IRES2014-013

IRES Working Paper Series

Pre- and Post-Delinquency Behavior: Cross-Neighborhood Variation in New York City

Kwan Ok Lee

May 2014

Pre- and Post-Delinquency Behavior: Cross-Neighborhood Variation in New York City Kwan Ok Lee Department of Real Estate, National University of Singapore 4 Architecture Drive, Singapore 117566, Singapore

Abstract

This study examines whether and how mortgage default risk and the resolution process of defaults interact with neighborhoods where homeowners reside. It uses a unique data that matches household information from the Center for New York City Neighborhoods (CNYCN) with a variety of census tract data. Findings suggest that loan-related triggers of mortgage distress (e.g. payment adjustment) are more significant in neighborhoods with higher price depreciation and higher vacancy rates. With respect to post-delinquency behavior, distressed homeowners residing in neighborhoods with higher house prices tend to underutilize CNYCN services, controlling for their distance to service providers. Along with strong housing market performance, the concentration of Black/African-Americans in a neighborhood is likely to contribute to less successful client outcomes. Keywords: Concentrated foreclosures; Triggers of mortgage distress; Cross-neighborhood variation; Foreclosure counseling; Counseling Receipts; Counseling outcomes

1. Introduction Foreclosure rates in the U.S. have consistently risen since mid-2007, with many parts of the U.S. seeing rates three and four time greater than they were before 2007 (RealtyTrac 2012). While New York City (NYC) has been hit less severely by foreclosure crisis, foreclosure and delinquency rates have increased considerably in some parts of NYC. As of December 2011, the average percentage of homeowner households in the formal foreclosure process1 of the top 10 NYC census tracts was 9.68%, far exceeding the statewide rate (4.64%) and nationwide rate (3.37%) (Corelogic 2012). Figure 1 further shows considerable variation in the distribution of mortgage delinquencies across NYC census tracts. The concentration of delinquent mortgages and subsequent foreclosure activities in certain neighborhoods should be of a concern to many policy makers and scholars because residents in these neighborhoods – often lower-income or minority-concentrated neighborhoods – are exposed to higher risk of foreclosures (e.g. Immergluck and Smith 2005; Bostic and Lee 2009). Another concern pertains to negative spillover effects at the neighborhood level. Concentrated foreclosures often lead to the concentration of deteriorated or vacant residential buildings in a neighborhood, and so bring negative impacts on neighborhood property values (Shlay and Whitman 2004) and accelerate the housing fileting process (Li and Morrow-Jones 2010). Other negative effects include crime, disinvestment, and associated municipal costs (Apgar et al. 2005; Immergluck and Smith 2006).

1

In New York, New York State Banking Department sends 90-day pre-foreclosure notices when borrowers are 2-3 months late with mortgage payments. These notices are designed to inform the borrowers that they are at-risk of foreclosure and what preventative steps are available to them. Therefore, these notices do not indicate that the formal foreclosure process has commenced or that all borrowers receiving a 90‐day pre‐foreclosure notice will enter into the foreclosure process. If a borrower does not become current on their mortgage within 90 days of receiving the notice, the lender then has the right to begin the formal foreclosure process by filing a lis pendens.

-2-

Despite the importance of understanding cross-neighborhood variation in the risk of delinquencies and foreclosures, such spatial analysis has been little done in the mortgage default literature. To explain why homeowners get into and how they respond to mortgage distress, past research has primarily focused on loan characteristics such as subprime loan and loan-to-value ratio and borrower characteristics such as race and income. Only recently, several studies started to suggest that after controlling for borrower and loan characteristics, mortgage defaults are disproportionately concentrated in certain neighborhoods, including central city neighborhoods (Immergluck 2009), neighborhoods with a higher share of subprime mortgages (Immergluck and Smith 2005), and predominantly Black/African-American neighborhoods (Doviak and MacDonald 2011; Chan et al. 2013). To the author’s current knowledge, no existing research has considered whether residing in a certain neighborhood influences the resolution process of mortgage defaults including receipts of foreclosure counseling and counseling outcomes. Building on the past literature, this study raises three research questions: 1) Do triggers of mortgage distress vary across neighborhoods?; 2) Does the probability of receiving foreclosure counseling vary by neighborhood characteristics such as median household income, racial composition, and house market conditions? ; 3) Is there any relationship between outcomes of foreclosure counseling and neighborhood characteristics such as the distance to service providers, housing market conditions, and the concentration of peer clients? By answering the first research question, this study attempts to shed additional light on a geographic disparity in mortgage performance and concentration of mortgage defaults. Then, this study explores whether the behavioral patterns of homeowners who have faced mortgage distress differ across NYC neighborhoods. The second research question addresses issues of lower utilization of available foreclosure assistance to eligible distressed homeowners that can hamper potential

-3-

success of this assistance (Russell et al. 2014). It investigates reasons for cross-neighborhood variation in the likelihood that distressed households utilize foreclosure assistance provided by the Center for New York City Neighborhoods (CNYCN), the largest network of non-profit service providers in New York City.2 Finally, lower success rates of counseling recipients in certain neighborhoods may contribute to elevated risks of delinquent properties to progress into foreclosures. Hence, the final research question tests whether the location of clients is related with the probability that clients successfully graduate from CNYCN services. In attempting to answer the above research questions, this study uses a unique data that matches household-level case management data with a variety of census tract-level information. Household data were collected from homeowners who sought foreclosure prevention assistance through the CNYCN’s network of providers between July 2008 and October 2010. These data contain information on key borrower and loan characteristics as well as geocodes of their residence. Census tract data include pre-foreclosure filings from New York State Banking Department, neighborhood characteristics from the 2005-2009 American Community Survey (ACS), and median house prices from DataQuick. To analyze pre- and post-delinquency behavior, this study uses a series of descriptive logit models of triggers of mortgage distress, linear regression models of counseling receipts, and logit models of counseling outcomes. Analysis results suggest that the mortgage distress associated with loans (e.g. payment adjustment and loan unaffordable from origination) is more significant in NYC neighborhoods with higher price depreciation and higher vacancy rates, controlling for client and loan characteristics. With respect to the post-delinquency behavior, homeowners residing in NYC

2

CNYCN is not the only counseling provider in New York City but it is the largest network of more than 50 agencies located across five boroughs and it is likely that the counseling cases reported by CNYCN represents the vast majority of foreclosure counseling in New York City (Chan et al. 2012).

-4-

neighborhoods with higher house prices are less likely to seek counseling or legal assistance provided by CNYCN, controlling for their distance to CNYCN service providers. Finally, the analysis predicts that along with strong housing market performance, the concentration of Black/African-Americans in the neighborhood is likely to contribute to less successful client outcomes: foreclosure, bankruptcy, or withdrawal. 2. Background To explain what triggers and how homeowners respond to mortgage distress, past research has primarily focused on loan characteristics such as subprime loan and loan-to-value ratio and borrower characteristics such as race and income. There has been little research on the spatial pattern of pre- and post-delinquency behavior although this is critical to explain why certain neighborhoods suffer the higher risk of mortgage defaults and foreclosures than other neighborhoods. Building on the past literature, this study addresses three important research questions: 1) Do triggers of mortgage distress vary across neighborhoods?; 2) Does the probability of receiving foreclosure counseling vary by neighborhood characteristics such as median household income, racial composition, and house market conditions? ; 3) Is there any relationship between outcomes of foreclosure counseling and neighborhood characteristics such as the distance to service providers, housing market conditions, and the concentration of peer clients? 1) Cross-Neighborhood Variation in Triggers of Mortgage Distress Several recent studies show that subprime lending and mortgage defaults are disproportionately concentrated in certain neighborhoods within a city. First, Bostic and Lee (2009) report the elevated likelihoods of foreclosure in lower-income neighborhoods. Second, Immergluck (2009) suggests that mortgage defaults are disproportionately concentrated in specific locations such as

