Upstart (UPST) Q2 2023 Earnings Call Transcript

Upstart (UPST 2.76%)
Q2 2023 Earnings Call
Aug 08, 2023, 4:30 p.m. ET

Contents:

• Prepared Remarks
• Questions and Answers
• Call Participants

Prepared Remarks:

Operator

Good day, and welcome to the Upstart second quarter 2023 earnings call. Today's conference is being recorded. At this time, I would like to turn the conference over to Jason Schmidt, vice president of investor relations. Please go ahead.

Jason Schmidt -- Vice President, Investor Relations

Good afternoon, and thank you for joining us on today's conference call to discuss Upstart's second quarter 2023 financial results. With us on today's call are Dave Girouard, Upstart's chief executive officer; and Sanjay Datta, our chief financial officer. Before we begin, I want to remind you that shortly after the market closed today, Upstart issued a press release announcing its second quarter 2023 financial results and published an investor relations presentation. Both are available on our investor relations website ir.upstart.com.

During the call, we will make forward-looking statements, such as guidance for the third quarter of 2023 relating to our business and our plans to expand our platform in the future. These statements are based on our current expectations and information available as of today and are subject to a variety of risks, uncertainties and assumptions. Actual results may differ materially as a result of various risk factors that have been described in our filings with the SEC. As a result, we caution you against placing undue reliance on these forward-looking statements.

We assume no obligation to update any forward-looking statements as a result of new information or future events, except as required by law. In addition, during today's call, unless otherwise stated, references to our results are provided as non-GAAP financial measures and are reconciled against our GAAP results, which can be found in the earnings release and supplemental tables. To ensure that we can address as many analyst's questions as possible during the call, we request that you please limit yourself to one initial question and one follow-up. Later this quarter, Upstart will be participating in the Goldman Sachs Communacopia + Technology Conference, September 7; and the Piper Sandler Growth Frontiers Conference, September 12.

Now, I'd like to turn it over to Dave Girouard, CEO of Upstart.

Dave Girouard -- Chief Executive Officer

Good afternoon, everyone. Thank you for joining us on our earnings call covering our second quarter 2023 results. I'm Dave Girouard, co-founder and CEO of Upstart. I told you last quarter that I was hopeful Q1 was transitional one for Upstart, and I continue to believe that's the case.

I'm pleased we delivered quarter-on-quarter growth in Q2 for the first time in more than a year, and more importantly, we achieved record high contribution margin and positive cash flow, a result of our efforts over the past year to improve efficiency and operating leverage in our business. This is despite in an environment where banks continue to be super cautious about lending. Interest rates are as high as they've been in decades and capital markets remain challenged. A close look at our financials in Q2 suggests that Upstart has the opportunity to grow quickly and profitably when we return to a normalized economy.

I'm also pleased to see clear signs that inflation is ebbing despite a continued strong labor market. Our lens on inflation is different from that of other. From our point of view, wage growth in excess of goods inflation is a new and positive development, particularly for the less affluent segments of the U.S. that we tend to serve.

The market is increasingly optimistic that the Fed can achieve their 2% inflation target without a serious recession. While the recession remains a possibility, our view is it's likely to be a shallow white collar recession, one less likely to result in significant unemployment for less affluent Americans. And unlike 18 months ago, the Fed now has readily available tools to handle a significant slowing of the economy. They can lower rates to spur growth once again.

We continue to be confident that our core personal loan risk models are properly calibrated and have been so since November of 2022. Thus, we expect these recent vintages to deliver at or above target returns. Funding markets remain cautious and risk averse. Banks and credit unions are generally focused on deposits and liquidity, while capital markets are beginning to show signs of normalization.

We added another committed capital partner in July and are in conversations with several more interested parties. We also completed a securitization after the close of Q2 with significantly tighter spreads than our prior deal earlier in the year. Meanwhile, we continue to manage Upstart cautiously, but optimistically in a funding-constrained environment. Every week I remind the Upstart team to focus maniacally on improving every aspect of our business, strengthening our company for a time when the markets will inevitably return to center.

As I said to you last quarter, I focus our team's energy on improving Upstart in four key dimensions. First, best rates for all. The core thesis of Upstart is that superior AI-enabled risk models will improve access to credit for all. And the company that can build superior risk models faster than anyone else stands to benefit from this dramatic transformation of the lending industry.

In this light, I want to share an exciting breakthrough, something we call parallel timing curve calibration. This is a technique aimed at accelerating the pace of model calibration and thus model development. The challenge with launching a new model in lending is that you have to wait many months to see how it performs in the real world. If you're originating three-year loans, then you need to originate some loans and then watch them perform for 36 months to have clear feedback on model calibration across the timing curve.

But with parallel timing curve calibration, the new model can be used to reunderwrite all in-process loans from the past, generating new predictions for how they will perform in their remaining months. This is not a back test. The new model is used to predict how all outstanding loans in the platform will perform in coming months, not how they performed to date. In this way, within a few months, you can have a clear signal as to calibration across all months in the timing curve.

This results in a dramatically faster calibration process for new models. From the point of view of our lending partners and credit investors, this is a giant win, because it means we provide a tighter and faster feedback loop regarding model performance. We're very excited about its potential to extend our leadership in AI lending and are in the process of patenting this technique. Next, more efficient borrowing and lending.

Last quarter, we reached an all-time high of 88% of unsecured loans fully automated. That means instant and automated approval with no waiting, no documents to upload, and no phone calls. This matters a lot because even the most accurate loan pricing model is useless if applying for a loan is too time consuming or laborious. The notion of building an entirely software-driven credit origination process, one that can run 24/7 in a fully lights-out environment has been with me since Upstart's founding and was inspired by my years at Google.

