Baird Fixed Income Insights: Convexity Pulse

AI-driven servicing efficiency and FHFA policy changes may accelerate mortgage prepayments, impacting MBS valuations.

28 min Listen to the original Episode page
Mortgage-Backed SecuritiesAI in FinanceFederal Housing Finance AgencyPrepayment RiskMonetary PolicyHousing Market

Summary

Speakers at Baird Fixed Income Insights discuss how AI adoption by mortgage servicers is reducing refinancing friction and accelerating prepayments, particularly for low-balance borrowers. They note that October Agency MBS net issuance slowed sharply to $9.7 billion, yet fair value models suggest 30-year and 15-year MBS remain cheap relative to predictions. The analysis highlights that AI tools, such as Rocket's voice platform and Pennymac's AI agents, are already operational in servicing workflows. Additionally, an FHFA executive order aims to reduce refinancing costs through title acceptance and appraisal waivers, which could disproportionately benefit smaller loans and increase prepayment risk.

Key takeaways

  1. 1
    Loans with appraisal waivers prepaid faster than those requiring appraisals: at a 50bp refinance incentive, waiver loans had a 21.2% CPR versus 15.7% for traditional loans, a gap that persisted to a 10.2% difference at a 100bp incentive.

    Unidentified speaker (speaker_1) · 23:29

Episode sections

  1. 0:17 – 2:22

    Intro and AI in Mortgage Servicing

    Speakers discuss how AI is being used by servicers to identify borrowers and streamline refinancing, potentially reducing friction and accelerating prepayments. They note that while this benefits borrowers, it may negatively impact investors by weakening call protection.

  2. 2:22 – 9:00

    MBS Market Outlook and Supply Dynamics

    The speakers highlight a sharp slowdown in October Agency MBS net issuance, with net issuance falling to $9.7 billion. They argue that despite higher rates, the supply picture supports a moderate overweight in MBS, as fair value models suggest the 30-year and 15-year MBS are cheap relative to predictions.

  3. 9:00 – 11:23

    Political Impact on Housing and Policy

    Following the US midterm elections, the speakers analyze the potential for divided government to impact housing policy. They suggest that while political gridlock may limit immediate fiscal expansion, modest declines in mortgage rates could improve affordability, though meaningful policy changes will take time to implement.

  4. 11:23 – 19:46

    AI-Driven Prepayment Risk and Portfolio Management

    The discussion shifts to how AI improves servicing efficiency, allowing servicers to predict and execute refinances faster. This increases prepayment risk for investors, particularly in higher coupon pools, and may alter the exercise of mortgage options for discounted loans through debt consolidation.

  5. 19:46 – 27:49

    FHFA Executive Order and Refinancing Friction

    An October 22nd executive order directs FHFA to expand title acceptance pilots and consider reducing appraisal and income documentation requirements. The speakers analyze how these potential reductions in fixed refinancing costs could disproportionately benefit low-balance borrowers and accelerate prepayments.

Claims worth verifying

The 30-year MBS basis is around 128 basis points versus a fair value prediction of 122 basis points, and the 15-year MBS is approximately 79 basis points versus a predicted 73 basis points, making both 1.5 to 2 standard errors cheap.

Unidentified speaker (speaker_0) ·3:02

How to verify: Check Baird Fixed Income Insights research reports or market data providers (e.g., Bloomberg, MSCI) for current MBS basis spreads and fair value model outputs for October 2026. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Gross agency fixed rate single family issuance fell to $81.4 billion in October, down 22% from September and nearly 30% from a year ago, while net issuance collapsed from $27.4 billion to $9.7 billion.

Unidentified speaker (speaker_1) ·4:35

How to verify: Review Federal Housing Finance Agency (FHFA) monthly mortgage market data or GSE (Fannie Mae/Freddie Mac) monthly production reports for October 2026. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Conventional 30-year mortgages generated $2.1 billion of positive net issuance in October, while shorter amortization products saw $6.9 billion of runoff, resulting in negative GSE net issuance of $4.8 billion.

