FTX, RIP

FTX, RIP

19th century business history is full of fun and now completely illegal maneuvers, like flooding the market with kerosene in order to bankrupt refineries, and then buying them out. It's not that all of these tricks worked well: at one point, two competing railroads owned by competing robber barons got into a pricing war that knocked the price of a cattle car down by 90% to drive the competing line into bankruptcy—so one of the competitors secretly bought cows to ship on the line and take advantage of artificially cheap freight instead. Bank runs were part of the fun here, too, and often featured contagion between equity markets and the rest of the financial system: before people fixated on the yield curve inverting, one of the symptoms of a market crash in the pre-Fed era was that margin rates would rise to over 100% annually. Those margin rates would suck liquidity out of the rest of the economy and into the market, which provided a widely-visible signal that financial conditions were not good.

These kinds of things are rarer today, because of a combination of formal regulation, changing institutions (financial institutions don't want to spend most of their time wondering if their counterparties are solvent, and defending their own solvency, so they create clearinghouses), and social norms.1

In a lightly-regulated financial market that hasn't had time to develop institutions and norms, things are sort of like the old days. Here's the brief timeline of FTX's last few weeks:

• September 21: FTX is in talks to raise money at a $32bn valuation, in line with its most recent round.
• November 1, 9:44am ET: Coindesk writes a story on a leaked Alameda balance sheet showing large exposure to FTX's FTT token.
• November 4, 5:38am: a Substack post asks if the company is insolvent, and rumors spread over the weekend.
• November 6th, 10:47am: Changpeng "CZ" Zhao, CEO of rival exchange Binance, announces plans to sell his FTT tokens.
• 16 minutes later: Alameda Research's CEO offers to buy them.
• November 6th, 6:28pm: Sam Bankman-Fried tweets that the exchange is solvent. Many people remember their Bagehot.
• November 7th, 7:38am: SBF reiterates that the exchange is fine. The Bagehot-posting intensifies.
• November 8, 6:00am: Tech news sites cover slow withdrawals at FTX.
• November 8th, 11:03am: SBF announces that FTX has signed a letter of intent to be acquired by Binance, which has been confirmed by CZ. By this time, total losses based on the positions in the Coindesk report are at least $2.3bn, mostly from FTT depreciation. (Those losses will roughly double in the next few hours as FTT continues to depreciate.)

FTX was apparently looking for rescue funding before the deal, though the amount they needed rose from $1bn initially to $5-6bn by the end.

The transaction doesn't affect Alameda Research, the crypto market-maker, nor does it involve FTX.us, which was about 5% of FTX's revenues. And acquisition price hasn't been announced, though distressed sellers who got rejected by buyers and got 500% more distressed over a period of less than 24 hours tend to have little negotiating power. One of the open questions is: why, exactly, did FTX need to sell? Granted, its associated token had gone down in value as exchange users rushed for the exits. But normal exchanges shouldn’t have any reason to take risk onto its own books. They are typically "levered" in the sense that its customers' assets are its own liabilities, but a properly-run exchange that has these liabilities on its books also has exactly matching assets.

But FTX isn't a normal exchange—it's a crypto exchange, and like most crypto exchanges, it has elements of an exchange, a clearinghouse, and a prime brokerage, all rolled into one. But unlike other crypto exchanges, it has yet another feature, an associated trading firm, Alameda Research.

One deep irony to all of this is that FTX's big competitive advantage was its risk management. Historically, exchanges offered margin trading (the people who like highly volatile crypto products like the extra volatility of leverage even more). Other exchanges would sometimes offer margin on a per-product basis rather than a portfolio basis, so someone could get liquidated on one position even if they had collateral for other positions.2 FTX's early investor materials talked about how some exchanges that offered leverage would repeatedly break trades: if there's one trader long a position with 10% equity, and another short the same position with 10% equity, then a sudden 20% move in either direction wipes one of these traders out, and the shortfall has to be covered by either the exchange or whoever was on the other side of the trade. And, historically, it's been the latter. Since crypto prices jump around, liquidating in the normal way is a problem. This margin situation can mean that customer withdrawals require more cash than the exchange immediately has on hand: customers will rarely be courteous enough to exit their Ethereum long futures positions at exactly the pace at which other customers are exiting short positions.

This also made FTX's relationship with Alameda Research a synergistic one. Alameda does systematic crypto trading, and while it doesn't have exactly the same ownership as FTX, there's lots of overlap in the form of SBF controlling both. And Alameda made up around 6% of FTX's exchange volume. Building a crypto fund and then realizing the ideal exchange for crypto funds doesn't exist yet is a nice example of a business pivoting into a higher-value complement.3 But the complementarity goes beyond that: any time Alameda's models tell it that some crypto product is oversold, the Alameda-and-FTX complex makes money in three ways—first from the trade itself, second from not needing to pay for liquidations, and third because traders come to view FTX as an exchange that provides more emergency liquidity than alternatives. And working with a derivatives-heavy exchange gives a market-maker some fun opportunities for repeatedly profitable trades; basis trades between crypto spot prices and crypto futures (i.e. buying the underlying asset shorting futures when futures are expensive and waiting for them to converge) can provide easy repeat income. But they also tie up capital, and when futures prices get more out of whack they tie up even more capital.4 Alameda is vague about what they do, whether out of the classic quant tradition of not giving up any more secrets than necessary or because what they do is so intimately tied with FTX's operations. So it's hard to say what specific trades may have gone wrong, but very easy to imagine a number of practices that could have started underperforming in response to traders exiting FTX.

