What actually failed at FTX?
Mt. Gox was a badly run startup. FTX was a large, well-known, heavily marketed exchange with institutional investors and a compliance department. That difference is the reason to study it: the second failure was not caused by the absence of sophistication.
The six days
FTX was founded in 2019 by Sam Bankman-Fried, who had earlier founded the affiliated trading firm Alameda Research.
On 2 November 2022, CoinDesk published a report on a leaked Alameda balance sheet showing that a large share of its assets consisted of FTT — the exchange token issued by FTX itself. On 6 November, Binance's chief executive announced his firm would sell its FTT holdings. A customer run began. On 8 November FTX halted withdrawals. On 11 November 2022, FTX Trading Ltd and around 130 affiliated entities filed for Chapter 11 bankruptcy.
John J. Ray III, the restructuring specialist appointed chief executive, wrote in a court declaration on 17 November: "Never in my career have I seen such a complete failure of corporate controls and such a complete absence of trustworthy financial information as occurred here."
Bankman-Fried was convicted on seven counts including wire fraud in November 2023 and sentenced to 25 years in March 2024. Senior colleagues at both firms pleaded guilty and cooperated.
What broke, mechanically
Customer assets were not segregated. Deposits were commingled and made available to the affiliated trading firm. Some customer deposits had been routed through a bank account held by a related entity.
The risk engine had an exemption. Every account on the exchange was subject to automatic liquidation when its collateral fell short — every account except Alameda's, which was configured to be exempt and carried an effectively unlimited line of credit. The single account large enough to sink the exchange was the one account the risk system could not act on. A control with an exception for the largest exposure is not a control.
The collateral was self-issued. A large part of the value backing the position was FTT, a token created by FTX whose price depended on confidence in FTX. This is circular collateral, and it is the most important idea in the case.
The books were not trustworthy. Ray's filings described the absence of an effective board, the use of off-the-shelf small-business accounting for a multi-billion-dollar group, expense approvals issued through chat messages, and no reliable list of the group's own bank accounts and entities.
Why circular collateral evaporates
Illustrative arithmetic. An entity holds 100 million units of a token it issued itself, quoted at $25. On the balance sheet: $2.5bn of assets. Now note that the token's genuine daily traded volume is a small fraction of that, and that most of the supply is held by the issuer and a handful of insiders.
The mark is a price at which nobody has ever bought $2.5bn of the token, and could not. When forced selling starts, the price falls, the collateral value falls with it, and it falls for exactly the same reason the position needed collateral in the first place. The asset and the liability are driven by one variable.
This is why "assets exceed liabilities" is meaningless until you ask what the assets are and who issued them. A solvency measured in your own token is a statement about your reputation, quoted in currency units.
The uncomfortable summary
Strip the crypto vocabulary and three of the four failures are ordinary financial-control failures: commingled client money, an exempted account, unreliable books. Any regulated broker-dealer supervisor would recognise all three, and prudential regimes exist largely to catch them.
What crypto contributed was speed and the absence of a supervisor watching in between. Six days from a published balance sheet to a withdrawal halt, and nine to the filing, is faster than most traditional bank runs, because the balance sheet was public, the market traded continuously, and nothing stood between the disclosure and the exit but the exchange's own decision to close the door.
In the data
The collateral test in step 1 can be run on FTX's own token, whose price history still holds the whole episode:
On 1 November 2022 FTT closed near $25.87 on about $62 million of trading. On 8 November it opened at $22.14, touched $3.15 and traded more than $3.3 billion. Turnover more than fifty times the ordinary days before it is what a mark becoming a price looks like, and it was the volume, not the closing price, that had told you all along the mark was never tradable at size.
Try it now
- For any token used as collateral anywhere, compare its market capitalisation with its actual daily traded volume, and estimate what fraction of a large position could be sold in a day without moving the price. That fraction, not the market cap, is what the collateral is worth under stress.
- For any leveraged arrangement you can read about, identify who issued the collateral and whether that issuer is the same entity whose solvency depends on it. Circularity is the fastest thing on this list to check.
- Write out the FTX timeline with dates — balance sheet published, withdrawal halt, filing — and note how many days a user actually had. Then compare it with the Mt. Gox timeline from the previous lesson.