Machine learning and statistical testing at vantagescore

Kimberly Adams, How one credit scoring company is thinking about financial inclusion, Marketplace Tech,  JULY 8, 2022. (more around July 2022)

Vantage Score, Protecting Consumer Credit Scores From Medical Debt Collections, MARCH 18, 2022

Feature selection:

Earlier today, the three national credit reporting agencies (CRAs), Equifax, Experian and TransUnion, announced a joint industry measure to remove nearly 70% of medical collection debt accounts from consumer credit reports.

The CRAs also announced that effective July 1, 2022, they will no longer include paid medical collection debt on consumer credit reports and will extend the time for unpaid medical collection debt from six months to one year, giving consumers more time to work with insurance and/or healthcare providers to address their medical debt before it impacts their credit.

VantageScore applauds this decision and shares in the commitment to help consumers recover from the COVID-19 pandemic.

In 2013, VantageScore made the empirically-driven decision to become the first national credit scoring company to entirely eliminate medical collection accounts that have been paid off. Since then, the company has taken a number of additional steps to reduce the impact that medical debt has on a person’s credit score.

“This is a major win for consumers. Medical accounts are different from other consumer debt because they often arise from unforeseen circumstances and complicated, opaque insurance and healthcare provider billing practices,” said Silvio Tavares, President & CEO of VantageScore. “Based on empirical research, these types of debts and collections that have been paid off are often not as predictive of a consumer’s creditworthiness. This is why VantageScore was a pioneer in eliminating their impact in our credit models nearly a decade ago, and it is one reason why more than 2,200 financial institutions use our scores.”

Vantage Score, Model Performance Assessment, 2021. 

A regulatory requirement for any credit scoring model is that there is no statistical bias with regards to any protected classes. Therefore, a key element of the annual model performance assessment process is to ensure that VantageScore 4.0 continues to be free from any such statistical bias by comparing default rates observed for a given credit score between consumer ethnicity groups.

Specifically, a Chi-Square based formal statistical bias test was run on protected ethnic class population segments, i.e., African-American (AOMC) or Hispanic-American (AOHC) populations, to assess if there are statistically significant differences in the actual default rates between these different ethnic groups and other groups. Census data at the zip codelevel was appended to the consumer data as a proxy for ethnicity. Bankcard and mortgage products were selected for this test based on their relative size and because they represented mainstream lending.