BISG and Other Proxies: What Is BISG and Why Proxies Matter to All Banks?
BISG, which is the CFPBs newest method of determining race and ethnicity proxies, is the acronym for Bayesian Improved Surname Geocoding. Like many other analyses in fair lending, BISG is borrowed from another research area. In this case BISG, in its latest version, comes from the medical field where large health care plans like Aetna and Kaiser Permanente use BISG to identify member race and ethnicity and measure disparities in health care outcomes. The researchers at the Rand Institute in Santa Barbara are leaders in the development of this analytical approach.
A Brief History of Fair Lending Proxy Analysis
Prior to the arrival of BISG, fair lending proxies have been determined in one of two ways: (1) location and (2) names. Location was used, for example, in determining whether or not an applicant was a minority by looking at the census tract where is applicant lived. If the applicant came from a census tract that was more than 80% minority, then for fair lending analysis purposes, that applicant was presumed to be a minority. Applicants from census tracts that are 80% or less minority are assumed to be non-minority. This approach works well for racial minorities that tend to live in concentrated neighborhoods. Blacks tend to live in such neighborhoods. Asians and Hispanics, for example, tend to live in racially more diverse neighborhoods.
Surnames have been used, for example, in determining whether an applicant is Hispanic. Based on surname lists from the Census Bureau, an applicant was considered to be Hispanic as long as his/her last name appeared on the Census Bureau list. Asians are another racial group where surnames are good indicators of race. Of course, these proxies depend on good surname lists.
Bayesian Improved Surname Geocoding
BISG combines the location and names approaches. It uses a list of 151,671 surnames gleaned from the 2010 Census along with each individual’s report race/ethnicity. Thus, for a name such as Smith, the Census Bureau list indicates what percentage of the population with the last name Smith are White, Black, Asians/Pacific Islander, American Indians/Alaskan Native, etc. This names data is then combined with the block group where the applicant lives to provide a probability as to the applicant’s race.
Why Use BISG versus Earlier Proxy Methods
While no proxy method is perfect, some studies, although not all, comparing the BISG method of producing proxies versus the earlier methods show the BISG method to be superior in identifying the race/ethnicity of the applicant because it combines two sources of information about an applicant’s race/ethncity. Of course errors still exist, but until race and ethnicity can be collected for credit products such as auto loans and secured and unsecured consumer loans, the evidence seems to suggest BISG is an improvement.
Why BISG and Other Proxies Are Important for Large and Small Institutions
Some individuals might argue BISG is only a concern for large institutions since they are not regulated by the CFPB. However, this sentiment ignores the fact that several recent fair lending settlements have involved consumer secured and unsecured lending at smaller institutions. The Fort Dodge settlement is a case in point. The analyses in these consumer cases require proxies and both the CFPB and DOJ used BISG. From a purely practical standpoint given the CFPB and DOJ use BISG, banks should at least know what the fair lending results using this proxy method are. Additionally, because of the way BISG works, the Bank is likely to see more prohibited basis applicants identified in their databases which may or may not provide analysis results that are beneficial to the Bank compared to older proxy methods.
Preiss&Associates has been doing custom fair lending analyses for more than 20 years. If you have fair lending questions, want to talk about your fair lending issues or have need for us to assist you with your fair lending program give us a call at 847-295-6881 or drop us an email at email@example.com.