Artificial intelligence-guided lender Upstart (UPST -2.70%) had a tough year, with its stock price down about 86%, as high interest rates dried up demand for loans originated through the company's platform. But Upstart is certainly not the only stock that got creamed this year. When a stock is struggling, the question to ask is whether the company's business model and thesis are still intact.
Upstart's investment thesis is that it can underwrite credit better than traditional underwriting methods such as Fair Isaac's (FICO -2.14%) FICO scoring system. Upstart largely originates unsecured personal loans, although the company does have plans to get into other lending verticals.
Following its third-quarter earnings results, let's take a look at how credit quality for loans originated through Upstart's platform is holding up.
Examining rising delinquencies
Loan defaults are on the rise among Upstart borrowers. During 2020 and 2021, when credit conditions were benign because of federal stimulus efforts, the ultralow interest rate environment, and high personal savings, Upstarts loan vintages outperformed expectations. But as interest rates jumped, stimulus programs ended, and consumers ran through their savings, defaults reverted to normal and started to underperform, rising beyond what was expected.
This increase in defaults followed a stimulus-driven period during 2020 where defaults were approximately 50% below expectations. A reversion to pre-stimulus default rates and a shift in credit risk profiles accounted for a large majority of the increase in late 2021 and beyond. *Loss variance represents actual In-Period Loss divided by Expected In-Period Loss. Note: Data is for total originations made via the Upstart platform as of Oct. 31, 2022. Prior Forecast is a reference to the Q2 2022 earnings presentation. Image source: Upstart.
Upstart recently revealed that it developed a metric called the Upstart Macro Index (UMI), which "is designed to estimate the level of default to expect in a time period, holding underwriting models and borrowers constant."
Upstart CFO Sanjay Datta said on the company's recent earnings call that in the third quarter, the UMI was at 1.7%, which means defaults were 70% higher than management would expect them to be "in a long-run normal macro environment." Furthermore, Datta implied that Upstart uses the UMI to price loans. So if the UMI is at 1.7, Upstart is now pricing loans at a UMI of 2.
It's a bit hard for me to wrap my head around this new data point. On one hand, it seems very useful, but on another, it also seems as if you have to be able to accurately predict economic conditions for this to work, which is obviously hard to do.
I mean, no one saw the current climate materializing earlier this year. Datta said there are some economic variables in regular bank stress testing that can estimate the UMI accurately, but it still seems as if the company is reacting retroactively to the economic environment.
Upstart vs. FICO
The other question is how much more effectively than traditional methods Upstart can assess borrower risk. A big traditional benchmark, especially for unsecured personal credit, is FICO, which examines the creditworthiness of a borrower based on credit information such as whether the borrower paid back previous debt and how much outstanding debt they might have.
Upstart long asserted that it can assess credit risk better than FICO and that its credit underwriting algorithms can find creditworthy borrowers with low FICO scores. In its third-quarter earnings presentation, Upstart included the following slide.
Upstart’s risk grades predict a steadily increasing default rate from left to right as the borrowers get riskier, as it should. But if you look at any column from top to bottom, there’s relatively little difference between default rates, regardless of the FICO score. Note: Upstart internal performance data as of Oct. 31, 2022. Consists of all originations made from Q1 of 2018 to Q2 of 2022. Image source: Upstart.
Upstart assigns grades to loans within specific FICO ranges, with A+ being the highest quality and E- being the lowest. According to this slide, the AI-led lender is able to more accurately separate creditworthiness within a specific FICO range, which is supportive of Upstart's thesis.
I question whether Upstart is finding enough A+ loans within a certain FICO range to make a meaningful difference in volume. Right now, with the personal savings rate in the U.S. low and revolving credit very high, there should be plenty of borrower demand.
But Upstart's conversion rate (loan transactions/legitimate loan inquiries) fell in the third quarter due to the UMI being higher, leading Upstart to do fewer loan transactions and get fewer loan inquiries in Q3.
Is the thesis still intact?
If the data provided by Upstart is accurate, the company is assessing risk more accurately than FICO, which is a plus for Upstart.
But I still wonder if what Upstart is doing is big enough. For one, it's hard to know if Upstart is finding enough new borrowers in the cracks to make a meaningful difference for banks, credit unions, and investors. And two, will its default rates outperform FICO by enough of a margin to also make a meaningful difference?
The company onboarded lots of banks and credit unions interested in the technology, but I'm not fully convinced it has proven its concept yet. It also still seems like the company is doing a lot of tinkering with its algorithms on the fly. So while the thesis isn't broken, I think it's still too early to invest in Upstart, especially with all of the other headwinds weighing on the business model right now.
