Heterogeneity in Income: Effects of Racial Concentration on Foreclosures in Los Angeles, California

ABSTRACT

The United States continues to be defined by racial concentration, where most racial/ethnic groups live apart from each other. For homeownership, neighborhoods with large proportions of racial minorities are often linked to negative outcomes for minority homeowners; this was particularly the case during the Great Recession. However, middle and upper income ethnic neighborhoods, or resurgent neighborhoods, have grown in numbers because of a concentration of immigrants, federal policies favoring professionals, ethnic-specific resources, and affluence. In 2007, about 37% of Los Angeles, California, Latino tracts were resurgent and 53% of Asian tracts were resurgent. This study finds that homeowners in resurgent neighborhoods had lower default/foreclosure rates and predicted probabilities than those in low-income neighborhoods. Asian resurgent neighborhoods had the lowest predicted probabilities of default or foreclosure, followed by Latino resurgent and White middle-class neighborhoods. There were also discrepancies among Asian neighborhoods based on nativity. Consequently, it is important to recognize that minority neighborhoods are heterogeneous, with differing impacts on homeownership opportunities when examined by class.

KEYWORDS:

• foreclosure
• ethnic neighborhoods
• immigration
• segregation

Acknowledgments

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author and do not necessarily reflect the views of the National Science Foundation. The author thanks the UCLA Center for Neighborhood Knowledge for article development support, in addition to the reviewers and editor for comments that improved the article.

Disclosure Statement

No potential conflict of interest was reported by the author.

Notes

1. There are instances of housing discrimination that formed historic ghettos, such as San Francisco’s (California) Chinatown in the 19th century (Kroll-Smith & Brown-Jeffy, 2013). Diaz (2005) details discriminatory policies that created or reinforced Chicano segregation before 1965. However, the United States passed several exclusionary laws throughout its history that banned migrants from Latin America and Asia. After 1965, the surge of Latino and Asian immigrants has helped to establish many contemporary Latino and Asian neighborhoods (Gibson & Jung, 2006; Logan & Stults, 2014).

2. Hing (1993) further describes the role of U.S. immigration policy and how it has shaped Asian American communities. While historical immigration laws sought cheap Asian labor in the 19th century, new laws in the 20th century favored Asians who are professionals with higher incomes, in large part because of the 1965 Immigration Act and the Immigration Act of 1990 (Hing, 1993). More recently, with the EB-5 program, Asian investors have been granted green cards after they substantially invest and create at least 10 jobs (Simons, Wu, Xu, & Fei, 2016).

3. Ethnic banks are an example of lending institutions in resurgent neighborhoods. Informal and formal ethnic lending has existed in minority and immigrant neighborhoods since early in U.S. history, to fill the credit needs of these groups that could not access mainstream banks (Hum, 2011). However, the growth of large Asian banks has contributed to the growth of resurgent neighborhoods with the globalization and deregulation of the financial sector (Dymski & Mohanty, 1999). Dymski and Mohanty (1999) argue that large formal Asian financial institutions have helped clients move from enclaves to ethnoburbs by supporting residents and businesses. Zonta (2015) traced the growth of Chinese and Korean banks into increasingly suburban areas in Los Angeles and New York, and found that Los Angeles Chinese banks have increasingly originated mortgages in wealthier suburbs.

4. DataQuick defines default as when a homeowner receives a legal Notice of Default document, which is at least 30 days after the lender contacted the homeowner about foreclosure avoidance assessment. A foreclosure is recorded as the date that a homeowner receives the Notice of Sale, or that the property can be sold at public auction. Homeowners who received multiple notices were counted only once unless the homeowner name changed in 2007, in which case they were then excluded. The study focuses on homebuyers of single-family homes, and does not cover condominiums, because single-family homes comprise the largest share of foreclosures (Foote, Gerardi, Goette, & Willen, 2008). Hartley (2009) also proved that single-family and multifamily housing markets are affected by foreclosure differently, in his analysis of supply and disamenity mechanisms. While there are exceptions (Foote et al., 2008; Rugh, 2015), most foreclosure studies focus on single-family homes (Biswas, 2012; Immergluck & Smith, 2006; Ong & Pfeiffer, 2008; Rugh & Massey, 2010; Schuetz, Been, & Ellen, 2008).

