There are a number of issues that keep popping up as risks for investors in Upstart. In my opinion, they are for the most part just noise.
So let’s look at the issue you just cited about Upstart’s AI model performance during a recession. Has it been tested during such times? And then you reference 2008 as the year that should be used as a source of data during a recession. But every recession is unique. Even if data from 2008 were incorporated to Upstart’s model (it may be, I don’t know if they reached that far back), how relevant would it be to the next recession? The 2008 recession resulted from the vastly over extended real estate bubble. That doesn’t seem to be a factor at present. If we experience a recession in the near term, it will most likely be the result of over-correction of the Covid hangover and the war in Europe currently driving inflation.
OK, let’s break this down a bit. Let’s start with what Upstart does with or without data representative of a recession. They have created an AI engine that does a superior job of assessing the true risk of a prospective borrower as compared to FICO.
FICO is a 30 year old tool that has zero scientific basis. It was the creation of the Fair Isaac Co. (hence the name, it’s an acronym). What Fair Isaac did was to codify the risk analysis that was in common practice among most US lenders. You can Google it, so I won’t go into details, but in a nutshell there are five weighted factors that go into assigning a FICO score. They were selected based on “common sense” and general availability.
Along comes Upstart with an AI based tool that has been trained on over 15 billion cells of performance data. Each Upstart assessment is based on the interactive relationships of well over 1,000 different factors specific to the borrower (I don’t recall the average number of factors, but 3,000 sticks in my head). The difference between an Upstart and a FICO risk assessment is enormous. Upstart’s assessment is based on data science versus FICO which is based on “common sense” and the gut reaction of a loan officer. About 70% of Upstart loans are fully automated which means the borrower never sees a loan officer (and is not burdened with document production). I might add that Upstart’s model is constantly getting smarter as new data is incorporated to the ML engine on an on-going basis.
Here’s what CEO, Dave Girouard wrote in the prospectus about risk assessment during recessions:
“What about those ugly recessions? Lenders would do well to put away their crystal balls and instead focus on predicting which borrowers will pay them back in any environment. While the probability of default for a borrower may vary by as much as 2x depending on the state of the economy, our experience suggests that risk associated with two borrowers with similar credit scores can vary by at least 10x.” He goes on to discuss credit shock due to Covid. https://ir.upstart.com/static-files/069cef16-cb41-462a-8059-…
In case it’s not obvious, my point is that when challenging Upstart’s risk assessment performance during a recession, it is important to keep in mind that this is a relative question that needs to answered with respect to the available alternatives, that being a field of one, FICO.
So, if “UPST has been given a downgrade in part because this analyst believes UPST has not proven itself in a “true recession” yet.” One should pause and ask exactly how well has FICO proven itself in a true recession? If you are going to downgrade Upstart based on macro-economic factors it must be evaluated with respect to the performance of the alternative under the same set of conditions. As Mr. Girouard noted, FICO has not performed all that well.
This is why I don’t pay much attention to the Wall Street analysts. They consistently fail to contextualize their “analysis.”
While I’m at it, let me address another commonly expressed risk for Upstart. That being the threat of a “BIG bank,” like JPMorgan, CITI, BofA, Wells Fargo, etc. creating their own competitive AI engine to assess borrower risk.
First off, this supposed threat is based on ignorance. The thinking is that a big bank with deep pockets could easily hire the expertise required to develop a competitive tool. The fact is nothing could be further from the truth. Upstart’s ML/AI software constitutes an extremely deep and wide moat that is incredibly difficult to emulate. And even if a would be competitor tried to attack Upstart’s IP, it’s important to understand that this is a time consuming and costly endeavor. Upstart went public in December 2020, but they have been in business since 2012. They have an enormous head start, and their lead keeps getting longer every day. Assuming that a competitive brain trust could be assembled and the requisite data could be made available, overcoming the lead that Upstart has established would be a near impossible task.
If you go back and read John Wayne’s excellent posts on Upstart from earlier last year you will find a post describing the efforts of a credit union (I forget which) that tried to develop their own ML/AI risk assessment tool. They found the task to be beyond their ability. They found that as they tried to incorporate more variables they code became increasingly fragile until they reached a point at which it simply failed to provide an output. It became obvious that it was easier, cheaper and more efficient to simply partner with Upstart and pay a reasonable fee in order to obtain an AI risk assessment.
But, OK, let’s concede that Jamie Dimon decides it’s a point of pride and his bank must have their own ML/AI tool. There’s 5,000 banks in the US, so that would leave 4,999 mostly local and regional banks for Upstart to partner with. You can bet that Jamie would not be sharing his tool. And along with the banks, there’s also 5,400 credit unions. This demonstrates Upstart’s really intelligent business model. Rather than competing with lending institutions, they provide a service available to all of them for a reasonable fee.