Fair Lending Studies Paint Incomplete Picture

There are three principal types of claims of discrimination in mortgage lending.  The first involves discriminatory rejection of mortgage applications, an issue receiving much scrutiny in the 1990s.  The second and third involve discriminatory assignment of loans to subprime rather than prime status and discriminatory mortgage terms apart from assignment to subprime status.

The latter two types of claims have predominated in more recent years and were subjects of the Department of Justice's recent $335 million and $175 million settlements of suits against Bank of America's acquired Countrywide Financial Corp. and Wells Fargo Bank, respectively.

I have previously discussed a statistical analysis problem that pertains to all three types of claims: specifically, efforts to adjust for racial differences in credit-related characteristics rarely, if ever, do so adequately.

I have also discussed an anomaly pertaining only to the first two types of claims.  In particular, for close to two decades, regulators have been pressuring or encouraging lenders to relax lending criteria and otherwise reduce the frequency of adverse lending outcomes. Regulators, however, remain unaware that reducing the frequency of adverse outcomes tends to increase relative differences in adverse outcome rates and continue to monitor lender practices on the basis of relative differences in adverse outcomes.  Thus, by responding to regulator pressures, lenders increase the chances that the government will sue them for discrimination. 

There is also a problem with fair lending analyses that pertains only to the latter two types of claims. Other issues  aside, analyses of differences in mortgage rejection rates at least examined the entire universe of people seeking a desired outcome.  Analyses of claims involving loan terms, including assignment to subprime, do not meet this criterion. Rather, analyses I have seen only consider borrowers who were offered a loan package they were willing to accept.  Applicants denied a loan and applicants who turned down loans are not considered at all.

Such analyses are therefore fundamentally unsound.  Moreover, they are unsound in a way that would tend systematically to find a disparity adverse to minorities irrespective of whether a lender has discriminated.  There are two aspects to the matter.

First, statisticians recognize that a group who on average has weaker measurable credit-related characteristics tends also to have weaker credit-related characteristics that cannot be measured or at least cannot be measured by a loan officer. This fact does not necessarily undermine an analysis where the lender is the only decision maker (as where the issue is loan approval or rejection), since the lender is required to act on the objective facts brought to its attention.  Indeed, the possibility that lenders would assume that, on average, minority borrowers have somewhat weaker credit-related characteristics than white borrowers who look the same on paper is one of the things that might make lending discrimination claims plausible.  The law prohibits a lender from drawing inferences on the basis of race even where the inferences have a sound statistical basis.

But it is another matter when the borrower is also a decision maker, as when he can refuse to accept a loan under terms he deems unsatisfactory.  In determining whether a particular offer is likely to be the best he can get, the borrower will take into account the additional information he has about his credit-related characteristics that the loan officer does not.  Such information may cause borrowers from groups with generally weaker credit-related characteristics to accept less favorable terms than borrowers of like paper qualification from groups with generally stronger characteristics.