Lehman vs FTX: How Audited Balance Sheets Hid $58 Billion
One was a 158-year-old Wall Street bank. One was a crypto exchange in the Bahamas. Both had clean audit opinions on the books the day they collapsed.
Lehman Brothers vs FTX Collapse: How $58 Billion Vanished From Audited Balance Sheets
Prefer to watch? This article is the written companion to the video above.
Two companies. One is a 158-year-old Wall Street institution, regulated by the SEC and audited by EY, one of the Big Four. The other is a crypto exchange run out of a penthouse in the Bahamas, audited by a firm whose lead partner, the SEC would later state, “fundamentally did not understand” the business he was signing off on.
Lehman Brothers moved up to $50 billion off its balance sheet, every quarter. Under the rules of the time, this was arguably permissible.
FTX took $8.9 billion of its own customers’ money. That was theft.
One legal, one criminal. But here is what they share, and it is the only part that should keep you up at night: both had clean audit opinions on the books the day they collapsed.
The same lie, 14 years apart
Lehman Brothers
158-year-old Wall Street bank · 2008
- The trick
- Booked its loans as sales — “Repo 105”.
- The hole
- ~$50.4bn moved off the books — every quarter.
- The collateral
- Real bonds, but mispriced.
- The gatekeeper
- EY knew about it — and signed.
- The verdict
- Legal at the time. No one charged.
FTX
Crypto exchange in the Bahamas · 2022
- The trick
- Printed its own token, FTT, out of nothing.
- The hole
- ~$8.9bn of customer money — taken.
- The collateral
- A token it printed itself.
- The gatekeeper
- Its auditor “didn’t understand it” — and signed.
- The verdict
- Theft. 25 years for its founder.
Both had a clean audit opinion on the day they collapsed.
Lehman: the quarterly vanishing act
End of fiscal year 2007. Lehman Brothers holds $691 billion in assets against $22.5 billion of equity, a leverage ratio of 30.7 to 1.
For every dollar of its own money, Lehman held thirty-one dollars of assets and borrowed the rest. Work out what that means: a 3.2% fall in asset values wipes out the entire equity. Not most of it. All of it.
The green sliver is everything Lehman actually owned. A 3.2% fall in the value of the rest wipes out the entire cushion — not most of it, all of it. That is what thirty-to-one leverage means, and it is why the balance sheet had to be made to look smaller than it was.
Lehman’s management knew the ratio looked terrifying to anyone who read it, so they made it look better. Not by reducing the risk, but by moving it somewhere the reader was not looking.
How Repo 105 worked
Banks borrow from each other constantly using repurchase agreements, or repos. You hand over bonds as security, receive cash, and buy the bonds back a few days later. Entirely ordinary, and it stays on the balance sheet as a financing, because that is what it is.
Lehman found a seam in the rules. Under the accounting standards of the day, if you handed over collateral worth at least 105% of the cash you received, the transaction could be treated not as a loan but as a sale. And a sale changes everything:
- The bonds disappear from the balance sheet
- The cash is used to pay down debt
- Assets fall, liabilities fall, and the leverage ratio looks healthier
Then, a few days after the quarter closed and the snapshot had been taken, Lehman borrowed fresh cash and bought it all back.
The only day that was published
Balance sheet size across a single quarter. Nothing about the risk changed between these three moments — only which one had a reporting date attached to it.
The balance sheet at its real size
Around $50bn moved off, booked as a sale
Bought back. Everything returns
Analysts, rating agencies and regulators all read the middle bar, because the middle bar was the only one with a date on it. The other two were never published anywhere. Every quarter, like clockwork.
Volume in the second quarter of 2008: $50.38 billion. Hidden from every analyst, every rating agency and every regulator who was reading the numbers Lehman published — which were, by construction, the only numbers there were.
The detail that gives the game away
No American law firm would sign off on it. They all reached the same conclusion: this is not a true sale. So Lehman routed the transactions through its London subsidiary and obtained an opinion from the English firm Linklaters, under English law.
Other major banks examined the same technique and declined to use it. Lehman used it alone.
That is the tell, and it generalises far beyond Lehman. When a structure only works in one jurisdiction, and only one firm in the market will put its name to it, the structure is telling you what it is.
