To be clear, we still don’t have a perfect understanding of what exactly happened at Alameda Research and FTX. However, at this point, I feel that we have enough information to get a grasp on the broad strokes. Through a combination of Twitter users’ investigations, forum anecdotes, and official news releases, the history of these two intertwined companies becomes progressively less hazy, slowly coalescing into something resembling a consistent narrative.

Of course, without witness testimonies and a full financial investigation, our claims only remain tentative at best. Any given piece of information may be flawed or even fabricated. However, if they are assembled together and put in context, they together lend credence to the following timeline:

SBF, Trabucco, and Caroline were (probably) initially well-intentioned but not especially competent at running a trading firm

Alameda Research made large amounts of book profits via leveraged longs and illiquid equity deals in the 2020-2021 bull market

Although Alameda was likely initially profitable as a market maker, their edge eventually degraded and their systems became unprofitable

Despite success with some discretionary positions, on net, Alameda & FTX jointly continued to lose large amounts of money and liquid cash throughout 2021-2022 as a result of excessive discretionary spending, illiquid venture investments, uncompetitive market-making strategies, risky lending practices, lackluster internal accounting, and general deficiencies in overall organizational ability

When loans were recalled in early 2022, an emergency decision was made to use FTX users’ deposits to repay creditors

This repayment spurred on increasingly erratic behavior and unprofitable gambling, eventually resulting in total insolvency

Details follow below. (Many thanks to all those who have contributed to this article, be it through private discussions or through public content that I’ve quoted or otherwise relied upon.)

Alameda Research probably lost >$15 billion dollars

To understand the FTX bankruptcy, we have to first understand the scope of the problem at hand. Most news accounts seem to portray the scale of the bankruptcy as relatively small. For example, the New York Times suggests that user deposits were used to make up for money that had gone into venture investments:

Meanwhile, at a meeting with Alameda employees on Wednesday, Ms. Ellison explained what had caused the collapse, according to a person familiar with the matter. Her voice shaking, she apologized, saying she had let the group down. Over recent months, she said, Alameda had taken out loans and used the money to make venture capital investments, among other expenditures.

Around the time the crypto market crashed this spring, Ms. Ellison explained, lenders moved to recall those loans, the person familiar with the meeting said. But the funds that Alameda had spent were no longer easily available, so the company used FTX customer funds to make the payments. Besides her and Mr. Bankman-Fried, she said, two other people knew about the arrangement: Mr. Singh and Mr. Wang.

Similarly, Matt Levine’s column seems to imply that the drop in the value of FTT used as collateral resulted in an enormous imbalance between assets and liabilities:

Now let’s add one more crypto element. If you are a crypto exchange, you might issue your own crypto token. FTX issues a token called FTT. The attributes of this token are, like, it entitles you to some discounts and stuff, but the main attribute is that FTX periodically uses a portion of its profits to buy back FTT tokens. This makes FTT kind of like stock in FTX: The higher FTX’s profits are, the higher the price of FTT will be. It is not actually stock in FTX — in fact FTX is a company and has stock and venture capitalists bought it, etc. — but it is a lot like stock in FTX. FTT is a bet on FTX’s future profits.

But it is also a crypto token, which means that a customer can come to you and post $100 worth of FTT as collateral and borrow $50 worth of Bitcoin, or dollars, or whatever, against that collateral, just as they would with any other token. Or something; you might set the margin requirements higher or lower, letting customers borrow 25% or 50% or 95% of the value of their FTT token collateral.

Both of these accounts miss a crucial part of the story. First, FTX is missing about $8 billion dollars’ worth of users’ collateral. Even if you consider the sum total of venture investments made by FTX and Alameda together, as well as a marginal drop in collateral value as a result of an FTT price decline, it simply does not make sense for FTX to be $8b in debt. The losses would be significant, yes, but they alone do not constitute a sufficient explanation for FTX’s bankruptcy.

In addition to that, it was popularly believed that FTX and Alameda together were enormously profitable, as a result of:

High trading fees on FTX combined with large user trading volumes

Extremely lucrative (or, depending on your perspective, predatory) venture deals for tokens such as SOL, MAPS, OXY, SRM, etc.

Highly probable collusion between Alameda and FTX to give Alameda an edge over other market makers on FTX

Although it is difficult to put an exact dollar value to their estimated profits, it was believed that these avenues, especially the lucrative venture deals, were responsible for at least $10b in profit between Alameda and FTX combined.

