Four different things in this post: Notes from an UPST auto loan webinar presentation, followed by a macro look of the personal loan market, my notes from the latest Upstart leaders in lending podcast, and a tidbit on discrimination in AI vs human based lending for PPP loans.
1. AUTO LOAN PRESENTATION
I watched an auto loan portfolio webinar presented to credit unions and given by the senior VP of biz dev of UPST (Jeff Keltner). It was uploaded a week ago:
The majority of this presentation didn’t provide all that much newer information compared to a separate past webinar on auto lending; that post is here: https://discussion.fool.com/upst-auto-refi-34917252.aspx
One thing that this newer webinar did have was a little more fleshing out of how Prodigy can benefit the consumer:
Direct lending often results in a poor buying experience. Recall that direct lending is when customers go directly to the bank/branch to get a loan, and then go to the dealership with that financing.
This is because when the customer gets a loan from a bank branch, then walks to the dealer, they might change their mind on the vehicle they were initially approved for and want a different vehicle.
Or they change their mind on the financing terms.
Or they want to be able to update their loan offer during shopping (for example, what if I add extra warranty or purchase servicing in advance and want to bundle it to my loan, what will the monthly payment become? Getting a loan ahead of time from a branch can’t help with this).
Direct lending is not flexible in this manner. Especially when consumers might get a hard credit pull from the bank in advance of going to the dealer.
Prodigy, instead, offers real time financing updates while shopping. (You can play around with it by going to UPST dealer partners like DGDG’s website to get a feel for its flexibility).
Then the webinar moved on to a Q/A session. I’ve paraphrased the answers to the questions as best I can:
How can UPST’s platform find an applicant’s vehicle without the applicant having to enter any vehicle information?
There’s a lot of integration to state level DMV. It can pull the info real time. This is a ‘wow’ moment to consumers that they don’t have to go look up their VIN on their car.
Can you elaborate on how AI can help marketing autorefi for credit unions?
Our marketing AI is predictive. It helps answer the two questions to decide which borrowers to market:
Would I make an offer to them?
Would they take the offer?
We do prescreened offers by mail, digital targeting, direct outreach to your current customers by email [wow, I can’t believe I never realized this before. When UPST partners with a bank/credit union, they can instantly give UPST access to their entire customer base and supply information about each customer…and this can permit UPST to market to them…I think that lowers marketing costs and improves profit margins from these borrowers!]
Also, most of the traditional lending products marketed in a physical branch aren’t actually marketed to the consumer. The consumer usually comes in looking for and asking for the product.
Expanding auto refi is different. You, instead, are actively looking for the consumers to make an offer to, rather than waiting for that small minority that decide to come in branch asking for one. Upstart can help with the right marketing strategies.
When do you do a soft versus hard credit pull?
We do a hard pull with our partner right before they sign the agreement at time of loan closing. Often the consumer will check their rate by a soft pull, leave and come back a week or two later after shopping around for other rates…we can limit risk because we can distinguish who had a drop in credit score due to shopping around vs an actual deterioration of their financial position (and thus shouldn’t be offered a loan).
How does UPST work if we want to use this on our credit union site and make it available to our existing members?
We have the UPST referral network where we market the name of Upstart nationwide and when borrowers come in we present an offer to one of our lending partners that fit their criteria…this is the easy way for most partners to put their deposits to work.
We can also offer the same digitized product via a white label experience totally branded in your name on your website. The fees for this are obviously lower than the referral network, as you are the one driving the consumer demand.
Is the AI/ML model built from a specific financial institution’s portfolio/performance or is the model built from UPST’s data?
Our AI/ML is a uniform machine learning model so that all partners can benefit from and contribute to, in terms of volume of data. Every one of our partners is giving the advantage of over $ 10 billion a month we done in terms of the intelligence of the model because it’s trained from all of that data. The loans you originate are also going back to train the model for all the lenders. It can be configured for each financial institution. It predicts the level of risk for any given offer for any consumer, the chance they pay back per month (either early, late, or not at all paid). This produces a pricing curve and so at the given level of risk, it tells the interest to charge.
[Can someone else confirm what I heard here? Timestamp is 43:09. $10 billion a month? That becomes $120 billion per year of loan data. UPST only helped to originate $2.8 billion in Q2…Is this because UPST is using the ‘past’ data of each bank partner to learn from, and not just UPST underwritten loans?]
Does your model answer why an application is denied?
The explainability of AI matters a lot. We always work with our lenders to explain any decision required. In the context of 1600 variables it can appear difficult to isolate which can explain a decision. We have a robust process that we can go over with you in depth; it’s a whole 45 minute presentation in itself, but we have thought a lot about it and can go over it in detail with anyone separately.
Before we launched any consumer lending product, we first went straight to the CFPB to determine how we should proceed with testing for these, and we have received now two No Action Letters and we have quarterly testing which includes the adverse action reason process.
I see in your slides that you have a lot of bank partners but not a lot of credit unions, what role do credit unions have in auto refi?
We as technologists, we did not first fully understand the outsized roles that credit unions play in the extension of credit to consumers. Although credit unions don’t have the largest assets percent wise in the nation, credit unions have the larger share of assets lending to consumers. Credit unions are the core part of our strategy. I actually have a podcast called Leaders in Lending where we brought on guys from NACFU and CUNA. We’re extremely excited to be partnering with credit unions and I think if you watch that site over the next 2-3 months you will see the dynamic of banks vs credit unions shift a little bit as we have a lot of credit unions partnerships going to go live in that space.
[Now THAT is what I like to hear. Sounds like more credit union partnerships announcements are still soon to come.]