Meanwhile, there's FTT: it's a token issued by FTX, which gives holders interest if they stake it, access to crypto airdrops, and a discount on commissions of up to 60%. FTX has a clever system where some of their commissions are used to buy and burn FTT, which makes it an equity-like: you could buy FTT because you expect a "dividend" from saving on commissions, or because you expect to profit from the "buyback" of automatic repurchases. One entity that owned a lot of FTT—$5.8bn as of June 30th—was none other than Alameda itself.

This could have been a completely cynical perpetual motion machine where Alameda makes FTX a better exchange by providing cheap liquidity, and the growth of FTX increases the value of the underlying FTT, giving Alameda more collateral to borrow against for providing more liquidity. It's certainly a complicated enough setup that the company had to go to some lengths to assert its credibility. A public persona, especially as a smart, no-nonsense trader, serves the same marketing function today that building a bank modeled after a Greek temple served in the late 19th century. It's a way to reassure investors that someone trustworthy is in control.

This "perpetual motion machine" is not necessarily a bad thing. It's also a description of the flywheel effect that many companies have when they monetize the same general thing in multiple ways: Disney, too, has a perpetual motion machine where movies and shows create foot traffic to theme parks, whose new rides sometimes inspire movies. There's nothing untoward about this (unless you're a parent keeping track of your large and rising naked short position in Disney-related toys and subscription costs).

But in this case, it meant that the entire FTX/Alameda entity was mostly long crypto volatility (higher transaction volume meant both more profitable trades for Alameda and more commissions for FTX) but with a position that would flip at some uncertain point to being net short volatility—if Alameda cushioned too many price swings, it would switch from mitigating volatility to transferring its own assets to irresponsible traders.

A long spell of low crypto volatility starts to get costly for this kind of operation. And that's also visible in prices: FTT started the year at $39.17 and was down to $26.25 at the end of October. If this meant that Alameda's own liquidity was gradually bleeding, then their ability to withstand a big move in crypto was weaker.

And that made them especially vulnerable to anything that a) made crypto suddenly spike in volatility, and b) that made FTT drop. CZ of Binance knew exactly what he was doing when he announced plans to liquidate FTT. (It's possible that any prudent exchange, as part of its risk management, would try to remain aware of the financial position of other exchanges, particularly those that, like FTX, had lots of derivatives exposure. No sense in Binance running the risk that it would have to liquidate or claw back some of its own traders because of problems somewhere else.)

Once FTT was dropping, the value of FTX was pretty binary: if they survived, they were worth less than before but still a lot. Call it $10bn, to account for either higher risk or the higher capital requirements to mitigate that risk. If FTT kept dropping, though, the entire formula would break, and at some point Alameda might be demanding liquidity rather than supplying it.

The pace of that decline was partly a function of whether or not speculators trusted FTX, and that was something CZ could partly control. (Not completely, as it turns out. FTT was trading at $15 just before the deal was announced, popped to $19 minutes later, and traded as low as $3.15 earlier today. It's now around $4.66.) Presumably, this deal went through at a minimal price. Matt Levine describes the expected offer as "We will buy your exchange, make sure that all your customers are made whole, and give you a Snickers bar in exchange for 100% of the equity."

Why do good traders cut losses? Why did SBF take the Snickers? A hypothetically perfect trader wouldn't do this, because that trader would have perfect information and wouldn't make losing trades. Traders cut losses because when the price moves against them strongly enough, it's clear that their mental model of the opportunity was wrong. And once you don't have a good mental model for where an asset should trade, your confidence interval for where it will trade is a lot wider. Combine that with leverage, and traders cut losses because past a certain point, any loss starts to imply insolvency risk. Given time, SBF might have been able to restructure things so Alameda used a bit less leverage and FTT was more widely-distributed. But another way to look at the stylized fact that all correlations go to 1 in a crisis is that crisis is a form of time travel in which every bad thing happens all at once. There's just no time to do anything but take an offer that stops the bleeding, and then regroup.

One of the hard questions interviewers sometimes had for Sam Bankman-Fried was: is this whole FTX thing just a utilitarian move to build a really great casino that will make money from gambling and donate it to worthy causes? We'll probably never know. An exchange can be functionally equivalent to a casino in normal circumstances, but there's a difference: if that exchange is also a clearinghouse that is closely tied to a market-maker, then it's possible to lose a sequence of ever-larger bets until you're effectively wiped out.