American Economic Review 101, 2108–2131. Chen, Zhuo., Cho, Seong.-Hoon., Poudyal, Neelam., Roberts, Ronald.K., 2009. Forecasting housing prices under different market segmentation assumptions. Urban Studies 46, 167. Goodman, Allen.C., Thibodeau, Thomas.G., 1998. Housing market segmentation. Journal of Housing Economics 7, 121–143. Grigsby, William.G., 1963. Housing Markets and Public Policy. University of Pensylvania Press, Philadelphia. Grigsby, William.G., Baratz, Morton., Galster, George.C., Maclenna, Duncan., 1987. The dynamics of neighborhood change and decline. Progress in Planning 28, 1–76. Immergluck, Dan., Smith, Geoff., 2006. The external costs of foreclosure: the impact of single family mortgage foreclosures on property values. Housing Policy Debate 17, 57–79. Leonard, Tammy., Murdoch, James.C., 2009. The neighborhood effects of foreclosure. Journal of Geographical Systems 11, 317–332. Lesage, James.P., 1997. Regression analysis of spatial data. The Journal of Regional Analysis & Policy 27, 83–94. Lin, Zhenguo., Rosenblatt, Eric., Yao, Vincent.W., 2009. Spillover effects of foreclosures on neighborhood property values. The Journal of Real Estate Finance and Economics 38, 387–407. Palm, Risa, 1978. Spatial segmentation of the urban housing market. Economic Geography 54, 210–221. Patton, Myles., Mcerlean, Seamus., 2003. Spatial effects within the agricultural land market in Northen Ireland. Journal of Agricultural Economics 54, 35–54. Rogers, William H., Winter, William, 2009. The impact of foreclosures on neighboring housing sales. Journal of Real Estate Research 31, 455– 479. Schuetz, Jenny., Been, Vicki., Ellen, Ingrid.Gould., 2008. Neighborhood effects of concentrated mortgage foreclosures. Journal of Housing Economics 17, 306–319. Watkins, Craig.A., 2001. The definition and identification of housing submarkets. Environment and Planning A 33, 2235–2254.
Contents lists available at SciVerse ScienceDirect
Journal of Housing Economics journal homepage: www.elsevier.com/locate/jhec
Housing submarkets and the impacts of foreclosures on property prices Arnab Biswas Department of Marketing and Economics, University of West Florida, 11000 University Parkway, Pensacola, FL 32514, USA
a r t i c l e
i n f o
Article history: Received 17 October 2011 Available online 29 May 2012 JEL classification: R D Keywords: Foreclosure Property values Submarket Spillover Neighborhood
a b s t r a c t The dramatic rise in the number of foreclosed properties since 2006 has come to assume the proportions of a national crisis. It is widely acknowledged that foreclosures hurt neighborhoods by devaluing the nearby properties through various channels. This paper offers a new way of conceptualizing and then estimating the potential effects of foreclosures on property values. Housing stock heterogeneity in the central city old neighborhood allows for the possibility that the impacts of nearby foreclosures may differ across types of housing. This study uses a dataset that covers twenty years of housing values from the City of Worcester (MA), and finds evidence that foreclosures of multi-family houses in close proximity influence the sales price of surrounding single-family properties after controlling for impact from foreclosure of nearby single-family houses. The most preferred estimate suggests that each multi-family foreclosure that occurs between 660 and 1320 feet away from the sale lowers the predicted sales price by approximately 3%. Nearby multi-family spillover impacts also persist over time. In addition, a new approach advocating for an alternative definition of housing submarket suggests that a distant foreclosure within the same submarket also lower sales price of a single-family home by 0.1%. Ó 2012 Elsevier Inc. All rights reserved.
1. Introduction The dramatic rise in the number of foreclosures since 2006 has come to assume the proportions of a national crisis. Foreclosure allows a lender to claim legal rights to the amount owed on a defaulted loan by selling or taking ownership (repossession) of the property securing the loan. Foreclosed properties are likely to sell at discount, both because they may have been physically damaged during the foreclosure process, and because lenders have an incentive to sell them quickly to reduce their holding costs. However, there is a widespread concern that foreclosures may negatively affect neighborhoods by lowering the prices of nearby properties. In the event of foreclosure, properties may sit vacant, reducing the visual appeal and more likely to be attracting vandalism and crime (Immergluck and Smith, 2006; Schuetz et al., 2008; Rogers and Winter, 2009; Campbell et al., 2011). Even the crime-free well maintained E-mail address: [email protected] 1051-1377/$ - see front matter Ó 2012 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.jhe.2012.05.002
vacant properties may depress nearby property values by adding to the local supply of available units. Foreclosures could also affect the price of ‘‘comparables’’ used to estimate neighboring property values (Lin et al., 2009). This study offers a new way of conceptualizing and then estimating the potential impacts of foreclosures on nonforeclosed property sales price. It examines how definitions of housing submarkets may influence the estimated impacts of foreclosures on non-foreclosed property sales prices. The traditional approach implicitly assumes homogeneous housing stocks, and narrowly defines submarkets spatially. It typically uses spatial proximity view to measure such spillover effects. In contrast, this study introduces a new approach, which accounts for housing stock heterogeneity (especially in the context of central city neighborhoods). This in turn firstly allows for the possibility that the impacts of nearby foreclosures may differ across types of housing. For example, multi-family foreclosures can affect nearby single-family properties. Secondly, it suggests that an alternative definition of a non-localized
236
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Table 1 Demographic comparisons across cities. City
Worcester (MA)
Springfield (MA)
Cincinnati (OH)
Providence (RI)
New Haven (CT)
Population density (per square mile) (%) White population (%) Black population Per capita income in 1999 dollars (%) Below poverty level Housing density (per square mile) (%) Occupied housing unit (%) Owner-occupied housing unit (%) Single-family unit (%) Multi-family unit Median block group level multi-family house (%) (%) Built before 1960 Median year structure was built Median gross rent as a % of household income in 1999 dollars Owner-occupied housing units: Median value
4592 77 7 18,614 18 1881 95 43 37 45 37 66 1946 25
4738 56 21 15,232 23 1906 93 50 49 35 24 67 1951 28
4239 53 43 19,962 22 2128 89 39 39 35 49 69 1948 25
9385 55 14 15,525 29 3671 92 35 26 59 51 70 1943 28
6541 44 37 16,393 24 2801 89 30 25 50 24 62 1951 30
118,400
86,500
93,200
101,700
104,300
Notes: Above information is based on the data from U.S. Census (2000). Multi-family units consider up to 19 units only. The median block group level measure is based on an adjustment made towards transforming housing units to number of houses. This measure represents the percentage of multi-family houses at the census block group level.
