UWM Holdings Corporation Class A

Good evening, wherever you may be.

My name is Damien Weldon. I lead the Business Unit for Valuation Analytics in Ice Mortgage Technology. I'd like to welcome you to this current instance of our webinar series. Today we're going to be talking about an upcoming event, an upcoming event for an ABM Federal Rule. Thank you for watching. Thanks for watching! Thank you for watching. Thank you for watching.

Please do reach out to us.

Thank you very much. ... ... ... ... ... ... Thank you. Thank you.

Thank you very much. excuse me, Jewett's AVM and Valuation Solutions customers. And that's a tool that we call AVM Model Monitor. So we're launching this today, initially rolling this out on a pilot basis. And again, we'll overview and demo that for you today. And again, I'm excited to be sharing this with our client base, in particular, our Encompass users. And then again, we'll wrap up with a Q&A, a little Q&A within our team here. And then we invite audience questions throughout the webinar. So if If you do have a question at any point when we're talking or demoing or illustrating anything here, please do pop your question in the Q&A section of the webinar interface here, and then obviously we'll get to those as we proceed to the webinar. Thank you very much. Download that and look at that. Again, if you have any questions, post the webinar. Please feel free to reach out to us at any point. We're here to serve you. We're here to meet your needs. So that's our coverage and our scope for today. Just by way of background, again, ABMs don't exist in a vacuum. They're part of a very comprehensive valuation solution suite that ICE has deployed and encompassed. and is also available in part on a standalone basis. Just a brief recap of our principal valuation products there. ABM is obviously right at the heart, but they also validate our data collection tool for on-site information collection from the homeowner. Thank you very much. Thank you very much. We have risk profiler available for appraisal review, PPO review, also that. And then right up front, we have what we call valuation selector, which is basically our recommendation engine for a particular valuation method, whether that be an AVM, an appraisal, inspection, and so on, just purely based on the, or solely based on the availability, the amount. and the quality of data available. And again, we work with a wide range of clients, originators and services principally, but also folks in the secondary and capital markets side of the house, as well as real estate brokerages and federal agencies also. Just to show a little bit of that solution in action, again, here's our standardized HELOC workflow within Encompass. We deploy Valuation Selector at to recommend the appropriate valuation method, again, in tandem with investor guidelines, whether from, you know, whether ultimately the destination of the credit is HELOC and the HELOC securitization or HEI securitization, or indeed if it's remaining on balance sheet, validate then obviously for the real-time data collection, generating property condition and quality, an AVM adjustment, and then that package then rolling forward, for example, into a desktop appraisal and that your AMC or your staff appraiser and for the larger institutions would then deploy. So solutions in action. Okay, I'm going to hand it over to my colleague Renji to talk about our ABM waterfall and the focus on the remaining segment of time we have together today will be on our ABMs and how our ABMs adhere to the new federal guidelines, but that also includes our ABM waterfall. So Renji, please take us through a little bit of overview of our PBM.

Yes, Demi, thank you. There are about 30 areas in the property valuation space, and I own the area of them. About three years ago, we did a deep dive analysis on the performance of each individual area, and we found that each area has its own unique strengths, but also comes with limitations. For example, one area might perform very well in California, but not so well in Texas. One might be good for a single-family home, Thank you for joining us. We call this selection process the ICE AVN waterfall, aka PVN. It's all about choosing the right AVN for the right use case. It took us about two years time to build using about 18,000 decision trees trained on roughly 2 million properties. As a result, it maximized not only PP10 accuracy, but also HRA coverage. To maintain the PVN accuracy and consistency, we use a 12-month rolling testing validation framework which allow us to monitor and tune the decision chain on a monthly basis. And we also run automated testing to track PVN and issue AVN performance in real production on a daily basis. So in a few minutes, my colleague and I will walk you through how we document and monitor the daily testing results using AVN model monitor. It's very cool stuff and you'll like it. For now, I'll hand it over to my colleague, Zeng. She will give you more detail on the ICE AVMs. Let me go to the next slide.

Thank you, Renji. Over to Zeng. Thanks, Renji. So as you can see, we have ICE AVM suite. Thank you for joining us. Thank you very much. Thank you. Thank you. Thank you. Each AVM confidence score range, which indicates how reliable the valuation is, the clients can also set your own confidence score threshold based on your comfort level. So you can choose to only use the valuations that meet your internal standard for reliability. So I will go into detail of how each ABM works. If you have any questions regarding how they work and how to integrate their results, feel free to bring them up during our Q&A section, and we're happy to walk through the details and help you clarifying anything. Next, I'll hand it over about our IS model risk management.

Thank you very much, Zen, and apologies. There was a little bit of interference on Zen's line there, so she was cutting out a little bit there, so apologies for that. And over to Anand, who's going to give us an overview of ICE's model risk management framework in general across our business. Thank you, Anand.

Thanks, Damien. What we have been able to do is being able to provide an independent entity with the ICE MRM team. Think of them as a second line of business unit that helps perform the Thank you for joining us. also perform the same set of service for the global clearing services, the fixed income and data services, and also for the mortgage technology team. They have been validating models independently for one of the biggest exchanges in the world, as ICE currently owns the New York Stock Exchange. And our models go through a very similar independence validation to be able to perform the The third party independence that we need for the performance validations of our AVMs. A lot of details is also mentioned in our methodology document where we are able to understand the whole governance process with our internal validation team and the ICMRM model governance team. Off to you, Damian.

