Who bears the risk of America’s 3% mortgages?

The homeowner keeps making the monthly payment. The investor holding the mortgage might prefer to get the money back sooner. The locked-in rate protects the household budget. For the investor financing the contract, slow repayments can become expensive.

On October 8, 2026, Freddie Mac reported an average rate of 7.40% on new thirty-year fixed-rate mortgages in its survey, up from 7.28% a week earlier. This benchmark covers home-purchase applications meeting the survey’s criteria. It measures conditions for entering the market, while older contracts retain their original rates. ^[1]

That coexistence deserves a closer look. A 3% mortgage shields the household budget, but commits another balance sheet to receiving a low contractual rate for an uncertain period. Mortgage securities, the hedges around them and the borrowing used to buy them connect a domestic financial decision to the wider bond market.

There is an important qualification. An old mortgage far below prevailing rates may already be priced on the assumption that repayment will be slow. The biggest new hedging adjustments may arise elsewhere, among loans closer to an attractive refinancing opportunity. The investigation therefore needs to follow the payment schedule, the price and the owner of the exposure. ^[6]

The valuable contract that is hard to take with you

Consider a hypothetical household that borrowed $300,000 for thirty years at 3%. After five years of regular payments, $266,719 remains outstanding. Its monthly principal-and-interest payment is still about $1,265. Replacing the remaining balance with a loan at 7.40% over the same remaining twenty-five years would require approximately $1,954 a month, an increase of $689.

This calculation excludes insurance, taxes and transaction costs. The 7.40% input comes from the official thirty-year benchmark; applying it to twenty-five years is an illustration, not an observed lender quote. At the original monthly payment, the household could finance only about $172,671 under the second set of assumptions.

Assumptions and method

Monthly payment = P × r / (1 − (1 + r)^(−n)), with r = annual nominal rate / 12. Balance after 60 payments = 300,000 × (1 + r)^60 − payment × ((1 + r)^60 − 1) / r. The replacement payment uses this same balance and 300 months at 7.40%. The observed 30-year survey rate is applied illustratively to 25 years. Current dollars; excludes fees, insurance and taxes.

The cost of giving up the old loan can influence a decision to move even when the replacement home is comparable. Most conventional fixed-rate loans described in Freddie Mac’s offering circular generally cannot be assumed by the buyer; FHA and VA loans operate under different rules. Selling therefore often means paying off the old mortgage and arranging new financing. ^[4]

FHFA has studied this mortgage lock-in effect. In its data through the second quarter of 2024, the average rate on the loans studied was 2.54 percentage points below the rate available on similar new loans. The research links that advantage to selling behaviour. It provides historical evidence for the mechanism, rather than a measurement of mobility in October 2026. ^[2]

Households still face competing priorities. A new job, a separation, an inheritance or the need for more space can outweigh the interest saving. For an investor, these different motives make it impossible to translate the rate gap into a certain repayment date.

One monthly payment, several recipients

A mortgage may remain on the original lender’s balance sheet or enter a pool used to issue securities. In a simple mortgage-backed security, or MBS, investors acquire rights to cash flows from a portfolio of home loans. The servicer collects payments and administers the loans. Part of the interest pays for servicing and the guarantee; the remainder is passed through to the security holder. ^{[3] [5]}

The borrower’s mortgage rate and the investor’s coupon are different numbers. In the portfolio model used here, loans paying 3% support a net coupon of 2.5%, after an assumed half-percentage-point deduction. These are calculation inputs, not an estimate of the fees charged on a particular commercial pool.

Assumptions and method

Functional diagram of an agency pass-through security. Arrows represent claims and payments without monetary scale. Guarantee and servicing follow their contracts; price and prepayment risks remain with the security. The 3% mortgage rate and 2.5% net coupon are hypothetical inputs used in the separate pool model.

An agency guarantee covers the contractual payments it specifies. Freddie Mac guarantees timely interest and scheduled principal on its UMBS, while its offering circular distinguishes that obligation from a guarantee by the United States. The investor still bears uncertainty about early principal repayments and the price at which the security could be sold. ^{[3] [4]}

When a homeowner repays early, investors receive their share of that principal. A buyer who paid above par may have less time to recover the premium through interest receipts. For a buyer who purchased at a substantial discount, receiving principal at par sooner can be beneficial instead. Entry price is as important as the coupon in understanding whose interests are served by an early repayment. ^{[4] [5]}

The same $100 million, returned at two different speeds

Now imagine $100 million of outstanding, otherwise identical mortgages, paying 3%, with twenty-five years left. Every borrower makes the required payment. The exercise changes only the amount repaid ahead of schedule, on top of normal amortisation.