Please bear with me as I share a short story. In 2003, I was interviewing for a role at Google. The company was still private at the time, so the world knew little about the financial giants that was growing in Mountain View. At one point in the process, the external recruiter said that I couldn't interview that week at Google because the entire company was skiing at Lake Tahoe.

I thought to myself, that's a great company. Then she said, but get this, the company is still making $7 million or $8 million a day in revenue. I thought to myself, no, that's an amazing company. And the idea has stuck with me since.

So, how have we done? In 2016, we began to pursue the goal of fully automated loans, zero human involvement from rate request to transfer of funds, approved in a matter of seconds, lights out. By Q2 2017, 29% of our unsecured loans were fully automated; in Q2 2019, it was 64%; in Q2 2021, it was 69%; and in Q2 2023, this past quarter, it was 88%. Automation doesn't just allow us to scale originations faster than headcount, it creates a wow moment for consumers who have never experienced such a fast and effortless loan application process. Recently, we began to brand this Fast Track, something we should probably have done years ago.

This breakthrough experience is a signature of Upstart and the lenders that we serve. Next, more resilient. In addition to ongoing initiatives to strengthen the funding side of our marketplace, we continue to optimize our fixed costs. This increases the leverage in our business so that Upstart can thrive across future economic cycles.

In Q2, we identified another $7 million in annual technical infrastructure costs that we can eliminate, bringing our total annual cost savings in tech expenses to nearly $17 million. We continue to hire very modestly and only in strategic situations. We also achieved a contribution margin of 67%, our best ever by a long shot. This is a sure sign that our focus on efficiency is bearing fruit.

A principal driver of this record contribution margin was our efforts to build a stronger relationship with our existing customers. As a result of these efforts, 38% of our originations in Q2 came from repeat borrowers, also a record for us. And as a consequence of that, we also saw record low acquisition cost per loan in Q2. Next, expanding our footprint.

We continue to make progress in our newer products and are excited to see the progress we'll make through the rest of 2023. In the second quarter, we made significant strides in our auto retail lending business. We expanded our footprint from 39 rooftops with Upstart lending implemented last quarter to 61 rooftops today. We also added 12 additional states we now support, covering more than 65% of the U.S.

population. We launched new risk models for both our auto refinance and retail lending products, delivering as much accuracy improvement as we've seen in the last year from our personal loan models. We continue to improve our auto recovery performance, reducing delays in recovery cycle by 75%. On the feature side, we launched a new device agnostic in-store application that expands access to desktops, laptops and tablet browsers.

We also brought on our second and third lending partners for auto retail, no easy feat given the current market environment. We're also making rapid progress on our small dollar relief loans. These loans start at just a few hundred dollars and are currently offered only to Upstart applicants who don't qualify for our mainstream personal loans. For this reason, they're entirely incremental to both our approval rates and our model training set.

Our first vintages have now fully reached maturity and our model is now fully calibrated with observed losses in line with expectations for our most recent model version. In a first for Upstart, we began using cash flow data as part of the risk model for small dollar loans. This incremental data has led to increased approval rates and will eventually become available for all our loan products. Our fully automated rate in Q2 for small dollar loans was 90%, an incredible achievement that demonstrates the power and impact of AI in lending.

In Q3, we'll move beyond offering this product exclusively to those declined for personal loans and will finally enable direct consumer applications. Last but not least, I'm happy to let you know that our home equity product is officially off the ground, with the pilot program in the state of Colorado. We expect a fast follow with the state of Michigan and also hope to be in a handful of additional states by the end of Q3. This is the first Upstart product specifically designed for prime borrowers, where a superior process enabled by automation is a richer source of differentiation than loan pricing itself.

As a reminder, I mentioned last quarter that we're targeting online approval in less than 10 minutes and a closing process of less than five days for an Upstart powered HELOC against an industry average closing time of more than a month. To wrap up, we're not yet certain the economy is headed to a better place. So, we continue to be cautious while investing for the long term. And you're now beginning to see the benefits of our disciplined approach.

Regardless of the economy's direction in the coming months, I'm confident that we're building a better, stronger enterprise for the future. We're in the pole position to lead the industry to an AI-enabled future, one that represents a giant leap forward for both borrowers and lenders. And we do this not because of fascination with AI, but because of what brought us here, the potential to dramatically improve access to credit for tens or even hundreds of millions of Americans. There are many dimensions along which you can weigh our efforts to make Upstart stronger: speed of model development, strength of unit economics, low fixed costs, demonstrable leverage in our business, improved funding supply, and growing product diversity.

But the dimension that gives me confidence more than any of these is talent density at both the executive and individual contributor level. For those excited about AI and passionate about its potential to improve lives, we know of no better place than Upstart to build a career. And while we're hiring strategically and with extreme caution, our digital-first approach is enabling us to hire top talent across the country. More than 90% of job candidates have accepted our offers in recent months, an incredible success rate.

While much of the world is debating how to return to the office full time, we're very happy with the results of our digital-first approach. I will close with a huge thank you to all Upstarters, as well as the family and the friends that support them. We're on an incredible mission together and it wouldn't be possible without each of you. Thank you.

And now I would like to turn it over to Sanjay, our chief financial officer, to walk through our Q2 2023 financial results and guidance. Sanjay?

