summary of data and II. Delinquencies and Foreclosures by Race sections

American Economic Journal: Economic Policy 2016, 8(1): 1–27 http://dx.doi.org/10.1257/pol.20140074

1

The Vulnerability of Minority Homeowners in the Housing Boom and Bust†

By Patrick Bayer, Fernando Ferreira, and Stephen L. Ross*

This paper examines mortgage outcomes for a large sample of individual home purchases and refinances linked to credit scores in seven major US markets. Among those with similar credit scores and loan attributes, black and Hispanic homeowners had much higher rates of delinquency and default in the downturn. These estimated differences are especially pronounced for loans originated near the peak of the housing boom. These findings suggest that black and Hispanic homeowners drawn into the market near the peak were especially vulnerable to adverse economic shocks and raise concerns about homeownership as a mechanism for reducing racial disparities in wealth. (JEL D14, J15, R23, R31, R38)

“We’re creating… an ownership society in this country, where more Americans than ever will be able to open up their door where they live and say, welcome to my house, welcome to my piece of property.”

— President George W. Bush, October 2004

Owning a home has long been viewed as an important mechanism for creating and storing wealth. Yet purchasing a home has traditionally required a substan- tial down payment, limiting access to those with a sufficient amount of initial capi- tal (Englehart 1996 and Brueckner 1986). Motivated by this perspective, a number of public policy programs have had an explicit goal of expanding homeownership and many politicians have embraced it as a means of upward mobility.1 Moreover, many commentators have argued that relaxing down payment constraints would help reduce large racial and ethnic gaps in homeownership and, ultimately, lead to a

1 Belsky, Retsinas, and Duda (2005); Herbert et al. (2005); and Quercia, McCarthy, and Wachter (2003) describe efforts to reduce down payment requirements for low-income borrowers, and Di and Liu (2007) find evidence of a reduction in the importance of household wealth in the transition to homeownership in recent years.

* Bayer: Department of Economics, Duke University and National Bureau of Economics (NBER), Box 90097, Durham, NC 27708 (e-mail: [email protected]); Ferreira: The Wharton School, University of Pennsylvania (and NBER), 1466 Steinberg Hall-Dietrich Hall, 3620 Locust Walk, Philadelphia, PA 19104 (e-mail: fferreir@whar- ton.upenn.edu); Ross: Department of Economics, University of Connecticut, 341 Mansfield Rd, Storrs, CT 06269 (e-mail: [email protected]). The authors thank the Ford Foundation, Research Sponsors Program of the Zell/Lurie Real Estate Center at Wharton, and the Center for Real Estate and Urban Economic Studies at the University of Connecticut for financial support. We thank Robert Avery and Erik Hurst for providing detailed com- ments on the paper, as well as the participants in the 2012 January meetings of the Weimer School for Advanced Studies in Real Estate, the NBER Summer Institute, and the Federal Reserve Bank of Atlanta. Gordon MacDonald, Kyle Mangum, and Yuan Wang provided outstanding research assistance. The analyses presented in this paper use information provided by Experian Information Solutions, Inc. Experian is a service mark and registered trademark of Experian Information Solutions, Inc. However, the substantive content of the paper is the responsibility of the authors and does not reflect the specific views of any credit reporting agencies.

† Go to http://dx.doi.org/10.1257/pol.20140074 to visit the article page for additional materials and author disclosure statement(s) or to comment in the online discussion forum.

2 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

reduction in racial and ethnic inequality (Rohe, Van Zandt, and MacCarthy 2002). In fact, racial and ethnic disparities in wealth dwarf those in income. In 1998, for exam- ple, the median wealth of black and Hispanic renters was less than $3,000 dollars and the seventy-fifth percentile wealth was below $10,000 (Herbert et al. 2005).2

The expansion of housing credit from the late 1990s to mid-2000s led to histor- ically high rates of homeownership in the United States for all households, includ- ing a peak of 49.7 percent in 2004:II for blacks and of 50.1 percent in 2007:I for Hispanics, and this increase in homeownership was largely cheered. But as the sub- sequent housing and economic crises developed, however, the risks of homeowner- ship became increasingly obvious. Delinquency and foreclosure rates rose sharply, especially in low-income neighborhoods, and many households not only lost sub- stantial housing wealth but also faced the prospect of lower credit scores (higher borrowing costs) for years to come. A comparison of mortgage delinquencies and foreclosures between 2005 and 2009 provides a particularly stark picture of the differential impact of the downturn by race and ethnicity. Figure 1 shows that while all homeowners had negligible 90-day delinquency and foreclosure rates in 2005 for our sample of seven major markets, high rates of delinquency and foreclosure and large racial and ethnic differences had emerged by 2009. More than one in ten black and Hispanic homeowners in our sample had a delinquent mortgage by 2009, compared to 1 in 25 for white households, and a similar pattern held for foreclosure rates. By 2010, homeownership rates had fallen to 45.6 percent for blacks and 47.5 percent for Hispanics.

While researchers have documented the greater exposure of minority house- holds to employment, income, and health shocks (see, for example, Smith 1995; Altonji and Blank 1999; Shapiro 2004; Shuey and Wilson 2008; and Hoynes, Miller, and Schaller 2012), much less is known about the differential impact of credit and financial shocks, especially in housing markets. We tackle this issue by examining mortgage outcomes by race during the last housing cycle in a diverse set of US housing markets. The main goal of our analysis is to properly estimate the differen- tial delinquency and default by minority homeowners in the housing market bust, accounting for observed differences in credit worthiness, loan characteristics, other demographics, house type, neighborhood, and lender type, such as prime and sub- prime lending.3

Moreover, we test another explanation for high rates of negative credit market outcomes for minority homeowners during the crisis, i.e., the selection of high-risk

2 Oliver and Shapiro (1997) also estimate that the financial assets and net worth of black households nearing retirement (aged 50–64) were only 14 and 20 percent of those of white households, respectively, while the analo- gous figure for income was 67 percent. Also see Deng, Ross, and Wachter (2003); Gyourko, Linneman, Wachter (1999); Duca and Rosenthal (1994); Wachter and Megbolugbe (1992); and Linneman and Wachter (1989) on the role of borrowing constraints in explaining the black-white homeownership gap.

3 The literature also suggests that subprime lending has been an important factor in explaining rising foreclo- sure rates in low income and minority neighborhoods. Gerardi and Willen (2009); Reid and Laderman (2009); Edminston (2009); and Wachter, Russo, and Hershaff (2010) show that the housing market crisis and the preced- ing growth of subprime lending have disproportionately affected low income and minority neighborhoods. More broadly, Mian and Sufi (2009) document large increases in foreclosure rates in neighborhoods that had a large volume of subprime loans and that this growth in subprime loans occurred primarily in neighborhoods with very low income growth.

VoL. 8 no. 1 3Bayer et al.: the VulneraBility of Minority hoMeowners

households into the housing market close to the peak of the housing cycle. If we recognize that households sort into owner-occupancy in part due to both the cost/difficulty of obtaining mortgage credit and the risk of exposure to future adverse events, then, for a given cost of credit, households with low risk of adverse events will be more likely to be homeowners. As the cost of credit falls or the availability of credit expands, more households will become homeowners, and the newest home- owners will be drawn from the remaining high risk pool. To the extent that wealth and liquidity gaps leave minority households especially vulnerable to negative eco- nomic shocks, we expect that those minority households drawn into homeownership following a major expansion of credit are especially likely to default in a subsequent economic downturn.4

There are significant empirical challenges to studying mortgage outcomes by race and ethnicity. Most directly, datasets linking home purchases and mortgage decisions by race to detailed risk factors and subsequent loan performance for a rep- resentative sample of homeowners have been essentially nonexistent. Most research documenting the distribution and performance of high cost or subprime loans have used one of two sources: Analyses using Home Mortgage Disclosure Act (HMDA) data (Calem, Gillen, and Wachter 2004; Avery, Canner, and Cooke 2005; and Apgar, Calder, and Fauth 2004) typically find significant racial and ethnic differences in the

4 For example, Charles and Hurst (2002) suggests that black households may have fewer family resources to draw on; particularly, they document that, conditional on credit score and income, young black adults are less likely to purchase homes due in part to the ability or willingness of parents to provide down payment assistance.

−0.05

0

0.05

0.1

0.15

0.2 U

n co

n d

iti o

n a

l r a

te s

2004 2005 2006 2007 2008 2009

Year

Foreclosures

−0.05

0

0.05

0.1

0.15

0.2

U n

co n

d iti

o n

a l r

a te

s

2004 2005 2006 2007 2008 2009

Year

Delinquencies

Whites Blacks Hispanics

Figure 1. Unconditional Rates of Mortgage Foreclosures and Delinquencies by Race and Year

notes: The figure presents the mean number of 90-day to 180-day mortgage delinquencies and the mean number of foreclosures within the last 12 months by group for each credit report year pooling both the home purchase and refinance samples. Each line represents a different group.

4 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

allocation of high cost credit, but these studies can only control for the extremely limited borrower and loan attributes in HMDA; and proprietary data aggregated across many lenders (Mayer and Pence 2008; Mayer, Pence, and Sherlund 2009; Reid and Laderman 2009; and Fisher, Lambie-Hanson, and Willen 2010) often capture only a select subset of the mortgage market and typically do not contain information on borrower race or ethnicity instead documenting differences at the neighborhood level.