We are therefore left with an even greater mystery. Somehow, it seems as though Alameda and FTX managed to burn through >$15 billion dollars’ worth of profits (likely more). This is an incredible shortfall, and, remarkably, no comprehensive account to date has emerged showing exactly how this came about!

We may never really know where all of the money went. However, we provide a number of separate hypotheses which, if combined together, could plausibly account for losses of $15 billion or more.

Alameda’s market-making edge decayed and they started punting longs

There has typically been a perception of Alameda as an extremely competent and profitable market maker. But is this perception actually accurate?

First of all, it’s worth noting that although the inner circle’s backgrounds (SBF and Caroline from Jane Street, Trabucco from SIG) are impressive, they are not exceptional. Having several years’ worth of experience at a trading firm does not make you a geniusーthe top tier of such firms hires well over a hundred new employees every year. In a podcast from 2019, SBF brags about making predictions on the timescale of 1-2 seconds:

Sam: Yes. A lot of exchanges have, if nothing else, volume-based features, and often we can get in the top tier there, and that does help a fair bit, especially when it comes to eeking out a business plan. […] That’s nothing. You made one basis point. If you guess randomly, maybe moderately right, about what Bitcoin is going to do in the next two seconds, maybe you make a basis point. [00:22:30] But if you can actually do that on one billion dollars a day of volume, then you can do the math, that’s actually a lot of money. We can’t always do that, but it sort of does mean that at some volume scale, you really do get down to, everything matters. Every little edge matters. So there is some of that. A lot of exchanges reach out to us about being liquidity fighters. We do do that on some exchanges.

While this might have been considered highly competitive in 2019-era crypto markets, it is a far cry from the microsecond-level precision of market making in traditional finance. (N.B.: Typically microsecond-level precision refers to a slightly different concept; however, even prediction of directional price movement takes place at timescales much lower than two seconds!) Such strategies may have served well three years ago, but as larger, extremely competent, and well capitalized market makers like Tower and XTX began to trade cryptocurrency markets, it is plausible that Alameda slowly lost its edge, much as Doug Colkitt suggests.

What do you do when:

You start losing your edge as market maker despite all the hard work you put into writing complicated code, and

Everyone around you is printing generational fortunes (at least on paper) by longing absolute dogshit?

Your natural inclination, of course, is to neglect your market-making activities and join the punters.

Several statements from Alameda’s executives themselves bolster this theory. For example, Trabucco describes a news-based trading strategy in April of 2021:

To be clear, he is literally describing taking on leveraged crypto market beta because of an emergent narrative about institutional adoptionーexactly the same reason why many retail traders invested in cryptocurrency throughout 2021.

He also describes longing DOGE over a timescale of multiple months because Elon Musk tweets about it a lot:

Again, let me be clear: this is the exact same sort of talk that you hear from your Uber or Lyft driver, especially close to the local tops of the bull market, about the $50 that they put into DOGE and SHIB turning into $200. This is not a sophisticated algorithmic trading strategy by any stretch of the imagination.

There’s no doubt that their BTC and DOGE longs, as described, made money. And there’s perhaps even an argument to be made that they were reasonably smart trades! The more important point, perhaps, is that this is simply just not “quant trading” and that Alameda is clearly expanding into territory where their edge, relative to other market participants, is much less quantifiable and seems relatively disconnected from their areas of expertise. They may have won on the trades they showcasedーbut what about trades where they didn’t win?

Turning to Caroline, she herself, a month ago, strongly implies that she prefers to punt on longs rather than pick up pennies in the algorithmic jungle:

Naturally, these strategies do very well in a bull market, when almost every long position goes up! Remarkably, Trabucco attributes their success to skill rather than simple market beta:

We can therefore speculate that their trading strategies were a combination of (1) negative-edge market making and (2) discretionary longs. Indeed, the more unprofitable their algorithmic trading was, the more tempted one imagines they were to make back all the losses with “brilliant” longs on BTC and DOGE. On net, it is likely that they were quite profitable throughout a good deal of 2020-2021, but as soon as the market began to reverse in late 2021, it is probable that their overall PnL suffered dramatically.

