Upstart is a Fintech company in the US that functions as a lending marketplace connecting borrowers to lenders and takes a fee for doing so. “Founded by ex-Googlers, Upstart goes beyond the FICO score to finance people based on signals of their potential, including schools attended, area of study, academic performance, and work history.- from their website”
Upstart went public in December 2020 with a $1.45B valuation, now has a valuation of $3.9B (May 20th). That however was not its peak valuation, if you look at the chart from Yahoo finance below.
How does it work?
Upstart operates on AI driven Credit decisioning algorithms that apportion out customers (for personal loans) to partner banks that state their eligibility criteria to Upstart based on FICO scores. The management recently announced plans to go big into Auto financing and during the Q1 analyst meeting showed considerable progress in this field with partnerships with a number of top dealerships for Auto financing. So far so good, right?
So why did the stock tank 50% after the Q1 results were announced by the CEO in May?
If you look at their AI model’s performance, it beats the predictability of FICO scores significantly. That is, the model works better in predicting the risk of a default by a customer. This is of course as per Upstart’s Q1 presentation based on historic performance of customer loans originated upto 48 months back.
Interpreting the chart:
For a FICO score of 700 or more, the average default is predcited to 3.4%. However, in Upstart’s Cell “D” with a FICO score of 700 or more, the default rate is predicted to be 6%.
Upstart’s AI model predicts chances of a credit default by a customer better than traditional UW using FICO scores. You can see how this can be incredibly attractive for Upstart’s banking partners. This is because FICO is not a great predictor of credit worthiness for all customers with a high FICO scores. Upstart’s AI model seems to have other parameters that better predict the default probability.
So, what could go wrong now?
This is the key issue: The model has been operating in a low interest rate regime that US enjoyed because of the Fed’s easy money policy.Here are my thoughts based on a study.
Upstart's AI model was trained on credit data from a low-interest rate regime and it is likely that in a high interest rate regime, the model will not function as well. Why do I say that? An important sub point- the US government issued stimulus checks during the COVID period- all that led to increasing the ability of the customers to repay loans. Not the best time to train an AI algorithm, would you agree?
Other issues that could have played a part:
- The loans that Upstart's partner banks do not write, Upstart takes those loans on its own balance sheet. The quality of these loans is possibly debatable and has spooked the analysts and markets alike.
- In a high interest-rate regime, borrowers may prioritize essential interest payouts (like mortgages), over Upstart originated personal loans, leading to a higher probability of default.
- Inflation pinches the interest servicing power of the borrowers, further causing a chance for defaults of their portfolio that sits on their balance sheet.
- As the portfolios go sour, their partner banks may not take in the loans originated by Upstart.
The moral(s) of the story:
- The upcoming high interest , high inflation regime is not an easy time for FinTechs to take on balance sheet risks. Stick to lead fees and origination fees.
- There is a time to grow when capital is cheap and plentiful and a time to moderate the growth projections (like in this current environment).
- AI is not a magic bullet and is still driven by the limitations of the data set that the models were exposed to in the past.