Risk Management Association says FIs need to validate the accuracy of their lending and other decisions
The transition to digital operations is a great opportunity for banks and other financial institutions (FIs) to streamline their operations, but not without risk.
As FIs speed up their drive to digitalize their businesses, there are some pitfalls relating to the models used to make decisions about lending and other matters, according to a report from the Risk Management Association (RMA).
"Banks increasingly use models to make decisions as well as to identify and measure risk, conduct stress tests, and perform other critical operations. This includes some powerful innovative practices like AI-driven models," said Edward J. DeMarco, Jr., the head of Non-Financial Risk at RMA. "While this benefits customers by speeding credit decisions—and banks by making it easier to detect fraud—banks need to fully understand models and validate their accuracy to avoid creating more risk for their institutions."
The report is based on respondents from a cross-section of the industry and found that the top two challenges to expanding validation capabilities were talent (72%) and cost (63%).
Seven in ten firms validate the models they use for lending and other decisions every one to two years, industry best practice, but the rest only do so every three years.
With a general shortage of talent affecting financial services and other industries, the RMA has convened a Model Validation Consortium (MVC) that allows FIs to pool skills for risk validation.
"The increase in modeling comes at a time of growing concern about talent shortages in the financial industry, particularly in areas requiring technical skills such as model validation," said Kevin D. Oden, founder and managing partner of Kevin D. Oden & Associates, a leading risk modeling validation company and RMA partner for RMA's Model Validation Consortium. "This creates challenges for all banks but especially for community and mid-tier banks, as they often rely on third-party vendors to create models. Banks must verify that those models meet business needs and do not propagate errors or bias."