UPST: Customers Bank, Lending Club earnings

Lending Club and Customers Bank released earnings reports today.

Main takeaway points:

1.) There doesn’t appear to be a macro slow down in fintech personal lending for Q3.
As we know, Customers Bank is a bank partner of UPST. They have contributed as much as 9.7% of the borrower pool in recent UPST securitization transactions. Customers Bank reported Q3 earnings today. Consumer installment loan balances grew 32% over a year ago. 78.3% of consumer installment loans are personal loans.
They also crossed over $1 billion in direct Customer Bank personal loan originations this quarter (cumulatively, I assume – not 1 billion originated within Q3 itself??).
So, I’m not sure how to estimate the total personal loan amount originated in Q3 specifically, but if someone could figure that out, it would be very helpful as it can be used to estimate Upstart’s Q3 personal loan numbers…

[Something I also want to point out here, is that Customers Bank has far more stringent credit score requirements than Finwise or Cross River. Their policy is a minimum score of 680. Their consumer installment loan average FICO is 740, but they do not provide a breakdown of their personal loan FICO average. They define consumer installment as a mix of personal loans, student loans, home improvement loans, specialty loans]

Meanwhile, Lending Club’s loan origination grew 14% quarter over quarter.…
Taken together, there doesn’t appear to be a macro slow down in the fintech personal lending market in Q3. That’s good, as we certainly expect Upstart to do well in Q3.

2.) UPST is likely to continue to grow much faster than Lending Club.
Lending Club growth rate for new borrowers continues to lag way behind UPST’s.
Lending Club stated on their earnings call today: “I mentioned we added about 100,000 new customers this quarter” and “average loan sizes of, call it, $14,000, $15,000” .
This implies about $1.5B of their $3.1B loan origination total for Q3 is coming from existing members.
So we can estimate about 50% of loans are to existing members (repeat borrowers), and 50% are new customers.
Lending Club was established in 2006, so they’ve had 15 years to build up their member base of 3.8 million currently.

In very stark contrast, we know UPST exceeded 100,000 quarterly new customers about a year ago, all the way back in Q4 2020!
UPST did 123,396 loans that quarter…there’s a GIGANTIC difference here in growth.
Upstart started personal loans in 2015. In just 5 years, they’ve surpassed Lending Club’s new quarterly borrower counts – something that Lending Club has taken 15 years to get to!

Not only that – but we know UPST did more than 100,000 new loans in a single month of June. At that run rate, UPST in Q2 was more than tripling what Lending Club could barely achieve in the 3 months of Q3 2021.

And we know UPST has said their repeat customer count is growing (repeat customer rate doubled from Q1 to Q2 2021), but still a small number.
That can only expand further over time - imagine when UPST has a ‘past borrower base’ matching that of LendingClub’s 3.8 million members.

3.) Lending Club is moving farther up the prime market segment and becoming less of a direct competitor to UPST in the nonprime market.
Why is Lending Club’s growth rate so slow versus Upstart?
The main reason, as I’ve shown in past posts on the board, is their lack of robust AI/ML underwriting models to properly price loans for below-prime borrowers. They are trying to avoid the high rate of below-prime borrower defaults they suffered in the past.

Lending Club has chosen to limit its lending to existing members/repeat borrowers, and only trying to acquire new borrowers within the above-prime market.

Per their call today: “And said another way, we don’t expect it to change much from there… just a reminder, what we’re holding is prime consumer loans, same loans we sell to a few dozen community and regional banks, [average] customer you’re talking about here is – average FICO is about 715. Income is… somewhere in the $90,000 to $100,000 range for individual income”.

In contrast - we know very well that UPST has the entire below prime and some of the prime market to target, while Lending Club is constraining their growth to the above prime market.

4.) Lending Club’s delinquency rates appear great on a quick look, but under the hood, there is nothing special. As I’ve posted before, Lending Club lacks robust AI/ML models that allow it to underwrite borrowers any better than the average fintech competitor.

On page 14 of LendingClub’s Q3 presentation (…), they boast about their Q4 2019 prime vintage delinquency rate being “50%” lower than fintechs overall.

Unfortunately, I’m not impressed with their Q4 2019 delinquency performance. They do NOT break out what percent of the Q4 2019 vintage are repeat borrowers.
A repeat borrower, who they know has already paid off a past loan with Lending Club, will be much, much less risky than a new borrower.
From combing through their past 2019, 2020 transcripts and presentations, it’s estimated that 30 to 50% of loans originated at the time were to repeat borrowers.

To give you an example of how repeat borrowers can make your overall losses look great, you can study another fintech competitor: LendingPoint.
LendingPoint’s securitized data for 2019-2 (11/25/2019 transaction date) reported 26.12% of their pool were for “renewal loans” and 73.88% were for new borrower loans.
With a 26% repeat borrower mix, and average weighted FICO of 669, they achieved 10% cumulative (net) loss as of 10/2021.
At first that looks very impressive for their FICO average (another fintech lender, Avant, had 12% CNL at similar months of seasoning with FICO average of 647) - but then you look at the breakdown in the KBRA report and you realize the NEW loan portion was hitting 17% cumulative (gross) loss rates, and the renewal loans at 12% cumulative (gross) losses!
The collateral pool of the better performing renewal loans are cushioning the actual losses from the new borrower loans, driving the cumulative (net) loss down.

The same is happening for Lending Club.
Again, they don’t tell us what percent are repeat borrowers and what percent are new borrowers, but it most certainly is at least 30% repeats in 2019.
One point of comparison we can do is with a competitor having similar FICO range: Marlette. They were founded in 2014, unlike Lending Club in 2006. Marlette does not have any significant repeat borrowers, versus Lending Club.

Lending Club’s 2019-P2 securitization (transaction date 9/2019) had 712 weighted avg FICO with 1.93% delinquency rate as of October 2021.
Marlette’s 2019-4 securitization (transaction date 10/2019) had 716 weighted avg FICO with 2.62% delinquency rate as of October 2021.

So, on the surface, it appears Lending Club is outperforming Marlette…but, since we know 30%+ of Lending Club’s borrowers here are repeats and Marlette’s are nearly all new borrowers, there’s absolutely nothing impressive here about Lending Club having only 26% lower delinquency rate.

Well, how does a ‘real’ AI/ML credit underwriting model perform?
We already know the answer - we can illustrate with Upstart.
Let’s compare UPST to a very fast growing fintech competitor, Upgrade.
Upgrade was founded in 2016. Both UPST and Upgrade have very insignificant fractions of repeat borrowers given their ‘newness’.
Upgrade does credit cards and personal loans, and are growing very very fast (but at what potential default costs in the future…?)…

Upstart’s 2019-1 securitization (transaction date 2/28/2019) had weighted avg FICO 687 with 2.25% delinquency rate as of October 2021.
Upgrade’s 2019-1 personal loan securitization (transaction date 2/21/2019) had weighted avg FICO 693 with 5.65% delinquency rate as of October 2021.

Here’s their securitizations’ FICO range breakdown:

            Upstart  Upgrade
Below 700   66.02%   62.4%
Above 700   33.99%   37.6%
**Above 720   16.6%   25.2%**
**Above 740   6.85%   15.31%**

That’s just incredible – Upgrade has nearly DOUBLE the delinquency rate of Upstart, even with a slightly higher FICO average.

And put another way, UPST’s AI underwriting is yielding 60% fewer delinquencies than Upgrade, and that’s also despite Upgrade having over TWO TIMES more borrowers above 740 FICO mixed in their pool.

I just can’t emphasize this enough - UPST’s AI/ML underwriting model is leagues ahead of all its fintech competitors.