Impacts of the Community Reinvestment Act on Neighborhood Change and Gentrification

Abstract

The Community Reinvestment Act (CRA) encourages bank lending in low- and moderate-income areas. We use a regression discontinuity design that exploits the relative-income threshold that distinguishes CRA-eligible from ineligible neighborhoods (census tracts) and find little evidence that CRA has contributed to neighborhood changes associated with gentrification in eligible areas. Over the 1989–1999 period, we find that eligible tracts had greater increases in mean income relative to ineligible tracts, but we find little evidence that the CRA caused decreases in the proportion of long-term residents or increases in the proportion of White or college-educated residents.

Keywords::

• CRA
• community reinvestment act
• gentrification
• housing
• low-income lending

Notes

^1. “If the loans are for homes or multifamily housing located in an area for which the local, state, tribal, or Federal government or a community based development organization has developed a revitalization or stabilization plan (such as a Federal enterprise community or empowerment zone) that includes attracting mixed-income residents to establish a stabilized economically diverse neighborhood, examiners may give more consideration to such loans, which may be viewed as serving the low-or moderate-income community's needs as well as serving those of the middle- or upper-income borrowers.” Such loans may also be considered if there is “other evidence of governmental support for a revitalization or stabilization project in the area” (Comptroller of the Currency, Federal Reserve System, & Federal Deposit Insurance Corporation, 2010)

^2. We thank an anonymous referee for suggesting the term neighborhood change and clarifying our thinking on this point.

^3. That is, “one or more [metropolitan statistical areas]…or one or more contiguous political subdivisions, such as countries, cities or towns” that include the census tracts “in which the bank has its main office, its branches, and its deposit-taking ATMs, as well as the surrounding [census tracts] in which the bank has originated or purchased a substantial portion of its loans” (Barr, 2005).

^4. An anonymous referee clarified for us that while community groups can file administrative challenges with financial-institution regulators, they cannot bring lawsuits in court.

^5. Among the information required are (1) the loan amount, (2) the purpose of the loan (home purchase, home improvement, refinancing), (3) the type of property involved (single-family, multifamily), (4) the location (state, county, MSA, census tract), (5) the race of the borrower, (6) the gender of the borrower, and (7) whether or not the loan was granted.

^6. Immergluck (2007) gives a discussion of qualitative and quantitative aspects of the tests.

^7. For the purpose of CRA evaluations, “moderate-income” households or neighborhoods are those with median incomes between 50% and 80% of the MSA median. A “low-income” household or neighborhood is one with a median income below 50% of the MSA median (B. Lee & Campbell, 1997).

^8. To assure that the census tracts were similar, Avery et al. (2003) controlled for tract characteristics These included (1) the starting rates of each of the dependent variables in 1990, (2) the proportions of the tract's population under 18 and over 65, (3) the number of people per household, (4) the percentage of the tract's population that was non-White, (5) the median income of the tract relative to the MSA, (6) the percentage change in total housing units, and (7) the percentage change in the tract's median income.

^9. See Berry and Lee (2007) for detailed explanation with an application to the CRA or Baum-Snow and Marion (2009) for an application to low-income housing credits. Thistlethwaite and Campbell (1960) provided the first description and application.

^10. Census tracts are used to demark neighborhoods.

^11. PMSA (primary metropolitan statistical area) values came from the American FactFinder database U.S. Census Bureau (2011). A PMSA is defined by the Office of Management and Budget (2010) as “a large urbanized county or a cluster of counties (cities and towns in New England) that demonstrate strong internal economic and social links in addition to close ties with the central core of the larger area.”

^12. This is the assumption of local independence (D. S. Lee, 2008).

^13. Nichols (2007) provides a detailed description and cites Cheng, Jianqing, and Marron (1997) to suggest that a triangle kernel has advantages for predictions at a boundary point.

^14. An alternate approach for conditioning on covariates is to obtain residuals from regressing the outcomes on covariates and to look for regression discontinuity in the residuals, as is done in Avery et al. (2003). D. S. Lee and Lemieux (2010) note that both approaches are legitimate.

^15. Geolytics (2011) describes the normalization process this way: “The basic procedure was to use geographic information system (GIS) software to overlay the boundaries of 2000 tracts with those of an earlier year. This allowed us to identify how tract boundaries had changed between censuses. We then used 1990 block data to determine the proportion of persons in each earlier tract that went into making up the new 2000 tract. For example, if a 1990 tract split into two tracts for 2000, the population may not have been divided evenly. Our method allows us to determine the exact weight to allocate to each portion.” For our purposes, it would have been better to conform the data to 1990 boundaries (because this is our baseline), but this normalization was not available to us.

^16. Tracts missing median family income data in either year studied were also dropped. This restriction reduced the sample size by less than 1% of the total sample.

^17. Several studies have used the 0.7 to 0.9 relative-income window to define the sample of tracts in studies of the CRA (Avery et al., 2003; Berry & Lee, 2007; Gabriel & Rosenthal, 2009). As robustness checks, Berry and Lee (2007) and Gabriel and Rosenthal (2009) also used a 0.75 to 0.85 relative income window. Gabriel and Rosenthal (2009) found that the 0.7 to 0.9 window produced somewhat stronger results than the full sample.

^18. Our earlier work also included real median family income. Results were similar to those for real mean family income.

^19. A common bandwidth standardizes the samples for the dependent variables to allow better comparisons. We chose the middle value across the dependent variables of the optimal bandwidths provided by Nichols's “rd” Stata command, explained above. Later we show alternate bandwidths.