Concentrated Foreclosure Activity and Distressed Properties in New York City

Abstract

Recent research suggests that foreclosures have negative effects on homeowners and neighborhoods. We examine the association between concentrated foreclosure activity and the risk of a property with a foreclosure filing being scheduled for foreclosure auction in New York City. Controlling for individual property and sociodemographic characteristics of the neighborhood, being located in a tract with a high number of auctions following the subject property’s own foreclosure filing is associated with a significantly higher probability of scheduled foreclosure auction for the subject property. Concentration of foreclosure filings prior to the subject property’s own foreclosure filing is associated with a lower probability of scheduled foreclosure auction. Concentrated foreclosure auctions in the tract prior to a subject property’s own filing is not significantly associated with the probability of scheduled foreclosure auction. The implications for geographic targeting of foreclosure policy interventions are discussed.

Keywords

foreclosure New York City multilevel modeling

From March 2011 through June 2012, nearly 3 million homeowners entered foreclosure across the United States, with nearly 1.1 million foreclosure sales during this same period (HOPE NOW 2012). This level of foreclosure activity signals the need to understand both the causes and consequences of foreclosure, particularly if such knowledge can help guide public investments and ensure that foreclosure mitigation efforts are effectively targeted. Place-based strategies, such as the federal Neighborhood Stabilization Program (NSP), use geographic risk factors to focus services in areas where foreclosure is likely to not only affect at-risk homeowners, but also serve to destabilize neighborhoods through such documented spillover effects as reduced property values or increased crime. Of primary interest here is if such measures are likely to reach those most at risk of foreclosure auction, recognizing that even among those who receive a foreclosure filing, many are able to avoid foreclosure auction. Using data from New York City, we examine whether concentrated foreclosure activity is associated with neighborhood conditions that increase the risk of future auctions and whether this lends support for particular forms of place-based intervention.

As former Assistant Secretary for Financial Stability at the U.S. Treasury Herbert Allison (2010) testified,

Although all stages of foreclosure are unpleasant, forced sale through foreclosure is the stage of the process that dislocates the family, disrupts the community, and destabilizes the local housing market. Importantly, this measure is rarely the focus of public discussion. Most attention is given to the number of foreclosure “starts” or to the total number of foreclosure referral notices sent to borrowers. In any given year, foreclosure starts greatly outnumber foreclosure sales . . . In the context of current policy, which focuses foreclosure avoidance efforts along the entire foreclosure process even if the foreclosure process has begun, foreclosure starts have become less relevant predictors of future foreclosure sales.

Policy makers attempting to “get ahead of the curve” often focus solely on the number of foreclosure filings without consideration for foreclosure auctions, conflate these different stages of the foreclosure process, or inappropriately rely on foreclosure filing as proxy for auction. These two phases of the foreclosure process may have substantially different impacts on neighborhoods and communities; foreclosure filings may not have a significant effect on the likelihood of other distressed properties reaching auction, whereas foreclosure auctions could be positively associated with the risk of auction for other nearby properties.¹ Increasing our knowledge of which properties are at greatest risk of being foreclosed upon, how they can be identified, and how neighborhood characteristics, including concentrated levels of both foreclosure filings and foreclosure auctions, are associated with risk, is particularly important for developing timely and effective foreclosure intervention strategies that are targeted to specific geographies that have experienced a high volume of foreclosure activity.²

New York presents an especially useful case in which to test our hypotheses about the association between different types of foreclosure activity and the outcome of a single financially distressed property in part because of the length of time between different stages in the process. Figure 1 provides a simplified model of the foreclosure process in New York (see Schuetz, Been, and Ellen (2008) for a thorough description of the legal foreclosure process in New York State). New York has a long process relative to many other states: The median time from foreclosure filing to scheduled auction among the foreclosure filings in New York City from the first quarter of 2007 (six months before our sample) was 365 days; the mean was 405 days.³ The current analysis focuses on the subset of foreclosure filings that are scheduled for foreclosure auction (those that reach both time points of public notice, point A, and point B in Figure 1) and seeks to determine factors that are associated with being scheduled for auction among properties that entered the foreclosure process in New York City. Outcomes for borrowers with a foreclosure filing that do not result in a scheduled foreclosure auction may be positive or negative—such borrowers may restructure or refinance debt and retain ownership, sell the property in a short sale, or turn over the deed in lieu of foreclosure. Even after a foreclosure auction has been scheduled, the borrower and lender may reach an agreement to satisfy the outstanding loan, and the scheduled auction will be canceled. Outcomes that occur between points B and C in Figure 1 are beyond the scope of the present analysis.

Although New York City’s housing market has fared better than many other markets, it has certainly not been immune to financial distress among its homeowner population: In 2007 and 2008, there were nearly 14,000 foreclosure filings annually, double the number in 2004. In 2009, the number increased to over 20,000. The number of foreclosure auctions, however, is much smaller. The annual count of foreclosure auctions varied between 3,000 and 4,500 in 2007, 2008, and 2009.

Foreclosure activity in New York City is highly concentrated in a few of the city’s 55 sub-borough areas (SBAs).⁴ From 2007 through 2009, the time period we analyze in this article, over half of the city’s foreclosure filings occurred in just nine SBAs. Foreclosure auctions were even more concentrated: During this same period over half of the city’s auctions took place in just six SBAs. Not all areas with the highest concentration of foreclosure auctions, however, are among the areas with the highest concentration of foreclosure filings; the two types of foreclosure activity are not concentrated in the same manner.⁵ This pattern makes it particularly important to assess the association of different types of concentrated foreclosure activity with the outcomes of other financially distressed properties to determine where interventions can be targeted to achieve the greatest marginal and/or cumulative result.

Foreclosure processes vary widely across geographies and are specific to local legal and market conditions. Comparing findings on the foreclosure experience of multiple geographies allows the research community to begin to distinguish between characteristics of foreclosure that are constant across place and time and those that are situational. In examining the process in New York, we must look beyond a decline in housing prices to assess how high cost lending concentrated in minority neighborhoods and other neighborhood characteristics may be associated with foreclosure in a predominantly rental housing market that did not experience substantial price declines (S&P/Case-Shiller Home Price Indices 2011).

We hypothesize that the number of foreclosure auctions surrounding a property that has received a foreclosure filing is positively associated with the likelihood of that property itself reaching auction, net of other factors. Auctioned properties may have negative spillover effects on their neighbors because they add supply to the market, reducing housing prices and neighbors’ ability to sell their properties or refinance. Unlike auctions, foreclosure filings do not necessarily involve a transfer of ownership and may not have as strong a negative effect as auctions may have. Yet owners of properties that have received a foreclosure filing may also defer maintenance, worsening the quality of the housing stock, and discouraging potential purchasers. A homeowner in financial distress who is unable to sell his or her property has fewer options to cure the delinquency and avoid foreclosure auction. A high volume of foreclosure auctions may also alter the choices that lenders make in deciding whether or not to modify a loan or follow through with a foreclosure in a saturated market. Regardless of the particular mechanisms, a feedback loop is established whereby the number of current foreclosure auctions is associated with increased risk of future foreclosure auctions, above and beyond property or neighborhood characteristics alone. We test this hypothesis using multilevel logit models that estimate the association between neighboring foreclosure activity and the probability of auction, controlling for a variety of property and neighborhood characteristics.