submarket may also be appropriate in terms of estimating the impacts of foreclosures. It allows for the possibility that houses that are not close by could be viewed as substitutes and those spatially separated foreclosed properties may exert downward pressure on the prices of non-foreclosed properties by adding to the supply of already available dwelling units in the entire submarket. This study is based on the housing market in the City of Worcester, Massachusetts, a mid-sized New England city. Worcester can be used as an example of a large number of middle-sized cities that have a spatial mix of housing rather than homogenous neighborhoods of one type or the other.1 Table 1 indicates that Worcester is indeed similar to other older cities based on a set of demographic characteristics. Especially, the median block group level measure reveals that half of the census block groups in each of these cities contain at least 25% of multi-family houses. This in turn gives an overall idea of housing stock heterogeneity in these older neighborhoods. The city had also experienced a faster rise in home prices followed by a more rapid fall in recent years and has been hit the hardest by foreclosure fallout, which is evident from the Fig. 2. It also shows that the incidence of multi-family foreclosure is a serious issue, and generally higher than that of the singlefamily foreclosures. In an effort to detect spillover effects of foreclosures, this study distinguishes between foreclosures on singleand multi-family dwellings that take place within 660 feet (one-eighth of a mile) and within 1320 feet (one-quarter of a mile) of each single-family transaction in the dataset. The coverage of the dataset is the period from 1991 to February 2008. There is evidence that properties in close proximity to foreclosures sell at a discount. The most preferred estimate suggests that within a 660 foot radius of the subject property, each single-family foreclosure reduces the value
1 Fig. 1 shows the basic facts about the spatial pattern of different types of housing in the City of Worcester.
of nearby single-family properties by approximately 1%. That result is consistent with the current literature. Immergluck and Smith (2006) find that each additional single-family foreclosure within one-eighth of a mile is associated with roughly a 1% decline in single-family property value in Chicago. Leonard and Murdoch (2009) report a similar price discount from every additional foreclosure within 250 feet of a sale of a single-family home in the Dallas County. Recently, an estimate by Campbell et al. (2011) indicates an average price discount of 1% due to a marginal foreclosure occurring within 0.05 mile of the subject property in the state of Massachusetts. In contrast, Lin et al. (2009) report a significant negative impact of up to 8.7% on neighborhood property values up to 0.9 km from the foreclosure, and up to 5 years after the foreclosure in Chicago PMSA. More crucially, this study finds evidence that foreclosures of multi-family houses in close proximity influence the sales price of surrounding single-family properties after controlling for impact from foreclosure of nearby single-family houses.2 Multi-family foreclosures have an impact up to one-quarter mile away from the subject property. The most preferred estimate suggests that each multi-family foreclosure that occurs between 660–1320 feet away from a subject single-family dwelling lowers the predicted sales price by approximately 3%. Such nearby multi-family spillover effects are also persistent over time. In addition to the local spillover effects, a distant foreclosure within the entire submarket for a single-family house has a substantial negative impact of about 0.1% on the property sales prices. The organization of the paper is as follows. The following section discusses the research hypotheses. Data and research methodology make up section three. Results are discussed in section four. Section five concludes the paper.
2
Foreclosure sales deed indicates the end of the foreclosure process.
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
237
Fig. 1. The spatial distribution of single- and multi-family housing in Worcester, Massachusetts. Source: GIS map for the city of Worcester and Warren data.
2. Research hypotheses Most estimates of the impact of foreclosures are based on geographic (spatial) proximity, and most importantly are confined to only one kind of housing stock (e.g. either single-family or multi-family or condominium). However, within a heterogeneous housing market setup, foreclosure of other types of housing may also exert negative spatial spillover effects on the neighboring property sales price. For example, multi-family houses (which are mostly renter occupied) often reflect lower level of maintenance, and perhaps are more prone to criminal activity and vandalism thereby yielding to differential impact of foreclosure on the nearby single-family property sales price. It is not clear whether the price impact of nearby single-family foreclosures is due to the deterioration in the quality of the neighborhood, or through supply channel, or comparable mechanism, or any combination of the above. In contrast, spillovers from multi-family foreclosures on single-family
property sales prices reflect pure neighborhood externalities effect. The possibility of such neighborhood interaction effects from multi-family dwellings leads to the following ‘‘spatial proximity’’ hypothesis. Hypothesis: Properties in close proximity to foreclosures sell at a discount, and those impacts of nearby foreclosures may differ across types of housing units. Another notable difference in this study is related to the heterogeneity of the housing stock by location and other attributes, which is consistent with a housing market that is not fully integrated. Instead, a housing market can be portrayed as a set of distinct but interrelated submarkets that incorporate dwellings differentiated by one or several alternative features. Studies however indicate that there is no single definition of a housing submarket.3 Some define submarkets as consisting of all dwellings within a specific 3 For detail see Bourassa et al. (1999, 2003) and Goodman and Thibodeau (1998), etc.
