![Sample our Health and Social Care journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5938/TOC-Subject-Sample-2021-Health-Social-Care.jpg"})
Although a rich literature exists on the determinants and effects of concentrated foreclosures, little is known about what drives variation in how long real estate owned (REO) properties take to sell and to whom, particularly in Latino communities heavily affected by the current foreclosure crisis. This research employs multilevel and event history modeling to assess the factors associated with recent REOs’ likelihood of sale, sale to an investor, and sale to a Spanish-surname household in a sample of majority Latino Southern California neighborhoods. Properties in inner-city and exurban Latino neighborhoods with larger black population shares were less likely to sell and more likely to sell to investors if they did, while those located in lower poverty, largely Latino communities were more likely to sell to Spanish-surname households. These results suggest that the crisis is both exacerbating existing patterns of inequality and segregation while enabling Latinos’ homeownership in potentially socioeconomic mobility-enabling areas.
Keywords:
Acknowledgments
The authors gratefully acknowledge the University of California, Santa Barbara's Chicano Studies Institute for providing funding for this research, as well as Dr. Paul Ong at the University of California Los Angeles for assisting with data purchase. They also would like to thank the two anonymous reviewers for their helpful critical feedback.
Notes
1. Rivera et al. (2008) estimate that Latinos will experience a net loss of at least $75.8 billion for subprime loans made between 1998 and 2006. In a study of 25 Latino households that foreclosed, Bowdler, Quercia, and Smith (2010) found that they experienced an average loss of about $89,000.
2. California foreclosure counts are provided primarily by two companies, DataQuick and RealtyTrac. DataQuick reports trustee's deeds (completed foreclosures) recorded, while RealtyTrac reports notices of trustee sale and REOs. DataQuick was a better source for this project because it included information on foreclosures sold at auction.
3. Conventional and Federal Housing Administration loan limits were raised from $417,000 and $362,790 to $729,750 for high-cost metropolitan areas. The Housing and Economic Recovery Act of 2008 set the credit at $7,500. The American Recovery and Reinvestment Act of 2009 increased it to $8,000 (see Internal Revenue Service, 2011).
4. Our choice to use these data over the more recent 2005–2009 American Community Survey data was driven by two factors. First, because of its small sample size, the ACS estimates have high margins of error at the neighborhood level, requiring aggregation at higher levels to enable greater accuracy. Second, we were concerned about issues with reverse causality arising from gauging the effect of 2005–2009 neighborhood conditions on 2007–2009 REO trajectories, with these outcomes possibly contributing to neighborhood change. While the 2000 U.S. Census data are the most appropriate at the census tract level for this study, neighborhood housing markets and demographics likely changed in regions affected by foreclosures during the 2000s, which may bias our estimates. In particular, communities with higher foreclosure rates may have experienced faster housing filtering and minority population growth, among other shifts (Lauria & Baxter, 1999; Y. Li & Morrow-Jones, 2010). We address this issue further in the conclusion.
5. Census tracts in Los Angeles, Riverside, and San Bernardino counties were matched to nearest elementary schools by entering the address associated with their centroid into the California School Finder search engine. Tracts were only matched to regular schools; charter, alternative, special education, and continuation schools were not included. School data for the 2006–2007 school year was compiled from Ed-Data and the California Department of Education. About 4% of properties in the Inland Empire were missing nearest elementary school data due to new school construction.
6. There is unresolved debate about the best way to identify urban or inner-city, suburban, and exurban communities, let alone whether such designations are useful (Nelson & Sanchez, 1999). Our definition loosely corresponds with Marcelli (2004)'s housing age-based delineation of the Southern California exurbs (defined as “suburbs” in his article). On average, 37% of the housing stock in the sample exurban tracts was built after 1980 compared to only 15% in the suburban and urban tracts.
7. Paul Willen at the Federal Reserve Bank of Boston and Stefano Giglio at Harvard University contributed to the keyword compilation. Real estate owned keywords included “bank,” “bk,” “loan,” “mortgage,” “servicing,” along with the names of major U.S. and international banks. If a buyer had a surname or included keywords such as “llc,” “llp,” “real,” “prop,” and “family trust,” it was identified as not a REO.
8. Family trusts were identified by coding for “family” and “trust” in the buyer name field. Lists subsequently were reviewed for consistency. Typical corporate investors included Marbury Park Group LLC, ACT Properties LLC, Seacliff Properties Inc., and Property Asset Management Inc., each of which bought 25 or more REOs in the studied communities.
9. Oftentimes, buyers’ middle initials were listed along with their first and last names. At times, spouses were listed as joint buyers. Mistaking buyers with the same name as the same person is rarer in these circumstances.
10. Although DataQuick does not provide information about owner-occupancy status, the fact that about one-fifth of the family trust investors and just under half of the LLC investors bought at least two properties in the sample Latino communities during the studied period suggests speculative intentions. One way to tease out occupancy status is by examining mortgage records. However, given the size of our sample, the possibility of buyers misrepresenting their status when applying for a mortgage, and the fact that one-third of the family trust investors and two-fifths of the LLC investors had no recorded loan information, we decided that pursuing this method would be unproductive.
11. See Appendix E Census List of Spanish Surnames in the “Downloads” section at http://fcds.med.miami.edu. The approximately 12,500 names on the list were culled from Spanish-surname lists developed by the U.S. Census and New Mexico's SEER's Guest Program, as well as underrepresented Caribbean and South American names present in the Miami phonebook.
12. This process involves converting the data from wide to long form, or from a file of foreclosures to a file of foreclosure months, with each row representing a foreclosure during a particular month that it was on the market. Properties that foreclosed in January 2007 may have up to 36 rows of data, while properties that foreclosed in June 2009 may have only one row of data. Properties that sold at auction, the omitted category, are counted as month 0. The dependent variable is dichotomous, accounting for whether or not a foreclosure sold in a given month, with sales receiving a 1. Once a property is sold, it is not included in subsequent months, as it is no longer at risk of being sold. The independent variables remain constant for each month that a foreclosure is in the analysis. See Treiman (2008) for more information on this type of discrete-time hazard rate modeling.
13. We estimated multilevel models, but they did not significantly improve model fit over clustered logistic regression. In turn, the effect of neighborhood percentage black and Latino and exurban location varied depending on a foreclosure's month at risk, with the negative effect of percentage of black and Latino on chance of sale declining with a property's months on the market and the negative effect of exurban location increasing. Yet, because these interactions detracted from, rather than added to, the model fit, they were excluded from the analysis (Bayesian information criterion difference of 144).
14. These models were improvements over clustered logistic regression models that controlled for intraclass correlation among properties in the same tract, and coefficients changed dramatically.
15. The Academic Performance Inxex measures spring test performance; it ranges from 200 to 1,000.
16. There were no majority Latino census tracts in exurban northern Los Angeles County in our sample.
17. Properties located in fast-growing Riverside and San Bernardino counties had an average square footage of 1,547, compared to 1,232 among properties located in the Los Angeles suburbs or the cities of Los Angeles, Long Beach, Riverside, San Bernardino, and Ontario.
18. It is important to note that our measures of median home prices at the census tract level are almost certainly biased by their inclusion of both foreclosure auction and REO sales that tend to depress prices. However, our measures of median sales price in 2007 and foreclosure rate 2007–2009 are not as highly correlated as one might expect (r = −0.35). Still, it is likely that changes in neighborhood foreclosure rates are driving price changes at this relatively small geography, making it difficult to tease out the relative effects of each of these factors.