15 September 2008. Chapter 11, with over $639 billion in assets. The largest corporate bankruptcy in American history, a record it still holds.
The principle: off-balance-sheet
Your landlord is coming tomorrow for the annual inspection and your flat is a disaster. So the night before, you rent a storage unit across the street and shove everything into it. Every box, every mess. The landlord walks in, sees a clean flat, signs off. The next day you carry it all back.
The mess was never gone. You moved it somewhere the landlord was not looking.
That is off-balance-sheet accounting: assets, liabilities and risks a company is genuinely exposed to, which do not appear on the balance sheet you are reading. Not all of it is fraud. Banks use these structures legitimately every day. But it is a spectrum, running from transparent to aggressive to Lehman, and beyond Lehman to what happened in the Bahamas.
There is a second concept you need, because it sits underneath both cases. When an asset has no observable market price, you value it with a model. Mark-to-model. Legitimate, and often unavoidable. But when the model’s assumptions quietly detach from reality, the output stops being an estimate and becomes fiction with a decimal point. Wall Street has a name for that stage:
Mark-to-market → mark-to-model → mark-to-myth.
Lehman’s mortgage products began as model estimates, and when housing collapsed the estimates became fantasy. FTX’s token was fantasy on day one.
FTX: the digital mirror
Fourteen years later, from Wall Street to a penthouse in the Bahamas.
FTX, one of the largest crypto exchanges in the world, created its own token, called FTT — out of nothing, which is what creating a token means. Its sister company Alameda Research, controlled by the same person, held billions of dollars of FTT on its books. Reportedly more of the token than the entire market said the token was worth.
Read that twice. They held more of a thing than the thing existed.
And then they pledged it as collateral, to borrow real money. Dollars. Bitcoin. Money belonging to FTX’s customers. A software setting inside the exchange allowed Alameda to withdraw beyond every limit that applied to everyone else, without tripping a single alarm.
Roughly $8.9 billion of customer funds, gone.
The structural parallel to Lehman is exact in one respect: both were relying on collateral that evaporates precisely when you need it. Lehman’s collateral was real but mispriced. FTX’s collateral was a token it printed itself, whose value depended entirely on the solvency of the entity pledging it.
Nine days from the leak of Alameda’s balance sheet to collapse. Larry Summers drew the sharper distinction at the time: FTX was not Lehman, it was Enron. Not a systemic accident but a crime. Sam Bankman-Fried was sentenced to 25 years in federal prison in March 2024.
Why the lie survives: the gatekeepers
The companies lied. That part is easy. But a lie only survives if the people paid to catch it look away.
EY and Lehman: they knew
The Examiner’s Report, the official court-ordered post-mortem, is not ambiguous. EY knew about Repo 105. They reviewed the policy. They never questioned the volume, and never questioned why it spiked with such perfect regularity at every quarter-end. A senior Lehman employee, Matthew Lee, went to EY as a whistleblower and warned them directly.
Big Four or no-name, it made no difference. EY saw the problem and looked away, because the client was paying a fortune. Prager Metis did not understand what it was looking at. Different reasons, identical result: the gatekeepers, whose entire professional purpose is to protect the person reading the numbers, signed off on both.
And this is not a story about two bad audits
Here is the part that changes how you should read any set of accounts. The body that inspects American audit firms publishes how often it finds that an auditor had not obtained sufficient evidence to support the opinion it had already signed.
How often the evidence was not there
Share of inspected audits where the regulator found the auditor had not obtained sufficient evidence to support the opinion it had already signed. United States, 2024, with 2023 for comparison.
audit roughly 80% of US listed market cap
six US firms
171 firms, over 800 audits reviewed
the category FTX’s auditor belonged to
Every one of these audits ended in an opinion. The deficiency is not that the numbers were wrong — it is that the auditor could not show it had done enough work to know either way. The rates fell in 2024, and the regulator still describes them, in its own words, as still high.
Look at the bottom row. Non-affiliated firms — small practices with no global network behind them, the category FTX’s auditor belonged to — sat at 52% in 2024, essentially unchanged from 53% the year before. Slightly worse than a coin flip.
The Big Four are far better at 20%, down from 26%, and they audit roughly 80% of the market capitalisation of US listed companies. But twenty percent is not nothing. It means that in one audit out of five at the best firms in the business, the regulator could not find enough evidence behind the signature.
