UPST: 2nd bank drops FICO req?

This post is about my thoughts/analysis gleaned from looking at securitized trust data. There’s two parts to it.

PART 1: Did a second bank partner drop credit score requirements for UPST?


**Year   Number of trusts**
2021      YTD 11
2020      6
2019      3
2018      2
2017      1


**Trust  Amount** 
21-ST9    $83M  Note1 (10/8/21)
21-4      493
21-ST8    80
21-ST7    74
21-3      486
21-ST6    82 
21-ST5    68
21-ST4    102
21-2      476   Note2 (5/11/21)
21-ST3    64    Note3 (4/9/21)
21-1      208
20-3      270
20-2      66
20-ST3    75
20-1      375
20-ST2    110
20-ST1    87
19-3      346
19-2      358
19-1      232
18-2      187
18-1      227
17-2      219
17-1      192

**Year      Securitized Amount ($ millions)**
2021 H1   2133
2020       983
2019       936
2018       414
2017       411

**Year      Loan transacted dollar amount ($ millions)**
2021 H1   4524
2020      3445
2019      2724

Note1
As of the newest UPST securitized trust issue report, dated 10/8/21, I noticed a new change today: the minimum credit score requirement for Finwise bank was decreased.
(Recall that, Finwise is the second largest bank partner of UPST). The KBRA report says in September 2021, the underwriting policy for FinWise originated loans was updated by lowering the minimum credit score from 580 to 300.
Correct me if I’m wrong, but the lowest credit score anyone can get is exactly 300, right? So a minimum requirement of 300 means there is effectively no longer a FICO score barrier!

We know from Q2 earnings call in August, that it was mentioned one bank had dropped their requirement…could it be possible that this is a second bank doing so?
Well, I suppose there is a chance that Finwise is actually that first unnamed bank who dropped their requirement and they only just implemented this new policy in September.

But either way, this is HUGE for 4 reasons:

  1. UPST’s AI model can learn/improve faster now that it’s exposed to more ‘edge cases’. The traditionally considered ‘deep subprime’ (defined as a score below 580) borrowers that typically nobody would ever touch, can now be run through UPST’s model.

  2. Possibly increased ‘network effect’. Other banks might be sold on the idea of getting rid of their FICO requirements if a second bank has dropped that with UPST. (caveat is that Finwise is actually just the first bank doing so, as noted above).

  3. Decreased concentration risk, in the sense that, Finwise clearly likes originating loans with UPST and trusts them a lot, so why would they want to stop partnering with UPST?

  4. Finwise bank already powers a VERY significant chunk of loans (according to the last 10Q, Finwise originates 32% of the total loan transaction volume https://discussion.fool.com/upst-10q-34901669.aspx?sort=postdate…).
    This means a very significant increase in potential revenue via expansion of the addressable market for UPST. And I mean that quite literally: expansion of the actual personal loan market.

Here’s what I mean: if you look at the last most recent LendingClub and Avant trust reports from KBRA, there are ZERO borrowers with a score below 660 for LendingClub, and ZERO borrowers with a score below 580 for Avant. UPST’s fintech competitors are just NOT willing to touch the deep subprime market! There is NO competition for this pool of borrowers. These borrowers are currently a very small fraction of the existing personal loan market as most lenders reject them outright!

Now, according to Experian and FICO, 16% of the total American population has a score between 300-580. That means 52 million Americans.
And according to 2017 data from Oliver Wyman, the 90+ days-past-due rates for score 300 to 570 ranged as low as 44.8% to 19.6% (see table below).
So, let’s just suppose very conservatively, that ~50% of these people ultimately default.
That leaves 26 million “hidden prime” borrowers in the FICO category of deep subprime, all for UPST to tap into! To me, this seems like almost a ‘greenfield opportunity’ that just opened up.

Now, keep in mind, not all deep subprime folks are those who have ‘damaged’ credit, many simply just have a ‘thin’ file with little credit experience.

