Fintech promised to revolutionize loans for those left out of the credit system. New research indicates it’s not living up to that promise

News Room

For years, the financial technology industry has touted its ability to spot good credit risks missed by big banks and traditional lenders, providing people with affordable loans they might not otherwise be able to access.

The technology companies and startups that make up the fintech world “rely heavily” on credit scores to price their loans and don’t incorporate other variables that could better predict default, according to a working paper distributed by the National Bureau of Economic Research earlier this year. As a result, fintech clients and customers with nonprime credit scores pay interest rates that are 45% more expensive than prime borrowers who are similarly at risk of default, the paper found. 

“If you believe that fintech is this really sophisticated machine that is able to identify risk better than traditional lenders, the surprising outcome of this paper” is that “it looks like it still has a ways to go to really price things according to risk,” said Mark Johnson an assistant professor of finance at Brigham Young University’s Marriott School of Business and one of the authors of the paper. 

For decades, lenders have relied on credit scores from companies like FICO to evaluate the risk that a borrower will default. Recently, those formulas have faced criticism from consumer advocates and lawmakers who say they use variables that bake in discrimination. As a result, these credit scores aren’t an accurate predictor of borrowers’ risk and are particularly faulty when it comes to low-income borrowers and borrowers of color, critics argue. Fintech lenders have said that by using the power of artificial intelligence and other tools to suck in a wide swath of data, they can find borrowers with bad credit scores who are actually good credit risks and lend to them accordingly. 

While these lenders may be using their models to provide credit to borrowers who historically have been considered too risky for traditional lenders, what Johnson’s paper suggests is that these borrowers are paying more for it than those who are similarly risky. These borrowers, who are paying $500 more on average in the first year of their loan, according to the paper, are more likely to be those with the least room in their budget to absorb the extra cost, said Matthew Bruckner, an associate professor of law at Howard University School of Law.  

“For low-income borrowers there is often a narrative around they’re higher risk, they are less likely to repay, they’re more financially marginal,” Bruckner said. The paper challenges that narrative, he said. “The loans themselves are what create the defaults,” he said. “If you’re being charged 10% instead of 4%, the payments are higher and it’s just harder to make them,” he added. 

‘A lot of mirror and smoke’

To understand how fintech companies are pricing their loans, Johnson and his co-authors analyzed aggregate loan data that covers about 70% of the fintech personal-loan market. The names of the companies weren’t included in the dataset. Some of the leading fintech personal loan originators include SoFi
SOFI,
-0.97%,
Lending Club
LC,
-0.64%,
Upstart
UPST,
+0.27%
and Prosper, according to S&P Global Market Intelligence. 

The paper’s authors, which include economists from the Securities and Exchange Commission, Georgia State University and Ohio State University, weren’t initially looking to the data to get a sense of fintech loan pricing, Johnson said. Instead, they wanted to know how sports gambling affected people’s use of debt. But what stuck out in their review of the data they obtained was that borrowers faced a large jump in interest rates when their credit score went even slightly below the 660 threshold, the cutoff for what’s considered a prime credit score. 

“That was a striking fact,” Johnson said. “You can imagine that people who have FICO scores of 659 are identical in terms of risk to people who have FICO 660.”  

That indicates that despite claims from fintech lenders that they harness information outside of traditional credit scoring to price loans, the “pricing still looks relatively unsophisticated,” Johnson said. The authors of the paper were able to determine that the fintech lenders’ pricing isn’t directly related to the risk a borrower will fall into delinquency or default by developing their own algorithm, which determined what interest rate a lender should charge a borrower if they were going to get a 5% return on the loan.  

What they found is that borrowers who are less risky, but have nonprime credit scores are overpaying for their debt and subsidizing riskier borrowers with prime credit scores. The authors speculate that there are two reasons why the loans in the data set aren’t being priced according to risk. The first is a lack of competition for borrowers with credit scores below 660. Because traditional lenders typically don’t work with this population, fintech lenders have more room to raise rates for this group. 

The other has to do with fintech lenders’ business model. These companies typically make loans and then sell them. Because banks won’t buy loans made to borrowers with credit scores below 660, due to regulation, fintech lenders have to look to other institutions to buy these loans.

“The hedge funds or other investors that take below 660 loans require higher returns so they have to charge higher interest rates,” Johnson said. “Interest rates don’t seem to be determined by risk, they seem to be determined, at least in part, by market forces that are outside of fintech’s modeling.”  

Marco Di Maggio, director of the Fintech Lab at Harvard Business School, said Johson’s paper indicates that in many cases fintech lenders’ promise to bring traditionally underserved people into the financial system is “a lot of mirror and smoke.” Instead, these companies are actually engaging in “a little bit of regulatory arbitrage,” Di Maggio said. 