Thank you very much indeed, Anand, for the overview there, and we'll come back to some of the testing as we proceed with our discussion today. Again, a little bit of context in terms of what's happening on October 1st. A group of interagency, an interagency group of six federal regulators, the NCUA, the CFPB, the FDIC, the Fed itself, and the OCC came together back in, originally back in 2022, Thank you very much. Back there in 2023, invited industry comment on that and then put forward the final rule and issued that to the six agencies there in June of last year. That was published in the Federal Register then in August last year, which set the clock for the implementation date, which now is October 1st. And there are five quality control standards articulated in that rule, confidence in AVMs, that there be no manipulation of data, No conflicts of interest, random sampling, and that the use of AVMs should be in accordance with the non-discrimination laws. In each case of those, ICE is committed to ensuring that we're in alignment and in adherence with those standards in order to be able to articulate with confidence to our clients that we are, and also that our clients can take that and codify that and capture that in their documentation Thank you very much. and so that has been the law for a little bit over 15 years at this point and so in a certain sense therefore no major change that those standards continue to be enforced on and after October 1st and what is new is the new standard relation in relation to non-discrimination and that is the first time that AVMs have had an explicit test for non-discrimination in the in their usage across protected classes for example race and ethnicity and we'll be talking a little bit more about that when we overview how we adhere to that and how we test for that within our model monitor framework. Again, we're not positioned to provide legal or compliance advice of any kind, but what I can say to you in your implementation of these standards, the scope of the rule is mortgage originators and secondary market participants, but obviously we're here to help. We're here to support and for you to optimize and maximize your use of AVMs in a compliant and best practices way. And this webinar is part of our education rollout in relation to that. Okay, at this point, I'd like to hand it back to Anand, who's going to overview our AVM model monitor tool, which speaks directly to many of these standards, and particularly the first one, which is about having confidence in AVMs and how to measure that and how to Monitor and surveil that over time. And then the fifth standard in relation to testing for non-discrimination on an analytical basis with our AVMs. And please do keep your comments coming in. I'm delighted to answer those once we move to the Q&A section of today. And again, any questions that we don't get to today, we'll be very happy to have an email exchange or a separate call with you afterwards. So please do keep those questions coming in. Over to you, Anand.

Can you move to the next screen? So what you guys are looking at is the main view of our AVM model monitor. On the left-hand side, we have the details from our ICE Mortgage Technology website, which will give you a great overview of our valuation suite. You will find details about all of our AVMs, the PVM, Validate, as well as all of our webinars that keep on conducting on how the PVM works, how do our combination of AVMs work, and how do we perform and validate are testing. On the bottom right-hand side of the corner, what you guys are looking at is the detail of our model methodology document. The section four here, we explicitly help everyone understand how ICE AVM supports the quality control standards. We have the details about our model risk management, the relationship that they have in terms of testing and validating the performance of our AVMs, and the overall details about the The model development, model governance, model optimization, model deployment, and model governance. The top right-hand side corner is what we will go into a lot of details right now on the AVM model monitor itself, where we provide validation and reporting on the performance of our AVMs in different sections. There is a section separately for outlier analysis and under-overvaluation. There is a section that specifically focuses on PPAT and NFSD. and they're going to go into deeper analysis or a walkthrough to our overall AVM section, the monthly performance AVM section and the geographical AVM sections to help understand how this tool reports and validates and helps measure the performance of our AVMs. So what you guys are looking at right now is the main overall AVM performance page. So hit rate and PPE10 have been considered to be the prime metrics or the performance metrics that are used by many of our clients to help understand AVM performance. At a very high level, the hit rate represents the model's capability to generate evaluation. and the PPE10 helps us understand the accuracy of the valuation where it gives us the measure on the number of properties that were under 10% error rate. The results that you guys are looking at is from our testing results done from in all of 2025. It's still August of 2025 and we have tested around 2.49 million samples and they're showcasing the performance of both Thank you for joining us. Thank you for joining us. The overall performance of our models, but given at a monthly level. So here what you guys are looking at is the same metrics, but now enhanced a little bit more at the monthly level. We feel the monthly view gives a great detail about the model stability, where all of our tests are forward blind testing, which means that we test all of the samples that we see before they enter our databases. It is a true forward blind test where the AVMs Thank you for joining us. Thank you for joining us. created cohorts in a way that are very relevant to help understand AVM performances in very specific cohorts, very specific lot sizes. And the race and ethnicity flag here is captured using the census tract and the census block information where we utilize the census information to assign a majority level census tract. and then capture and look at the performance of all of our models at the monthly level, daily level and optimize for these specific cohorts. The next view that we are looking at is now a geographical view where, as you guys can imagine, state level view depending on the different volumes in the transaction, every state gets different number of samples that are captured. but we do provide for every state, for every AVM, for every month performance at the state level, at the CBSA level, at the county level, and then we'll look at the SIPP level as well as we are going down granularity in Thank you very much. The last page that I wanted to bring everyone's attention to is our historical testing data sets. What we're giving here is a 10-year view of some of our six of the eight AVMs have a 10-year history and more of testing performance information. So we help understand the hit rate, the median error, and the PPE10 and sample sizes to give everyone a great history of how the AVMs have performed given to different Macroeconomic conditions, interest rate changes, the COVID environment changes where these models have been performing relatively stable across a lot of different economic cycles. And we utilize all of this testing performance and validation to improve and enhance the performance of our AVMs. For two of the models, as you can see, some of them started in 2016 and 2018. We utilize a lot of this information to understand our seasonal behavior and help improve the performance with a greater baseline. Off to you, Damien.

Great overview there, Anand. I appreciate it. And again, I encourage our audience, please do feel free to use our Q&A section of our webinar interface here if you have any questions either on any of the content or any of the remarks or comments or information that you've heard so far in our webinar today.