When Genius Failed

Long-Term Capital Management, founded in 1994 by the former Salomon Brothers bond trader John Meriwether, ran a strategy called convergence arbitrage: buy a bond that trades cheap for a reason unrelated to its promise to pay — usually a loss of liquidity — and short its near-identical, more liquid twin, betting the price gap closes. In 1994 the fund found a textbook case. A Treasury bond issued in February 1993 was trading at a yield 12 basis points above an almost identical bond issued six months later, purely because traders preferred the newer, more liquid “on-the-run” issue. The price gap on a pair of $1,000 bonds came to $15.80.

Twelve basis points is nothing. Long-Term made it something by scaling the trade to a size that mattered: it bought $1 billion of the cheap “off-the-run” bonds and sold $1 billion of the expensive on-the-run bonds. The firm’s own risk model held that owning one and shorting the other was “one twenty-fifth as risky as owning either bond outright” — so it reckoned it could “prudently leverage this long/short arbitrage twenty-five times.” The model was the entire justification for the size, and the size was the entire reason a few basis points turned into millions of dollars.

The leverage was financed almost for free. Long-Term bought the off-the-run bonds, immediately loaned them out to a Wall Street firm for cash collateral, then used that same cash as collateral to borrow the bonds it needed to sell short — repo financing, running in both directions at once. The trade barely touched the fund’s own capital. That was the appeal. It was also the mechanism by which Long-Term’s balance sheet grew far faster than its equity did.

Long-Term’s partners included Myron Scholes and Robert Merton, who would share the 1997 Nobel Prize in economics for option-pricing theory — the closest thing finance had to a scientific pedigree. By the end of 1995 the firm’s equity capital had roughly tripled to $3.6 billion on new investor money, while its balance sheet grew even faster, to $102 billion in assets. That put Long-Term’s leverage at 28 to 1 — a ratio that did not count the derivatives book, which added exposure the balance sheet never recorded at all.

Leverage of that size compresses volatility almost to nothing until it doesn’t. On a $102 billion balance sheet, Long-Term’s actual return on total capital — what it would have earned investing only its own money — came to roughly 2.45 percent. The 59 percent headline return investors saw that year was leverage applied to that thin real number, not a larger edge the model had found.

Model-driven confidence extended into positions no risk system could have made liquid on demand. The trader Haghani once bet $2.3 billion — half long on Shell, half short on Royal Dutch — on a small mispricing between two shares of what was structurally the same company. “A position that large was totally illiquid.” Illiquidity is not itself a flaw; a patient holder can wait it out. A leveraged holder cannot always choose to be patient.

The same faith in a model, and the same blind spot for what happens once the model’s assumptions stop holding, is the subject of The Physics of Wall Street — written years after Long-Term collapsed, describing the identical failure mode in the funds that came after it.

By September 1998, Long-Term’s positions were an open secret. Banks that had lent the fund money, cleared its trades, or simply talked to its traders for years had pieced together a rough map of what it held — and once a fund that size is known to be losing money and unable to sell, other desks can profit by selling the same things first. One Goldman trader watched his own firm work against the fund it was supposedly trying to save: “As they hammered away at Long-Term’s trades, Leahy felt sick, as though the firm’s competitors were liquidating Long-Term’s own positions for it.”

Lawrence Hilibrand, the partner most responsible for the firm’s decade of secrecy about its own portfolio, put the lesson in one line: “when you bare your secrets, you’re left naked.” Long-Term’s edge had always depended on no one else knowing what it owned. Once its size and its distress were both visible, that same knowledge became a weapon pointed the other way.

Goldman Sachs’s Jon Corzine put the ordinary logic of it on the record: Goldman traders “did things in markets that might have ended up hurting LTCM. We had to protect our own positions. That part I’m not apologetic for.” No conspiracy was required. Ordinary self-preservation by a counterparty who can see your position is enough — which is what counterparty risk means once a book is exposed: not that a counterparty might fail to pay you, but that a counterparty who can see you are weak has every reason to trade against you before you can save yourself.

The same instinct toward secrecy, and the same vulnerability once secrecy fails, runs through Inside the Black Box, Rishi Narang’s account of how quant funds actually operate day to day.

The trigger was external and, on paper, contained. On August 17, 1998, Russia declared a moratorium on $13.5 billion of its own ruble-denominated debt — a government defaulting on obligations in its own currency, which Lowenstein notes had not happened even at the depths of Latin America’s debt crisis. Long-Term had bet, along with much of Wall Street, that a nuclear power simply would not default. It was wrong about that call. But the moratorium alone did not break the fund.