-5-

central city neighborhoods. These two analyses use the neighborhood level data so fails to account for individual-level heterogeneity. Finally, Doviak and MacDonald (2011) and Chan et al. (2013) report that homeowners residing in the neighborhood with a higher share of Black/African-American population are more likely to default regardless of their own race. While they control for individual-level factors in analyzing the effect of neighborhood racial composition on foreclosure rates, they focus on observable loan characteristics and the race of borrowers. This study attempts to advance past research by further disentangling underlying causes of the spatial concentration of delinquencies and foreclosures. A basic assumption is that if certain neighborhood characteristics are associated with residents’ mortgage troubles, the trigger of mortgage distress should differ among households that reside in different neighborhoods, even after controlling for differences in household- and loan-specific attributes. Using household-level case management data collected from the interaction with homeowners experiencing mortgage distress in NYC, it is able to investigate what actually triggers the concentration of delinquencies and foreclosures in certain neighborhoods. For example, it tests whether households residing in NYC neighborhoods with a higher share of minority residents are more likely to report loanrelated reasons as the trigger of their mortgage distress compared to those residing in neighborhoods with a lower share of minority residents. This study also pays attention to geographic dynamics of housing markets. While housing prices have declined over the last several years in almost all areas in the U.S., the degree of this decline varies across and within cities (Glaeser and Gyourko 2008). For example, Manhattan has seen a lower decline in housing prices than Brooklyn or Queens due to its tighter housing markets and more inelastic supply. A testable hypothesis is whether individual economic

-6-

hardship or familial events will play a more distinct role to mortgage distress in neighborhoods with lower price depreciation rather than risks associated with alternative mortgage instruments. 2) Cross-Neighborhood Variation in Receipts of Foreclosure Counseling As shown in Figure 1, geographic distribution of mortgage delinquencies does not overlap with the location of clients who receive CNYCN services in many parts of New York City. For example, while several neighborhoods in Manhattan and Queens show a moderate level of concentrated foreclosures, there are almost no CNYCN clients in these neighborhoods. As suggested by Quercia et al. (2004), non-profit financial counselors can play an important role in intermediating delinquent borrowers and their lenders since many borrowers tend to avoid a direct contact with their lenders (Apgar et al. 2005). Roper Public Affairs and Media (2005) suggests that three-quarters of the borrowers experiencing mortgage defaults desire to receive services from counseling agencies. While counseling efforts have rapidly expanded since the subprime crisis in many cities, previous research is silent on cross-neighborhood variation in the receipts of foreclosure counseling. This study therefore attempts to identify whether and why the participation in the CNYCN foreclosure counseling program varies across NYC neighborhoods. This cross-neighborhood variation may involve the presence of a supply-side constraint or differences in household demand. First, it is natural to assume that defaulted homeowners residing farther from CNYCN service providers are less likely to seek these services. Next, it would be reasonable to expect that local housing markets can play a role in demand for service receipts. As Weinstein (1980) suggests, myopia may lead households residing in the neighborhoods with relatively strong housing markets to overestimate future home price appreciation and underestimate the foreclosure risk of their properties. Therefore, households

-7-

experiencing mortgage distress in these neighborhoods may have lower demand for CNYCN services than those in other neighborhoods. 3) Cross-Neighborhood Variation in Outcomes of Foreclosure Counseling Most previous studies evaluating foreclosure counseling provide evidence that it has helped many clients avoid or cure foreclosures (e.g. U.S. HUD 1983; 1975; Mayer et al. 2009).3 In a more rigorous statistical manner, Agarwal et al. (2010) report substantially lower default rates among the graduates from a voluntary counseling program as well as persistence of positive effects of counseling over time. Using the loan performance data merged with administrative data from a nationwide counseling agency, Collins and Schmeiser (2013) consistently find that telephone counseling reduces the likelihood of losing a home to foreclosures. In predicting the outcomes of foreclosure counseling, previous studies tend to rely on client characteristics (e.g. race, income, education, FICO score) and loan characteristics (e.g. adjustable rate mortgage, loan balance, debt-to-income ratio). More recently, several researchers account for the importance of service characteristics to explain variation in these outcomes. Quercia and Cowan (2008) suggest that for each additional hour spent with foreclosure assistance, probabilities that clients avoid foreclosure increase by 10%. They also report that distressed homeowners who receive budget/credit counseling are twice as likely to avoid foreclosure. On the other hand, Collins and Schmeiser (2013) find that the length or intensity of counseling does not matter for outcomes. In terms of the mode of counseling, Collins (2007) finds that clients are more likely to attend additional counseling sessions after receiving face-toface counseling compared to phone counseling.

3

There is a concern about potential flaws in research design, methods, and data that previous evaluation studies used (Collins and O’Rourke 2011).

-8-

This study explores whether outcomes of the CNYCN foreclosure program significantly vary across NYC neighborhoods after controlling for the above client, loan, service characteristics that differ across individual clients. It hypothesizes that the distance between clients’ residence and CNYCN service providers is likely to have a positive relationship with the probability that clients withdraw from CNYCN services. It also test for a relationship between client outcomes and the concentration of CNYCN clients in a neighborhood. Presumably, neighbors could go to receive services together and give a positive influence on motivations and attitudes for counseling, so a positive relationship is expected. 3. Data and Methods 1) Data This study combines several data to analyze effects of neighborhood characteristics on pre- and post-delinquency behavior of individual homeowners. The main data are individual-level case management data4 from the Center for New York City Neighborhoods (CNYCN), a non-profit organization that partners with more than 50 agencies and coordinates foreclosure prevention and intervention services in all five boroughs in NYC. According to Chan et al. (2012), although it is not the only counseling agency in NYC, it is likely that counseling cases reported by CNYCN represent the vast majority of foreclosure counseling in New York City. CNYCN sends its counseling service information to those who receive 90-day pre-foreclosure notices from the New York State Banking Department (NYSBD). The delinquent borrowers in NYC may directly contact CNYCN or call 3115 and are directed to CNYCN. The CNYCN data consist of over

4

The data are based on service providers’ case notes of their interactions with homeowners so there is a potential limitation that some information gets lost or misinterpreted, especially given potential errors that service providers make and clients’ likely emotional distress. 5 It is the phone number for government information and nonemergency services in NYC and incoming calls will be directed to CNYCN helpline services

-9-

10,000 individual case records of homeowners who received face-to-face foreclosure counseling or legal services through CNYCN’s network of providers between July 2008 and October 2010. CNYCN data contain useful information including: 1) client characteristics such as race, primary language they use, household income, FICO score, and family composition, 2) loan characteristics such as the loan amount, monthly payment, interest rate, whether the loan is adjustable rate mortgage (ARM), debt-to-income ratio (DTI), date of loan origination, and 3) service characteristics such as hours of legal and counseling services and duration of being a CNYCN client. Since CNYCN data contain geocodes of clients’ residence, several census-tract level datasets can be matched with their individual data to assess the neighborhood conditions that could potentially influence pre- and post-delinquency behaviors. First, it uses the data of preforeclosure filings from the NYSBD for the period of April 2010-February 2011.6 To account for the extent to which neighborhoods experience foreclosures in a different stage, these preforeclosure filings are categorized as those delinquent for less than 90 days, 90-120 days, and 120+ days. Because census tracts vary significantly in terms of owner-occupancy rates, delinquency rates are calculated per unit of owner-occupied housing basis.7 Second, this study obtains information from DataQuick on resale median prices of single family homes in every ZIP code of New York City for the period of 2006-2008.8 With this

6

New York State did not require the pre-foreclosure filings (PFFs) before February 2010. This study assumes that mortgage default incidences in one neighborhood after this requirement would have been similar to those for the period of 2008-9. 7 For the robustness check, delinquency rates were calculated per unit of owner-occupied housing with mortgages outstanding. Results appeared to be consistent. 8 To reduce a measurement error, the analysis excludes ZIP codes with less than five transactions. The majority of the client intakes took place between 2006 and 2009 and these clients have likely observed the local housing market during the period of 2006-8.