We overcome these issues by first assembling a unique panel dataset that links a representative sample of HMDA data on home purchase and refinance mortgages originated between May and August in the years between 2004 and 2008, to public records data on housing transactions and liens for approximately 270,000 home- owners in seven distinct metropolitan housing markets: Chicago, IL Consolidated Metropolitan Statistical Area (CMSA); Cleveland, OH Metropolitan Statistical Area (MSA); Denver, CO MSA; Los Angeles, CA CMSA; Miami-Palm Beach corridor; San Francisco, CA CMSA; and Washington DC-Baltimore MD suburban corridor. These data contain information on all liens as well as the name and address of the individual purchasing the housing unit or refinancing their mortgage and in many cases the name of the individual’s spouse, in addition to information about house prices, housing characteristics, and neighborhood or census tract of the housing unit.

We then provided this rich sample to one of the major credit rating agencies. The credit rating agency used the name and address to match borrowers to archival credit reporting data, providing in each year a Vantage credit score plus detailed credit line information from each individual’s report. This matched mortgage-housing unit-credit data provides detailed information on the financial circumstances of each homeowner at the time of mortgage origination, the terms of the mortgage includ- ing whether it was a high-cost loan, and the performance of the mortgage over the subsequent years.5

Our data has three critical advantages relative to other data that has been ana- lyzed in the recent literature. First, all of the information is matched at the indi- vidual rather than neighborhood level. This allows us to examine how the pricing and performance of loans varies with the race of the homeowner separately from the racial composition of the neighborhood. Second, our dataset includes a repre- sentative sample of all mortgages that were originated in these markets in the time period. Finally, our dataset contains controls for all of the standard risk factors that are typically considered in mortgage underwriting and pricing models.

Our empirical results show that black and Hispanic households are more likely to become delinquent and default on their mortgages than white households with similar credit scores, loan characteristics, demographics, house type, neighborhood, and lender. Blacks are 3.0 and 1.5 percentage points more likely to enter foreclo- sure than whites in our home purchase and refinance samples, respectively; while Hispanics are 3.1 and 2.1 percentage points more likely to enter foreclosure in the

5 Bhutta and Canner (2013) provide the one exception in the literature that documents racial and ethnic differ- ences in foreclosure controlling for borrower credit scores by merging HMDA data with data from a credit reporting agency, but they are only able to control for the overall amount of the primary mortgage because they do not observe either subordinate debt or the value of the home securing that debt.

VoL. 8 no. 1 5Bayer et al.: the VulneraBility of Minority hoMeowners

same samples. These differences are substantial relative to the unconditional 5.2 and 3.7 percent average foreclosure rates in the home purchase and refinance samples, and represent approximately one-third of the unconditional racial and ethnic differ- ences in the home purchase samples and one-half of the unconditional differences in the refinance sample. Most of this decline in the unconditional racial and ethnic dif- ferences arise from including general controls for observable mortgage-risk factors in our econometric model, as opposed to controls associated with subprime lending.

Next, we include in our model a series of contemporaneous controls that represent information that may not have been available to the lender at the time of mortgage origination. These controls include county by credit year fixed effects, contempora- neous measures of negative equity based on county-level price indices, and the inter- action of the negative equity variables with race and ethnicity specific measures of county employment rates. These controls decrease conditional racial differences in foreclosure by approximately 50 percent and decrease conditional ethnic differences in foreclosure by between 10 and 20 percent. The decline in racial and ethnic differ- ences arises entirely from the use of race specific employment rates. Controlling for county-level employment rates does not erode the racial and ethnic differences in foreclosure. The foreclosure and delinquency differences are explained by the lower rates of employment among blacks, and to some extent Hispanics, and the higher rates of foreclosure among all households with negative equity and in counties with low employment rates for their group.

We then attempt to identify which borrowers are responsible for the observed racial and ethnic differences in delinquency and foreclosure. For African American homebuyers, we find that all racial differences are concentrated among borrowers with subprime credit scores and high debt expense to income ratios. Ethnic differ- ences are somewhat higher for high debt expense to income ratio loans, but sub- prime credit does not seem to matter. Furthermore, ethnic differences in foreclosure are concentrated among Hispanic borrowers in counties with low employment rates for Hispanics. While as noted earlier within county employment rate differences between Hispanics and whites does not explain ethnic differences, the effect of exposure to low employment rates has a much larger effect on Hispanic homeowners than on white homeowners. A similar, but more modest phenomenon, is observed for black homeowners. The larger impact of county-level employment rates on the foreclosure risk of minority borrowers is consistent with those borrowers being less able to manage the economic distress created by economic downturns, even after controlling for differential exposure to loans with risky terms and contemporaneous levels of negative equity and employment risk.

We already noted that recent home purchases have higher foreclosure and delin- quency rates than the refinance sample—which by definition are composed of homeowners that were drawn into the housing market much earlier. As a final test of our hypothesis concerning the vulnerability of recent minority homebuyers, we examine whether the timing of the selection into the housing market has an effect on delinquency and foreclosure. We find that racial and ethnic differences are larg- est for home purchase originations in 2006, the peak of the housing boom accord- ing to the Case-Shiller price index, even after controlling for the subsequent higher rates of negative equity for borrowers who purchased near the peak of the housing

6 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

market. The pattern is similar, but less strong, for the refinance sample. We also examine racial and ethnic differences for a subsample of refinance mortgages that were originally purchased between 1998 and 2008 and subsequently refinanced in our sample period. For this subsample, racial and ethnic differences in foreclosure are tiny for homes that were originally purchased from 1998 to 2000, but substantial for homeowners who originally purchased their homes between 2001 and 2007, i.e., those drawn into the market during the credit expansion, with the largest effects arising between 2004 and 2007. These home purchase timing effects are largest for minority borrowers with high debt-to-income ratios and who are exposed to low employment rates.

Taken together, our results provide strong evidence that minority households drawn into homeownership late in the recent housing market boom were especially vulnerable in the subsequent downturn in ways that are not explained by borrower characteristics observed at origination, exposure to different lenders or loans, and performance of local labor and housing markets. These results call into question the idea of encouraging homeownership as a general mechanism for reducing racial disparities in wealth. To the extent that increases in homeownership are driven by the entry of especially vulnerable households into the owner-occupied market, such a push may backfire, leaving vulnerable households in a difficult financial situation and adversely affecting their wealth and credit-worthiness for years.

There are important caveats to our results. First, we do not directly observe house- hold savings or wealth, which might help to explain why some households are able to manage adverse economic shocks more easily than others. Future work could more explicitly measure whether the estimated differences in mortgage outcomes by race reported here are primarily due to differences in wealth. Another limitation of our analysis is that our study period only covers the recent housing crisis. While Hoynes, Miller, and Schaller (2012) show similar patterns by race for negative labor market shocks in a number of recent recessions, there is a chance that the recent housing market boom and bust may be unique in history.

The remainder of this paper proceeds as follows. The next section presents the data used in the analysis. Section II presents the main results from our analyses of credit outcomes, and Section III shows heterogeneity in those estimates. Section IV presents additional analyses by the year of mortgage origination or home purchase. Section V concludes.

I. Data

Our dataset is based on public Home Mortgage Disclosure Act (HMDA) data from between 2004 and 2008 and proprietary housing transaction/lien and asses- sor’s databases purchased from DataQuick.6 We begin with a convenience sample of seven major housing markets where DataQuick has information on refinance mort- gages going back to at least 2004: Chicago, IL CMSA; Cleveland, OH MSA; Denver,

6 Data provided by DataQuick Information Systems, Inc. www.dataquick.com. The property transaction data is collected by DataQuick or by intermediaries from county assessor’s offices and contains a population of all sales and liens of all types including refinance mortgages, home improvement loans, and home equity lines of credit.

VoL. 8 no. 1 7Bayer et al.: the VulneraBility of Minority hoMeowners

CO MSA; Los Angeles, CA CMSA; Miami-Palm Beach corridor; San Francisco, CA CMSA; and Washington DC-Baltimore MD suburban corridor. We restrict our HMDA data to home purchase or refinance mortgages on owner-occupied, 1–4 fam- ily properties. In the DataQuick sample, we eliminate non-arm’s length transactions, transactions where the name field contains the name of a church, trust, or where the first name is missing, and transactions where the address could not be matched to a 2000 census tract or the zip code was missing (this last step eliminates very few records due to the high quality of the name and address records in the assessor files). The HMDA and DataQuick data are then merged based on year, loan amount, name of lender, state, county, and census tract. We obtain high quality matches for approx- imately 50 percent of our HMDA sample. As we show in Table 1, this criteria does not impact the representativeness of our final sample.

Next, we draw a sample of mortgages to provide to a credit-reporting agency. These mortgages were sampled from May through August so that the March 31 archival credit report for the year of the mortgage provides appropriate informa- tion on the borrowers’ credit quality prior to obtaining the mortgage. We oversam- ple mortgages to minority borrowers, mortgages to white borrowers in minority or low-income neighborhoods, and high cost mortgages as designated in HMDA as high rate spread loans. In order to maximize the number of minority loans given the likelihood of sample saturation, we first draw the following oversamples based on race and ethnicity: 500 in each site, year, and group (400 for 2004)7 selected ran- domly from mortgages to black borrowers, mortgages to Hispanic borrowers, and mortgages to white borrowers in minority or low-income neighborhoods. We then split the remaining sample into rate spread and nonrate spread loans drawing 1,000 borrowers associated with rate spread loans in each year and site (800 for 2004) and 2,714 borrowers (2,286 for 2004) from the nonrate spread sample in each year and site. Weights are developed based on the probability of selection, and each site receives equal weight in the pooled sample.8

This sample is provided to Experian Information Solutions, Inc.,9 who matches the name and address of each borrower and co-borrower to archival credit report data from March 31 preceding the mortgage transaction and March 31 for every year that follows this transaction through 2009. Our match rate for the pre-mortgage archive is 81.4 and 84.5 percent in the home purchase and refinance samples, respectively. For years following the mortgage, the match rate rises by 4 to 5 percentage points. In many cases, these individuals also may not have been found by the credit reporting agencies when the lender requested a report, in which case lack of a score matches

7 The smaller sample in 2004 is driven by budgetary restrictions because costs depend upon the number of records, and earlier years imply more records because we follow every borrower from the origination year until 2009.

