The Big Picture - Macro Perspective on the Capital Markets, Economy, Technology, and Digital Media

---

The Subprime Crisis: Is Government Housing Policy to Blame?

---

Keywords: Community Reinvestment Act, CRA, government sponsored enterprises, affordable housing goals, mortgages, subprime crisis.

Abstract:

---

I. INTRODUCTION

Increased homeownership has been a goal of federal policy for decades. Towards this end, several initiatives have aimed to expand access to mortgage credit, particularly to low- and moderate-income borrowers. However, experiences following the subprime crisis – particularly the loss of wealth through house price declines and the large number of foreclosures – have led some to question whether facilitating homeownership actually promotes the welfare of lower-income households.

Others have gone beyond questioning whether promoting homeownership is beneficial and have suggested that government efforts to promote homeownership may, in fact, have been a primary cause of the crisis. Peter Wallison, one of the ten members of the Financial Crisis Inquiry Commission (FCIC), issued a 100-page dissent from the FCIC’s majority report in which he identified government housing policy as the “sine qua non of the financial crisis” (Wallison, 2011, p. 2). In particular, Wallison focuses on two programs as the culprits: the Community Reinvestment Act (CRA) and the affordable housing goals imposed on Fannie Mae and Freddie Mac, the government-sponsored enterprises (GSEs). Wallison argues that these two programs, which encourage lending to lower-income borrowers, caused lenders to reduce their underwriting standards. The lower standards inflated the housing bubble and, when the bubble ultimately burst, manifested themselves in sharply higher mortgage delinquency rates. Similar arguments about the role of the CRA and GSE goals in the subprime crisis are increasingly being echoed by others.¹

Many of the studies that argue that the CRA and GSE goals played a central role in precipitating the subprime crisis – as well as those papers that have argued against this view – have not relied on hard empirical evidence. Instead, they have pointed to a general association between the existence of the CRA/GSE goals and the overall increase in lending to lower-income borrowers and neighborhoods during the buildup to the crisis (Wallison, 2009; Liebowitz, 2008). For example, some papers compare aggregated time series of loan volumes and pricing in areas favored by these regulations with areas that are not. Loan volume differences by themselves, however, are insufficient to “prove” that the regulations contributed to the elevated mortgage delinquency observed during the crisis. Instead, a link from regulation to loan performance is necessary and here, with few exceptions, the evidence is scant.

In this paper, we examine whether a link exists between these programs and subsequent mortgage performance. Our analysis relies on two empirical approaches. The first approach, which focuses primarily on the CRA, examines whether loan outcomes across low-to-moderate income (LMI) census tracts varied according to lending activity in the tract. Census tracts differ in the composition of lenders that have historically operated within the tract and these differences tend to persist over time. Since the CRA only affects some institutions, this provides a quasi-natural experiment. If the CRA caused depository institutions to reduce their underwriting standards in LMI tracts, then LMI tracts that have been disproportionally served by CRA-covered lenders historically should have experienced worse outcomes than otherwise similar tracts. Our first approach tests this conjecture by examining the relationship between activity by CRA-covered lenders and loan outcomes.

The second approach takes advantage of the fact that both the CRA and GSE goals rely on hard geographic rules that were fixed for most of the past ten years. These regulations favor loans made to borrowers in census tracts where the median family income is below a fixed threshold. If these regulations provided an incentive for – or perhaps even required – loans to be made that otherwise would not have been granted, then one might expect loans in the favored neighborhoods to perform worse, all else equal, than loans made in areas that were not favored by these regulations. Using a regression discontinuity design, we test this conjecture in the region immediately surrounding the relevant thresholds for these regulations, where each regulation’s impact should be easiest to detect.

The outline of the remainder of the paper is as follows. In the next section we provide background information about the CRA and the GSE goals and discuss the literature that has examined the relationship between these regulations and the subprime crisis. We set up our empirical tests in section 3 and present our results for the two empirical approaches in the following two sections. Section 6 discusses the conclusions we draw from our analysis.

II. BACKGROUND

The Community Reinvestment Act (CRA), passed in 1977, encourages commercial banks and savings associations to meet the credit needs of their local community in a manner consistent with safe and sound operation.² Under the CRA, the federal banking supervisory agencies assess each covered institution’s record of meeting the credit needs of its entire community, including lower-income neighborhoods. The financial institution itself is given the ability to define its “community,” or the areas in which its performance will be assessed. These “assessment areas” generally correspond to the counties in which an institution has deposit-taking offices. The financial institution is permitted to achieve its goals directly, by loan origination, or indirectly, by purchasing loans originated by others.

