The Full Intersection between student loans and mortgage

Paid Subscriber Only Post through Thursday (Thank you for your continued support)

Mandatory repayment of the SAVE student loans isnt just looming a large payment shock for a big chunk of FHA borrowers. Its looming a large payment shock for roughly 4.8 million loans across the FHA, VA, FNMA and FHLMC portfolios when student loan repayment restarts for SAVE borrowers in early 2026 assuming the student loan overhaul in the “One Big Beautiful Bill” remains largely unchanged from how it passed the House. It is unclear how that shock will be absorbed in those households, be it reduced savings/investment funds, reduced consumption of goods/services, missed credit card/revolving repayments, missed auto payments, or missed mortgage payments, but something will have to give and with wage garnishment in play for defaulted student loans its very unlikely to be missed student loan payments.

Background

The SAVE (Saving on a Valuable Education) student loan repayment plan was introduced by the Biden administration in August 2023 as a replacement for the REPAYE plan, part of a broader effort to make student loan repayment more affordable. SAVE was an income-driven repayment (IDR) plan that bases monthly payments on a borrower’s income and family size. Comparted to other IDR plans it was particularly generous both excluding 225% of the applicable poverty level for the borrowers family size from income (vs. 150% for other IDR plans) and capping payments at 5% of discretionary income for undergraduate loans/10% for graduate loans or a weighted average if both exist. It also did not accrue interest if a borrower’s payment didnt cover the full interest, and had generous forgiveness terms.

Many borrowers chose SAVE because it offered the lowest monthly payments available and effectively eliminated negative amortization. The problem was the plan was legally enjoined in July 2024 after multiple states challenged the administration's authority to implement such a sweeping change to IDR plans without congressional approval. That injunction was affirmed earlier this year. As a result of the injunction SAVE borrowers were placed in administrative forbearance, so they did not need to make monthly repayments as the courts sorted it out. To be sure, these SAVE borrowers were subject to repayment for the roughly 9 month period between student loan resumption in October 2023 and SAVE injunction in July 2024 and even with negative consequences of non-payment (collections, credit reporting etc.) suppressed during the 12 month student loan resumption on-ramp period, many likely were albeit, and importantly, at the much lower payment amount of the SAVE plan. As of March 31st, 7.84 million borrowers were still in SAVE forbearance

Income Driven Repayment Plans worked for more than just low income borrowers

When you hear income driven, it evokes an image of relief tailored primarily to low-income borrowers struggling to make ends meet, and many such borrowers were indeed in these programs. But IDR plans like SAVE also made strategic financial sense for higher-income borrowers with large balances. Because monthly payments are capped at a fixed percentage of discretionary income, high-balance borrowers could limit their payments even if interest accumulated, and ultimately qualify for loan forgiveness after 20 or 25 years. For borrowers in public service, this timeline shortened to 10 years under PSLF. In effect, IDR shifted the question from "how fast can I repay this loan?" to "how little can I pay on this loan before it is forgiven?" As a result, these plans often made the most sense for mortgaged homeowners who also carried large student loan debt, and not just FHA borrowers.

Indeed the studentaid.gov data (as of March 31st) bears out that the SAVE plan in particular, but all income driven repayment plans in general (highlighted in yellow) carried high student loan balances.

What portion of Mortgagors carry student loan debt?

As I wrote in my first article on the intersection of FHA loans and student debt, the FHA in a 2021 press release cited that 45% of their first time homebuyers (~80% of their borrowers) carried student loan debt, but that is slightly contradicted by their Mortgage Insurance Funds annual report to Congress in 2021 which cited 40%, which I now assume is the more correct figure. (https://www.hud.gov/sites/dfiles/Housing/documents/2021FHAAnnualReportMMIFund.pdf)

What about the other portfolios? FNMA reported last November that about 18% of applicants carried student loan debt (https://www.fanniemae.com/research-and-insights/perspectives/providing-greater-certainty-through-enhanced-risk-management)

I am unable to find any literature on FHLMC but given how similar the FNMA and FHLMCs portfolios are it is likely to be pretty similar.

Surprisingly, given the Education benefits available to Veterans, a large chunk of VA borrowers also carry debt. 27% of Veterans took out student loans in the 2015-2016 school year. (https://www.pew.org/en/research-and-analysis/articles/2021/09/13/veteran-student-loan-debt-draws-new-attention?)

Notably the two datapoints I can anchor to are the (something a little less than) 40% of FHA borrowers who originated in 2021 and the something close to 18% of FNMA borrowers who originated in 2023.

Extending my “student loan detection” algorithm to the other portfolios?

Unlike FHA with annual MIP data, the GSE MBS disclosure data does not provide the amount of PMI a borrower is paying. Thus I calculate PMI on all applicable GSE loans using a representative rate sheet accounting for all the major factors (LTV credit score etc.) so I get an accurate picture for each GSE loan what the non housing debt load is.

Armed with anchoring figures of ~40% for FHA in 2021 and 18% for FNMA in 2023 (with a rough expectation to find 25-27% I applied my algorithm originally explained here:

and tweaked here

But that revealed a problem, I was finding way more student loan debt than I was expecting in the VA, FNMA, and FHLMC portfolios.

Mo Income, Mo Debt

Because the old algorithm was detecting more student loan debt in the GSE portfolios than the target 18%, it seemed likely there was a stronger relationship between income and monthly non housing related debt load (car, student loan, cards and other debt) than I was accounting for (I only escalated expectations on auto debt for higher income folks). Indeed there is, a somewhat strikingly consistent one.

the consistency of which becomes even more apparent when you view it as the per borrower incremental increase over the last 10k of income.

and even moreso when you take the avg of the last 5 10k increments. The higher income increments smooth out really well since there just arent that many borrowers at these income levels so subject to more volatility at an individual bracket.

