The FTX weekend: which pegs broke

FTX was the second-largest crypto exchange and Alameda Research was its affiliated trading firm and one of the market's dominant liquidity providers. The trigger was a 2 November 2022 CoinDesk report on Alameda's balance sheet: of roughly $14.6B in assets, a large block — about $5.8B — was FTX's own FTT token, an illiquid asset Alameda itself could not sell at quoted prices without crashing it. The disclosure exposed that FTX's market-making arm was effectively capitalized by paper it minted.

On 6 November Binance announced it would liquidate its FTT, and the resulting loss of confidence became a classic bank run. FTX paused withdrawals on 8 November as FTT fell more than 80% to around $4, a Binance rescue letter of intent collapsed within a day, and on 11 November FTX, Alameda, and roughly 130 affiliates filed for Chapter 11. Sam Bankman-Fried was later convicted of fraud.

Critically, none of this was a reserve failure at a stablecoin. No issuer's cash, Treasuries, or collateral pool was impaired by the FTX bankruptcy itself. The stablecoin stress that accompanied the weekend came entirely from the collapse of a counterparty, which is why the pegs that moved cannot be diagnosed from their own mechanics alone.

Because Alameda was a top-tier market maker and lender, its forced deleveraging hit whatever it touched: positions it held, collateral it had pledged, and liquidity pools it traded against. As the firm scrambled for cash, on-chain observers and exchanges saw stablecoin selling that pushed individual pegs off the dollar — not because those pegs were unsound, but because a very large, distressed seller was unwinding into them over a thin weekend.

This is the missing failure mode that the algorithmic and reserve-banking case studies don't capture. In Terra the mechanism destroyed itself; in the SVB episode a reserve became temporarily unreachable. Here the stablecoins were largely fine internally — the shock arrived from outside, through the order book, and selected its targets by who was nearest the failing entity. A solvent peg can still print below par if the seller is big enough and the venue is shallow enough.

The practical lesson is that proximity is itself a risk dimension. Pharos tracks counterparty and venue concentration alongside reserve quality precisely because exposure to a single dominant trading firm or exchange can transmit stress that no amount of clean collateral prevents in the moment.

HUSD was a fiat-backed dollar tied to the Huobi exchange ecosystem. Its decline began before FTX's final days: after Justin Sun took a senior position at Huobi, the exchange moved to delist HUSD and shift trading volume into Sun's own stablecoin, USDD. With its home venue actively winding it down and helping users swap into USDT, HUSD lost the one thing a custodial dollar depends on — a credible redemption path.

On 31 October 2022 HUSD fell about 72% from its peg, touching roughly $0.28, and its market capitalization had already collapsed from more than $1B in 2021 to around $63M. There was no run on reserves in the usual sense; the issuer's ecosystem simply chose to stop supporting it, and the token had no independent demand to fall back on. It never reclaimed the dollar and now sits in the Pharos cemetery, recorded as a counterparty failure rather than a mechanism failure.

HUSD is the cleanest illustration of the thesis. It was not the weakest stablecoin design in the market — it was a custodial dollar whose fate was decided by the entity that controlled it. When that entity pivoted to a competing product, the peg died, and the FTX-weekend stablecoin turmoil that pulled USDD toward $0.96 was the same Sun-and-USDD nexus that had already finished HUSD off.

USDD was the more genuinely strained survivor. Around 9-10 November it slid to roughly $0.96 on several exchanges, which Justin Sun publicly attributed to Alameda selling USDD to fund FTX. At a market cap near $725M, USDD's collateral looked ample on paper but was largely staked and illiquid; counting only liquid reserves, its effective coverage was much tighter. The Tron DAO Reserve announced plans to buy up to $1B of USDT to reinforce backing, and the peg recovered, though it had clearly absorbed real damage.

USDT's deviation should be read more skeptically. A ~$0.96 quote appeared on a subset of venues and data aggregators around 10 November, but other venues showed lows nearer $0.97, Tether characterized it as a price-discrepancy issue rather than a reserve problem, and at least one aggregator later acknowledged a data fault. Tether did separately freeze USDT tied to FTX, and there was visible speculation that Alameda was shorting USDT, but the largest fiat-backed stablecoin reclaimed par within the session and within about two weeks was trading at $0.999.

Both assets stayed live and both reclaimed the dollar, which is why neither is the subject of this study. They are the control cases: they show how shallow and recoverable a contagion deviation is when the underlying reserves are sound and an intervention or arbitrage closes the gap.

The peg that breaks isn't always the weakest design — it's the one nearest the failing counterparty. FTX impaired no stablecoin's reserves, yet it killed HUSD and dented USDD and USDT, because contagion travelled through Alameda's positions and the venues it traded on. Counterparty and venue proximity belong in a peg's risk picture alongside collateral quality.

A custodial dollar is only as durable as the entity that runs it. HUSD had no mechanical flaw; it died because its controlling ecosystem chose to stop supporting it and pointed users at a competing token. Concentration of an issuer's distribution and redemption in a single exchange is a structural fragility, not a convenience.

Distinguish a data print from a depeg. The USDT ~$0.96 quote was shallow, venue-specific, and disputed, and it reclaimed par almost immediately — a different category of event from USDD's real-collateral strain or HUSD's terminal abandonment. Treating every off-par tick as equivalent overstates contagion; reading the source, depth, and recovery of each deviation is what separates noise from a genuine break.