Crucial new mortgage study: "Liar's Loan?"

Here are excerpts from an important new paper on the causes of the Mortgage Meltdown, "Liar's Loan?" by three academic economists from Columbia, Indiana, and Yale. They obtained records on 721,767 mortgages handed out from January 2004 to February 2008 (including deliquencies up through January 2009) by a big, very aggressive national mortgage bank. The economists managed to match most of them up with ethnicity info from the Home Mortgage Disclosure Act database.

They don't say who the bank is (or, more likely, was) but it septupled the number of mortgages it gave out from the first half of 2004 to the second half of 2006. It gave out a lower percentage of subprime loans than the national average, but it did specialize in low-documentation (liar loans) mortgages.

The results aren't terribly surprising. Low-document (i.e., liar loans) have higher delinquency rates than full document loans, and loans originated by local mortgage brokers who sold them to the bank did worse than the loans originated by the bank itself. The model, therefore, is one of predatory securitizing from the ground up.

The usual ethnic differences in delinquency are observed:

"In the full sample, the ranking of delinquency rates by race/ethnicity is as follows: white (24.7%), Asian (27.1%), black (37.4%), and Hispanic (40.2%)."

That's not as great as the huge ethnic differences (3.3x for blacks vs. whites, 2.5x for Hispanics, and 1.6x for Asians) seen even in credit score and income-adjusted foreclosure rates from the state of California in the San Francisco Fed study by Laderman and Reid. (I've filed a Freedom of Information Act request to get the crucial raw data for California from the SF Fed).

The narrower delinquency rates seen in this one bank study are not surprising, since the bank was more or less pursuing a certain class of borrowers (i.e., iffy ones).

Liar’s Loan?

Effects of Origination Channel and Information Falsification on Mortgage Delinquency

Wei Jiang, Ashlyn Aiko Nelson, Edward Vytlacil

This Draft: August 2009

ABSTRACT This paper presents a comprehensive predictive model of mortgage delinquency using a unique dataset from a major national mortgage bank containing all of its loan origination information from 2004 to 2008. Our analysis highlights two major agency problems underlying the mortgage crisis: an agency problem between the bank and mortgage brokers that results in lower quality broker-originated loans, and an agency problem between banks and borrowers that results in information falsification by borrowers of low-documentation loans--known in the industry as Alt-A or “liars’ loans”--especially when originated through a broker. We also document significant differences in loan performance by race/ethnicity that cannot be explained by observable risk factors or loan pricing.

...The crisis that started from the mortgage market quickly spread to other financial markets and throughout the economy...

We use the experience of a major national mortgage bank to uncover the determinants of the mortgage crisis and the evolution of the crisis at a micro level.... Loans issued by the bank since the beginning of 2004 reached a cumulative delinquency rate of 28% by early 2009; approximately half of these delinquent loans were in the state of short sale or foreclosure. Finally, the borrowers and properties underlying the bank’s loans during our sample period have fair representations in all 50 states. Therefore, lessons from this particular bank have general implications for the national mortgage market. The proprietary data set represents the most comprehensive, detailed, and disaggregated data set in the mortgage loan literature. Our data set consists of all 721,767 loans that the bank originated between January 2004 and February 2008. ...

We provide evidence of borrower information falsification at both individual variable and aggregate levels. ...

Finally, we document significantly higher delinquency rates among Hispanic and black borrowers. The differences in delinquency rates--4 to 11 percentage points higher for Hispanics and 3 to 4 percentage points higher for blacks, relative to white borrowers--are not explained by the full set of individual risk factors collected at loan origination, or by differences in loan pricing. Our analysis--which includes far more detailed data than that used in prior research on the relationship between race/ethnicity and credit-- does not support a finding of discrimination, whereby minorities are subjected to higher lending standards or higher pricing for given financial products. Rather, the findings suggest that systematic differences between white and minority borrowers--such as information and experience disparities resulting from a lack of prior home buying experience or exposure to mainstream financial institutions--may explain these delinquency differences. ....

The housing boom welcomed many first-time homebuyers to the mortgage market. In early 2004, only 7.6% of borrowers in the sample were first-time homebuyers, a figure that climbed to 18.1% by late 2006. As the housing market collapsed and lenders tightened standards, the percent of first-timers fell to 12.7% by the end of 2007. During the sample period, black and Hispanic borrowers gained a significantly higher share of new loan originations. In early 2004, they represented 4.5% and 7.5% of the borrower population, respectively; by early 2007, the percentages were 8.9% and 23.3%. More strikingly, the proportion of blacks and Hispanics who were first-time borrowers increased from 10.3% in early 2004 to more than 25% in late 2006. The national mortgage market experienced a similar increase during the same period in the percentage of first-time homebuyers and the expansion of credit to minority households, who were disproportionately first-time homebuyers. According to national HMDA data on home purchase loans,8 6.6% (10.8%) of borrowers were black (Hispanic) in 2004; the numbers increased to 8.7% (14.4%) in 2006.