Sanjay Datta -- Chief Financial Officer

Thanks, Dave, and thanks to all of you for joining us today. We're pleased with our return to sequential growth and EBITDA profitability this past quarter, as our work to unlock committed funding, rationalize our fixed cost base and expand margins begins to bear fruit. We accomplished these objectives despite ongoing macro challenges for our lending partners, and despite a U.S. borrower whose recovery from the stimulus-driven effects of 2021 and 2022 has yet to fully materialize.

Our best measure of borrower delinquency trends, the Upstart Macro Index has tread water over the past few months, and in fact, even seen a more recent seasonal increase versus earlier '22 and '23 levels, as we came off of the favorable tax refund seasonality that ran through April. Despite a continuing recovery in the disposable income, stemming from the ever strengthening labor market, any incremental earnings over the past quarter has been directed almost entirely toward higher consumption, which has continued to increase in lockstep. And consumer balance sheets have not benefited from incremental savings as they had earlier in the year. Despite these latest dynamics, we believe our underwriting models remain well calibrated to this environment, and we are expecting our vintages since late 2022 to deliver or exceed their targets.

On the funding side of the ecosystem, banks remain conservative in managing the asset side of their balance sheets, generally seeking to rationalize loan positions and conserve cash in aggregate. The ongoing supply of loans on offer in the secondary markets by sellers anxious for liquidity contributes to a challenging market dynamic. With loan books being sold at bargain prices, and creating no shortage of buying opportunities for selective investors. Our view is that it will take some time for the market to work its way through this surplus of cheap available yield.

Despite this, we continue to pursue a number of promising discussions with prospective funding partners, aimed at bringing more committed capital to the platform and believe that we will be well positioned once the loan market returns to a more traditional state of pricing equilibrium. With these items as context, here are some financial highlights from the second quarter of 2023. Revenue from fees was $144 million in Q2, comfortably above our guided expectation of $130 million, aided by a beneficial mix shift toward institutional funding, as well as ongoing take rate optimization. Net interest income was negative $8 million, largely owing to higher-than-expected discount rates and unrealized fair value adjustments on some existing assets, as well as the impact of rising borrower charge-offs, particularly in our legacy R&D portfolio.

Taken together, net revenue for Q2 came in at $136 million, slightly above guidance and representing a 40% contraction year over year. The volume of loan transactions across our platform in Q2 was approximately 109,000 loans, up roughly 30% sequentially and representing over 43,000 new borrowers. Average loan size of $11,000 was up 4% versus the same period last year, but down sequentially due to growth in small dollar loans. Our contribution margin, a non-GAAP metric, which we define as revenue from fees minus variable costs for borrower acquisition, verification and servicing, as a percentage of revenue from fees came in at 67% in Q2, up 20 percentage points from 47% last year and 7 percentage points above our guided expectation for the quarter.

Continued investment in loan processing automation and fraud models have led to a new high in fully automated rates at 87%, bringing down loan onboarding costs and improving the conversion efficiency of our marketing costs. Higher numbers of repeat borrowers have similarly improved our overall cost per acquisition. And a mix shift toward institutional funding has benefited our take rates. Taken together, our contribution margins are stronger than they have ever been.

Operating expenses were $169 million in Q2, down 35% year over year and 28% sequentially, as workforce restructuring initiatives announced in Q1 are now translating into reduced operating burden. In addition, as Dave alluded to, we have done a significant amount of work to improve the efficiency and decrease the overall expense of our technical infrastructure, which represents a large portion of our fixed cost base. Declines in other categories, such as sales and marketing and consumer -- customer operations, were largely in line with the decline in the loan volumes that drives them. Altogether, Q2 GAAP net loss was $28.2 million and adjusted EBITDA was positive $11 million, both comfortably ahead of guidance.

Adjusted earnings per share was $0.06 based on a diluted weighted average share count of 91.0 million. We ended the quarter with loans on our balance sheet of $838 million, down sequentially from $982 million the prior quarter. Of that amount, loans made for the purposes of R&D principally within the auto segment represented $493 million of the total. Just after quarter close, we completed a one-off $200 million ABS transaction, funded entirely from our own balance sheet.

As you may recall, we traditionally sponsor ABS transactions on behalf of our loan buyers who are usually the principal economic agents and loan contributors to the transaction. In this case, we took the unusual step of funding a deal from the Q2 vintages accumulated entirely on our own balance sheet. We did this both to reset the market understanding for how our more recent vintages should be expected to perform, as well as to serve as a visible signal to the market of our confidence in the adjustments that have been made by our own underwriting models and adapting to the new environment. Our corporate liquidity position at the end of Q2 remains strong with $510 million of total cash on the balance sheet, and approximately $558 million in net loan equity at fair value.

Looking ahead, while there remain good reasons to be optimistic about the general longer-term direction of the U.S. consumer, in the short term, we remain circumspect about the timing of the recovery of borrower delinquency trends and the recovering health of the funding markets more broadly. In particular, until we see a definitive inflection and reversal in the trajectory of UMI, we will continue to err on the side of being very conservative in our assessment and pricing of borrowers. With this context in mind, for Q3 of 2023, we expect total revenues of approximately $140 million, consisting of revenue from fees of $150 million and net interest income of approximately negative $10 million; contribution margin of approximately 65%; net income of approximately negative $38 million; adjusted net income of approximately negative $2 million; adjusted EBITDA of approximately $5 million; and a diluted weighted average share count of approximately 84.5 million shares.

Notwithstanding the promising direction this past quarter, there is still much work to be done to restore our business to the scale and growth that we aspire to. We have made encouraging recent strides in execution, operational discipline, technological innovation and deal making. And while we await emergence from the combined jetwash of the funding macro, and the borrower delinquency trends that are running their course, we will continue to push for further progress in all of these areas. When we are finally clear of the environmental turmoil around us, we are convinced that our business will be as formidable as ever.