There is definitely a lot of discretionary trading that Alameda did do correctly; buying Solana at low prices, accumulating massive quantities of low-float Solana ecosystem tokens, boosting the Solana ecosystem overall, buying oversold liquidations, and so on. Ultimately, though, one gets the sense that perhaps they overextrapolated from the ambient bull market and ended up overrating their own trading ability, with consequent losses later down the line. Their strategy of borrowing against illiquid ecosystem coins (discussed in a later section), rather than selling them more consistently, may also have worked to their detriment.

To the astute reader, one part of the above narrative may seem rather odd. Suppose that you look at your internal PnL graph and you see that your market making is in the red every single day; wouldn’t you simply… turn off your quoter? Well, not necessarily. It is easy to imagine that they justified the market making in terms of the inflated trading volumes it generated on FTX, which in turn justified higher venture valuations for the exchange. Beyond that…

Alameda was an incredibly disorganized, poorly run trading firm

Working at the top trading firms for a couple years clearly does not mean you are a top trader. It also does not mean that you are good at organizing business practices.

An ex-Alameda employee on the Effective Altruism forum shared this telling account of Alameda’s internal practices:

This comment is merely an anecdotal account based on the commenter’s personal memories; however, certain of the details mentioned are corroborated by personal accounts that have been communicated to me privately, which makes me inclined to believe that this is a legitimate and largely accurate portrayal of Alameda’s internal operations. It describes a truly sorry state of affairs, with incredible amounts of money lost to poor bookkeeping, arbitrary discretionary trades under SBF’s direction, internal mismanagement, and incredibly poor organization.

Again, this picture is consistent with a number of stories I have heard via trusted sources, both firsthand and secondhand, all painting a picture of extreme organizational dysfunction, fragmented information, and poor capital management. For example, a personal friend whose company received venture funding from FTX remarked that despite falling very behind on a prior commitment to providing monthly status updates, nobody from Alameda ever bothered to follow up with them. Other accounts are quite similar, and all describe a company culture best described as “flying by the seat of one’s pants.”

If a firm is incapable of scaling back a $3m/year AWS bill or even tracking whether or not millions of dollars’ worth of payments have arrived or not, what are the chances they have accurate tracking of their PnL, let alone a clear view of the state of their assets and liabilities at any given moment? I would say those chances are quite poor. Even if they saw their market making strategies lose money every day, it is not difficult to imagine them saying to themselves: “Well, we may be down a hundred thousand dollars today, but our locked tokens or long positions have appreciated millions of dollars’ worth of value; even applying a large haircut to be conservative, we are still ahead. Why turn off the quoter and jeopardize liquidity at FTX? Instead, we can just throw an intern at the algos and see if they can fix them. If not, no huge loss.”

When you only have a vague understanding of your company’s books, especially considering the incredible amounts of money that FTX expended on advertisements, branding deals, and other discretionary expenses, it is likely that they did not realize the severity of their situation until their loans started being recalled after the LUNA implosion. When you actually have to come up with money and find yourself short, it is at that point which you have to confront your enormous losses, and likely tempted SBF et al. to shore up what they hoped was a temporary deficit with customers’ deposits from FTX. That, in turn, prompted yet more degenerate, riskier forms of gambling, digging the hole even deeper…

Sam Bankman-Fried was erratic, rash, and potentially incompetent

At the end of the day, though, isn’t it weird for a group of ex-traders to engage in such reckless behavior? Although we know little about Caroline and Trabucco personally, we thankfully have a surplus of anecdotes about SBF as an employer and manager.

It is important to note that we should take any personal anecdotes quoted below with a grain of salt. It is always easy to come out after the fact and say that someone who was accused of a crime clearly showed signs of criminality before the fact. Nevertheless, they are consistent in the behavior they describe as well as consistent with non-anecdotal observations made below.

It’s clear that SBF had an “enormous appetite for risk,” as an ex-FTX employee commented:

In fact, SBF is on the record as denying the applicability of the Kelly criterion for bet sizing:

To be entirely clear, this simply does not make sense in the case of repeated bets. As Matt Hollerbach points out, this misunderstanding arises from conflation of the arithmetic and geometric growth rates of wealth over time. SBF is provably wrong about this, yet refuses to reconsider his logicーa curious quality for someone who worked at Jane Street and studied physics at MIT!

Why might SBF be so insistent on intentionally oversizing his bets? Purely because of intellectual confusion about what strategy is optimal for long-run growth? An alternative hypothesis is that, beyond any natural proclivity to risk-taking, he may have been taking dopaminergic drugs (prescribed for for Parkinson’s disease) as a nootropic. These drugs as well known to cause risky behavior such as compulsive gambling or shopping sprees!