5. While surnames are useful when race/ethnicity is unavailable in a data set, there are several issues to consider when implementing surname methods. First, the Census Bureau surname list can lead to higher false positives when used to identify racial groups because it was designed to preidentify groups (Abrahamse, Morrison, & Bolton, 1994). Additionally, surname dictionaries are more effective for some segments of the population, such as foreign-born individuals (Eschbach, Kuo, & Goodwin, 2006), men and older people (Wong, Palaniappan, & Lauderdale, 2010), and groups that have more discernible surnames, including Latinos and some Asian groups (Fischella & Fremont, 2006).

6. I conducted robustness tests using 2000 census-tract data to examine whether there were differences in the results. Of the six regressions, the coefficients for neighborhood typology, household factors, and tract owners were similar in magnitude and signs to models using 2010 decennial data to determine resurgent neighborhoods. These tests indicate consistency among data sources.

7. I created these categories based on Ong et al.’s (2014) findings, in which Los Angeles County home prices rose slowly until 2003, whereas between 2003 and 2006 home prices increased by more than 50%. Los Angeles home prices began to fall after they peaked in 2006. Purchase year is important because there were dramatic changes in home prices that affected negative equity. Palmer (2015) found a 6.5% difference in default rates when simulating 2006 homebuyers as if they had the same price path as 2003 borrowers. Also, in 1999, the housing market was still recovering from the first boom in high-risk lending and recession from the 1990s (Immergluck, 2009). Thus, I focus on homeowners who purchased homes during the housing cycle that began after the smaller recession during the 1990s and ended before the most recent Great Recession. By focusing on homeowners in 2007, my sample excludes those who may have purchased during the 2000s housing bubble and moved before 2007 or lost their home. Among homeowners who were unable to prevent foreclosure, they were more likely to have had high-risk loans (Avery, Brevoort, & Canner, 2008).

8. The dictionary uses the national decennial census to estimate racial probabilities. However, the dictionary may underestimate the probabilities for certain racial/ethnic groups for smaller geographies with varying racial composition. For example, the 2000 census dictionary indicates that 40% of individuals with the surname “Lee” are White, 17% are Black, and 38% are Asian or Pacific Islander. These probabilities are based on the national population, which was 75% White, 12% Black, and 4% Asian in 2000. However, Asian Americans are overrepresented in Los Angeles; in 2000, they comprised 12% of the county, compared with about 43% Whites and 10% Blacks. If the census dictionary were constructed for Los Angeles, it is likely that the probability that the surname “Lee” belongs to an Asian or Pacific Islander would be higher than 38%.

9. I conducted a sensitivity analysis on income using the top quartile rather than the 50% cutoff in household income and home value. I found that the top 75% income threshold produces qualitatively similar results to the 50% criteria in direction and magnitude of coefficients. The analysis produced lower odds ratios across the neighborhood typologies, which is expected because homeowners who live in areas within the 50% to 75% upper threshold in income were then categorized as enclaves and communities of constraint.

10. I tested whether there were differences in results when using the county proportion of foreign born by race to test an alternative classification of enclaves. According to the ACS 2009 5-year estimates, about 68% of Asian Americans and 44% of Latinos were foreign born. Among tracts with a concentration of Latinos, a foreign-born rate of 44% was approximately the top quartile threshold in nativity. There were no Asian tracts with more than 68% foreign-born residents. Thus, I used the top quartile of foreign-born residents among Asian neighborhoods (41%). With this new criterion of 44% foreign-born for Latino enclaves and 41% for Asian enclaves, there were no qualitative differences in results in terms of statistical significance, direction, and magnitude, which supports the consistency of my findings.

11. Clark et al. (2015) found similar trends in Los Angeles, where the magnitude of Latino and Asian residents corresponded to a decrease in homogeneous White and Black neighborhoods and an increase in mixed neighborhoods.

12. Restrictive housing covenants had limited Black homeowners in the central city. After they were outlawed in the late 1940s, middle-class Black residents moved to farther western parts of the city to access improved public amenities (Chapple, 2010). However, the county has seen a decline of Black residents, with many moving to counties farther inland, particularly since the 1965 Watts Riots and recently because of housing affordability (Pfeiffer, 2012).