What broke Long-Term was that it could not get out. “Long-Term knew it had to reduce its positions, but it couldn’t.” Its book had grown to a “mind-boggling sixty thousand individual positions,” many large enough that selling them at all — let alone quickly — would move the market against the very trade it was trying to close. A convergence trade only pays off if you can hold it to convergence. A convergence trader forced to sell before the spread closes locks in the loss the model said would disappear.

The losses did not arrive as a single shock. They compounded daily as more counterparties who could see the fund’s distress traded ahead of it: “Day after day we had massive losses, and they didn’t stop.” In August alone Long-Term lost $1.9 billion, 45 percent of its capital, leaving $2.28 billion in equity against $125 billion in assets — a leverage ratio near 55 to 1, before the derivatives book was even counted.

The mechanism is the one Ray Dalio maps from the other side, at the level of an entire economy, in Big Debt Crises: a leveraged position does not fail because the underlying view was wrong. It fails because falling prices force a sale, and the sale itself drives prices lower.

On September 23, 1998, fourteen banks met at the Federal Reserve Bank of New York — not because regulators believed Long-Term’s trades were unsound, but because Long-Term’s failure risked breaking the market it traded in. The fund still held roughly $100 billion in assets, and Wall Street executive Herbert Allison reminded the room it carried $1 trillion in notional derivative exposure on top of that. A disorderly collapse would have forced sixty thousand positions onto the market at once, and no one at the table wanted to learn what that did to every other leveraged book sitting next to Long-Term’s.

The consortium that recapitalized the fund with $3.65 billion was not rescuing a good trade; it was containing a liquidation. Lowenstein is explicit about the motive: “the consortium’s only interest was to get its money back.” The bankers’ own priorities, in order, were “reduce the fund’s risk level, return capital to the new investors, and — last — try to realize a profit. To a man, the bankers would be happy just to get out whole.” No one in the room was betting the trades were right. They were buying time to unwind them without a fire sale.

The popular memory of Long-Term Capital Management is that the trades were sound all along — that the banks who took over the portfolio eventually made a fortune on positions the market had temporarily mispriced under panic. Lowenstein’s own numbers say otherwise.

“Recapitalized with a fresh $3.65 billion, Long-Term continued to plummet, like a parachutist who yanks the rip cord but keeps falling anyway. In its first two weeks, the consortium lost $750 million.” The rescue did not stop the bleeding on contact. It took over a portfolio that kept losing money under new ownership for weeks after the deal closed.

A year is enough time to check a thesis, and the check did not go the way the popular version remembers it. “In the first year after the bailout, the fund earned 10 percent — hardly a dramatic recovery.” Through the losses, the partners kept insisting “that spreads were too wide, that now was the time to invest, that opportunities had never looked better. The models said so!” A year after the bailout, Lowenstein notes, “they had yet to be proven right.”

The fund was wound down rather than ridden back to its old returns, closing by early 2000. Departing staff signed away their right to ever discuss what had happened: “Departing staff members were cajoled into signing termination agreements in which they repeated a pledge never to say a word about the fund.” The silence is part of why the “the trades were right all along” version survived as long as it did — the people who knew the actual numbers were contractually barred from correcting it.

The lesson that outlasts the specific trades: leverage does not only multiply a loss once a position turns against you. Once a leveraged position is large enough, and its holder’s distress is visible enough, disclosure and distress together give every counterparty a reason to sell into it before you can — and that pressure operates whether or not the original analysis was correct. A book being dumped by a forced seller is not, by itself, proof the seller was wrong. It is also not, by itself, free money for whoever buys it. Long-Term’s own recovery numbers are the record of exactly how slowly a forced liquidation actually resolves.

Everything above is Lowenstein on Long-Term Capital Management in 1998. It is here because of a book filed in 2026. Situational Awareness LP grew from $255m to $13.7bn in six quarters, ran a long AI-buildout book against a short semiconductor hedge, and in July 2026 sold the bulk of its public positions after margin calls. The filings, priced forward against our own sessions, are here.

The parallel is not that both managers were wrong. Long-Term’s convergence spreads did eventually converge, and the AI names Situational Awareness held for fifteen months were the ones it was most right about. The parallel is the mechanism in the epilogue above: a position large enough to matter, visible enough to identify, held by someone visibly under pressure. At that point the market can trade against the holder rather than against the thesis, and being right becomes irrelevant to whether you survive.

One difference is worth stating plainly, because it cuts against the analogy. Long-Term’s banks knew its book because they cleared its trades. A 13F is a snapshot filed up to 45 days late, reports long positions only, and never shows leverage — so what the public could read about Situational Awareness in July was far less than what Long-Term’s counterparties could read in 1998. Whatever pressure the fund came under, a quarterly filing is not the channel that delivered it.