HMDA: Using HMDA Data
The 2004 HMDA data is finally out and the numbers show what we expected and feared: Black Americans fare less well in the mortgage market than non-Hispanic whites. Now what?
The reports are likely to generate the highest level of loan analysis activity we have seen in some time. Many of the analyses will be sound. Some may be driven by specific agendas. No matter who does the analysis and what the study purports to find, it is essential that financial institutions understand the data at several levels: industry-wide, market-wide, and proprietary.
The federal financial regulatory agencies have undertaken lending data analysis for some time - 30 years to be precise. The first data was collected in 1974 from banks and thrifts. Credit Unions were not part of the experiment as they were not making mortgages at the time.
The four bank and thrift regulatory agencies collected data on loan applications using three different models. Each model was used in six cities across the country, selected for size, mortgage market activity and racial diversity. One model was strikingly similar to the early HMDA collection, looking at aggregate lending patterns. Another model collected loan-specific information including the race of the applicant and the loan decision. The third model collected some applicant-specific information, such as income.
Having collected this information, the agencies then had to do something with it. After a few months of metaphysical meditations, the analysis began. All three collection systems showed the same thing: Black loan applicants were turned down three times as often as white loan applicants. The turndown experience of Hispanics and Asians varied, apparently affected as much by location as by any other factor. For example, Asians in Buffalo, NY earned higher mortgage approval rates than whites, while Asians in San Francisco earned lower approval rates than whites.
The differences in approval rates were too significant to be attributed to chance or something equally quirky. The agencies faced the necessity of explaining the differences in treatment by race. The explanations offered relied on information that was not collected, such as credit history or employment stability. All explanations offered were simply speculative as no information had been collected to support them.
Only the third approach had any information about the applicant. Hoping that different income levels would explain the racial discrepancy, the agencies ran an analysis controlling for income. The result: income did not explain the discrepancy. High income Blacks suffered the same fate as lower-income Blacks.
Progress? What is striking about a comparison of the 1974 analysis with the 2004 analysis is that the numbers are not very different. We still find the numbers showing applicants sorted by race in the same relative positions. In fact, the data found that Blacks are still turned down almost three times as often as non- Hispanic whites. Hispanics still don't fare as well as Whites and Asians do well in some areas and not in others.
What is wrong with this picture? For almost 30 years, since the first data collection and analysis, the banking industry has worked hard at fair lending. Efforts have included affirmative marketing, flexible underwriting, training, outreach, and self-testing. Some institutions have even built in rewards or motivators for loan officers to make loans to affirmatively targeted groups or areas. But the numbers don't change.
Federal Reserve System fair lending experts Bob Avery, Glenn Canner, and Bob Cook, studied the 2004 HMDA data and have published their analysis and findings in the Summer 2005 issue of the Federal Reserve Bulletin. Their findings are interesting. Some are not new, but simply observations based on the data. However, some observations raise some intriguing ideas and may shed some light on what the industry can do to actually change the numbers.
According to the study's analysis of denial rates, gender discrimination did not appear to occur. While some variations were found by location, product, and price, the differences were not significant and often went in different directions.
Racial and ethnic differences are a different matter. In fact, the raw 2004 HMDA numbers show denial rates by race or ethnicity that are not very different from the denial rates in 1974. The authors of the study took several steps to control for borrower-related factors which reduced the differences by more than half.
The HMDA data also shows racial differences by loan product, making it difficult to draw general conclusions about lending patterns as a whole. However, the data does enable lenders to look more closely, by product type, location, borrower and delivery system to identify weaknesses in fair lending.
In the study, the authors acquired data outside of HMDA to control for borrower qualification characteristics. This additional data significantly reduced the racial and ethnic differences - but some differences remain and are at this point unexplained.
Denials vary by product type, with home purchases having the lowest denial rates. About 15% of the applications to purchase a site-built home are denied. Applications for loans to be secured by the borrower's current home are denied at more than twice that rate Approximately a third of applications for home improvement loans and refinancings are denied. This finding could be interpreted in several ways. One is that home purchasers are better prepared and have a more accurate idea of the value of the security property. Another interpretation would be to note that home improvement loans are usually a staple of neighborhood reinvestment. High denial rates could therefore raise concerns about CRA.
Loan pricing has been the long- expected (and feared) star of the 2004 HMDA reports. The 2004 data reports some pricing information but does not include most of the factors that are used to set prices. These factors could explain variations in the reported loans, but the information was not available to the authors of the study. Any explanations raised must be institution by institution. This means that you must know your lending pattern and your pricing factors.
Loan pricing factors include the cost of funds when the loan was made, credit risk presented by the applicant and the application, prepayment risk, overhead expenses, servicing costs, and discretionary pricing. In addition, variations in loan pricing may result from effective negotiation by the applicant or the route through which the applicant came to the lender.
Brokered applications are getting close attention under RESPA's Section 8. In addition, mortgage broker fees are finance charges that have the effect of increasing the APR. The questions raised by the authors of this study are likely to add a dimension to the mortgage broker issue. Who uses mortgage brokers and what brokers charge their customers now becomes a CRA and fair lending question.
One interesting finding is that the point spreads on higher-priced loans tended to be larger for both conventional home improvement loans and government-backed junior liens. This finding is likely to call attention to home improvement lending, including pricing, advertising, and product choices.
Geographic patterns are tied to the pricing findings. The study found that higher priced loans tended to be more concentrated in the southern region of the country. The study also found a close association between higher-priced loans and the proportion of individuals in an MSA county with low credit scores. The appropriate remedy in this situation might be customer financial education.