Finally, we should distinguish that having no minimum FICO score requirement is different than a scenario of a borrower who has NO credit score at all (there are about another 26 million Americans in this category). Those fall under different requirements (see Note2 and Note3 below)


FICO       90+ DPD rate
571-590    15.5%
551-570    19.6%
530-550    23.7%
511-530    29.0%
491-510    33.4%
471-490    37.4%
451-470    39.9%
300-450    44.8%

(sources: https://www.oliverwyman.com/content/dam/oliver-wyman/v2/publ… and https://www.fico.com/sites/default/files/inline-images/Scree…)

Note2 and Note3
The 2021-2 UPST trust (report issued May 11, 2021) and 2021-ST2 UPST trust (report issued April 9, 2021) both showed a new change in requirements by Cross RIver Bank.
CRB lowered its minimum credit score requirement from 620 to 600 in February 2021.
CRB also raised the maximum allowed Debt to Income ratio to 50% from 45%, and loosened the ‘no bankruptcies permitted within the last 12 months’ rule to within the last 3 years.

I think part of the reason why criteria is being relaxed by CRB, is they are also seeing the positive results of their retained loans and are trusting UPST’s AI models more and more over time.

(For Finwise, ever since becoming a bank partner in 2019, they never had debt to income ratio maximums, bankkruptcy or max inquiry requirements. Finwise appears to have been fully on-board with trusting the AI models way more than CRB!)

PART 2: Can we predict Q3 results with knowledge of UPST trust data?

I’ve been following the UPST securitized trust reports from KBRA closely. Since Q2 earnings I’ve tried to see if I could use the KBRA reports to ‘predict’ how business is performing.

Unfortunately, although we see the volume of trusts and size of trusts have grown considerably (see tables above) over time, I’ve realized it’s not possible to use this information to forecast Q3 performance.

This is for a number of reasons, such as variable dates of loan statistical cutoffs, the average weighted months seasoned of loans included in each trust is always 4 months (which will always overlap two quarters) and also the fact that if you do the math, not all loans are securitized. There is a huge chunk that are going to whole loan sales, the comprehensive data of which is not possible as a layperson.
(Whole loans sales are when institutional portfolio managers will buy up entire borrower loans from UPST. In this way, the borrower loan was not packaged into a securitized trust where thousands of other similar loans are pooled together.)

Here’s the boring math, if anyone cares to read:

If you read the 10Q reports, you find that "in the year ended December 31, 2020, 21% of the loans funded through our platform were retained by the originating bank and 77% of loans were purchased by institutional investors through our loan funding programs.
Our institutional investors and buyers that participate in our loan funding programs invest in Upstart-powered loans through whole loan purchases, purchases of pass-through certificates and investments in asset-backed securitizations."

The 10Q filings show $3.445B was originated in 2020, and $4.524B in 2021 H1.
The filings also show 2.652B and 3.414B respectively were purchased for resale (these are the loans destined for whole loan sales or securitizations/pass through trusts).
We also know 69.9M and 42.55M were held on UPST’s balance sheet for research per the 10Q.

So you can deduce 723M and 1.067B retained by banks respectively (this matches the 21% retained rate in 2020 stated above, and yields 23% bank retention in 2021 H1). So at least we know the trend is more and more banks are retaining the loans directly, rather than selling back to UPST for immediate securitization)

Using this information, if you subtract from the 2020 total dollar amount of trusts (1.191B) it means in 2020, about 1.461B went to whole loan sales.

It’s hard to know what that whole loan sale figure is for 2021, due to the various loan cutoff dates for the trusts.
As such, we can’t really estimate any meaningful information to help us predict Q3 results.
All we know is that securitization amounts and number of trusts has accelerated in 2021 versus 2020, which is a good thing of course, indicating higher and higher institutional demand (see tables above)

Also, shoutout to a person on twitter who messaged me with Bloomberg terminal info that the UPST consumer loan trusts have also had upsizing in offerings increase over time, thanks to rise in demand (see below)

PRICED: Upstart $671.36M Consumer Loan ABS
2021-09-17 17:22:24.925 GMT
(Bloomberg) – Increased to $671.36m from $493.456m.

Versus, from earlier in the year the upsizing of offerings were smaller:

“PRICED: Upstart $475.9M Consumer Loan ABS via Goldman; Upsized
2021-05-04 20:09:12.316 GMT
(Bloomberg) – Increased to $475.875m from $412.011m.

112 Likes

Oops, forgot to write in a part for the Note2, and Note3 section of the post.