Here are the raw number of loans for each bracket in my dataset (loans originated between 2018/2019 and 2023 which I can confidently match to the HMDA origination record) to get a sense for the income levels in the different housing portfolios

Armed with this insight I simplified the methodology to just calculating the housing debt each month (principal, interest, taxes, insurance, mortgage insurance (MIP or PMI), HOA), and then calculating estimated non-housing debt as a straightforward 110 dollars (a bit above the avg FHA rise) per month per each 10k/year of income, So a single borrower making 45000 a year would have an expected 495/month in non housing debt and a loan with 2 borrowers making 95000 a year would have an expected 2090/month in non housing debt. The “expected” non housing debt creates a threshold where if the loan carries more I assume that “excess” monthly debt is student loan debt. Essentially the algorithm interprets anomalously high non-housing debt on an individual mortgage as student loan debt. When I apply that 110 per borrower/per 10k of income algorithm across the FHA loans originated in 2021 the algorithm detects student loans on 38.78% of loans and across the FNMA loans originated in 2023 detects student loans on 17.99% of the loans originated.

Then applying the algorithm across every loan originated since 2018 (FHA/VA) or 2019 (FNMA/FHLMC) that remains in MBS pools as of April 2025, it finds

Consistency in SAVE

The simplified algorithm for detecting student loan debt identifies loans with particularly high monthly non housing debt. Whats interesting is how much higher than the threshold for student loan detection most of the loans are. For example when I run the VA portfolio using an expected non housing debt per borrower per month of 90 dollars/month per 10k income, then 100 and 120 dollars a month, I get the following results.

Once above 100 dollars a month per borrower per 10k of income of expected non housing debt, the percentage of loans detected with SAVE student loans only drops 2 pts from 100 to 120, whereas the percentage of loans detected with Other student loans drops 6 pts

Projecting to the full portfolio

As I wrote in my last post on the subject, extending this to the full portfolio is a little more difficult because I don’t have visibility into the origination characteristics of loans originated prior to 2018/2019. Older loans that originated with the borrowers carrying student loan debt are more likely to be student debt free in 2025 (paid off or forgiven) and also more likely to have lower balances now. To account for this, I adjust down the percentages of each portfolio also having student loan debt to

FHA: ~30%

VA: ~24%

FNMA/FHLMC: ~17% each

and the portion of SAVE borrowers in each portfolio to

FHA: ~19%

VA: ~15.5%

FNMA/FHLMC: ~9%

which results in full portfolio numbers of roughly

FHA: 8 million loans, 2.4 million with student loan debt, 1.52 million in SAVE plan

VA: 3.7 million loans, .89 million with student loan debt, .573 million in SAVE plan

Each GSE: 15 million loans, 2.55 million with student loan debt, 1.35 million in SAVE plan.

All portfolios: 41.7 million loans, 8.4 million with student loan debt, 4.8 million in SAVE

I have no visibility into the ~10 million loans that FHFA classifies as “Other Conventional” comprising bank held loans, private securitizations etc. but I assume its roughly similar.

Calculating the Payment Shock

The payment shock (for a more full discussion on how I calculate new IDR payments and SAVE plan payment amounts for each loan see my prior two articles linked above) that will hit the SAVE plan loans needs to be expressed as both the full payment shock as well as the payment shock over SAVE. The full payment shock is the increase from what the borrower in forbearance is paying, zero, and what the borrower will be paying under the new IDR when they are transitioned over to it. The shock over SAVE is the new IDR payment amount less the SAVE amount the borrower was previously expecting to pay (and may have been actually paying from September 2023 - July 2024. ) Those average payment shocks for each portfolio are:

FHA: Avg Full Shock $520/month || Shock over SAVE $375/month || Avg student loan balance $53646

VA: Avg Full Shock $694/month || Shock over SAVE $471/month || Avg student loan balance $68117

FNMA: Avg Full Shock $1019 /month || Shock over SAVE $645/month || Avg student loan balance $100814

FHLMC: Avg Full Shock $988 /month || Shock over SAVE $629/month || Avg student loan balance $97565

Distribution of the Payment Shocks

Since the New IDR payment as well as SAVE payment are heavily dependent on borrower income, the payment shocks are smaller for borrowers with smaller incomes and larger for those with larger incomes. The following tables show the payment shock in each origination income bracket up to 400k for each portfolio.

FHA

VA

FHLMC

FNMA

How this plays out

As a reminder from prior posts, Once the “One Big Beautiful Bill” becomes law (and of course assuming that the student loan overhaul within it is largely unchanged, or the loan repayment portion of it anyways) the timeline starts. As required in the legislation, all existing income driven repayment plans (including SAVE) must be converted to the new IDR plan within 9 months. Given the time required for rulemaking and servicer implementation the actual repayment requirements for SAVE borrowers likely wont hit until the end of that 9 months, so in the first few months of 2026.

Whats next?

GNMA MBS data for May should drop early this week so I will have the data for that out the day after it drops, I will also have an update to the Total Treasury model out soon since the Social Security Trust fund adjustment payment should hit tomorrow, solidifying another variable in just how much runway the Congress has to pass the “One Big Beautiful Bill”

As always, Thanks for Reading and your continued support.

Best,

John