In recent posts, we have focused on redlining as one of the more prominent fair lending risks that institutions are facing today. The agencies have clearly made redlining a top priority, and recent fair lending enforcement actions have confirmed this. There is, however, more to the fair lending risk management story.
The Many Facets of Fair Lending Risk, Examination, and Enforcement
While redlining is a significant risk for institutions, it is important to always bear in mind that fair lending risk management is multifaceted and involves a host of differing pressure points. An effective program, therefore, must be comprehensive, flexible, and adaptable, encompassing all possible risk areas. This also requires effective prioritization from a monitoring-resource perspective.
It is important to note as well that fair lending laws and regulations cover virtually every aspect of lending transactions.
This includes the level of treatment received from the moment the borrower makes contact with the institution all the way through the decision process, pricing of the loan, and servicing. Such a wide net is why effective risk prioritization is so important. Wise lenders will take steps to minimize these risks in critical stages of the transaction through policy and product structuring instead of relying on monitoring alone.
With regard to monitoring, however, there is one key feature of the fair lending landscape that is more and more becoming entrenched in both the regulatory examination and enforcement context, and that is the use of statistics and statistical analyses, including multivariate regression analyses.
In all recent enforcement actions, for example, statistical analysis is relied on in terms of both measuring consistency and treatment of target and control groups and, perhaps more critically, for quantifying and determining amounts of restitution that may be due borrowers.
Similarly, regression analyses are also becoming more prevalent in routine fair lending examinations and have been widely used by exam teams for larger institutions for a number of years. Institutions that have significant volumes of lending must have statistical analyses, including regression analysis, as a layer of their fair lending monitoring and risk mitigation efforts.
When Regression Should Be Used
Setting aside redlining concerns for the moment, what should be analyzed from a statistical perspective and, more specifically, a fair lending regression standpoint?
The “big ticket” items are loan pricing and loan underwriting, which would compare approval/denial rates and interest rates charged to borrowers between protected and non-protected groups. Such an analysis requires multivariate regression.
In order to accurately test differences in treatment, the factors that affect pricing and underwriting must be accounted for, or “held constant” so only differences in class status are truly measured. For example, if loan pricing is based on credit score, then we would expect pricing to vary based on the borrower’s score. It is then necessary to account for credit score to determine if protected and non-protected groups were treated differently.
Similarly, with underwriting, the same principle applies. We would expect the credit decisions to vary based on the criteria the institution uses in the credit decision. These need to be accounted for in order to fully assess fair lending implications of these decisions. Regression is a key tool to be used in this regard.
Why Regression Analysis is Important
From both the individual institution and the regulatory perspectives, regression analysis is an important, and often necessary, option for fair lending monitoring and evaluation.
First, large volumes of data can be analyzed. As opposed to the traditional side-by-side file review approach, in which only samples of loan applications can be reviewed, regression allows analysis of an entire group of loans.
Second, it allows for quantifiable measurement of differences in treatment between groups, i.e. disparities.
Third, and importantly, it allows for multiple factors to be accounted for simultaneously, thus controlling for the effects of these with respect to treatment of borrowers. For example, if score, loan amount, and term are all considered in loan pricing, and these data points are available to be used in the analysis, the regression equation will control for the effects of each factor and allow for the differences between groups to be tested independent of these influences.
Regression vs File Review
Statistical analysis and, more specifically, regression analysis mathematically accomplishes what would be done in a traditional file review but on a complete dataset: isolate the effects of class status on loan outcomes.
In the ideal file review, target and non-target group applicants with differing loan outcomes would be compared where they were identical on all loan attributes EXCEPT class status, therefore, isolating the effects of group status. If a target applicant was denied that was identical to a control group applicant that was approved with respect to relevant loan characteristics, that would suggest evidence of disparate treatment. In that case, other explanations have been ruled out since the key loan attributes were identical.
The regression equation accomplishes something similar except it estimates these treatments on average for all target and control group applicants in the sample. This allows testing of all applicants across a large sample while still isolating the effects of class status and, therefore, estimating differences in treatment.
As stated earlier, regression also allows for quantification of these effects. In other words, if one group on average is charged higher rates after controlling for loan pricing criteria, the size of these differences are further estimated. In the case of evidence of disparate treatment, or discrimination, any appropriate restitution can then be quantified.
Summary
Regression analysis can be a powerful tool and is now being more commonly used in fair lending examinations and inquiries. For financial institutions with large volumes of data, using regression as part of fair lending monitoring is a necessity.
It is important to note that these are complex methods subject to a litany of nuances with regard to analysis and interpretation. In order to have the full benefit, the application of statistical methods in general, and regression analysis in particular, should be conducted by knowledgeable and experienced practitioners.
If your institution wants to explore the application of statistical analysis in its fair lending risk management, we’ve been working with banks like you for almost 30 years. Reach out to us below to start the conversation.