238
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
geographical area defined by a census tract, a zip code, a school attendance zone, or political boundaries (see Bourassa et al., 1999; Palm, 1978; Watkins, 2001). In general, submarkets are defined to the extent that the housing stock is homogeneous within such geographic limits. Most studies estimating the impact of foreclosures implicitly define housing submarkets in such a manner where submarkets are based on geographic (spatial) proximity. Such an approach implicitly assumes homogeneous housing stocks (even within each housing type, e.g. single-family), and narrowly defines submarket spatially. Grigsby et al. (1963, 1987) offers an alternative view of housing submarkets that are defined regardless of location. Instead, a submarket is defined as a collection of dwellings that are close substitutes for one another, but relatively poor substitutes for dwellings in other submarkets in terms of offering perceived packages of housing services. This view suggests that spatially separated houses and neighborhoods may be considered close substitutes by market participants. This is because households purchase a bundle of housing attributes and neighborhood characteristics and they do not restrict their search necessarily to one neighborhood. Thus, submarkets are often comprised of non-contiguous pieces of real estate. For example, even two adjacent single-family houses may not be perfect substitutes for each other and hence may belong to different quality submarkets. Such an expanded view of a submarket suggests that as the foreclosed homes come to the market by adding to the supply of existing inventory, distant foreclosures may also negatively affect the price within the same submarket.4 This leads to the following ‘‘crowding-in’’ hypothesis. Hypothesis: Foreclosures may lead to excess supply and hence reduce value within an entire housing submarket whether or not properties are immediately adjacent to each other. This study will test both the hypotheses within the context of Worcester housing market.
Table 2 Descriptive statistics of the variables. Variable
Mean
Standard deviation
Median
Price Bathrooms Lot Size ( 1000) Interior Sqft (1000) Total Rooms Fireplaces Age ( 10) Assault Rate. Dist. to Railroad ( 100)ft. Number of observations
165896 1.594 10.327 1.434 6.296 0.473 5.389 0.018 44.676
92778.35 0.619 16.343 0.549 1.395 0.499 3.366 0.012 40.863
148500 1.5 8 1.3 6 0 5.45 0.014 30.784
18270
18270
18270
Notes: The property characteristic data on the city of Worcester are provided by the Warren Group. Neighborhood characteristics are taken from GIS of Worcester and the Worcester Police Department.
The property sales data for the City of Worcester are provided by the Warren Group, which provides information on real estate transactions for New England. For each transaction, the Warren Group data record basic characteristics of houses including the number of rooms, the lot size, the interior surface area, the style, and the year of construction. It also provides information about whether a property is a single-family house or one of several other kinds of residential housing (such as a two-family, threefamily or condominium). In this study, properties are defined as multifamily if the properties in the dataset are classified either as a two-family, a three-family, a 4–8 unit apartment building or a building with nine or more apartments. The Warren data also provide information on the
sales price of each house, the date of each transaction, and the names of the corresponding buyers and sellers. The data are cleaned by removing transactions that appear to be intra-family transfers of ownership (Campbell et al., 2011).5 Duplicate transactions that reflect intermediation or corrections of public records are also dropped. Finally, transactions with a sales price lower than first percentile ($10,000) are excluded to deal with the lower bound extreme values. The Warren data also record different types of deeds, including a notation that the deed indicates the end of the foreclosure process (foreclosure deed). Since it is assumed that foreclosure effects are not contemporaneous with the date of the sale, the sales data are restricted to 1993–2008 (18,270 single-family transactions), the years for which data on foreclosures during the prior two years are available. To identify the likely effects of foreclosures on each sale, information on only single- and multi-family foreclosures between 1991 and 2008 is considered. Condominium foreclosure cases are excluded. The sales dataset is merged with the GIS for the city of Worcester and with a dataset on crime rates. Data on distance of a property from railroads are also calculated in ArcGIS. Table 2 provides descriptive statistics for the variables used in the analysis. The median singe family house has 1300 square feet of finished area, six rooms, 1.5 bathrooms, no fireplace and is located on an 8000 square foot lot. The median age of the house is 55 years which is higher than the mean age. The median house sells for a nominal price of $148500. The median assault rate is 1.4%.6 The median distance to a railroad is 3078 feet (approximately 0.6 mile or 1 kilometer). Distance from the railroad is a rough measure of other neighborhood disamenities. Since industrial development followed railroads, neighborhoods near railroads are more likely to include abandoned or underused factory buildings, brown field sites and other neighborhood disamenities associated with the pattern of development of an older industrial city such as Worcester.
4 Empirically, it is hard to separate out supply effect from the neighborhood externalities effect of foreclosure locally. Consequently, the negative price effects of those spatially separated foreclosures can be better explained by supply channel, and not through any undesirable neighborhood activities.