Thanks to all of the teams at Upstart, who continue to execute ahead of expectations. This has obviously not been an easy past few quarters, but I am confident that we are pointed in the right direction and that we have the right people in place to seize the once-in-a-generation opportunity that remains before us. Or as we like to say internally, "We are under the Maple Tree." With that, Dave and I are happy to open the call to any questions. Operator?

Questions & Answers:

Operator

Thank you. [Operator instructions] We will take our first question from Simon Clinch with Atlantic Equities. Please go ahead.

Simon Clinch -- Atlantic Equities -- Analyst

Hi, guys. Thanks for taking my question. I appreciate it. I'm really interested in, I guess, what levers you -- or perhaps how you respond when you start to see the UMI flatten or improve? What are the levers that you would be able to pull? And how should we think therefore about the speed at which you can start to get a recovery in your conversion rates and your general loan growth and your market share gains as a result of that?

Dave Girouard -- Chief Executive Officer

Thanks, Simon. This is Dave. Basically, we are watching UMI trends and effectively maintaining what we hope to be a buffer between the assumptions that are in a new loan model or in the loans that are being produced in any particular time to the trend in UMI. So, as we will hopefully see over time UMI trending back down, at some point, the underlying assumptions in the models will stick with that.

And again, maintaining -- always aiming to maintain a buffer, but it should trend down. So, UMI, as you see it really is the output of what's going on out there in the economy and we always want to stay ahead of it. And the best thing we can really do is make sure UMI is as accurate and as recent as possible, that's why we're sort of continuing to innovate on it. But it will be a good indication of when the assumptions that will go into the next loans that are being originated.

Simon Clinch -- Atlantic Equities -- Analyst

OK. I understand. And just as a follow-up then, I was wondering, maybe, Sanjay, you could talk through the slide on the long-term funding commitments and your share of the risk and just make sure we understand sort of what's going on that -- in that particular situation as an example.

Sanjay Datta -- Chief Financial Officer

Yeah. Hey, Simon. This is Sanjay. Happy to do it.

Yeah, so we have a new slide in our materials that's meant to sort of pull together the punch line for, essentially, what we have at risk as part of our committed capital partnerships. They're a little bit hard to pull together in the financials directly because the contracts are all a little bit different and they're all accounted for a little bit differently. But I guess at a headline level, as part of those deals, we have invested or co-invested to date on the order of $40 million. So that, I guess, you can think of as the maximum exposure we have.

That $40 million over time will be worth as little as zero or as much as around $80 million or $83 million, I guess, depending on the timing extent to which these loans over or underperform. Our best current estimate given the trends is that that $40 million, we believe, is on track to be worth about $52 million. So, some marginal over performance there. But that's something that we will obviously forecast and track for you guys over time.

Simon Clinch -- Atlantic Equities -- Analyst

OK. Thanks, guys.

Operator

We will take our next question from Lance Jessurun with BTIG. Please go ahead.

Lance Jessurun -- BTIG -- Analyst

Hey, thanks for taking my question, guys. First one is just around kind of the governor on your growth right now, which is at 36% APR. Any commentary or color around how many you kind of have to turn back right now? Because the risk model is turning them above 36%. And how should we kind of think about that going forward in the coincidences of where you see Fed funds going?

Sanjay Datta -- Chief Financial Officer

Hey, Lance. Let's see. So, I guess that 36% APR cap, with the current levels of UMI is a significant constraint on our business. In fact, I think, as we stand today, it's probably the bigger constraint on growth on our platform.

It's very punitive, from the perspective of our approval rates, which are quite low right now. How to think about this evolving going forward? Well, the 36% APR cap will not change. That's something that is sort of sacred to us. As UMI declines, sort of as Simon said in his question, as our observed -- the observation of UMI declines over time because presumably the macroeconomies coming back into equilibrium.

We will accordingly reduce the forward assumptions being fed to the model, and that will essentially reflect lower loss estimates over time. And that will pull borrowers who were currently excluded from our credit box, they'll pull them back into the approval universe. So that's the mechanism by which we would expect to grow as loss rates in the economy subside and as UMI comes back down.

Lance Jessurun -- BTIG -- Analyst

Got it. Thank you. And then, in terms of July data, I know there's been some alternative data sources out there, but anything you can kind of frame around July data and where it's come in and how do you kind of see that in terms of a quarter -- or in terms of a monthly run rate through August and September?

Sanjay Datta -- Chief Financial Officer

So, you're asking about July data, Lance?

Lance Jessurun -- BTIG -- Analyst

Yes.

Sanjay Datta -- Chief Financial Officer

I don't -- we're not in a -- really in a position to comment at all on July numbers. Hopefully, we'll have a good report for you when we talk about Q3.

Lance Jessurun -- BTIG -- Analyst

Got it. Thanks.

Operator

We will take our next question from Ramsey El-Assal with Barclays. Please go ahead.

John Coffey -- Barclays -- Analyst

Hi, thanks. This is John Coffey on for Ramsey. I was wondering, Sanjay, if you could tell me a little bit more about the loan balances you have on the balance sheet. I know some came off in Q2 I think from some of your new partners.

And clearly, those got built up a little bit again. Could you just maybe talk briefly about how we should think about the cadence of those loans in the balance sheet for the remainder of the year? And if that like $1 billion mark is still sort of like maybe sort of informal high watermark that you won't go beyond or have you rethought that a little bit?