8 The sampling is explicitly based on eight strata for each site: black borrowers, Hispanic borrowers, white borrowers in minority or low-income neighborhoods, and all other borrowers divided into rate spread and non- rate spread loans. All loans from the same strata and year receive equal weight. We have a convenience sample of housing markets so it would be inappropriate to weight based on the number of mortgages. In any stratified sam- pling scheme, Los Angeles, which dominates our sample in terms of total number of HMDA mortgages, would be selected with certainty while housing markets like Denver and Cleveland would be assigned to a stratum with other similarly sized and located metropolitan areas, and if chosen, would receive a higher weight (offsetting the smaller number of mortgages) based on the probability of being selected from the stratum.

9 Experian is a service mark and registered trademark of Experian Information Solutions, Inc.

8 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

the information that the lender would have had when approving and pricing the loan, but lenders can enter by hand additional information that is not available to us such as social security number or previous addresses.10

Table 1 illustrates the impact of our match process on the sample mean on race and ethnicity of the borrower, whether the primary mortgage is a high-cost loan or rate-spread loan, which is defined as a 3 percentage point or more spread between the mortgage Annual Percentage Rate (APR) and the interest rate on ten-year treasury notes (the APR includes both the interest or note rate on the loan and the effect of closing costs on the cost of credit), loan amount, family income of the bor- rower, and census tract variables including median income, percent black, Hispanic, and Asian residents, and percent of properties owner-occupied. The first column shows the mean for the entire HMDA sample for our seven sites where each site receives equal weight in the mean. The second column shows the mean for our HMDA-DataQuick match, and the third column restricts our sample to mortgages between May and August. The fourth column shows the weighted mean for the sam- ple of mortgages that was provided to the credit reporting agency. The last column in Table 1 shows the weighted means on these common variables for just the subsam- ple where the name and address was matched to the minimum amount of credit line data in order to generate a record. The sample composition is quite stable except for a moderate decline in share white and moderate increase in loan amount between

10 For home purchase mortgages, we only observe the address of the new housing unit, but in practice this does not present a major problem for the credit data match because the archival data can be matched based on current and several past addresses and in practice we observe only a small difference between the home purchases and refinance match rate.

Table 1—Sample Selection of HMDA Variables

Variable Mean

  HMDA

full sample

  DQ high quality match

  DQ high quality match

May–Aug.

  Sample, weighted

  Experian matched sample, weighted

Black 0.111 0.116 0.114 0.114 0.112 Hispanic 0.174 0.194 0.193 0.192 0.185 Asian 0.075 0.086 0.085 0.086 0.089 White 0.678 0.601 0.605 0.605 0.611 Loan amount (in 1,000s) 247 271 274 274 278 Applicant income (in 1,000s) 107 105 105 106 106 Tract median income (in 1,000s) 59.1 59.6 59.7 59.7 60.4 Tract pct. Black 0.126 0.116 0.115 0.115 0.113 Tract pct. Hispanic 0.169 0.165 0.164 0.165 0.163 Tract pct. Asian 0.063 0.065 0.065 0.065 0.066

Observations 9,345,709   4,002,996   1,459,468   273,589   238,785

notes: This table presents the means for all Home Mortgage Disclosure Act (HMDA) home purchase and refinance mortgages in our seven market areas between 2004 and 2008 where each market is given equal weight in the means. The second column presents the means for the subsample where we have a high quality match between HMDA and the housing transaction file based on loan amount, type of loan, census tract and lender name, and the third column is based on further restricting the sample to mortgages originated between May and August. The fourth column presents means for the stratified sample that was merged to the credit history data weighted by the inverse of the sampling probabilities again with equal weights for each market, and the fifth column presents weighted means for the subsample that was successfully merged to an established credit history prior to mortgage origination. Standard deviations are available upon request.

VoL. 8 no. 1 9Bayer et al.: the VulneraBility of Minority hoMeowners

columns 1 and 2 associated with the difficulty of matching lender names between HMDA and the DataQuick provided assessor files. While our HMDA-DataQuick match algorithm loses 50 percent of the HMDA mortgages, the composition of the match sample is quite similar to the composition of the population of mortgages, and the other aspects of our sample construction have virtually no impact on the composition of mortgages over key attributes.11

Table 2 shows the weighted means for our final samples of post-mortgage credit reports for the home purchase and refinance subsamples that were successfully merged to pre-mortgage credit report data.12 The first two columns show the mean and standard deviation for our sample of home purchase mortgages, and the last two columns show these values for refinance mortgages. From the contemporaneous credit history data, we obtain number of 90-day to 180-day mortgage delinquencies and foreclosures in the last 12 months. From the pre-mortgage data, we obtain these same outcomes prior to the mortgage and the borrowers’ (or if unavailable co-bor- rower’s) Vantage score. The first credit report observation following the mortgage is used to obtain monthly mortgage payment, which when combined with HMDA income is used to calculate the mortgage payment to income ratio.13 The monthly mortgage payment is combined with debt payments from the pre-mortgage credit data and HMDA income to calculate debt payment to income ratio. Finally, age is observed for many borrowers and co-borrowers in the credit history files.

The HMDA data contains whether the loan is a high cost or exceeds a standard rate spread above treasury rates, borrower race and ethnicity, gender, loan amount, applicant income, and whether a co-borrower is present. We use the loan amount to calculate whether the loan is nonconforming or a jumbo loan, i.e., too large to be purchased by the Government Sponsored Enterprises; and based on census tract, we observe tract racial and ethnic composition, income, poverty rate, share owner-occupied, and the value of median rents to median home price, which is viewed as a proxy for anticipated housing price appreciation. From the match with transaction data, we observe the presence and size of subordinate liens, whether the liens are fixed or variable rate mortgages, the loan-to-value ratio based on sales price for home purchase mortgages and on an estimated value based on either pre- vious sales price14 or assessed value for refinance mortgages when a previous sale is unobserved,15 and detailed property attributes including whether single family home, a condominium, and number of units on the property.

11 The key factor limiting the match rate is the lender name because the lender of record in the local assessor’s data often differs from the respondent in HMDA. Less restrictive match criteria can yield a match rate around 90 percent, but in order to be conservative, we restricted ourselves only to instances where we successfully match on lender name.

12 The sample size in Table 2 is substantially larger than the sample size in the final column of Table 1 because each mortgage in Table 1 will have one post mortgage credit report observation for every year between the year after origination and the final year of our data in 2009.

13 The mortgage payment for the current mortgage is only observed in the credit line data from the year follow- ing the mortgage. However, in most instances, borrowers who are matched by the credit reporting agency prior to the mortgage are also matched in the following year.

14 We use our extensive housing transaction data to develop both a hedonic and repeat sales quarterly price index for each county. When we observe a previous sale of the property, we simply adjust that earlier sales price to estimate current value based on the hedonic index. However, the repeat sales index yields quite similar estimates.

15 When a previous sale is not observed, we use the county assessment and adjust that value by the average ratio of sales price to assessed value for that county and quarter, see Clapp, Nanda, and Ross (2008). In California, our

10 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

refinance sample is restricted to mortgages where a previous purchase is observed because property assessments are uninformative as to the value of the underlying property. This restriction is feasible because the DataQuick data in California contains transactions back to the late 1980s.

Table 2—Descriptive Statistics

  Purchase sample   Refinance sample

  Mean SD Mean  SD

credit data Mortgage delinquency 90–180 days 0.035 0.214 0.030 0.196 Foreclosure 0.053 0.319 0.037 0.250 Delinquency prior to mortgage 0.001 0.025 0.002 0.046 Foreclosure prior to mortgage 0.001 0.037 0.006 0.082 Vantage score 781.3 104.0 775.0 109.8 Mortgage payment to income ratio 0.256 0.280 0.247 1.064 Debt payment to income ratio 0.321 0.336 0.348 1.191 Borrower age 27.1 23.1 34.3 24.8

HmDA data   Rate spread 0.151 0.358 0.168 0.374 American Indian 0.003 0.054 0.003 0.056 Asian 0.097 0.296 0.086 0.280 Black 0.089 0.284 0.126 0.331 Hispanic 0.195 0.397 0.182 0.386 White 0.616 0.486 0.604 0.489 Male 0.643 0.479 0.643 0.479 Female 0.355 0.478 0.355 0.478 Loan amount (in 1,000s) 105.7 106.1 100.6 108.8 Applicant income (in 1,000s) 285.9 210.1 261.6 199.2 Co-borrower present 0.354 0.478 0.476 0.499 Jumbo loan 0.280 0.449 0.189 0.392 Tract median income (in 1,000s) 60.04 23.43 60.44 23.54 Tract share black 0.100 18.340 12.273 21.652 Tract share Hispanic 0.162 20.210 17.116 20.573 Tract share Asian 0.065 10.252 7.107 10.756 Tract share owner occupant 0.683 23.892 69.202 22.749 Tract share in poverty 0.077 7.273 7.962 7.491 Tract rent/price 0.0045 0.0019 0.0044 0.0017

DataQuick data   Loan-to-value ratio 0.885 0.242 0.576 0.416 Subordinate lien 0.428 0.495 0.016 0.127 First lien adjustable rate 0.515 0.500 0.478 0.500 Condo 0.218 0.413 0.139 0.346 Mobile 0.001 0.035 0.001 0.034 Single family 0.774 0.418 0.837 0.369 Lot size (sf in 1,000s) 15.67 575.97 15.05 397.86 Unit square feet (in 1,000s) 1.80 26.19 1.77 20.12 Number of bathrooms 1.99 6.60 2.00 1.15 Number of bedrooms 2.17 6.85 2.04 1.77 Number of stories 1.17 1.58 1.22 1.33 Units in building 1.41 14.83 1.49 19.88

Sample size 327,124   306,213

notes: The first two columns contain means and standard deviations for our sample of post-mortgage credit reports for each home purchase mortgage borrower for which a credit score was observed prior to mortgage origination. The last two columns contain the same information for the post-mortgage reports of mortgage refinancers.