Unidentified speaker (speaker_1) ·6:01

How to verify: Analyze FHFA monthly mortgage market data broken down by loan type and amortization period for October 2026. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Through September 2026, Fannie Mae and Freddie Mac increased their combined agency MBS holdings by approximately $51 billion, while the Fed's MBS portfolio declined by $143 billion.

Unidentified speaker (speaker_1) ·7:10

How to verify: Check Fannie Mae and Freddie Mac quarterly financial statements or H.15 Federal Reserve statistical release for H.3/H.4.1 data on agency MBS holdings. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Rocket reported that its AI voice platform handled 65% of inbound servicing calls in Q3 2026.

Unidentified speaker (UNKNOWN) ·15:14

How to verify: Review Rocket Companies (RKT) Q3 2026 earnings release or investor presentation materials. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Pennymac developed 22 AI agents across origination workflows.

Unidentified speaker (UNKNOWN) ·15:14

How to verify: Review Pennymac (PNC) Q3 2026 earnings release or investor presentation materials. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

An October 22nd executive order directs FHFA to expand a title acceptance pilot and consider reducing appraisal requirements, income documentation burdens, and loan-level pricing adjustments (LLPA).

Unidentified speaker (UNKNOWN) ·19:46

How to verify: Locate the text of the Executive Order issued on October 22, 2026, via the Federal Register or White House archives. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

FHFA estimated title acceptance savings of $500 to $1,500 per eligible refinance transaction, with a $1,100 saving representing 73 basis points of a $150k loan.

Unidentified speaker (speaker_0) ·21:50

How to verify: Review FHFA public statements or reports regarding the title acceptance pilot program and associated cost savings estimates. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Loans with appraisal waivers prepaid faster than those requiring appraisals: at a 50bp refinance incentive, waiver loans had a 21.2% CPR versus 15.7% for traditional loans.

Unidentified speaker (speaker_1) ·23:29

How to verify: Analyze historical GSE prepayment data comparing loans with appraisal waivers to those with appraisals, controlling for refinance incentive spreads. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

GSEs charge lenders 187 basis points upfront fee for a purchase loan with a credit score of 690 and LTV of 84, and 250 basis points for a refinanced loan in the same bucket.

Unidentified speaker (speaker_0) ·22:23

How to verify: Review GSE (Fannie Mae/Freddie Mac) fee schedules or pricing guides for loans with credit scores of 690 and LTV of 84%. For recent events, verify 2026-08-16 through 2026-11-16. Use another period only when the speaker explicitly names it.