A common market measure is CPR (conditional prepayment rate), an annualised prepayment rate applied to the principal still exposed. A 12% CPR does not mean that $12 million leaves an original $100 million portfolio every year: the balance declines as payments arrive. Our model converts the annual assumption into a constant monthly prepayment rate and applies it after that month’s scheduled principal repayment.

At an assumed 12% CPR, the pool has returned $54.9 million of principal after five years. At 3%, it has returned $26.6 million. The principal-weighted average repayment date moves from 5.90 to 10.95 years. This weighted average life describes when principal arrives. Duration, which measures price sensitivity to interest rates, describes something different.

Assumptions and method

Homogeneous $100 million pool, 3% mortgage rate, 2.5% net coupon, 300 months remaining. SMM = 1 − (1 − CPR)^(1/12). In month t, the surviving pool’s scheduled payment is B × r / (1 − (1 + r)^(−(301 − t))), where B is the opening balance and r = 0.03 / 12. Scheduled principal = payment − B × r; prepaid principal = (B − scheduled principal) × SMM. Coupon = B × 0.025 / 12. Weighted average life = sum(t × principal paid) / ($100 million × 12). Present value = sum(month-t cash flow / (1 + 0.055 / 12)^t). Constant CPR, full prepayment by a homogeneous fraction, no defaults or transmission delays. Total principal equals $100 million under each scenario. The scenarios isolate timing without calibrating effective convexity or an option-adjusted spread.

To give the wait a financial value, discount both payment streams at the same flat 5.5% rate, using the 2.5% net coupon. The calculated value falls from $86.34 million to $77.61 million. The second pool still returns its entire $100 million principal, but later, while paying less interest than the return used to discount the cash flows.

These are model values, with constant prepayment assumptions, no defaults and simplified monthly payments. The calculation isolates a change in the repayment schedule; it does not produce a full price curve or an option valuation. Actual collection-to-distribution delays are also omitted. Freddie Mac UMBS have a contractual payment delay. ^[3]

The $8.73 million difference is therefore neither an observed October loss nor a discount to apply to every old mortgage. It is the difference between two assumptions maintained through maturity. An investor buying after a repricing can pay an already adjusted price and seek a satisfactory return. A seller’s loss and a new buyer’s opportunity can describe the same security at different points in time.

The next refinancing threshold depends on the loan

The attraction of refinancing depends on expected savings, fees, borrower eligibility and how long the homeowner expects to keep the property. Around that decision boundary, a rate movement can sharply change expected repayments. Farther away, it may primarily change the value of cash flows already expected to last a long time. ^{[4] [6]}

A small numerical test makes the distinction visible. Suppose, purely for this example, that a borrower considers refinancing once the new rate offers a saving of at least half a percentage point. At a new rate of 6.4%, an existing 7.5% loan clears that hurdle. At 7.4%, it no longer does. An existing 3% loan remains far below the hurdle in either case.

Assumptions and method

Illustrative indicator in percentage points = existing rate − new rate − 0.5 point. The 0.5-point threshold is chosen for the exercise, without representing a lender rule or a prepayment probability. New rates of 6.4% and 7.4% are hypothetical. Actual refinancing costs and eligibility constraints are not estimated.

The half-point requirement is an illustrative convention, not a lending rule. It shows why “old cheap mortgage” is an inadequate description of the next change in borrower behaviour. As the gap becomes very large, the refinancing option can lose much of its marginal sensitivity even while the security remains highly exposed to a general increase in yields.

