FICO: Poorly designed credit risk models can harm consumers, lenders, investors and the overall financial system. Fintech companies that circumvent regulations may be putting their customers and themselves at risk
The best credit risk models have five key attributes – predictive power, broad coverage, fairness, transparency and the ability for the consumer to dispute the process when mistakes are made
Recent trouble at marketplace lenders such as Lending Club has led many in the industry to say the fintech bubble is bursting. Whether or not this proves to be true, it is important to explore the unintended dangers posed by some of the new ways fintech companies are attempting to assess risk, as Dr. Andrew Jennings, senior vice president for Scores, posted on the FICO Blog.
A large body of regulation has built up in the U.S. to provide consumer protections related to the use of personal data, underwriting criteria and marketing practices. “Fintech business models that circumvent or violate these regulations could pose significant unintended dangers, including negative impacts for consumers, lenders, investors and the overall financial system,” Jennings said.
“In the lending industry, bad credit-risk models often lead to bad lending decisions. We need look no further than the recession of 2008-2009 for a reminder. Bad decisions can be devastating for consumers who get overextended. The decisions can also result in massive losses for investors, unsustainable business conditions for lenders, and a weakening of the financial system. “For example, should regulators (or the lenders themselves) today take a closer look at risk models in China where delinquency rates are near 25 percent in the P2P lending market? By comparison, delinquencies on credit cards in China are in the 3-5 percent range. At the very least, this indicates a possible need for new risk tools to guide underwriting policies in P2P lending.”
The best credit-risk models, Jennings said, have five key attributes – predictive power, broad coverage, fairness, transparency, and the ability for the consumer to dispute the process when mistakes may have been made.
“It isn’t always possible to build all of these attributes into every model, but it is important that data scientists try to include all of them in their models,” Jennings wrote. “It is particularly difficult to reach this gold standard when models have opaque variables, rely on data that could be easily gamed, or are so complex that problems can’t be diagnosed when the models fail. These are the dangers with many of the scoring models being touted by fintech companies.”
For example, transparency is important in credit markets because it helps ensure greater fairness and accessibility. If a consumer receives a low FICO® Score, a reason code like “balances too high” is clear and actionable. “But if, for example, a score uses data on a person’s educational background, what can be done when the score is low?” Jennings wrote. “Tell that person that should have attended a better college or read a different subject or attained a higher grade?”
“Such data is often predictive but would be problematic on many of the key requirements of a best-in-class risk model. Designers of alternative credit-risk scores may not intend to illegally harm anyone, but the unintended dangers of their models may result in unreliable lending decisions and/or widespread bias against certain groups of people.”
Care must be taken to ensure the process works for all participants by embracing innovations that spur a healthy marketplace while avoiding the dangers of poorly designed risk models.
For discussion of this and related topics from Dr. Jennings and other FICO experts, subscribe to the FICO Blog.
Source: FICO Press Release