Although many loan types can be used to satisfy the requirements of the CRA, residential mortgage lending plays a prominent role. In part, this is because of the public availability of loan-level data on mortgage originations and purchases collected under the Home Mortgage Disclosure Act (HMDA). Since the mid 1990′s, federal bank examiners have relied upon a series of numerical measures to help evaluate compliance with the CRA. These measures include the share of loans originated (or purchased from other lenders) in LMI census tracts or made to LMI borrowers.

A census tract is designated as an LMI tract when its median family income is less than 80 percent of the median family income of the surrounding area at the last Decennial Census. For urban tracts, the surrounding area is the metropolitan statistical area (MSA) and for rural tracts it is the non-metropolitan area of the state. Borrowers are designated as LMI, regardless of the characteristics of their census tract, when their contemporaneous income is less than 80 percent of the median family income for the surrounding area, as estimated annually by the Department of Housing and Urban Development (HUD). Loans reported under HMDA are typically used for these calculations and analyses are restricted to loans within an institution’s assessment area. Based on the examiner’s evaluation, which often involves comparing an institution’s lending and purchases with those of its peers, an institution is assigned a public CRA rating of “outstanding,” “satisfactory,” “needs to improve,” or “substantial noncompliance.” Most institutions receive a satisfactory rating. These CRA ratings are considered by federal banking agencies when assessing an institution’s application for a charter, deposit insurance, branch or other deposit facility, office relocation, merger, or acquisition.

The CRA only applies to commercial banks and thrifts. Independent mortgage banks or credit unions, which together originated about 30% of all loans reported in HMDA in 2008, are not covered. Moreover, more than half of all loans originated or purchased by CRA-covered institutions are made outside of their assessment areas and thus are not considered in their CRA evaluations.³

The GSE affordable housing goals were imposed by Congress on Freddie Mac and Fannie Mae as part of the Federal Housing Enterprises Financial Safety and Soundness Act of 1992 (also called the “1992 GSE Act”). Similar to the quantitative lending activity requirements of the CRA, the GSE goals establish annual percentage of business requirements for the GSEs in terms of their purchases of mortgages falling into three categories: loans to LMI borrowers, loans to underserved areas, and loans to special affordable populations.

These terms are defined using similar concepts as the CRA. In urban areas, an LMI borrower is defined for GSE purposes as one whose income is below the median family income of the MSA (estimated, as above, by HUD). Similarly, a census tract is designated as an underserved area if the median family income of the tract is less than 90 percent of the median family income of the MSA. A tract with a median family income of up to 120 percent of the MSA median is also considered underserved if more than 30 percent of the population in the tract is minority. Finally, special affordable populations are defined based on a borrower’s income relative to the MSA median family income. Borrowers with incomes below 60 percent of the MSA median family income, or who have an income that is below 80 percent of the median and reside in census tracts with median family incomes below 80 percent of the MSA median, are considered special affordable populations. Similar, but slightly more flexible, guidelines are applied to rural areas.⁴

The numerical target levels for GSE lending goals are set in advance each year by the GSEs’ regulator (originally HUD and now the Federal Housing Finance Agency). The targeted ratios for all three of the GSE affordable housing goals have been rising over time. In assessing the GSEs’ performance in meeting these goals, non-conforming or jumbo loans (loans above a certain size), subprime loans, and government-backed loans (FHA and VA) are generally not considered.⁵

In thinking about how the CRA and GSE goals might influence the activities of mortgage lenders, one can imagine several distinct possibilities. First, the CRA and GSE goals may have little or no effect on the activities of the regulated institutions. Banking institutions may not need to undertake special activities to serve adequately the credit needs of their communities and the GSEs may be able to meet the housing goals through their normal course of business. In this scenario, neither regulation would result in more than minimal changes in the volume, pricing, or sources of credit in any area.

