Commentators on both the right and left seem to agree on one aspect of the recent mortgage crisis: government policy was at the heart of it. But they disagree on which particular government policy is at fault. The theory from the left is that financial deregulation allowed mortgage lenders and securitizers to exploit both mortgage borrowers and the investors in mortgage-backed securities. On the right, the thinking is that the government instituted policies and programs that were designed to increase credit availability and expand homeownership—policies that induced lenders to make massively risky loans.
To test these theories, researchers must identify a specific change in government policy and then explain the effects this policy change should have. They must then turn to the data to show that the predicted effects occurred soon after the government policy was instituted. A new paper by Sumit Agarwal, Efraim Benmelech, Nittai Bergman, and Amit Seru weighs some evidence related to one government policy that has long been controversial in conservative policy circles: the Community Reinvestment Act (CRA). In particular, the authors claim that the CRA played a role in the mortgage crisis by encouraging banks to make risky loans. How does this research project hold up?
History of the CRA
Before we discuss the details, here is some background on the CRA, which was enacted in 1977. In a 2003 retrospective on the law, William C. Apgar and Mark Duda noted that the CRA "was built on the simple proposition that deposit-taking banking organizations have a special obligation to serve the credit needs of the communities in which they maintain branches" (Apgar and Duda 2003, 169). The act instructed regulators to conduct periodic CRA examinations to make sure that banks were meeting the credit needs of their deposit bases. To enforce compliance, regulators had to take a banking institution's CRA record into account whenever the institution applied to consolidate with some other institution or to expand its operations with new branches (Apgar and Duda 2003, 172).
What economic effect should we expect the CRA to have? For banks, the act changes the economics of mortgage lending. In effect, it adds an extra "shadow return" to each CRA-eligible loan, over and above the loan's usual financial return. For example, the risk-adjusted return on a particular mortgage loan may be 5 percent without the CRA, but this return could rise to 6 percent after the bank factors in the benefit of the loan to its CRA compliance—and by extension, to its ability to perform a profitable merger or open a profitable branch. Simple economic theory implies that after the CRA, banks should make more and riskier loans in CRA-eligible locations, all else being equal.
The Agarwal et al. paper provides evidence that the CRA did indeed lead to more lending and also to riskier lending. The authors argue that in the three quarters before and the three quarters following a CRA examination, the average lender would make more loans and riskier loans in CRA-eligible areas.
In principle, the evidence in the paper seems consistent with the theory: government policy to encourage more lending encouraged more lending. However, other researchers have raised strong objections to the paper's empirical design. Most notably, the University of North Carolina Center (UNC) for Community Capital published a paper that claims to rebut the evidence put forth in the Agarwal et al. study, asserting that the study's entire identification strategy is invalid and therefore the results are spurious. In our reading of the paper, we found three significant issues that make us skeptical of the authors' interpretation that the CRA played a significant role in the crisis by increasing the amount of risky lending during the housing boom.
First issue: Time periods do not correspond
As the authors of the UNC paper note, the six-quarter window that Agarwal et al. use to identify the causal impact of CRA examinations "rarely corresponds to the actual period that is covered by the CRA exam." Instead, the CRA examiners typically analyze loans originated well before the actual exam date. To illustrate the issue, the UNC paper looks at a CRA exam of JPMorgan Chase that occurred in June 2011. The authors, who obtained their information from the public record of the exam (the CRA Performance Evaluation), find that the exam covered mortgage originations from January 2007 through December 2010. In contrast, the Agarwal et al. six-quarter window would have run from October 2010 to March 2012, implying an overlap of only one quarter. And even that one quarter of overlap is unlikely—the authors point out that the CRA examiner evaluated only JPMorgan's market share of lending through 2009, as the 2010 data generated to comply with the Home Mortgage Disclosure Act (HMDA) were unavailable at the time of the exam. The implication is that JPMorgan would have had no incentive to increase CRA-eligible mortgage originations in the three quarters before the examination period, since the CRA examiner was not going to consider those loans anyway.
