Each measure against the sector total it belongs to
Leverage is usually quoted as an absolute number, which makes almost anything look small next to a $58tn equity market. That comparison is wrong in kind: market capitalisation is a price, and these are claims. Measured against the sector totals they actually belong to, the picture is different.
| Measure | Size | Share | Of | Note |
|---|---|---|---|---|
| US listed long-term debt | $4.20tn | 18.6% | nonfinancial corporate liabilities | the measure everything else is usually compared to |
| Hyperscaler off-balance-sheet commitments | $2.63tn | 11.6% | nonfinancial corporate liabilities | five companies: purchases plus leases signed but not started |
| Bank lending to non-bank financial institutions | $2.00tn | 10.1% | all bank credit | +243% since 2020 |
| Margin debt | $1.50tn | 7.1% | household debt | highest share since the series began in 1997 |
| US listed lease obligations | $1.33tn | 5.9% | nonfinancial corporate liabilities | leases are leverage the debt line does not show |
| Sector total | Value | As of |
|---|---|---|
| Households and nonprofits, total assets | $204.54tn | 2026-01-01 |
| Households and nonprofits, net worth | $182.98tn | 2026-01-01 |
| All sectors, debt securities and loans | $115.56tn | 2026-01-01 |
| All sectors, total debt securities | $66.16tn | 2026-01-01 |
| Federal debt outstanding | $39.07tn | 2026-01-01 |
| Federal government, marketable Treasury securities | $30.64tn | 2026-01-01 |
| Nonfinancial corporate business, total liabilities | $22.63tn | 2026-01-01 |
| All domestic sectors, total mortgages | $21.90tn | 2026-01-01 |
| Households and nonprofits, total debt | $21.07tn | 2026-01-01 |
| All commercial banks, bank credit | $19.75tn | 2026-07-29 |
| All sectors, corporate and foreign bonds | $17.07tn | 2026-01-01 |
| Nonfinancial business, credit market instruments | $14.45tn | 2026-01-01 |
| Households, home mortgages | $13.85tn | 2026-01-01 |
| Agency and GSE-backed pools, debt securities and loans | $12.54tn | 2026-01-01 |
| Nonfinancial corporate business, debt securities and loans | $8.98tn | 2026-01-01 |
| Nonfinancial corporate business, corporate bonds | $8.04tn | 2026-01-01 |
| Consumer credit outstanding | $5.17tn | 2026-06-01 |
| Federal government, nonmarketable Treasury securities | $3.82tn | 2026-01-01 |
| State and local governments, municipal securities | $3.56tn | 2026-01-01 |
| Bank loans to nondepository financial institutions | $2.00tn | 2026-06-01 |
| Issuers of asset-backed securities, total liabilities | $1.88tn | 2026-01-01 |
| Motor vehicle loans owned and securitised | $1.57tn | 2024-10-01 |
| Security brokers and dealers, customer receivables | $622bn | 2026-01-01 |
Every measure on this page is a level, and a level cannot say when anything happens. One widely used indicator tries to, and it is not flashing in a single country. That is the strongest evidence against reading this page as a warning, and it is published here for exactly that reason. It also has a blind spot, and the blind spot happens to be the two places where this cycle's borrowing actually grew, so both halves belong in the same breath rather than one after the other.
The credit-to-GDP gap is private borrowing as a share of the economy minus its own long-run trend. The Bank for International Settlements built it after finding that a gap above about ten percentage points preceded most post-war banking crises by two to three years, and it is written into the Basel rules as the reference for how much extra capital banks should hold in a boom. It compares each country to its own history, never to another, which is why a country can carry a great deal of debt and still read as calm.
No country in the sample is above the warning level. The highest reading is Saudi Arabia at 7.0 points, and 38 of 44 countries are below their own trend, many of them a long way below. The United States sits at -11.5. On the one measure built specifically to see a credit crisis coming, the private sector of the developed world is carrying less debt relative to the economy than its own history would predict.
That deserves to be taken seriously rather than explained away. It is the strongest available evidence against reading this page as a warning, and it is why nothing here is written as a forecast.
It also has a specific blind spot, and it happens to be the two places this page found the growth. The gap measures private non-financial credit: it excludes governments, whose borrowing multiplied nearly sevenfold, and it excludes the financial sector, where lending to non-banks grew 6.2 times in eleven years. An indicator built to catch households and companies over-borrowing, as they did before 2008, is looking in neither of the places where this cycle's leverage actually accumulated.
Quarterly, 44 countries, latest 2025-Q4. The trend is a one-sided filter re-estimated at each point, so recent readings get revised and the direction matters more than the last value. Ireland and Luxembourg read as extreme outliers because their GDP is inflated by multinational accounting, not because their borrowers deleveraged.
