The Bond Rallied 15 Basis Points. The Mortgage Took 8.
A mortgage rate is not one number. It is a government bond yield plus a fee for the work of turning a 30-year household loan into something a bond fund will buy. That fee is the primary-secondary spread: the gap between what a lender pays to fund the loan and what it charges the borrower, quoted in basis points on the rate. One basis point is a hundredth of a percentage point. With the 10yr at 5.05% and the gap at 200bp, a borrower pays 7.05%. With the gap at 219bp, the same borrower pays 7.24%. For most of the last year that fee sat near 200bp, and it moved in the borrower's favour: when the bond rallied in the autumn of 2025 the mortgage fell nearly twice as far, 44.0bp against a 23.7bp Treasury move. That has stopped. Over the last 2 survey weeks the 10yr fell 15.4bp and the mortgage fell 8bp, a pass-through of 0.52. The rest was kept as fee. This matters for the midterm because the election cannot move that fee and can still be blamed for what it costs. Congress writes the fiscal path, the fiscal path sets what the long end charges, and the long end arrives at a kitchen table divided by whatever the spread happens to be that week. The vote sets the numerator. The joint sets the multiplier. This column prices the joint.
The instrument is the fee, not the bond
The instrument is the 30yr mortgage rate less the 10yr Treasury over the same survey week. Freddie's survey is collected Monday to Wednesday and stamped Thursday, so the bond leg here is the 3-session average of that window rather than Thursday's close. Pairing Thursday against Thursday instead reads 236.0bp for the week of 22 Oct; the 17.3bp gap between the 2 conventions is the survey's own collection lag, not a market move. Both are reported, and the stricter one is the one used.
On that measure the spread closed at 218.7bp in the week of 22 Oct, from 220.0 the week before and 211.3 a fortnight earlier. The 52-week range is 181.0 to 220.7 against a median of 199.7, and only 3 of 53 weeks closed wider: 23 Oct at 220.7; 26 Nov at 220.5; 15 Oct at 220.0. The widest is 52 weeks old and the second widest 47 weeks old, and the third closed 1 week before this one - which is the point: the extreme is not a spike, it is a level being held.
That is a 1-year extreme, not an all-time one, and the distinction is the difference between a story and a number. Since 1990 the survey has printed 283 weeks at or above 218.7bp - 14.5% of 1921 observations - and the record is 312.5bp in the week of 1 June 2023, when the regional banks were failing and the mortgage market had no bid at all. Today's print is a 94th-percentile week inside a market that is still trading. The fee is re-pricing, not the market closing.
Pass-through went from 1.86 to 0.52
Between 7 Aug and 23 Oct 2025 the 10yr fell 23.7bp and the mortgage fell 44.0bp. Pass-through was 1.86: the mortgage moved nearly twice the bond, because the spread was narrowing at the same time, from 241.0bp to 220.7bp. A borrower got the rally plus 20.3bp of fee compression on top of it. That is what a functioning primary market looks like from the outside.
Across the last 2 survey weeks of 2026 the arithmetic runs backwards. The 10yr fell 15.4bp; the mortgage fell 8. Pass-through is 0.52, and 7.4bp of the rally was absorbed by the spread, which widened +7.4bp over the same fortnight. On the median new home price that is $2,161 a month against $2,087 if the whole Treasury move had arrived.
A pass-through below 1.00 is the normal state of a mortgage market with a wide basis. Lenders price off the forward curve plus a spread, and a wide spread is sticky because it is not a rate: it is a risk premium for holding convexity - the asymmetry that a borrower refinances the moment rates fall and stays put when they rise, so the lender's asset shortens exactly when it is least convenient. Above 1.00 is the exception, and the autumn of 2025 was the exception: the spread was compressing off a 2025 wide and the mortgage chased the bond down.
The buyer who used to take the other side stopped
A spread is the price of a balance sheet agreeing to hold the loan. The largest balance sheet in that business is the commercial bank, and it has been leaving. Banks held $2,740bn of agency MBS in the week of 13 April 2022; on 21 Oct they held $1,895.2bn. That is 845bn gone in 4 years, 187.4bn of it in the last 12 months, and the lowest weekly reading since 10 June 2020.
Nothing replaced it at the same price. The marginal holder is now a mortgage real estate investment trust or an overseas buyer, both of which fund at a floating rate and therefore demand a wider spread to hold a fixed one. Rate volatility is the input they charge for. The MOVE index, which prices Treasury volatility, closed at 94.55 and the VIX at 16.91. Modest volatility, a 219bp basis: the basis is not pricing panic, it is pricing the absence of the old buyer.
This is the part of the chain with no vote in it. No congressional majority restores 4 years of bank portfolio demand in a quarter, and no fiscal package changes what a mortgage REIT pays to fund a position in agency MBS.
What it costs, in the only number a household reads
The median new home sold for $393,300 in the quarter ending 1 Jul, down 4.89% on the quarter and 4.10% on the year - the lowest quarterly print since 1 April 2021. Builders are cutting price. Starts were 1,221k annualised in 1 Sep, 7.43% below a year earlier. Both are the volume channel clearing through price, and both are consistent with a market where financing, not willingness to pay, is the binding constraint.
