For the last decade the regulatory and enforcement agencies have been increasingly using statistical methods such as regression to evaluate fair lending compliance.
With the passage of Dodd-Frank and the new emphasis on modeling and quantification, there has been a fervor to apply econometric techniques to a wide array of issues in the financial industry. This includes not only fair lending but stress-testing for large institutions and the new requirements that will be soon implemented under CECL (Current Expected Credit Losses).
In the industry’s eagerness to embrace more sophisticated methods, the terms “model” and “modeling” have been broadened to encompass nearly any process that includes numbers. Many software applications, including popular ones such as Excel, can perform any number of statistical techniques on data with a mouse click, including regression analysis.
There exists a significant and growing gulf, however, between the appropriate use of regression analysis and how it often is applied. This space is not exempt from challenges in this regard. There is often a great deal of misunderstanding in terms of the appropriate use of regression analysis.
We plan to expand on this topic further in a forthcoming white paper, but for now we will just dispel a few myths:
- Regression is magic – This is a little “tongue-in-cheek”, but it often seems as if it is thought of this way. Most practitioners involved in fair lending understand the basic concept of regression and know some of the jargon. It is many times falsely believed that a regression analysis can be performed on any data with no real thought about the assumptions and necessary conditions that must be met to produce a valid analysis. Instead they believe data can just be read in, hit a button, and out comes the answer. The reality is much more complicated.
- Sample selection does not matter – What seems always lost is that statistical analysis is only used to draw a conclusion about a larger population. If this were not the case, there would be no need for the use of statistics. The sample, then, is absolutely critical in order to have a reliable answer to the questions being asked.
- Because the model “controls for variation”, segmentation of the sample is unimportant – We have addressed this in part in a white paper and in previous post. It is essential and part of the modeling process to understand and address accordingly the composition of the data to be analyzed.
- The results and conclusions drawn are only affected by factors included in the model – This assertion could not be more wrong. A fundamental assumption of the regression model is that it has been properly specified and contains all relevant variables. Since a model is just a simplified version of reality, more often than not this just does not hold true. The results, therefore, may be more affected by what has not been included than what has been included. It is surprising how often this fact is simply ignored.
- Regression analysis is always an appropriate tool for fair lending analysis – Regression analysis is a powerful tool and very effective for fair lending analyses. However, it is not appropriate in all situations, and there are times when a more reliable analysis can be conducted by other means. In fact, in some cases a regression analysis could produce biased results.
Employing the use of quantitative methods to problems or research questions suggests a level of sophistication that often adds credibility; and it should, because these are scientific methods. However, scientific methods applied unscientifically do not result in sound conclusions and instead may reinforce erroneous conclusions.
How to cite this blog post (APA Style):
Premier Insights. (2017, December 21). Proper Application Of Regression For Fair Lending Analysis [Blog Post]. Retrieved from https://www.premierinsights.com/blog/proper-application-of-regression-for-fair-lending-analysis