Second, CRA-covered institutions may extend more credit to neighborhoods receiving greater weight in CRA performance evaluations, but accomplish this through enhanced staff training, greater community outreach and marketing, or similar activities without changing the pricing of loans or underwriting standards. Such a response to the CRA might alter the sources of mortgage credit in targeted areas (as banking institutions take origination market share from institutions not covered by the law), without resulting in a net change in lending activities at the market level. The GSE goals could produce a similar effect if the GSEs can purchase more from goal-rich sources without having to alter their underwriting standards or pricing. Again, one would expect a higher percentage of goal-satisfying loans to be purchased by the GSEs, with little or no impact on the amount of lending in a market.

Third, banking institutions may respond to the CRA by offering financial incentives to borrowers from targeted neighborhoods (or sellers of mortgages from these areas) by reducing prices for credit (including transaction costs), easing credit standards, or undertaking more costly underwriting to identify applicants who are creditworthy but not obviously so. Similarly, the GSEs may opt to pay lenders more for qualifying loans or to accept loans they otherwise would not in response to the affordable housing goals. These responses, as above, will increase the share of lending accounted for both by CRA-covered institutions and the GSEs in communities favored by these regulations. If lenders respond by lowering loan prices to borrowers or by engaging in more costly and effective underwriting without modifying existing credit standards, the amount of mortgage credit extended will increase, potentially raising home values and inducing borrowers to borrow more than they otherwise would have. However, if lenders also respond by lowering their credit standards, higher rates of default and foreclosure could result.

Much of the literature on the CRA and GSE affordable housing goals has focused on the effect of the regulations on market share and loan volumes. For example, Bhutta (2010); Avery, Canner, and Calem (2003); and the Joint Center for Housing Studies (2003) examine how CRA targets affect lending activity. Similarly, Bhutta (2008); Gabriel and Rosenthal (2009); Bostic and Gabriel (2006); and Conley, Porter and Zhong (2010) examine the effects of the GSE goals. However, as noted above, demonstrating that the regulations impacted market share is insufficient to show a causal link between regulation and the subprime crisis.⁶ It is also necessary to establish that the regulations affected the quality of loans that were underwritten.

Here, there have only been a few studies. Avery, Bostic and Canner (2005) look at the impact of the CRA on bank profitability, but do so during a period in which there was little distress in the housing market. The most applicable evidence comes from Laderman and Reid (2009) who compare the performance of loans originated in California by CRA-covered lenders with otherwise comparable loans originated by others. The data used in their analysis was constructed by matching HMDA data (used to determine the lender) with performance information from the LPS/McDash database (a sample of loans serviced by 19 top mortgage servicers). Laderman and Reid find no evidence that CRA-covered loans were lower quality; indeed, they find that such loans performed better than non-CRA loans.

III. EMPIRICAL APPROACH

An ideal test of the role of the CRA and GSE goals in the subprime crisis would focus on lending activities that would not have taken place in the absence of the regulations. Since identifying such loans is virtually impossible with available public data, we rely on two indirect approaches: analyzing variation in lending and purchase activity by lender type and a regression discontinuity examination of loan outcomes around the geographic thresholds designated by the CRA and GSE goals.

In both of these approaches, the unit of analysis is the census tract, as defined by the 2000 Decennial Census. This unit has been used by regulators in evaluating the CRA and GSE goals from 2003 to the present. We restrict the sample to census tracts with a constant classification – that is, GSE goal- and CRA-qualifying or not – over the eight year period 2001 to 2008.⁷ We also limit the sample to tracts in counties that were in MSAs for the entire period, since HMDA reporting requirements for rural areas are less comprehensive. We further require that at least three home purchase and three refinance loans be originated in each tract in each year and limit the construction of all HMDA-based statistics to first-lien loans for owner-occupied properties.⁸ Finally, to account for the role of significant cross-market variation in performance and lending patterns, all of our analysis is “within market.” That is, we either express variables as deviations about MSA means or add MSA fixed effects to all of the estimated equations.

Our primary outcome measure is the percentage of mortgage borrowers in a census tract who were 90 or more days past due on at least one mortgage obligation at the end of 2008,⁹ as determined from the records of Equifax, one of the three national credit bureaus.¹⁰ Other outcome measures are used in supplementary analyses. These include the share of first-lien mortgage loans originated in a tract during 2004-2006 that had estimated front-end payment to-income ratios (PTIs) exceeding 30 percent, generally considered marginal in underwriting, and the share that were reported as higher-priced in HMDA, which is often used as a proxy for high-risk or subprime lending activity (Avery, Brevoort, and Canner, 2007).¹¹ These outcome measures focus on lending activity during the period 2004 to 2006, because this was the high-water period for the subprime market, before the market collapse that began in 2007. Finally, we use estimates of house price changes from 2001-2006 and 2006-2008 as additional outcome measures to explore whether the CRA or GSE goals contributed to house price appreciation in the earlier period or depreciation in the later period. Tract-level house price appreciation is estimated using median home purchase loan sizes from HMDA in each tract over time.