Sanjay Datta -- Chief Financial Officer

Hey, John. I guess, a specific comment and a general comment. The comment specific to Q2 would be, yeah, it is true that while the net loan balance from Q1 to Q2 went down, we sort of went down further and then built it back up. One of the things we were doing very deliberately was to build up some very, very fresh collateral to be able to put into an ABS deal.

And that ABS deal, as I said in my remarks, happened after the end of the quarter. So those loans sort of sit on our balance sheet at the end of the quarter and shortly after we put them into an ABS deal. It's a bit unusual for us to run an ABS deal off of our own balance sheet, as I said, but there's a couple of important reasons for why we wanted to do it. We wanted to put fresh collateral into the securitization reporting, so that people could see how the models have changed and how they might expect freshly originated loans to perform.

So that's maybe one dynamic that's specific to the quarter. More generally, I think our construct has not changed. We're comfortable going up to a number around $1 billion or so in loan assets. And that's the number that leaves us with still, I think, a comfortable amount of cash to continue running the business and add some safety stock.

And within that parameter, we'll sort of flux up and down as we believe benefits the business.

John Coffey -- Barclays -- Analyst

All right. Thanks, Sanjay.

Operator

We will take our next question from Rob Wildhack with Autonomous Research. Please go ahead.

Rob Wildhack -- Autonomous Research -- Analyst

Hi, guys. Just another question about the committed capital co-investment. I do appreciate the detail in the slides there. How are changes to that reflected in the income statement? And does that mean that there's a plus-$11.5 million impact in the second quarter from the mark up over the $40.2 million?

Sanjay Datta -- Chief Financial Officer

Yeah. Hey, Rob. So, the short answer is, no, that impact is not in the P&L. I wish it were that simple.

The reality is we've done a couple of different deals. And contractually, they'll have their nuances. I think the result of that is that they're all being accounted for in slightly different ways. Some of it is showing up on the balance sheet as a beneficial interest.

Some of it is showing up on the balance sheet under our restricted assets. Some of it is being fair valued, some of it is being carried at cost. So, I think the difficulty of trying to pull all those different accounting treatments together and create a clear picture is the reason why we're just going to put it on one slide for you. But the short answer to your ultimate question is, no, it's not really hitting the P&L in a way where fair value is being recognized in the net interest income line.

Rob Wildhack -- Autonomous Research -- Analyst

OK. Thanks. And then, of the $2 billion longer-term funding commitment you announced in May, how much of that was funded in the second quarter?

Sanjay Datta -- Chief Financial Officer

I mean, I would say approximately -- if you sort of remove the back book component that was a component of that original deal, roughly a quarter.

Rob Wildhack -- Autonomous Research -- Analyst

OK. So, one quarter in excess of the $352 million that you sold?

Sanjay Datta -- Chief Financial Officer

Got that, Rob.

Rob Wildhack -- Autonomous Research -- Analyst

No, I think you dropped from me quickly.

Sanjay Datta -- Chief Financial Officer

Oh, sorry. I just said that that's correct. The statement you made beyond the back book sale, it was about a quarter of the remainder.

Rob Wildhack -- Autonomous Research -- Analyst

Great. Thanks.

Sanjay Datta -- Chief Financial Officer

Sorry about that.

Operator

We will take our next question from James Faucette with Morgan Stanley. Please go ahead.

James Faucette -- Morgan Stanley -- Analyst

Thank you very much. Just hoping to get a little bit of clarification really quickly here. And maybe I missed it. In the guide, you're looking for around minus $10 million net interest income in the third quarter.

Is that being driven by loans on the balance sheet or the way that the funding commitments work through the P&L? Just trying to make sure I understand that mechanism.

Sanjay Datta -- Chief Financial Officer

Yeah. Hi, James. Yeah, I would say a major factor in that number is the fact that by and large, our balance sheet now particularly after the ABS deal we completed, which -- where we cleared out most of our recent sort of core personal loans, the remaining balances are R&D and a lot of it is very seasoned. And some of those older vintages of, say, auto loans, auto loans from year-and-a-half years ago, their charge-offs are starting to grow, both because they were originated at a time where our models were still sort of in calibration and as well, they were originated in an environment that was -- and has rapidly deteriorated.

So, on some of the -- maybe you could think of this partly as some of the cost of doing R&D we've got a book of R&D loans now and the charge-offs are elevated.

James Faucette -- Morgan Stanley -- Analyst

OK. Great. And then, I guess just as related to demand and originations, like how are you thinking about the impact from higher take rate? And how should we think about how those could move and how overall origination should trend at least on a sequential basis through the rest of this year and into '24? Thanks.

Dave Girouard -- Chief Executive Officer

Hey, James. For sure, our rates are kind of as high as they've ever been, I believe. And that -- part of that is a function of UMI being super high, part of it is the return demanded by the market is much higher, of course, related to Fed rates, and also our take rate is high. And all three of those contribute to the rates being high.

We would anticipate those things kind of coming off together. So, as you see UMI trend down, and maybe at some point interest rates will trend down as well, but also our take rate is probably higher than we would see it being in a kind of normalized scenario. So, all three of those are very high, driving price very high. And for sure, the reason -- I mean, the counter to that is that the market for the loan demand is very strong.

And for that reason, that's why you're seeing very low CAC. And so, that's kind of the place where we are today. Prices for credit are super high. Demand remains super high, and that sort of nets out to where we are right now.

James Faucette -- Morgan Stanley -- Analyst

OK. Thank you.

Operator

We will take our next question from Reggie Smith with JPMorgan. Please go ahead.

Reggie Smith -- JPMorgan Chase and Company -- Analyst

Hey. Thanks for taking my question, guys. I guess kind of a follow-up to the last question. It sounds like, obviously, pricing is up right now.