The graph in Figure 3 in the report shows that early 2007 vintage loans had radically higher delinquency rates than early 2004 loans: 31% of the early 2007 loans were delinquent by early 2009 compared to 12% of early 2004 loans. One reason was the lower prices in 2004 than in 2005-2007; another was that the bank was scraping deeper and deeper into the bottom of the barrel, dragging up ever more dubious borrowers.

... Column 1 of Table 4 Panel A indicates that the following variables predict a higher likelihood that a borrower will obtain a loan from a broker rather than from the bank: high debt level, original purchase (as opposed to refinance), first lien, first-time owner, owner-occupied, low income, low credit score, female borrower, minority borrower, young borrower, short employment tenure, and self-employed. All non-white racial groups favor the Broker channel in comparison to whites. Most of these characteristics (except perhaps the first-lien and self-employed variables) are associated, on average, with lower financial sophistication, less experience with mortgages, and lower credit quality. This relation calls attention to the issue of irresponsible lending--lending without due regard to ability to pay, to poorly informed borrowers--as analyzed by Bond, Musto, and Yilmaz (2008) and Inderst (2006).

The government spent years telling the mortgage industry to diversify its broker forces. When it finally did, that just greased the skids for Spanish-speaking and black mortgage brokers to dredge up bad borrowers who didn't have a clue what they were getting themselves talked into by their co-ethnics.

The variables that predict choosing a low-doc loan have the following contrasts with those that predict choosing a broker. First, borrowers with low loan-to-value (LTV) ratios but high loan size are more likely to choose low documentation. Second, first-time owners and those purchasing owner occupied properties are less likely to choose low documentation. Third, borrowers with high income and credit scores tend to choose low documentation. Fourth, black borrowers do not appear disproportionately in low documentation loans, while Hispanic and Asian borrowers do.

An examination of credit scores by race reveals that average credit scores are highest among Asian and white borrowers, and lowest among Hispanic and black borrowers. Hispanic borrowers who obtain loans directly from the bank have credit scores that are comparable to those of white borrowers, but those who obtain loans through a broker have credit scores that are on average 2-5 points lower. Black borrowers have average credit scores that are 14-27 points lower than white borrower credit scores, across all subsamples.

Remember, this is just for borrowers from this dubious bank, not for the overall population.

Section V offers a more detailed analysis of these race/ethnicity effects. Last, the time trend of credit scores, as shown by the year dummy variable coefficients, is informative; while Bank loans saw steady improvement in credit scores over time from 2004-2008, credit scores for Broker loans deteriorated from 2004-2007, and only recovered in 2008. The findings provide evidence that the bank pursued a growth strategy which relied on penetrating marginal borrowers through the broker channel.

A "Don't Tell Us, We Don't Want to Know" relationship between the bank and the mortgage brokers.

V. “The Color of Credit:” Race/Ethnicity and Loan Performance There is a large body of research dedicated to exploring disparate impact on minorities in credit markets and in the mortgage market in particular. A common challenge in this line of research is distinguishing between the effects of disparate impact and discrimination, because most researchers pursuing this question do not have access to the full set of variables to predict loan pricing and performance (see Ross and Yinger (2002) for a full analysis of challenges in identifying racial discrimination in the mortgage market). As an example, the landmark Boston Fed Study (Munnell, Tootell, Browne, and McEneaney (1996)) found that race strongly predicted loan approval among applicants even after controlling for a long list of personal characteristics and individual risk factors, though their estimated race effects were smaller than those found in earlier studies employing a smaller set of control variables. Yet their study did not include other important covariates--such as credit score--which strongly predict loan performance, and did not have information on ex post loan performance. Thus, the study was unable to conclude whether the disparate loan approval rates across race resulted from legitimate economic considerations or from discrimination.

Our findings complement this line of prior research by including additional covariates and by relating loan performance to race/ethnicity. In the full sample, the ranking of delinquency rates by race/ethnicity is as follows: white (24.7%), Asian (27.1%), black (37.4%), and Hispanic (40.2%). Controlling for observable characteristics, the black-white (2.8 to 5.2 percentage points) and Hispanic-white (5.9 to 8.3 percentage points) differences are statistically significant at the 1% level in all four subsamples, while the Asian-White differences (-1.1 to 1.1 percentage points) are not significant even at the 10% level. Notably, the difference in the delinquency rates between white and black/Hispanic borrowers is more than 50% higher in the Broker subsamples than in the Bank subsamples.

It would be very interesting to track delinquency rates by the race of the broker.