VoL. 8 no. 1 11Bayer et al.: the VulneraBility of Minority hoMeowners

II. Delinquencies and Foreclosures by Race

In this section, we present empirical estimates for models of two housing market outcomes, delinquencies, and foreclosures. Each mortgage origination contributes one or more observations based on the origination date. Specifically, the 2004 origi- nations contribute housing market outcome observations in 2005, 2006, 2007, 2008, and 2009, while each 2008 origination contributes only a single post-mortgage out- come for 2009. We write those outcomes d for individual i , of race r , at time of origination s , calendar time t , and MSA m , as:

(1) d irstm = β r ir + γ X is + δ st + θ m + ε irstm ,

where r includes dummies for blacks, Hispanics, and Asians, so all estimates are relative to white households. X is includes the set of pre-mortgage origination vari- ables reported in Table 2.16 Our design relies on the fact that we are controlling for almost all characteristics of borrowers, houses, and mortgages that were observable by lenders at time of origination. Especially noteworthy is the information about credit scores, which in principle means that we will compare housing outcomes between blacks and whites, for example, with similar creditworthiness and with similar choices of house type, neighborhoods, and loan characteristics. Finally, the model also includes year of origination by year of the credit profile indicators δ st to deal with common trends, as well as housing market fixed effects.

Next, we expand our set of underwriting controls X is to include a variety of vari- ables intended to capture the potential influence of the subprime mortgage market on observed racial and ethnic differences. The extended model includes a dummy for whether the loan is a high-cost loan or rate-spread loan, as defined in the data section above. Since subprime lending shares are highly concentrated in a subsam- ple of lenders, we include lender fixed effects, and we also include neighborhood or census tract fixed effects because subprime lending is also potentially concentrated in certain cities and neighborhoods. Finally, we include additional controls to rec- ognize that the impact of key loan terms on borrower outcomes may vary between prime and subprime borrowers. We identify borrowers with Vantage scores below 701 as subprime borrowers17 and then interact the subprime dummy with dummy variables associated with key thresholds of loan-to-value ratio, debt-to-income ratio, and mortgage payment to income ratio,18 as well as with the presence of subordinate debt and whether the primary mortgage has a fixed or adjustable interest rate.

16 The loan-to-value ratio is included as intervals or bins below 0.6, 0.6 to 0.8, 0.8 to 0.84, 0.85 to 0.89, 0.90 to 0.94, 0.95 to 1.00, 1.00 to 1.04, and 1.05 and above. The Vantage scores are included as a series of dummy variables based on 20 point bins. The mortgage payment and debt-to-income ratios are also divided into bins. The bins vary in size. For mortgage payment to income ratios, the smallest bins are 0.02 around the traditional secondary market criteria of 0.28, and for total debt payment to income ratios the smallest bins are 0.03 around the threshold of 0.36.

17 The credit reporting agencies that developed the Vantage score algorithms describes scores below 701 as nonprime. Further, a Vantage score of 701 is comparable to a FICO score of 660 in that in both cases approximately 30 percent of individuals have credit scores below these thresholds.

18 The loan-to-value thresholds used are 0.80, 0.90, 0.95, and 1.00; the debt-to-income thresholds used are 0.36 and 0.45; and the mortgage payment to income ratio thresholds used are 0.28 and 0.33.

12 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

We also consider a model that includes contemporaneous controls in order to capture changes in the economic environment during the sample period

(2) d irstm = β r ir + γ X ist + δ st + θ mt + ε irstm .

Specifically, the model is extended to include housing market by current year fixed effects, contemporaneous controls for negative equity, and controls for employ- ment risk. The controls for negative equity are based on original loan-to-value ratio scaled by changes in the price level based on a county-level hedonic price index that was estimated using all single family transactions in the county during the sample period. We create a dummy for whether the current loan-to-value ratio is between 1.0 and 1.1, 1.1 and 1.3, 1.3 and 1.5, and above 1.5. The literature on mortgage foreclosure and default suggests that the likelihood of foreclosure and the effect of negative equity on foreclosure will vary with the employment outcomes of home- owners. We do not have individual information about employment. Instead, we cre- ate measures of local employment and unemployment rates for prime age (25–54) males for each county, year, and racial/ethnic group using the public use sample of the American Community Survey, where the employment rate is calculated as the fraction of males in that group who reported being employed last week. We interact this employment measure with the negative equity dummy variables. One caveat is that Gyourko and Tracy (2013) demonstrate substantial attenuation bias in credit outcome models when county-level employment rates are used to proxy for actual future employment outcomes. In our case, however, we find that county employment rates (at least as a proxy for the risk of unemployment) have strong explanatory power for differential rates of mortgage delinquency and foreclosure.

The first and fifth columns of Table 3 present the unconditional racial and ethnic differences in 90–180 day mortgage delinquency and in foreclosure filing, respec- tively, for the samples of home purchase and refinance mortgages. The estimates for the home purchase sample are shown in panel A and the estimates for the refi- nance sample are shown in panel B. We find massive differences for both black and Hispanic borrowers of 8.6 and 3.0 for black home purchase and refinance borrowers and of 9.1 and 4.3 for Hispanic borrowers in the probability of facing a foreclosure filing, as compared to the 5.2 and 3.7 percent average foreclosure rates in the home purchase and refinance samples, respectively. Both overall delinquency and fore- closure rates and racial and ethnic differences in those rates are larger in the home purchase sample. One possible explanation for this difference is that, on average, the homes in the refinance sample were purchased earlier. Those homeowners may have had more time to accumulated housing equity prior to the housing market downturn, or were generally less vulnerable to housing shocks. We will explore the timing of house purchase in more detail in Section IV.

The next two columns of estimates for the delinquency and foreclosure models include first the detailed underwriting variables, and then in the next column those controls plus the additional controls for subprime lending, including the lender and neighborhood fixed effects (see Appendix Tables A1 and A2 for a complete list of parameter estimates). Blacks are 3.0 and 1.5 percentage points more likely to enter foreclosure than comparable whites in our home purchase and refinance samples,

VoL. 8 no. 1 13Bayer et al.: the VulneraBility of Minority hoMeowners

respectively; while comparable Hispanics are 3.1 and 2.1 percentage points more likely to enter foreclosure than whites in the same samples after controlling for risk factors observable at the time of the loan. Comparing columns 1 and 5 to columns 3 and 7 indicates that these differences represent approximately one-third of the uncon- ditional racial and ethnic differences in the home purchase sample and one-half of the unconditional differences in the refinance sample. For both samples, both out- come variables, and both groups, the main erosion of racial and ethnic differences in credit market outcomes arises from the inclusion of standard underwriting controls. The inclusion of lender and neighborhood fixed effects and additional controls for subprime lending had almost no effect on the estimated racial and ethnic differences with one exception. The control for whether the loan was high cost or a rate spread loan accounts for the entire reduction in the estimates between columns 2 and 3 and columns 6 and 7 (see Appendix Table A2). Significantly, the effect associated with having a high cost loan cannot simply arise from the higher mortgage payments associated with these loans because our model contains detailed controls for the housing and total debt expense to income ratios faced by households. Estimates for Asians are also statistically significant, but substantially smaller than the differences observed for African Americans and Hispanics and not economically meaningful.

Table 3—Estimated Differences in Mortgage Outcomes—Delinquency and Foreclosure Rates

Delinquency Foreclosure

Race Unconditional Underwriting Subprime Contempor-

aneous Unconditional Underwriting Subprime Contempor-

aneous

Panel A. Home purchase sample Black 0.060*** 0.030*** 0.023*** 0.015*** 0.086*** 0.042*** 0.030*** 0.017***   (0.002) (0.002) (0.002) (0.003) (0.003) (0.004) (0.004) (0.004) Hispanic 0.054*** 0.023*** 0.019*** 0.015*** 0.091*** 0.038*** 0.031*** 0.025***   (0.001) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.003)

Sample size 331,608 331,608 331,608 330,912 331,608 331,608 331,608 330,912 r2 0.013 0.057 0.098 0.108 0.016 0.074 0.122 0.137

Panel B. refinance sample Black 0.030*** 0.016*** 0.012*** 0.005** 0.030*** 0.017*** 0.015*** 0.007**   (0.001) (0.002) (0.002) (0.002608) (0.002) (0.002) (0.003) (0.003) Hispanic 0.031*** 0.015*** 0.013*** 0.011*** 0.043*** 0.024*** 0.021*** 0.019***   (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002)

Sample size 309,137 309,137 309,137 308,459 309,137 309,137 309,137 308,459 r2 0.005 0.046 0.090 0.096 0.005 0.045 0.095 0.102

notes: This table presents the estimates for number of 90-day and 180-day mortgage delinquencies and number of foreclosures within the last 12 months for the samples described in Table 2 based on the home purchase and refi- nance samples, panels A and B, respectively. The first column for each outcome presents unconditional differences, the second column presents results conditional on detailed controls for pre-origination Vantage credit score, loan to value ratio, mortgage and total debt payment to income ratios, whether interest rate is adjustable, presence of sub- ordinate debt, whether a jumbo loan, borrower income, race, ethnicity, gender, age, presence of coborrower, census tract demographics and detailed unit attributes, as well as origination year by credit year fixed effects. The third col- umn presents estimates after adding controls for rate spread loans, lender fixed effects, tract fixed effects, and addi- tional subprime controls, and the fourth column presents estimates after adding county-by-year fixed effects plus contemporaneous controls for negative equity and employment rates. Standard errors (in parentheses) are robust to heteroskedasticity, and clustered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