The distinction is between a large interest-rate exposure and a rapid change in that exposure. Convexity describes how price sensitivity itself changes as rates move. In the region where rising rates slow prepayments, the cash flows extend and can deepen the price decline. That is negative convexity. Its strength depends on the loan’s position relative to the refinancing decision boundary. ^[6]

A market rate is also an incomplete description of human behaviour. A 2024 preprint by Perotti, Grzelak and Oosterlee explicitly considers behavioural uncertainty and the limits to hedging it. Loan sizes, original rates, seasoning and borrower circumstances all vary. A portfolio average can conceal several distinct areas of responsiveness. ^[11]

The hedge that has to be rebuilt

A manager may seek to keep interest-rate exposure around a chosen target. If slower repayments increase the duration of the securities, the existing hedge may become insufficient. Selling Treasuries, or entering a swap that pays fixed and receives floating, can reduce that exposure. Falling rates can require the opposite adjustment. ^{[6] [8]}

Consider a separate hypothetical portfolio, with market value held constant at $1 billion to isolate the calculation. Its estimated duration increases from four to six years. A local approximation puts its sensitivity to a one-basis-point movement, or 0.01 percentage point, known as DV01, at $400,000 before the revision and $600,000 afterwards. The manager has an additional $200,000 of sensitivity to offset if the original target is to be maintained.

A hedging instrument with an eight-year duration would require an additional short position equivalent to $250 million of market value. That amount describes a hedge exposure, not a $250 million expense or loss. It depends on the assumed durations and ignores non-parallel movements in the yield curve.

Assumptions and method

DV01 ≈ market value × duration × 0.0001. For a constant $1 billion market value, 4 then 6 years produce $400,000 then $600,000 per basis point. Equivalent short exposure = 200,000 / (8 × 0.0001) = $250 million. Local approximation for a parallel shift; no observed trade or loss. This portfolio is separate from the seasoned-mortgage model.

This exercise is independent of the seasoned 3% pool introduced earlier. It does not assign an observed duration to that pool. It illustrates the consequences of a sensitivity revision, whatever combination of mortgages causes it.

If several managers move in the same direction, their sales can weigh on bond prices and push yields higher. The effect depends on order size, timing and the ability of other buyers to absorb the positions. Reducing risk in one portfolio can require an intermediary or another investor to finance and hold more of it.

An insurer matching long-dated liabilities, an index fund and a heavily leveraged investor face different targets and liquidity constraints. An amplification loop requires participants who actually adjust their positions. The existence of prepayable loans alone reveals neither the quantity sold nor the effect on rates.

The identity of the holder matters

The Federal Reserve still held $1.890706 trillion of residential MBS on October 7, 2026. The amount published in H.4.1 the following day is the securities’ remaining principal, rather than market value. On its own, it does not reveal their coupon distribution. ^[7]

A central-bank portfolio operates under different constraints from an investor funding securities through repo, a loan collateralised by those securities. A valuation loss does not automatically produce a private margin call on the Fed’s holdings. The interest received and the cost of its liabilities still have economic consequences. Public ownership changes the constraint; it does not abolish the financial value of time.

The servicer has another type of exposure. A mortgage servicing right, or MSR, provides income in exchange for administering loans. Faster prepayments cut short those revenues. Slower repayments can extend a remunerated relationship, other things equal, while defaults increase servicing costs. Treating servicing rights and mortgage securities as though all mortgage assets respond in the same direction misses this distinction. ^[10]

Hedges can also work as intended. For the second quarter of 2026, Dynex reported book value per share rising from $12.60 to $12.90. It attributed the outcome partly to tighter spreads and hedges that offset much of the adverse rate effect. This is the company’s account of its own portfolio, but it provides a concrete counterexample to the idea that every holder mechanically suffers the same loss. ^[12]

A price decline can arrive before a missed mortgage payment

Risk becomes more urgent when the investor borrows to buy the securities. A repo lender accepts the MBS as collateral after applying a safety discount, or haircut. If the security’s price falls, the collateral is worth less. The lender can demand a partial repayment or additional securities to restore the agreed coverage. AGNC describes this process in its repurchase-agreement disclosures. ^[8]

In a hypothetical position containing $100 million of securities financed with $95 million of repo, initial equity is $5 million. A 3% price decline takes collateral value to $97 million. At an unchanged 5% haircut, it supports only $92.15 million of borrowing. The investor must provide $2.85 million of cash, or additional collateral acceptable to the lender.

Assumptions and method

Hypothetical figures in millions of dollars: assets 100, repo 95, equity 5. After a 3% price fall, assets are worth 97. Maximum funding at a constant 5% haircut = 97 × 0.95 = 92.15; required repayment = 95 − 92.15 = 2.85. Residual equity before an injection = 97 − 95 = 2. The valuation loss of 3 and cash call of 2.85 are not two losses to add together. Hedges, costs and other resources excluded.