- 10 -

information, the rate of home price appreciation is calculated.9 This performance of local housing markets is likely to be related with both the trigger and resolution process of mortgage distress. Third, the 2005-2009 American Community Survey (ACS) provides all other neighborhood information at the census tract level.10 Median household income can be an important factor, as lower-income households will generally be more vulnerable to the delinquency risk and less capable to overcome this risk. While lower-income households are less capable to overcome the delinquency risk, they may also have less information of CNYCN services or less time to access services. Because mortgage performance and financial sophistication have both been found to be correlated with race, the racial or ethnic makeup of a community may be also correlated with the delinquency risk and the likelihood of counseling receipts. Finally, using the Geographic Information System (GIS) technique, the distance between each census tract and each CNYCN agency is computed. The models for the postdelinquency behavior control for the distance between census tracts of clients and their CNYCN service providers. 2) Methods This study use three distinct approaches to address three research questions indicated earlier. First, to investigate potential reasons for the concentration of delinquencies and foreclosures in certain neighborhoods, this study runs a series of descriptive logit models with one dichotomous dependent variable among three different triggers. To do so, reasons of mortgage distress selfreported by CNYCN clients are combined into three categories: economic hardship (business

9

The author acknowledges that using median house prices as a measure of appreciation is not perfect because the homes were not constant quality and reflect only homes that were sold. 10 2010 Census do not provide the census-tract level data.

- 11 -

failure, loss of income, difficulty meeting utility payments), loan-related reasons (payment adjustment, loan unaffordable from origination, fraud), and others (inability to sell, death in family, incarceration, military service, marital problems, medical issues).11,12 Logit models measure the effect of residing in the neighborhoods with different characteristics on different reasons for mortgage distress. Key explanatory variables that identify characteristics of the census tract where distressed homeowners reside include: median household income, a share of minority population, house price appreciation, vacancy rates, and delinquency rates. Following Doviak and MacDonald (2011) and Chan et al. (2013), these models also incorporate a wide range of controls including client characteristics (e.g. race, primary language,13 income, family status) and loan characteristics (e.g. loan amount, monthly payment, ARM, interest rate, DTI, FICO score, delinquency status, and the year of origination). Second, to analyze whether and why delinquent homeowners in certain neighborhoods are less likely to receive foreclosure counseling provided by CNYCN, linear regressions are estimated at the census tract level. A dependent variable is the ratio of CNYCN clients to preforeclose filings (PFFs) in each census tract.14 Independent neighborhood variables fall into several categories: homeownership (the share of PFFs, median house price, price appreciation rate, monthly owner cost, owner cost as a percentage of income), income (poverty rate, median household income), demographic (% Black/African Americans, % Hispanics/Latinos)15, and

11

The CNYCN client is told to pick up one biggest reason for mortgage distress out of 18 reasons that are mutually exclusive at the time of intake. 12 In addition to separate logit models, this study also estimates a multinomial logit (MNL) model with a categorical dependent variable with these five categories. Here the base category is “others”. The results are presented in Appendix 1. 13 Since the nativity information is not available, the primary language is used as a proxy. 14 To further analyze whether any of these regressors are associated with the higher incidence of foreclosures but not with the probability of counseling receipts, additional linear regressions are estimated with a dependent variable of the fraction of owner-occupied units with PFFs. 15 These two categories are not mutually exclusive. Since Hispanic Blacks and Hispanic Asians account for only 1.3 percent of the entire sample, however, most households defined as Hispanics are Hispanic Whites.

- 12 -

housing market (homeownership rate, vacancy rate, multi-family housing rate, and rate of housing built before 1970). While CNYCN services are available in all five boroughs, some census tracts may be located at a longer distance to CNYCN partner agencies than other tracts. Hence, the model is controlled for the distance between the census tract and the location of the closest CNYCN service provider as well as borough dummies. Finally, to predict the outcome of CNYCN clients, a logit model is estimated. To do so, the primary client outcomes are categorized by success (brought mortgage current, mortgage modified, reverse mortgage obtained, satisfied mortgage, credit repaired, extended homeowner or tenant’s tenure in property, obtained clear title to property, obtained partial claim loan from FHA lender, secured charitable grant), withdrawal from services, and others (foreclosure, bankruptcy, executed deed-in-lieu, relocation, referral, property sold, short sale, advised rights and options).16,17 This model focuses on a relationship between these client outcomes and neighborhood characteristics such as the distance to a service provider, the concentration of CNYCN clients, racial composition, and house price appreciation. The model is controlled by client and loan characteristics at the household level. It also accounts for characteristics of services that clients have received including hours of counseling and legal services and duration of being a CNYCN client.18 Since CNYCN is a consortium of more than 50 agencies, there can be variation in resources, community relationships, and staff capacities across agencies and this

16

The CNYCN counselors answered the primary outcomes of their clients out of 18 outcomes that are mutually exclusive and the data excludes those who are currently in counseling. While there is a non-successful outcome such as foreclosure, it accounts for a very small fraction of the data (less than 0.5%) so the failure is not separately categorized in this analysis. 17 Withdrawals are categorized separately since those withdrawn from CNYCN counseling services may have resolved their distress without the aid of counseling services or have sought counseling services from other agencies. 18 Some clients may have a pre-existing relationship with CNYCN (e.g. taking homeownership counseling) prior to their foreclosure counseling/legal assistance. Homeowners with this relationship may be more likely to seek out the CNYCN foreclosure assistance and there may be greater potential for a successful outcome. Because the date when clients contacted the CNYCN for the first time is known, the model is controlled for the duration of being a CNYCN client.

- 13 -

variation may influence the outcome of CNYCN services. To account for this variation, the analysis includes agency dummies. The above models pool CNYCN clients from all of the NYC neighborhoods. Since this sample may include multiple clients at a single census tract, these models may violate the standard assumption that observations are the independent. This issue is handled by clustering standard errors at the census tract. The standard errors clustered in this way appear on average about 2-2.5 times greater than those computed without clustering. 4. Results 1) Profiles of CNYCN clients Table 1 compares household characteristics among regular borrowers, borrowers with 90-day pre-foreclosure filings (PFFs), and CNYCN clients in NYC.19 As expected, borrowers with 90day PFFs are more likely to be low-income than regular borrowers. Also evident is that CNYCN clients are much more likely to be lower-income even compared with other distressed borrowers in NYC. Moreover, despite their lower income status, these clients tend to take out a larger amount of loan. Their loans are also riskier representing a significantly higher portion of ARM borrowers and relatively higher interest rates. With respect to the race, CNYCN clients are much more likely to be minorities, compared to other NYC borrowers with PFFs. In particular, a significant portion of CNYCN clients are Blacks/African Americans. Table 1 also compares CNYCN clients’ neighborhoods with the average of all NYC neighborhoods. CNYCN clients do not live in particularly lower-income or lower-cost

19

The information of other NYC borrowers is from the matched data of Home Mortgage Disclosure Act data (HMDA) in 2004-2008 and Pre-foreclosure Filings (PFFs) in 2010. This attempts to report borrowers who defaulted and those who did not default in 2010. Readers should note that since the CNYCN data has been recorded at the time of counseling intake which varies from 2004 to 2010, the time frame is quite comparable but may not be perfectly equivalent with the HMDA-PFF data.