I looked over the earnigns results from banks that reported this week to get a sense of the macro environment for personal loans. What I found:
-Bank of America does not offer personal loans.
-JPMorgan Chase does not offer personal loans.
-Goldman Sachs personal loans (Marcus) appears largely unchanged 0% QoQ and decreased by 25% YoY to $3 billion.
-Wells Fargo personal loan average balance for 3Q21 increased 1% QoQ but decreased 16% YoY to $4.974 billion.
-Both Citigroup and PNC Bank do not break down their amounts of personal loan. They mix personal loans, lines of credit and small business loans within a category of “other”. So I don’t think we can glean much from that measure to isolate their personal lending business, although for Citigroup the entire segment fell 20% YoY.
The takeaway: macro environment continues to show headwinds for the personal lending business, as consumers continue to deleverage - at least, it seems, for the big banks.
Back of the napkin comparison here: Upstart facilitated $2.8 billion of personal loan originations in Q2 2021. Multiply by four and that could be $11 billion originated in a year; even accounting for early payoffs, UPST is definitely facilitating way more in personal loan balances than any one ‘big bank’.
Looking ahead into Q4 for UPST (even though UPST has yet to report Q3 next month) we should also keep in mind that there is seasonality to personal loans, and loan originations tend to be weaker in the Q4 period than other parts of the year. But for Q3, I remain confident that UPST will show they are brushing aside these macro challenges, similar to how they smashed Q2.
3. HOME EQUITY IN UPSTART’S FUTURE?
THis week’s Upstart Leaders In lending podcast was with the Head of Retail Lending at Liberty Bank, discussing the topic of home equity.
My take away points:
Consumer Bankers Association (CBA) is the only member driven trade association focused exclusively on retail banking. Basically, it advocates for the interests of retail banking. CBA has a home equity lending committee, which serves as an advocate for the home equity lending industry. This committee is chaired by a senior vice president/director of consumer lending at Liberty Bank in Connecticut ($7B assets). Upstart is one of many companies that has an associate membership with CBA.
Jeff Keltner, senior VP of biz dev at Upstart, has an established relationship with the chair for the past couple years and had him as a guest on this week’s podcast.
Home equity balances continue to run off and fall in decline, from the 2008-2009 time frame they peaked at 700B and now today it is at around 250B.
Apparently, before the GFC, home equity products were approved in 3 to 5 days instead of 45-60 days today.
[personal comment here: I was in high school during the GFC so I’m shocked to hear HELOCs used to be done in 3-5 days?? I opened a HELOC this past year and wow I was really disappointed with the very lengthy and ridiculous application process ]
Thanks to an overreaction to the GFC, automated valuation models (AVM) have largely fallen into disuse. This means full appraisals are usually demanded before approving a home equity product. The home equity process now mirrors the mortgage process. As a result, while home equity balances have come down since the GFC, personal loans have gone up (especially driven by fintech lenders) thanks to personal loans’ much faster/easier approval processes.
The discussion noted that around 15 years ago, non-bank/fintechs had about 20% of the share of mortgage originations, and today they have over 60% share; this is because they are 100% focused on the quality of the experience, which traditional banks have fallen behind.
It was said in the podcast: “So there’s really I think two things that lenders can do: one, you could either kind of give up on home equity or look at other products like personal loan; or two, you can just look at the home equity experience and try to get back to where we were pre-recession, not necessarily from a credit risk appetite perspective but just from a process perspective.”
To sum it up, home equity is definitely a lending market ripe for ‘disruption’ by Upstart in the future; and we know from recent CEO/CFO podcasts/fireside chats that they have mentioned home lending repeatedly; plus the one job posting mentioning mortgages several weeks back; all of these are strong clues that home lending is the next flavor of credit they are pursuing. Will UPST announce an acquisition (the way they did with auto/Prodigy) to expedite a foothold into the market? We’ll see.
In any case, I believe traditional banks would be very happy to partner with someone like Upstart that can revive this lending segment for them. Although, a future resurgence of home equity may cannibalize/eat into the personal loan market, if consumers begin to switch back to HELOCs/home loans.
4. BIAS AND AUTOMATED LENDING IN PPP LOANS
Score one for AI based lending over traditional human underwriting.
A working paper published on Oct 11 and another published Sep 2, showed that most black borrowers received PPP loans from fintech companies and not traditional banks. “Even after controlling for a firm’s zip code, industry, loan size, PPP approval date, and other characteristics, Black-owned businesses were 12.1 percentage points (70% of the mean) more likely to obtain their PPP loan from a fintech lender than a traditional bank.”
The automated loan processing systems utilized by fintechs as opposed to manual human reviews by traditional banks is to credit for this. It’s really the small banks to blame; many of the better resourced big banks had automated processes and exhibited no racial differences.
One author said “the human brain is a much scarier black box than any machine-learning algorithm. You can constrain an algorithm to meet fair-lending standards, and you can ensure the data it trains on isn’t biased. That may be hard to do, but it’s a clear and objective possibility. Whereas when you have a human loan officer who is in front of someone and making a decision, you can never do that.”
Remember that the PPP program was risk free to the lenders. If the borrower defaulted, the government would still repay the lender. In theory, any lender should have been willing to lend to any qualified applicant. That really is crazy to see racial bias/discrimination show up in a risk free environment!
With UPST’s AI/ML models that have proven to increase minority inclusion in the personal lending world, I’m sure whenever UPST enters small business lending in the far future, they’ll do even better than the low bar already set by small traditional banks. I would think that small banks will see the advantage of partnering with UPST to improve their fairness and avoid getting targeted by regulators.