Upstart 2022-1 PreSale Report from KBRA 03/23/22

For anyone that is following Upstart. KBRA published their pre-sale report for tranche 2022-1 dated March 23, 2022 that includes a chart that lists many other securitization from Upstart. That report included a column on page 13 that I hadn’t seen before. “Expected Base Case CNL (for Current Period)”. You can use this new column and compare against the “Current CNL” column. For tranches 2021-ST4 thru 2021-ST10 and 2021-4 & 2021-5 the CNL (cumulative net loss) exceeds what KBRA was projecting at this point in the life cycle of these securitizations. I have confirmed that I can compare these columns in this manner.

From my understanding, I interpret this as a negative. I like Upstart but can’t hold it since this would give reason to believe their AI model is under performing in more recent securitizations and I assume that the market will see that unfavorably given this is one of their paramount benefits. Upstart hasn’t responded to me yet on a request to understand this new information.

I have held shares until recently when I found this information. I will follow Upstart closely because I like what they are doing. I’m hopeful they will be a better solution in the future and I hope to step back in as that unfolds.

I don’t believe this has anything to do with the recent discussions around higher losses in various Upstart past securitizations/tranches. That was centered around how tranches had faired better because of gov’t stimulus and as that came to an end those tranches were experiencing higher losses but still within expectation.

This recent information is based on KBRA’s base case and how CNL has started to diverge now compared to where KBRA expected them to be now. In short, the current CNL’s appear to be worse than KBRA expected. Please go and read the report for yourself. I’m happy to be wrong. I’d like to be a shareholder in this company.

15 Likes

I like Upstart but can’t hold it since this would give reason to believe their AI model is under performing in more recent securitizations and I assume that the market will see that unfavorably given this is one of their paramount benefits.

This seems to keep coming up, despite my having the impression that this has been long since explained that they are intentionally opening the market to more people with lower scores, so one would expect the loss to go up, but that this is more than offset by the increased revenue from granting more loans.

45 Likes

This seems to keep coming up, despite my having the impression that this has been long since explained that they are intentionally opening the market to more people with lower scores, so one would expect the loss to go up, but that this is more than offset by the increased revenue from granting more loans.

Well, I think it would be more accurate to say that the increased loss is offset by higher interest charged on the loans.

I think what CodedFlat is pointing out is that the higher interest rates are determined by the expected CNL rates. If the actual CNL rates are higher than the expected rates, then the interest rates were not properly set.

In other words, the loaner’s income from the loans will not be as high as the loaner’s projected income. This, obviously, is not desirable.

However, I am not convinced that this number, taken in isolation, is that significant. Since, as tamhas suggested, Upstart is also increasing the number of potential borrowers, and the loaners value these new people as potential long-term customers.

9 Likes

Is this a pre-sale report, how can we know actual loss vs. expected loss now?

A piece of context that is missing is what cumulative net loss rate Upstart expected to achieve?

for example:

If upstart expected 15% CNL and the actual CNL is 15% that would be a fantastic result.

If upstart expected 8% CNL and the actual CNL was 12% that would be very bad.

If upstart expected 10% CNL but the actual CNL was 8% that would also be a bad result.

The deviation from the expected value + or - is what makes it bad. Being wrong on the + side means you could have given lower rates, being wrong on the - side ( loosing too much) means you shouldn’t have lent money or have charged a higher rate. Neither scenario are good. Banks can also tune the risk they would like to have ( minimum FICO scores,etc).

6 Likes

The more negative piece of news from that report is that the share of Class E loans is shrinking relative to the other loans.
The class E loans are the more risky borrowers. I thin this means they aren’t lending to as many sub-prime or fair prime borrowers.

A possible reason would be the bank partners don’t want to take that risk, but it does suggest the loss ratio’s aren’t where Upstart expects them to be yet. Upstart will probably spend more of their own capital to develop this further.