- 14 -

neighborhoods. However, their neighborhoods have experienced a significantly higher drop in house prices between 2006 and 2008, compared to other NYC neighborhoods. This can be one of potential reasons why CNYCN neighborhoods also show the higher rate of pre-foreclosure filings in 2010. In terms of demographics, CNYCN neighborhoods are more concentrated with Black/African American residents. This is consistent with the literature that higher foreclosure rates are associated with the higher share of Blacks/African Americans at the neighborhood level (Chan et al. 2013). 2) Triggers of Mortgage Distress Table 2 presents the main reason why CNYCN clients experience mortgage distress. Economic hardship appears to be the most significant trigger followed by loan-related reasons and others such as familial events. As suggested by Reinhart and Rogoff (2009), the aftermath of the subprime crisis is closely associated with a profound decline in employment and subsequent economic hardship of families. The unemployment rate of New York City rose from 4.2% in November 2006 to 10.5% in January 2010 (New York State Department of Labor). To these homeowners who lost their jobs or experience an unexpected income shock, economic hardship may play the most critical role in their mortgage distress. Table 2 shows whether triggers of mortgage distress vary across neighborhoods where they reside. First, residents’ mortgage troubles in predominantly Black/African American neighborhoods are more likely to be associated with loan-related reasons. A higher drop in house prices and higher vacancy rates also increases the risk associated with loan-related reasons. This is consistent with Foote et al. (2008) suggesting a close relationship between local housing markets and the risk of defaults. With respect to racial variation among CNYCN clients, African Americans are more likely to experience loan-related problems. On the contrary, Hispanic/Latino

- 15 -

clients are more likely to report economic hardship as a reason for mortgage distress. This raises interesting questions as to the racial difference in mortgage performance among minorities, and whether any difference in the treatment of mortgage industry may explain the disparate findings. While low-income households (50-80% of Area Median Income) are more likely to report economic hardship, very low-income households (below 50% of AMI) are less likely to do so. Next, each column of Table 3 presents results of a logit model with one dichotomous dependent variable among three different triggers: economic hardship, loan-related reasons, and others.20 Beginning with neighborhood characteristics, Table 3 provides evidence that mortgage distress is less likely to be triggered by loan-related reasons in neighborhoods with higher price appreciation (Column 2). Economic hardship or other reasons are more significant in these neighborhoods (Column 1 and 3). Also evident is that after controlling for differences in client and loan characteristics, homeowners residing in the census tract with a higher vacancy rate are more likely to experience mortgage distress due to loan-related triggers (Column 2). These results strongly suggest the importance of housing market dynamics to mortgage default risks (Foote et al. 2008).21 Economic status of the neighborhood, racial composition, and the concentration of PFFs appear to be insignificant to the triggers of mortgage distress. Table 3 also shows that Hispanics/Latinos residing in neighborhoods with a large Hispanic/Latino population are much more likely to report loan-related triggers (Column 2). This finding mirrors previous results that racial discrimination in mortgage lending could lead to riskier loans for minorities residing in minority-concentrated areas. For example, Kain and Quigley (1975) and Yinger

20

Analysis results are consistent with the results of a multinomial logit (MNL) model with a categorical dependent variable with three triggers of mortgage distress (Appendix 1). 21 R square decreases significantly if excluding price appreciation especially in logit models shown in Column 2 and 3. This suggests that price appreciation does have strong predictive power for loan-related triggers and other triggers.

- 16 -

(1978) provide evidence that minority households tend to pay higher price for equivalent housing compared to Whites. Courchane and Nickerson (1997) and Black et al (2003) also suggest that redlining may increase loan prices for applicants from minority-concentrated neighborhoods. These logit models are controlled for a wide range of client and loan characteristics and some results of these controls are worth to mention. Table 3 reports that loan-related default risk is more significant among the homeowners using non-English as a primary language (Column 2). Since immigrants who are new to homeownership have more likely experienced foreclosures for their home purchase mortgages compared to native-born households (Allen 2011), it is important to understand that their mortgage distress is primarily caused by loan-related reasons.22 Results also show that higher risk loans such as ARMs or those with higher monthly payment significantly increase loan-related mortgage distress (Column 2), as would be expected. Clients who have been delinquent at least 90 days are less likely to report the loan-related triggers while other reasons are more significant among them (Column 2). A potential explanation is that other issues could grow as mortgage delinquency continues even though it was triggered with loanrelated problems in the beginning. 3) Receipts of Foreclosure Counseling Figure 2 distinguishes areas where the ratio of CNYCN clients to PFFs is lower than the average citywide rate and areas where this ratio is higher. It shows that most areas with concentrated PFFs in the northern portion of Brooklyn and the eastern portion of Queens are light or dark green, suggesting that CNYCN has drawn clients from these areas at a rate higher than the overall rate for New York City. The red areas where CNYCN has drawn clients at a lower rate

22

Allen (2011) specifically compares Hispanic, foreign-born homeowners with white, native-born homeowners.

- 17 -

include South Brooklyn, some scattered areas in Queens, and several neighborhoods in Manhattan and Bronx. It is noteworthy that only less than 1 percent of clients using the face-to-face counseling services provided by CNYCN live in Manhattan. Among incoming calls to the CNYCN helpline service, only 4.15 percent are Manhattan residents.23 Given that 19 percent of NYC’s owneroccupied units and 12 percent of NYC’s PFFs are located in Manhattan in 2010, this is extremely lower demand for face-to-face and helpline services. One potential explanation is that Manhattan residents with mortgage distress may fail to realize their situation because of the scarcity of foreclosures in Manhattan as indicated by Erskine Kennedy in New York Times (2008). It is also possible that Manhattan homeowners in mortgage distress may be in a better position than those in other areas because of relatively stable housing markets, and in turn, have lower demand for assistance. Finally, supply constraints may play a role if most of CNYCN counseling agencies are located outside Manhattan.24 To formally investigate potential reasons why homeowners residing in certain neighborhoods with a higher risk of mortgage delinquencies do not participate in CNYCN counseling services, linear regressions are estimated at the census tract level with two comparable dependent variables: delinquency rates (the fraction of houses with PFFs) and service receipt rates (the ratio of CNYCN clients to PFFs). Using the pooled sample of NYC census tracts, it tests whether and how neighborhood characteristics such as the performance of

23

CNYCN offers helpline services (direct call or from 311) as a pre-counseling tool to refer potential clients to one of CNYCN service providers. This is a separate dataset from the main data used for this analysis and contains no outcome information. 24 This is not likely because there are three CNYCN agencies located in Manhattan. Still, to address this concern of supply constraints, the distance to the closest CNYCN agency from each census tract is computed and included in the regression model.