5 Although in the absence of any additional information it is very difficult to establish any family relationship between two individuals with the same surname. 6 The assault rate is calculated as the number of assaults per thousand residents in the police statistical area.
3. Data and research methodology
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
239
Fig. 2. Incidence of foreclosure over time. Source: House price index data for Worcester is taken from Office of Federal Housing Enterprise Oversight (OFHEO).
3.1. Econometric specification A variation of a hedonic regression model is chosen to test the negative impact of foreclosures on non-foreclosed housing prices: LnðProperty Valueijt Þ ¼ b0 þ Xij b1 þ Fijt b2 þ Tij b3 þ C ijt
ð1Þ The model’s dependent variable is the natural log per unit sales prices of single-family non-foreclosed property i in Police Statistical Area (PSA) j in year t.7 The log transformation is commonly used in hedonic studies to reflect the non-linearity of bundling characteristics. Xij is a vector of structural characteristics of the subject property i, including the number of rooms, lot size, living area, age etc. The variable Fijt is a vector of the count of foreclosures within a given time and distance interval of the subject property i; Tij is a set of PSA-year fixed effects, which allow for house price variation over time at the PSA level. PSAs are a good proxy for local neighborhoods as they control for time invariant amenities and characteristics of the same. Finally, the regression includes quarterly dummies to account for the seasonality of housing prices. C ijt is an error term. 3.2. Spatial proximity This study specifies the foreclosure measures (Fijt) in two ways. The spatial proximity measure of foreclosure takes count of all nearby foreclosures in different categories (single- and multi-family units) for the twelve months prior to the property sale. Two distance buffers are used to identify those nearby foreclosures around each subject property, consistent with the literature on the effects of proximate phenomena on property values. Fig. 3 provides 7
PSA subdivides all of Worcester into 58 small areas.
a schematic representation of the general spatial model of nearby foreclosures where two buffer areas are drawn, one with a radius of 660 feet, and the other with a radius of 1320 feet. Table 3 shows that the average count of foreclosures near each property sale varies noticeably across different time–distance intervals and types of foreclosure. Around each non-foreclosed single-family house, the concentration of multi-family foreclosed houses on average is less than the single-family foreclosed ones. The average number of multi-family foreclosures also increases with distance from the subject property. Table 4 presents the distribution of property sales by percentage of foreclosure. Almost 90% of the property transactions have no singlefamily foreclosures within 660 feet throughout the year before sale while 96% of the property sales have no multi-family foreclosures.
3.3. Expanded submarket view This study empirically derives a classification of housing submarkets, where the submarket is defined on the basis of substitutability across all characteristics, regardless of the location. It uses a statistical clustering technique to identify housing submarkets. The clustering approach is based on the principle of minimizing the dissimilarities in dwelling characteristics within a cluster, and maximizing the across cluster heterogeneity. Variables of structural and neighborhood characteristics are used in the clustering method.8 The assault rate and distance from railroad are the two variables that constitute the neighborhood variables. While the former controls for the crime rate in the locality, the latter one is important in this context as it captures the basic facts about the spatial pattern of different types of housing in the City of Worcester. Fig. 1 clearly shows that 8
For detail, see Chen et al. (2009).
240
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Fig. 3. A general spatial model of the impact of a foreclosure on property value.
Table 3 Average number of foreclosures.
Table 4 Distribution of property sales by percentage of foreclosures.
0–660 ft Buffer
660–1320 ft Buffer
Period
Mean
Mean
Std. dev.
Count
Single-family housing 1 Year before 1 Year after 1–2 Years before 1–2 Years after
foreclosures 0.124 0.379 0.130 0.391 0.115 0.364 0.120 0.376
0.256 0.267 0.247 0.246
0.554 0.573 0.551 0.556
Multi-family housing 1 Year before 1 Year after 1–2 Years before 1–2 Years after
foreclosures 0.057 0.314 0.057 0.303 0.055 0.288 0.058 0.317
0.139 0.140 0.139 0.137
0.544 0.542 0.532 0.558
Std. dev.
Notes: The distance buffers are drawn with the help of ArcGIS software. Foreclosure data on the city of Worcester are from the Warren Group and Worcester Registry of Deeds.
the multi-family houses are mostly located along the railroads, whereas single-family houses on average are away from the same. In particular, this study uses the two-step clustering method that results in three definable submarkets: high, medium, and low. Table 5 provides the distribution of the single-family house and neighborhood characteristics as well as the foreclosure measures by submarket- which in turn somewhat helps determine the ranking of those submarkets on a set of quality parameters. These submarkets however show a good deal of geographic overlapping, which is evident from Fig. 4. On average, the properties in the high submarket are located at the greatest distance from railroads.
0–660 ft Buffer
660–1320 ft Buffer
1 year before
1–2 Years before
1 Year before
1–2 Years before
Single-family 0 89.1 1 9.57 2 1.17 3 0.13 4 0.02 5 _
89.86 9 1.02 0.09 0.02 0.01
79.22 16.86 3.21 0.63 0.07 0.01
79.95 16.32 3.11 0.49 0.08 0.06
Multi-family 0 95.76 1 3.24 2 0.69 3 0.18 4 0.1 5 0.03
95.67 3.46 0.63 0.18 0.05 0.01
90.8 6.63 1.51 0.6 0.19 0.28
90.82 6.47 1.65 0.56 0.31 0.2
Notes: The distance buffers are drawn with the help of ArcGIS software. Foreclosure data on the city of Worcester are from the Warren Group and Worcester Registry of Deeds.
