CPR and Prepayment Risk in Agency MBS

The Borrower Decides When You Get Your Money Back

A US mortgage borrower can repay the whole loan at any time, without penalty, for any reason. Buy a pool of those loans and you have bought a bond whose repayment date belongs to somebody else, and that somebody will hand the money back exactly when you least want it. That is the risk this page is about, and CPR is how the market measures it.

CPR (conditional prepayment rate) is the share of a pool’s remaining balance that pays off over a year: refinancings, house sales, and the occasional borrower simply clearing the debt. Agency mortgage REITs like Annaly and AGNC own tens of billions of dollars of these pools, funded with short-term borrowing. CPR decides how long today’s spread is actually earned before the principal comes back and has to be reinvested at whatever the market is paying then.

The CPR Formula

Freddie Mac defines CPR from the single monthly mortality rate (SMM):

CPR = [1 − (1 − SMM)^12] × 100

SMM is the percentage of the remaining pool balance that prepays in one month. Compounding twelve months gives the annualised CPR. FHFA uses the same compound annual prepayment rate concept for uniform MBS disclosure.

Term	Meaning
SMM	Single monthly mortality: monthly prepay speed
CPR	Annualised prepayment rate derived from SMM
Experienced CPR	What the portfolio actually prepaid in the period
Projected CPR	Management’s forward long-term or life-of-pool assumption

Q4 FY2025 Portfolio Speeds

Annaly and AGNC disclosed quarter-end CPR figures in FY2025 supplements. No full-year average CPR was filed for either name.

Company	Experienced / actual CPR	Projected CPR	As-of
NLY	9.7% (quarter experienced)	10.8% (long-term)	Q4 2025
AGNC	9.7% (Q4 actual portfolio)	9.6% (portfolio-life)	31 Dec 2025

Both portfolios prepaid at roughly 9.7% in Q4. Annaly’s long-term projection (10.8%) sits slightly above AGNC’s portfolio-life CPR (9.6%). Small differences in pool seasoning, coupon distribution, and model assumptions explain the gap; the headline is that both books are prepaying at high-single-digit speeds, not at the 20%+ CPRs seen in refi waves.

Negative Convexity: The Holder Loses on Both Sides

When rates fall, an ordinary bond rises in price, because its fixed coupon is now above the market. A mortgage pool cannot rise nearly as much, because borrowers refinance into the cheaper rate and hand the principal back at par. The bond is effectively called away at 100 cents precisely at the moment it would otherwise have appreciated, and the money returns to be reinvested at the new, lower yield.

When rates rise, the pool falls in price like any other bond. But now prepayments dry up: nobody refinances a 5% mortgage when the market is at 8%, and nobody with a cheap loan wants to move house. The principal you would like back in order to buy the higher coupons stays locked in the below-market bond for longer. That is extension, and it arrives exactly when the holder would rather have their money.

Upside capped, downside not. The technical name for that asymmetry is negative convexity, and it is the single feature that separates an agency MBS from a Treasury of the same duration. The extra yield an agency mREIT earns over Treasuries is, in large part, payment for wearing it. Swaps and TBAs can hedge the pool’s duration; they do not hedge the convexity, because the option to prepay belongs to the borrower and there is no cheap way to buy it back.

Agency books are long-duration assets funded with short-term repo. A CPR shock that shortens the assets without a matching move on the funding side compresses net interest spread, and the mark runs through book value.

Whether Speed Helps or Hurts Depends on What You Paid

Faster is not automatically worse. Prepaid principal comes back at par, 100 cents on the dollar, whatever the pool cost. So the direction of the damage is set by the purchase price.

Bought at	What a prepaid dollar returns	Effect of fast CPR
A premium, say 103	100 against a 103 carrying value	Hurts. You lose the 3, and the unamortised premium is written off faster, which drags reported yield down
Par, 100	100 against 100	Neutral on price; you still face reinvestment risk
A discount, say 95	100 against a 95 carrying value	Helps. The pull to par you expected over years arrives now

High-coupon pools bought after rates rose trade above par, and that is where prepayment risk concentrates. Annaly reports its net interest spread both with and without a premium amortisation catch-up for this reason: a change in the projected long-term CPR forces a retrospective true-up of how much premium should already have been written off, and it can move a quarter’s reported spread without anything happening to the underlying loans.