Autism Capital recently shared an account from an ex-FTX employee recounting how SBF encouraged extreme use of stimulants:

This is, of course, consistent with SBF’s self-admitted usage of stimulants as performance enhancers:

What is particularly notable, however, is SBF’s usage of “patch[es] for … stimulants.” In a follow-up Tweet, Autism Capital reports that the patches are Emsam (US brand name for selegiline), a MAO-B inhibitor used to treat Parkinson’s which increases levels of dopamine in the brain. This was confirmed with some excellent follow-up detective work, also from Autism Capital, which pinpointed a particular video frame from a video showing SBF at his desk:

If you zoom in, you can see what appears to be wrappers for some form of medication:

These match perfectly with Emsam’s commercial packaging!

Dopaminergic medications for Parkinson’s disease have long been associated with extreme risk-taking, such as in this study of pathological gambling in PD patients taking medication:

Many other such references exist in the literature. Remarkably, there is even a case study of selegiline-induced hypomania, as well one reporting hypersexuality and paraphilia! Overall, it seems very likely that SBF’s propensity to take extreme levels of risk was elevated to preternaturally high levels via heavy, habitual usage of amphetamines and selegiline, causing him to rationalize nonsensical strategies as optimal via convoluted “linear wealth utility” reasoning.

Note that PD medications have also been strongly associated with the development of compulsive shopping. FTX’s massive spending on advertisements and branding partnerships may very well have been a ploy to attract deposits, but it may also have been driven, in part, by SBF’s continual abuse of these drugs. For example, there are reports that a $12 million ad campaign with famous Japanese baseball Shohei Ohtani was killed after a single day on air. Similarly, FTX acquired naming rights to the e-sports organization TSM for a stunning $210 millionーfar out of line with comparable deals in the e-sports industry. Even his property acquisitions are stunning, with a reportedly $200m real estate portfolio in the Bahamas. These cannot be rationalized as risky bets with positive expected value, and only make sense in the context of executive management which is either incompetent or, as we suggest, literally drugged up and on a multi-year shopping spree.

One can also argue that there are clear signs of SBF’s deficits in the realm of overall competence and cognitive ability. For example, his personal blog, Measuring Shadows, is incredibly dullーit is a compendium of overly long posts about effective altruism, moral philosophy, and baseball statistics. In comparison, Caroline Ellison’s blog is far superior, demonstrating a level of thoughtfulness and wit that surpasses SBF by many leagues. Perhaps more tellingly, the Financial Times reported that SBF is remarkably bad at League of Legends:

While this may initially seem like a frivolous observation, it is actually shocking that SBF was unable to rank higher than the Bronze or Silver league after years of regular play across hundreds if not thousands of individual games. It is reflective of an incredibly impaired level of cognition, much like someone who was unable to learn how to ride a bicycle after an entire month of practice or someone who never progresses past the level of an average elementary school student after five years’ of daily piano practice. To me, this is one of the strongest indications possible that SBF is, to be blunt, “not entirely there.”

Just in the last 24 hours, SBF has been tweeting out the phrase “What HAPPENED” letter-by-letter, with gaps of hours between some tweets:

This is, frankly, just absurd and incomprehensible. It makes absolutely no senseーwhat is he doing? What could be possibly hope to accomplish with this? I struggle to imagine the state of his mind at the moment. Does he think that this is an elaborate joke? Is he hopped up on a metric ton of obscure drugs and beginning to lose his sanity? Neither of those bode well for the integrity of his cognition. (It was originally speculated that this was some bizarre scheme to avoid having deleted tweets show up on deletion trackers, but that theory has since come under criticism.)

How do we reconcile these observations with SBF’s MIT degree, Jane Street pedigree, successful execution of international Bitcoin price arbitrage, and the founding of Alameda Research and FTX (which, even if ultimately failures, still required some degree of executive function to actually start)? It is not clear, but one could speculate that excessive drug usage may have literally “fried his brain” to some extent. Ultimately, it is challenging to speculate about why his cognitive performance has degraded to such a poor level, but the fact remains that it does appear to be quite poor.

One cannot help but wonder if this general deficit in intellectual ability is why the FTX balance sheet (provided by SBF earlier last week) is so farcically simplistic:

Frankly, this can barely be considered a serious attempt at accounting. It is so absurd that one’s initial assumption is that it is some sort of insane attempt at the financial equivalent of “making up an answer when the teacher calls on you.” But what if this is actually reflective of the quality of his thought.