Another relevant example (obtained from correspondence with economists from the Federal Reserve Board of Governors) is the June 2006 exam of Citibank. That exam used 2004 HMDA data for the market share analysis and used data through 2005 to compute the bank's distribution of loans to low- and moderate-income borrowers or neighborhoods. Thus, there is almost no overlap between the data used by the CRA examiners and the window employed by Agarwal et al. If the UNC paper is correct in its assertion that CRA examiners often consider loans that are outside of the six-quarter window used by Agarwal et al., then their claim that institutions were ramping up their CRA-eligible lending in order to pass their CRA examinations is flawed.
Second issue: CRA treatment effects possibly overestimated
Agarwal et al. find an increase in lending resulting from the CRA in non-CRA-eligible census tracts for both high- and low-income households. Specifically, they stratify their sample based on income terciles and find that origination rates to borrowers in the bottom-income tercile in non-CRA-eligible tracts increased by 6 percentage points around the initiation of CRA exams. This result supports their interpretation because banks would obtain CRA credit for loans to these borrowers. However, the results also show that origination rates for borrowers in the highest-income tercile in the non-CRA-eligible tracts increased by almost 4 percentage points around the initiation of CRA exams. Since these loans did not count toward fulfilling CRA obligations, the effect cannot be interpreted as a CRA treatment effect. Rather, a reasonable interpretation of this estimate is that it is picking up an unobserved factor that happens to be correlated with the timing of the CRA examinations (that is, a spurious correlation). If this is the case, then the true CRA treatment effect in CRA-ineligible tracts is really the difference between the increase in origination rates for the borrowers in the bottom income tercile and the borrowers in the top income tercile, which is an economically small 2 percentage points. Furthermore, by lending to high-income borrowers in non-CRA-eligible tracts, banks would tilt the distribution of their lending away from areas targeted by the CRA, which would end up hurting them in a CRA exam. Thus, it's difficult to imagine a scenario in which banks would target these borrowers for CRA-related purposes.
Issue 3: Securitization an unlikely explanation for effects
Agarwal et al. argue that they find significant CRA effects on lending in the 2004–06 and 2007–09 periods but not during the 1999–2003 period, and they find significant CRA effects on default rates only in the 2004–06 period. The authors' explanation for this pattern is that 2004–06 was the period in which the securitization of mortgage loans peaked, and "banks are more likely to originate loans to risky borrowers around CRA examinations when they have an avenue to securitize and pass these loans to private investors after the exam" (p. 21).
There are at least three problems with this line of reasoning. First, private securitization markets shut down in the 2007–09 period, so they couldn't possibly explain the increase in lending in CRA-eligible tracts during that period. The GSEs were very active in securitizing mortgages during this period—but they were also very active in the early 2000s, so agency securitization doesn't seem like an adequate explanation either.
Second, while it is true that securitization could alter the risk-return tradeoff for mortgage lending—it does so by allowing mortgage originators to offload their credit risk by selling their loans into mortgage-backed securities—securitization would make this offloading possible and appealing to many mortgage originators, not just CRA lenders. The result could easily be a decline in mortgage lending by depository institutions in CRA-eligible areas rather than an increase, thanks to increased competition from nondepository institutions. In fact, we would argue that the empirical evidence supports this interpretation more than Agarwal et al.'s interpretation. The late 1990s and early 2000s saw the emergence of nondepository institutions that specialized in originating subprime mortgages and selling them to securitizers. These aggressive subprime lenders were typically not subject to CRA requirements, a fact that is consistent with the shrinking footprint of CRA institutions, which we discuss in more detail below. According to Bhutta and Canner (2009), only about 6 percent of subprime loans made in 2005 and 2006 were made to CRA-targeted populations by CRA-regulated lenders. In effect, one of the consequences of the dramatic rise in private-label securitization volume was that it created lots of competition among the riskier segments of the mortgage market. This situation likely resulted in less lending by banks in CRA-eligible areas rather than more.
Finally, perhaps a more fundamental reason to doubt that securitization explains the timing of the paper's effects is that securitization has been around a long time. Laws needed to be changed before securitization could take off, but these legal changes occurred in the 1980s. So if the CRA and securitization together formed a lethal combination for the mortgage market, then why did the crisis occur in the late 2000s rather than the late 1980s?