In both components of the analysis, we use a common set of tract-level control variables. These variables include a set of “baseline” controls that are limited to variables that can truly be considered as exogenous and measured well before the loans that contributed to the subprime crisis. Primarily these are Census 2000 variables, but also include the relative income of the tract in the 1990 Census and the mean credit score of mortgage borrowers in the tract which is calculated from data from Equifax at the end of either 2000 or 2004.¹²

For the delinquency outcome variable we estimate an additional equation which includes a set of “expanded” controls. The expanded controls are calculated from HMDA data over 2004-2006. These controls capture information about the characteristics of the borrowers and the loans that they took out over this period. The expanded controls include the share of loans extended in each tract in 2004-2006 that were reported as being higher-priced, had high PTI ratios, were underwritten without income, or involved a “piggyback loan,” which is a junior-lien loan that was taken out at the same time as the first lien. We also include several measures of borrower income in the expanded controls to account for the potential impact that the borrower-based CRA and GSE preferences may have had. Two of the expanded controls, the share of loans with a high PTI and the share that were reported as higher priced, are also used as outcome measures in supplementary analyses.

In some estimations, we use only the baseline controls because of concerns that the expanded controls might not be exogenous, and thus their inclusion might skew our results. For example, if the CRA caused banks to lend to more low-income borrowers and these borrowers were more likely to become delinquent, then controlling for the share of low-income lending might inappropriately reduce the estimated effect on loan outcomes that is attributed to the CRA. On the other hand, if the expanded control variables are independent of the lending effects induced by the CRA or GSE goals, then the inclusion of these variables in the estimated delinquency equations improves the precision of our tests.

IV. APPROACH 1: VARIATION BY LENDER TYPE

Our first approach examines differences in loan outcomes associated with variations in the type of lender serving census tracts eligible for both the CRA and GSE underserved goals. If CRA-covered lenders reduced their lending standards as a result of the regulation, then those tracts with relatively more CRA-covered lending activity should have experienced worse outcomes than similar tracts with fewer covered lenders. If the GSE goals had a similar effect on lending, then those tracts that have proportionally more loan sales to the GSEs should have experienced worse outcomes.

We divide lending activity in each census tract into the share accounted for by six different institution types:

1) Depository institutions lending outside of their assessment area;¹³

2) Depository institutions lending within their assessment area;

3) Affiliates of depository institutions lending outside of their assessment area;

4) Affiliates of depository institutions lending within their assessment area;

5) Credit unions; and

6) Independent mortgage companies.

If the CRA caused lenders to loosen their underwriting standards, we would expect tracts with a larger share of within-assessment-area lending by depository institutions, or their affiliates, to have experienced worse outcomes.¹⁴ We include these loan shares as independent variables in the estimations in this section, with the loan share of independent mortgage companies serving as the omitted group.

In addition to originations, lenders can also meet their CRA requirements by purchasing loans. To account for the possibility that depository institutions may have purchased loans to satisfy the requirements of the CRA and GSE goals, we also include the share of loans originated in a tract that were purchased by each of the six institution types.¹⁵ If the CRA caused depository institutions to purchase low-quality loans, then we would expect those neighborhoods with more purchases by CRA-covered institutions (or their affiliates) to have experienced higher delinquency rates. We also include the share of loans in the tract that were sold to the GSEs to determine whether a higher share of loan sales to the GSEs was associated with worse outcomes. Because of our concerns about exogeneity, we measure these “share of lending” and “purchase” variables at two points in time. These “control periods” include 2001, which we select because it is safely before the start of the housing boom, and 2004-2006, which captures market activity during the height of the subprime market. Each model is estimated “within-MSA” (uses MSA fixed effects) using 2000 Census tracts as the unit of analysis.