14 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

The final pair of columns, columns 4 and 8, in Table 3 presents the estimates for models that include a series of contemporaneous controls that represent information that may not have been available to the lender at the time of mortgage origination. These controls further erode the delinquency and foreclosure differences decreasing racial differences in foreclosure by approximately 50 percent and decreasing ethnic differences in foreclosure by between 10 and 20 percent. The decline in racial and ethnic differences arises entirely from the use of race-specific employment rates. Racial and ethnic differences are essentially unchanged in models that simply con- trol for county by year fixed effects or control for the overall county-level employ- ment rate.19 The foreclosure and delinquency differences that were not explained by standard mortgage-risk variables are in substantial part explained by the lower rates of employment among blacks during the recent crisis, and to some extent among Hispanics, and the higher rates of foreclosure among all households with negative equity and in counties with low employment rates for their group.20

As noted above, racial and ethnic differences in delinquency and foreclosure in the refinance are smaller than the comparable differences in the home purchase sample. Racial and ethnic differences continue to be higher in the home purchase sample even after controlling for mortgage-risk factors. Further, in columns 4 and 8, the larger differences in the home purchase sample cannot be explained simply by the increased time available to homeowners in the refinance sample to accumulate housing equity prior to the crisis because these models explicitly control for current loan-to-value ratios. On the other hand, since foreclosure rates in general are higher in the home purchase sample, the unexplained differences are a larger share of the unconditional racial and ethnic differences in the refinance sample than in the home purchase sample. In columns 3 and 7, where we control for mortgage-risk factors, racial and ethnic differences are about one-third of the average rate of foreclosure in the home purchase sample, and one-half of the average rate of foreclosure for the refinance sample. For the models that control for contemporaneous factors, the pat- tern is the same, but the relative magnitude of the unexplained differences are much smaller for the home purchase sample.

III. Heterogeneity in the Race Differentials

We first present the time pattern of these race differentials in Figure 2. We esti- mate these parameters by interacting the race dummies with calendar year indica- tors. The top panel of the figure shows racial and ethnic differences for the home purchase sample and the bottom panel of the figure shows these differences for the refinance sample. Each figure contains four lines representing the racial differ- ences in each year unconditionally (circle), conditional on traditional risk factors

19 Whether race-specific or overall-county employment rates are used in the model, the estimates indicate that negative equity and employment rates have powerful and interactive effects. The estimates on the negative equity dummy variables are large and positive, while the estimates on the interactions with employment rate are nega- tive and similar in magnitude suggesting that being in negative equity has little or no effect on foreclosure when employment rates are high, but becomes increasingly important as employment rates fall. See Appendix Table A3.

20 The r2 for these models are not very high in large part because future adverse outcomes are heavily driven by events that happen after the borrower obtains their mortgage, but also because the fit of linear models tends to be lower when events are infrequent.

VoL. 8 no. 1 15Bayer et al.: the VulneraBility of Minority hoMeowners

(diamond), conditional on subprime controls (square), and conditional on contem- poraneous controls (triangle). While the inclusion of additional controls erodes the level of racial and ethnic differences, the pattern is the same overall for all models and both groups. Racial and ethnic differences are relatively small or nonexistent beginning in 2005, and then grow rapidly as the crisis proceeds, typically peaking at the end of our sample in 2009.

Unconditional model

Add subprime controls

Add risk factor controls

Add contemporaneous controls

−0.05

0

0.05

0.1

0.15

0.2 P

ro b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2005 2006 2007 2008 2009

Year

Blacks

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2005 2006 2007 2008 2009

Year

Hispanics

Panel A. Purchase sample

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2005 2006 2007 2008 2009 Year

Blacks

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2005 2006 2007 2008 2009 Year

Hispanics

Panel B. Re�nance sample

Figure 2. Black and Hispanic Probability of Foreclosure Relative to Whites by Calendar Year, Home Purchase, and Refinance Samples

notes: The figure presents estimated group differences in the number of foreclosures within the last 12 months by credit report year separately for the home purchase and refinance samples. Each line represents the estimated differ- ences for a unique model based on a specific set of regression controls as discussed in relation to Table 3.

16 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

Next we attempt to identify which borrowers are responsible for the observed racial and ethnic differences in delinquency and foreclosure. In order to accom- plish this, we reestimate the final models in Table 3, columns 4 and 8, including interactions with key risk factors. We consider three mortgage-risk factors by inter- acting race and ethnicity with dummy variables for whether the borrower has a subprime credit score (Vantage score below 701), whether the loan is a high cost or rate spread loan, and whether the debt expense to income ratio is above levels that were required during this time to be considered for purchase by the Government Sponsored Enterprises, FreddieMac, and Fannie Mae (a ratio above 0.45). We also estimate models interacting the race and ethnicity dummies with one minus the group specific, county-level employment rate described above. The models use one minus the employment rate so that the estimated coefficients on the black and Hispanic dummy variables are for racial and ethnic differences where employment rates are 100 percent, as opposed to 0 percent, which would be far out of the sample range.

These estimates are shown in Table 4 with the results for the home purchase sam- ple presented in panel 1 and for the refinance sample in panel 2. Columns 1 and 4 present the baseline results from columns 4 and 8 in Table 3. Columns 2 and 5 pres- ent the results where race and ethnicity is interacted with the mortgage-risk factors. For blacks, racial differences in foreclosure and delinquency are higher in the home purchase sample for borrowers with subprime credit scores and high debt-to-income ratios and in the refinance sample for borrowers with rate spread loans and high debt-to-income ratios. Significantly, no racial differences remain for black borrow- ers who do not have any of these risk factors. For Hispanics, ethnic differences in delinquency and foreclosure are larger among borrowers with high debt-to-income ratios in both samples, but these effects are smaller than the effects for blacks, and most of the ethnic differences observed in Table 3 remain among Hispanic borrow- ers who do not have any of these risk factors.

Columns 3 and 6 present the results where race and ethnicity are both inter- acted with one minus the prime age male employment rate, a general unemploy- ment rate that accounts for both unemployment and labor force participation. For both groups, racial and ethnic differences are concentrated among borrowers with higher county-by-group level unemployment rates. However, the ethnic differences in the impact of employment rates are significantly larger than the racial differ- ences. In order to give a sense of the magnitude of these effects, we estimate the racial and ethnic differences for borrowers at low risk of unemployment, i.e., when the employment rate among prime age males as a share of all prime age males is 96 percent. At this high employment level, ethnic differences in foreclosure are near zero at 0.005 and 0.006 in the home purchase and refinance samples, respectively, but racial differences in foreclosure remain significantly higher at 0.029 and 0.011 for the two samples.

IV. Heterogeneity in Mortgage Outcomes by Year of Origination

This section examines the pattern of delinquency and foreclosure risk based on year of origination. First, we restrict the sample to 2008 and 2009 delinquencies and

VoL. 8 no. 1 17Bayer et al.: the VulneraBility of Minority hoMeowners

foreclosures in order to focus on the crisis period and to avoid confounding the direct effect of origination year with differences that arise because earlier origination years allow for foreclosure and delinquency during the pre-crisis period. Table 5 presents the unconditional racial and ethnic differences by origination year with 2004 origi- nations as the omitted category. For home purchase originations, there is a substan- tial increase in the racial and ethnic differences in foreclosure and delinquency for originations between 2005 and 2007 relative to 2004 with the largest differences arising for 2006 originations. The pattern for refinance mortgages is less clear with substantially smaller and less significant differences in the estimated racial and eth- nic effects between originations from 2005 to 2007 and originations in 2004, and no differences at all in the racial foreclosures differences for refinance mortgages.

As in Figure 2, Figure 3 presents racial and ethnic differences in foreclosure for all four model specifications presented in Table 3, except Figure 2 presents racial differences by origination year. The top panel of Figure 3 shows the results for the home purchase sample. In all models, we get a peak in racial and ethnic dif- ferences in 2006 at the peak of the housing market. For the final model indicated

Table 4A—Race and Ethnicity Interactions—Home Purchase Sample

  Delinquency Foreclosure

Race Baseline Risk factor Employment Baseline Risk factor Employment

Black 0.015*** −0.002 −0.005 0.017*** 0.000 −0.001   (0.002) (0.003) (0.005) (0.004) (0.005) (0.007) Hispanic 0.015*** 0.016*** −0.032*** 0.025*** 0.020*** −0.056***   (0.002) (0.002) (0.003) (0.003) (0.003) (0.005) Black × Subprime   0.018***     0.035***       (0.005)     (0.008)   Hispanic × Subprime   0.000     0.008       (0.004)     (0.007)   Black × Rate spread   0.006     −0.011       (0.004)     (0.007)   Hispanic × Rate spread   −0.006     0.002       (0.003)     (0.005)   Black × High DTI   0.041***     0.058***       (0.006)     (0.009)   Hispanic × High DTI   0.010**     0.021***       (0.004)     (0.007)   Black × Unemp rate     0.570***     0.754***       (0.065)     (0.092) Hispanic × Unemp rate     0.908***     1.521***       (0.066)     (0.089)

r2 0.102 0.108 0.109 0.137 0.137 0.138

notes: This table presents estimates for delinquency and foreclosure based on the home purchase sample. The first column for each outcome presents the final column of estimates from Table 3. The second column presents estimates based on the interaction of the dummy variables for race and ethnicity with whether the borrower has a Vantage credit score below 701, a loan-to-value ratio above 0.95, or a debt-to-income ratio above 0.45. The third column presents estimates using interactions of race and ethnicity with one minus the race-by-year county average employment rate for prime age adults. Standard errors (in parentheses) are robust to heteroskedasticity, and clus- tered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

18 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

by the triangle, these differences arise even after controlling for contemporaneous housing prices, and so these results cannot simply be driven by the fact that 2006 home purchase mortgages exposed homebuyers to the largest declines in housing prices relative to the purchase price. Rather, minority homebuyers in 2006 appear to be more vulnerable to the economic downturn than homebuyers in earlier years or homebuyers after the onset of the crisis, even after conditioning on detailed credit risk factors associated with the borrower and the mortgage. Panel B of Figure 2 shows the results for the refinance sample. As noted above, racial differences in foreclosure do not increase for originations in or near 2006. Ethnic differences in foreclosure for the refinance sample follow the same pattern as was observed in the home purchase sample, but the effects for 2006 are substantially smaller in magnitude.

