UPST: new KBRA report blow out


The new KBRA surveillance report on UPST, released 8/19/2021, is an absolute blow out, just like UPST’s Q2 2021 earnings last week.
If you want to access the actual report you will need an account (which is free):…

I’d like to reference you to my initial post digging into KBRA data on UPST and its competitors, where UPST’s data in that post was drawn from what was available (KBRA reports from 10/13/2021):…
Specifically, look at the screenshot showing KBRA’s base case cumulative net loss (CNL) projections for UPST:

Now, compare that to what we see in this new report.
Screenshot of UPST CNL and delinquencies:

KBRA’s base case CNL has been revised down across the board on all vintages, at an average reduction of 30% from base case CNL at closing (a range of 24.4% to 40.5% reduction in CNL).
This is likely a conservative projected reduction, as we can see the actual current CNL for some vintages is over 50% lower than the KBRA’s expected base case % for the current period! A future report may very well revise the CNL even lower.

We can also see UPST total delinquencies across nearly all vintages are below 2.00%.
And this is with UPST borrowers sitting at an average weighted FICO in the 670s.

For comparison, I looked at KBRA’s report on competitor MFT (Marlette, also branded as Best Egg), which was published 8/13/2021.
Marlette brags about concentrating on a prime loan business and rejecting most non-prime applications. They sport an average weighted FICO of 721 on their most recent trust (the last four trusts have averaged FICO 720-724) - yet data shows total delinquencies across some vintages at double those of UPST, up to 4.72%!
Screenshot of Marlette’s delinquencies:

And finally, here is what a typical non-prime lender with traditional underwriting yields, for another comparison. Avant lends to borrowers with a weighted average FICO of 650. Most up-to-date information is from KBRA’s last review of Avant in May 2021. Their delinquencies and expected CNL are far greater than what is projected for UPST - see screenshot:

All of this is reinforcing the idea that UPST’s underwriting algorithms have truly figured out how to find the “hidden prime” borrower. We should not be surprised that, as according to last week’s conference call, more than 150 institutions are now buying Upstart-powered loans or bonds, up from 100 institutions. Combined with their regulatory and compliance advantage with the CFPB No Action Letter, I am increasingly viewing their AI/ML technologies as a true moat, and not just a first mover advantage in this space.