The remainder of the article is organized as follows. The section “Background” of the article synthesizes and discusses the existing literature on both the causes and consequences of foreclosure. We then present our data and methodology and explain our choice of models in the section “Data and Method.” The section “Findings and Discussion” presents findings, section “Implications” discusses implications of our findings, and conclusions follow in the section “Future Research.”

Background

We draw upon two somewhat distinct bodies of literature for this article. The first seeks to identify the factors that affect the various outcomes of mortgage loans and include borrower, loan, lender, and other variables. The second set of studies estimates the effects of foreclosures on neighborhoods and communities above and beyond their effect on individual households. Taken together, there is substantial evidence that foreclosure auctions may act as both cause and effect. Within the broad categories of contributor and consequence, previous research has focused on two areas: individual borrower and loan-level factors that contribute to foreclosure, and the impacts of foreclosure on neighborhoods and broader geographies. While there are some studies that have considered geographic patterning of foreclosures, previous research has failed to adequately account for all of the community sources of risk for homeowners in distress. We contribute to the literature by examining the association between neighborhood foreclosure activity and individual-level risk.

Causes of Foreclosure

A growing body of work analyzes the outcomes of mortgage loans, such as prepayment, delinquency, default, and foreclosure, and evaluates different phases in the life of the loan. The reasons why loans enter the foreclosure process—receive a foreclosure filing—are not explored in our analysis, but the research reviewed in this section highlights the association between loan, borrower, locational characteristics, and legal frameworks and foreclosure activity. Existing research has identified mortgage characteristics that are associated with negative loan outcomes. Adjustable-rate and broker-originated mortgages, high loan-to-value ratios, and investor-owners are positively associated with loan default (Berkovec et al. 1994; Calhoun and Deng 2002; Ding et al. 2011; Strom and Reader 2013). Among properties with subprime loans, Quercia, Stegman, and Davis (2007) find that prepayment penalties, balloon loans, and low/no-documentation loans are positively associated with entering foreclosure (receiving a filing).

Locational characteristics add explanatory power to analyses that consider the impact of borrower and loan characteristics on mortgage outcomes. Studies examining local housing market characteristics find associations between neighborhood-level lending and housing values and foreclosure activity. In Chicago from 1995 to 2002, Immergluck and Smith (2005) find that neighborhood-level subprime lending is positively associated with foreclosure filings, while in the rapidly growing, high house price appreciation context of Phoenix from 2000 to 2007, Agarwal et al. (2012) show that concentrations of low/no-documentation and hybrid adjustable rate loans are positively associated with foreclosure. Foreclosure of a defaulted loan is more likely than reinstatement or sale in areas with housing price depreciation while the loan is in default (Ambrose and Capone 1998). Goodman and Smith (2010) find that housing values are positively associated with neighborhood foreclosure rates, but negatively associated with the density of real estate owned (REO) properties. In a study of Cuyahoga County, Ohio, Coulton et al. (2008) find that the borrower’s race and whether the loan was subprime have the greatest statistical impact on the likelihood that a loan receives a foreclosure filing; however, neighborhood characteristics including higher neighborhood poverty rates, lower median sales prices, and each additional foreclosure filing within a 500 foot buffer of the property also increase the likelihood that a loan results in a foreclosure filing. Providing more evidence that neighborhood housing market characteristics predict foreclosure, Chan et al. (2013) show that concentrated lis pendens filings and REO stock are positively associated with default in New York.

Population characteristics such as race/ethnicity, employment, and income are also associated with foreclosure activity. Berkovec et al. (1994) find that the proportion of neighborhood population that is Asian or Hispanic is negatively associated with the likelihood of default (lender foreclosure and deed-in-lieu of foreclosure), while proportion of the population that is Black is not associated with likelihood of default, after controlling for a wide range of homeowner and loan characteristics. Other studies, however, do find that the proportion of the population that is Black is positively associated with foreclosure rates (Chan et al. 2013; Molina 2012; Rugh, Albright, and Massey 2015) and REO (Goodman and Smith 2010). Unemployment rate at the town level and percentage minority at the zip code level are positively associated with properties sold at foreclosure auction, and median income is negatively associated with foreclosure deeds (Foote, Gerardi, and Willen 2008).

In addition to loan and neighborhood demographic and housing market characteristics, state foreclosure laws are also associated with foreclosure activity. The length of the foreclosure process matters: Foreclosure of a defaulted loan is less likely than reinstatement in states with short (under three months) and long (over six months) foreclosure process (Ambrose and Capone 1998). Controlling for zip- and metropolitan area-level loan and demographic variables, state-level legislation that increases costs of foreclosure to lenders and places stringent regulations on predatory lending is associated with lower rates of foreclosure (Goodman and Smith 2010).

This research shows the variation in factors hypothesized to increase the likelihood of foreclosure. Individual, property, and neighborhood characteristics and legal frameworks are associated with foreclosure activity. For our analysis, it is important to acknowledge these other mechanisms that potentially mediate the association between concentrated foreclosure as a neighborhood characteristic and a financially distressed property’s likelihood of reaching auction.

Effects of Foreclosure

There is substantial evidence that foreclosures have negative spillover effects on neighboring properties and have the potential to disrupt local housing markets. In this analysis, we examine the association between foreclosure activity and the likelihood of a property with a foreclosure filing being scheduled for foreclosure auction. Much of the existing research, including what we review in this section, focuses on foreclosure’s effects on property values.

Immergluck and Smith (2006) find that each conventional completed foreclosure within an eighth-mile of a single-family home corresponds to a 0.9% decline in value; this effect diminishes with distance from the foreclosed home. Schuetz, Been, and Ellen (2008) expand upon Immergluck and Smith (2006) by controlling for additional neighborhood characteristics, and observe the impact of foreclosure filings, rather than foreclosure sales. Schuetz, Been, and Ellen (2008) observe a smaller decline in value than Immergluck and Smith (2006), which they attribute to New York City’s rapidly appreciating housing market. In addition, they find evidence of a threshold effect where one to two foreclosure filings within 250 to 500 feet have no significant effect on sales prices, but three or more filings do have a negative effect; within 500 to 1,000 feet, the threshold is six filings.

Lin, Rosenblatt, and Yao (2009) include time and distance measures in their models of the effect of foreclosure sales on neighborhood property values in Chicago and find that each foreclosure sale within five years and 0.9 kilometers of a nonforeclosure sale can decrease neighborhood property values by up to 8.7%; the effect diminishes at greater time and distance. This is a substantially higher estimate than what Immergluck and Smith (2006) find in Chicago, which may be attributable to the inclusion of time in their model, as foreclosure sales closest in time to nonforeclosure sales have the biggest impact. Rogers and Winter (2009) perform a similar analysis of housing price effects of foreclosure sales using data from St. Louis County and estimate the marginal impact of foreclosure sale on neighboring house prices to be 1% or less, similar to Immergluck and Smith (2006), and the marginal impact of foreclosure sales on housing prices declines as foreclosure sales increase. Following Lin, Rosenblatt, and Yao (2009), Kobie and Lee (2011) include time and distance in their models of the effect of foreclosure activity on sales prices, but specify distance using a face block measure of proximity. In their analysis of Cuyahoga County, Ohio, they find that foreclosure filings within the face block more than a year old are associated with a 1.7% reduction in sales prices. Kobie and Lee distinguish between foreclosure filings and sales, as Phillips and VanderHoff (2004) recommend, finding that foreclosed properties sold at sheriff sales are associated with a sales price reduction of nearly 3%. Sales price discounts resulting from neighboring foreclosures, however, may not be uniform across all types of neighborhoods (Whitaker and Fitzpatrick 2013).