1 Like

Link to the April 4 report as none has linked it yet
https://www.kbra.com/documents/report/62309/upstart-securiti…

Like the original post by CodedFlat said, the Expected CNL is a new field that isn’t available in other Upstart reports prior to the March 15 one. Lendingpoint’s New Issue Reports don’t have this field either. I couldn’t find a glossary on this, So I emailed both KBRA and Upstart IR asking about what Expected CNL means.

Here’s the reply from Upstart IR, on the same day:

CNL stands for cumulative net loss. The expected base case CNL is a KBRA estimate based on the risk characteristics of the securitized loans and our past loss performance. Please let me know if there is anything else I can help with.

KBRA did not get back to me.

Str1der’s point in #84075 on both +/- deviation from expected is a room for improvement is a great insight: the goal is to build a machine to assess credit risk -— to price loans —- accurately.

Because Expected CNL metric is computed by KBRA from past Upstart trusts, it was neither a plus or minus for me on Upstart as an investment. If anything, actual CNL in the period being close to Expected CNL aggregated from past Upstart pass through trust’s performances seems to be a good sign? It sounds like the performances of current trusts is trending in the same direction as past trusts. Happy to hear other interpretations.

24 Likes

Other interesting tidbits from the 2022-1 report.

On page 26 they compare Upstart vs lending point loan pools.

Upstart has 30.48% 619 or lower fico scores vs 13.54% for lending point yet the cumulative net loss assumption is only .55% higher for Upstart.

Borrowers in the UPST 2022-1 pool have a weighted average FICO score of 654 and the majority of loans (58.5%) have a FICO score of less than 660.

7 Likes

The more negative piece of news from that report is that the share of Class E loans is shrinking relative to the other loans.

Not true. The dates on the graph run right to left. Actually the Class E have increased from 34% to 39%.

And, as has been pointed out, these are KBRA’s projections, not Upstart’s.

However, I note that the CNL’s that exceeded KBRA’s were for loans originated during the “fraud attack”. The KBRA report states:
in Upstart’s third quarter 10Q for 2021, the Company experienced a large and coordinated fraud attack. The attack did not affect any outstanding securitizations and did not have a significant impact on the Company’s financial results..

Perhaps KBRA is wrong and the increased CNL is a result of the attack.

KC, very long UPST

9 Likes

We might be reading from different reports:
This was the report I was referencing.
https://www.kbra.com/documents/report/65123/upstart-pass-thr… ( Date is April 5th)


From the above report: 
Deal Name         UPSPT 2022-ST3  UPST 2022-1 UPSPT 2022-ST2 USPTT 2022-1A
Transaction Date  4/12/2022*       4/4/2022     3/15/2022       3/10/2022
Total             $83,384,000    $435,480,000  $102,342,000  $476,989,000
Grade E%             34%             39%          41%              42%        
Grade F%              0%             2%           1%               1%  

Notice that 2022-ST3 is the most recent transaction, and 2022-1A is the earliest. So grade E% is actually declining.
(The massive piece of context that is missing is what happened to comparable loans over time. Was there a broader risk-off that is separate from Upstart? )

despite this, the 2022-1A and the 2022-1 tranches are much larger, so the dollar amount of E loans is still high.

In terms of why, I would suspect some of Upstart’s bank partners might want to take on less risk right now given the macro-uncertainty, so that specific securitization (2022-ST3) might specifically be for those bank partners.

Given that there is a much larger tranche before that has a larger percentage of Grade E and F loans, I don’t think it has a lot to do with Upstart’s algorithm.

The 2% of Grade F loans in 2022-1 is also interesting. This an encouraging sign for algorithm improvements.

(Long UPST)

2 Likes

Str1der,

Yes, that is a different report than what I looked at. These are pass-through and the one I referenced was for ABS or a different flavor of ABS. I suspect that we have wrung out all the juice that the board managers will allow on this subject. I will send you a “reply to author” message and if you want, we can discuss all this in more detail. I am very interested in the details of these and have sent a question off to UPST IR.

