Revised Impact of Student Loan Repayment on FHA portfolio

Some ways a bit more, some ways a bit less

On Tuesday I unlocked a previously paid post on the upcoming impact of student loan repayment on the FHA portfolio, which largely hasnt hit yet because most of the FHA borrowers who have student loans were in the still in forbearance SAVE plans. If you havent read that post yet (and free subscribers probably havent since substack didnt seem to email you when I converted the post from paid to Everyone), I highly suggest reading that first, though be cautioned, the loans wont be going to the Repayment Assistance Plan but rather a redefined income driven plan as I describe below.

Some Incorrect Mechanics

Unfortunately, I misinterpreted the pending student loan overhaul legislation and incorrectly described that SAVE plan borrowers would be forced into the new Repayment Assistance Plan (RAP). They wont, instead, initially, all the student loans in an income driven repayment (IDR) plan which includes currently SAVE plans, PAYE plans as well as the statutory Income Based Repayment plan (IBR) will be forced into a new IDR plan and then have the option of moving to RAP when it eventually gets implemented. So the real sequence is

(SAVE and PAYE and IBR) → New IDR → (optionally for existing borrowers) RAP

With a likely timeline of:

1. Aug 1, 2025 - “Big Beautiful Bill” passes, student loan overhaul becomes law

2. Nov 1 ,2025 - Interim Final Rule is issued on move to New IDR plans (The legislation waives negotiated rulemaking so it can move quick)

3. May 1, 2026 - Servicers are required to have switched over all borrowers (possible some move earlier)

4. May 1, 2026 - Interim Final Rule is issued for new RAP plan'(legislation provides 18 months but I doubt it takes that long with no negotiated rulemaking)

5. Feb 1, 2027 - Servicers must implement RAP and borrowers have option to switchover

RAP implementation could be longer or shorter but it doesnt matter to the payment shock because that will all be felt in the new IDR plan initially and the legislation requires switchover within 270 days of enactment.

What payment will the new IDR plan require?

The new IDR plan is an update of the existing Income Based Repayment plan defined in section 493C of the Higher Education Act. The proposed legislation alters it to make the payment 15% of Discretionary Income (AGI minus 150% of poverty line for family size) for all borrowers. The current version of this IBR is 10% for borrowers who started borrowing after summer 2014 and 15% before.

Rerunning the loan level calculations on the 3 million FHA loans I can link to HMDA (and the 40% of whom I calculate has student debt burden), the new IDR payment is on avg a bit higher than what RAP will offer increasing the payment shock even more.. It varies by borrower situation, lower income borrowers with larger families (who thus benefit more from the 150% poverty line offset) may see a decreased payment, while single borrowers with no children above $100,000 will see a much larger payment vs. RAP.

It also means that existing PAYE borrowers and the post summer 2014 IBR borrowers will also see payment increases upon transition to the new IDR. My calculations indicate there are not too many of these borrowers in the FHA portfolio but they too will face some payment shock

Payment Shock from this transition

SAVE: Goes from $0/month → avg 640/month

Post 2014IBR and PAYE Borrowers: Goes from $425/month → avg 640/month

Updates to the algorithm explained in the earlier post

I am continuing to refine the algorithm described in the post unlocked on Tuesday to more accurately “deduce” student loan debt from the origination DTI and income. It will never be perfect, but the better it becomes, the more accurate picture it will paint.

1. Algorithm now includes the annual MIP in the housing related DTI component for each FHA loan

2. Algorithm now imputes an HOA component in the housing related DTI. It determines the monthly cost as 0 for homes with origination property values < $200,000 and $50/month per hundred thousand dollar property value after that. Coarse, but should help reflect that higher value homes are more likely to have HOAs.

3. I scale the avg 400-450 (depending on origination year) car payments by .9 for borrowers with income less than 50,000, at scale between 50-100k, 1.15 for borrowers between 100-150k, 1.3 for borrowers between 150k-200k and 1.45 above 200k. Ive done this to reflect the likelihood that higher income borrowers in aggregate will drive nicer cars with higher payments.

Is 2.5m FHA borrowers really the right number in SAVE?

I explain my methodologies in detail in my posts so if readers see assumptions they think are off based they can understand how those assumptions would impact the conclusion and adjust their thinking accordingly. Sometimes those readers include future me.