None of that makes those companies frauds, and most of them were perfectly fine. What it does mean is that a clean opinion is weaker evidence than almost everyone assumes it to be. It tells you an auditor signed. It does not, on its own, tell you an auditor looked.
Two things people believe that are wrong
“Lehman was a one-off. Dodd-Frank fixed it.” Off-balance-sheet risk did not disappear. It migrated, into private credit and shadow banking, where supervision is thinner and disclosure is optional.
“FTX was a crypto problem. Doesn’t touch my portfolio.” The mechanics — self-referential collateral, opaque balance sheets, captured or incompetent auditors — are not a crypto problem. They are a financial-system problem, and they can appear in any asset class that permits them. The same “someone else is holding it, trust us” risk hides inside products that feel perfectly safe: a synthetic ETF carries a counterparty you never see.
What you can actually do
CHECK — read the footnotes, not the brochure
If you own individual stocks, learn to open the 10-K on the SEC’s EDGAR database. Not the glossy annual report; the filing.
Go to the footnotes and search for “Variable Interest Entities.” These are one of the ways risk is parked outside the main balance sheet, and heavy reliance on them without full consolidation is a question you should be able to answer.
Then compute total assets ÷ total equity and compare it with the company’s peers. A dramatically higher ratio is not proof of wrongdoing. It is a reason to keep reading.
AVOID — two structures
Any platform that creates its own token and uses that token as collateral for its own borrowing. That is the FTT model, a death spiral with a countdown timer, and the timer is invisible from outside.
Funds where most assets have no observable market price. Mark-to-model with no liquid secondary market. The returns look wonderful precisely because nobody is testing them.
BUILD — diversify across custodians, not just assets
If you use a crypto exchange, demand proof of reserves and proof of liabilities. Reserves alone are meaningless: a claim about what you own says nothing about what you owe.
And check who audits the platform. Given the numbers above, a multi-billion-dollar business with a no-name auditor should be treated as unverified rather than as verified.
Most importantly:
Thirty assets at one broker that commingles customer funds is not diversification. It is concentration wearing a disguise.
Fifteen assets across three custodians is a genuinely different risk. It is the same custody blind spot behind a quieter fact almost nobody checks: in most cases you don’t legally own the shares you think you do.
And if you think this is history
Non-bank financial intermediation — shadow banking — reached $256.8 trillion in 2024, according to the Financial Stability Board. That is 51% of all global financial assets, growing at roughly double the pace of the banking sector. Private credit is a large and fast-growing piece of it.
And the FSB’s own report notes severe limitations in the data available to measure it.
That last sentence is not a footnote. It is the entire lesson. The balance sheet lie did not disappear; it moved somewhere nobody is required to look.
Primary sources
- 01Report of Anton R. Valukas, Examiner — In re Lehman Brothers Holdings Inc. (Repo 105, Vol. 3) — United States Bankruptcy Court, Southern District of New York
- 02Spotlight: Staff Update on 2024 Inspection Activities (March 2025) — aggregate Part I.A deficiency rate 39% in 2024 against 46% in 2023; Big Four 20% against 26%; US non-affiliated firms 52% against 53%; 171 firms inspected and portions of over 800 public company audits reviewed — Public Company Accounting Oversight Board (PCAOB)
- 03Audit Firm Prager Metis Settles SEC Charges for Negligence in FTX Audits (Press Release 2024-133) — U.S. Securities and Exchange Commission
- 04FSB reports continued growth in nonbank financial intermediation in 2024 to $256.8 trillion — 51% of total global financial assets — Financial Stability Board
- 05Global Monitoring Report on Non-Bank Financial Intermediation 2025 (full report) — Financial Stability Board
Questions people actually ask
What was Repo 105?
An accounting treatment Lehman Brothers used to move assets off its balance sheet at quarter-end. In an ordinary repurchase agreement, a bank hands over bonds, receives cash, and buys the bonds back days later; it stays on the books as a financing. Lehman exploited a rule under which, if you handed over collateral worth at least 105% of the cash received, the transaction could be booked as a true sale. The bonds then vanished from the balance sheet, the cash was used to pay down debt, and the leverage ratio looked healthier — until a few days after the quarter closed, when Lehman borrowed and bought everything back.