How do you guys think about managing adverse selection? Maybe talk a little bit about some of the sensitivity people may exhibit to price? I see that your approval or conversion rate is down. And I know there's two components to that. I would imagine approvals are down, but maybe you could talk a little bit about the acceptance of these offers as well to kind of give us some color there. And I have a follow-up.

Thank you.

Sanjay Datta -- Chief Financial Officer

Yeah. Hey, Reggie. Sure. So, let's see a little bit about adverse selection.

Adverse selection tends to be an impact that is greater where the market or the segment is more competitive. So when there's a lot of sort of competing alternatives and you try to raise your rates, you'll typically suffer from adverse selection. And of course, when segments are less competed, adverse selection is less of an effect. In our case, for a lot of the segments where we tend to learn a lot of volume, even with higher rates, we tend to still have the best rates available.

So, even with higher take rates and higher loss assumptions, we, in many instances, are still significantly below what you might think it was the market clearing rate out there based on the credit scores. And so, in that instance, there's not a lot of discernible adverse selection. And if we were to try and raise our take rate significantly in very highly competed segments, very prime borrowers, then it's something that we would be thinking about a lot more.

Reggie Smith -- JPMorgan Chase and Company -- Analyst

Understood.

Sanjay Datta -- Chief Financial Officer

And your second question, Reggie, was about acceptance rates?

Reggie Smith -- JPMorgan Chase and Company -- Analyst

Yes.

Sanjay Datta -- Chief Financial Officer

Yes. I mean, it's very simple. It's a pretty classic sort of supply and demand construct, where we raise our rates. And not only do our approval rates go down because of the 36% APR cut off, but for those who remain approved, they'll be less likely to take a loan.

And typically, at least what we've observed in our data is that people who don't take loans with us don't necessarily take them from a competing source. The majority of them just don't take the loan. So it causes people's demand to reduce.

Reggie Smith -- JPMorgan Chase and Company -- Analyst

Sure. Got it. And one quick follow-up on the co-investment, and I appreciate the disclosure. How should we think about that book? Are these loans subordinated to the rest of the structure? Is there a certain ratio that you must hold relative to what you run through these -- through the committed facility? Any color you could share there would be helpful for kind of modeling and thinking about that.

Thank you.

Sanjay Datta -- Chief Financial Officer

Sure. Yeah. I mean, they do tend to sit in the equity part of the stack, if you will, on a loan transfer where you have sort of senior money and maybe mezzanine money and sort of equity or residual money, this tends to be sort of at the equity side of the stack. There's no -- there's not necessarily any sort of specific ratios.

I think each time there is an investment made by one of our committed capital partners, we tend to have a co-investment that varies depending on the relationship.

Reggie Smith -- JPMorgan Chase and Company -- Analyst

Understood. So I guess there's no way to kind of think about or know how large that could be? Like do you have expectations over the next few quarters, like how big that could be?

Sanjay Datta -- Chief Financial Officer

None that we're talking about explicitly. But I guess I would say, like we consider this to be a part of the overall sort of risk budget of the business, and we've talked about sort of making sure we stay at or under a level of about $1 billion in asset risk. And this is part of that envelope. So, I would anticipate over time, more of our capital sort of falling into this category, maybe less under direct loan assets, where we can sort of use our capital to unlock broader pools of capital and co-invest, as opposed to having loans directly on the balance sheet.

But I think what we're going to try to do with you guys is have a conversation about the overall sort of risk position and risk budget of the company, and we'll certainly have some views on how big we would ever want that to get.

Reggie Smith -- JPMorgan Chase and Company -- Analyst

Got it. Understood. Thank you so much.

Operator

We will take our next question from Giuliano Bologna with Compass Point. Please go ahead.

Giuliano Bologna -- Compa

Hello. Thanks for taking my questions. One thing I'm curious about is you obviously put in that new risk sharing disclosure. What I would be curious about -- and I realize this period is probably somewhat different because you did a loan sale and you had some of the forward funding agreements funding in the quarter.

I'm curious if you can provide a rough sense of how much principal that risk, $40 million or so of up-down risk-sharing covers.

Sanjay Datta -- Chief Financial Officer

Hey, Giuliano. Yeah, on the order of $800 million.

Giuliano Bologna -- Compa

That sounds good. Very helpful. Then, the next thing I'd be curious about is just thinking about from a funding perspective. You obviously built up the balance sheet and then disclosed some during 3Q with the ABS deal.

I'd be curious if you have a rough sense of how much balance sheet -- how much the volume was that flowed through the balance sheet during the second quarter was? Thank you.

Sanjay Datta -- Chief Financial Officer

Yeah. I don't think we have that explicitly broken out, but I think that's something that you could probably deduce from the cash flow statement. We can point you in the right direction when we chat later.

Giuliano Bologna -- Compa

That's sounds good. It'll probably also come out in the Q when that comes out as well. And the only question, I think, which is a little bit of a follow-up. Obviously, on the forward funding agreement side, you had mentioned in the past, roughly $500 million per quarter.

You mentioned an additional -- some additional partners. I'd be curious if that number changes. And obviously, that will impact kind of where the risk sharing goes, because you won't have -- may not have large loan sales going forward. I'd be curious just to get your perspective on that.

Sanjay Datta -- Chief Financial Officer

Sure. So you're asking if the amount of committed funding per quarter expectation has grown?

Giuliano Bologna -- Compa

Yeah, if that has changed? Last quarter, what was mentioned was roughly $500 million per quarter. I'm curious if that has changed with the new funding agreements and based on the current arrangements in place.