We next use the refinance sample in order to examine home purchase origina- tions over a longer timeframe. Specifically, we match refinance mortgages with the original home purchases back to 1998, which is the earliest year for which we observe home purchases in all seven of our markets. This exercise is imperfect for two reasons. First, the sample only considers a select subsample of home purchase

Table 4B—Race and Ethnicity Interactions—Refinance Sample

Delinquency Foreclosure

Race Baseline Risk factor Employment Baseline Risk factor Employment

Black 0.005** −0.004 −0.019*** 0.007** 0.008** −0.014*** (0.003) (0.003) (0.004) (0.003) (0.003) (0.005)

Hispanic 0.011*** 0.011*** −0.035*** 0.019*** 0.014*** −0.032*** (0.001) (0.002) (0.003) (0.012) (0.002) (0.004)

Black × Subprime 0.004 −0.013**   (0.004) (0.005) Hispanic × Subprime 0.004 0.020***   (0.004) (0.005) Black × Rate spread 0.010*** 0.002   (0.003) (0.004) Hispanic × Rate spread −0.009*** −0.007*   (0.003) (0.004) Black × High DTI 0.016*** 0.009*   (0.004) (0.005) Hispanic × High DTI 0.007** 0.010**   (0.003) (0.005) Black × Unemp rate 0.651*** 0.631***   (0.066) (0.071) Hispanic × Unemp rate 0.874*** 0.956***   (0.065) (0.073)

r2 0.096 0.096 0.097 0.102 0.102 0.103

notes: This table presents estimates for delinquency and foreclosure based on the refinance sample. The first col- umn for each outcome presents the final column of estimates from Table 3. The second column presents estimates based on the interaction of the dummy variables for race and ethnicity with whether the borrower has a Vantage credit score below 701, a loan-to-value ratio above 0.95, or a debt-to-income ratio above 0.45. The third column presents estimates using interactions of race and ethnicity with one minus the race-by-year county average employ- ment rate for prime age adults. Standard errors (in parentheses) are robust to heteroskedasticity, and clustered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

VoL. 8 no. 1 19Bayer et al.: the VulneraBility of Minority hoMeowners

mortgages back to 1998, namely the mortgages for borrowers who chose to refinance during our sample period. Second, when borrowers refinance, they have the oppor- tunity to remove equity using a cash-out refinance, and a large share of refinance mortgages were used to extract equity in the lead up to the crisis (Demyanyk and Van Hemert 2011). While we can do nothing to address the first concern, we did conduct supplementary analyses where we split the refinance sample based on our best proxy for equity extraction, i.e., the ratio of the total mortgage amount taken out during the refinance to the outstanding mortgage debt observed in the March 31 credit report right prior to the refinance. The results of those estimations are shown in Appendix Tables A4 and A5.

Table 5—Estimated Differences in Mortgage Outcomes by Origination Year

Race and ethnicity Delinquency Foreclosure

interactions Black Hispanic Black Hispanic

Home purchase sample Level estimate 0.061*** 0.061*** 0.086*** 0.104***   (0.005) (0.005) (0.010) (0.009)

2008 −0.031*** −0.026*** −0.055*** −0.066***   (0.007) (0.006) (0.011) (0.010)

2007 0.017** 0.038*** 0.000 0.035***   (0.007) (0.006) (0.011) (0.012)

2006 0.054*** 0.051*** 0.091*** 0.108***   (0.008) (0.007) (0.014) (0.012)

2005 0.027*** 0.034*** 0.033** 0.057***   (0.008) (0.007) (0.013) (0.011)

Sample size 0.027 0.035 r2 207,723 207,723

refinance sample Level estimate 0.038*** 0.044*** 0.051*** 0.059***   (0.005) (0.005) (0.007) (0.006)

2008 −0.022*** −0.023*** −0.051*** −0.046***   (0.006) (0.006) (0.007) (0.007)

2007 −0.002 0.015** −0.027*** −0.002   (0.006) (0.006) (0.008) (0.008)

2006 0.020*** 0.018*** −0.009 0.035***   (0.007) (0.006) (0.009) (0.009)

2005 0.010 0.007 −0.003 0.024***   (0.007) (0.006) (0.008) (0.009)

Sample size 0.011 0.012 r2 192,953 192,953

notes: This table presents estimates for a model where mortgage origination year is interacted with race and ethnicity to explain delinquency and foreclosure for the home purchase and refi- nance samples. The first column for each outcome presents the estimates for origination year interacted with whether the borrower is black, and the second column presents the estimates for origination year interacted with whether Hispanic. The row labelled level estimate presents the estimate on the dummy variable for black or Hispanic. The omitted category is all mort- gages underwritten in 2004. Standard errors (in parentheses) are robust to heteroskedasticity, and clustered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

20 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

In Table 6, we present the unconditional racial and ethnic differences in foreclo- sure by home purchase year relative to mortgage refinancing for homes purchased prior to 2008. While the sample represents only home purchases that selected into refinance, and our econometric specification must control for credit scores observed after the home purchase, our estimates demonstrate the significance of purchase year for a representative sample of refinance mortgages. Notably, the origination

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2004 2005 2006 2007 2008

Origination year

Blacks

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2004 2005 2006 2007 2008

Origination year

Hispanics

Panel A. Purchase sample

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2004 2005 2006 2007 2008

Origination year

Blacks

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

2004 2005 2006 2007 2008

Origination year

Hispanics

Panel B. Re�nance sample

Unconditional model

Add subprime controls

Add risk factor controls

Add contemporaneous controls

Figure 3. Black and Hispanic Probability of Foreclosure Relative to Whites by Loan Origination Year, Home Purchase, and Refinance Samples

notes: The figure presents estimated group differences in the number of foreclosures within the last 12 months by mortgage origination year separately for the home purchase and refinance samples. Each line represents the esti- mated differences for a unique model based on a specific set of regression controls as discussed in relation to Table 3.

VoL. 8 no. 1 21Bayer et al.: the VulneraBility of Minority hoMeowners

year coefficients are relatively flat for the first three years of the sample 1998 to 2000. After 2000, we see substantially larger racial and ethnic differences in foreclo- sures for homes that were originally purchased closer to the boom, and the largest differences arise between 2004 and 2006. Unexpectedly, racial and ethnic differ- ences drop for home purchases in 2006 relative to 2005 and 2007, except for the racial differences in delinquency. We do not have any explanation for this anomaly. Figure 4 presents the racial and ethnic differences by home purchase year for all four models. As before, the basic pattern of results is robust to including controls for mortgage-risk factors, contemporaneous employment risk, and levels of negative equity. The findings are also robust across subsamples based on the ratio of the new mortgage amount to the outstanding original mortgage balance prior to the refinance.

Table 6—Estimated Differences in Mortgage Outcomes by Home Purchase Year—Refinance Sample

Delinquency Foreclosure

Ethnicity interactions Black Hispanic Black Hispanic

Level 0.034*** 0.037*** 0.035*** 0.052*** (0.003) (0.004) (0.004) (0.005)

2008 0.038 −0.034 −0.051*** −0.052*** (0.009) (0.003) (0.009) (0.019)

2007 0.037 0.041* 0.046 0.088** (0.003) (0.025) (0.036) (0.041)

2006 0.017 0.049*** 0.01 0.015 (0.015) (0.014) (0.022) (0.017)

2005 0.027** 0.038*** 0.029** 0.050*** (0.011) (0.009) (0.014) (0.011)

2004 0.037*** 0.028*** 0.020* 0.057*** (0.009) (0.007) (0.011) (0.010)

2003 0.021** 0.023*** 0.018 0.024** (0.009) (0.007) (0.011) (0.010)

2002 0.015* 0.025*** 0.015 0.019* (0.009) (0.008) (0.010) (0.011)

2001 0.035*** 0.004 0.006 0.016 (0.012) (0.007) (0.010) (0.011)

2000 0.004 −0.003 −0.004 −0.001 (0.008) (0.008) (0.009) (0.011)

1999 0.008 −0.009 0.014 −0.009 (0.008) (0.008) (0.011) (0.010)

1998 0.009 0.003 −0.008 −0.005 (0.011) (0.010) (0.010) (0.011)

Sample size 192,953 192,953 r2 0.012 0.012

notes: This table presents estimates based on the refinance sample for a model where home purchase year is interacted with race and ethnicity to explain delinquency and foreclosure. The first column for each outcome presents the estimates for home purchase year interacted with whether the borrower is black, and the second column presents the estimates for origi- nation year interacted with whether Hispanic. The omitted category is all homes purchased prior to 1998. Standard errors (in parentheses) are robust to heteroskedasticity, and clustered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

22 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

Finally, following up on the analyses in Table 4, we examine the effect of home purchase year for borrowers with key risk factors by interacting year of home pur- chase with either a dummy variable for whether the borrower has a high debt-to- income ratio or with the race and year specific unemployment rate for the county in which the borrower resides. These results are shown in Table 7 for the home purchase sample and in Table 8 for the refinance sample. In the home purchase sample, both blacks and Hispanics have larger racial and ethnic differences in delinquency and foreclosure in 2005 and 2006 if they had high debt-to-income ratios with the effects being largest in 2006, and Hispanics have larger ethnic differences in delinquency and foreclosure after 2004 if they were exposed to high unemployment rates with the effects being largest in 2006 and 2007. Similarly for the refinance sample, the effect of having a high debt-to-income ratio on racial and ethnic differences in delin- quency and foreclosure tends to be largest for homes purchased between 2001 and 2005, and the effect of exposure to high unemployment rates on racial and ethnic differences tends to be highest for homes purchased between 2002 and 2007.