The data reported for 2004 included 24,594 loans subject to the special HOEPA rules because of pricing. Not surprisingly, HOEPA loans tended to be smaller than lower-priced loans. The average first-lien HOEPA loan was $98,650 compared to $173,125. Junior lien HOEPA loans showed similar differences. The interesting fact to emerge from the data was that HOEPA loans do not appear to be targeted to low- and moderate-income borrowers. About 75% of the HOEPA loans were extended to middle- and higher income borrowers. Most of the HOEPA borrowers were non-Hispanic whites.
This indicates that assumptions about the relationship between HOEPA loans and predatory lending may have been incorrect. In fact, high cost loans may not target the most vulnerable borrowers but instead target those with capacity to pay.
In the context of CRA and fair lending compliance, one of the interesting findings is that institutions that make high-cost loans tend to do so outside of their assessment area. The study found that pricing spread differences by group within the lender's assessment area were about one third of those for loans outside of the assessment area. The overall incidence of higher priced lending for all groups was lower within the lender's assessment area.
The study's authors hypothesize that this assessment area pricing difference could be driven by the availability of loan channels. This, of course, turns into a powerful argument in support of the bricks-and-mortar demands of community groups. The data appears to indicate that the impact of having a lending branch or office in the neighborhood is felt not only in available services but also in pricing.
There is another possible cause. Banks and thrifts know that their lending practices within their designated assessment areas will be closely scrutinized while lending outside of the assessment area gets less attention. No matter which interpretation you find more likely, both argue the importance of CRA.
HMDA data doesn't tell us everything. In fact, it doesn't include enough information to enable us to draw conclusions about whether and how discrimination occurs. Part of the HMDA debate has always been whether other information should be included. For 2004, the reports took the giant step of including pricing data and more detailed information about the loan transaction. However, the only information about the applicant is income. Thus, evaluations of applicant qualifications is not possible.
Applicant qualifications remains part of the great debate about the cause of differences in decisions by race. The study's authors therefore tapped a data source in addition to the HMDA reports in an attempt to determine the role of applicants' credit qualifications.
The additional data was supplied to Georgetown University's Credit Research Center by eight lenders that specialize in subprime lending. Additional data elements included credit history scores, loan-to-value ratios, appraised values, whether the loan was direct or broker originated, and several loan terms, including prepayment penalties.
The augmented data still showed differences based on race and ethnicity, but these differences were much smaller. The study first looked at the data base from the CRC using only information available through HMDA and then looked at the same data with the additional borrower information. Using information about borrower qualifications, the differences in decisions by race or ethnicity was reduced by a third.
The use of the CRC data provides some support for the claim that borrower qualifications account for the differences in decisions that are revealed by HMDA. However, the support only goes a short distance - one third of the distance, to be exact. The authors conclude that more applicant-specific and loan-specific information would be needed to support firm conclusions about racial or ethnic discrimination.
Ideas for the Future
One obvious change to see would be more HMDA data. However, the burden of collection and reporting would be massive. There are other steps the industry could take to improve the numbers. In fact, taking additional steps may be the only way to prevent the expansion of HMDA data.
Lending in lower income areas clearly needs more attention. Home improvement loans can be the life-blood of an older neighborhood. It would make good strategy to consider both pricing and availability of home improvement loans and junior liens.
In setting strategies to target credit needs in older and lower-income neighborhoods, lenders should give careful attention to pricing. The data do show pricing patterns, some of which make lenders vulnerable to fair lending challenges.
The lending patterns shown in the 2004 data appear to support the proposition that delivery channels appropriate to the borrower are essential to being an active lender to that group or location of borrowers. In this context, the bricks-and-mortar debate is alive and well. Consider whether branches are conveniently available to your entire assessment area. If not, other methods of efficient delivery of lower-priced products may be available.
Clearly, discrimination based on race and ethnicity must continue to get intense attention. The differences in lending results won't go away by themselves. In addition to offering the right products, in the right way, in the right place and at the right price, both lenders and borrowers need to know more. Education for both lenders and borrowers is probably a key factor in changing the picture. The more customers know about managing their finances and purchasing and maintaining a home, the smaller the gap should be.
Even though the study didn't identify gender discrimination, don't ignore the possibilities that it could occur. Any group can be given differential treatment before applications are taken without differences showing up in an analysis of decisions. This means that training for all customer contact staff remains important.
Finally, keep CRA and fair lending on the front burner. In almost every way that the HMDA data can be interpreted, CRA and fair lending are significant elements.
- Get a copy of the Federal Reserve Bulletin article. It is available at www.federalreserve.gov under publications.
- Study your own HMDA report and compare your numbers with the findings in the study. Understand or seek explanations for any differences.
- Take a quick look at originations and denials by race, ethnicity, income level and location. Look for any patterns that could indicate fair lending or CRA concerns.
- Take a careful look at all the home improvement products you offer. Make these the subject of your next fair lending analysis.
- Review application decisions and loan pricing to identify areas that would benefit from financial education such as Money Smart.
- Now evaluate your delivery channels. Consider how effective they are in reaching all potential borrowers and in delivering consistent prices to all borrowers.
- Compare HMDA reports with your CRA program elements and goals. Use HMDA to evaluate the success of lending strategies and to identify program elements that need adjustment.
Copyright © 2005 Compliance Action. Originally appeared in Compliance Action, Vol. 10, No. 12, 10/05
First published on 10/01/2005