The indicator above is a model. This one is a price. The gap between what a weaker borrower pays and what the safest pays is the market's own opinion of credit risk, quoted daily by people with money at stake, and it goes back 1960.
It is 0.48 percentage points, against a median of 0.89 over 798 months and 3.38 at the peak in 2008-12. That places it in the 4th percentile of its own history: lenders have charged more than this for weaker credit in roughly 96% of the months since 1960.
This is the third measure on this page that is not flashing, after the credit-to-GDP gap and the collateral money market funds accept. It is also the most direct: the others are constructed, and this is what someone actually paid. Any argument that credit is dangerously priced has to explain why the people pricing it disagree.
Before treating that as reassurance, it is worth asking what this same indicator said last time. The series is long enough to check, and the answer is that it said almost nothing. At the dot-com peak it stood at 0.69 points, the 24th percentile of its history to that date, and over the preceding two years it had moved +0.09 points. It reached 1.41 eventually, 31 months after the top. At the 2007 equity peak it stood at 0.82 points, the 44th percentile of its history to that date, and over the preceding two years it had moved -0.13 points. It reached 3.38 eventually, 14 months after the top.
So the spread is not an early warning. It is a coincident-to-lagging measure that confirms trouble once trouble has arrived, and in the two years running into 2007 it was falling. A calm reading today is therefore evidence about the present price of risk and close to worthless as a forecast, which is the opposite of how a fourth-percentile number usually gets read. The same applies to the margin-debt chart below: it peaked in the same month as the market in 2000 and three months ahead of it in 2007, which is not warning, it is company.
If the priced measures were quiet into both tops, the fair question is whether anything moved earlier. Two things did, and neither is a price. Neither is much of a warning either, which is the part worth reading to the end of.
The lead is much shorter than it first looks, and the earlier version of this section overstated it. Measuring from a trough to the market top gives thirty months or more, but the trough is a date chosen after the event. Measured from the crossing, the point at which more banks were tightening than easing, the warning arrived 17 months before the 2000 top and 3 months before 2007. Both series then reached their highest readings after the peak rather than before it. They are quantities rather than prices, and a lender tightening is acting on its own book before any of it reaches a spread, but "before" here is months, not years.
And then it cried wolf. Over the two years to 2023-07, lending standards swung 83 points, from -32.4 to +50.8. That is a larger move than before either of the episodes above, and the largest reading in the series outside 2008 itself. No credit crisis followed. An indicator is not an early warning if its loudest signal in forty years was wrong, and this section would be dishonest without saying so.
So the honest position is that nothing here has a clean record. The priced measures never led. The survey led twice and then misfired once, harder than it had ever led. Today it is back at dead neutral and bank mortgage delinquency is off its low but nowhere near its 2007 climb, while the measures moving hardest in this cluster sit at the government-guaranteed margin rather than in the bank aggregate. That is what the start of a cycle looks like. It is equally what a problem confined to one segment looks like, and nothing here separates them.
Lending standards are the net share of banks reporting tighter business credit, so zero means as many easing as tightening rather than no change in the level of standards. The two episodes are dated from the equity peak, which is a choice: dating from the credit event instead would lengthen every lead.
Monthly, 1960-01 to 2026-06. A percentile compares this measure only to itself, which is the point: it says whether today is unusual for this series, not whether the level is objectively safe. A spread can be historically tight because risk genuinely is low, or because it is mispriced, and no percentile can separate those two.
The usual account of this cycle is that credit left the banks and moved somewhere less visible. It is worth testing directly. Below is each part of the US credit system, set to 100 at the end of 2007, the quarter before the last crisis. Anything above 100 has grown in nominal terms since then.
Nothing moved between banks and the bond market. Bank credit grew 2.36 times and the whole stock of debt securities grew 2.37 times, which is as close to identical as this data gets. The movement is at the two ends. Government borrowing multiplied nearly sevenfold. And the private securitisation machine whose failure defined 2008 is smaller today than it was then, at 0.41 times its 2007 size, before adjusting for eighteen years of inflation. Whatever the next accident is, it is not a rerun of the last one, because that market never came back.
One line is missing from the chart because it cannot honestly be drawn on the same axis: bank lending to non-bank financial firms is only reported from 2015. Measured over its own life it grew 6.2 times in eleven years, 17.3% a year, faster than anything above. That is the section below.
These are segment totals and not a partition of the whole, so they overlap: agency pools hold mortgages that also count in total mortgages, and corporate and foreign bonds include paper issued by financial firms. They are not summed here for that reason. Nominal levels throughout, not adjusted for inflation or for the size of the economy.
Margin debt is money investors borrow against the shares they already own. It does not start a fall, but it decides how far one travels, because a margin call forces selling regardless of what the seller thinks. As a share of the economy it is now above every previous peak in the record, which begins in 1997.