Primary source

Do your own research

  • Verify current 30-year and 15-year MBS basis spreads against the stated fair value predictions of 122bp and 73bp respectively using primary dealer quotes or Bloomberg data.
  • Check the latest monthly GSE issuance reports from Fannie Mae and Freddie Mac to confirm the October net issuance drop to $9.7 billion and the $51 billion increase in their combined holdings.
  • Review Rocket Mortgage's Q3 earnings reports or investor presentations to verify the claim that AI handled 65% of inbound servicing calls.
  • Examine FHFA's official documentation on the October 22nd executive order to assess the timeline and scope of title acceptance pilots and appraisal waiver expansions.
  • Analyze recent prepayment speed data from GSEs to validate the CPR differential between appraisal-waived and traditional loans at various refinance incentive levels.
  • Assess the impact of the Fed's $143 billion MBS portfolio decline on market liquidity and pricing dynamics compared to the GSEs' increased holdings.
  • Investigate whether AI-driven debt consolidation refinancing is material enough to alter prepayment models for deeply out-of-the-money loans.
  • Confirm the basis point savings calculation for a $1,100 title acceptance saving on $150k vs $650k loans using standard mortgage fee structures.
Full transcriptexpand
0:17
Unidentified speaker (speaker_0): And welcome back to Baird Fixed Income and Sides Convexity Pals Podcast. I am still Kirill Carol Kryloff, and I am here today. once again with my dearest comrade in arms, Stephen Scheer. Steven, I think we owe our listeners an apology for missing
0:37
Unidentified speaker (speaker_0): last week. But I think we have a very good reason and explanation to that. We were indeed kidnapped by Halloween monsters uh after the Halloween itself, who required demanded us to surrender our deep secrets and our prepayment model.
0:57
Unidentified speaker (speaker_0): We bravely refused and were able to escape their captivity. But that hopefully will absolve us of responsibility for missing last Saturday. Also, Stephen Steven, today is the national fast food Day. Switching gears on you, by the way.
1:17
Unidentified speaker (speaker_0): Which I think is remarkably appropriate given what we've been researching about artificial intelligence and mortgage refines.
1:26
Unidentified speaker (UNKNOWN): Hmm,
1:27
Unidentified speaker (speaker_1): I'm almost afraid to ask where you're going with that one.
1:31
Unidentified speaker (speaker_0): Well, mortgage companies, Steven, are using AI to identify bars, process applications. and remove all sorts of refinancing obstacles. Pretty soon, refinancing your mortgage may resemble ordering a Big Mac at the drive-thru.
1:49
Unidentified speaker (speaker_1): Oh, okay. Well, uh, I guess that means that uh
1:53
Unidentified speaker (UNKNOWN): would
1:53
Unidentified speaker (speaker_1): you like fries with that transitions to would you like to cash out another 100k?
1:58
Unidentified speaker (speaker_0): No, no, no, no, no, Steven. Unlike McDonald's, servicers. will already know the answer before you even pull up to the window.
2:08
Unidentified speaker (speaker_1): Well that's great for the borrower, but probably less exciting for the investor who thought the mortgage would you know be outstanding for another 10 years.
2:22
Unidentified speaker (UNKNOWN): All
2:23
Unidentified speaker (speaker_0): Alright right, Steven. After October's impressive recovery, mortgages finally encountered some turbulence, hotter inflation, higher rates, and rising volatilities push spreads predictably wider. But here is an interesting part. Even though, Stephen,
2:43
Unidentified speaker (speaker_0): generic OS still really don't look particularly generous, we have actually moved to a moderate overweight thanks to the supply picture. And our basis models, fair value models, are finally cooperating. The 30-year basis is around
3:02
Unidentified speaker (speaker_0): 128 basis points versus 122 basis points of our fair value prediction, while 15-year MBS is approximately 79 basis points versus. 73 basis points predicted value. So both are approximately one and a half to two standard
3:22
Unidentified speaker (speaker_0): errors cheap.
3:24
Unidentified speaker (speaker_1): Yeah, I mean, those are definitely significant numbers. And the supply picture, I think you're right, is uh really important at this point. Mortgage applications Locations barely responded to the recent rate rally, and we're heading into the seasonal slowdown.
3:39
Unidentified speaker (speaker_1): Meaningful new supply may remain scarce until probably March, really. Under those circumstances, we don't necessarily need the banks to return aggressively. Even modest demand from REITs or money managers can improve spreads considerably.