If the above observations about SBF’s personality and competence are even half true, that would go a great deal toward explaining Alameda’s losses. (Even though he formally stepped down as CEO, he of course maintained a close relationship with Caroline and Trabucco.) In particular, once customer deposits were raided, he may very well have tried to “make it all back” in increasingly desperate attempts that just dug the hole deeper and deeperーhardly unsurprising if he indeed was actually operating as a mildly brain-addled compulsive gambler.

Collusion between Alameda and FTX caused huge losses from algo failures

It has long been suspected, and corroborated by multiple reports, that FTX and Alameda were essentially operating in total collusion, with Alameda frontrunning token listings and potentially having special privileges to bypass risk checks. (The famously slow latency experienced by API traders on FTX may have been essentially engineered as an artificial barrier that only Alameda could skip.)

Doug Colkitt speculated that this could actually have caused an “algo meltdown,” similar to a well-known incident with Knight Capital Group in traditional markets:

Reports communicated privately suggest that this may have actually been the case, resulting in losses potentially in excess of $1 billion dollars! And, of course, there may have been other incidents not known to me.

Without full access to FTX’s records, it is challenging to really confirm or deny these claims, although the sources are credible, and I am inclined to believe them. If true, the scale of these losses is more than enough to be a major contributor to FTX & Alameda’s overall shortfall.

Loans collateralized by FTT/SRM resulted in reflexive liquidations

A trusted friend (non-insider) anonymously shared the following theory with Autism Capital, which I will reproduce here in full as it is clearly and concisely written:

The theory is quite convincing, and boils down to an increasingly desperate series of attempts to backstop a series of reflexive liquidations of loans backed with illiquid shitcoins (FTT/SRM). Because of token emissions, even maintaining their prices at a constant level requires a constant inflow of capital, continually increasing Alameda’s exposure to these coins and reducing their cash reserves. Ultimately, they ended up in a vulnerable position where they simply ran out of the money required to meet demand for customer withdrawals while preventing a collateral liquidation cascade.

In isolation, I am not sure if this theory fully explains the losses suffered by FTX and Alameda. However, positioned in the context of the other theories advanced above, it is certainly plausible that this theory explains a fair portion of the shortfall.

This theory dovetails well with Caroline’s admission that FTX transferred over customers’ deposits after the LUNA crash in order to repay recalled loans. In particular, it provides a reason as to why FTX was essentially compelled to bail out insolvent lenders like Voyager and BlockFi. On top of loans recalled by other entities, such as Genesis, this would represent a substantial and unanticipated short-term demand for cash. If Alameda was already operating with poor bookkeeping, and especially if Alameda were substantially exposed to the LUNA crash itself, it becomes increasingly easy to imagine that SBF et al. felt they had no other option than to dip into customers’ deposits.

Summary

So, what have we really learned? In the end, we don’t actually have a great idea of exactly how Alameda and FTX burned through as many billions of dollars as they did. However, there are sufficiently many competing theories available, each with a sufficient amount of supporting evidence, that I suspect that the reality of the matter looks something much like a combination of all of the above factors.

We can try to give a very rough accounting of their potential losses:

Voyager/BlockFi acquisition: 1.5b

LUNA exposure: 1b

KCG-style algo crash: 1b

FTT/SRM collateral maintenance: 2b

Venture capital: 2b

Real estate, branding, other frivolous spending: 2b

FTT drop from $22 to $4: 4b

Discretionary longs going bad: 2b

Total: 15.5 billion

This is, of course, very rough. I have no idea how much they lost in LUNA or how much they might have blown on longing shitcoins. I have not bothered to actually calculate out the total of their venture investments, nor have I gone and added up every single one of their branding partnerships. Yet the overall picture is clearーwe have enough potential sources of losses that even if some of the numbers are off here and there, it is now at least conceivable how they might have arrived at such staggering losses. You can make up your own numbers, but the point is that there are many reasonable combinations of figures above that could result in losses equaling or even vastly exceeding $15 billion.

I make no claim to the ultimate veracity of the above information. Instead, my purpose is largely to compile these here into something of a patchwork narrative, which I do not believe exists at present. The reader is invited to draw their own conclusions!

(BY 0xfbifemboy）