3.4. Issues with error term Errors are no longer i.i.d. due to two reasons. First, the errors are not identically distributed in the presence of age related heteroskedasticity. Second, the errors may be correlated over the houses within a neighborhood (cluster). In the presence of such intra-cluster correlation of disturbances as well as heteroskedasticity, this study adopts a robust approach that corrects the variance of
241
A. Biswas / Journal of Housing Economics 21 (2012) 235–245 Table 5 Descriptive statistics of the variables for each housing submarket. High
Medium
Low
Variable
Mean
Std. dev
Median
Mean
Std. dev
Median
Mean
Std. dev
Median
Price Bathrooms Lot size (1000) Interior sqft(1000) Total rooms Fireplaces Age (10) Assault rate Dist. to railroad (100) ft. 1 Year before foreclosure Number of observations
261142 2.484 19.187 2.429 8.392 0.968 5.269 0.013 56.279 3.899 2241
139486 0.732 34.994 0.668 1.450 0.175 3.873 0.010 34.422 3.366
245000 2.5 12 2.061 8 1 5.7 0.014 49.70 3
168450 1.539 9.983 1.384 6.183 1 4.999 0.016 53.606 12.654 6475
74055 0.487 15.922 0.327 0.973 0 2.638 0.012 46.289 7.125
154000 1.5 8.393 1.355 6 1 5.3 0.014 36.929 11
141825 1.423 8.482 1.233 5.881 0.0001 5.682 0.020 35.903 32.399 9554
73965 0.479 5.475 0.354 1.173 0.010 3.638 0.013 36.066 20.132
127000 1.5 7.328 1.159 6 0 5.7 0.017 25.670 32
Notes: The property characteristic data on the city of Worcester are provided by the Warren Group. Neighborhood characteristics are taken from GIS of Worcester and the Worcester Police Department.
the linear regression estimator. The standard errors are cluster-corrected at the PSA-year level. There can be another problem in interpreting the estimated nearby impacts since both house prices and foreclosures may be influenced by a common local economic shock. This may result in the correlation of the error term with the measures of nearby foreclosures. For example, if a plant that provides employment to a neighborhood shuts down, it could prompt localized unemployment and foreclosures. Moreover, foreclosures are endogenous to house prices to some extent. With a fall in the house price, homeowners are more prone to default as they face lower or even negative equity. Preferably, separating out these two effects warrants an instrumental variable that would be correlated with foreclosures, but orthogonal to property prices. However, this study is unable to find such an instrument. One way to correct the problem is to use the equivalent of the difference-in-difference approach described in Campbell et al. (2011). This approach includes foreclosures occurring after the sale as a control for a localized higher probability of a foreclosure. The future foreclosures are considered to have at least as much explanatory power for property sale prices as prior foreclosures. In such a framework, the estimated impact of a foreclosure on the sales prices of surrounding properties should be interpreted as the difference between the prior and future foreclosure coefficient estimates. The following equation illustrates the difference-in-difference approach:
LnðProperty Valueijt Þ ¼ b0 þ Xijt b1 þ Fijt-k b2 þ Fijtþk b3 þ Tijt b4 þ C ijt
ð2Þ
The variable Fijt k is a vector of foreclosure count occurring within a given distance interval and within k years prior to the sale of the subject property i. The variable Fijt + k is a vector of foreclosure counts that occur within the same given distance interval and within k years after the sale of the subject property i. The coefficient difference b2–b3 captures the net marginal (difference-in-difference) impact of foreclosures on the sales prices of nearby properties.
Spatial autocorrelation is another common source of imperfection in house price modeling. It can take two forms namely, spatial error dependence and spatial lag dependence. A weight matrix approach is commonly used to model the spatial pattern in error term due to omitted variables. However, there is little agreement on the choice of a proper weight matrix. Several forms are commonly used depending upon the separation distance (such as distance decay and sharing common borders). In the presence of spatially clustered foreclosures, this study uses a strategy that reflects a more direct form of spatial interaction than spatial auto models, since it does not impose any restrictions on the pattern of spatial autocorrelation. In contrast, spatial lag dependence refers to the assumption of correlated errors as they occur between the dependent variables. In this case, the price history in the immediate vicinity of a given property plays a significant role in terms of determining its market value. The consequences of ignoring spatial lag dependence are more severe than the consequences of ignoring spatial error dependence as the former is related to theoretical considerations while the latter is related to a statistical one (LeSage, 1997; Patton and McErleans, 2003). Within this context, the regression model includes spatially weighted average price of non-foreclosed property sales within a radius of 1320 feet and within a year before the sale of the property. The weight is defined as 1320 less the distance between the subject property and a particular non-foreclosed house, divided by 1320. This assigns a maximum weight of 1 if the house is next to the subject property, and a minimum weight of zero to a house 1320 feet away (i.e. on the boundary). A dummy is also introduced to control for the cases where no transaction takes place within 1320 feet during the previous year. 4. Results of estimation This section presents the results of the tests of the spatial proximity and the expanded submarket hypotheses. Table 6 reports the base hedonic regression results without the foreclosure variables. The coefficients on all of the structural variables have the expected signs and
242
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Fig. 4. Single-family housing submarkets. Source: GIS map for the city of Worcester and Warren data.