A complete listing of the variables used in this phase of the analysis, along with their definitions and sample means, is presented in table 1. Table 2 provides the results of our estimation using the delinquency rate of mortgage borrowers in the tract at the end of 2008 as the dependent variable. Columns (1) and (2) use the baseline controls and the share of lending variables calculated using the 2001 and 2004-2006 control periods, respectively. Column (3) presents the results of the estimation using the set of expanded controls, with 2004-2006 as the control period.

The results presented in table 2 suggest that within-assessment area lending, by either depository institutions or their affiliates, was associated with lower 2008 delinquency rates than similar tracts that had less lending by these institutions and more lending by independent mortgage banks (the omitted group). A comparison of the impact of in- and out-of-assessment area lending (the coefficients in the first four rows of the table) also supports the view that CRA lending is associated with better, not worse, loan quality. In all but one case, the within-assessment area coefficient is more negative than the comparable out-of-assessment area coefficient, although the difference is generally not statistically significant. GSE sales are also negatively associated with delinquency, though generally not at significant levels.

The evidence regarding the share of loans purchased by depository institutions within their assessment areas is mixed. Within-assessment-area purchases by CRA-covered institutions are positively associated with 2008 delinquency rates when 2001 is used as the control period, but negatively associated when 2004-2006 is used. This suggests that CRA-covered lenders shifted their within-assessment-area purchases towards less risky census tracts during the middle of the decade, which appears inconsistent with the CRA having induced depository institutions to purchase riskier loans during the run up to the subprime crisis. In addition, the magnitude of the effect found for 2001 is quite small. Since within-assessment-area purchases by CRA-covered lenders represented only 3 percent of loan originations during 2001, this implies that, on average, loan purchases were associated with delinquency rates that were 0.12 percentage points higher. The magnitude of this effect appears inconsistent with CRA-related purchases having played a large role in elevating delinquency rates.

A possible explanation for the lack of a clear relationship between either lending or purchases by CRA-covered institutions and subsequent delinquency is that only those few institutions that choose to pursue an “outstanding” CRA rating need to alter their behavior, whereas most other institutions can achieve a “satisfactory” rating through their normal course of business. In this case, worse outcomes from the CRA would only be associated with lending activity from outstanding-rated institutions. To test for this, we subdivide the share of lending and purchases by depository institutions and their affiliates operating within their assessment areas into the share accounted for by outstanding-rated institutions and by satisfactory-rated institutions.¹⁶ The results from these estimations, shown in table 3, exhibit only small differences between satisfactory- and outstanding-rated institutions. In each estimation, within-assessment-area lending by outstanding-rated institutions was associated with significantly better, not worse, loan performance than within-assessment-area lending by satisfactory-rated institutions. These results also continue to show mixed evidence of loan purchases by within-assessment-area depository institutions, though it is notable that the positive effect of loan purchases observed earlier when 2001 was used as the control year derive entirely from purchases by outstanding-rated depository institutions.

Another possibility is that our analysis may rely on too high a level of aggregation and obscure the fact that the subprime boom took on very different forms in different parts of the country. In particular, the CRA and the GSE housing goals may only have had an effect in those markets where lending activity grew the most, perhaps in response to local economic conditions or house price dynamics. To examine this possibility we divide the sample into three groups of states. These groups include the “sand states” of Arizona, California, Florida, and Nevada which experienced very rapid rates of loan growth; the “rust belt” states of Indiana, Michigan, Illinois, Wisconsin and Ohio which were relatively stagnant markets; and all other states.

The estimations for these three state subgroups are presented in columns (1) through (3) of tables 4A and 4B (for control years 2001 and 2004-2006, respectively). Results for the three state subgroups continue to show that lending by CRA-covered institutions was generally associated with lower levels of delinquency. The lone exception to this is the positive coefficient on depositories in their assessment areas in the sand state estimation that uses 2004-2006 controls. This coefficient is not statistically significant and is lower than the coefficient on lending by depositories outside of their assessment areas, suggesting that the positive effect likely reflects differences in the business models of institutions rather than the CRA. Coefficients in the other two state groups are consistent with those of the overall regressions. Results for loan purchases by depository institutions within their assessment areas continue to produce mixed results, though in the estimations by geography the coefficients are generally insignificant.