Even if CRA encouraged risk, would it really say much about government policy?
With these caveats in mind, what would a finding that the CRA encouraged risky lending really tell us? In our opinion, a finding that the CRA encouraged risky lending would probably tell us little about the role of government in the financial crisis.
The focus needs to be on quantitative magnitudes. The question of whether or not the CRA led to an increase in risky lending of any size may not be that interesting because it is hard to imagine a world in which the CRA would not have done so. Economists begin with the premise that banks are profit-maximizing entities, so they should make all the loans that increase their expected profits. If a loan is not made, then that is because the bank must have judged the loan to reduce expected profits rather than raise them. As we described above, the CRA increases the risk-adjusted return for certain loans, so that some of the loans a bank deems unprofitable in the absence of the CRA (because of risk-adjusted returns that were too low) become profitable with CRA. Because these are marginal loans in risk-adjusted returns, then risk must be increasing.
If we start with the assumption that the CRA leads to more risky lending, the more interesting question is how much risky lending is encouraged. As it happens, the quantitative magnitudes of the estimates in the Agarwal et al. study are quite small. For example, if we assume that the appropriate CRA treatment effect should only be measured using the difference in the increase in lending between CRA-eligible census tracts and CRA-ineligible tracts, then magnitudes are trivial. Specifically, the paper finds that the CRA increased origination rates in CRA-eligible census tracts relative to CRA-ineligible census tracts by somewhere between 1 and 3 percentage points, depending on the specific quarter around the initiation of the CRA exam. When one considers that the average origination rate in the Agarwal et al. sample is 72 percent, and only 15 percent of loan originations in the sample came from CRA-eligible tracts, this is an extremely small effect. The effect becomes even smaller if you adjust the baseline estimates to take into account the likely simultaneity bias that we discussed above in the subsection titled "Issue 2."
The CRA passed long before the crisis
In concluding, we should point out that the CRA went into effect in 1977, 30 years before the financial crisis. If the CRA did shift the risk-return tradeoff for mortgage lending, then why didn't risky lending take off in 1978 rather than 2003? Moreover, the footprint of CRA-regulated institutions in the mortgage market has shrunk dramatically since the law passed. Figure 1 (taken from Foote, Gerardi, and Willen [2012]) shows that nondepository mortgage companies—which generally are not covered by the CRA—accounted for only 15 percent of mortgage lending when the CRA was passed in 1977. By the late 1990s, however, these non-CRA entities had grown to nearly 60 percent of the mortgage market. If the CRA is so toxic to the mortgage market, then it is puzzling why the act had no effect soon after its enactment, when it covered 85 percent of the mortgage market, yet led to an explosion of risky lending 25 years later, when it covered only 40 percent of the market.
Indeed, any attempt to link the recent crisis to government policies aimed at expanding mortgage credit and homeownership faces an uphill struggle. The basic problem is that the federal government has been deeply involved in housing and mortgage markets since at least the end of World War II. In particular, the Federal Housing Authority (FHA) and Veterans Administration (VA) loan programs began at about that time and were explicitly designed to extend homeownership to underserved populations. As figure 2 (also from Foote, Gerardi, and Willen, 2012) shows, the FHA and VA pioneered no and low down payment loans in the 1950s and 1960s. And as figure 3 shows, FHA loans accounted for 40 percent of loans outstanding in the 1970s and had default rates that were an economically massive 100 percent higher than non-FHA loans. In their size and their effect on housing markets, the FHA and VA were literally orders of magnitude more important than the CRA. Did government lead to risky lending? Yes! But it did so 30 years before CRA and 60 years before the recent financial crisis.
Chris Foote, senior economist and policy adviser at the Federal Reserve Bank of Boston,
Kris Gerardi, financial economist and associate policy adviser at the Federal Reserve Bank of Atlanta, and
By Paul Willen, senior economist and policy adviser at the Federal Reserve Bank of Boston