So before reading a rising CPR as bad news, check the weighted average price of the book. On a discount portfolio, the same print is a tailwind.

What Actually Drives CPR

The level of mortgage rates is not the thing. What moves prepayments is the gap between the rate a borrower is already paying and the rate they could get today. A pool of 3% loans is untouchable at 6.5%; a pool of 7.5% loans written last year would empty out fast at the same 6.5%.

Our planning assumptions for the rate side:

Rate mark	Planning level
30-year fixed mortgage	6.50%
10-year US Treasury	4.50%
Agency MBS current-coupon spread over Treasuries	100 bps

At around 6.5%, a Q4 2025 CPR near 9.7% is what you would expect: most outstanding US mortgages carry lower rates than that, so refinancing costs the borrower money. A material fall in mortgage rates would push CPR well above that and test hedge books. A sustained rise slows CPR and extends duration; MSR-heavy hybrids like Rithm gain when that happens, while a pure agency book gets fewer prepayments but also fewer chances to recycle principal into higher coupons.

Rates set the incentive. Four other things decide who actually acts on it.

Seasoning. A brand-new loan barely prepays. Nobody refinances a mortgage they closed three months ago, and few people move house immediately. Speeds ramp over roughly the first two or three years, then settle. Two pools with the same coupon can prepay at very different rates simply because one is older.

Burnout. A pool that has already lived through a refinancing window has lost the borrowers who were willing and able to act. What remains is the population that did not move when it clearly paid to, so the same incentive applied a second time produces a much weaker response. Burnout is why a prepayment model fitted to the last cycle overshoots in the next one.

Loan size. Refinancing costs a few thousand dollars in fees whatever the balance. On a $600,000 loan a one-point rate saving repays that in months; on a $90,000 loan it takes years. Small-balance pools therefore prepay slowly and predictably, which is why agency mREITs pay more than the generic pool price for “specified pools” screened on loan size, geography or credit profile. That pay-up is what buying back some convexity costs.

Friction. Some borrowers never refinance, whatever the arithmetic says. Damaged credit, too little equity, self-employment paperwork, a house worth less than the loan, or plain inertia. There is also a floor working the other way: people sell, divorce, die and relocate regardless of rates, which keeps a few points of CPR running even when no refinancing makes sense.

Worked Mini-Example: SMM to CPR at 9.7%

Reverse-engineer the monthly speed implied by a 9.7% CPR to show the compounding:

1. Target CPR = 9.7% = 0.097
2. Solve: 0.097 = 1 − (1 − SMM)^12
3. (1 − SMM)^12 = 0.903
4. 1 − SMM = 0.903^(1/12) ≈ 0.9915
5. SMM ≈ 0.85% per month

Note that twelve months at 0.85% does not give 10.2%. Each month’s prepayment shrinks the balance the next month’s rate applies to, so the twelve monthly speeds compound down to 9.7%, not up. Multiplying SMM by twelve overstates CPR, and the error grows as speeds rise.

Under one per cent of the pool leaving each month does not sound like much. On a $100B agency book it is roughly $10B of principal a year coming back and needing somewhere to go. At NLY’s 5.6× economic leverage, the equity underneath that reinvestment decision is a fraction of the headline asset size; see leverage definitions.

Now set NLY’s experienced 9.7% against its projected 10.8% long-term CPR. The model expects the pool to run about a tenth faster than the quarter actually delivered. A persistently faster speed shortens the expected life of the assets, pulls premium write-off forward, and moves the hedge ratio, none of which needs rates to do anything. Compare experienced with projected every quarter: a gap that keeps widening means the model is behind the pool.

Linking CPR to the Agency Screen

CPR sits between spread and book value in the agency workflow:

1. Spread (NLY ~1.45% FY avg, AGNC 1.92% FY2025) funds the dividend.
2. CPR determines how long spread is earned on the current pool.
3. Leverage (NLY 5.6× economic, AGNC 7.2× at-risk) scales BVPS sensitivity to reinvestment and mark moves.
4. BVPS (NLY +5.5%, AGNC +5.6% FY2025) is where the quarter shows up in price-to-book.

Two quarters can show the same spread and a different book outcome, and dividend yield will not explain the gap. CPR is usually where the difference is hiding: one quarter’s principal came back and was reinvested at a worse yield, and the other’s did not.