Community Reinvestment Act.
* "Did the Community Reinvestment Act (CRA) Lead to Risky Lending?" by Sumit Agarwal, Efraim Benmelech, Nittai Bergman, and Amit Seru. December 2012. NBER #18609.
In previous Working Papers columns (Spring 2011 and Fall 2012) I described papers that examined evidence of the effects of the Community Reinvestment Act on the housing bubble. The research design employed in those papers is called a "discontinuity design"; the papers compare individuals and census tracts that just qualify for credit under CRA affordability goals and those that just miss qualification. The assumption is that the legal distinction of qualification is arbitrary and thus plausibly exogenous, so simple regressions and differences in descriptive statistics are adequate tests of the CRA's effect. Loan frequency should be arbitrarily higher for those individuals and census tracts for which eligible CRA institutions receive "CRA credit" relative to loan frequency for individuals and census tracts for which CRA institutions just miss receiving credit. If the CRA effect is real, subsequent defaults should be higher for those loans that receive credit.
The behavior of the banks in those papers was not consistent with the CRA effect hypothesis because loan performance was better or no different in CRA credit areas than in non-CRA areas. In addition, troublesome "exotic mortgages" such as interest-only, negative-amortization, or teaser-rate mortgages were used by higher-income people ($141,000 average income) with high credit scores (only 7 percent of borrowers had a score below 620) to purchase more expensive houses in areas with high population growth and no price decline in the last 10 years. None of those stylized facts are consistent with the hypothesis that CRA loan affordability goals were causally important in the housing bubble.
Agarwal et al. use a different test to determine the CRA's effect. They compared banks in a particular census tract undergoing CRA exams over the years 1999-2009 with banks not undergoing exams in the same month. The loan origination rate for banks in the treatment group increased by 4 percentage points from the 72 percent average in the sample. Lending increased in CRA-eligible tracts by 8.2 percentage points. The 90-day delinquency rates increased by 0.1 percentage points from an average rate of 1.2 percent to 1.3 percent; the increase was 0.4 percentage points for loans in CRA census tracts. In the six quarters surrounding CRA exams, loan originations increased by 5 percent and loans defaulted 15 percent more often.
There is plausible exogeneity in this research design because under the CRA, small banks are reviewed every five years while large banks are reviewed every two years. The results stem from the random, calendar-driven review of some banks in an area and not others.
How can we reconcile the results of this paper with the others? In this paper, banks under scrutiny by regulators are compared to banks not under scrutiny. In the previous papers, all loans to eligible people and eligible tracts are compared to all other loans. In my view the research design of the previously reviewed papers is more useful in determining the effect of the CRA during the recent housing bubble. Thus the CRA can affect lending behavior as described in the current paper without overturning the no-effect result of the previous papers.