- 18 -

neighborhood housing markets and racial composition are related with the likelihood that distressed residents receive CNYCN counseling services.25 Tables 4 indicates that key neighborhood characteristics explaining delinquency rates are the house price level and housing burdens of homeowners in a neighborhood; both lower median prices and higher owner costs with mortgages are related with higher PFFs incidences (Column 1). The data also point to the importance of housing market dynamics to the likelihood of receiving foreclosure counseling; after controlling for the distance to the closest CNYCN service provider, the likelihood of seeking CNYCN’s counseling services falls significantly if the median price is higher or housing burdens are lower in the neighborhood (Column 2). Interestingly, the extent to which these housing market factors affect rates of counseling receipts is much greater than the extent to which they influence delinquency rates. It is also notable that the rate of PFFs in the earlier stage appears to decrease the probability of counseling receipts (Column 2). A potential interpretation is that homeowners may not yet realize that mortgage delinquency has begun to be serious in their neighborhoods (Weinstein 1980). Tables 4 then provides evidence that distressed homeowners in higher poverty neighborhoods are less likely to seek CNYCN counseling services. Hence, these distressed homeowners may be more susceptible to the progress into foreclosure by not receiving foreclosure counseling that can help resolve mortgage distress. The analysis also shows that PFF incidences are greater in neighborhoods with a greater presence of Black/African American and Hispanic/Latino households (Column 1) and this is consistent to evidence provided by previous research (Allen 2011; Doviak and MacDonald 2011; Chan et al. 2013). Fortunately, controlling

25

Although CNYCN is not the only counseling agency in NYC, it is meaningful to analyze the CNYCN take-up rates because it is likely that the counseling cases reported by CNYCN represent the vast majority of foreclosure counseling in NYC (Chan et al. 2012).

- 19 -

for the default risk in the neighborhood, distressed homeowners residing in these minorityconcentrated neighborhoods are more likely to seek CNYCN counseling services (Column 2). This raises an interesting question as to the behavior of distressed minority homeowners across different stages of mortgage delinquency. On the one hand, a lower degree of financial sophistication could lead them to decide to buy a house without proper evaluation of their own repayment ability or to pay more than is necessary for their credit (Courchane et al. 2004). On the other hand, they are less likely to have a satisfactory safety net to overcome a trigger event such as job loss or the monthly mortgage payment changes associated with instrument risk (An and Bostic 2009) so they tend to seek immediate counseling or legal assistance to resolve mortgage distress. 4) Outcomes of Foreclosure Counseling While some CNYCN clients successfully modify their delinquent loans and bring mortgage current, other clients still experience the progress into the foreclosure stage or withdraw from CNYCN services. Table 5 shows whether client outcomes vary by client characteristics and characteristics of their neighborhoods. On average, 38.37% of clients have successfully modified their mortgages or refinanced while 23.38% have withdrawn from CNYCN services. Clients residing in predominantly Black/African American neighborhoods are more likely to lead to the successful outcomes. However, they also show a higher probability of the withdrawal. With respect to client characteristics, Black/African Americans and those whose primary language is not English are slightly less likely to be successful while Hispanic/Latino clients are more successful. Lower-income clients have a higher probability of non-successful outcomes and withdrawal. Consistent with Collins (2007) and Quercia and Cowan (2008), clients who receive counseling or legal assistance longer than the average are much more likely to successfully

- 20 -

modify loans or bring mortgage current. Similarly, longer hours of counseling or legal assistance significantly reduce the probability of withdrawal. Table 6 presents results of logit models that estimate the probability of successful outcomes26 with the sample of all clients that finished CNYCN services (Column 1) and with the subsample of clients excluding those that are withdrawn from services (Column 2). As mentioned in the earlier section, these models are controlled for potential variation in service quality and staff capacity between different CNYCN agencies by adding agency dummies.27 In addition to main logit models, Appendix 3 presents results of a logit model of the likelihood that clients withdraw from CNYCN services, using same explanatory variables shown in Table 6. First, results suggest that strong housing market performance has a negative relationship with client outcomes. Clients residing in the neighborhood with higher price appreciation are less likely to successfully overcome their mortgage distress (Table 6, Column 1 and 2) and more likely to withdraw from CNYCN services (Appendix 3). On the one hand, as suggested by Weinstein (1980) and Nofsinger (2012), optimism bias may have led some clients to underestimate the foreclosure risk of their properties, and in turn, negatively affect their efforts toward CNYCN services. On the other hand, financial institutions holding distressed loans in neighborhoods with higher price appreciation may see a greater advantage in foreclosing than modifying a loan. In addition, clients withdrawn from CNYCN services do not necessarily have

26

These outcomes include brought mortgage current, mortgage modified, reverse mortgage obtained, satisfied mortgage, credit repaired, extended homeowner or tenant’s tenure in property, obtained clear title to property, obtained partial claim loan from FHA lender, and secured charitable grant. 27 According to Appendix 2, the rate of successful client outcomes significantly vary by counseling agencies, ranging from 11.5% (JASA - Legal Services for the Elderly in Queens) to 94.4% (Neighborhood housing Services – Staten Island). Similarly, withdrawal rates range from 0% to 71.88% (MHANY Management Inc.). Even if differences in demographic and economic characteristics of clients can account for some variation in their outcomes, variation in the probabilities of success and withdrawal still seems to be very large. This raises an important question whether the distribution of resources has been unequal to different services providers (in fact, the hours receiving counseling and legal assistance significantly differ between service providers). Unfortunately, CNYCN was not able to provide the answers to this question.

- 21 -

negative perceptions about services.28 As we do not have sufficient information of such strategic foreclosures or homeowners’ motivations for withdrawing from services, it is premature to draw any conclusion. Second, Table 6 shows that a higher share of Black/African American population has a negative relationship with the probability of success (Column 1 and 2). This suggests that existing higher delinquencies and foreclosures in predominantly Black/African American neighborhoods in NYC (Chan et al. 2013) may be further exacerbated by lower success rates of foreclosure counseling among their residents. Third, a longer distance to service appears to increase withdrawal rates (Appendix 3), indicating that accessibility and related transaction costs play a role in continuing to participate in counseling but not necessarily in success of counseling outcomes. This is consistent with Rusell et al. (2014) that geographic proximity to intake agencies increases the probability of completing the application for counseling. Finally, there is no significant relationship between the concentration of CNYCN clients and outcomes of foreclosure counseling. Several results of client and service characteristics are noteworthy and may have important implications for service providers. Hispanic/Latino clients are more likely to overcome their mortgage distress (Table 6, Column 1) and less likely to withdraw from CNYCN services (Appendix 3). This is consistent with Chan et al. (2013) that find lower default hazards of Hispanic white borrowers.29 On the contrary, clients who use non-English as their primary language show the lower probability of success and higher probability of withdrawal. Given foreign-born homeowners are at higher risk for foreclosure (Allen 2011), this suggests that special attention should be given to these homeowners in organizing counseling efforts. CNYCN

28

Those withdrawn from CNYCN counseling services may have resolved their distress even without the aid of counseling or have sought counseling from other agencies 29 Because Hispanic Blacks and Hispanic Asians account for only 1.3 percent of the current sample, most CNYCN clients defined as Hispanics are Hispanic Whites.

- 22 -

clients that have previously failed loan modification are less likely to lead to successful outcomes (Table 6, Column 1) and more likely to withdraw from the services (Appendix 3), as one would expect. Also evident is that the duration of being a CNYCN client has a significant, positive effect on the probability of success (Table 6, Column 2). However, there is no consistent evidence with respect to the direct relationship of client outcomes with legal or counseling hours. This is consistent with evidence that the length of counseling makes no difference in outcomes (Collins and Schmeiser 2013). 5. Conclusion Given significant variation in delinquencies and foreclosures between neighborhoods, an important question arises whether and how mortgage default risk and the resolution process of mortgage defaults interact with neighborhoods where homeowners reside. Using the householdlevel case management data, this study explores the pre- and post-delinquency behavior of homeowners residing in different census tracts in New York City. Unlike other studies that focus only on final outcomes of distressed mortgages (e.g. foreclosure vs. modification), therefore, this study is able to document the dynamic process of mortgage delinquencies: what triggers the mortgage distress, how homeowners treat this distress, and whether they successfully overcome mortgage distress. An important finding regarding the underlying mechanism of mortgage distress is that the default risk associated with loans is more significant in neighborhoods with higher price depreciation and higher vacancy rates, controlling for differences in household-specific economic and loan characteristics. Housing market performance may also matter for the resolution process of mortgage distress. Median house prices have a negative relationship with the probability of counseling receipts while higher price appreciation appears to decrease success