Using national data, Harding, Rosenblatt, and Yao (2009) find a “contagion discount”: The decline in sales price associated with proximity to a home sold at foreclosure auction grows from the onset of financial distress through the foreclosure sale and then stabilizes. The authors suggest that it is likely deferred maintenance and disinvestment incurred during the foreclosure process that affects the value of surrounding properties. Lambie-Hanson (2015) provides evidence for this proposed mechanism, finding that properties in financial distress are subject to more complaints about their physical condition than comparable properties not in default, and that the likelihood of a property receiving a complaint or code violation increases throughout delinquency, peaking while the property is REO. In a qualitative study, neighbors of foreclosed properties report neglect and trash accumulation that could dampen surrounding sales prices (Graves 2012). Comparing contemporaneous home sales in the same small geographic area across 15 metropolitan areas, Gerardi et al. (2015) find small spillover effects of distressed properties on sales prices, although these effects diminish over time and are sensitive to the condition of REO properties. Based on these results, it seems likely that a foreclosure filing that cures before reaching foreclosure auction will have a lesser impact on surrounding properties. Schuetz, Been, and Ellen’s (2008) findings may show a smaller effect of foreclosure filings if the effect is averaged across many foreclosure filings that cure (which, if sold, could have varying effects on surrounding property values depending on the type of owner who takes possession of the property) and others that terminate in foreclosure auction (where sources of neighborhood distress such as vacancy and/or deferred investment may be more likely (Lambie-Hanson 2015)). We posit that properties with foreclosure filings are not necessarily visibly distressed and may or may not be available for sale; thus, foreclosure filings in and of themselves may have a smaller effect on the local market than other types of foreclosure activity such as auctions.

Research shows substantial variation in measurement techniques used to estimate foreclosure’s effect on housing sales prices by including time, distance, and phase of the foreclosure process. These articles also vary by year and geographic focus (and, thus, housing market characteristics), as some include single cities or counties and others use national data. Most of these articles estimate sales price effects similar in magnitude to Immergluck and Smith’s (2006) seminal work. We contribute to this body of research by considering a different dependent variable: We estimate the association between foreclosure activity and the probability of other financially distressed properties (those that have received foreclosure filings) being scheduled for foreclosure auction. In addition, we include time measurement in multilevel models. Although much research focuses on estimating sales price effects of foreclosure activity, measuring the probability of auction is important in itself as an indication of household and neighborhood instability.

Data and Method

We investigate the differential association between varying types of concentrated foreclosure activity at a tract level and a property’s risk of being scheduled for foreclosure auction. This article relies on tracking a sample of all residential foreclosure filings filed across New York City’s five boroughs from the third quarter of 2007 (July 1, 2007 to September 30, 2007) through October 2009. Our foreclosure filing data come from PropertyTrac™, and scheduled foreclosure auction data come from PropertyShark™; these private vendors collect data from New York City’s clerk recorders.⁶ Before assembling our dataset, we eliminated any filings that were not a result of mortgage or tax liens (e.g., liens from mechanics or condominium/cooperative boards) as well as properties with incomplete identification or geographical information. Foreclosure filings from July 1, 2007 to September 30, 2007 were linked to one or more auctions scheduled from July 1, 2007 to October 31, 2009.⁷ Our outcome is whether a foreclosure filing was linked explicitly to one foreclosure auction by index number (dichotomous for whether or not the foreclosure filing resulted in a scheduled auction).⁸ Our final analytic sample consists of 4,154 foreclosure filings, of which 1,004 resulted in auction.

We use random-intercept multilevel logit models to estimate the association between property- and census tract-level characteristics and the likelihood of being scheduled for auction resulting from a foreclosure filing. Our property-level, or level 1, equation is as follows:

y_{i j} = β_{0 j} + β_{1 j} X_{i j} + r_{i j},

where $X_{i j}$ is a vector of property-level characteristics including building age (coded as a dummy, where 1 indicates built 1969 or later, and 0 indicates earlier), number of residential units in the building, and the number of emergency housing code violations in the year prior to the foreclosure filing.⁹ We include number of emergency housing code violations in the year prior to the foreclosure filing as a proxy for physical distress and potential neglect that may alter the ability of an owner to short-sell or otherwise cure a property prior to auction.¹⁰

We estimate random intercepts using our neighborhood, or level 2, equation:

β_{0 j} = γ_{00} + γ_{01} X_{j} + u_{0 j},

where $X_{j}$ is a vector of census tract-level characteristics including demographic and socioeconomic (compositional factors) and housing market conditions (contextual factors). Specifically, we include race/ethnicity (percentage of residents who are White), nativity status (percentage foreign born), median household income, the total number of residential units in the tract, the share of high cost residential mortgage loans originated in 2005, the total number of residential property transfers from May through November 2007, and squared terms for high cost lending, percentage White, median household income, and percentage foreign born.

The race, nativity, income, and residential unit data are from the 2000 Decennial Census. The 2005 Home Mortgage Disclosure Act (HMDA) is the source for each tract’s rate of high cost lending (our proxy for subprime lending).¹¹ Property transfers data come from the New York City Department of Finance, which maintains records on transactions. We sum property transactions in each tract from May 2007 through November 2007, choosing this time period to control for general market conditions in the neighborhood at the time of our sample filings.¹² When combined, equations (1) and (2) form the basis for our estimation strategy.

Recent studies of mortgage default have used duration and proportional hazard models to estimate the probability of various foreclosure activity, controlling for a combination of borrower, loan, property, and/or neighborhood characteristics (Coulton et al. 2008; Deng, Quigley, and Van Order 2000; Foote, Gerardi, and Willen 2008; Gerardi and Willen 2009). As with other single-level analytic strategies, such as ordinary least squares (OLS), hazard analysis assumes that subjects (properties) behave independently, overlooking the potential for subjects within a particular context (neighborhood) to behave more alike (Barber et al. 2000). If this assumption is violated, then the standard errors may be underestimated, resulting in Type I errors (Diez Roux 2000).

Our hierarchical logit estimation strategy allows us to nest property-level observations within a higher level geography, an advantage over proportional hazard models because hazard models allow observations to be clustered, but not nested, thus potentially failing the assumption of independence.¹³ Although hierarchical logit models do not give us the same flexibility regarding censored outcomes as do hazard models, based on exploratory analysis of a complementary sample, we find that our data are at very low risk of being right-censored.¹⁴ The foreclosure filings and scheduled foreclosure auctions we analyze here occurred prior to the implementation of the federal government’s Home Affordable Modification Program (HAMP). Our time frame ensures that HAMP’s assistance to homeowners does not affect our sample. In addition, we are less interested in the length of survival (the analytic goal of proportional hazard models), than the association between concentrated neighborhood-level foreclosure activity and our property-level outcome and, therefore, select an estimation strategy that best addresses our analytic aims and multilevel data structure.

