Rebutting allegations of unfair lending: here's a basic primer on how to build a powerful defense against charges of unfair lending using origination data and statistics.
An example of a price discrimination allegation would involve the Federal Deposit Insurance Corporation (FDIC) maintaining that, in 2009, a bank charged higher interest rates to black borrowers as compared with white borrowers who were refinancing second-lien loans on owner-occupied one-to-four-family dwellings.
In a second example, the Office of Thrift Supervision (OTS) might point to a savings association's first-half 2010 pattern of Hispanics paying higher rates than non-Hispanic whites for first-lien purchase loans for manufactured housing.
Regulators often focus on narrowly defined products that reflect the effect of screens that most types of loans passed. But if the regulatory agency refers a matter to the Justice Department, DO} can expand the scope of investigation beyond the area of initial concern.
Interest rates examined may be the loan's basic note rate, its annual percentage rate (APR) or an overage measure of the difference between the rate of interest listed on a rate sheet and the note rate or APR charged. Because HMDA databases do not include any interest-rate information for some borrowers, regulatory agencies' conclusions would have to be based on additional data submitted by lenders. Along with interest rates, lenders commonly supplement HMDA information on loan amount, applicant income, product details and race/ethnicity designations with data on other determinants of interest rates such as credit score and history, debt profile, loan term and loan-to-value (LTV) ratio.
Regression analysis is the computational tool used to measure racial (or ethnic) mortgage rate disparities while taking into account other factors. The regression equation's measured race effect indicates how much more, on average, one group pays than another, holding constant (i.e., net of the impact of) all the other included regression variables.
For example, a regression limited to a bank's 30-year fixed-rate first-lien loans on owner-occupied one-to-four-family properties might reveal that after taking into account each borrower's credit score, loan amount, LTV and whether the loan financed a new purchase or a refi, being black rather than white is associated with an APR that is 23 basis points higher. Accompanying a regression's basis-point difference for two groups is the disparity's level of statistical significance.
When rate sheets determine mortgage rates with absolutely no discretion allowed, a regression equation will fit a database perfectly and the race effect will equal zero.
Even with some discretion for individual loans, measured race effects are often insignificant when factors such as credit scores and loan-to-value ratios arc major determinants of mortgage rates. It the disparity in a credible regression equation does not reach statistical significance, agencies generally will not pursue the matter further.
Sometimes, particularly with a large number of loans, regulators may consider statistically significant but very small basis-point differences not worth further investigation. How small is small has not been clearly delineated, but it is safe to state that, for example, a black-white difference of a single basis point would be regarded as trivial even if highly statistically significant.
Because individual loan officer discretion commonly affects mortgage rates and because all information reflecting legitimate determinants of mortgage rates is rarely available for analysis, regression is not. a panacea and explanatory variables may only narrow, but not reduce, calculated racial pricing disparities to insignificance.
In the hope of producing statistical defenses that rebut unfair lending allegations, lenders should collect and retain more rather than less data on borrowers and their loans. Moreover, statistical modeling should reflect the underlying lending process. For example, if mortgage banker ABC is concerned only whether credit scores are 1) below 600; 2) between 600 and 700; or 3) higher than 700; while banker XYZ considers every single point, the credit-score variable should be strikingly different in regressions for the two lenders.
Among the rarely used but potentially powerful variables that could explain interest rates are borrowers' employment profiles. A loan to a self-employed individual or one with frequent unemployment spells is generally riskier than a loan to an otherwise similar borrower in a secure salaried job, and the additional risk would be reflected in a higher mortgage rate. Thus, including employment data in a regression would be expected to increase the explanatory power of the regression and to narrow measured racial disparities.
Loan characteristics can produce similar regression effects. An example would be the presence and extent of prepayment penalties, associated with lower interest rates than an otherwise similar mortgage without the penalties.
Other potential explanatory variables, such as the condition of the mortgage market, are not directly associated with individual borrowers or their loans. Even within a quarter of a calendar year, the mortgage market can tighten and loosen. Indicator variables for the month in which each loan was locked reflect this variation to some extent. More precise measures of prevailing mortgage rates are collected weekly in the Freddie Mac Primary Mortgage Market Survey, publicly available data.
Although the inclusion of additional variables will tend to increase the explanatory power of a regression equation and, more often than not, narrow racial disparities, not all possible variables should be considered.
For example, along with a variable indicating whether a borrower is black or white, a second race variable could be added measuring the percent of the population that is African American in the Census tract in which each borrower resides. Given residential segregation, the race of borrowers will be highly correlated with the racial characteristics of their Census tracts, and including both variables will blur the effect of each and improperly create a misleading impression that race is less correlated with interest rates than is in fact the case.