3:57
Unidentified speaker (speaker_0): Well, you'll you always look at the heart of things. Demand is the key here and has been the key.
4:03
Unidentified speaker (speaker_0): But supply, right? We've spent so much time, Stephen, and our recent podcast discussing who will buy that very next mortgage. Perhaps we should also ask how many mortgages investors will actually have to buy and can buy. October issuance slowed
4:23
Unidentified speaker (speaker_0): sharply, and I really mean sharply. And I think the composition. is even more interesting than the sort of headline. What caught your imagination, Steven, when you look at the issuance numbers?
4:35
Unidentified speaker (speaker_1): Well, let's start with the headline because the slowdown was huge. Gross agency fixed rate single family issuance fell to 81.4 billion, and that's down 22% from September and nearly 30% from a year ago. But I think the more important figure for investors is net
4:54
Unidentified speaker (speaker_1): issuance, which collapsed as well from about $27.4 billion in September to just $9.7 billion in October. So it's a substantial reduction in incremental supply. But we do have to remember that year-to-date year to date net issuance is still running slightly ahead of last year.
5:13
Unidentified speaker (UNKNOWN): Wow,
5:14
Unidentified speaker (speaker_0): you're you nailed a step. us Steven. You are a GOAT. By the way, I just learned from my kids that this is a term of endearment and ultimate respect. So I just want to clarify it. Because in
5:24
Unidentified speaker (UNKNOWN): my
5:25
Unidentified speaker (speaker_1): depends on how it's being used.
5:27
Unidentified speaker (UNKNOWN): That's right.
5:27
Unidentified speaker (speaker_0): Well in my attempt to kind of sound and useful and cool, I I certainly want to make sure I don't offend you because you you're neither as useful or cool as me. So you may take it as a as outright an insult but but you are a GOAT. By the way speaking of speaking of supply I think agency split is kind of fascinating don't you think
5:47
Unidentified speaker (speaker_0): Steven? GSC net issuance turn uh negative at what minus 4.8 billion while GDMA added 14 and a half billion. But the conventional number is somewhat misleading eh isn't it Steven?
6:01
Unidentified speaker (speaker_1): Yeah very much so because Conventional 30-year mortgages still generated uh 2.1 billion of positive net issuance. The negative aggregate came from 6.9 billion of runoff in the shorter amortization products, like 15-year, 20-year, 10-year. And that isn't really just an
6:21
Unidentified speaker (speaker_1): October phenomenon. Year to date, in fact, nearly 63 billion of positive.
6:27
Unidentified speaker (speaker_1): conventional 30-year supply has been almost perfectly offset by around 63 billion of runoff in those shorter amortization collabs.
6:36
Unidentified speaker (speaker_0): Yeah, sure. Slow down, love. We need to pause here, I think. We should not interpret negative GSC net issuance as It's evidence that conventional 30-year supply has disappeared, right? Meanwhile, genies can continue generating substantial positive net issuance, almost entirely, in fact,
6:56
Unidentified speaker (speaker_0): from 30-year collateral. And Steven, there's another part of the supply equation that I personally find quite interesting. How much of this issuance actually needs to be absorbed by private investors?
7:10
Unidentified speaker (speaker_1): There's where the Fed and GSCs GSEs enter the picture. Through September of this year, Fannie and Freddie increased their combined agency MBS holdings by approximately $51 billion. That provides a meaningful incremental demand. On the other hand, the Fed's MBS portfolio declined by
7:30
Unidentified speaker (speaker_1): $143 billion over the same period. So the GSC GSE bid offsets part of the runoff, not all of it.
7:36
Unidentified speaker (speaker_0): So here, Steven, we almost have to buy. borrow something from the police.
7:41
Unidentified speaker (UNKNOWN): Every
7:42
Unidentified speaker (speaker_1): breath you take,
7:43
Unidentified speaker (speaker_0): we really do need to watch every move these borrowers make because the supply that reaches the market really depends on both issues and changes in quote unquote official sector holdings.
7:58
Unidentified speaker (speaker_1): Well, Sting, I can't say you're wrong there. And uh the outlook is becoming more interesting. You know, mortgage rates rose sharply during September and early October, while the purchase and refi applications weakened. And that suggests gross issuance could remain subdued through the year end. But
8:17
Unidentified speaker (speaker_1): slower refinancing also means fewer paydowns, so net That issuance may not fall as dramatically as gross production.