Our data do not allow us to test many of the hypothesized individual-level factors that may lead to increased risk of auction (such as loan-to-value ratio, credit score, homeowner income, or loan product). Much foreclosure research faces similar limitations, particularly because of the lack of publicly available data (for discussion of data availability, see Coulton et al. 2008; Newman 2010). We acknowledge this, while stressing that our primary analytic goal is to estimate the association between concentrated foreclosure activity and the probability that a foreclosure filing results in scheduled auction, controlling for relevant property and neighborhood characteristics. Although others have identified negative spillover effects of foreclosure activity that may interfere with a homeowner’s ability to avert auction later on (such as declining home values or increased social and physical disorder that deters potential buyers), there has been limited research on how nearby foreclosure activity affects the risk of future foreclosure auction and even less that accounts for multiple stages of foreclosure activity within the same analysis. Estimating the associations between foreclosure activity and the risk of other financially distressed properties being scheduled for foreclosure auction follows and informs current policy initiatives that focus on stabilizing neighborhoods and is, therefore, a valuable complementary research strategy to the articles cited above.

New York State’s judicial foreclosure process ensures a long lag between a foreclosure filing and auction (about 12 months). We find that the majority of properties that enter the foreclosure process do not sell at auction and that the rate at which foreclosure filings are scheduled for auction varies substantially across subareas within New York City. Based on a sample of foreclosure filings in 2007, we find that less than 30% of foreclosure filings in New York City are scheduled for auction within two years of the filing, and only 13% of properties with foreclosure filings are sold at auction. This is corroborated by a report by the Furman Center for Real Estate and Urban Policy that finds that 14% of properties that received a foreclosure filing in 2007 were sold at auction by the end of June 2009 (Furman Center for Real Estate and Urban Policy 2010). These findings suggest that it is inappropriate to treat the density of foreclosure filings and foreclosure auctions in an area equally, and perhaps more so to use one as a proxy for the other.

Our estimation strategy distinguishes between two types of foreclosure activity, foreclosure filings and scheduled foreclosure auctions, to improve the precision with which we are able to measure our hypothesized association—namely, increased risk of auction for surrounding properties. We also add a temporal dimension, distinguishing between activity that occurs prior to the subject property’s own foreclosure filing and that which occurs after filing, as these contextual factors are likely to alter the risk of auction in substantively different ways depending on the particular phase of the foreclosure process at which they are experienced.

To achieve this, we construct group mean centered measures of concentrated foreclosure activity surrounding the property using the scheduled foreclosure auction and foreclosure filing data described above. We sum the number of scheduled foreclosure auctions and foreclosure filings occurring on other properties in the same census tract as the subject property in the 360 days before and 360 days after the subject property’s own foreclosure filing, respectively, excluding the scheduled foreclosure auction of the property itself when appropriate. Each of the four counts—foreclosure filings on other properties in the 360 days before the subject property’s foreclosure filing, scheduled foreclosure auctions on other properties in the 360 days before the subject filing, foreclosure filings on other properties in the 360 days after the subject property’s foreclosure filing, and scheduled foreclosure auctions on other properties in the 360 days after the subject filing—is group centered at the tract level (individual counts are subtracted from the mean number of foreclosure filings or scheduled foreclosure auctions within the 360 days preceding or following the sampled filing(s) in a particular tract) to provide a relative measure of the property’s exposure to particular levels of concentrated activity. Group-centered terms are included in level 1 and the average of each of our four individual property-level measures of concentrated foreclosure activity within a given tract are included in level 2, giving us the following full model:

\begin{array}{l} y_{i j} = γ_{00} + γ_{01} X_{j} + γ_{10} (A_{i j} - {\bar{A}}_{. j}) + γ_{20} (B_{i j} - {\bar{B}}_{. j}) \\ + γ_{30} (C_{i j} - {\bar{C}}_{. j}) + γ_{40} (D_{i j} - {\bar{D}}_{. j}) + γ_{50} X_{i j} + u_{0 j} + r_{0 j}, \end{array}

where $A_{i j} - {\bar{A}}_{\cdot j}$ is our group-centered term for the number of scheduled auctions in the tract 360 days following the subject filing, $B_{i j} - {\bar{B}}_{\cdot j}$ is our group-centered term for the number of scheduled auctions in the tract 360 days prior to the subject filing, $C_{i j} - {\bar{C}}_{\cdot j}$ is our group-centered term for the number of foreclosure filings in the tract 360 days following the subject filing, and $D_{i j} - {\bar{D}}_{\cdot j}$ is our group-centered term for the number of foreclosure filings in the tract 360 days prior to the subject filing. ${\bar{A}}_{\cdot j},$ ${\bar{B}}_{\cdot j},$ ${\bar{C}}_{\cdot j},$ and ${\bar{D}}_{\cdot j}$ are our tract-level means for each of these four counts, respectively, and represent the primary predictors of interest for estimating the association between concentrated foreclosure activity and likelihood of being scheduled for auction.¹⁵

Descriptive statistics for the level 2 variables are shown in Table 1. The first panel shows the mean and standard deviation for the group mean measures of foreclosure concentration at the neighborhood level. The neighborhood-level demographic, socioeconomic, and housing market variables are presented in the second panel. Not every New York City census tract contained a property receiving a foreclosure filing in the third quarter of 2007 (our sample period); weighted means for only those tracts included in the sample are shown in panel 2. In general, neighborhoods in our sample have more home sales, have a lower proportion of White non-Hispanic residents, and have a greater share of high cost lending than tracts citywide. These differences are in line with previous research on the geographic characteristics associated with foreclosure activity (Immergluck and Smith 2005; Newman and Wyly 2004).

Table 1.

Descriptive Statistics.

Panel 1. Foreclosure Activity Variables	Full Sample (n = 4,154)
Panel 1. Foreclosure Activity Variables	M	SD
Average number of prior scheduled auctions in tract	5.3	6.0
Average number of prior foreclosure filings in tract	18.0	14.7
Average number of post scheduled auctions in tract	7.9	9.8
Average number of post foreclosure filings in tract	22.9	18.5
Panel 2. Neighborhood Variables	Full Sample^a (n = 4,154)
Panel 2. Neighborhood Variables	M	SD
Residential units in neighborhood	1,420.5	923.8
Home sales in neighborhood (May−Nov 2007)	21.9	48.7
% of population that is White	27.3	28.8
% of population that is foreign born	33.7	14.7
Median household income (1999 $)	40,054.7	14,896.1
% high cost loans	41.5	15.9

Nested in 1,177 census tracts.

Findings and Discussion

Odds ratios and standard errors for all models are shown in Table 2. Testing the association between property characteristics and neighborhood demographic and socioeconomic characteristics and the likelihood of scheduled foreclosure auction (model 1), we find that buildings constructed in 1969 or later have 26% greater odds of reaching scheduled auction than buildings constructed before 1969.¹⁶ High cost lending, the share of the population that is White, and median household income are all positively associated with the risk of scheduled auction. The share of the population that is foreign born is negatively associated with the risk of scheduled auction.

Table 2.

Odds Ratio Estimates of Foreclosure.