Other variables, rarely used, could save the day for mortgage bankers. The first is loan points.
If a white borrower pays 2 points at settlement to receive a 5.125 percent 30-year mortgage, while a similar black borrower pays no points for a 5.5 percent mortgage, a simple comparison of the rates will suggest that the African-American borrower is being overcharged by 0.375 percent.
But if points data were readily available, a banker could determine the zero-points mortgage option offered to each borrower. Zero-point note rates--or APRs calculated based on these note rates--would provide a fairer comparison than the actual mortgage rates charged. Lenders are responsible only for the array of mortgages they offer, not for borrowers' choices from the menu.
In addition, some loan applications include borrower information--such as education, occupation and investments held--that suggests financial sophistication. Educated people tend to be better loan shoppers and more financially aware.
Certain occupations, such as commission-based sales jobs, are associated with relatively high negotiating skills. And borrowers who've picked individual stocks are likely to be savvier than those who invest only in mutual funds or certificates of deposit.
Because minorities tend to have lower education levels and occupational status than non-Hispanic whites, incorporating education and occupation into a regression equation may explain away some of the observed white-black and white-Hispanic disparities that regulators interpret as discrimination.
Types of investments held by individuals of different backgrounds may have effects similar to those of education and occupation. A common model of lending discrimination presumes not that lenders exploit an animus against minority borrowers, but rather that they can negotiate with more success against minority borrowers and do so for the purely pecuniary motives of increasing commissions and enhancing their sales record.
The list of variables that may explain why mortgage rates vary across borrowers discussed in this article is suggestive rather than exhaustive, and can be augmented following discussions with loan officers.
When a regression reveals significantly higher interest rates paid by African Americans or Hispanics, it may be worthwhile to identify which minority borrowers were most disadvantaged and examine what characteristics they have in common. Perhaps this investigation will lead to another regression variable, previously overlooked.
Alternatively, these borrowers may predominantly have been served by one loan officer whose decisions are responsible for a lender's unfortunate overall interest-rate pattern. Even if no commonality is apparent among these borrowers, the argument that removing a handful of minorities from the database eliminates or sharply reduces calculated disparities may favorably influence regulators. For example, suppose a bank's statistical profile reveals that African Americans paid significantly higher interest rates than non-Hispanic whites. Suppose further that removing the 10 percent of black borrowers who overpaid the most leaves a remaining pool of borrowers with no racial disparity. This pattern supports an argument that damages could be limited to the 10 percent group.
The basic statistical approach to pricing allegations outlined here--with regressions as the mainstay analytical tool--also holds for broker fees, steering and loan denials, although there are important conceptual distinctions. Technical approaches, beyond the scope of this article, also vary across--and for that matter, within--issues investigated.
Broker fees, steering and loan denial
Regulators have alleged unfair lending when broker fees for wholesale loans originated by a lender are significantly higher for blacks or Hispanics than for non-Hispanic whites. Although these fees can be expressed in dollars, they have more often been measured as a percent of loan amount, commensurate with interest rates.
The legal theory supporting allegations of broker fee discrimination presumes that by accepting or rejecting borrowers whose broker fees may contribute to disparate patterns, mortgage bankers are responsible for the fees charged by third-party brokers. Minority borrowers can be charged statistically higher fees than non-Hispanic whites even when no single broker discriminates, as in the hypothetical example where brokers with only white clients charge low fees, brokers serving only blacks high fees and brokers with a biracial clientele an intermediate fee that is the same for all whites and blacks.
Even under this legal theory, which presumably eventually will be seriously challenged by attorneys defending mortgage bankers, statistical defenses can be employed.
First of all, the components of broker fees--typically yield-spread premiums and direct fees--can be analyzed separately, with the possible result that only one component is responsible for any disparities. Attention can then be limited to that element.
Second, the explanatory variables that affect pricing are not necessarily those that affect broker fees. In particular, fees may depend more on the amount of time brokers had to spend with or on behalf of their clients than on the borrower characteristics that fit pricing regressions.
When neither broker practice nor available data furnishes potential variables to explain broker fee variation, regulators may focus on disparities in raw broker fees, uncorrected for any variables except loan amount, implicit in the percent measure.
Another area of concern has been disproportionate steering of minority borrowers into subprime loans. Steering analysis is plagued by the difficulty of distinguishing consumer choice from heavy lender influence over--or even outright restriction of--available loan products. Nonetheless, regulators often attribute strikingly divergent demographic patterns of prime versus subprime mortgages to lender steering.