- 23 -

rates of counseling. Therefore, housing market dynamics may not only elevate the default risk but also influence how homeowners respond to mortgage distress. Although it is possible that distressed homeowners residing in the neighborhoods with relatively strong market performance may underestimate foreclosure risk and underutilize foreclosure counseling, it is premature to draw any conclusion given a lack of further information. In addition to housing market dynamics, neighborhood racial characteristics play an important role in post-delinquency behavior. Controlling for the client’s own race, the concentration of Black/African-American population in a neighborhood increases the probability of service participation but reduces the probability of successful outcomes. Finally, the longer distance to the service provider may hinder continuing participation in foreclosure counseling. These findings have important implications for both public policy and counseling practices. They suggest that loan-related triggers of mortgage distress are more sensitive to local housing market dynamics, compared to economic or other triggers. If certain neighborhoods experience higher delinquencies that are caused by loan-related reasons, then neighborhoodbased policies should focus on housing market dynamics to lower these risks in the future. Analysis results also suggest that the degree of utilization and outcomes of counseling and legal services significantly differ geographically. If distressed homeowners in certain neighborhoods underutilize or end up withdrawing from these services, they may lose the opportunity to resolve their delinquencies and make more rapid progress into foreclosures. This negative resolution process of mortgage defaults could lead to the concentration of foreclosures in certain neighborhoods. Therefore, counseling agencies would need to pay attention to attract more clients from neighborhoods that used to show strong housing market performance but newly

- 24 -

experience mortgage defaults or neighborhoods that have higher share of Black/AfricanAmerican population as well as improve their services targeting such neighborhoods. Despite significant contributions, this study has several shortcomings. First, it uses the sample of CNYCN clients residing in New York City, so the results of this study have the limited capacity of generalization. Second, the universe of CNYCN clients is self-selected and it does not have a comparison sample of households who have delinquent loans but do not receive assistance from the CNYCN service providers. While there are several other counseling agencies that provide similar legal or counseling assistance to the distressed NYC homeowners, this study was not able to account for the utilization of these agencies and their client outcomes due to data limitations. Among CNYCN clients, those who receive a specific type of assistance are also selected by the service provider and by the intake procedures. Thus, the analysis findings are to be interpreted in light of a self-selection problem.

- 25 -

References Allen, R. (2011). “The relationship between residential foreclosures, race, ethnicity, and nativity status”, Journal of Planning Education and Research 31(2): 125-142. An, X., and Bostic, R. (2009). “Policy incentives and the extension of mortgage credit: Increasing market discipline for subprime lending”, Journal of Policy Analysis and Management 28(3): 340-365. Apgar, W. C., Duda, M., and Gorey, R. N. (2005). “The Municipal Cost of Foreclosures: A Chicago Case Study”, Homeownership Preservation Foundation Housing Finance Policy Research Paper Number 2005-1. Agarwal, S., et al. (2010) “Learning to cope: Voluntary financial education and loan performance during a housing crisis”, The American Economic Review 100(2): 495-500. Black, H. A., Boehm, T. P., and DeGennaro, R. P. (2003). “Is there discrimination in mortgage pricing? The case of overages”, Journal of Banking & Finance 27(6): 1139-1165. Bostic R., and Lee, K.O. (2009). “Homeownership: America’s Dream?”, in Insufficient Funds: Savings, Assets, Credit, and Banking among Low-Income Households (Rebecca Blank and Michael Barr, editors), Russell Sage Foundation. Chan, S., Sharygin. C., Been, V., and Haughwout, A. (2012). “Pathways after Default: What Happens to Distressed Mortgage Borrowers and Their Homes?”, The Journal of Real Estate Finance and Economics: 1-38. Chan, S., Gedal, M., Been, V., and Haughwout, A. (2013). “The Role of Neighborhood Characteristics in Mortgage Default Risk: Evidence from New York City”, Journal of housing Economics: 22(2), 100-118. Collins, J. M. (2007). “Exploring the design of financial counseling for mortgage borrowers in default”, Journal of Family and Economic Issues 28(2), 207-226. Collins, J. M., and O’ Rourke, C. M. (2010). “Financial education and counseling—Still holding promise”, Journal of Consumer Affairs 44(3): 483-498. Collins, J. M., and Schmeiser, M. D. (2013). “The effects of foreclosure counseling for distressed homeowners”, Journal of Policy Analysis and Management 32(1): 83-106. CoreLogic. (2012). “CoreLogic Reports More Than 860,000 Completed Foreclosures Nationally in the Last Twelve Months”, March 15, http://www.corelogic.com/about-us/news/corelogicreports-more-than-860,000-completed-foreclosures-nationally-in-the-last-twelve-months.aspx. Courchane, M., and Nickerson, D. (1997). “Discrimination resulting from overage practices”, Discrimination in Financial Services: 133-151. Courchane, M. J., Surette, B. J., and Zorn, P. M. (2004). “Subprime borrowers: Mortgage transitions and outcomes”, The Journal of Real Estate Finance and Economics 29(4), 365392. Doviak, E. and MacDonald, S. (2011). “Who Defaults? Who Goes into Foreclosure?”, http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1802102. Foote, C., Gerardi, K., Goette, L., and Willen, P. (2008). “Subprime Facts: What (We Think) We Know about the Subprime Crisis and What We Don’t”, Public Policy Discussion Papers, Federal Reserve of Boston. Furman Center for Real Estate and Urban Policy. (2007). http://furmancenter.org/files/FurmanCenterHMDAAnalysis_000.pdf Glaeser, E., and Gyourko, J. (2008). “The Case Against Housing Price Supports”, Economists’ Voice, October. - 26 -

Immergluck, D., and Smith, G. (2005). “Measuring the Effect of Subprime Lending on Neighborhood Foreclosures: Evidence from Chicago”, Urban Affairs Review 40(3): 362-389. Immergluck, D., and Smith, G. (2006). “The External Costs of Foreclosure: The Impact of Single-Family Mortgage Foreclosures on Property Values”, Housing Policy Debate 17(1): 5779. Immergluck, D. (2008). “From the Subprime to the Exotic: Excessive Mortgage Market Risk and Foreclosures”, Journal of the American Planning Association 74(1): 1-18. Immergluck, D. (2009). “The Foreclosure Crisis, Foreclosed Properties, and Federal Policy: Some Implications for Housing and Community Development Planning”, Journal of the American Planning Association 75(4): 406-423. Kain, J., and Quigley, J. M. (1975). Housing Markets and Racial discrimination A Microeconomic Analysis. New York: National Bureau of Economic Research. Li, Y., and Morrow-Jones, H. A. (2010). “The impact of residential mortgage foreclosure on neighborhood change and succession”, Journal of Planning Education and Research 30(1): 22-39. Mayer, N. S., et al. (2009). National Foreclosure Mitigation Counseling Program Evaluation: Preliminary Analysis of Program Effects. Washington, DC: The Urban Institute. Report Prepared for NeighborWorks America. New York Times. (2008). “Foreclosure Makes Its Move on Manhattan”, September 5, http://www.nytimes.com/2008/09/07/realestate/07cover.html?pagewanted=all. Nofsinger, J. R. (2012). “Household behavior and boom/bust cycles”, Journal of Financial Stability 8(3): 161-173. Quercia, R., Cowan, S, and Moreno, A. (2004). The Cost-Effectiveness of Community-Based Foreclosure Prevention. Joint Center for Housing Studies of Harvard University. www.jchs.harvard.edu Quercia, R. and Cowan, S. M. (2008). “The Impacts of Community-based Foreclosure Prevention Programs”, Housing Studies 23 (3): 461–483. RealtyTrac. (2012). “2011 Year-End Foreclosure Report: Foreclosures on the Retreat”, press release, January 9, http://www.realtytrac.com/content/foreclosure-market-report/2011-yearend-foreclosure-market-report-6984. Reinhart, C. M., and Rogoff, K. S. (2009). “The Aftermath of Financial Crises”, American Economic Review 99(2): 466-72. Roper Public Affairs and Media. (2005). Freddie Mac, Roper survey asks why more delinquent borrowers don’t call lenders for help. http://www.freddiemac.com Russell, B. D., Moulton, S., and Greenbaum, R. T. (2014). “Take-up of mortgage assistance for distressed homeowners: The role of geographic accessibility”, Journal of Housing Economics. Shlay, A. B., and Whitman, G. (2004), “Research for Democracy: Linking Community Organizing and Research to Leverage Blight Policy”, http://commorg.wisc.edu/papers2004/shlay/shlay.htm United States Department of Housing & Urban Development, Region IX. (1975). Evaluation Report: Effectiveness of Home Ownership Counseling. United States Department of Housing & Urban Development, Office of Policy Development and Research. (1983). Report to Congress on Housing Counseling. Weinstein, N.D. (1980). “Unrealistic optimism about future life events”, Journal of Personality and Social Psychology 39(5): 806–820.

