Revisiting the subprime crisis: The dual mortgage market and mortgage defaults by race and ethnicity

ABSTRACT

The impacts of the foreclosure crisis have been widespread, catalyzing the worst economic downturn since the Great Depression and leading to dramatic declines in housing equity and wealth. However, Black and Hispanic households and communities have been disproportionately affected by the crisis, contributing to a tightening of credit standards and a retrenchment of lending in these communities. This study uses a unique, national data set of purchase mortgages originated between 2004 and 2007 to examine the racial and ethnic dimensions of subprime lending practices that were prevalent during the boom and explores the role that these lending practices had on subsequent rates of default. This study thus contributes to the debate on the causes of the housing crisis and its implications for racial inequality in America.

Acknowledgments

The authors thank colleagues at the Center for Community Capital and the Center for Responsible Lending, as well as the editors of this journal and three anonymous reviewers for their helpful comments and critiques. Debbie Bocian’s and Wei Li’s work on this article was conducted while they were employed at the Center for Responsible Lending.

Notes

1. One limiting factor in mortgage data collected under the HMDA is that many of the underwriting factors used by lenders, such as borrower credit score, are not included (Avery et al., 2007).

2. Because LPS and BlackBox have ZIP codes and HMDA has census tracts, we used the MABLE/Geocorr12: Geographic Correspondence Engine to allocate loans in ZIP codes to corresponding census tracts (Missouri Census Data Center, n.d.).

3. For full details on the three data sets, as well as the probabilistic matching technique, see Appendix A in Bocian et al. (2011).

4. A loan is defined as higher priced if it has an interest rate 3 or more points above the federal treasury rate at origination, as identified in the HMDA data. The HMDA pricing data, which have been included in 2004, were added to increase understanding of lending activity in the subprime, or higher priced, segment of the mortgage market. Though there is not a one-to-one correspondence between higher priced and subprime loans, many researchers use the terms higher priced and subprime interchangeably.

5. Because of the variety of hybrid ARM structures, we limit our hybrid ARM definition to ARMs with an initial interest rate period of less than 5 years, which captures the majority of the subprime hybrid ARM products.

6. Low-income borrowers are those whose income falls below 50% of the area median, moderate-income borrowers are those between 50 and 80% of area median, middle-income are those between 80 and 120%, and high-income are those with incomes above 120% of the area median. To provide some context for these ratios, in San Francisco, California, a moderate-income family earning 80% of the median in 2000 would have an annual income of $60,000; in Cleveland, Ohio, a moderate-income family would have an income of approximately $44,000.

7. Zillow (n.d.) is a web-based real estate company that tracks more than 110 million homes in markets across the United States. For a complete description of its house price index methodology, including comparisons to Case-Shiller, see http://www.zillow.com/research/zhvi-methodology-6032/.

8. Note that this metric includes the period of “recovery” in house prices postcrisis. Many borrowers experienced considerably larger house price declines between the month of origination and the month of delinquency.

9. This in part reflects the evolution of the subprime market and lending practices during the boom; borrowers who bought during the 2000s boom—a large share of whom where Hispanic and Asian—were more affluent and suburban than the targets of subprime lending in inner-city neighborhoods in the 1990s (Rugh, 2015; Schafran & Wegmann, 2012).

10. Though this can also be achieved through interaction variables (as in Table 2), interaction effects in logistic models can be hard to interpret (Ai, & Norton, 2003).

11. The marginal effect represents the difference in probability between having a loan originated by a broker (majority/minority, sand, rust) versus not, with the rest of the controls in the model being held at their means. The models are run separately for each racial/ethnic group and contain all of the controls presented in Table 2.

12. These findings should be interpreted with some caution. Research has shown that there were high levels of income overstatement on mortgage applications between 2002 and 2005 and that the degree of overstatement was higher in low-credit neighborhoods than in high-credit neighborhoods (Mian & Sufi, 2015).

13. There is also a rich discussion in the mortgage literature about the advantages and disadvantages of using a multinomial logit versus a proportional hazards approach to modeling competing risks in the mortgage market (An & Qi, 2012; Clapp, An, & Deng, 2005). We use the multinomial approach here; robustness checks using the proportional hazards approach revealed that the substantive conclusions do not differ significantly between the two methods.

14. Because restructuring the data greatly increases its size (because there is one observation for every month the loan is in the data), we take a 10% random sample of the data used in the first section of the article.

15. An alternate specification using completed foreclosures as the terminal event does not substantially change the findings. Using 90-day delinquency is common in the literature to account for the influence of state foreclosure laws that can significantly affect the timing between a foreclosure filing and default.

16. Results from the prepayment model are available from the authors upon request.

17. As mentioned earlier, the lack of data on the borrower’s combined loan-to-value ratio means that we are likely understating the effect of equity position on the likelihood of default.

18. The lack of a strong effect on income may be due to the fact that income is measured at origination, not at the time of delinquency or default; the effects of a homeowner who has lost his or her job or experienced a drop in income post-origination is not captured in our model.

19. To simplify exposition, we create a combined measure of nontraditional loan terms that equals 1 if the loan had either a prepayment penalty, had a hybrid ARM structure, was an interest-only loan, or had a balloon payment. All of these entered individually are also positive and significant in the model.

Additional information

Notes on contributors

Carolina K. Reid

Carolina K. Reid is an assistant professor in the Department of City and Regional Planning at the University of California at Berkeley and the faculty research advisor for the Terner Center for Housing Innovation. Carolina specializes in housing and community development, with a specific focus on efforts to expand access to credit and homeownership, including the Community Reinvestment Act and the impact of the foreclosure crisis on communities of color. Before joining the faculty at UC Berkeley, Dr. Reid served as the research manager for the Community Development Department at the Federal Reserve Bank of San Francisco and as a senior researcher with the Center for Responsible Lending. She received her PhD from the University of Washington, Seattle, and her undergraduate degree from Stanford University.

Debbie Bocian

Debbie Bocian is a senior associate in the Social and Economic Policy division of Abt Associates, where she specializes in the consumer financial protection, financial capability, and economic inclusion fields. Prior to Abt, Debbie served as a principal researcher at the Center for Responsible Lending (CRL), where she designed, managed, and executed projects on how interactions with consumer credit markets impact the wealth and financial security of vulnerable families and communities. Debbie holds an MPP from the Kennedy School of Government at Harvard and a BA in economics from Haverford College.

Wei Li

Wei Li is a senior research associate in the Housing Finance Policy Center (HFPC) at The Urban Institute, where his research focuses on the social and political aspects of the housing finance market and their implications for urban policy. His research led to the creation of the HFPC Credit Availability Index and the real denial rate. Li’s work has been published widely in various academic journals and has been covered in the Wall Street Journal, the Washington Post, and the New York Times, as well as in other print and broadcast media. Li is a quantitative research methodologist with expertise in cost–benefit analysis, program evaluation, and causal inference. Before joining Urban, Li was a principal researcher with the Center for Responsible Lending. Li received his MA in statistics and his PhD in environmental science, policy, and management from the University of California, Berkeley.

Roberto G. Quercia

Roberto G. Quercia is Harris Distinguished Professor in the Department of City and Regional Planning and Director of UNC Center for Community Capital, University of North Carolina at Chapel Hill. He has published widely on the topics of low-income homeownership, affordable lending, affordable housing policy, and homeownership education and counseling.