Surviving the Statistical Shift in Fair Lending Examinations

In the modern banking regulatory environment, fair lending examinations have taken an increasingly data driven approach. For compliance officers and bank executives of large institutions, the days of manual file reviews have largely been replaced by high-stakes mathematics. Fair lending exams have become statistical in nature, and understanding this shift is no longer optional—it is a survival skill.

When Things Heat Up

In some cases, an examination may move beyond the initial review to a secondary and elevated phase in which the examination team believes it has encountered a pattern or practice of discrimination. At this juncture, an institution may be facing a DOJ referral, and a robust defense is absolutely critical. This phase is more often than not statistical in nature.

In such an examination, regulators use multivariate regression analysis to apply scientific methods to fair lending, moving beyond simple observation to establish potential causal links between a prohibited basis (such as race) and lending outcomes.

Here is a breakdown of how regulators employ this methodology:

1. Testing for Correlation and Controlling Variables: The fundamental question regulators ask is whether a fair lending pressure point, such as pricing or underwriting (approvals/denials), is correlated with a "prohibited basis". However, they recognize that a simple relationship between two variables does not prove causation, as other background factors are likely involved.

Therefore, regulators use multivariate regression to "control" for variance related to legitimate business factors. This method provides a mathematical way to simulate a "perfect world" comparison, where an analyst would compare loans that have EXACT characteristics on ALL factors with the exception of prohibited base designation.

2. Quantifying Disparities: A typical scenario would involve a communication such as a “15 Day Letter from the FDIC” or “Preliminary Findings Letter” in the case of the Fed which would present and quantify the findings.

• Dependent Variable: The regulators measure a specific outcome variable, such as the interest rates charged on loans or loan denial.
• Target vs. Control: They compare a target group (e.g., Black borrowers) against a control group (e.g., non-Hispanic white borrowers).
• Calculating the Difference: The staff economist calculates a specific disparity after controlling for legitimate pricing factors. This step "winnows out" the variance in rates that can be explained by legitimate criteria.

3. Establishing Statistical Significance: It is not enough to simply find a disparity; regulators use probability theory to determine if the finding constitutes "proof".

• Ruling Out Chance: They test for "statistical significance," which determines if the measured differences are large enough to rule out the possibility that the result is just random variation.
• The Threshold: For example, if a disparity is significant at the "1% level," it means the probability that the difference between the target and control group in the outcome variable is the result of random variation in the dataset is less than 0.01.

If the regulators conclude through these methods that there is a pattern or practice of discrimination, they are required to refer the matter to the U.S. Department of Justice.

The Proper Response

A bank facing a statistical fair lending inquiry—such as a "15 Day Letter" or even an initial data request for each an exam—should treat the situation as a high-stakes event with potential legal consequences. Since a finding of discrimination requires a referral to the U.S. Department of Justice, the sources recommend a coherent strategy focused on rigorous data analysis.

1. Expert Assistance & Resources: The sooner experts are retained, including legal counsel, the more quickly the bank can move toward a resolution and enhance the chances of a favorable outcome. Bank management should expect to devote significant resources to navigate this successfully. Because the outcome of these exams can lead to a mandatory referral to the Department of Justice, taking short-cuts is high risk.
2. Conduct Independent Analysis: A bank should never wait for the regulators to reveal the results of the statistical analysis. It is critical that the bank knows what its data may reveal. When conducting this internal analysis, the bank should be wary of relying solely on "canned" software packages. Reliable modeling requires considerable time, effort, knowledge, and expert judgement to accurately assess the bank's lending environment and data.
3. Look Beyond the Examination: The bank should seek to identify the sources of concern for examiners and formulate strategies to mitigate these risks. If the inquiry highlights issues, or to prevent future inquiries, the bank should use the same statistical methods used by regulators to improve its own processes. This includes:

• Developing mechanisms to collect and maintain data.
• Using data insights to refine policies and lending criteria.
• Continuing to test and refine outcomes to lower fair lending risk on an ongoing basis.
• Have a written and quantified plan and be poised to implement immediately

4. Estimate Potential Impact & Remediation: One of the reasons examinations utilize statistics is that impacts of practices on protected groups can be quantified. This should be a critical part of this process and management should be prepared to propose restitution if there is difficulty in demonstrating that practices were non-discriminatory in nature.

Conclusion

The world of statistics can be intimidating, filled with jargon and complex theories, and fraught with risk. However, to avoid issues, banks must treat fair lending compliance as an ongoing management cycle, not a one-time exam event.

The implication is simple: you cannot afford to wait for an exam to find out if your data suggests discrimination. By adopting a proactive, data-driven approach, you can identify and correct disparities before they become a federal case.