- 27 -

Yinger, J. (1978). “The black-white price differential in housing: Some further evidence”, Land Economics 54(2): 187-206.

- 28 -

Figure 1. Spatial Distribution of CNYCN Clients and Pre-Foreclosure Filings

Source: New York State Banking Department and CNYCN Data Note 1. This shows the distribution of pre-foreclosure filings (PFFs) in all stages during April 2010-February 2011.

Figure 2. Ratio of the CNYCN Clients to Pre-Foreclosure Filings

Source: New York State Banking Department and CNYCN Data

- 29 -

Table 1. CNYCN Clients vs. Other NYC Borrowers NYC Borrowers without preforeclosure filings1 Household Characteristics Loan amount > $250,000 Income < $80,000 Interest Rate >= 8% Adjustable Rate Mortgage (ARM) Asian Black Hispanic FICO Score Monthly Payment Debt-to-Income Ratio Loan delinquent at least 90 days

NYC Borrowers with 90-day Preforeclosure filings1

43.00% 48.10% 10.11% 12.52% 5.41% 9.48% 7.78%

56.00% 43.80% 16.40% 18.70% 4.96% 15.28% 11.36%

All census tracts in NYC Census Tract Characteristics Median Household Income Median House Price in 20002 Median House Price in 2008 Price Appreciation Rate (2006-8) 3 Delinquency Rate4 Vacancy Rate % Blacks % Hispanics

$54,487 $261,291 $552,593 -1.04% 3.59% 8.26% 27.08% 25.67%

CNYCN Clients

82.84% 82.01% 14.01% 35.46% 6.37% 54.16% 22.49% 557.97 $2,513 0.86 58.05% Census tracts of CNYCN clients $53,153 $248,548 $442,046 -7.34% 4.38% 7.85% 34.37% 25.16%

Source: Author’s calculation based on the merged data from various sources Note. 1. Data source: Matched data of Home Mortgage Disclosure Act data (HMDA) in 2004-2008 and Preforeclosure Filings (PFFs) in 2010. 2. Data source: 2000 Census and in 2008$. 3. Dataquick has provided information on the median house price in every ZIP code in New York City for 2006-8. 4. It is the rate of the number of pre-foreclosure filings that have been delinquent over 90 days per owner-occupied units in the census tract during April 2010-February 2011. 5. All other neighborhood characteristics are obtained from the 2005-2009 American Community Survey (ACS).

- 30 -

Table 2. Variation in Triggers of Mortgage Distress by Neighborhood/Client Characteristics Whole sample Predominantly black neighborhoods (%black > 34.37%) Neighborhoods with a higher drop in house price (appreciation rate < -7.34%) Vacancy rate > 7.85% Black clients Hispanic clients Very low-income clients (below 50% of AMI) Low-income clients (50-80% of AMI) Loan delinquent at least 90 days

Economic Hardship 57.75%

Loan-related

Other

13.05%

29.20%

66.95%

17.38%

15.67%

58.71%

13.99%

27.30%

66.64% 55.45% 62.85%

17.48% 13.92% 11.49%

15.88% 30.63% 25.66%

55.86%

12.44%

31.70%

60.69% 58.82%

11.17% 15.35%

28.14% 25.83%

Source: Author’s calculation based on a merged data of CNYCN clients, Dataquick, and 2005-2009 ACS Note. 1. AMI stands for the Area Median Household Income. In this case, the area indicates New York City.

- 31 -

Table 3. Logit Models of the Triggers of Mortgage Distress

Neighborhood Characteristics ln(median household income) % Blacks % Hispanics House price appreciation rate (2006-8) ln(2008 median house price) Vacancy rate Delinquency rate1 Client Characteristics Black client (yes=1) Hispanic client (yes=1) Black client*%Blacks Hispanic client*%Hispanics Primary language is non-English (yes=1) ln(household income) Have a child (yes=1) Married (yes=1) Loan Characteristics ln(loan amount) ln(monthly payment) ARM loan (yes=1) Interest rate at origination Debt-to-income ratio FICO Score Loan delinquent at least 90 days (yes=1) Year dummies of loan origination Pseudo R2

(1) Economic Hardship

(2) Loanrelated

(3) Other

-0.105 -0.250 -0.175 0.406* -0.195 -0.272 1.318

-0.385 0.141 -0.493 -0.843** 0.541 3.565** -3.091

-0.949 1.597 -0.828 0.180*** -2.565 -1.505 11.250

-0.252 0.080 0.435 -0.200 -0.071 -0.286*** -0.044 0.663***

0.736* -0.704* -1.047 1.987* 0.592* 0.235 0.265 0.016

4.119** 2.297** -3.242* -5.498 0.660 0.479 -0.578 -0.294

0.164 0.059 -0.424*** -0.092*** -0.014*** 0.000 -0.167 YES 0.0410

-0.074 0.552* 1.478*** 0.056 0.099 0.000 -0.686*** YES 0.1218

-0.357** 0.979 -0.114 -0.047 -0.837 0.007* -0.226 YES 0.1911

Note. 1. It is defined by the rate of the number of pre-foreclosure filings that have been delinquent over 90 days per owner-occupied units in the census tract. 2. *, **, *** indicate significance at the 0.1, .05, .01 levels, respectively. 3. To account for unobserved heterogeneity between census tracts, all standard errors are clustered at the census tract. 4. All other neighborhood characteristics are obtained from the 2005-2009 ACS.

