Agregue a sus temas de interés Cerrar
lunes, 12 de agosto de 2019
lternative data that evaluate creditworthiness are gaining traction, but policy makers grapple with fairness issues
The Wall Street Journal
Should borrowers be denied new loans because they didn’t finish college?
That sort of question is vexing policy makers as they seek to encourage lenders to use new types of data and computer-driven models to allow more borrowers to qualify for loans and at lower prices.
Such efforts seek to address mounting criticism of an existing credit-evaluation system that relies on past loan-repayment history but also raises questions about fairness and accessibility to credit.
Several bills have been introduced in the House of Representatives this year by lawmakers from both parties to improve the credit-scoring system. Last month, a bipartisan House financial-technology task force held a hearing on the use of alternative data.
Rep. French Hill (R., Ark.), the task force’s top Republican, said alternative credit criteria has “the potential to widen the universe of borrowers and provide greater access to affordable credit.” Citing a report by credit-reporting company TransUnion , Mr. Hill said two in three lenders were able to lend to more borrowers with the use of alternative data.
Sen. Kamala Harris (D., Calif.), a Democratic presidential candidate, has proposed including on-time rent payments and cellphone bills in credit scores as part of her campaign pledge to boost black homeownership.
Some financial-technology companies are already using these techniques. At Upstart Network Inc., an online lender founded by former Google Inc. employees, loan applicants are asked to provide the highest education degree obtained, the names of university or colleges attended and areas of study, as well as employment history.
“The use of occupational history and educational background generates a significantly more accurate credit model,” says Dave Girouard, Upstart’s chief. He adds the company’s model approves 27% more loan applicants for its personal loans than the traditional credit-scoring model and results in 16% lower average interest rates for approved loans.
Other lenders are also using educational history or employment status to screen borrowers, with similar results, along with more widely accepted data such as rent and utility payments and cash-flow records from borrowers’ bank statements.
Despite the growing embrace of such measures, some consumer advocates say the trend will hurt lower-income or minority borrowers. “It entrenches and perpetuates inequality in such an obvious and stark way,” says Chi Chi Wu, an attorney for the National Consumer Law Center, who adds that the existing system already makes it harder for people with lower income and fewer assets to qualify for cheaper loans.
The ratio of people with at least bachelor’s degrees is 33% among whites and 54% among Asians, compared with 23% for African-Americans and 16% among Hispanics, according to census data.
Policy makers are hopeful, though, that alternative data will allow lenders to reach some of the 45 million Americans, or 19.3% of the adult population, who currently don’t have access to credit, according to the Consumer Financial Protection Bureau.
Given their reliance on people’s loan-repayment history, traditional scoring models handicap young people with no or limited credit history, as well as those who may have had previous financial difficulties.
But determining what types of data should be permitted in credit scoring is a complex and contentious task. The Equal Credit Opportunity Act of 1974, the main law used to ensure credit access and prevent discrimination, prohibits lenders from using information such as race, sex, national origin and age as factors in determining credit availability. The use of other types of information, such as income and assets, are permitted.
Whether the use of data related to education and occupation is appropriate is debatable. While serving as a strong indicator of people’s creditworthiness, the data show similar correlation to prohibited factors. A 2015 study by Federal Reserve economists showed the average student-loan delinquency rate among those who didn’t finish college was 44%, compared with 11% for bachelor’s degree holders.
“America’s 1970s-era antidiscrimination law is not well suited to deal with today’s big data reality,” says Aaron Klein, policy director for Brookings Institution’s Center on Regulation and Markets. He added change must start with an honest conversation about which borrowers deserve special protection and how to compensate lenders for serving riskier consumers.
The uncertainty largely has kept traditional banks from using alternative data. Even financial-technology lenders in the U.S. typically stick to noncontroversial information such as bank-account data, and stay clear of information gleaned from social media.
“There is yet a framework that makes it perfectly safe to [use such information] without facing an existential threat,” said David Klein, chief executive of CommonBond, an online student-loan provider.
The Government Accountability Office, the independent watchdog arm of Congress, recommended in December that financial regulators provide clearer guidance to lenders about how to use alternative data in the underwriting process, something that hasn’t yet happened.
While most lenders tightly guard their lending models, the experience of Upstart is revealing. The company has developed and operated its lending model under the supervision of the CFPB for nearly two years. It is the only company that has been approved for its innovation program. On Monday, the CFPB unveiled detailed data on how Upstart’s model has expanded credit access, and encouraged other lenders to explore ways to lend to more people with no or limited credit history.
Upstart’s Mr. Girouard says the company’s model provides higher approval rates and lower interest rates for “every traditionally underserved demographics.”
It initially tried to use the borrowers’ test scores and grade-point averages, but abandoned it four years ago because “efforts to collect and verify the information was not worth the predictability it provided,” he said.
By Yuka Hayashi