Another possible explanation of our results is that CRA regulators may have been more concerned with lending to minority populations than to low- or moderate-income borrowers (although there are no explicit racial targets in the CRA regulations). In this case, the CRA and GSE housing goals may only have induced risky behavior by lenders in neighborhoods with high minority population shares. To test this possibility, we restrict the sample to those census tracts that had minority population shares that exceeded 30 and 50 percent. The results from these estimations are shown in columns (4) and (5), respectively, of tables 4A and 4B. These results provide no evidence that either the CRA or sales to the GSEs are associated with higher delinquency rates in these census tracts. The coefficient on in-assessment area purchases by depository institutions remains positive and significant when 2001 is used as the control year and negative and statistically insignificant when 2004-2006 is used, both with magnitudes that remain quite small.

The results that we have presented thus far are based on the 2008 delinquency rate as the outcome variable. It may be that insufficient time had elapsed between the subprime loan buildup and 2008 to allow the full impact of lower lending standards to be reflected in delinquencies. Thus, as a robustness check, we also conducted similar analyses using more direct measures of loan quality during the peak 2004-2006 lending years. The alternative measures, both of which are included in our expanded controls, include the share of loans that had a high PTI ratio and the share that were reported as higher priced in HMDA.

The results of these estimations, which are shown in columns (1) and (2) of tables 5A and 5B, are consistent with our earlier findings. Tracts with more within-assessment area lending by depository institutions had less high-PTI lending and fewer loans reported as higher priced, using independent mortgage companies as the control group. The coefficients from the high-PTI estimation suggest that this negative relationship was weaker than the relationship between outside-of-assessment-area lending and delinquency, though the difference is very small and a similar relationship is not found for affiliates or when the share of higher-priced lending is used as the dependent variable. Within-assessment-area purchases by depository institutions are positively associated with high-PTI lending and the share of higher-priced lending, though as with the results for delinquency the size of the coefficients suggest the magnitude of this effect is small.

So far we have focused on indicators of loan quality. As discussed earlier, the CRA and GSE housing goals could have affected the mortgage market not by lowering underwriting standards but by inducing lower mortgage rates in favored areas which had the effect of increasing the demand for housing in these areas. Such an increase in demand may have contributed to the increase in house prices observed during the boom period of the decade, and then potentially to price declines at the end of the period if the earlier increases were unsustainable.

To test this possibility, we use house price changes over two periods as outcome measures in our regressions. Tract-level house-price changes are calculated for 2001-2006 and 2006-2008 using the median size of home purchase loans reported in HMDA. These measures rely on the assumption that loan-to-value relationships remained constant over these periods. Because of concerns about endogeneity, we restrict ourselves to the baseline controls and use 2001 as the control period.

Column (3) of tables 5A and 5B uses the tract-level change in house prices between 2001 and 2006 as the dependent variable and columns (4) and (5) use the change from 2006 to 2008. The estimation reported in column (5) includes the lagged 2001-2006 price increase as a control, and thus the equation can be interpreted as measuring the change in house prices appreciation rates.

Within-assessment area lending by depository institutions appears to be positively associated with house price changes during the 2001-2006 period. A positive association is observed for 2006-2008 as well. This suggests that, to the extent the CRA induced higher lending volumes that contributed to house price appreciation, the resulting price increases were sustainable. Indeed, all depositories (including credit unions), are less associated with price declines during the 2006-2008 period than the less-regulated independent mortgage banks. GSE sales are negatively related to house price appreciation during both periods, although the measured relationship in the latter period is small and insignificant.

The share of loans purchased by depository institutions within their assessment areas is positively associated with house price changes from 2001-2006 and negatively associated with changes from 2006-2008. This result is consistent with loan purchases by CRA-covered institutions having contributed to the boom and bust in house prices on both ends; however, neither of these effects is statistically significant at the 5 percent level.

In sum, there is little in the results presented in this section to suggest a link between the share of lending accounted for by CRA-covered lenders and either lower loan quality or house price appreciation that may have contributed to the subprime crisis. Indeed, the evidence suggests that, all else equal, LMI tracts served by CRA-covered lenders show fewer, not more, loan delinquencies in 2008 than tracts served by lenders not subject to the CRA. Our results also provide little or no evidence that within-assessment-area loan purchases by depository institutions contributed to the subprime crisis and no evidence of a statistically significant relationship between loan sales to the GSEs and delinquency.

Read the rest of this entry »