Understanding the Limitations of Fair Lending Regression Analysis

»  Understanding the Limitations of Fair Lending Regression Analysis

As we have discussed in previous posts, fair lending regression analysis has become widely used for evaluation of lender practices.

While a comprehensive discussion on the subject is unnecessary certainly beyond a blog post, it is important to understand limitations when such analyses are applied to such a critical issue as fair lending.

Regression is used by regulatory and enforcement agencies as well as lending institutions. Consequently, upper-level and executive management of lending institutions often find themselves reviewing results of regression and other statistical methods applied to their data.

In the last few decades advances in technology have produced easy accessibility to software applications that can readily produce various forms of regression as well as other statistical calculations. However, proper application of statistical methods and drawing conclusions from such is more complex than just pushing a button.

The Expansion of Econometrics

Since the marginalist revolution in the late 19th century which birthed the application of mathematical methods to economic issues, the field of econometrics has expanded dramatically. As is the case with all advances and modernization, the discipline has become highly specialized.

There are many different fields of econometrics that have their own range of individual techniques. This is largely due to the data challenges that are often unique to each field of analysis. Fair lending is no different and has become a specialization in and of itself.

When Scientific Methods are Applied Unscientifically

What is often the case, however, is a very “loose” application of regression analysis. Statistical techniques, including econometrics are sophisticated methods but rest on a fundamental set of assumptions. These assumptions are both theoretical as well as mathematical, and the conclusions drawn are only as good as the validity of the assumptions. Although this is fundamental, this is more often than not overlooked. The result is scientific methods being applied unscientifically.

The more troubling aspect here is that the statistical component creates an illusion of sophistication that conveys validity that may not exist because the fundamental component of the analysis was flawed. The best example of this in recent memory is the financial crisis which was due almost entirely to flawed assumptions regarding risk mitigation from securitization of debt.

Researchers and scientists that use such methods are keenly aware of the limitations and are extremely cautious in drawing conclusions absent of careful qualification. This is, unfortunately, often lost in the world of fair lending. We have highlighted some of these issues in previous posts and plan to expand on these in the coming months.