8:25
Unidentified speaker (speaker_0): Which, Stephen, Steven, leaves us. with a fairly nuanced conclusion here. October's supply slowdown improves the technical backdrop. But the benefit, the benefit isn't uniform across the sectors. Conventional 30-year production remains positive, shorter amortization collateral continues
8:44
Unidentified speaker (speaker_0): running off, and genie supply, well, that is still a substantial factor.
9:00
Unidentified speaker (speaker_0): Alright, my dear Steven, the midterms are behind us. Did you know that? And did you also know the Democrats have taken control of both chambers of Congress, and Washington is preparing for another two years. of divided government. Depending on whom you ask, this is
9:20
Unidentified speaker (speaker_0): either a wonderful development or the beginning of apocalypse. Personally, Steve and I think REM, good old REM, had a very appropriate soundtrack. It's the end of the world as we know it. And I feel fine. But let's leave. politics
9:40
Unidentified speaker (speaker_0): aside. What does this actually mean for housing and mortgages? Steven, I know you have all the answers.
9:48
Unidentified speaker (speaker_1): Well, the immediate market reaction I think was fairly predictable. The tenure Treasury rallied modestly as investors interpreted gridlock as reducing the possibility of additional fiscal expansion. But I wouldn't confuse that with a structural change in interest rates. The deficit,
10:08
Unidentified speaker (speaker_1): inflation, and term premium haven't disappeared. For housing, however, a modest decline in mortgage rates can definitely improve affordability and potentially encourage some existing homeowners to give up their low mortgages and put more homes on the market.
10:24
Unidentified speaker (speaker_0): And once again, you're right on the money. Affordability was clearly one of the issues driving voters' frustration. So I suspect we will hear considerably more about first-time buyer assistance, down payment support, and increasing housing supply. But Steven, how much of that
10:44
Unidentified speaker (speaker_0): can realistically happen with Congress and the White House controlled by all different parties?
10:50
Unidentified speaker (speaker_1): Yeah, that's the complication. Congress can introduce proposals immediately, but implementing meaningful housing programs requires legislation, funding, and sometimes changes to underwriting or construction capacity. Those things take time. So lower mortgage rates could affect housing activity relatively
11:10
Unidentified speaker (speaker_1): quickly while the policy response may take considerably longer. And if additional buyer assistance arrives without more housing supply, some of the benefit could simply be reflected in higher home prices.
11:23
Unidentified speaker (speaker_0): Which brings us to the mortgage investor views. So here it goes. The first 25 to 50 basis points of a rate rally could be reasonably supportive for agency MBS, no doubt. Especially if lower volatility. and better technical can help spreads. Now, also let's
11:43
Unidentified speaker (speaker_0): remember most borrowers still aren't close enough to refi. But if the rally continues, the situation will change dramatically. All of a sudden, sixes and six and a halves that investors accumulated to gain income and shorten duration will begin facing meaningful call exposure. What
12:03
Unidentified speaker (speaker_0): would you, Steven, emphasize from a portfolio management perspective?
12:08
Unidentified speaker (speaker_1): Well, uh, our favorite, two-sided complexity, but with more attention to the call protection. And importantly, protection has to be economically priced. Paying an enormous premium for the strongest historical prepayment story isn't necessarily the right answer. I'd also caution about generic
12:28
Unidentified speaker (speaker_1): higher coupon exposure, particularly if increasingly sophisticated servicing technology makes refinancing easier during the next rally.
12:36
Unidentified speaker (speaker_0): Well, and there's a related implication to the one you mentioned, Steven, right? It is all about genies. If Congress eventually expends Expense first-time buyer assistance, even if you're right, it will take time, well then disproportional share. of incremental mortgage production could flow through FHA as well as
12:56
Unidentified speaker (speaker_0): other government programs.
12:58
Unidentified speaker (speaker_1): That could mean additional Genny May supply, particularly in production coupons. And if rates subsequently decline, those same mortgages may face substantial refinancing exposure, including the streamlined refinancing programs. So investors could face greater relative supply. when the mortgages are originated and more call
