Regardless of the regulator, Community Reinvestment Act and fair-lending examinations are growing in sophistication. When these exams started, denial rates were the focal point.

We hear less and less about denial rates as the regulators have become more advanced in their thinking about discrimination testing. At this point, prudent financial institutions can no longer wait for the examiners to ask questions based on the results of their pre-exam and inbank performance assessments. They must anticipate regulatory questions by doing these analyses before the regulators arrive.

Regulators prepare several types of reports prior to entering an institution in an effort to focus their exams. The newest of these procedures uses logit regressions. Thus, regulators have moved beyond denial rates as a test for discrimination to a specialized form of multiple regression.

This is the same kind of regression analysis that was used in the Boston Federal Reserve study and in the Decatur Federal investigation.

Initially, logit testing was only used by the Department of Justice when it decided to do an in-depth investigation.

In June, the Federal Reserve announced logit analysis would be a regular part of its fair-lending examination procedures, particularly at larger institutions.

In addition to being a more rigorous and thoughtful way to test for discrimination on the basis of race, sex, and marital status than prior methods, the logit process is also more cost-efficient because it first identifies if there is the likelihood of discrimination before spending additional resources on a full-blown investigation. In problematic situations it then highlights the potentially troublesome loan files.

In view of the advantages of a better and cheaper process, other regulators are likely to adopt similar changes in their procedures.

Many financial institutions do not understand logit analysis and how it can be used by the regulators in discrimination testing. Yet, it can be of significant benefit to bankers to see the results of these tests and review the relevant loan application files before the regulators come knocking.

Logit analysis is a specialized form of multiple regression analysis. The name "logit" comes from the 1ogistic function, a math formula that describes a curve resembling an elongated S. The output from a logit equation can be interpreted as the probability that an applicant with a given set of financial (income, monthly expenses, loan value, etc.) and nonfinancial (employment and credit history, etc.) characteristics will be approved for a home mortgage.

The logit testing process used by regulators can be broken into two parts. The Level I logit testing uses publicly available HMDA data that financial institutions submit each year.

Armed with this information, prior to entering your institution, the regulators will estimate a logit regression. If the race, sex, or marital status variables in this equation are shown to be important to the accuracy of the equation, the regulator has reason to believe that there may be a pattern or practice of discrimination occurring in the institution.

Using this finding but recognizing that he does not have all the important loan decision variables in his regression model, the regulator will enter your institution and ask for a sample of loan files from which additional data will be collected. The additional data may include the following kinds of information about an applicant: the ratio of mortgage payment to income, dollar value of liquid assets, education attained, ratio of total monthly debt to monthly income, number of years on current job, credit history, and selfemployment.

Utilizing this enhanced data base the regulator proceeds to Level II testing. A new' logit regression will be estimated over the expanded database. If any of the discrimination variables are still statistically significant, loan application files will be pulled for intensive review.

Which files are likely to be reviewed can be determined from the output of the logit model since the software allows applications to be broken into two groups.

The first group is those applications approved by the bank but whose model score indicates they should have been rejected. Some bankers call these loans "brag loans." That may be true, but the model is warning that these loans entail more than the normal credit risk and perhaps should be monitored more closely.

The second group of applications the model identifies are those that were denied by the institution but had model scores that were high enough to warrant their being approved. These applications will be of particular interest to the regulators looking for discrimination. In addition, they should be of particular interest to the bank because they represent foregone business.

In one actual case, the number of these applications amounted to 10% of the total number of loans. Thus, the institution had passed up an incremental 10% more mortgage business.

The regulators are most likely going to want to look at this second group of application files. There may well be or should be documentation in these application files that fall outside the purview of the logit regression to justify the bank's actions.

If there is not appropriate justification in the files, bank management should find out if it is an isolated instance or part of a larger pattern. Reviewing these files for a commonality such as coming from a particular branch or from a particular loan officer should tell the institution whether or not the discrimination is an isolated instance.

In the case of the denied loans being part of a larger pattern, the bank ought to develop and implement corrective action plans. Regulators have indicated a proactive stance on the part of the institution is much better than ignoring the existence of potential problems and letting the examiners discover your problems.

Today's CRA and fair-lending regulatory environment is getting tougher. Counterbalancing that trend is the fact that we know more about the testing the regulators are doing.

Sensible management will take advantage of this knowledge and get prepared before the regulators arrive.

The process is relatively simple and can be done internally or by an outside consultant.' Such actions have large paybacks in terms of reduced potential regulatory fines and diminished softer costs such as image, management and personnel time, and foregone profits associated with any possible investigation.

Testing sophistication by the regulators is growing, but so is the ability of the financial community to anticipate the new tests. Furthermore, there may be a silver lining in that the tests done for regulatory purposes can also be used to reduce costs and generate incremental business.

For financial institutions, the incentives are on the side of taking action now. Logit testing can help keep regulatory surprises to a minimum while reducing the cost of certain aspects of compliance.

It can do this because it highlights the loan files that are likely to need attention. In addition, the information generated in the compliance process can be used elsewhere in the bank to enhance underwriter productivity, improve loan quality, and generate additional revenue.

Subscribe Now

Access to authoritative analysis and perspective and our data-driven report series.

14-Day Free Trial

No credit card required. Complete access to articles, breaking news and industry data.