Independent Variables	1	2	3	4
Property-level variables
Building age dummy (new = 1)	1.260* (0.163)	1.211 (0.147)	1.217 (0.156)	1.231 (0.157)
Group-centered post scheduled auctions		0.747*** (0.031)	0.806*** (0.029)	0.746*** (0.031)
Group-centered prior scheduled auctions		0.859** (0.053)		0.859** (0.053)
Group-centered prior foreclosure filings		1.170*** (0.041)		1.170*** (0.041)
Group-centered post foreclosure filings		1.118*** (0.039)		1.117*** (0.039)
Neighborhood-level variables
Average number of post scheduled auctions in tract		1.260*** (0.024)	1.180*** (0.021)	1.219*** (0.023)
Average number of post scheduled auctions in tract²		0.997*** (0.000)	0.998*** (0.000)	0.998*** (0.000)
Average number of prior scheduled auctions in tract		1.015 (0.017)		1.028* (0.017)
Average number of prior foreclosure filings in tract		0.931*** (0.011)		0.944*** (0.011)
Average number of post foreclosure filings in tract		1.006 (0.008)		1.000 (0.009)
High cost lending	1.046*** (0.017)		1.019 (0.016)	1.036** (0.017)
High cost lending²	1.000 (0.000)		1.000 (0.000)	1.000 (0.000)
% of population that is White	1.031*** (0.008)		1.031*** (0.008)	1.020*** (0.008)
% of population that is White²	1.000*** (0.000)		1.000*** (0.000)	1.000* (0.000)
Median household income (1,000s)	1.136*** (0.024)		1.058*** (0.021)	1.067*** (0.021)
Median household income (1,000s)²	0.999*** (0.000)		1.000** (0.000)	0.999** (0.000)
% of population that is foreign born	0.930*** (0.017)		0.954*** (0.017)	0.952*** (0.016)
% of population that is foreign born²	1.001*** (0.000)		1.001*** (0.000)	1.001*** (0.000)
Residential units in tract	1.000 (0.000)		1.000 (0.000)	1.000 (0.000)
Home sales in tract	1.001 (0.002)		0.988** (0.005)	0.997 (0.005)
Constant	0.00407*** (0.003)	0.168*** (0.018)	0.0240*** (0.015)	0.0238*** (0.015)
Intraclass correlation	.266	.177	.181	.158
Observations	3,993	4,013	3,993	3,993
Number of groups	1,177	1,190	1,177	1,177

Note. Standard errors in parentheses.

p < .1. **p < .05. ***p < .01.

Model 2 examines the association between our foreclosure concentration variables and scheduled foreclosure auction, distinguishing between both types of activity (foreclosure filings vs. scheduled auctions) as well as timing (before vs. after the subject foreclosure filing). The odds ratio for the tract average number of scheduled foreclosure auctions in the 360 days after the filing is significant and positive. Holding all other variables at their means, the predicted probability of scheduled auction for a property in a tract with eight scheduled auctions, the mean for our sample, is 24%. The predicted probability of scheduled auction for a property in a tract with nine scheduled auctions is 27%. We find evidence of a threshold in the concentration of tract-level scheduled auctions where, once exceeded, the marginal increase in association between each additional surrounding scheduled auction and the odds of scheduled auction for the subject property is reduced. The odds ratio for the tract average number of foreclosure filings in the 360 days before the subject foreclosure filing is significant and negative, with each additional foreclosure filing associated with a reduction in the odds of scheduled auction of nearly 7%. Neither concentrated number of scheduled auctions prior to or foreclosure filings following the subject foreclosure filing are significantly associated with risk of scheduled auction.

Model 3 considers our focal foreclosure activity measure, the tract average number of surrounding scheduled foreclosure auctions in the 360 days following the subject foreclosure filing, along with key compositional and contextual neighborhood characteristics. These variables control for underlying rates of high cost lending and other characteristics that have been shown to correlate with foreclosure activity. The count of scheduled foreclosure auctions in the 360 days after the subject foreclosure filing is still significant and positively associated with the likelihood of scheduled auction for the subject filing. Holding all other variables at their means, the predicted probability of scheduled auction for a property in a tract with the mean number of scheduled auctions (eight) is 23%. For a property in a tract with nine scheduled auctions, the predicted probability of scheduled auction is 25%. High cost lending is not significant in model 3.

Model 4 includes all property, neighborhood, and foreclosure activity variables. We find the association between neighborhood-level foreclosure filings and the likelihood of the subject property being scheduled for auction to be distinct from the association between neighborhood-level scheduled foreclosure auctions and the likelihood of the subject property being scheduled for auction once we account for the number of foreclosure filings and scheduled foreclosure auctions in the 360 days prior to and after the foreclosure filing of interest. Holding all other variables at their means, the predicted probability of a property in a tract with eight surrounding scheduled auctions being scheduled for foreclosure auction itself is 22%. An increase of one scheduled auction in the surrounding tract, going from eight to nine, is associated with an increase in the predicted probability of scheduled auction to 25%. The number of foreclosure filings in the 360 days prior to the foreclosure filing of interest is negatively associated with the likelihood of scheduled auction, with each additional foreclosure filing in the surrounding census tract associated with a reduction in the property of interest’s likelihood of scheduled auction of approximately 5%. High cost lending has a positive, significant association with the risk of scheduled foreclosure auction in this model, but this association is mediated by the inclusion of all four of our concentrated foreclosure activity variables.

Although the association between the number of scheduled foreclosure auctions on surrounding properties after the foreclosure filing and the probability of a single foreclosure filing resulting in scheduled foreclosure auction is robust, its magnitude differs across models. Figure 2 shows the predicted probability for models 2, 3, and 4, with the solid line representing the full model of property and neighborhood characteristics (model 4).¹⁷ In all three models, the graph shows a clear threshold, with probability of scheduled auction escalating exponentially with each additional scheduled foreclosure auction on a surrounding property in the 360 days after the foreclosure filing up to about 20 scheduled foreclosure auctions in models 2 and 4. For between 20 and 35 surrounding scheduled auctions, the probability of scheduled auction increases with each subsequent scheduled foreclosure auction, but at a much lower rate. Controlling for neighborhood characteristics but excluding other measures of foreclosure concentration produces a decline in probability at the highest levels of foreclosure concentration in model 3. In models 2 and 4, the reduction in probability is not apparent at the highest concentrations, but the predicted probability plateaus at around 80% for neighborhoods with more than 35 scheduled foreclosure auctions.

Figure 3 shows predicted probability of foreclosure based on neighboring foreclosures in the 360 days after filing along with the predicted probability of foreclosure based on neighboring foreclosure filings in the 360 days prior to the subject foreclosure filing (from model 4). The negative association between prior foreclosure filings and probability of scheduled auction shown in Figure 3 explains part of the difference between the model 3 and model 4 curves in Figure 2, where model 3 does not control for prior foreclosure filings. This finding emphasizes the importance of recognizing that different phases in the foreclosure process may be differentially associated with the outcomes of neighboring properties in financial distress. Models that consider these different types of concentrated foreclosure activity in isolation provide biased estimates of their associations with the outcomes of individual properties.

The negative association between nearby foreclosure filings and risk may indicate the impact of foreclosure filings on lender behavior. If, for example, in a specific neighborhood, there are a large number of properties that have already received a foreclosure filing, lenders may be less likely to foreclose on subsequent filings. They may also be reluctant to increase supply and competition for other distressed properties they are trying to sell. This finding merits further investigation that is not addressed here because of data limitations.

These models do not include borrower- and loan-level characteristics that may be associated with the likelihood of a property with a foreclosure filing being scheduled for auction. As explained above, we are limited to publicly available data for this analysis and are most interested in exploring the association between neighborhood-level foreclosure activity and the probability of being scheduled for auction. While imperfect, these publicly available data are the most widely accessible to policy makers and local agents working in foreclosure mitigation services. We recognize that our analysis omits key borrower-level variables. A study of mortgage default in New York City (Chan et al. 2014) includes borrower, loan, property, and neighborhood characteristics and finds a positive association between the neighborhood rate of foreclosure filings and the likelihood that a property with a foreclosure filing results in auction. In separate analyses that examine the association between only foreclosure filings and the likelihood of reaching auction (available upon request), we find a similar positive association, although smaller in magnitude, between neighborhood-level foreclosure filings and the likelihood that a property with a foreclosure filing is scheduled for auction. This is only the case when we exclude other types of foreclosure activity included in our main models. That these findings are in the same direction to those found by Chan, Sharygin, Been, and Haughwout but smaller in magnitude suggests that our results may be robust to the omitted variables and may provide a conservative estimate of the association between neighborhood-level foreclosure activity and likelihood of scheduled auction.