13:18
Unidentified speaker (speaker_1): risk later,
13:19
Unidentified speaker (speaker_0): which is a rather unfortunate combination, I'd say. The politicians promise more affordable housing, borrowers hopeful lower payments, and mortgage investors discovered that that embedded option hasn't really gone anywhere.
13:34
Unidentified speaker (speaker_1): Perhaps that's the main takeaway here. The election may change the policy. conversation, but it doesn't eliminate the fundamental trade-offs in mortgage investing.
13:44
Unidentified speaker (UNKNOWN): Well,
13:44
Unidentified speaker (speaker_0): let's conclude on a high note. Let's say this: Washington can change the rules, but repealing negative convexity? Well, that is going to be tough.
14:02
Unidentified speaker (UNKNOWN): All right,
14:02
Unidentified speaker (speaker_0): Steve. Steven, you mentioned the increasing sophistication of servicing technology just a moment ago. And I think it deserves some more attention and more digging on our part. You and I spend decades, the best decades of our life, Steven, I must say, modeling mortgage prepayments
14:22
Unidentified speaker (speaker_0): using interest rate incentive, loan balances, credit score seasoning. and all the usual and unusual suspects. But I wonder whether we are approaching a point where the technology used to originate a mortgage becomes almost as important as the borrower's characteristics.
14:42
Unidentified speaker (speaker_0): Artificial intelligence, ever heard of it, Stephen, is changing how servicers identify and approach borrowers. What do you think, Steven, that does to the traditional prepayment modeling?
14:56
Unidentified speaker (speaker_1): Uh, quite a big change here. Servicers already know their customers' mortgage rates, balances, payment histories, and property characteristics. AI allows them to combine that information with market data to build a broader picture of the borrower.
15:14
Unidentified speaker (UNKNOWN): Using
15:14
Unidentified speaker (speaker_1): predictive models, they can more accurately identify which borrowers are more likely to jump on a refi when rates dip, or which will be looking for cash-out opportunities, even if overall rates don't provide an incentive. And in either case, AI means finding them faster and automating much more of
15:34
Unidentified speaker (speaker_1): the origination process. Rocket, for example, reported that in an AI voice platform handled 65% of inbound servicing calls in the third quarter. And Pennymax said it's developed 22 AI agents across origination workflows. These aren't hypothetical capabilities anymore.
15:54
Unidentified speaker (speaker_0): That, Stephen, that reminds me of the minority report movie. Tom Cruise was trying to predict crimes. before they even happened. Mortgage services are trying to predict refinancing before borrowers even realize they want one. Here is the question I find
16:14
Unidentified speaker (speaker_0): really interesting. If Rocket uses AI to capture a borrower who would otherwise refinance Finance with uh, let's say Wells Fargo, the mortgage still prepays. Nothing really changes. from the perspective of MBS investors. So where does the incremental prepayment risk actually come from,
16:34
Unidentified speaker (speaker_0): Steven?
16:35
Unidentified speaker (speaker_1): That's a really interesting point. And I think one of the keys is timing. The borrower may have eventually refinanced with another lender, but the technology arms race here means that borrowers could be reached much sooner, meaning they're their prepayment comes much sooner to MBS pools. But importantly, I
16:55
Unidentified speaker (speaker_1): think this also means some services are going to be a lot better at identifying borrowers who would not have otherwise refinanced. In the past, a short-term dip in rates may have left many borrowers remaining in their loans because they didn't react quickly enough, or lender capacity constraints left them hanging too long. If
17:14
Unidentified speaker (speaker_1): servicers can reach them sooner and process the loan faster, the efficiency of the refi response that MBS investors have to model is much higher. If AI reduces processing costs and improves outreach and identifies financing alternatives that borrowers hadn't
17:34
Unidentified speaker (speaker_1): considered, it can potentially change the exercise of the mortgage option itself. So prepayment models that have been trained on historical data trends, well, they need to be adjusted to reflect the higher risk now.
17:47
Unidentified speaker (speaker_0): Well, that, Steven, would create very different implications across the coupon stack. In premium mortgages, more efficient Tradition refinancing can obviously weaken traditional coal protection stories.
