Fair Lending Regression Analysis: Some Frequently Asked Questions
Whether you use regression analysis in your fair lending monitoring program or you are thinking about using it, a few regression related questions are asked regularly. This blog will discuss four of the more popular questions.
-
What data do I need?
Depending on whether you are going to do a credit or pricing decision analysis, start with your underwriting guidelines or price sheets. They will contain key data elements. Aside from those underwriting and pricing elements, there may be some other factors that can be important, e.g. promotion data and regional or branch break outs. Because each institution is different, the useful data elements will vary by bank.
-
How many applications do I need to do a regression analysis?
The FFIEC Fair Lending Exam Procedures say that for credit decisions you need a total of 200 applications broken out as follows: 50 approved prohibited basis applicants, 50 denied prohibited basis applicant, 50 approved non-prohibited basis applicants and 50 denied non-prohibited basis applicants. A pricing decision analysis requires 100 applicants, 30 of which should be from the particular prohibited basis group you are analyzing and 30 applicants should be from the prohibited basis comparator group. The remaining 40 applicants can be from other prohibited basis and non-prohibited basis groups.
-
How many equations do I need to estimate?
A rule of thumb is that, for a pricing analysis, if there are different rate sheets, then you should estimate a separate pricing equation for each different rate sheet. Similarly, if there are structurally different underwriting guidelines then separate underwriting equations should be estimated. Of course, because this is a general rule, there are going to be exceptions. One exception is that different geographies may require separate equations even though the same factors are used in both geographies.
-
Should I use proxies for government monitoring information and, if so, which ones?
The answer to the first part of the question is Yes. Your regulator is going to use proxies, so you should also. After all, you want to understand what kind of results they will get if they analyze your data. The answer to the second part of the question is that, despite the faults associated with the various proxy methods, use the method your regulator uses.
Do you have other regression or statistically related questions that have been bothering you? If so, please call 847-295-6881 or email (rpreiss@preissco.com), and we would be glad to answer your question. We’ll also add it to the list of questions above for the benefit of all interested compliance officers.