Alternative models show that other property-level characteristics, such as the number of residential units and the number of emergency housing code violations on the property, do not contribute substantially to the model, nor alter the magnitude of our primary associations. Neither of these variables is significant in the property-level or nested models. Neighborhood homeownership rate was also not significantly associated with scheduled foreclosure auction once other neighborhood characteristics were accounted for in the model.

These findings demonstrate the importance of neighborhood foreclosure activity as a factor associated with the risk of foreclosure auction for individual properties. Our analysis differs from those who focus on foreclosure filings’ association with housing price declines (Schuetz, Been, and Ellen 2008), foreclosure filings (Coulton et al. 2008), and foreclosure auction (Chan et al. 2014). We argue that the difference in findings result from the different loan population and foreclosure activity measured in these articles. Chan et al. (2014) are looking at the outcomes of subprime loans whereas our analysis includes all types of loans—prime and subprime—that received foreclosure filings across New York City in the third quarter of 2007. In addition, all three of these articles only include one type of foreclosure activity—foreclosure filings—as an independent variable. Our main argument is that different types of foreclosure activity may be associated with different outcomes for financially distressed properties.

Implications

Policy responses to foreclosure generally focus on two goals: to keep homeowners in their homes and stabilize neighborhoods. These goals are sometimes discussed interchangeably (Swanstrom, Chapple, and Immergluck 2009), but they are achieved using two distinct strategies: Counsel and provide legal services to homeowners to keep them in their homes and/or concentrate resources in the most geographically affected areas to mitigate negative spillover effects (e.g., declining property values, crime) from foreclosures. Our findings corroborate the association between concentrated foreclosures and property-level outcomes that may affect local market conditions and justify geographic targeting to ensure the effectiveness and efficiency of programming. These findings suggest, however, that the impacts of foreclosure intervention strategies may differ by the level of concentrated foreclosure activity within the neighborhood in which they are implemented.

That scheduled foreclosure auctions are associated with increased risk of foreclosure auction for nearby properties suggests that place-based intervention strategies that seek to reduce overall residential foreclosure activity in a given geography may not only help to stabilize neighborhoods, but also indirectly reduce the risk of foreclosure auction for other financially distressed properties, if appropriately targeted. The nonlinear association between scheduled foreclosure auctions and other properties’ probability of scheduled foreclosure auction raises questions for policy makers about where to intervene. Distinguishing between programming that aims to help individual homeowners and that which attempts to stabilize neighborhoods is important for understanding the implications of our findings.

Policy makers deciding where and how to intervene should also pay close attention to the measure of foreclosure activity they are using to determine which areas are the most in need of foreclosure prevention or mitigation programming. As we demonstrate in this analysis, both concentrated foreclosure filings and scheduled foreclosure auctions are associated with the likelihood of a financially distressed property being scheduled for foreclosure auction, but in different directions. Concentrated foreclosure filings are associated with a lower likelihood of being scheduled for auction while concentrated foreclosure auctions are associated with a higher likelihood. These findings suggest that conflating foreclosure filings with foreclosure auctions could potentially reverse the implications of what a given policy is intended to affect. Disaggregating and distinguishing among different types of foreclosure activity is important for policy and is, thus, a substantive part of what we investigate here.

Should foreclosure intervention policies target only the neighborhoods with the highest concentration of foreclosure auctions? This depends on the goals that the policies are intended to accomplish. If the express goal of a policy intervention is to keep homeowners in their homes, policy makers could consider implementing a program such as HAMP. This federal program lowers monthly mortgage payments for homeowners at risk of losing their homes to foreclosure. Homeowners seeking assistance through HAMP have to meet a number of eligibility criteria including when they obtained their mortgage, how much they owe on the mortgage, demonstrating financial hardship, and documenting sufficient income to support a modified payment.¹⁸ HAMP does not target specific geographies in its application process and does not consider the potential spillover effect that stopping one foreclosure could have on nearby homeowners or the neighborhood at large. Our findings suggest that a program such as HAMP that serves homeowners without regard for their geographic location may well achieve the goal of keeping homeowners in their homes without necessarily promoting broader neighborhood stabilization.

If, alternatively, the priority for a foreclosure intervention policy is to stabilize neighborhoods rather than keep homeowners in their homes, our findings suggest that targeting neighborhoods directly or homeowners within specific neighborhoods may be an effective strategy. Policy makers could design a program such as NSP that accounted for the differences in state-level foreclosure processes. Geographic targeting in NSP was based on two formulas: a state-level formula that included the state’s foreclosure rate, subprime rate, default rate, and vacancy rate and a substate formula that included an estimated city-level foreclosure rate and vacancy rate. Foreclosure processes differ by state, and there is no national-level statistic that represents the same phase of the foreclosure process in each locality. Our analysis shows that different types of concentrated foreclosure activity have different kinds of associations with foreclosure on individual properties. Thus, we would recommend that a policy that targets assistance geographically be designed at the state level and use indictors of foreclosure activity reflecting each state’s unique foreclosure process and timeline.

The type of goal policy makers choose should affect where they intervene. Where neighborhood stabilization is the goal and geographic targeting the strategy, our findings suggest that directing resources to a neighborhood with the highest concentration of foreclosure activity would not necessarily have the same impact as targeting neighborhoods with less foreclosure activity. In low-concentration neighborhoods, our findings suggest that the marginal effect of each additional scheduled auction may be greater yet the overall risk of auction is relatively low. Intervening here would help individual homeowners, but not necessarily an entire neighborhood. In contrast, targeting resources to neighborhoods with high numbers of foreclosure auctions may help to stem a spillover effect if the total number of auctions can be reduced considerably. The overall risk for neighboring properties facing auction is likely to remain substantial, however, and the marginal effect on homeowners still at risk may be nil. Concentrating resources in moderately affected tracts may address both policy goals: Homeowners will receive benefits directly, and focusing on these neighborhoods may also produce substantial downstream impacts in the form of reduced probability of foreclosure auction for nearby homeowners in the early phases of the foreclosure process.

With limited resources, public agencies presumably want to prioritize assistance to homeowners who would be at the greatest risk of experiencing the worst outcomes but for the intervention. Those at the greatest risk can be defined in two ways that reflect the two goals and strategies outlined above: an individual homeowner at risk of foreclosure auction regardless of the level of concentrated foreclosure auctions in the surrounding neighborhood, or a neighborhood in which homeowners are at higher risk because of heavily concentrated foreclosure activity. Our findings suggest that strategies that seek to address the needs of both individual homeowners and neighborhoods are likely to be more impactful in neighborhoods with a moderate concentration of foreclosure auctions where both total volume of auctions and individual risk could be ameliorated substantially by reducing the number of foreclosures.¹⁹ These strategies could include a borrower-focused program such as HAMP or a targeted program such as state-specific NSP. Research showing increased neglect and reduced sales prices as a consequence of foreclosure and vacant and abandoned homes (Gerardi et al. 2015; Han 2014; Lambie-Hanson 2015) highlights the tension between lengthening the foreclosure process to allow homeowners more time to become current on their mortgage and swiftly foreclosing and transferring a property to a new owner who can presumably afford upkeep. Ultimately, however, the decision about which goal and accompanying strategy to prioritize belongs to the policy makers.