For 2021-ST2 UPST trust (report issued April 9, 2021), Cross River Bank also removed criteria for borrowers who had no credit score at all. Prior to that securitized trust, CRB would only consider ‘no credit score borrowers’ if they had at least an associate degree or higher level of education. But since then, that’s not necessary any longer.

The pool of folks who have no credit score at all are typically the young generation in college or fresh out of college/higher education who’ve never accessed credit. There’s at least 26 million Americans in this category. And there’s another 19 million Americans who have credit reports that are so limited or out of date that they are unscorable. So that’s total 45 million Americans with no scores at all.

And not only that, but Blacks and Hispanics have higher rates of credit invisibility or unscoreable reports. About 15 percent of Blacks and Hispanics are credit
invisible (compared to 9 percent of Whites and Asians) and an additional 13 percent of
Blacks and 12 percent of Hispanics have unscored records (compared to 7 percent of
Whites).
The source is 2021 data from the CFPB: https://files.consumerfinance.gov/f/201505_cfpb_data-point-c…

Now, what’s great about UPST, is that they have the first ever Spanish language lending platform, and has already proven to the CFPB in its No Action Letter that they approve more minorities than a traditional underwriting platform!

See UPST’s press release: https://ir.upstart.com/news-releases/news-release-details/up…
"In 2020, Upstart’s model approved 27.2% more Hispanic borrowers than a traditional model with 10.5% lower interest rates, despite offering its service only in English.

See also an UPST blog post: https://www.upstart.com/blog/why-i-joined-upstart
“2019 results showed that Upstart’s model increased approval rates for African-American applicants by more than 45% with 21% lower APRs compared to a traditional credit model. That’s a big win for responsible financial inclusion.”

66 Likes

Sorry for another correction. Received a few emails now mentioning the typo mistake I made:

CRB also raised the maximum allowed Debt to Income ratio to 50% from 45%, and loosened the ‘no bankruptcies permitted within the last 12 months’ rule to within the last 3 years.

Should actually be:

CRB also raised the maximum allowed Debt to Income ratio to 50% from 45%, and loosened the 'no bankruptcies permitted within the last 3 years’ rule to ‘within the last 12 months’.

Screenshot of the source: https://ibb.co/Ht0mchF

10 Likes

I went back to Q2’s earnings call to find that comment about the first bank partner dropping FICO score requirements. I thought the full quote was interesting enough to add to the discussion’s backdrop:

“For the first time, one of our bank partners decided to eliminate any minimum FICO requirement for their borrowers. To us, this demonstrates both a commitment on behalf of this bank to a more inclusive lending program, as well as an increasing confidence in Upstart’s AI-powered model. While credit scores can be useful, hard cutoffs based on a three-digit number invented 30 years ago leaves far too many creditworthy Americans out in the cold. We’re hopeful a second bank partner will make a similar decision in the near future.

Obviously this isn’t conclusive evidence that Finwise was specifically the first or second bank partner. But it does seem very possible that a second bank partner has already joined the ranks by now, given both these comments and your research. Any further opinion from me would just be pure speculation though.

16 Likes

jonwayne,

Thanks for all your detailed posts about UPST. I am a few days behind in reading obviously. In your post of Oct-9, you wrote that CRB has loosened the rule about “no bankruptcies” from 3 years to 1 year.

That one worried me a bit. Wouldn’t declaring bankruptcy recently be a relatively strong indicator that we don’t want to loan money to that person?

But then I thought deeper. Perhaps the issue here is simply predictability. It doesn’t much matter if a borrower is a “good borrower” or “bad borrower”, it is about prediction accuracy. If we know rate of defaults of a certain group, and we can price the interest rates properly across the group, then its about enlarging the number of borrowers to the maximum, each priced accordingly, and not weeding out “bad borrowers”. I never thought about it that way before.

From that perspective, if you would agree with it, it sounds a lot like insurance.

Thank you for sharing.

Rob

2 Likes

“From that perspective, if you would agree with it, it sounds a lot like insurance.”