13. The sale price was used because of data challenges in identifying home values. First, county tax assessor information does not have home values under California’s Proposition 13. This Proposition was passed in 1978 and limits property assessment to no greater than 2% each year since the home’s 1975 base year value of assessment. The only time that a home property value is reassessed is when the homeowner changes or if new construction occurs. Zillow offers estimates of home value, but only recently offered its microdata to the public. However, Zillow has been critiqued because of user-contributed data (see Gelman & Wu, 2011; Hagerty, 2007). Thus, sale price offers the closest approximation of home value in Los Angeles. A small number of outliers were above 1.1 (0.4% of homeowners in the cohort), which may result from clerical errors or individuals with significant assets put toward their loan. Without additional data, it is impossible to differentiate between artificial or real outliers. If there are some individuals with significant loans, these are the exception and represent a small percentage of homeowners, who are not the focus of the analysis.

14. Although Home Mortgage Disclosure Act (HMDA) data do include interest rates, there are a number of issues with joining HMDA and DataQuick. Other studies have linked individual data with Home Mortgage Disclosure Act data on credit and subprime loans for the metropolitan statistical area level (see Rugh & Massey, 2010), subprime lending per 10,000 homeowners (Ong & Pfeiffer, 2008), or tract level (Ong et al., 2014). However, HMDA does not include individual property identification variables that can be linked to DataQuick. Instead, HMDA has individual loans with the tract geography. Thus, the only way to link HMDA and DataQuick would be to use the loan values, which is fraught with challenging data-link errors such as differences in reporting loan, and houses that may have similar loan amounts in the same tract. Also, HMDA recently required lending institutions to report loan borrower debt, but the rule was not enacted until January 2018 (Consumer Financial Protection Bureau, 2017). Newman (2010) outlines additional issues with linking foreclosure data to HMDA.

15. Sale price was also taken out of the model to understand whether adjusting to 2013$ added statistical significance. Although sale price did not inflate over time, the coefficients were similar and sale price added statistical power, as assessed by likelihood ratio tests.

16. The majority of homeowners in the cohort (85%) live in their home. Those who do not live in their homes were also included, for two reasons. First, homeownership is the significant driver of asset building among minorities (Taylor et al., 2011). Thus, it is important to include these properties for homeowners, even if they are purchasing homes for reasons other than residence, to understand how investments are affected because they are located in these areas. Also, previous studies have found that minority neighborhoods have a larger share of investment properties than nonminority areas do, which may affect nearby properties and the odds of default or foreclosure outcomes in enclaves and communities of constraint (see, e.g., Ellen, Madar, & Weselcouch, 2013; Hwang, 2015; Pfeiffer & Molina, 2013).

17. The surname method also undercounts Black and White homeowners based on surname, in large part because of the historical legacies of slavery (Inscore, 1983). According to the 2009 ACS 5-year estimates, about 7% of homeowners in the county were Black, and 47% were non-Hispanic White. Many surnames that are associated with Whites or Blacks do not meet the threshold over 70%. For example, Williams has a 49% chance of being White and 47% of being Black. Thus, interpretations of household racial identification for Blacks and Whites should be analyzed with caution in the logistic regressions.

18. Logistic regression models that test household or neighborhood characteristics that comprise the final model are available upon request from the author.

19. Additional tract variables were included in the original models, such as educational attainment, unemployment rates, racial composition, vacancy, and loan delinquency rates from ACS and U.S. Housing and Urban Development. Nearest school Academic Performance Index scores were also included. The full model had nonsignificant likelihood ratio tests relative to the more parsimonious model that is used in the article. These tract variables did not add statistical power because these variables are endogenous with the data used to construct the neighborhood typology.

20. These are logistic regression model factors that influence the probability of an outcome. These regressions also produce odds ratios, which compare the odds that one subgroup of homeowners will default/foreclose with the odds of another subgroup. If the odds ratio is 1, then the two subgroups are equally likely to default/foreclose. Odds ratios of less than 1 suggest that when other factors are equal, the odds of default/foreclosure for the first subgroup is less than the odds of default/foreclosure for latter subgroup, whereas odds ratios greater than 1 show the opposite pattern.

21. Preliminary analyses identified 14 Indian, 12 Korean, 37 Chinese, and one Vietnamese resurgent tracts.

Additional information

Funding

This work was supported by the National Science Foundation Graduate Research Fellowship Program [DGE-1144245].

Notes on contributors

C. Aujean Lee

C. Aujean Lee is an Assistant Professor at the University of Oklahoma in the Regional and City Planning program. She received her PhD from the University of California, Los Angeles.