Fair Lending Gets Fancy

Steve Cross, Deputy Comptroller for Compliance Management, addressed the students attending ABA's National Graduate School of Compliance Management and discussed several developments in fair lending that compliance managers should be tracking.

Regression Analysis
The most significant new element in the examination procedures is a discussion of regression analysis. Logit regression analysis used statistical techniques to evaluate the extent to which many variables predict the outcome of one variable, the dependent variable. In fair lending examinations, a prohibited basis would be the dependent variable and the bank's underwriting standards would be the variables used in the analysis.

Examiners using regression analysis will begin by looking at the variables identified by the bank as important in its underwriting. The bank uses the variables because the bank believes or has found that the variables are important to predict the applicant's performance. The examiner's analysis adds a prohibited basis as a dependent variable to determine whether that factor, e.g., race, has predictive value as to outcome.

The OCC has used regression analysis side by side with file analysis in same institutions to make sure they get the technique right. Because the circumstances and market issues for each bank vary, there is no single analysis model. Examiners customize the models to the specific examination. Cross believes that this is a valuable improvement to how the OCC does fair landing analysis. What matters in this approach is whether a finding is statistically significant.

Cross notes that the OCC has done many fair lending exams using regression analysis with no finding of discrimination. Should there be a positive finding, the OCC would then study the analytical model to make sure that there were no flaws in the design. The starting question would be: does the analysis make sense?

Credit Scoring
The new OCC procedures discuss evaluation of credit scoring in the context of fair lending. Cross warns that credit scoring is neither a safety zone nor a panacea. First, there may be problems with data input. For example, if two similar applicants each earn $40,000 plus a $10,000 bonus, will different staff enter the income the same way? If data entries are not consistent, the credit score will only be as good as the model and data input to the model.

Second, there may be abuses in how the scoring system is used. Cross warns that if you use credit scoring in making underwriting decisions, use it properly. Overrides to the system may be appropriate, but banks must monitor overrides carefully. Loan officers are reluctant to let the "machine" make the decision for them and are tempted to change the decision. The review program should make sure that the override pattern does not occur on a prohibited basis. Also, the review should look for any patterns or disparities in override rates.

Finally, there is the potential for disparate impact with credit scoring. Any system is based on the data that was used to develop the system. In purchasing, developing, or using a system, the bank should look at what information went into system and how the system uses information. Look especially for information that is being questioned in the underwriting process as having a possible discriminatory impact.

Cross advises that the OCC is in the early stages of a research project on the potential for disparate impact in credit scoring systems. This is worth watching. Any studies performed by OCC, other regulatory agencies, or FNMA and FHLMC should provide the industry with valuable (and free) guidance on the use of credit scoring.

Matched Pair Testing
The regulatory agencies are seeing a rise in discrimination complaints against banks based on matched pair testing. Cross points out that discrimination testing should be based on certain principles to be sure that a finding is actually based on discrimination. First, there should be only one variable - such as gender, race, or age - that is a prohibited basis. Everything else in the test should be constant or slightly favor the minority tester.

Second, there must be enough tests to support a finding. Many of the tests used as the basis of discrimination complaints have involved only one to three matched pair tests at one lender. While this can provide sufficient information to instigate a complaint against the bank, it is not usually enough information for supporting a conclusion.

Finally, the tests should be controlled for location and loan officer. The tests OCC has seen didn't always control for the loan officer or branch of the bank being tested. Therefore, quality of tests coming in has not been very good. OCC has been working with HUD, the federal agency funding many of the testing exercises, to try to tighten oversight of projects to ensure quality results.

Matched pair testing, used properly, is still the best way to look at pre-application discrimination. There is a growing likelihood that you banks - especially in metropolitan areas - will be tested. Cross advises banks to make their front line staff aware of how to treat customers in a consistent and non-discriminatory way. He also warns that banks should be aware that there may be complaints based on scanty evidence.

Proposals
Cross also encouraged the students to comment on the FRB's announcement of proposed rulemaking on Regulation B. His priority list includes the issues of collecting monitoring data on all loans, distinguishing between inquiries and applications, the responsibilities of banks involved in indirect lending, documentation required for business credit, and coverage of preapplication marketing.

ACTION STEPS

Evaluate your methods of self-assessment for fair lending. Think about whether you are using techniques that are effective in your bank and your market. Then make sure you are using them correctly.
Read up on trends in matched pair testing. Also check HUD's website to find out whether HUD has awarded any testing grants in your market.
Find out whether your bank uses any form of credit scoring. First determine whether the system considers age and therefore must comply with Regulation B.
Look closely at how the credit scoring system is being used and monitored.
Evaluate the factors used to develop the system. Question whether any could be challenged for having a disparate impact.