Table 6 Standard hedonic regression result. Variable
Coefficient
Robust standard error
Bathrooms Lot size (1000) Interior sqft (1000) Total rooms Fireplaces Age (10) Age squared Age cubed N R-square
0.060⁄⁄ 0.001⁄ 0.206⁄⁄ 0.025⁄⁄ 0.092⁄⁄ 0.010 0.004⁄⁄ 0.0002⁄⁄ 18,270 0.642
0.007 0.0003 0.012 0.004 0.007 0.008 0.001 0.0001
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value Notes: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Standard errors are clustered at PSA-year level.
reasonable magnitudes, and are consistent with the literature. A larger number of bathrooms, lot size, living area, and number of rooms all add to the value of the property. For example, each additional bathroom raises the house price by about 6%. A one thousand square foot increase in living area is associated with a 21% increase in the sales price. Coefficients on the squared and the cubed terms of the age variable turn out to be significant, which indicates the presence of non-linear relationship between dwelling price and age. 4.1. Tests of spatial proximity hypothesis Table 7 reports the results of estimating the hedonic regression that includes two alternative sets of specifications for the spatial spillover variables. The first set of specifications (columns 1 through 3) includes information on the combined total of nearby foreclosures of both single- and
243
A. Biswas / Journal of Housing Economics 21 (2012) 235–245 Table 7 Impacts of nearby foreclosures on the sales price of single-family homes: 1 year window. Variable Near before 0–660 ft Far before 660–1320 ft
(1)
(2)
0.033⁄⁄ (0.008) 0.012⁄ (0.005)
All 0.029⁄⁄ (0.008) 0.009+ (0.005)
(3) 0.027⁄⁄ (0.008) 0.009+ (0.005)
(4)
(5)
(6)
0.023⁄⁄ (0.008) 0.002 (0.006)
All 0.018⁄ (0.008) 0.002 (0.006)
0.017⁄ (0.008) 0.002 (0.006)
Controls for potential endogeneity 0.019⁄⁄ (0.006) 0.003 (0.004)
Near after 0–660 ft Far after 660–1320 ft
0.009 (0.007) 0.007 (0.005) 0.029+ (0.017) 0.031⁄⁄ (0.010)
Near before 0–660 ft Far before 660–1320 ft
Multi-family only 0.030+ (0.017) 0.030⁄⁄ (0.010)
0.027 (0.017) 0.029⁄⁄ (0.010)
Controls for potential endogeneity Near after 0–660 ft Far after 660–1320 ft
N R-square
0.029⁄ (0.014) 0.005 (0.010) No 18270 0.642
Yes 18270 0.645
Spatial dependence control Yes 18270 0.646
No 18270 0.643
Yes 18270 0.646
Yes 18270 0.646
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value. Note: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Foreclosure variables are the number of foreclosures that occurred within a year of the sale and within two buffer distances (0–660 ft and 660–1320 ft) of the subject property. Standard errors are clustered at PSA-year level.
multi-family houses that took place within a year before the sale of the subject property. In order to test for an additional effect of multi-family foreclosures, the final set of specifications (columns 4 through 6) includes only the number of multi-family foreclosures as separate explanatory variables. Additional controls for spatial dependence are included in specifications 2, 3, 5 and 6. Finally, the third and sixth specifications include counts of foreclosures that occur within a year after the sale of the subject property. The results in Table 7 indicate that recent neighborhood foreclosures have a significant negative impact on the price at which a house will sell. Each foreclosure within 660 feet of the subject property and that occurs within a year prior to a sale lowers the sales price of a single-family home by 3.3% (column 1). The impact falls to 2.9% (column 2) when the average level of recent unforced sales prices in the neighborhood is included as a regressor. In contrast, foreclosures of properties in the outer ring (660–1320 feet) result in a discount of 0.9–1.2%. Column 3 adds the ‘‘aftersale’’ foreclosures, and only the coefficient on the ‘‘aftersale’’ variable within 660 feet of the sale is negative and statistically significant, suggesting that the occurrence of future foreclosures is indeed correlated with current conditions and property values.9 The magnitude of the coefficient on the ‘‘after-sale’’ variables is smaller than the ‘‘before-sale’’ counterparts, and decreases as the distance 9 Although the reported p-values are based on two-tailed t-test, statistical significance of foreclosure coefficients should be read based on onetailed t-test critical value. Since it is assumed that foreclosures negatively affect the sales prices of nearby houses, the alternative hypothesis is H1: bForeclosure < 0.
from the sale increases. Within this framework, the difference-in-difference approach that takes the differential impact of a foreclosure yields a net impact of about 1% on the sales price for foreclosures occurring within one-eight mile of the subject property. The final three columns explicitly show that the impact of multi-family foreclosures is indeed different from that of single-family foreclosures, since the coefficients on the multi-family variables are mostly statistically significant. The results in specification (4) suggests that each singlefamily foreclosure within 660 feet of the subject property and within a year before the sale lowers the sales price of a single-family home by 2.3% in column 4, or 1.8% in column 5. In contrast, each multi-family foreclosure within 660 feet of the subject property lowers the sales price of a single-family home by 4.8–5.2% (2.3% + 2.9%). The multi-family foreclosure drives the results in the outer ring. For example, impact of single-family foreclosure in the outer ring is not statistically significant whereas the price discount from an additional multi-family foreclosure is about 2.9% in column 4, or 2.8% in column 5. In the presence of ‘‘after-sale’’ foreclosures, the difference-in-difference approach from column 6 suggests that each multi-family foreclosure within 660 feet of the subject property results in a 0.6% (4.4%–3.8%) decline in the sales price of a single-family home while the discount from a single-family foreclosure is 0.8% (1.7%–0.9%).10 Again,
10 The ‘‘before’’ coefficient on multi-family variable within 660 feet of the sale is still statistically significant at 10% level of significance based on onetailed t-test critical value.
