The American Bankers Association has published a compendium of essays from prominent economists exploring the proper use of statistics in proving lending bias.
In Fair Lending Analysis the economists argue that most studies produce misleading results by relying on faulty statistical models.
In the foreword, Federal Reserve Board Governor Lawrence B. Lindsey explores the public policy impact that statistical studies can have. Excerpts of his remarks follow.
No issue is more important to the smooth functioning of any market economy than access to credit. Discrimination in mortgage lending not only tears at the fabric of our society, it also tears at the underpinnings of capitalism and the free market. In those instances where market participants discriminate on the basis of an irrelevant personal characteristic - race, gender, religion, marital status, etc. - appropriate intervention by government can increase both social fairness and economic efficiency.
In the past few years the government and banking regulatory agencies have greatly expanded their efforts to detect and eliminate bias in mortgage lending. Many of these efforts involve extensive reliance on statistics as the basis for fair-lending enforcement.
Simple answers are not possible because the issues involved are so complex. Caution should be used before jumping to any conclusions about the extent and nature of the problem.
Regulatory and enforcement agencies have their work cut out for them in the years ahead as they seek to develop appropriate investigative tools and procedures to detect discrimination and to craft appropriate and effective remedial actions where needed.
Research indicates that blatant discrimination in the credit market is very rare today. Clearly qualified applicants of all backgrounds are approved, and clearly unqualified applicants are rejected. However, econometric evidence suggests that discriminatory practices may occur in cases of marginally qualified applicants. How much discrimination exists in the credit markets continues to be hotly debated. How to eliminate it is equally problematic.
In this debate, however, we must not lose sight of two important truths. One, discrimination will ultimately be eliminated not by government agencies at the national level but by individuals working together within their own communities. And two, we must make sure that in the process of eradicating this problem, the cure is not as debilitating as the disease.
Of late, statistics have played a major role in our drive to eliminate discrimination. To carry the medical analogy a bit further, many people consider statistics to be the antidote to the poison that is discrimination. There is no doubt that statistics play an important role. But are they a cure? Not likely. For statistics come with their own problems - side effects if you will. Let me explain.
The use of statistics in lending enforcement exploded onto the national scene as the result of some high-profile cases where discrimination was detected through the use of a statistical technique called regression analysis.
Regression analysis is a mathematical tool used to measure the degree to which two or more phenomena are related, such as loan approval, income, and credit history. Using regression analysis, one can compute a probability that any individual applicant will have a loan approved.
The use of regression analysis as a tool for detecting lending discrimination has two parts. First, it can be used to assess whether the race of a large group of applicants is statistically related to the lender's decisions to approve or deny applications after controlling for various measures of creditworthiness.
Second, it can be used to compute the probability of approval for minority applicants using the treatment of nonminority applicants as a baseline.
I must admit to being somewhat troubled by the amount of faith the enforcement process is placing in interpretations of model results. A probability-based model like regression analysis cannot be used to say anything conclusive about an individual.
To date, we regulators have not been able to provide bank management with a list of practices, which if followed, guarantee that their institution is in compliance with the law. Indeed, with statistically based enforcement, such a list will always be impossible to generate. Statistical modeling, by definition, means that each loan decision will always be judged on a relative basis, not on any absolute standard involving the behavior of a particular loan officer.
We should also be concerned that the climate created by improper statistical analysis may have had the unintended consequence of limiting the expansion of credit. Consider, for example, the widespread reporting of the ratio of rejection rates for different demographic groups. Lenders are now experimenting with easing various underwriting standards as a means of expanding lending to historically under-served groups. Such experimentation entails risks that may be mitigated by more careful screening of marginal applicants. It would be unfortunate if such experimentation was discouraged in the name of "fair lending" by associating fair-lending compliance with low rates of applicant rejection.
Under current policy conditions, I would expect credit-scoring-type procedures to be dominant by the end of the decade.
Over the longer term, I am very concerned that the use of statistics in enforcement will ultimately lead to statistically based loan decisions. In fact, such statistics-based appraisal systems already exist. These credit- scoring models will continue to gain broader use as regulators seek ever- more-sophisticated statistical means to detect discrimination.
The concept that I am trying to convey is that we humans also have a sense of justice that transcends the statistical sense of fair treatment provided by the computer. That is why I am so troubled by the policy dilemma our country now faces in the area of lending discrimination.
As is so often the case, it may be that cleaning up the unintended consequences of well-intentioned policy actions taken today will be the biggest challenge for tomorrow's policymakers.