One such assumption I made was scaling the ~1 million borrowers I have calculated are likely in SAVE over the ~3 million FHA borrowers I have the qualifying income for in order to perform the calculation to the rest of the 7.95 million full FHA portfolio., I am having second thoughts on whether that is a safe assumption for two reasons:

1. Per the February FHA performance trends report, there are ~3m loans currently in the full FHA portfolio that were originated prior to 2018, 1.2m before 2012. While the percentage of those loans that also have student loans may well be consistent with the 40% anchor number, the balance characteristics of those earlier loans are likely considerably less due to lower costs of education back then and thus also much less likely to have chosen the SAVE plan. I dont have a great methodology for trying to estimate how many loans this removes but am penciling in 500k to 700k for now.

2. I have started to apply the same assumptions and methodology for determining student debt load to the other housing portfolios (VA, FNMA, FHLMC) and found more evidence of student loans that were likely in SAVE than I was initially expecting. Given that the total across the housing portfolios cant possibly exceed the 7.84m the department of education reported were in the program, it also argues for a lower than 2.5m count in FHA

When I have completed the loan level analysis looking for student loan debt in the other housing portfolios I will have a more confidence in the estimated number of FHA borrowers in SAVE (along with the other portfolios), but that will take some processing time to achieve.

Regardless even if the final number is only say 1.5m, its a materially large number facing a very significant payment shock, and while the borrowers in the other portfolios are likely somewhat less susceptible to it, it looks like there will be materially large numbers in those portfolios as well.

Will the Senate force changes on the student loan overhaul?

It is certainly a possibility and it seems likely that other aspects of the “big beautiful bill” will see some revision

https://www.politico.com/live-updates/2025/05/20/congress/johnson-woos-the-senate-00360400

I doubt though that the loan repayment provisions in the student loan overhaul will be altered much since if anything the rumblings seem to be more spending cuts desired by the Senate not less and the CBO scored the savings from loan repayment at close to 300b over the next 10 years. If that gets reduced then there would have to be deeper cuts to Medicaid/SNAP or less SALT deductions or more taxes somewhere else to keep the deficit impact of the full bill the same. In the various gives and takes I doubt that Republican Senators will focus here for changes. I just dont think they likely perceive political peril if they dont. Perhaps they should given the knock-on effects when these payment shocks hit, but I doubt they do. You will have to judge for yourself.

How did I misinterpret the proposed legislation?

Simply put, by not reading it closely enough and assuming that various articles from news organizations and student loan advocacy groups correctly reported on what was in the bill. Of course I didnt rely on just those sources. I read portions of the bill as well (e.g. the RAP mechanics) and had extensive sessions with ChatGPT about it. Interestingly, the news organizations/advocacy groups and ChatGPT all missed the nuance of the RAP rollout timing, the creation of the new IDR program and what is actually going to happen to all borrowers currently in an IDR plan post enactment. It was the CBO’s scoring of the bill that alerted me to what I had missed. From that scoring (https://www.cbo.gov/system/files/2025-05/HEDWork_Reconciliation2025.pdf)

Horrified at the prospect I had published incorrect information I went back to the proposed legislation and confirmed it. Interestingly when I then went back to chatGPT and informed it of the new IDR plan nuance it caught on and was able to describe the mechanics of how section 493C of the Higher Education Act is being amended, which is where the existing IBR plan is statutorily proscribed.

Using LLMs in the research process, power and perils

I regularly use ChatGPT to both rapidly bring me up to speed on subjects I dont already know and to help answer specific questions about what is said in large documents. In my experience it is a phenomenal tool for that purpose. I NEVER though just trust it, I always confirm either with other sources or detailed reads of the source documents. Unfortunately both my sources were incorrect and I failed to detect that. Its a cautionary tale for the weak point in using LLMs. On that note though, would it kill Congress to make proposed legislation available in a form that shows what the revised part of the US Code will look like. Deciphering language like, such and such provision in such and such act is modified to insert the words “blah blah blah” after the word “blah” in paragraph (d)(2)….. is brutal. Its worse than reading a Word document with Track Changes on thats been modified 50 times in Showing all changes.

Conclusion

Anyways I own and explain my mistakes. Fortunately this one did not materially impact the core message of my earlier post. The student loan repayment impact on the FHA loan portfolio borrowers both has not hit much yet and it likely spans a materially large population of loans (1.5m - 2m) with an average materially large payment shock ($640/mo)

I will provide a final update on all the housing portfolios in the next week or two once I have had the time to process them.

Thanks for reading,

John