- 32 -

Table 4. Linear Regression Models of Delinquency and Counseling Receipts (1) Delinquency rate (Ratio of PFFs2 to owner-occupied units) Homeownership Characteristics Rate of PFFs (less than 90 days) Rate of PFFs (90-120 days) Rate of PFFs (120+ days) ln(2008 median house value) House price appreciation rate (2006-8) ln(Median monthly cost with mortgage) ln(Median monthly cost without mortgage) Median owner cost as a percentage of household income (with mortgage) Median owner cost as a percentage of household income (without mortgage) Income Characteristics Poverty rate ln(Median household income) Demographic Characteristics % Blacks % Hispanics Housing Characteristics Homeownership rate Vacancy rate Multi-family housing rate Housing rate built before 1970 Geographic Characteristics Distance to the closest CNYCN agency Borough Dummies (Omitted = Staten Island) Manhattan Bronx Queens Brooklyn Adjusted R2

(2) Rate of counseling receipts (Ratio of the CNYCN clients to PFFs2)

-0.007** 0.004 0.009*** 0.000 0.017**

-0.729*** -0.055 0.597*** -0.020** 0.010 0.027** 0.006 0.072*

0.014

0.051

0.028** 0.004

-0.097** -0.003

0.047*** 0.029***

0.129*** 0.114***

-0.022*** 0.074*** -0.019*** -0.014***

-0.040 0.041 -0.066** -0.026 -0.009***

-0.012*** 0.003 0.007*** 0.006** 0.4459

-0.086*** -0.045** -0.045*** -0.021 0.1761

Note 1. *, **, *** indicate significance at the 0.1, .05, .01 levels, respectively. 2. It is the number of pre-foreclosure filings that have been delinquent over 90 days in the census tract during April 2010-February 2011. 3. To account for unobserved heterogeneity between census tracts, all standard errors are clustered at the census tract.

- 33 -

Table 5. Variation in Outcomes of Foreclosure Counseling Whole Sample

Success

Withdrawal

Other

38.37%

23.38%

38.25%

Predominantly black neighborhoods (%black > 34.37%)

44.32%

32.18%

23.50%

Neighborhoods with a higher drop in house price (appreciation rate < -7.34%)

39.88%

23.78%

36.34%

37.86%

23.00%

38.42%

36.89%

24.69%

33.39%

Hispanic clients

45.48%

21.13%

41.26%

Primary language is not English

36.43%

22.21%

41.36%

Lower-income Client (below 80% of AMI)

33.95%

24.79%

22.52%

Counseling Assistance > 6.94 hrs (average)

68.57%

8.91%

41.90%

Legal Assistance > 3.45 hrs (average)

54.53%

3.57%

50.06%

Neighborhoods with a higher concentration of CNYCN clients1 (> 6.23%) Black clients

Source: Author’s calculation based on a merged data of CNYCN clients, Dataquick, and 2005-2009 ACS Note 1. It is the ratio of the number of CNYCN clients to the number of pre-foreclosure filings that have been delinquent over 90 days in the census tract. 2. About 50% of the clients in the data are currently in counseling so are excluded for this analysis.

- 34 -

Table 6. Logit Models of Counseling Outcomes (1) Success = 1 Neighborhood Characteristics Distance to service provider ln(median household income) % Blacks % Hispanics House price appreciation rate (2006-8) ln(2008 median house price) Delinquency rate1 Concentration of CNYCN clients 2 Client Characteristics Black (yes=1) Hispanic (yes=1) Primary language is non-English (yes=1) ln(household income) Have failed modification (yes=1) Have been defaulted (yes=1) Loan Characteristics ln(loan amount) ln(monthly payment) ARM loan (yes=1) Interest rate at origination Debt-to-income ratio FICO Score Loan delinquent at least 90 days (yes=1) Service Characteristics Hours of legal services Hours of counseling services Duration of being a CNYCN client

(2) Success = 1 (excluding clients withdrawn from the CNYCN program)

0.018 -0.001 -0.619** -0.171 -0.413** 0.191 0.002 2.837

0.033 -0.090 -1.093*** -0.472 -0.186* -0.274 -0.070*** 3.773

0.202 0.764*** -0.706* 0.032 -0.621* 0.207

-0.296 0.369 0.235 0.021 -1.070 0.497*

0.012 0.133 -0.076 0.064 -0.179 0.000 0.391*

-0.200 0.351 -0.010 0.084 -0.398 -0.001 0.044

-0.470** -0.003 0.027

-0.459 -0.005 0.093***

Counseling agency dummies Year dummies of loan origination Pseudo R2

YES YES 0.1722

YES YES 0.3317

Note 1. It is defined by the rate of the number of pre-foreclosure filings that have been delinquent over 90 days per owner-occupied units in the census tract. 2. It is defined by the rate of the number of the CNYCN clients to the number of pre-foreclosure filings that have been delinquent over 90 days in the census tract. 3. *, **, *** indicate significance at the 0.1, .05, .01 levels, respectively. 4. All standard errors are clustered at the census tract.

- 35 -

Appendix 1. Multinomial Logit (MNL) Model of Triggers of Mortgage Distress

Neighborhood Characteristics ln(median household income) % Blacks % Hispanics House price appreciation rate (2006-8) ln(2008 median house price) Vacancy rate Delinquency rate Client Characteristics Black client (yes=1) Hispanic client (yes=1) Black client*%Blacks Hispanic client*%Hispanics Primary language is non-English (yes=1) ln(household income) Have a child (yes=1) Married (yes=1) Loan Characteristics ln(loan amount) ln(monthly payment) ARM loan (yes=1) Interest rate at origination Debt-to-income ratio FICO Score Loan delinquent at least 90 days (yes=1) Year dummies of loan origination Pseudo R2

(1) Economic Hardship

(2) Loanrelated

-0.249 -0.294 -0.378 0.290 -0.112 0.847 0.646

-0.598 -0.146 -0.756 -0.580* 0.412 4.239*** -2.426

-0.134 -0.105 0.264 0.410 0.174 -0.052 0.015 0.833***

0.541 -0.794* -0.779 2.292* 0.770** 0.246 0.266 0.680***

0.182 0.030 -0.032 -0.099 0.254 0.000 -0.332

Note 1. *, **, *** indicate significance at the 0.1, .05, .01 levels, respectively. 2. Baseline outcome is other reasons. 3. All standard errors are clustered at the census tract level.

- 36 -

0.080 0.503 1.422*** -0.031 0.351* 0.001 -0.658*** yes 0.0699

Appendix 2. Variation in Client Outcomes by Service Providers JASA - Legal Services for the Elderly in Queens

Neighborhood Housing Services Staten Island

MHANY Management Inc.

% Asian clients Mean client income Mean Hrs of counseling assistance

11.50% 0.97% 51.03% 19.97% 4.60% $43,852 3.85 hrs

94.44% 0.00% 26.50% 21.27% 8.53% $48,139 25.29 hrs

26.56% 71.88% 56.67% 25.99% 3.29% $60,992 1.11 hrs

Mean Hrs of legal assistance

2.95 hrs

0 hrs

0 hrs

% Clients with Successful outcomes % Clients withdrawn % Black clients % Hispanic clients

- 37 -

Appendix 3. Logit Model of Counseling Outcomes II Withdraw = 1 Neighborhood Characteristics Distance to service provider ln(median household income) % Blacks % Hispanics House price appreciation rate (2006-8) ln(2008 median house price) Delinquency rate Concentration of CNYCN clients Client Characteristics Black (yes=1) Hispanic (yes=1) Primary language is non-English (yes=1) ln(household income) Have failed modification (yes=1) Have been defaulted (yes=1) Loan Characteristics ln(loan amount) ln(monthly payment) ARM loan (yes=1) Interest rate at origination Debt-to-income ratio Loan delinquent at least 90 days (yes=1) Service Characteristics Hours of legal services Hours of counseling services Duration of being a CNYCN client Counseling agency dummies Year dummies of loan origination Pseudo R2

0.017* -0.261** -0.169 -0.656 0.261* -0.126 -1.296 0.000*** -0.249 -0.847*** 0.919*** 0.035 0.507 0.030 -0.065 0.148 0.132 -0.049 0.091 -0.463*** -0.024*** -0.002 -0.008 YES YES 0.2434

Note 1. Credit score is omitted in this analysis because of colline

- 38 -