![Sample our Social Sciences journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110822/TOC-Subject-Sample-2021-Social-Sciences.jpg"})
Abstract
The subprime boom and subsequent foreclosure crisis highlighted risk associated with pursuit of the American Dream of homeownership. People of color and those living in segregated areas were particularly harmed by the dramatic rise and fall of the housing market. Almost a decade after the economy’s collapse, questions remain about racial and spatial disparities in access to mortgage credit. I leverage Home Mortgage Disclosure Act data to explore mortgage application outcomes in 2014. Well into the economy’s recovery, minority borrowers remained at a disadvantage in the mortgage approval process. Whereas 71% of White applicants were approved for home loans, approval rates were lower for Asians (68%), Latinos (63%), and Blacks (54%). Black and Latino borrowers were also significantly more likely to receive higher cost loans than Whites, a practice that has accelerated since the foreclosure crisis. Results suggest that segregation exacerbated racial disparities as lenders funneled expensive credit into isolated minority communities. Furthermore, the differences between White and minority outcomes were largest in census tracts where subprime lending was common in 2006 and foreclosures accumulated during the Great Recession. Together, these findings indicate how spatially organized markets have racialized consequences in a highly segregated society.
Acknowledgments
The author would like to thank Jackelyn Hwang, Elizabeth Roberto, and Jacob Rugh for their valuable contributions to this project.
Notes
1. Only 54.3% of applications submitted by Blacks and 62.5% of those submitted by Latinos were approved, compared with 68.4% and 70.9% among Asians and Whites, respectively.
2. Findings were substantively identical using the full sample regardless of MSA racial populations.
3. Average prime rates were published weekly and varied between 4.60 (on January 6) and 3.85 (on December 22) for 30-year, fixed-rate loans made during 2014 (Federal Financial Institutions Examination Council (FFIEC), Citation2016).
4. Missing data regarding applicant race have long been a concern with HMDA, particularly when evaluating the relationship between spatial inequalities (e.g., racial segregation) and borrower characteristics (Wyly & Holloway, Citation2002). When No Race applicants are included in this article’s statistical models as a dummy variable, the existing relationships do not change. Compared with White applicants, No Race applicants are significantly disadvantaged. Specifically, the magnitude of the negative coefficients for the No Race dummy variable in models predicting loan approval are substantively similar to the coefficients for the Black dummy variable. No Race loan candidates are slightly less likely than Whites to originate high-cost loans, although the magnitude of the relationship is very small (i.e., less than 1% less likely than Whites to originate high-cost loans).
5. Although the U.S. 2010 crosswalk file was designed for data aggregated at the tract level, I adapted its structure to estimate 2010 census-tract assignment for mortgage application data organized with 2000 census tracts. Specifically, I assigned each application in the 2006 HMDA data set the 2010 census tract that most overlapped with the 2000 tract based on the weights provided in the US 2010 data set. Although most 2000 tracts overlap almost completely with 2010 tracts (the median weight is 0.999), this method introduces noise into my estimates of tract-level 2006 subprime lending rates. To the extent that this imprecision biases my results, the direction of that bias should be toward zero (e.g., if subprime loans are mistakenly assigned to neighboring tracts), which does not threaten the main findings of this article.
6. For Freddie Mac data, LTV below 6% and above 105% were excluded, whereas those below 0% and greater than 97% were excluded from the Fannie Mae data set.
7. Appendix Table A4 shows that results from probit models (presented as odds ratios) are substantively identical to the linear models presented in the article.
8. To address the concern that results may be biased by census tracts in which very few loans were approved in 2006—thereby potentially introducing very high or very low subprime lending rates—I estimated the main results excluding census tracts in which fewer than 10 loans were approved in 2006. The findings in this smaller sample are presented in Appendix Table A5 and are almost identical to those presented in the main text and tables.
9. Rugh and Massey (Citation2010) used annualized change in HPI from 2000 to 2006 normalized by the annualized change between 1980 and 2000. My findings were substantively identical using this alternative measure. I used 2000 to 2006 raw change because missing historical HPI data (especially before 1991) reduced my analytical sample.
10. My results hold when the sample is restricted to home purchase loans, although the relationships between both outcomes and ecological characteristics are slightly stronger.
11. Estimates from interacted models of loan approval are available in Appendix Table A1, whereas predicted probabilities of loan approval across borrower race and residential segregation are available in Appendix Figure A1.
12. Coefficients available in Appendix Table A2.
13. Segregation findings were consistent when estimated using an instrumental variable approach used in previous work: the total number of local governments within each MSA and two-staged least squares (e.g., Cutler & Glaeser, Citation1997; Hyra et al., Citation2013). I gathered county-level local government data from the 2012 Census of Local Governments. Results from these estimates are available in Appendix Table A3.