The valuation loss is $3 million. The $2.85 million liquidity call is a different operation: a reduction in permitted borrowing. Adding them together as two losses would overstate the damage. Before any outside support, the residual equity in the position falls from $5 million to $2 million. Other assets, hedges or available cash may allow the investor to ride out the shock.

Every household keeps paying in this scenario. The urgency comes from the valuation and financing of the collateral. It could force a sale if available resources are inadequate. Conversely, a liquid buffer or a hedge that generates cash in time can prevent an unfavourable disposal.

At AGNC, repos financing investment securities had a weighted average remaining maturity of thirteen days on June 30, 2026. This measures neither the duration of the mortgages nor an obligation to sell the whole portfolio in thirteen days. It indicates how frequently funding must be renewed. The company also disclosed $7.5 billion of unencumbered cash and Agency MBS, a buffer that includes securities rather than consisting entirely of bank cash. ^{[8] [9]}

Hedging rates leaves another price to watch

An MBS can lose value relative to the instrument used to hedge it. Investors may demand more compensation for its prepayment option, its liquidity or uncertainty around valuation. That additional yield relative to reference rates is the spread. A hedge calibrated to general interest rates does not automatically offset a change in this relative price. ^{[5] [8]}

AGNC publishes two separate tests for its June 30, 2026 portfolio, including hedges. A parallel 25-basis-point increase in interest rates, with mortgage spreads held constant, would reduce tangible net book value per common share by an estimated 2.3%. A 25-basis-point widening in MBS spreads, with interest rates held constant, would reduce it by 12.2%. These are static company model estimates, without assigned probabilities, rather than realised losses. ^[8]

Assumptions and method

AGNC’s static simulations for its June 30, 2026 portfolio, including hedges. Metric: estimated change in tangible net book value per common share. A +25-basis-point parallel rate shift with spreads unchanged: −2.3%. A +25-basis-point widening in MBS spreads with rates unchanged: −12.2%. Separate tests, without addition, realised loss or assigned probability.

We do not add the two tests to create a combined-shock result. That would require a fresh, consistent valuation of assets and hedges. Their contrast instead points to the useful question: which price movement has been hedged, and which remains with the investor?

Another line in AGNC’s release illustrates the importance of definitions. Its projected CPR for the remaining life of the portfolio was 8.6%, while actual CPR during the quarter was 13.0%, expressed as an annualised rate. One is an assumption about future years; the other summarises recent experience. Treating them as identical observations would manufacture an acceleration or slowdown. ^[9]

The limits of the October diagnosis

There is recent market testimony. On May 21, 2026, Reuters reported that market participants attributed part of Treasury selling to mortgage hedging. That supports the plausibility of the channel, without allowing an effect to be quantified for October. ^[14]

Measuring that contribution would require positions by coupon and holder, changes in duration, executed hedges and market depth when orders arrived. The public documents examined here do not provide that reconstruction. The 7.40% mortgage survey rate, the Fed’s weekly balance sheet and June company disclosures illuminate different layers. Putting them next to each other does not establish a daily causal chain.

Repayment timing also needs to be distinguished from borrower credit quality. In its May 2026 Financial Stability Report, reflecting data through April 23, the Fed still described overall mortgage delinquency rates as low by historical standards and home-equity cushions as substantial, while noting particular weaknesses in FHA loans. That spring assessment is not a substitute for October data. ^[13]

The result is a more precise allocation of risk than a universal mortgage-crisis narrative allows. The homeowner keeps the favourable payment. The holder accepts that principal may return later. Some investors carry that wait on a stable balance sheet; others must refinance at short intervals and adjust their hedges. New buyers may acquire the exposure at a price that compensates them better.

The crucial interaction is between a repayment schedule that can change and a funding obligation that falls due on a fixed date. To locate the pressure point, ask who needs cash next, which assets can be mobilised and how much of the changing schedule was already reflected in the purchase price. The rate printed on the homeowner’s contract answers none of those questions by itself.

The guide Reading a consumer-credit securitisation explains cash flows and tranches. From the credit card to the annuity follows the owners of risk in a different US consumer-credit channel.

Sources and method

Research closed on 11 October 2026. Published figures retain their original periods and scopes. The hypothetical models use current dollars, with formulas disclosed beneath each figure. They measure neither realised losses nor hedges executed in October. AGNC sensitivities are the company’s simulations for its June portfolio. The Reuters report was read in Investing.com’s attributed Reuters republication; direct access to Reuters remained unavailable.