Sanjay Datta -- Chief Financial Officer

I would say not in a meaningful enough way to sort of reannounce it, if you will. As Dave mentioned, we did -- we are working with a new partner in the committed capital world, and they're now contributing to our funding. But some of our constraints now are on the borrower side. And so, I think given that, we're probably in a similar ballpark.

Giuliano Bologna -- Compa

That's very helpful. I appreciate the answers to the questions, and I will jump back in the queue. Thank you.

Operator

We will take our next question from David Scharf with JMP Securities. Please go ahead.

David Scharf -- JMP Securities -- Analyst

Hi. Good afternoon, and thanks for taking my questions. A lot have been asked, I guess these are kind of extensions of what's already been offered up. But Sanjay, just back to, I guess, the funding side and how it impacts kind of how you're thinking about originations.

I know you've reiterated sort of the comfort level of that $1 billion ceiling on retained assets. But is there any sort of targeted roadmap for year-end where you'd like to get the balance sheet contracted to? I mean, should we be thinking about all these new funding partners as vehicles for stepped-up asset sales in the second half? Just trying to get a sense how we ought to think about the on-balance sheet exposure going out six months, and ultimately, how much room you have to reaccelerate volumes when the macro environment dictates it.

Sanjay Datta -- Chief Financial Officer

Hey, David. Sure. Let's see. I mean, a lot of it is dependent on the environment, of course.

I mean, in an ideal world, what we'd love to do is reduce the balance sheet significantly, apply it to R&D and begin to maybe apply more of it very surgically to these committed capital-type co-investments where we can unlock much larger pools of third-party capital. That's the ideal. Now, we're in an environment where borrower demand is high, funding markets are tight. And frankly, the collateral has a lot of excess value.

I think that they're priced for very high yields, and we can provide a lot of value to the business with one extra dollar origination from the perspective that we obviously unlock a lot of take rate. And then, similarly, I think these loans are priced for good yield. So, it's a rational economic decision in the current environment. We love a normalized environment, where there is plenty of third-party capital to satisfy the borrower demand and we weren't in a position of using ours.

But that will require essentially a normalization of the UMI and a normalization of the funding markets.

David Scharf -- JMP Securities -- Analyst

Got it.

Sanjay Datta -- Chief Financial Officer

It's hard to say whether that will happen by the end of the year or not.

David Scharf -- JMP Securities -- Analyst

Understood. Maybe a follow-up. Regarding the negative fair value marks again this quarter, were the downward revisions, are they mostly related to kind of discount rate, prevailing rates? Is it more credit performance-related? Or is it just based on -- with so many underperforming loans for sale out there, just kind of prevailing market data points you're seeing? And kind of related to that, is it sort of evenly distributed, those fair value decreases to personal loans in auto? Or is it more concentrated in auto?

Sanjay Datta -- Chief Financial Officer

Sure. Yes. So in Q2, there's two things of about equal magnitude. One was on the unrealized fair value side, where some of our assets, notably, some of the co-investments we made in some of these deals.

They were -- they had applied to them a much cheaper discount rate than we were anticipating by the third-party valuators. So that was sort of an unrealized value reduction. And the other was what I mentioned earlier, which is -- well, I mean, our R&D portfolio, which is predominantly auto, they are getting to a level of charge-offs, and all that are bringing down the NII line. I think if you're thinking about Q3 and forward, it's predominantly the latter.

It's old loans in the auto segment that were originated in a very different environment with a model that was still undergoing calibration. So now that I think we've learned a lot in our calibration of the auto product in terms of the scale and timing of that R&D effort. I think there's some great learnings we've had and how we will apply it differently to the next set of products that we're going to calibrate. But the auto book that we have, that's really at the root of a lot of charge-offs that are coming through the P&L.

David Scharf -- JMP Securities -- Analyst

Great. Thank you so much.

Operator

We will take our next question from Vincent Caintic with Stephens. Please go ahead.

Vincent Caintic -- Stephens Inc. -- Analyst

Hey. Good afternoon. Thanks for taking my questions. First question, just kind of a big picture question about your funding partners.

When we last spoke about the Castlelake $2 billion. I think that was a proof of concept. So just wondering if you could talk about kind of the conversations, the appetite you've been having from potential partners and kind of the mix between institutional investors versus the banks and other guys. Presumably, this co-invest -- this capital co-investment side is very helpful.

And seeing that the assessed value is higher than the initial capital invested, so presumably, that implies that the performance is better than kind of the initial agreement. So just sort of wondering if you could talk about the appetite on your discussions with your partners. Thank you.

Sanjay Datta -- Chief Financial Officer

Sure. Yeah. So a couple of questions in there. This is Sanjay.

I'll sort of talk through them. Then first question related about the nature of the discussions around these sort of -- some of the capital-type deals. We've done a couple of deals now. And they are obviously different flavor, but they all sort of get to the same thing, which is it's a counterparty that has the wherewithal to spend through a cycle and for some committed period of time.

And in exchange, we think there's some pretty attractive return profiles for them. And as we've shown, we're willing to put our skin in the game alongside. And I think that is, on the one hand, a very attractive conversation right now. And I think there's a lot of people who are engaging.

On the other hand, as I said, it's a market with a lot of distraction in it right now. And so, these conversations are making progress, but I think they are -- there's a lot of company. And any investor who's selective right now has a lot of interesting options available. Some are ad hoc and one-off, others are more programmatic, like the ones we're talking about with them.