18:01
Unidentified speaker (UNKNOWN): A
18:01
Unidentified speaker (speaker_0): borrower who let's say, previously ignored a modest rate incentive might suddenly find the transaction almost effortless. But I think the discount story is even more intriguing. Consider a homeowner with a 3% mortgage, let's say, substantial accumulated
18:21
Unidentified speaker (speaker_0): equity, and some credit card debt costing him 20%. Conventional refinance models say that the borrower is deeply out of the money. But from the household perspective, accessing equity to consolidate expensive debt could still make economic sense, right?
18:40
Unidentified speaker (speaker_1): Yeah, for sure. I mean, I think that HELOC throws a bit of a wrench into that argument, but it's not always a better choice and really depends on how deeply out of the money the more. mortgage is and what kind of alternative financing is available to borrowers. But in general, yes, AI could make identifying those opportunities
19:00
Unidentified speaker (speaker_1): considerably easier. And for discounted MBS, it can mean faster principal recovery at par without a conventional rate-driven refinancing incentive.
19:09
Unidentified speaker (speaker_0): Oh Well now, now, Steven, you've got me thinking about the Terminator. The machines.
19:15
Unidentified speaker (speaker_0): aren't necessarily coming for humanity, but they may be coming after our prepayment modeling assumptions.
19:24
Unidentified speaker (speaker_1): So this time I'll be back means your principal is coming back in par.
19:29
Unidentified speaker (speaker_0): Precisely, precisely. The technology is advancing and the relative value implication is potentially significant. Whether our prepayment models are keeping up. well, that's another question.
19:46
Unidentified speaker (UNKNOWN): All
19:46
Unidentified speaker (speaker_0): right now, Steven, we've been discussing how AI could make mortgage refinancing easier. But you know what? Washington may be preparing to remove some of the obstacles themselves. An October 22nd executive order directs FHFA to expand a title acceptance pilot and consider
20:05
Unidentified speaker (speaker_0): reducing appraisal requirements, income documentation burdens, and loan-level pricing adjustments, lovingly known as LLPA. The state of objective here, well, is cheaper refinancing for homeowners. I am wondering, however, if you think mortgage
20:25
Unidentified speaker (speaker_0): investors could end up paying part of the bill.
20:29
Unidentified speaker (speaker_1): This is a great question. And let's start with distinguishing what has actually been directed and what remains under consideration. The order doesn't establish a comprehensive GSE streamlined refi program, but it does direct the FHFA to expand that recent title acceptance pilot to a
20:49
Unidentified speaker (speaker_1): nationwide audience and expand. the LTV eligibility criteria. Of course, it is subject to safety and soundness determinations. And from an LLPA's perspective, the order directs FHFA to consider LLPA reductions and appraisal and income verification
21:08
Unidentified speaker (speaker_1): waivers. But it also notes that these decisions should be made when enterprise data supports support.
21:13
Unidentified speaker (UNKNOWN): them.
21:14
Unidentified speaker (speaker_1): And I think unlike the FHA and VA streamlines, conventional refinances generally remain subject to federal ability to repay verification requirements. So it isn't a situation where borrowers suddenly refinance without any underwriting.
21:30
Unidentified speaker (speaker_0): Fair enough. But even reducing transaction costs could matter, right? FHFA estimated title acceptance savings of $500 to $1,500 and $1,100 per eligible refinance transaction. The reduction in the fixed cost of refinancing could be most
21:50
Unidentified speaker (speaker_0): substantial for low loan balance borrowers, right? $1,100 represent about 73 basis points of $150K mortgage balance compared to only 17 basis points.
22:03
Unidentified speaker (speaker_0): for 650k loan, right? And LLPA, Steven, could also be consequential, especially for low credit score and high LTV borrowers trying to refine. For example, the GECs currently charge lenders 187 basis points upfront fee for
22:23
Unidentified speaker (speaker_0): purchase loan with a credit score of 690 and LTV of 84. They charge 250 basis points for a refinanced loan in the same bucket, implying a 63 basis points roughly in refire premium. The order does not specify how exactly L LPA
22:43
Unidentified speaker (speaker_0): should change. So it is not clear if a potential streamline streamlined program would eliminate them entirely. or just eliminate the Refi premium relative to purchase loan pricing. What is clear is the overall direction of the order, which is to find ways to reduce Refi
