12.4 million government-guaranteed mortgages, and the $238bn of distress the guarantee absorbs
Money market funds lend 678 billion dollars against agency mortgage bonds, which is the largest single block of collateral in the overnight funding market. It is genuinely safe collateral, and the reason it is safe is worth following, because the risk did not disappear.
These bonds carry a US government guarantee. If the borrower stops paying, the lender is still repaid, and the loss lands on the federal housing agencies instead. So the question is not whether the collateral is sound. It is how much borrower distress the guarantee is absorbing on its behalf.
That can be counted exactly, because every loan behind those bonds is disclosed individually. Across 12,423,780 mortgages carrying $2.50 trillion, 1,117,508 are behind on payments: 9.0% of loans and 9.5% of the balance, which is $238 billion. Of that, 304,042 loans and $72 billion are six months behind or worse.
This is the government-guaranteed end of the mortgage market: first-time buyers, small deposits, thin credit files. The gradient is what a mortgage book looks like when it is doing the job it was created for. Borrowers below a 580 credit score fall behind at 22.2%, roughly 14 times the rate of borrowers above 740. None of that reaches the investor holding the bond.
So the answer to where this risk went is: onto the public balance sheet, and the amount is $238 billion. That is the honest way to read the safety of the collateral in the repo market. The repo lender is protected, the money market fund is protected, and the exposure was moved rather than removed.
Every loan in Ginnie Mae pools, 2026-07, streamed in full rather than sampled. The published record layout is behind a script-driven control and could not be retrieved, so fields were identified structurally and each checked against something external: the balance column sums to $2.5 trillion against Ginnie's known outstanding, the delinquency column reproduces the distribution Ginnie publishes, and the credit score column produces a monotonic gradient no other field does. The six-month bucket is capped at "six or more", so it holds more loans than the three, four and five buckets combined.
A snapshot says who defaults. It cannot say whether the position is deteriorating, which is the more useful question. Here are the same pools every month for 39 months.
Both are rising, and the serious end is rising much faster. Loans with any payment missed went from 6.6% to 9.4%, up 2.8 points, wandering on the way because missed payments are seasonal and most borrowers catch up. Loans six months or more behind went from 1.05% at their low in 2024-05 to 2.46%, a rise of 134% over 25 months. In money that is $26bn to $72bn, close to a trebling.
The climb is not uninterrupted: the serious measure fell in 5 of those months, and paused through the middle of 2025 before steepening again. A shorter window would have hidden that pause and made the rise look cleaner than it is.
The distinction matters because the two measure different things. A missed payment is often a bad month. Six months behind is a borrower who is not coming back, and it is the stage at which the guarantee actually pays out. The number that looks calm is the one that catches up later; the number that is moving is the one that costs money.
Every loan in Ginnie Mae pools at each report month, 202304 to 202606, read from the quarterly performance files in full rather than sampled. These files are ordered by pool, so reading the first portion of one is not a random draw: doing that gave delinquency a full point and a half too high. The six-months bucket is capped at "six or more".
Between falling behind and losing the house sits forbearance: a formal agreement that pauses payments. A loan in forbearance is delinquent and cannot be foreclosed on, so this is the valve between the two numbers above. It is reported loan by loan every month since 2020-05.
The pandemic programme has almost entirely unwound, from 1,334,969 loans to 113,305, down 92%. The interesting part is what happened after it emptied out.
Forbearance stopped falling and started climbing again. On the old reporting basis it bottomed at 27,318 in 2024-04 and reached 80,411 by 2024-11, a rise of 194%. On the new basis it bottomed at 92,384 in 2025-08 and reached 146,453 by 2025-12, a rise of 59%. Two separate measurements, both pointing the same way.
Of the 113,305 loans in forbearance today, 26,081, or 23%, are there under the category reserved for borrowers who have lost their job rather than under an ordinary payment plan.
