No matter how hard developers try to identify and avoid discriminatory variables in models, these variables can still find their way in.
Even seemingly innocuous lifestyle questions, such as which brand of deodorant or shampoo someone prefers, can potentially determine a person’s gender, which is a prohibited basis group. These models often are used by marketing departments and can lead to unintentional discriminatory segmentation if they aren’t closely monitored.
A model review team needs to have a good understanding of how models are developed, how model variables work, and whether the weighting of each variable could have an unintentional fair lending impact. Otherwise, models might unintentionally affect prohibited basis groups in disparate ways, including:
- Excluding majority-minority census tracts or low- and moderate-income areas
- Weighting female applicants less favorably than male applicants
Not fully understanding the models in use can result in penalties, consumer reimbursements, and increased legal and reputational harm.
The need for a fair lending compliance program in models is just as important as the need for validation. How organizations approach fair lending review alongside model validation can significantly influence levels of risk and efficiency.
Yet at many organizations, the model risk management and legal and compliance departments collaborate less than they should – or they don’t collaborate at all. In some cases, the fair lending topic might enter the conversation so late in the process that extensive rework is necessary for compliance.
Addressing disconnects and improving communication between model validation and fair lending compliance teams can save organizations substantial time and money and improve productivity.