Ever wonder how one can conduct a fair lending review of your bank’s consumer lending products without having government monitoring information like race, gender, and ethnicity available? Enter the proxy.
Why Use Proxies?
In the world of fair lending, HMDA reportable applications have traditionally been the focus of regulatory reviews. And for good reason – government monitoring information (GMI) including race, gender, and ethnicity is collected and readily available as part of the HMDA data. This allows easy classification of applications for any type of qualitative or quantitative fair lending review.
In the last several years, however, consumer loans of different types have become the subject of such reviews resulting in an increasing number of regulatory enforcement actions. Since financial institutions are prohibited from capturing this type of information on consumer loans, how can a fair lending review of consumer products be conducted?
The answer is through the use of a “proxy” by which to classify the target and control groups. We will explain the different methodologies used to do this in a future series of posts, including some problems with the application of these methodologies. Here we focus on a simple explanation.
So What is Proxy?
In a typical consumer loan dataset, race, ethnicity, and gender of the applicant is not identifiable; therefore, these attributes are unobservable. By applying a proxy methodology, an indicator can be developed that takes the place of the actual attribute that cannot be identified.
For example, although race and gender is unknown, if a factor can be found that is significantly correlated with the unknown attribute we can come up with an estimate of the applicant’s race, ethnicity, or gender. This can then be used instead of the actual attribute.
Simply, a proxy is a variable that is correlated with an attribute that is unobservable and is used as an indicator….
The diagram below will help illustrate. In this example, we are interested in examining differences in treatment of loan applicants by race. Since race is unknown, we identify a variable that is closely correlated with race. That variable then gives us an indication of the race of each applicant.
How Should a Proxy Be Used?
Because the proxy that is used in place of race only provides an estimate of the person’s race, the application of proxies is best suited for statistical analysis with large datasets.
Once the proxy is determined it can then be used in a regression analysis to test differences in treatment between a target and control group. Because the proxy is not 100% accurate, that is, an indicator of race for example will not always correctly predict the applicant’s race, it is not ideal to use proxy data for transactional reviews unless the samples are very large.
Thus, the use of proxies is more suited for a quantitative fair lending review.