23:03
Unidentified speaker (speaker_0): friction for borrowers that are already GSC guaranteed.
23:09
Unidentified speaker (speaker_1): Yeah, I agree. And we have some historical data that might give us a clue as to how. reducing upfront costs could impact speeds. The DSEs already have an appraisal waiver program that has allowed many borrowers to refinance without needing to pay the $400 to $700 typical fee.
23:29
Unidentified speaker (speaker_1): Historically, loans originated with appraisal waivers have prepaid faster than comparable loans requiring appraisals, particularly when refinancing refinance incentives. become meaningful. For newer loans at a 50 basis points refinance incentive, those with waivers prepaid at 21.2
23:48
Unidentified speaker (speaker_1): CPR versus 15.7 for traditional appraisal loans. And the gap extends to about a 10.2 CPR difference at 100 basis points.
23:59
Unidentified speaker (speaker_0): That's actually fascinating. Another thing I found interesting is that the waiver versus appraisal gap was much more significant in many of the specified pool stories than in generic collateral, which probably intuitive to some extent. At 50 basis points in the money incentive, generic pool cohort showed only about 4.5
24:19
Unidentified speaker (speaker_0): CPR faster speed for waiver loans. But loans in 85k max uh category were 10 10.5 and a
24:27
Unidentified speaker (speaker_0): CPR faster when they had waivers. That implies the savings and fixed cost of the refi were more important to the lowest loan balance borrowers.
24:39
Unidentified speaker (speaker_1): It's important that we don't assume the waiver itself caused the entire difference. The GSC program The itself introduces some bias in borrower selection, origination channels, seasoning. and other characteristics that can influence the results. Still, the evidence suggests that refinancing friction and borrower
24:58
Unidentified speaker (speaker_1): characteristics interact in ways that traditional specified pool assumptions need to capture.
25:04
Unidentified speaker (speaker_0): Which means it's time for another movie reference, Steven. The Italian job. The entire operation depends on removing obstacles between you and. something valuable. In this case, policymakers are potentially making it easier for borrowers to exercise their
25:23
Unidentified speaker (speaker_0): refinancing option. But Steven, does that mean investors should abandon low loan balance protection?
25:32
Unidentified speaker (speaker_1): No, certainly not. It means evaluating what portion of the protection comes from fixed refinance. refinancing costs versus more durable borrower constraints. And I think the PIW analysis even shows there might be a bit of a sweet spot in the middle of the loan balance flavor stack. In a rally,
25:52
Unidentified speaker (speaker_1): lower transaction costs could weaken some historical prepayment advantages, but right now, with most legacy low coupons still deeply out of the money, the immediate impact of any type of changes that the uh FHFA makes due to this order would probably be limited.
26:10
Unidentified speaker (speaker_0): So then the broader message is that lower refinancing costs may help homeowners, but they could also change the value of spec pool pay apps.
26:21
Unidentified speaker (speaker_1): Exactly. And until we know the eligibility rules, speech changes, and implementation. details, the size of that effect remains uncertain.
26:36
Unidentified speaker (UNKNOWN): Well,
26:37
Unidentified speaker (speaker_0): Steven, happy National Fast Food Day. I do think we have inflicted enough mortgage analytics on our listeners for one week. We've covered housing policy, artificial intelligence. and an increasingly complicated businesses of refinancing. Anything else you would like to add
26:57
Unidentified speaker (speaker_0): before we call it quits?
26:59
Unidentified speaker (speaker_1): Maybe just sum it up that uh between AI and Washington, borrowers may soon have more ways to refinance than investors have ways to model them.
27:08
Unidentified speaker (speaker_0): Well that's really reassuring Steven. Alright, enough with bad news. We will be back next Monday with more mortgage mysteries puzzles in our forecasts. Until then, may the pulse be with you.
27:23
Unidentified speaker (UNKNOWN): I
27:24
Unidentified speaker (speaker_0): am Kirill Kryloff
27:25
Unidentified speaker (UNKNOWN): and
27:25
Unidentified speaker (speaker_0): I'm Steven Cheer. Thank you for listening and take
27:29
Unidentified speaker (UNKNOWN): care.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.

More from Baird Fixed Income Insights: Convexity Pulse

All insights