Thanks for the clarification,

KC, long UPST

Sorry, Str1der, MF is not functioning for me WRT “reply to author”… :frowning:

KC

We are interested in the discussion.

3 Likes

Short term changes can be useful for future trends, but the long-term trend is also important.

Current projections as of February, 2022

                                                                                                            
Loan Group       WTD Average FICO  % 619 and Lower    % 659 or Lower  Initial KBRA Base CNL Projection Current KBRA Base Case CNL projection     Pool Balance  
UPST 2018-1           688             0.01%           19.77%               22.10%                                 13.00%                            $251,701,368
UPST 2018-2           689             -----           19.44%               19.30%                                 12.75%                             $208,470,093
UPST 2019-1           687             0.48%           23.28%               21.10%                                 14.50%                            $257,678,015 
UPST 2019-2           691             0.82%           21.30%               18.25%                                 12.25%                            $421,627,118
UPST 2019-3           690             0.91%           21.98%               17.90%                                 12.25%                            $418,208,138 
UPST 2020-1           688             1.28%           23.43%               16.55%                                 11.50%                            $449,514,017 
UPST 2020-2           687             0.00%           24.37%               17.60%                                 11.00%                             $95,819,850 
UPST 2020-3           691             1.62%           20.98%               17.80%                                 11.00%                            $330,878,704
UPST 2021-1       **678**            4.63%           34.30%               18.85%                                 17.00%                            255,904,982   
UPST 2021-2           674             9.22%           39.01%               17.35%                                 17.00%                            $566,012,815
UPST 2021-3           669            12.94%           42.48%               17.15%                                 17.15%                            $556,685,730 
UPST 2021-4           670            12.68%           41.76%               14.90%                                 14.90%                            $776.587.550
UPST 2021-5           670            14.27%           41.81%               15.15%                                 15.15%                            $603,242,439
UPST 2022-1           654         **30.48%**          58.54%           14.50%-16.50%                         N/A                                    $503,636,059                                           

WOW! Impressive jump from 2021-5 to 2022-1.

The 30% of 619 loans and lower should be good for their training data.

Sources:

2022-1 → 2021-5 Data = https://www.kbra.com/documents/report/62309/upstart-securiti… ( also used the individual KCAT files)
2021-1 → 2019-3 Data = https://www.kbra.com/documents/report/43679/upstart-securiti… ( also used the individual KCAT files)
2019-3 → 2018-1 FICO distribution Data = https://www.kbra.com/documents/report/26768/abs-upstart-secu…

This report contains the 2018-1 to - 2022-1 CNL projections
https://www.kbra.com/documents/report/62309/upstart-securiti…

4 Likes

I’ll replay to multiple posts here.

  1. This seems to keep coming up, despite my having the impression that this has been long since explained that they are intentionally opening the market to more people with lower scores, so one would expect the loss to go up, but that this is more than offset by the increased revenue from granting more loans.

I noted that I wasn’t speaking to this. As I understand it, what I brought up is distinct from this prior, and resolved, concern.

  1. However, I am not convinced that this number, taken in isolation, is that significant. Since, as tamhas suggested, Upstart is also increasing the number of potential borrowers, and the loaners value these new people as potential long-term customers.

KBRA takes all this into account when the project their “expected CNL”. They are giving us this column so we can compare apples to apples.

  1. Is this a pre-sale report, how can we know actual loss vs. expected loss now?

This pre-sale report for tranche 2022-1 includes current data on past tranches.

  1. A piece of context that is missing is what cumulative net loss rate Upstart expected to achieve?

This would be something that Upstart could provide but since KBRA produces this report, and for objectivity, KBRA needs to produce what they expect the loss to be. What gives the report merit is the 3rd party opinion.

  1. Because Expected CNL metric is computed by KBRA from past Upstart trusts, it was neither a plus or minus for me on Upstart as an investment. If anything, actual CNL in the period being close to Expected CNL aggregated from past Upstart pass through trust’s performances seems to be a good sign? It sounds like the performances of current trusts is trending in the same direction as past trusts. Happy to hear other interpretations.