And from that perspective…

A lot of talk seems to focus a lot on UPST eventually moving into mortgages…

That doesn’t seem to me to be the best use of UPST’s data… and a direction I’m not sure we’d want them to go anyway. Giving large mortgages to even slightly less stable borrowers worries me that an economic downturn could be another Countrywide and financial crisis…

But where UPST’s AI risk modeling seems to make a lot more sense to me would be first moving to life insurance… with the thousands of data points UPST can gather about a person, seems they’d be better able to price life insurance… and then maybe next into car insurance… seems the same thoughts would apply… when you know everything about a person, you could more accurately price their risk driving a car. Seems like AI could shake up the insurance industry the same way it will shake up the banking industry… and would align better with the founders goals than mortgages would…

Maybe partner up with some insurers and wrap insurance into the prodigy software… new car buyers could walk out of the dealership with a new car and cheaper insurance than their old insurer!

Any thoughts from all you folks smarter than me???

4 Likes

In your post of Oct-9, you wrote that CRB has loosened the rule about “no bankruptcies” from 3 years to 1 year.
That one worried me a bit. Wouldn’t declaring bankruptcy recently be a relatively strong indicator that we don’t want to loan money to that person?

Hi Rob, thanks for your post reply. Well, I believe the whole point is to trust UPST’s AI/ML.

I think you answered that question yourself here!

If we know rate of defaults of a certain group, and we can price the interest rates properly across the group, then its about enlarging the number of borrowers to the maximum, each priced accordingly, and not weeding out “bad borrowers”. I never thought about it that way before.

I agree. If the models are learning from which borrower who had a previous bankruptcy will repay, and who doesn’t, then it will only learn to be smarter in the future. It needs to ingest that data or it will never know - the same way Upstart’s AI/ML models need to learn how to underwrite deep subprime borrowers, those who have credit scores below 580. Almost every other fintech lender and big bank avoid the deep subprime category. Upstart has to be given a chance to be fed that data, or it’ll never expand that specific market.

10 Likes

So I came across this. An official Nerdwallet review article on Upstart that was last updated/written on August 26, 2021.

https://www.nerdwallet.com/reviews/loans/personal-loans/upst…

How to qualify for an Upstart loan
Minimum credit score: None.
Minimum credit history: None; borrowers with credit histories too limited to produce a FICO score may be accepted.
Minimum gross income: $12,000.
Employment: Full-time job, full-time job offer starting in six months, a regular part-time job or another source of regular income.
Must have a U.S. address where the borrower resides (unless military personnel on active duty).
Must be at least 18 years old.
Valid email account required.

The fine print of these reviews:

NerdWallet’s review process evaluates and rates personal loan products from more than 30 lenders. We collect over 45 data points from each lender, interview company representatives and compare the lender with others that seek the same customer or offer a similar personal loan product. NerdWallet writers and editors conduct a full fact check and update annually, but also make updates throughout the year as necessary.

And recall that the KBRA report said:

“The Upstart Program currently offers three, five, and seven year monthly amortizing unsecured installment loans to borrowers who generally have a minimum credit score (FICO or VantageScore) of 600 or no credit score; however, Loans originated through FinWise may, as of September 2021, have a minimum credit score of 300.

So if the review author discovered that Upstart no longer had a min FICO score before September…then could this help ‘confirm’ that Finwise is indeed the second bank to drop a FICO minimum score requirement?

19 Likes

My Humble Submission

$SOFI

We are excited to leverage SoFi’s sophisticated tech platform, strong brand, and consumer appeal to originate loans through Pagaya’s AI network, extending its business to a broader audience, so more people can access credit and achieve their financial goals," said Anthony Noto, CEO of SoFi.

THIS IS IT. THIS CHANGES EVERYTHING. This is going to give competition to UPST

I reserve the right to be wrong. I am submitting this without any bias just that many of you are sitting on huge profit in UPST and this might give some competition in the very long term.

I can be wrong in my assessment.

3 Likes

humble submission with some hyperbole…

I actually first brought the board attention to Pagaya when they announced their SPAC intentions. Their deck was full of Upstart references, essentially listing them as a competitor. Upon further review, they are quite different.

Does this mean it is not a competitor at all? Probably not. As many have mentioned, one of the main risks with Upstart is that the others are watching how lucrative their space is and will definitely want to get a piece of the pie. SoFi also sees this potential.

1 Like