Future Research

This article finds a positive association between concentrated foreclosure auctions and the likelihood that a New York City property that has received a foreclosure filing will result in scheduled auction. We add to the existing literature on foreclosure activity by showing that it is important to consider different types of foreclosure activity—both foreclosure filings and foreclosure auctions—in studying the association between foreclosure activity and property- and neighborhood-level outcomes. Future research examining the association between foreclosure activity and risk of foreclosure auction will need consistent, complete data on borrower, loan, property, and neighborhood characteristics. Ideally, as Newman (2010) argues, these data would be available to academic and policy researchers so that the details of the foreclosure process specific to individual municipalities could be compared with each other.

Additional research is necessary to make our estimation of foreclosure risk more precise and further interrogate some of our findings. The share of foreclosure filings that result in scheduled foreclosure auction varies considerably across boroughs, from less than 10% being scheduled for auction in Brooklyn to more than 40% in Queens. Including variables that account for institutional and/or jurisdictional differences, if available, may help identify why the rate of foreclosure auction varies so dramatically across boroughs in New York City.

The foreclosure process in New York State differs from other states, however, making it imperative to test these hypotheses in other jurisdictions, especially those such as Las Vegas and Detroit that have experienced high levels of foreclosure activity in settings far different from New York City, to design effective foreclosure prevention and response policy at the federal, state, and local levels. Hammel and Shetty (2013) provide an account of the complexity of the foreclosure process in Ohio, for example. Variation in foreclosure processes across jurisdictions necessitates replicating this research across geographies. Only from multiple analyses of different jurisdictions facing different challenges can we better understand common phenomena versus location-specific characteristics of foreclosure. Our analysis showing differential probability of foreclosure based on neighborhood characteristics highlights the challenges that existing policies face in targeting resources for foreclosure prevention.

Future research should also consider a broader set of potential outcomes of financially distressed properties. We have examined one possible outcome in this article, the likelihood of a property with a foreclosure filing being scheduled for auction, that we feel is particularly valuable for informing policy discussions around foreclosure mitigation. As is indicated above, there are other positive and negative potential outcomes for financially distressed properties including restructuring or refinancing debt to retain ownership, selling the property in a short sale, or turning over the deed in lieu of foreclosure (see Chan et al. 2014). In designing policy and determining where to intervene, it would be useful to know more about how concentrated foreclosure activity may be associated with the likelihood of these other outcomes.

This article helps to establish a relationship between concentrated foreclosure activity and the risk of other financially distressed properties reaching scheduled foreclosure auction and underscores the importance of considering different types of foreclosure activity in our understanding of homeowner and neighborhood risk and foreclosure mitigation policy. Our findings highlight how sensitive analyses can be to the way foreclosure activity is conceptualized, namely, the phase of the foreclosure process used to represent foreclosure activity. This finding applies to both research and policy. Our work suggests that it may be possible to address homeowner and neighborhood stability simultaneously by directing resources to distressed homeowners in neighborhoods with moderate concentrations of scheduled foreclosure auctions. Policy makers should consider the specific goals they are seeking to achieve and recognize that effective intervention strategies may differ for individuals and communities.

Footnotes

Acknowledgements

The authors would like to thank Sean Reardon, Eric Hedberg, Danielle Wallace, Kathleen Cagney, and Carly Knight for their input on methodology; Jihae Hong and Emily Baierl for their assistance preparing this manuscript; Paavo Monkkonen for providing thoughtful feedback on a preliminary draft; and the Urban Affairs Review editors and anonymous reviewers for their comments. Eric Smith also contributed valuable work to the data gathering and analysis that preceded this particular article.

Authors’ Note

The views represented here are those of the authors and do not necessarily reflect those of the Department of Housing Preservation and Development or the City of New York. Preliminary versions of this article were presented at the 2010 meetings of the Urban Affairs Association and the Association of American Geographers.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Kristin L. Perkins is a PhD candidate in sociology and social policy at Harvard University. Her research examines inequality and social stratification with a focus on families and neighborhoods. She completed this research while program evaluator at the New York City Department of Housing Preservation and Development.

Michael J. Lear completed this research while deputy director of Housing Policy Research and Program Evaluation at the New York City Department of Housing Preservation and Development.

Elyzabeth Gaumer is assistant commissioner for the Division of Research and Evaluation at the New York City Department of Housing Preservation and Development.

References

Agarwal

Sumit

Ambrose

Brent W.

Chomsisengphet

Souphala

Sanders

Anthony B.

2012. “Thy Neighbor’s Mortgage: Does Living in a Subprime Neighborhood Affect One’s Probability of Default?” Real Estate Economics 40:1–22.

Allison

Herbert

. 2010. “Foreclosure Prevention: Is the Home Affordable Modification Program Preserving Homeownership?” Testimony Before the House Committee on Oversight and Government Reform. Accessed January 8, 2016. http://www.treasury.gov/press-center/press-releases/Pages/tg608.aspx.

Ambrose

Brent W.

Capone

Charles A.

1998. “Modeling the Conditional Probability of Foreclosure in the Context of Single-Family Mortgage Default Resolutions.” Real Estate Economics 26:391–429.

Barber

Jennifer S.

Murphy

Susan

Axinn

William G.

Maples

Jerry

. 2000. “Discrete-Time Multilevel Hazard Analysis.” Sociological Methodology 30:201–35.

Berkovec

James A.

Canner

Glenn B.

Gabriel

Stuart A.

Hannan

Timothy H.

1994. “Race, Redlining and Residential Mortgage Loan Performance.” Journal of Real Estate Finance and Economics 9:263–94.

Calhoun

Charles A.

Deng

Yongheng

. 2002. “A Dynamic Analysis of Fixed- and Adjustable-Rate Mortgage Terminations.” Journal of Real Estate Finance and Economics 24:9–33.

Chan

Sewin W.

Gedal

Michael

Been

Vicki

Haughwout

Andrew

. 2013. “The Role of Neighborhood Characteristics in Mortgage Default Risk: Evidence from New York City.” Journal of Housing Economics 22:100–18.

Chan

Sewin W.

Sharygin

Claudia

Been

Vicki

Haughwout

Andrew

. 2014. “Pathways After Default: What Happens to Distressed Mortgage Borrowers and Their Homes?” Journal of Real Estate Finance and Economics 48:342–79.

Cheah

Ban Chuan

. 2009. Clustering Standard Errors or Modeling Multilevel Data? New York: Columbia Univ.

10.

Coulton

Claudia

Chan

Tsui

Schramm

Michael

Mikelbank

Kristen

. 2008. “Pathways to Foreclosure: A Longitudinal Study of Mortgage Loans, Cleveland and Cuyahoga County, 2005–2008.” Center on Urban Poverty and Community Development, Mandel School of Applied Social Sciences, Case Western Reserve Univ, Cleveland.

11.

Cutts

Amy Crews

Merrill

William A.

2008. “Interventions in Mortgage Default: Policies and Practices to Prevent Home Loss and Lower Costs.” Freddie Mac, Working Paper No. 08-1(http://www.freddiemac.com/news/pdf/interventions_in_mortgage_default.pdf).

12.

Deng

Yongheng

Quigley

John M.

Van Order

Robert

. 2000. “Mortgage Terminations, Heterogeneity and the Exercise of Mortgage Options.” Econometrica 68:275–307.