That record depends on the denominator, and the honest version is worth stating before someone else does. GDP is a flow, while margin debt is a stock borrowed against market value, and US equity has roughly doubled relative to output since 2000. So part of the record is mechanical: more listed equity per unit of economy rather than more borrowing per unit of equity. Measured against the value it is actually lent against, margin debt is 2.04% today, against 1.84% at the dot-com peak and 2.23% in 2007. That is the 79th percentile of 355 months and below the 2007 reading. Elevated on either measure. A record on only one of them.
Earlier episodes are not directly comparable. Broker loans were around 8% of output in 1929, on a different definition and when buying on 10% margin was legal. The market-value denominator has its own distortion in the other direction: its highest readings come when equity prices collapse faster than the borrowing against them unwinds, which is why the peak on that measure sits in late 2008 rather than at any market top.
The common claim about this cycle is that credit has moved off bank balance sheets, so banks carry less of it than they did in 2008. That is at best half true. Banks did not step back from this lending; they moved one rung up the chain and now fund the funds that make the loans.
| Month | Bank loans to non-bank financials |
|---|---|
| 2020-01 | $581bn |
| 2022-01 | $820bn |
| 2024-01 | $1.00tn |
| 2025-01 | $1.41tn |
| 2026-01 | $1.88tn |
| 2026-06 | $2.00tn |
That is a US series, and this site is written for European investors, so the obvious question is whether euro-area banks did the same thing. They did, and at a third of the pace.
Both indexed to 2015-01. US banks multiplied their lending to non-bank financial firms 6.2 times; euro-area banks 1.8 times. The starting point is the part worth pausing on. In 2015 euro-area banks lent EUR0.95tn to these borrowers against $0.32tn in the United States, roughly three times as much. Today the US is the larger of the two at $2.00tn against EUR1.69tn. Europe did not start this; it was already there, and America overtook it in eleven years.
Indexed rather than converted, because one series is in dollars and the other in euros and running eleven years of monthly stocks through a spot rate would put exchange-rate movement into a chart about credit growth. The definitions are close but not identical: the Fed publishes one bucket for nondepository financial institutions, while the ECB splits the same territory finely and the line here is financial corporations other than banks, insurers and pension funds. Adding insurers moves the euro-area total by about EUR150bn and the multiple barely at all.
A bank lending to a private credit fund is still funding the loan. What changes is where the first loss sits, and that depends on advance rates and collateral quality this series does not show.
Every measure above stops at a border. US bank lending, euro-area bank lending, US mortgages, US money funds: each is bounded by a country, and that boundary is where the measurement ends rather than where the money does.
There is credit denominated in dollars, owed by non-bank borrowers who are not in the United States. It is real debt in a currency whose central bank did not create it for that borrower, and it appears in no national credit total. It now stands at $14.74 trillion, against $5.89 trillion before the last crisis.
Indexed to the quarter before the crisis, because the useful question is which parts of the credit system came back afterwards. Dollar credit offshore is 2.5 times its 2008 level, and euro credit outside the euro area 2.2 times. Set that against private-label securitisation at 0.41 times, from the chart further up this page. The market that caused 2008 never recovered. This one more than doubled.
It is also the cleanest example on this page of the pattern running through all of it. The borrowing did not stop, and it did not stay where the last crisis taught everyone to look. It moved somewhere with no single supervisor, no national total, and no early-warning indicator pointed at it.
Each currency is shown in its own units and never converted or summed. Yen credit is quoted in yen. Running twenty-six years of quarterly stocks in three currencies through a spot rate would put exchange-rate movement into a chart about credit growth. Non-bank borrowers only, and credit means loans plus debt securities rather than loans alone.
Everything above measures borrowing from the outside: totals, schedules, prices. This is the lenders' own view of it. Euro-area banks classify every loan by whether its credit risk has risen significantly since it was written, and the resulting bucket is the closest thing there is to a supervised early warning. Here is corporate lending in 2026-Q1.
Germany flags 23.2% of its corporate loan book, against a median across the 15 of 9.9%. Banks are more worried about companies than about households: the corporate share is the higher of the two in 10 of the 14 countries reporting both, and in Ireland the gap is 11 points. The household side of the same return is on the mortgage page.
This is a bank's judgement rather than a payment record, and how conservatively it is made differs by country, supervisor and accounting practice, so a high reading can mean cautious lenders as easily as struggling borrowers. It is comparable between these countries because they report it on one definition; it is not comparable with anything reported on another.
What these numbers are not. Every figure here is a level, and a level describes how far a fall travels, never when one starts. None of this is a forecast. The commitments are multi-year and undiscounted; the bank series is an aggregate that cannot show which banks or against what collateral.