At 20% down that price is a $314,640 loan. At 7.32% the monthly principal and interest is $2,161. At 7.13% - the 10yr's 5.133% plus the 52-week median spread of 199.7bp - the same loan pays $2,121. The fee is $41 a month, $486 a year, on a house that has already given up 4.10% of its price to get the buyer in the door.
Equivalently, at the median spread the same $2,161 payment would carry a house priced at $400,812 - 1.9% more house than the builder is selling. That gap is the spread, translated into the only unit the buyer negotiates in. It is also why the price cuts and the volume decline can coexist with a payment that has barely improved: the builder and the lender are both absorbing, from opposite ends, the same +19.0bp.
Where the midterm actually reaches this
Three links, and only the first is on a ballot. Congress and the administration set the fiscal path; the fiscal path sets how much duration Treasury has to place in the long end; the long end sets the 10yr. That chain is real and it is already visible in the curve: the 2s10s is 38.0bp on the 28 Oct par close, the 20yr at 5.41% still above the 30yr at 5.37% by 4bp, and the 30s10s at 32.0bp. A curve that charges more at 20 years than at 30 is a curve where the marginal buyer is being paid to take duration, not one where duration is scarce.
Then the chain hits the joint. A gridlock outcome that leaves the fiscal path unchanged leaves the 10yr roughly where it is and the payment where it is. A sweep that adds fiscal impulse moves the 10yr by whatever the market decides, and that move arrives at the payment multiplied by the current pass-through of 0.52. The same 25bp fiscal premium costs the household 13bp of financing under today's pass-through and 46bp under last autumn's - which sounds like an improvement and is not, because the other 19bp of spread is already charged whether the bond moves or not.
The lever that would actually move the fee is administrative, not legislative: how the government-guaranteed channel is priced, what the insurance premium is on a government-insured loan, and the capital rule that decides how much agency MBS a bank may hold against a dollar of equity. None of the 3 is set by a vote on 3 November. That is the asymmetry worth trading. The market will spend the week after the election arguing about the numerator, and the multiplier will be exactly where it is.
Where this is wrong
The call is that the fee holds at or above 215bp into year-end and pass-through stays below 1.00, so a Treasury rally delivers about half of itself to a household payment. It is wrong at 195bp. If the survey-window spread closes below 195 - 4.7bp under the 52-week median of 199.7 - with the 10yr still at or above 5.05%, then the fee is compressing without the bond's help, bank demand has returned, and pass-through should be expected to run above 1.00 again. That is a clean falsification and it is checkable every Thursday at 10:00 ET.
The second falsifier is the absorber. If commercial-bank MBS holdings rise for 4 consecutive weekly prints while the spread stays above 225bp, the balance-sheet explanation is wrong: the banks came back and the fee did not care, which would put the cause in convexity and volatility rather than in who is holding the loan. The MOVE index then becomes the instrument, and 94.55 is the level to watch - a break below 85 with the spread unmoved would say the same thing from the other side.
The third falsifier is arithmetic rather than market. If the median new home price falls another 4.9% next quarter, builders are absorbing the fee in price and the payment channel is not the binding constraint at all - the volume channel is, and the political read moves off the mortgage rate and onto starts, where 1,221k annualised is already 7.43% below a year ago.
Notes
1. Mortgage rate: Freddie Mac Primary Mortgage Market Survey, 30yr fixed, weekly (FRED MORTGAGE30US). Week of 22 Oct = 7.32%; week of 8 Oct = 7.40%. Survey is collected Monday to Wednesday of the stamped week.
2. Bond leg: 10yr Treasury constant-maturity, 3-session average of the matching Monday-Wednesday window (FRED DGS10) = 5.133% for the week of 22 Oct and 5.287% for the week of 8 Oct. Thursday close for the same week: 4.96%, which pairs to 236.0bp. Both conventions are given; the spread series throughout uses the survey-window average.
3. Par curve levels: US Treasury daily par yield curve, 28 Oct close - 2yr 4.67%, 3yr 4.80%, 5yr 4.85%, 7yr 4.95%, 10yr 5.05%, 20yr 5.41%, 30yr 5.37%, 2s10s 38.0bp, 30s10s 32.0bp.
4. Bank MBS holdings: FRED WSHOMCB, all commercial banks, Wednesday-of-week levels. $1,895.180bn on 21 Oct; $2,740.180bn peak on 13 April 2022. Series begins December 2002; the last reading below today's was 10 June 2020. Nominal dollars.
5. Median new home price: FRED MSPUS, $393,300 for the quarter ending 1 Jul, 413,500 the quarter before. Starts: FRED HOUST, 1,221k annualised in 1 Sep, 1,319k a year earlier.
6. Volatility and credit: MOVE 94.55, VIX 16.91 from the 2026-10-29 snapshot; high-yield OAS 3.18% and investment-grade OAS 0.88% as of 27 October.
7. Payment arithmetic: 360-month fixed loan, 20% down, no taxes, insurance or mortgage insurance. Principal and interest only. It is arithmetic on the Q3 price print at quoted rates - not a forecast of either rate or price.
Housing Is Adjusting Through Volume, Not Defaults — the volume channel and the delinquency data; this piece takes the financing channel and the fee on it
The Long End Has Already Voted. It Voted for Gridlock. — the fiscal premium and the 30s10s; here the question is what that premium costs once it reaches a payment
The Curve Peaks at 20 Years, and the Dealer Is the Buyer There — who is forced to hold the duration that sets the 10yr leg of this spread