“We can get funding from the banks and do riskier loans, securitize these loans and sell them to investors,” Di Maggio said of the system many fintech lenders use. “Everybody in the current system is using these new companies as a way of creating some value, but probably not really achieving financial inclusion.”  

‘There is hope for fintech’

Other research from Di Maggio indicates that at least one fintech lender, Upstart, does expand access to credit to borrowers who might struggle to get a loan from a traditional lender. An analysis of Upstart data by Di Maggio and co-authors (including an Upstart data scientist) found 

that borrowers with credit scores below 640 who received approval for loans through the company had a 60% probability of being rejected by traditional lenders. 

DiMaggio, who has no financial affiliation with Upstart, said he and his co-authors’ research on Upstart’s loans indicates that the company is pulling in more data than traditional credit scoring in evaluating potential borrowers to the benefit of nonprime consumers. They’re more likely to get credit and cheaper credit from Upstart than from a traditional lender. 

Johnson’s research doesn’t contradict the findings about Upstart, he said. For one, that paper was only looking at Upstart data and he and his co-authors are able to paint “a broader picture of fintech.” 

“It could be that Upstart is different from other fintechs, we of course can’t rule that out,” he said. “If that is the case, then that’s a really hopeful picture of where fintech is headed.” Another possibility is that doing the exact same analysis on the Upstart data would also indicate that borrowers with lower credit scores pay a higher interest rate than those with similar risk, he said. 

Paul Gu, Upstart’s co-founder and chief technology officer, said in an interview that his company’s model is indeed different from the median fintech model. “Just because the industry average looks like the bank average, it does not mean that every player in the industry looks like the traditional bank average,” he said. 

Gu noted that the Upstart model doesn’t place any special emphasis on traditional credit scores in determining whether a borrower is a good credit risk and how to price their loan. Still, the paper from Di Maggio and his co-authors indicate that Upstart borrowers with a credit score of 650 or below are more likely to pay interest rates of about 30% or higher than those with credit scores of 651 or above. 

FICO is still correlated with risk, so it’s not surprising that borrowers with a higher FICO score would have a lower rate even in Upstart’s model, Di Maggio said. What the paper suggests is that borrowers with a similar FICO score, including those with nonprime scores, can get a different and often lower rate from Upstart because FICO is just one of many inputs, he said. 

Di Maggio and Johnson’s papers also have findings that overlap  — most critically, that there are borrowers out there who are decent credit risks, but are being ignored by traditional lenders because they have low credit scores. How you spin the findings depends on the baseline you’re comparing fintech lenders to, Bruckner said. 

On the one hand, borrowers with low credit scores who might be denied credit by traditional lenders are getting loans at lower interest rates than they would if they were turning to other types of alternative credit, like a payday loan. Nonetheless, these borrowers are paying more for their debt than would if they were being evaluated simply based on their risk of default, Johnson’s paper found. 

“I see a problem that people who are similarly situated are being treated differently,” Bruckner said, though it’s not clear to him whether there’s a good regulatory solution to this problem. Instead, he said the entities that oversee fintech firms, including the Consumer Financial Protection Bureau, the Federal Trade Commission and state financial regulators should monitor the dynamic closely. 

“In this developing technological space, we should allow regulators to identify gross problems and solve those problems, but not necessarily to make rules that are binding for now and forever,” he said. “Supervision might be better, lawsuits in individual cases, policing the boundaries might be a better solution for now until these industries mature a little bit.” 

In the past, fintech companies have pushed back against proposals that would subject them to interest rate caps and regulations similar to traditional brick and mortar banks. For example, some criticized a law that Colorado legislators passed this month that would require fintech lenders who partner with out of state banks to lend to Colorado borrowers to comply with the state’s interest rate cap. The idea behind the law was to prevent fintech lenders from charging interest rates that are higher than what’s allowed in Colorado by partnering with out-of-state banks that have higher or no interest rate caps. 

Some fintech companies have justified the push for lighter touch regulation in part “by the claim that they’re offering access to lower cost credit to people with lower credit scores,” said Ellen Harnick, the executive vice president for the western region at the Center for Responsible Lending. “What this paper indicates is that in general it’s simply not true,” she said of Johnson’s findings. 

Though “there’s hope” that using more sophisticated models to evaluate credit risk could increase access to credit among those who have traditionally been underserved by the lending system, there’s also “reason for caution,” Harnick said. 

Consumer advocates have raised concerns that some of the alternative data fintech lenders say they use could reproduce racial inequalities. 

“There is a lot of black box about what actually drives these algorithms,” Harnick said. “There is a real risk that these methods of algorithmic underwriting, once again, may benefit some people and not benefit — or really disadvantage — others and that yet again there will be a racial component to how the benefits and burdens are spread.” 

Read the full article here

Share this Article
Leave a comment