Overall, these results corroborate the intuition presented in the introduction. As credit expanded during the housing boom, households with higher unobserved risk of defaults and delinquencies entered the housing market and, empirically, black and Hispanic borrowers were much more likely to be in this group. The selection of these especially high-risk households into the market translated into much higher rates of delinquencies and defaults once the recession started, especially among minority borrowers.

−0.1

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

1996 1999 2002 2005 2008

Purchase year

Blacks

−0.1

−0.05

0

0.05

0.1

0.15

0.2

P ro

b a b ili

ty o

f fo

re cl

o su

re r

e la

tiv e t o w

h ite

s

1996 1999 2002 2005 2008

Purchase year

Hispanics

Re�nance sample

Unconditional model

Add subprime controls

Add risk factor controls

Add contemporaneous controls

Figure 4. Black and Hispanic Probability of Foreclosure Relative to Whites by Home Purchase Year of Refinance Loan

notes: The figure presents estimated group differences in the number of foreclosures within the last 12 months by the year of home purchase for the refinance sample. Each line represents the estimated differences for a unique model based on a specific set of regression controls as discussed in relation to Table 3.

VoL. 8 no. 1 23Bayer et al.: the VulneraBility of Minority hoMeowners

V. Summary and Conclusion

In this paper, we identify large racial and ethnic differences in the likelihood of mortgage delinquencies and foreclosures in the recent housing market bust and asso- ciated recession. Substantial differences remain after controlling for the borrower, home, loan, and market attributes, including the individual’s credit score, that would have been observed by lenders ex ante; differences in lenders, neighborhoods, or the significance of loan attributes that might be associated with subprime lending; and ex post measures of exposure to local housing and labor market shocks. Collectively, these results imply that the relatively poor mortgage outcomes for minority bor- rowers are not simply a function of greater participation in the subprime sector or greater exposure to neighborhood housing price declines or unemployment rates.

Table 7—Estimated Differences in Mortgage Outcomes by Origination Year and Risk Factors

Race and ethnicity Delinquency Foreclosure

interactions Black Hispanic Black Hispanic

Panel A. Home purchase sample—debt-to-income interactions 2008 −0.029 0.033** −0.097*** −0.080***   (0.020) (0.015) (0.035) (0.023) 2007 −0.016 0.028* −0.079** −0.006   (0.020) (0.015) (0.035) (0.025) 2006 0.050** 0.061*** 0.179*** 0.158***   (0.023) (0.017) (0.043) (0.033) 2005 0.032 0.045*** 0.085* 0.076***   (0.026) (0.018) (0.046) (0.028)

Sample size 207,241 207,241 r2 0.139 0.184

Panel B. Home purchase sample—unemployment rate interactions 2008 0.011 0.213 −0.286 0.498***   (0.113) (0.142) (0.182) (0.187) 2007 −0.101 0.960*** −0.198 1.195***   (0.114) (0.153) (0.184) (0.207) 2006 −0.128 0.498*** −0.003 1.212***   (0.126) (0.155) (0.213) (0.233) 2005 −0.218* 0.291* -0.151 0.426**   (0.124) (0.156) (0.204) (0.204)

Sample size 207,241 207,241 r2 0.139 0.182

notes: This table presents estimates based on the home purchase sample for a model where the interactions between home purchase year and the race and ethnicity dummy variables are interacted with mortgage risk factors, high debt-to-income ratio dummy (panel A) or exposure to unemployment rates by race by current year-by-county (panel B). The first column for each outcome presents the estimates for home purchase year interacted with whether the borrower is black and with the risk factor, and the second column presents the estimates for origination year interacted with whether Hispanic and the risk factor. The omitted category is homes pur- chased or mortgages originated in 2004. Standard errors (in parentheses) are robust to het- eroskedasticity, and clustered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

24 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

Finally, these differences are concentrated among borrowers who may have dif- ficulty making mortgage payments during the crisis. Blacks and Hispanics with high debt payment to income ratios have larger racial and ethnic differences in delinquencies and foreclosures, and in fact, black borrowers who do not have risk factors of this type are no more likely to have a delinquency or foreclosure than white borrowers. Also, blacks and Hispanics experience much larger increases in delinquency and foreclosure rates as county employment rates fall during the crisis than are experienced by equivalent white borrowers. These effects are largest for Hispanics, and we do not find any ethnic differences in delinquency and foreclosure in counties and years with relatively high employment rates for Hispanics.

A further decomposition of our main findings by origination year reveals that the large estimated racial and ethnic differences in mortgage outcomes are concen- trated in mortgages that were originated in 2005 through 2007. That is, black and

Table 8A—Differences by Home Purchase Year and DTI—Refinance Sample

Race and ethnicity Delinquency Foreclosure

interactions Black Hispanic Black Hispanic

2008 −0.052 0.015 −0.084* −0.21   (0.088) (0.072) (0.045) (0.131) 2007 0.015 0.041 −0.035 −0.143**   (0.065) (0.059) (0.086) (0.067) 2006 0.036 0.047 0.02 0.037   (0.035) (0.029) (0.044) (0.030) 2005 0.077*** 0.055*** 0.045 0.063***   (0.026) (0.018) (0.033) (0.024) 2004 0.049** 0.022 0.004 0.061**   (0.020) (0.014) (0.023) (0.026) 2003 0.042* 0.007 0.004 0.033   (0.024) (0.016) (0.022) (0.024) 2002 0.017 0.016 0.04 0.026   (0.020) (0.019) (0.026) (0.028) 2001 0.057* 0.019 0.018 0.047*   (0.030) (0.017) (0.024) (0.029) 2000 −0.002 −0.006 −0.033 0.020   (0.019) (0.019) (0.021) (0.023) 1999 −0.049*** −0.026* 0.009 −0.016   (0.017) (0.015) (0.024) (0.022) 1998 0.014 0.046** −0.028 0.028   (0.023) (0.023) (0.020) (0.027)

Sample size 192,444 192,444 r2 0.131 0.144

notes: This table presents estimates based on the refinance sample for a model where the inter- actions between home purchase year and the race and ethnicity dummy variables are inter- acted with mortgage risk factors, high debt-to-income ratio dummy. The first column for each outcome presents the estimates for home purchase year interacted with whether the borrower is black and with the risk factor, and the second column presents the estimates for origination year interacted with whether Hispanic and the risk factor. The omitted category is homes pur- chased prior to 1998. Standard errors (in parentheses) are robust to heteroskedasticity, and clustered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

VoL. 8 no. 1 25Bayer et al.: the VulneraBility of Minority hoMeowners

Hispanic households drawn into homeownership at the peak of the credit expansion and housing boom were especially likely to subsequently become delinquent and default on their mortgages. This finding is again robust to the inclusion of a broad set of controls for borrower, lender, loan, and neighborhood attributes, as well as controls for contemporaneous risk factors. This finding is also observed over a lon- ger time period when examining racial differences by year of home purchase in the refinance sample. And these effects are more pronounced for borrowers with high debt-to-income ratios or with exposure to lower employment rates.

The simple idea discussed earlier that forward-looking borrowers sort into home- ownership in part based on their risk of experiencing adverse future events high- lights a potential unified explanation for the full set of empirical results. African Americans and Hispanic borrowers appear to be much more vulnerable to changes in market employment rates, declines in housing prices and having originally high debt-expense-to-income ratios. African American and Hispanic homebuyers near

Table 8B—Differences by Home Purchase Year and DTI—Refinance Sample

Race and ethnicity Delinquency Foreclosure

interactions Black Hispanic Black Hispanic

2008 0.287 0.012 0.360 0.161   (0.425) (0.161) (0.464) (0.298) 2007 0.758* 0.138 0.915 −0.447   (0.395) (0.833) (0.686) (1.081) 2006 0.165 0.751** 0.803* 0.387   (0.241) (0.332) (0.418) (0.250) 2005 0.276 0.858*** 0.728*** 1.061***   (0.189) (0.215) (0.268) (0.257) 2004 0.459*** 0.380** 0.320* 0.570***   (0.157) (0.169) (0.176) (0.193) 2003 −0.040 0.472** 0.251 0.131   (0.186) (0.206) (0.169) (0.216) 2002 0.108 0.158 −0.103 0.609**   (0.136) (0.208) (0.166) (0.269) 2001 0.335 −0.083 −0.102 0.140   (0.305) (0.179) (0.174) (0.242) 2000 0.116 0.118 0.081 0.334   (0.150) (0.241) (0.179) (0.249) 1999 0.063 0.040 0.158 −0.124   (0.175) (0.188) (0.188) (0.228) 1998 −0.184 −0.004 −0.010 0.531*   (0.184) (0.252) (0.157) (0.293)

Sample size 192,444 192,444 r2 0.131 0.144

notes: This table presents estimates based on the refinance sample for a model where the inter- actions between home purchase year and exposure to unemployment rates by race by current year-by-county. The first column for each outcome presents the estimates for home purchase year interacted with whether the borrower is black and with the risk factor, and the second col- umn presents the estimates for origination year interacted with whether Hispanic and the risk factor. The omitted category is homes purchased prior to 1998. Standard errors (in parenthe- ses) are robust to heteroskedasticity, and clustered at tract-credit year level.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.