We must also control for loan pricing in order to attribute these delinquency differences to race/ethnicity. If certain racial/ethnic groups pay higher interest rates conditional on other characteristics, then the heavier payment burden could cause higher delinquency. Such a concern is warranted by prior research on consumer financing. Charles, Hurst, and Stephens (2008) show that blacks pay significantly higher rates when financing a new car, in large part because blacks are more likely to use more expensive financing companies. Similarly, Ravina (2008) finds that black borrowers in an online lending market pay rates that are over 100 basis points higher than comparably risky white borrowers. Much of the difference can be attributed to favorable interest rates obtained in same-race lender-borrower pairings and the underrepresentation of black lenders. In the context of mortgage lending, price differences could occur by pricing a given product differently for borrowers from different demographic groups, but more likely occurs through steering uninformed borrowers into more costly products, such as subprime loans, when more attractive products are available; or through aggressive negotiation strategies used by brokers to enhance their fees and commissions (known in the industry as yield spread premiums).

We find no evidence that black or Hispanic borrowers pay higher initial or current interest rates on bank-originated loans, conditional on observable individual risk factors. However, among broker originated loans, black borrowers appear to pay higher rates, on the order of 10-16 basis points, while there is no clear evidence that Hispanic borrowers are subject to higher loan pricing. While the coefficients on Black are significant and positive in the Broker subsamples, the magnitudes are much lower than those documented in other credit markets (e.g., Charles, Hurst, and Stephens (2008) and Ravina (2009)). The estimated gender effect is insignificant throughout, both in terms of loan pricing and loan performance. Our results are closer to findings in Haughwout, Mayer, and Tracy (2009) that there is no adverse pricing by race, ethnicity, or gender in either the initial rate or the reset margin based on a sample of subprime loans originated between 2004 and 2006.

Our data suggest that loan pricing is an unlikely explanation for the higher delinquency rates observed among black and Hispanic borrowers. Black borrowers exhibit higher delinquency rates relative to white borrowers, even for bank-originated loans for which we find no evidence of unfavorable pricing. The average (median) unpaid balance on loans among black borrowers is $185,000 ($150,000). Thus, the estimated black-white difference in interest rate among broker-originated loans--10-16 basis points-- amounts to an additional monthly payment of $15-$25 (or $13-$20) using the mean (or median) balance. It is unlikely that such a difference could be pivotal in loan delinquency.

Moreover, Hispanic borrowers exhibit the highest delinquency rates in our sample among all demographic groups, although there is no evidence that they face unfavorable interest rates in comparison to other groups.

Previous work sheds light on the unobserved risk factors that are correlated with race/ethnicity variables. First, blacks and Hispanics have lower savings rates on average than whites of similar age, education and income (Blau and Graham (1990), Charles, Hurst, and Roussanov (2007)). As a result, they accumulate less wealth (often difficult to measure), making them more vulnerable to adverse economic shocks. Second, minorities are less likely to have family or relatives who can help when they have trouble meeting their mortgage payments (Yinger, 1995).

Third, Guiso, Sapienza, and Zingales (2009) offer an interesting explanation for the highest delinquency rates observed among Hispanic borrowers. Based on survey data, the authors find that Hispanics are much less likely (between 18 and 27 percentage points) than blacks or whites to feel morally or socially obligated to continue paying their mortgages when the equity value is significantly below zero.

Wow. Don't tell the SPLC or they'll come after Wei Jiang, Ashlyn Aiko Nelson, Edward Vytlacil with pitchforks.

Historically, policymakers and researchers concerned with mortgage lending discrimination have focused on two key issues: unequal access to credit (i.e., disparities in loan approvals and denials) and pricing disparities. While we do not examine differences in mortgage approvals by race, our analysis suggests that the housing boom fueled a rapid expansion of credit among Hispanic and black borrowers. Moreover, the share of first-time borrowers among black and Hispanic households grew from 10% in early 2004 to 25% in late 2006. In addition, we find little evidence of pricing discrimination as a cause for loan delinquency.

Taken together, the findings suggest that market dynamics and credit expansionary practices during the sample period may have alleviated some of the inequalities in credit access and pricing. Yet the ex post loan performance data suggests that such credit expansion was achieved largely through lowered lending standards, particularly among brokers originating low-documentation loans. The persistence of Hispanic and black race effects in the delinquency models raises further questions, including whether such borrowers were well-informed about the mortgage process and possessed the requisite experience and knowledge to continue making their mortgage payments in full and on time.

In other words, the government finally got what it had been asking for, and got it good and hard.

Our analysis raises the question of why this major mortgage bank—as well as other market players—allowed such deterioration in borrower and loan quality to persist before tightening its lending standards. A plausible explanation is that the expansion of the secondary mortgage market and the ease of loan securitization weakened the bank’s incentive to screen borrowers by allowing the bank to offload risk.

Predatory securitizing.

My published articles are archived at iSteve.com -- Steve Sailer