13.

Diez Roux

Ana V.

2000. “Multilevel Analysis in Public Health Research.” Annual Review of Public Health 21:171–92.

14.

Ding

Lei

Quercia

Roberto G.

Wei

Ratcliffe

Janneke

. 2011. “Risky Borrowers or Risky Mortgages Disaggregating Effects Using Propensity Score Models.” Journal of Real Estate Research 33:245–77.

15.

Chau

Paley

Irina

. 2007. “Explaining the Growth of Higher-Priced Loans in HMDA: A Decomposition Approach.” Journal of Real Estate Research 29:441–77.

16.

Foote

Christopher L.

Gerardi

Kristopher

Willen

Paul S.

2008. “Negative Equity and Foreclosure: Theory and Evidence.” Journal of Urban Economics 64:234–45.

17.

Furman Center for Real Estate and Urban Policy. 2010. “Foreclosed Properties in NYC: A Look at the Last 15 Years.” Furman Center Fact Sheet, New York Univ, New York City. Accessed January 8, 2016. http://furmancenter.org/files/publications/Furman_Center_Fact_Sheet_on_REO_Properties.pdf.

18.

Gerardi

Kristopher

Rosenblatt

Eric

Willen

Paul S.

Yao

Vincent

. 2015. “Foreclosure Externalities: New Evidence.” Journal of Urban Economics 87:42–56.

19.

Gerardi

Kristopher

Willen

Paul

. 2009. “Subprime Mortgages, Foreclosures, and Urban Neighborhoods.” The B.E. Journal of Economic Analysis & Policy 9:Article 12.

20.

Goodman

Allen C.

Smith

Brent C.

2010. “Residential Mortgage Default: Theory Works and so Does Policy.” Journal of Housing Economics 19 (4): 280–94.

21.

Graves

Erin M.

2012. “What Do the Neighbors Think? Assessing the Community Impact of Neighborhood Stabilization Efforts.” In New England Community Developments, edited by Federal Reserve Bank of Boston, 1–8. Boston, MA: Federal Reserve Bank of Boston.

22.

Hammel

Daniel J.

Shetty

Sujata

. 2013. “Complexity and Change in the Foreclosure Process in Toledo, Ohio.” Housing Policy Debate 23:35–58.

23.

Han

Hye-Sung

. 2014. “The Impact of Abandoned Properties on Nearby Property Values.” Housing Policy Debate 24:311–34.

24.

Harding

John P.

Rosenblatt

Eric

Yao

Vincent W.

2009. “The Contagion Effect of Foreclosed Properties.” Journal of Urban Economics 66:164–78.

25.

HOPE NOW. 2012. “Industry Extrapolations and Metrics (June 2012).” Accessed January 8, 2016. http://hopenow.com/industry-data/2012-08-07-HOPENOW%20Full%20Report%20(June).DRAFT.PDF.

26.

Immergluck

Daniel

. 2010. “Neighborhoods in the Wake of the Debacle: Intrametropolitan Patterns of Foreclosed Properties.” Urban Affairs Review 46:3–36.

27.

Immergluck

Daniel

Smith

Geoff

. 2005. “Measuring the Effect of Subprime Lending on Neighborhood Foreclosures: Evidence from Chicago.” Urban Affairs Review 40:362–89.

28.

Immergluck

Daniel

Smith

Geoff

. 2006. “The External Costs of Foreclosure: The Impact of Single-Family Mortgage Foreclosures on Property Values.” Housing Policy Debate 17:57–79.

29.

Kobie

Timothy F.

Lee

Sugie

. 2011. “The Spatial-Temporal Impact of Residential Foreclosures on Single-Family Residential Property Values.” Urban Affairs Review 47:3–30.

30.

Lambie-Hanson

Lauren

. 2015. “When Does Delinquency Result in Neglect? Mortgage Distress and Property Maintenance.” Journal of Urban Economics 90:1–16.

31.

Lin

Zhenguo

Rosenblatt

Eric

Yao

Vincent W.

2009. “Spillover Effects of Foreclosures on Neighborhood Property Values.” Journal of Real Estate Finance and Economics 38:397–407.

32.

Molina

Emily Tumpson

. 2012. “Reversed Gains?: African-Americans and Foreclosures in the Los Angeles Metropolitan Area, 2008–2009.” In Black California Dreamin’: The Crises of California’s African-American Communities, edited by Banks

Ingrid

Johnson

Gaye

Lipsitz

George

Taylor

Ula

Widener

Daniel

Woods

Clyde

, 127–141. Santa Barbara: UCSB Center for Black Studies Research.

33.

Newman

Kathe

. 2010. “Go Public! Using Publicly Available Data to Understand the Foreclosure Crisis.” Journal of the American Planning Association 76:160–71.

34.

Newman

Kathe

Wyly

Elvin K.

2004. “Geographies of Mortgage Market Segmentation: The Case of Essex County, New Jersey.” Housing Studies 19:53–83.

35.

Phillips

Richard A.

VanderHoff

James H.

2004. “The Conditional Probability of Foreclosure: An Empirical Analysis of Conventional Mortgage Loan Defaults.” Real Estate Economics 32:571–87.

36.

Quercia

Roberto G.

Stegman

Michael A.

Davis

Walter R.

2007. “The Impact of Predatory Loan Terms on Subprime Foreclosures: The Special Case of Prepayment Penalties and Balloon Payments.” Housing Policy Debate 18:311–46.

37.

Rogers

William H.

Winter

William

. 2009. “The Impact of Foreclosures on Neighboring Housing Sales.” Journal of Real Estate Research 31:455–79.

38.

Rugh

Jacob S.

Albright

Len

Massey

Douglas S.

2015. “Race, Space, and Cumulative Disadvantage: A Case Study of the Subprime Lending Collapse.” Social Problems 62:186–218.

39.

S&P/Case-Shiller Home Price Indices. 2011. “August 2011 Seasonally Adjusted Home Price Index Levels.” Accessed January 8, 2016. http://us.spindices.com/index-family/real-estate/sp-case-shiller.

40.

Schuetz

Jenny

Been

Vicki

Ellen

Ingrid Gould

. 2008. “Neighborhood Effects of Concentrated Mortgage Foreclosures.” Journal of Housing Economics 17:306–19.

41.

Strom

Elizabeth

Reader

Steven

. 2013. “Rethinking Foreclosure Dynamics in a Sunbelt City: What Parcel-Level Mortgage Data Can Teach Us About Subprime Lending and Foreclosures.” Housing Policy Debate 23:59–79.

42.

Swanstrom

Todd

Chapple

Karen

Immergluck

Daniel

. 2009. “Regional Resilience in the Face of Foreclosures: Evidence from Six Metropolitan Areas.” Institute of Urban and Regional Development, Univ of California, Berkeley.

43.

U.S. Department of Housing and Urban Development. 2008. Methodology for Allocation of $3.92 Billion of Emergency Assistance for the Redevelopment of Abandoned and Foreclosed Homes.” Accessed January 8, 2016. http://hud.gov/offices/cpd/communitydevelopment/programs/neighborhoodspg/nspfa_methodology.pdf.

44.

Whitaker

Stephan

Fitzpatrick

Thomas J.

2013. Deconstructing Distressed-Property Spillovers: The Effects of Vacant, Tax-Delinquent, and Foreclosed Properties in Housing Submarkets.” Journal of Housing Economics 22:79–91.