The step in 2025-05 is not a surge. The disclosure changed the source it draws forbearance from in that month, and the level jumps about 45% across the join. That is why the chart breaks the line there and why the two climbs above are measured separately, inside each basis, instead of as one number running through it. Quoted end to end the rise would read as more than 400%, and most of that would be the change in what is counted.
Everything above measures a state, and a state can cure. This measures an outcome. Every month, each loan that leaves a pool is recorded with a reason it left, which separates a borrower who refinanced from a borrower who was foreclosed on. The record runs 203 months, back to 2009-09, so it is much the longest series here.
The collapse in the middle is not credit, it is policy. The foreclosure moratorium covered exactly these loans, and what it closed is the exit this chart measures. The low of 354 came in 2021-10, a few months after the moratorium lifted, because a foreclosure already in train still takes months to finish. Set against an average of 1,238 a month in the two years before, that dip records what was permitted rather than how borrowers were faring.
Since then the rate has tripled. Over 2023-04 to 2026-06 the number of loans in these pools grew 12.4%, while completed foreclosures grew 241%. Per hundred thousand loans that is 7.6 a month rising to 22.9. Both ends of that comparison have a counted book behind them, which is what makes it a rate rather than a tally.
Why there is no comparison with 2008 here. The latest month is the highest count in the whole record, and that fact on its own is close to meaningless: this book has roughly tripled in size since 2009, so an equal count today is a much lower rate. The count is shown because its shape is the point. The only rate stated is the one where the size of the book is known at both ends. A chart of distressed exits as a share of all exits would be worse still, because payoffs are the denominator and payoffs follow mortgage rates, so that line climbs whenever refinancing stops even if not one extra borrower is in trouble.
Everything above is American, because that is where loan-level mortgage data is published. Europe does not disclose its mortgage books that way, but it does ask households directly whether they are behind on their housing payments, every year since 2003. Here is 2025.
The spread is the first thing to notice. Greece at 10.6% sits more than thirty times above Croatia at 0.3%, and only 8 of the 27 are above the euro-area figure of 3.4%, shown in amber. There is no European mortgage market to speak of, only twenty-seven of them, and an average conceals almost everything worth knowing.
The direction also runs the other way to the American series above. Euro-area households behind on housing payments have fallen from 5.0% in 2014 to 3.4% in 2025. Over broadly the period in which US government-backed delinquency has been climbing and completed foreclosures have tripled, European household payment stress has been easing.
There is a second European measure, and it is the administrative one the survey is not. Euro-area banks classify every loan by whether its credit risk has risen significantly since it was written. That bucket, stage two, is the bank's own warning about loans that are still being paid, and it is reported under a common supervisory definition every quarter since 2020-Q3. Here it is for household lending in 2026-Q1.
It produces an almost completely different ranking. Germany tops it at 16.5% of household loans flagged, while its own households reported arrears of just 2.2%, near the bottom of the previous chart. Across the 14 countries in both, the rank correlation between what banks flag and what households report is 0.10. They are close to unrelated.
That gap is worth more than either number alone. Stage two is a judgement, and how conservatively it is made differs by country, by supervisor and by accounting practice, so a high reading can mean cautious banks rather than struggling borrowers. Arrears are what households say about themselves. Neither is wrong, and anyone who picks one and calls it European credit stress can have whichever answer they came for.
One split does survive across measures. Loans already gone bad, rather than merely flagged, sit disproportionately at the smaller banks: nationally supervised institutions average 3.25% non-performing against 1.59% at the large ones the ECB supervises directly, roughly double.
These are not the same measurement and they never share an axis. The American figures are administrative records: every loan in a pool, with a servicer reporting whether a payment was missed. This is a household survey, self-reported, and it counts rent as well as mortgages, so a household that has never had a mortgage can appear in it. The levels cannot be compared in either direction, and putting them on one chart would manufacture a comparison the data does not support. What can honestly be set side by side is which way each one is moving.
Delinquency is not loss. Many of these borrowers cure, and the guarantee means the bondholder is repaid either way. What the number measures is how much distress the public balance sheet is carrying, not how much money anyone has lost.