I think the jury is still out on this. KBRA has been giving far more comprehensive data in recent months. Also, we know that Upstart’s tranches over performed because of gov’t stimulus and so I wouldn’t find confidence in looking too far back for data on how these tranches have progressed historically.

It’s been reported via @stockmarketnerd that KBRA started conservative in their loss projection but have loosed up their CNL projections because Upstart finally had enough loan data (think duration of various tranches/securitizations). What KBRA thinks should happen is based on Upstarts historical performance and isn’t “conservative” like in the past. IF, it’s trending above that level now, meaning in more recent tranches, then it’s under performing what they believe is Upstarts model. How much worse can it perform and still be better than a FICO based method? I don’t know but it raises concerns.

I do believe this is something that deserves more understanding because frequently there is a lot of nuance in these details.

  1. I have reviewed the 2022-ST3 Pre-Sale report dated April 5, 2022. It continues to show CNL’s that are elevated compared to what is expected at this point. I’ve taken the last 3 snapshots (02/04/22, 03/23/22 and 04/05/22) and plotted them against each other. I’m only looking at relatively recent tranches because we know that Upstart performed well during the pandemic when the gov’t stimulus helped the tranches over perform. Upstart has noted this several times. So I have focused on the more recent for this reason and others.

In the “ST” tranches everything is doing worse than expected BUT that margin has narrowed. AND, these are new tranches that have months not years of performance so it’s early. It is still between 9%-167% worse than expected. Let me describe what that means. I took the CNL and divided by ECNL. 0.8% / 0.3% = 167%.

The non “ST” tranches have done much better. None of them are doing worse than expected. But their over performance has narrowed just a tad.

In discussing this with others it sounds like we shouldn’t draw hard conclusions yet BUT we should certainly monitor this. The trend is still above expectation and that’s negative. The positive is that the under performance has narrowed between 02/04 and 04/05.

I haven’t taken the time to understand if there is a difference between ST and non ST tranches. If anyone knows that would be potentially insightful.

  1. Also, and I’d give credit to the individual but not sure if they want to be cited, it was told to me that the “two columns…to focus the most are the ‘Initial Base Case Loss Projects’ and ‘Current [KBRA Base Case Loss Projection]’.” Compare those two and keep an eye on them. You don’t want them to increase the loss projection and hopefully decrease it as has happened for multiple tranches.

What the CNL v ECNL shows is an early trend and if it stays elevated, without trending back to expectation, would likely lead to a change in the Current KBRA Base Case Loss Projection. That would be a big negative for at least that tranche and raise questions about why that happened.

I really like Upstart and though I’ve stepped out of the investment for the time being I am watching it closely because it has many characteristics that I like in an investment. A strong mngt team, a huge TAM and hopefully a large moat. If that moat can be confirmed, and assuming the economic backdrop is stable, then I’m back in.

11 Likes

What the CNL v ECNL shows is an early trend and if it stays elevated, without trending back to expectation, would likely lead to a change in the Current KBRA Base Case Loss Projection. That would be a big negative for at least that tranche and raise questions about why that happened.

I am not sure that is the case. KBRA computes CNL based on risk characteristics and past performance of UPST loans. Recently UPST loans have a much lower weighted average FICO suggesting the risk characteristics have increased substantially. But if KBRA is not sufficiently discounting for the loss of stimulus money this year then their projected ECNL will be a lot lower because they would still be basing it on past performance of UPST loans during Covid when there was stimulus money.

The more important question which we don’t know is how well UPST AI models have adjusted for the loss of stimulus money. If they did adjust very well then their internal ECNL will be higher than what KBRA is projecting for the recent loans. This is what the banks and other institutional investors will be going by imo. As long as the banks and other investors are getting a CNL that is lower than this internal ECNL they are making money and should be we willing to lend. That is my read. Happy to be corrected if my interpretation is off.

Long UPST 8%

5 Likes