And so, I think that it's proving to be an interesting and attractive option, but in a crowded field right now. The split between institutional money and bank money, I would say, as we said in our remarks, some of our take rates benefited this quarter because there was a further shift toward institutional money. It's no secret that banks have challenges right now with liquidity. A lot of them, as we said, are looking to harvest cash.

And so, that was sort of a mix shift that happened in Q2. It wouldn't surprise me if that trend held or continued in Q3. As for the -- sort of the committed capital that we outlined in our investor slide and how to think about it, I guess, first of all, we're saying that the -- a large part of that initial investment of capital on our part wasn't just new originations. As you recall, there was a back book transaction.

That was a part of the Castlelake deal that you mentioned. And so, there was a sort of a one-time sort of retention of basis in that deal. That's part of the $40 million. I would say most of the new originations that we produced under the guise of these committed capital deals in Q2 are sort of on track and at par, if you will.

The majority of the upside that we have between the $40 million that we invested and the $50 million that we're sort of assessing or forecasting, it has to do with how the back book itself was priced and how it was expected to perform. And it is, in fact, beating those expectations. So it's less, I would say, a reflection of new origination, more a reflection of the back book, which is a pretty sizable portion of that $40 million. I guess the other implication is, as you think about how that might grow in the future, it certainly won't grow at a clip of $40 million a quarter based on the agreements we have in place.

Vincent Caintic -- Stephens Inc. -- Analyst

OK. That's super helpful. Quick follow-up. So the take rates, presumably from the mix shift in funding, it wouldn't be unreasonable to assume that the take rate holds or even gets better going forward? Just want to confirm that.

Thank you.

Sanjay Datta -- Chief Financial Officer

I guess, in aggregate, you can see that we've sort of guided to a relatively flat contribution margin next quarter, maybe marginally lower. And so, I think you could probably infer from that that our take rates, we're assuming that they're going to be roughly stable to this quarter. In the medium to longer term, as Dave said, we would expect with a normalizing economy for our take rates to slowly subside.

Vincent Caintic -- Stephens Inc. -- Analyst

OK. That's very helpful. Thank you.

Sanjay Datta -- Chief Financial Officer

Thank you.

Operator

And we have time for one more question, and we will go to Simon Clinch with Atlantic Equities. Please go ahead.

Simon Clinch -- Atlantic Equities -- Analyst

Hi, guys. Thanks for taking my second batch of questions. I was actually wondering, maybe, Dave, if you could talk a bit more about the parallel time and curve calibration. Because I guess, I'm not the most tech-savvy person, and it sounded an awful lot like back testing to me, but you say it isn't.

So I'll be really interested in sort of how it differs or why isn't it back testing? And why you -- I guess, what this really means in terms of your ability to capture the next wave of -- well, actually manage through the next cycle that we see in a much better fashion than you have done in the last couple of years.

Dave Girouard -- Chief Executive Officer

Yes. Sure, Simon. I mean, it ultimately comes down to being able to calibrate a model as quickly as possible, which really just leaps you into developing the next model. And so, at the heart of it, it's about model development speed.

But the way it generally works is -- a back test is when you apply a new model to a bunch of old loans and see if it can accurately predict their outcome. I mean, that's sort of the way that models are developed in the first place, is something like a very large back test. It's essentially what AI training is. But in this case, what we're actually doing is not predicting the past, but predicting the future for loans that different models had originated.

And so, what that really means, as I said earlier, normally, if you have 36-month loans, if you want to get accuracy, see the performance of the model to the entire timing curve, you need to wait 36 months, of course. But if you have a diverse set of loans that were originated in all sorts of times in the past several years, even in the first month, you're getting observations into all 36 months of the timing curve. And that's because you have reunderwritten model -- loans that were originally done under a different model. But you're not predicting the past, you're predicting what they'll do in the next month and the month after that.

So that's what's really unique about it. It is a true model. It is predicting future performance of loans, but it's predicting the future of loans that it didn't originally -- was not originally used to originate. And that's kind of the magic of it, is it just gives you a lot more data about the performance of your loans, specifically about the timing curve than you could get -- which would normally, again, take much longer time.

And that's the heart of it. It's a very novel notion. It helps you have a very clear understanding of how your model is performing very, very quickly. And that all goes toward speed.

Simon Clinch -- Atlantic Equities -- Analyst

OK. Great. Thanks. I might have to go back to school on that one.

Thank you.

Dave Girouard -- Chief Executive Officer

We might try to write something up on this to get -- for the nerds who really want to dig deep into how this -- something like this works, we'll probably try to put something out there.

Operator

And that concludes today's question-and-answer session. I will turn the conference back to Dave Girouard for any additional or closing remarks.

Dave Girouard -- Chief Executive Officer

Thanks to everybody for joining us today. I'm confident that financial services and lending in particular will be one of the bright shining stars for AI in the coming years and in the coming decades. And we believe there's no company better positioned to lead that transformation than Upstart. So thanks for joining us today.

We'll see you next time.

Operator

[Operator signoff]

Duration: 0 minutes

Call participants:

Jason Schmidt -- Vice President, Investor Relations

Dave Girouard -- Chief Executive Officer

Sanjay Datta -- Chief Financial Officer

Simon Clinch -- Atlantic Equities -- Analyst

Lance Jessurun -- BTIG -- Analyst

John Coffey -- Barclays -- Analyst

Rob Wildhack -- Autonomous Research -- Analyst

James Faucette -- Morgan Stanley -- Analyst

Reggie Smith -- JPMorgan Chase and Company -- Analyst

Giuliano Bologna -- Compa

David Scharf -- JMP Securities -- Analyst

Vincent Caintic -- Stephens Inc. -- Analyst

More UPST analysis

All earnings call transcripts