244
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
Table 8 Impacts of nearby foreclosures over time: Two year window.
1 Year Before
1–2 Year Before
1 Year Before
1–2 Year Before
Variable
(1)
(2)
Near before 0–660 ft Far before 660–1320 ft Near before 0–660 ft Far before 660–1320 ft
All 0.028⁄⁄ (0.008) 0.009+ (0.005) 0.021⁄⁄ (0.007) 0.005 (0.004)
All 0.017⁄ (0.008) 0.003 (0.006) 0.012 (0.008) 0.001 (0.006) Multi-family only 0.044⁄⁄ (0.017) 0.025⁄⁄ (0.008) 0.036⁄ (0.016) 0.009 (0.007)
Near before 0–660 ft Far before 660–1320 ft Near before 0–660 ft Far before 660–1320 ft
N R-square
Spatial dependence control Yes Yes 18270 18270 0.643 0.646
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value Notes: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Foreclosure variables are the number of foreclosures that occurred within two year windows before the sale and within two buffer distances (0–660 ft and 660–1320 ft) of the subject property. Standard errors are clustered at PSA-year level.
only multi-family foreclosures are driving the results in the outer ring with a significant negative price impact of about 3%. Two main points can be drawn from the last three specifications. First, multi-family houses seem to be more likely to be affected by a common spatial shock (as the ‘‘after’’ coefficient is statistically significant for the inner ring). Second, multi-family foreclosures have an impact up to 1320 feet away from the subject property whereas the impact from single-family foreclosure is restricted within a 660 foot radius of the subject property. The final assessment of the spatial proximity hypothesis addresses an issue that relates to the lasting impact of foreclosures. Results from Table 8 indicate that the marginal impacts of both types foreclosure from between one and two years prior to the sale decrease over time. While these estimates do not control for future foreclosures, a persistent effect is evident in the results for multi-family dwellings located within one-eight mile of a subject single-family home.
Table 9 Proximity versus within-submarket effects of foreclosures.
Variable
Semi-log (1)
Semi-log (2)
Elasticity (log–log) (3)
Near before 1 year 0–660 ft Far before 660–1320 ft
0.017⁄ (0.008) 0.002 (0.006)
All 0.015+ (0.008) 0.004 (0.006)
0.026⁄ (0.013) 0.004 (0.011)
Near after 1 year 0–660 ft Far after 660–1320 ft
Controls for potential endogeneity 0.009 0.009 0.015 (0.007) (0.007) (0.012) 0.007 0.006 0.011 (0.005) (0.005) (0.010)
Near before 1 year 0–660 ft Far before 660–1320 ft
0.027 (0.017) 0.029⁄⁄ (0.010)
Near after 1 year 0–660 ft Far after 660–1320 ft
Controls for potential endogeneity 0.029⁄ 0.029⁄ 0.055⁄ (0.014) (0.014) (0.023) 0.005 0.005 0.011 (0.010) (0.010) (0.016) Crowding-in effect 0.001⁄⁄ 0.020⁄ (0.0004) (0.009)
Submarket effect Before 1 year
N R-square
Multi-family only 0.027 0.040 (0.017) (0.027) 0.028⁄⁄ 0.046⁄ (0.010) (0.018)
Yes 18270 0.646
Spatial dependence control Yes Yes 18270 18270 0.646 0.646
+p < 0.10, ⁄p < 0.05, ⁄⁄p < 0.01; p values are based on two-tailed t-test critical value. Notes: Dependent variable is the log of the sales price of single-family properties. All models include a set of property characteristics (see Table 6) and PSA-year fixed effects. Additional quarterly dummy variables control for seasonality of housing price. Foreclosure variables are the number of foreclosures that occurred within a year of the sale and within two buffer distances (0–660 ft and 660–1320 ft) of the subject property. It also adds the number of foreclosures that occurred within 1 year before the sale at the submarket level. Standard errors are clustered at PSA-year level.
4.2. Test of the crowding-in hypothesis
column (6) from Table 7; columns (2) and (3) include the number of foreclosures occurring within the relevant submarket for the year prior to the sale of the subject property. In a specification that already includes controls for nearby foreclosures, the submarket variable captures the pure supply effect of (only) distant foreclosures within the same submarket. The final column reports the elasticity coefficient estimates for the purpose of comparing different types of impacts in a standardized form.11 The results of column (2) indicate that each distant foreclosure that occurred within a year before the sale of the subject property lowered the sales price of a single-family home within the same submarket by about 0.1%.12 Although the
In the context of a heterogeneous stock of housing, an alternative view of housing submarkets suggests that there will be additional effects of distant foreclosures on the sales price of homes as foreclosure leads to crowding-in of excess dwelling units within an entire submarket. Table 9 includes the key results of the test of the submarket hypothesis. Column (1) reproduces the results found in
11 The number of foreclosures at the submarket level is expected to be much higher, so the associated coefficient is going to be much lower. Elasticity coefficients are reported so that it will be easier to see which type of foreclosure is really driving the results. 12 This submarket impact is robust across different sets of submarket delineated using alternative neighborhood variables such as area unemployment rate, median household income, mean travel time to work, percentage of people having bachelor degree etc.