26 AmEricAn Economic JoUrnAL: Economic PoLicy fEBrUAry 2016

the peak of the housing market and the credit market expansion have the highest foreclosure rates during the crisis, and these high rates cannot be explained by the higher rates on negative equity experienced by borrowers who took out loans when housing prices were at their peak.

Our results complement a recent literature that aims to understand the causes and consequences of the last housing cycle. For example, while Mian and Sufi (2009) demonstrate a significant role for subprime lending in explaining overall neighbor- hood level foreclosure rates, our analysis implies that subprime lending can explain at most a modest fraction of observed racial and ethnic differences in credit market outcomes. In addition, tighter underwriting standards and increased financial over- sight arising from recent financial reforms are unlikely to address these concerns because the observed differences arise after controlling for all traditional under- writing variables, are based on comparisons within lenders and neighborhoods, and occur across a broad spectrum of minority borrowers.

Finally, regardless of the ultimate explanation for the observed higher rates of negative mortgage outcomes, our study raises serious concerns about homeowner- ship as a vehicle for reducing racial-wealth disparities. Our findings suggest instead that homeownership may be especially risky for households with a low initial level of wealth (savings) or fewer family resources on which to draw when hit with an adverse economic shock. Because delinquencies and default have consequences that go beyond the direct loss of housing equity/wealth, they can contribute substan- tially to perpetuating the wealth gap across generations. The increased cost of all subsequent borrowing (through the lower credit scores), in particular, makes future wealth accumulation much more difficult.

REFERENCES

Altonji, Joseph G., and Rebecca M. Blank. 1999. “Race and gender in the labor market.” In Handbook of Labor Economics, Vol. 3C, edited by Orley C. Ashenfelter and David Card, 3143–3259. Amster- dam: North-Holland.

Apgar, William, Allegra Calder, and Gary Fauth. 2004. credit, capital and communities: The impli- cations of the changing mortgage Banking industry for community-Based organizations. Cam- bridge: Harvard University Joint Center for Housing Studies.

Avery, Robert B., Glen B. Canner, and Robert E. Cooke. 2005. “New Information Reported under HMDA and Its Implications in Fair Lending Enforcement.” federal reserve Bulletin (Summer): 344–95.

Bayer, Patrick, Fernando Ferreira, and Stephen L. Ross. 2016. “The Vulnerability of Minority Home- owners in the Housing Boom and Bust: Dataset.” American Economic Journal: Economic Policy. http://dx.doi.org/10.1257/pol.20140074.

Belsky, Eric S., Nicolas Retsinas, and Mark Duda. 2007. “The Financial Returns to Low-Income Homeownership.” In chasing the American Dream: new Perspectives on Affordable Homeowner- ship, edited by William M. Rohe and Harry L. Watson, 191–212. London: Cornell University Press.

Bhutta, Neil, and Glenn B. Canner. 2013. “Mortgage Market Conditions and Borrower Outcomes: Evidence from the 2012 HMDA Data and Matched HMDA: Credit Record Data.” federal reserve Bulletin 99 (4): 1–58.

Brueckner, Jan K. 1986. “The Downpayment Constraint and Housing Tenure Choice: A Simplified Exposition.” regional Science and Urban Economics 16 (4): 519–25.

Calem, Paul S., Kevin Gillen, and Susan Wachter. 2004. “The Neighborhood Distribution of Subprime Mortgage Lending.” Journal of real Estate finance and Economics 29 (4): 393–410.

Clapp, John, Anupam Nanda, and Stephen L. Ross. 2008. “Which School Attributes Matter? The Influence of School District Performance and Demographic Composition on Property Values.” Journal of Urban Economics 63 (2): 451–66.

VoL. 8 no. 1 27Bayer et al.: the VulneraBility of Minority hoMeowners

Charles, Kerwin, and Erik Hurst. 2002. “The Transition to Home Ownership and the Black-White Wealth Gap.” review of Economics and Statistics 84 (2): 281–97.

Demyanyk, Yuliya, and Otto Van Hemert. 2011. “Understanding the Subprime Mortgage Crisis.” review of financial Studies 24 (6): 1848–80.

Deng, Yongheng, Stephen L. Ross, and Susan M. Wachter. 2003. “Racial Differences in Homeowner- ship: The Effect of Residential Location.” regional Science and Urban Economics 33 (5): 517–56.

Di, Zhu Xiao, and Xiaodong Liu. 2007. “The Importance of Wealth and Income in the Transition to Homeownership.” cityscape: A Journal of Policy Development and research 9 (2): 137–52.

Duca, John V., and Stuart S. Rosenthal. 1994. “Borrowing Constraints and Access to Owner-Occupied Housing.” regional Science and Urban Economics 24 (3): 301–22.

Edminston, Kelly D. 2009. “Characteristics of High-Foreclosure Neighborhoods in the Tenth District.” federal reserve Bank of Kansas city Economic review Q2: 51–75.

Engelhart, Gary V. 1996. “Consumption, Down Payments, and Liquidity Constraints.” Journal of money, credit and Banking 28 (2): 255–71.

Fisher, Lynn M., Lauren Lambie-Hanson, and Paul S. Willen. 2010. “A Profile of the Mortgage Crisis in a Low-and-Moderate-Income Community.” Federal Reserve Bank of Boston Public Policy Dis- cussion Paper 10-6.

Gerardi, Kristopher S., and Paul Willen. 2009. “Subprime Mortgages, Foreclosures, and Urban Neigh- borhoods.” B. E. Journal of Economic Analysis and Policy 9 (3): 1–37.

Gyourko, Joseph, Peter Linneman, and Susan Wachter. 1999. “Analyzing the Relation among Race, Wealth, and Homeownership in America.” Journal of Housing Economics 8 (2): 63–89.

Gyourko, Joseph, and Joseph Tracy. 2013. “Unemployment and Unobserved Credit Risk in the FHA Single Family Mortgage Insurance Fund.” National Bureau of Economic Research (NBER) Work- ing Paper 18880.

Herbert, Christopher E., Donald R. Haurin, Stuart S. Rosenthal, and Mark Duda. 2005. Homeowner- ship Gaps among Low income and minority Borrowers and neighborhoods. Washington, DC: U.S. Department of Housing and Urban Development.

Hoynes, Hilary, Douglas L. Miller, and Jessamyn Schaller. 2012. “Who Suffers during Recessions?” Journal of Economic Perspectives 26 (3): 27–48.

Linneman, Peter D., and Susan M. Wachter. 1989. “The Impacts of Borrowing Constraints on Home- ownership.” American real Estate and Urban Economics Association 17 (4): 389–402.

Mayer, Christopher J., and Karen Pence. 2008. “Subprime Mortgages: What, Where and To Whom?” National Bureau of Economic Research (NBER) Working Paper 14083.

Mayer, Chris, Karen Pence, and Shane M. Sherlund. 2009. “The Rise in Mortgage Defaults.” Journal of Economic Perspectives 23 (1): 27–50.

Mian, Atif, and Amir Sufi. 2009. “The Consequences of Mortgage Credit Expansion: Evidence from the U.S. Mortgage Default Crisis.” Quarterly Journal of Economics 124 (4): 1449–96.

Oliver, Melvin L., and Thomas M. Shapiro. 1997. Black Wealth, White Wealth: A new Perspective on racial inequality. New York: Routledge.

Quercia, Roberto G., George W. McCarthy, and Susan Wacther. 2003. “The Impacts of Affordable Lending Efforts on Homeownership Rates.” Journal of Housing Economics 12 (1): 29–59.

Reid, Carolina, and Elizabeth Laderman. 2009. “The Untold Costs of Subprime Lending: Examining the Links among Higher-Priced Lending, Foreclosures and Race in California.” Federal Reserve Bank of San Francisco Working Paper 2009-09.

Rohe, William M., Shannon Van Zandt, and George McCarthy. 2002. “Social Benefits and Costs of Homeownership.” In Low-income Homeownership: Examining the Unexamined Goal, edited by Nicolas P. Retsinas and Eric S. Belsky, 381–406. Washington, DC: Brookings Institution Press.

Shapiro, Thomas. 2004. The Hidden cost of Being African American: How Wealth Perpetuates inequal- ity. Oxford: Oxford University Press.

Shuey, Kim M., and Andrea E. Wilson. 2008. “Cumulative Disadvantage and Black-White Disparities in Life-Course Health Trajectories.” research on Aging 30 (2): 200–225.

Smith, Jeffrey. 1995. “Racial and Ethnic Differences in Wealth in the Health and Retirement Study.” Journal of Human resources 30 (S1): 158–83.

Wachter, Susan M., and Isaac F. Megbolugbe. 1992. “Impacts of housing and mortgage market dis- crimination racial and ethnic disparities in homeownership.” Housing Policy Debate 3 (2): 353–70.

Wachter, Susan M., Karl Russo, and Jonathan E. Hershaff. 2010. “Subprime Lending: Neighborhood Patterns Over Time in US Cities.” Unpublished.