Mortgage rates hit 6.76%. The MBS buyback bought 16 basis points.
Fannie and Freddie added $38B of agency bonds since December. The spread they target has gone sideways since January. The 10-year moved 99 points.
The Editors · 9 min read ·
The 30-year fixed mortgage averaged 6.76% in the week of 10 September 2026, up from 5.98% in late February. Almost none of that increase was about housing. The 10-year Treasury note closed at 4.96% on 11 September, its highest since October 2023 and 99 basis points above its February low of 3.97%. The gap between the two, what the market calls the mortgage spread, barely moved at all: 1.89 percentage points in January, 1.93 in the first week of September.
That gap is the one number US housing policy has been aiming at all year. In early January, Fannie Mae and Freddie Mac said they would speed up purchases of agency mortgage bonds, and the White House put a figure on it: up to $200 billion. The idea was simple enough. Buy the bonds, push their yields down toward Treasuries, and the rate a borrower pays falls without the Federal Reserve doing anything.
It worked for about a month. The spread compressed roughly 16 basis points in January and has gone sideways ever since. Over the same eight months the 10-year added 99. If you're waiting on Washington to lower your mortgage rate, the number you're actually waiting on is the 10-year.
What the spread is, and why it became a policy target
A 30-year mortgage is priced off the 10-year Treasury, not off the Fed funds rate. Most 30-year loans get paid off or refinanced in under a decade, so the 10-year note is the closest thing to a maturity-matched risk-free benchmark. Lenders and MBS investors then add a margin on top for prepayment risk, servicing, credit, and the cost of holding the paper.
That margin is the spread, and for three decades it was remarkably stable. From 1990 through 2021 it averaged 1.67 percentage points, computed weekly from Freddie Mac's survey rate against the 10-year constant maturity yield. Then it blew out. It peaked at 3.13 points in the week of 1 June 2023, as the Fed shrank its MBS book and rate volatility made mortgage paper expensive to hedge.
The 1.46 points of extra spread at that 2023 peak cost a $400,000 borrower close to $380 a month. That is why it became a policy target. Compressing the spread is the one lever that lowers mortgage rates without lowering rates for everyone else, and without waiting on the Fed.
The spread moved in January, then stopped
Here is the annual average, computed from Freddie Mac's weekly survey minus the 10-year constant maturity yield for the same week.
The year-over-year drop is real. 2026 is averaging 1.96 points against 2.30 in 2025. What the annual number hides is when it happened. The monthly series:
| Month | Spread | 30-year | 10-year |
|---|---|---|---|
| Sep 2025 | 2.23 pp | 6.35% | 4.12% |
| Dec 2025 | 2.04 pp | 6.19% | 4.15% |
| Jan 2026 | 1.89 pp | 6.10% | 4.22% |
| Mar 2026 | 1.94 pp | 6.18% | 4.24% |
| Jun 2026 | 2.02 pp | 6.49% | 4.47% |
| Aug 2026 | 1.99 pp | 6.67% | 4.68% |
One step down in January, then eight months of noise between 1.82 and 2.13. Scotsman Guide's reporters watched the same thing happen in real time and described it in February: spreads "dipped to the low- or mid-180s after the MBS announcement yet have returned to the 190 to 200 range in the six weeks since." Two different methods, the same answer.
The other thing the monthly table shows is that the spread was already falling before anyone announced anything. September 2025: 2.23. December 2025: 2.04. That's 19 basis points of compression in a quarter, with no program running. January's 16-point step looks less like a break in the series and more like one more month of a normalization that started in 2023 and has been grinding lower every year since.
What Fannie and Freddie actually bought
The monthly volume summaries are public, and they settle the question of scale.
Fannie Mae's July 2026 summary puts agency securities in its retained portfolio at $99.7 billion, against $71.5 billion at the end of December 2025. Net add: $28.2 billion. Freddie Mac's July volume summary shows $54.7 billion against $44.6 billion in December. Net add: $10.1 billion.
Combined, that is $38.3 billion of net agency bonds in seven months, against a headline of up to $200 billion.
The direction since spring is more telling than the total. Fannie's agency book peaked at $110.5 billion in April and has fallen every month since, down to $99.7 billion in July. Freddie's peaked at $56.3 billion in May and has slipped to $54.7 billion. Together the two are $12.2 billion below their April high. Whatever the buyback is now, it isn't a bid that grows every month.