A. Biswas / Journal of Housing Economics 21 (2012) 235–245
effect seems small, the elasticity measure indicates that foreclosures at the submarket level have substantial impact on the sales price of the subject property within the same submarket. Column 3 indicates that a one percent increase in submarket foreclosures leads to a 0.02% decrease in the predicted sales price.13 An interesting way of assessing the quantitative significance of these results is to evaluate the impact of the surge in foreclosures that began in 2006 and continued in 2007. Consider the change in the average foreclosure measure between two years: 2005, a year of low foreclosure activity at the height of the housing bubble, and 2007, which experienced the largest number of foreclosures most likely observed since the 1930s. The increase in foreclosures from 2005 to 2007 on average reduces the predicted sales price of a 2005 single-family house by 5.5% by 2007.14 For example, given the average sales price of $252,425 in 2005, a 5.5% reduction corresponds to the loss of almost $14,000 by 2007. This predicted impact of changes in the housing market is associated only with impacts of foreclosures on the sales price. 5. Conclusion This study offers a new way of conceptualizing and then estimating the potential spillover effects of foreclosures on the sales prices of non-foreclosed single-family houses. Using traditional time-distance intervals, the results confirm the occurrence of substantial additional spillover effects from multi-family foreclosures on the sales prices of neighboring single-family properties. While the impact of single-family foreclosure is relevant within 660 feet of a sale, the discount is substantial (about 3%) for multi-family foreclosures occurring in the outer ring (660–1320 feet), and these multi-family effects do persist over time. Such persistence suggests that foreclosures do not just temporarily depress the nearby property prices but they may have negative physical impact on the immediate surroundings which lasts for some time. When a more expansive definition of a housing submarket is used, results suggest that a distant foreclosure within the entire submarket also has a substantial negative impact on property sales prices. The evidence suggests that the total estimated impacts of all types of foreclosure are much higher than what is reported in the current literature. The results of this study may offer some useful policy implications against the backdrop of recent surge in foreclosure. Evidence calls for government intervention as foreclosure impact is not limited only to those who are directly involved but also the entire community faces the consequences. Especially in the context of old central city
13 This paper also estimates the foreclosure impacts by three quality submarkets. Impacts are found to be limited to high and low submarket. In the high submarket, the nearby foreclosures of both types of housing matter, but not the foreclosures at the submarket level. In contrast, the nearby multi-family foreclosures along with foreclosures at the submarket level have significant impact on the sales prices in the low submarket. Results are available from the author upon request. 14 Estimation uses coefficient measures from column (2) of Table 9.
245
neighborhood with a heterogeneous stock of housing, policymakers should consider the possibility of a heightened impact of multi-family foreclosures. For the policy makers, it will be also worth considering the possibility of additional impacts due to within-submarket effects of distant foreclosures in the context of city neighborhood like Worcester where the phases of housing development took place at different points in time unlike the relatively new suburbs. Acknowledgment The author would like to thank John Brown, Tom Mayock, Junfu Zhang, Wayne Gray, Rahul Rakshit, and Aniruddha Mitra for helpful comments. Special thanks to John Brown and Jacqueline Geoghegan for providing the dataset. References Bourassa, Steven.C., Hamelink, Foort., Hoesli, Martin., MacGregor, Bryan.D., 1999. Defining housing submarkets. Journal of Housing Economics. 8, 160–183. Bourassa, S., Hoesli, M., Peng, V.S., 2003. Do housing submarkets really matter? Journal of Housing Economics. 12, 12–18. Campbell, John.Y., Giglio, Stefano., Pathak, Parag., 2011. Forced sales and housing prices. American Economic Review 101, 2108–2131. Chen, Zhuo., Cho, Seong.-Hoon., Poudyal, Neelam., Roberts, Ronald.K., 2009. Forecasting housing prices under different market segmentation assumptions. Urban Studies 46, 167. Goodman, Allen.C., Thibodeau, Thomas.G., 1998. Housing market segmentation. Journal of Housing Economics 7, 121–143. Grigsby, William.G., 1963. Housing Markets and Public Policy. University of Pensylvania Press, Philadelphia. Grigsby, William.G., Baratz, Morton., Galster, George.C., Maclenna, Duncan., 1987. The dynamics of neighborhood change and decline. Progress in Planning 28, 1–76. Immergluck, Dan., Smith, Geoff., 2006. The external costs of foreclosure: the impact of single family mortgage foreclosures on property values. Housing Policy Debate 17, 57–79. Leonard, Tammy., Murdoch, James.C., 2009. The neighborhood effects of foreclosure. Journal of Geographical Systems 11, 317–332. Lesage, James.P., 1997. Regression analysis of spatial data. The Journal of Regional Analysis & Policy 27, 83–94. Lin, Zhenguo., Rosenblatt, Eric., Yao, Vincent.W., 2009. Spillover effects of foreclosures on neighborhood property values. The Journal of Real Estate Finance and Economics 38, 387–407. Palm, Risa, 1978. Spatial segmentation of the urban housing market. Economic Geography 54, 210–221. Patton, Myles., Mcerlean, Seamus., 2003. Spatial effects within the agricultural land market in Northen Ireland. Journal of Agricultural Economics 54, 35–54. Rogers, William H., Winter, William, 2009. The impact of foreclosures on neighboring housing sales. Journal of Real Estate Research 31, 455– 479. Schuetz, Jenny., Been, Vicki., Ellen, Ingrid.Gould., 2008. Neighborhood effects of concentrated mortgage foreclosures. Journal of Housing Economics 17, 306–319. Watkins, Craig.A., 2001. The definition and identification of housing submarkets. Environment and Planning A 33, 2235–2254.