Because the sharp increase in subprime loans in 2005-2008 has been followed by a severe decline, steering cases are likely to be relatively uncommon in the next couple of years, given statutes of limitation in fair lending laws.
Lender and borrower decisions are clearer when investigating loan denial because of the HMDA category of loans offered but not accepted. Factors explaining prime versus subprime borrowing, or loan acceptance versus denial, should be similar to those explaining loan pricing.
Most obviously, credit score alone sometimes delimits prime and subprime loans or determines loan eligibility. The fact that information on rejected applicants will almost always be less complete than for approved borrowers will inevitably constrict loan denial investigations. In any event, because of lender accommodation, loan denials have been rare in the last few years. Perhaps in the near future, loans will be denied to applicants who formerly would have borrowed subprime mortgages, and loan denial allegations will become more common than claims of disparate steering.
Presentation is important for all statistical studies, but especially in cases of steering and loan denial that focus on differential incidence rather than basis points.
When 80 percent of black borrowers and 90 percent of white borrowers are offered loans, bankers prefer to emphasize the similarity of acceptance rates. On the other hand, regulators might highlight that 20 percent of blacks but only 10 percent of whites were denied loans and conclude that the denial rate for blacks was twice that for whites.
Fortunately, proper statistical methods--in this instance focusing on the 10 percent simple difference in both acceptance and denial rates--are immune to this sort of manipulation.
Redlining analysis differs markedly from the methods discussed here. It is based on Census and HMDA data, uses individual loans only to compose aggregates and does not involve regression analysis.
The basic redlining comparison considers the proportion of a bank's loans in minority areas relative to those of comparable lenders. Possible definitions of minority areas, geographic regions and comparable lenders and loans will suggest several redlining yardsticks.
When regulators present a study that purports to demonstrate a target bank's sub-par lending activity in minority neighborhoods, it will often be possible to construct an equally plausible comparison with the opposite conclusion. The inherent imprecision in this exercise is enhanced when, as now, the Census is well out of date. Until the 2010 Census tabulations are fully available in 2013, redlining studies will be based on Census data more than a decade old.
Consider an allegation that a small bank originates too few mortgages to nearby potential black borrowers. The redlining analysis might begin by determining the Census tracts whose populations in 2000 were majority black--that is, 50 percent or more African American in the metropolitan statistical area (MSA) in which the bank does all its business. Suppose HMDA databases reveal that 10 percent of the subject bank's 2009 mortgages, but 25 percent of all other 2009 mortgages originated in that MSA, were lent to borrowers (black or not) residing in the MSA's majority-black Census tracts, and that this difference is statistically significant. A regulator could find this result strong evidence of redlining.
But all the assumptions supporting the redlining conclusion can be challenged to produce alternative benchmarks.
The African-American proportions in the Census tracts could be based on blacks and non-Hispanic whites only or on all residents. Fifty percent is not the only meaningful measure of substantial representation of African Americans, who comprise less than 15 percent of the national population with substantial variation across regions. A bank may do all its business within an MSA, but none or hardly any in the most outlying counties, which could therefore be excluded from the calculations. For that matter, the target bank's explicitly defined market or Community Reinvestment Act (CRA) assessment area could be used, so long as these were not gerrymandered to avoid minority areas.
Furthermore, the bank under scrutiny could be compared against lenders of similar size or with comparable loan volume rather than with all institutions originating loans in the same area. Less-common products such as loans on manufactured housing or to investors could be included or excluded. Offers of loans not taken by applicants could be added to the mix.
If the target bank appears to have originated too few loans in African-American Census tracts under a variety of assumptions, then a statistically based redlining allegation would be well founded. But if the conclusion is very sensitive to alternative assumptions, a redlining allegation should be easy to counter.
The keys to rebutting unfair lending allegations are maintenance and use of rich databases, proper statistical methods that reflect bankers' lending practices, and consideration of reasonable alternative assumptions. The same tools can be applied proactively in internal monitoring processes that seek to identify disparities regulators deem evidence of discrimination. With sufficient lead Lime, bankers can alter policies and practices to narrow these disparities and ensure that eventual data submissions will present patterns of fair lending.
Farrell Bloch is an economist and statistician in private practice in Washington, D.C. He can be reached at email@example.com.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Fair Lending|
|Comment:||Rebutting allegations of unfair lending: here's a basic primer on how to build a powerful defense against charges of unfair lending using origination data and statistics.(Fair Lending)|
|Date:||Aug 1, 2010|
|Previous Article:||From properties to pets.|
|Next Article:||Why mortgage fraud still exists: mortgage fraud was on the rise in 2009, according to LexisNexis Mortgage Asset Research Institute. The latest...|