How much did Lehman hide, and was it legal?
The Examiner's Report records Repo 105 volume reaching $50.38 billion in the second quarter of 2008. And here is the uncomfortable part: under the accounting rules as they stood, it was arguably permissible. The offence was not obviously a crime — it was that the balance sheet the market was reading was not the balance sheet Lehman actually had. No individual was criminally charged over it.
Why did Lehman route Repo 105 through London?
Because no US law firm would provide the true-sale opinion the treatment required. Lehman therefore executed the transactions through its London subsidiary and obtained an opinion from the English firm Linklaters under English law. Other major banks looked at the same technique and declined to use it. That is the tell: when a structure only works in one jurisdiction, and only one firm in the market will touch it, the structure is telling you something about itself.
What is off-balance-sheet accounting?
Assets, liabilities or risks a company is genuinely exposed to that do not appear on its main balance sheet. Think of renting a storage unit the night before the landlord's inspection: the mess is not gone, it is simply somewhere the inspector is not looking. Not all of it is fraud — banks use off-balance-sheet structures legitimately every day. But it is a spectrum, and Lehman sat a long way along it.
What is mark-to-myth?
When an asset has no observable market price, a firm values it using a model — mark-to-model. That is legitimate and often unavoidable. It becomes mark-to-myth when the model's assumptions have quietly detached from reality, and the resulting number is fiction dressed as accounting. Lehman's mortgage products started as model estimates; when the housing market collapsed the estimates became fantasy. FTX's token was fantasy from the first day.
What did FTX actually do?
FTX created its own token, FTT, out of nothing. Its sister firm Alameda Research — controlled by the same person — held billions of dollars of that token on its books, reportedly more than the entire market said the token was worth. Alameda then pledged this self-printed token as collateral to borrow real money — dollars and bitcoin — belonging to FTX's customers, via a software setting that let it withdraw without triggering the exchange's own risk limits. Roughly $8.9bn of customer funds went missing.
Was FTX the same thing as Lehman?
No, and it is important to be precise. Lehman exploited an accounting rule; the practice was, at the time, arguably legal, and it was disclosed to nobody because it did not have to be. FTX committed theft. Larry Summers made the sharper comparison at the time: FTX was not Lehman, it was Enron — not a systemic accident but a crime. Sam Bankman-Fried was sentenced to 25 years in federal prison in March 2024. What the two cases share is the mechanism of concealment and the failure of the gatekeeper, not the culpability.
Did the auditors know?
In Lehman's case, yes. The Examiner's Report found that EY was aware of Repo 105, reviewed the policy, and never questioned why the volume spiked precisely at quarter-end. A senior Lehman employee, Matthew Lee, went to EY as a whistleblower and warned them directly. EY continued to issue clean opinions. In FTX's case the failure was different in kind: the SEC found that the engagement partner at Prager Metis 'fundamentally did not understand FTX, or the crypto asset markets in which it operated'.
What happened to the auditors?
Prager Metis paid $1.95 million to settle SEC charges of negligence in the FTX audits and of violating auditor independence rules. EY settled with the New York Attorney General for $109 million — a figure worth holding next to the roughly $150 million Lehman had paid EY in audit fees over the preceding years, and against EY's global revenue, which was measured in tens of billions. No criminal charges were brought against anyone at either firm.
How do I check whether a company I own is doing this?
Read the 10-K on the SEC's EDGAR database — the actual filing, not the glossy annual report — and go to the footnotes. Search for 'Variable Interest Entities': these are one way risks are parked outside the main balance sheet. Then compute the leverage ratio, total assets divided by total equity, and compare it with the company's peers. If it is dramatically higher, that is not proof of anything, but it is a question worth being able to answer.
Has the problem been fixed since 2008?
It has largely moved. Non-bank financial intermediation — shadow banking — reached $256.8 trillion in 2024 according to the Financial Stability Board, or 51% of total global financial assets, growing at roughly double the pace of the banking sector. Private credit is a significant part of that, and the FSB itself notes severe limitations in the data available to measure it. That last point is the whole point: the risk did not disappear, it migrated to where the reporting is thinner.