Capacity isn't the constraint either. Freddie's portfolio is capped at $225 billion under its agreement with Treasury, and the balance that counts against the cap was $160.9 billion at the end of July. There is room. It just isn't being used.
The 10-year did the work
Run the two moves side by side from the February trough:
| 27 Feb 2026 | 11 Sep 2026 | Change | |
|---|---|---|---|
| 10-year Treasury | 3.97% | 4.96% | +99 bp |
| 30-year mortgage | 5.98% (26 Feb) | 6.76% (10 Sep) | +78 bp |
| Spread | 1.96 pp | 1.93 pp | -3 bp |
The mortgage rate tracked the Treasury almost one for one, and the spread absorbed a rounding error. The last three sessions make the point at a smaller scale: the 10-year went 4.83%, 4.95%, 4.96% across 9, 10 and 11 September, 13 basis points in three days, as the Bureau of Labor Statistics reported August CPI at 3.4% year over year with core at 2.4%, and the market moved into the FOMC's 15-16 September meeting.
Thirteen basis points in three days. The entire remaining distance between today's spread and its thirty-year norm is 26. The bond market moves half the policy prize in an ordinary week, and the same supply pressure shows up everywhere else in the long end, from the corporate calendar competing for the same buyers to what the Fed's own inflation gauges are signalling into this meeting.
What it's worth to a borrower
Take a $400,000 loan, 30 years, principal and interest only.
| Scenario | Rate | Monthly | vs today |
|---|---|---|---|
| Today | 6.76% | $2,597 | |
| Spread back to its 1990-2021 norm | 6.47% | $2,520 | -$77 |
| 10-year back to February, spread unchanged | 5.82% | $2,352 | -$245 |
Finishing the spread job is worth $77 a month. Getting the 10-year back where it was in February is worth $245. That ratio is the whole argument. The bond market holds roughly three times the money that housing policy does here, and it moves faster.
There's a version of this for savers too, which runs in the opposite direction: the same long yields that are raising your mortgage quote are not showing up in what banks pay on deposits.
What would prove this wrong
Three things, honestly.
The counterfactual is unobservable. If the GSE bid is what's holding the spread at 1.95 instead of letting it drift back toward 2.3, then the program is working and the evidence looks exactly like nothing happening. Wolf Street made this argument on 3 September: without the buybacks, mortgage rates might be over 7%. Nobody can price that claim either way.
Net balances are not gross purchases. The agency-securities line in the monthly summaries is a stock, not a flow. Both firms buy and let paper run off at the same time, so a flat balance is consistent with steady buying against steady liquidations. It rules out a large growing bid. It doesn't rule out a bid.
The direction can flip. If the FOMC hikes on 16 September and the long end reads it as a credible inflation fight, the 10-year can fall and pull mortgage rates with it. That would still be the bond market doing the work, which is the point, but it would make the headline rate move down instead of up.
What to watch now
The August volume summaries land in late September and will show whether Fannie's agency book fell for a fourth straight month. Watch the 10-year through the FOMC meeting and the projections that come with it. And watch whether the spread does anything at all outside its 1.82 to 2.13 band, because nine months of data say it won't.
One last thing for anyone sitting on a purchase waiting for relief. The spread has already given back most of what it has to give. What's left is 26 basis points, and the last three weeks of Treasury selling took more than that. Price the loan against the bond market, because that is what's pricing you.
Sources
- Freddie Mac Primary Mortgage Market Survey, 10 September 2026
- US Treasury daily par yield curve rates, September 2026
- US Treasury daily par yield curve rates, February 2026
- Fannie Mae Monthly Summary, July 2026
- Freddie Mac Monthly Volume Summary, July 2026
- Bureau of Labor Statistics, Consumer Price Index, August 2026, released 11 September 2026
- Federal Reserve FOMC calendar
- FRED: 30-Year Fixed Rate Mortgage Average (MORTGAGE30US) and 10-Year Treasury Constant Maturity, weekly (WGS10YR), used for the 1990-2026 spread series
- ResiClub Analytics on the $200 billion directive, 9 January 2026
- Scotsman Guide on early spread movement, 24 February 2026
- Wolf Street on the stuck spread, 3 September 2026
This is not financial advice.
