The second-quarter results from the artificial intelligence (AI) lending platform Upstart (UPST 0.96%) were disappointing and included wide-scale funding issues. But there was one bright spot in the earnings report.
The company showed updated data that indicates its AI-powered loan-underwriting algorithms have been assessing risk significantly better than Fair Isaac's (FICO 1.82%) traditional FICO credit scoring. FICO has been one of the main ways lenders evaluate the credit worthiness of individual borrowers for some time now.
This updated data seem impressive, but I'm still a bit skeptical, and I think the market is, too.
Banks are pulling back
Upstart's whole business model is built on the premise that its technology can assess the creditworthiness of loan applicants better than traditional methods like FICO. This would enable the company to find additional borrowers that haven't historically had access to traditional credit products but are creditworthy while helping banks increase their loan issuance without increasing their default rates. Based on data Upstart provided in its second-quarter earnings materials, its models are outperforming FICO by a wide margin.
Image source: Upstart.
As you can see above, Upstart evaluates loan applicants and assigns them a grade, while Fair Isaac's evaluation assigns a numerical score. What the chart shows is those loan applicants who received a high FICO score (deemed more creditworthy) but a low grade from Upstart (less creditworthy) tended to default on loans at a much high rate than applicants who got a low FICO score but a high Upstart grade. Likewise, those with a top grade from Upstart didn't default all that much no matter what FICO score they received (the default rate ranged from 0.7% to 1.3%). The outperformance is significantly noticeable in the 639-and-below range where Upstart A+ grades have a default rate of 1.3% and E- grades have a default rate in excess of 14%.
Despite this data indicating Upstart can better assess creditworthiness, the number of loans originated by banks and credit unions through Upstart's platform and then retained on their balance sheets really fell off in the second quarter of 2022, with total transaction volume down more than 15% from the first quarter. That was surprising. I was under the impression that most banks and credit unions were using Upstart to originate loans largely to prime borrowers and above. For instance, one of Upstart's first and oldest banking partners, Customers Bancorp (CUBI 1.55%), disclosed information about its personal loan portfolio in the first quarter of this year and only 1% of the portfolio had a FICO score below 680.
It surprises me that banks and credit unions would pull back so heavily on originating loans to prime buyers and above because credit quality is still quite good among this group, and when you consider how widely Upstart is outperforming. There was also plenty of loan demand in the second quarter. Perhaps the 15% drop represented loans to near-prime borrowers and below but I didn't think banks and credit unions were doing this amount of subprime lending with Upstart because they still tend to be quite conservative.
The bond market has no faith right now
The other interesting thing about Upstart is that professional investors apparently have very little faith in the company. In August of 2021, Upstart issued $575 million of senior unsecured convertible notes that are due in 2026. The notes bore an annual interest rate of 0.25% and convert into stock at $285.26 per share.
Recently, however, due to the falling share price of Upstart and how far away it is from its conversion price, the yield to maturity on the bond has risen to 13.9% on Aug. 10. Bond yields and prices are inversely related, so as the price of the bond has lost value, the yield to maturity has skyrocketed far away from the bond's coupon rate because investors have seemingly lost faith in the model. It's a bit hard for me to believe that professional investors could see Upstart's credit outperformance, believe it, and still place a yield-to-maturity-rate this high.
Additionally, current default rates in Upstart's securitization vintages (loans that have been issued long enough with enough on-time payments made that the risk of default is lower.) issued in 2021 and early 2022 are now underperforming their target gross returns at this point in time. Management has said that this is largely due to the termination of stimulus programs, which is no doubt a significant contributor.
But 2021 vintages are performing way worse than pre-pandemic vintages, so it's hard to say this is just a reversion to normal. The later vintages are filled with loans to borrowers lower on the credit spectrum, and they really just don't seem to be performing well at all.
It might still be too early to tell
I am by no means a data scientist, so I cannot definitively say one way or the other whether Upstart's AI underwriting models are outperforming FICO. What I can say is that the market doesn't seem to be buying it right now when you consider the actions of bank partners, institutional investors, the bond market, and the beaten-down stock price.
Upstart could end up outperforming FICO, but I still think it is too early right now, and therefore I don't know if the data the fintech company provided is enough to make a determination about the true effectiveness of the company's technology.
