LTCM: How Two Nobel Laureates Nearly Broke the System
A fund run by Nobel laureates hit 250:1 leverage and lost $4.6bn in five weeks. The risk model that missed it still runs — inside your broker's margin system.
LTCM Explained: The Nobel Price Hedge Fund Collapse
Prefer to watch? This article is the written companion to the video above.
Two Nobel Prize winners. A former Vice Chairman of the Federal Reserve. The most feared bond traders Wall Street had ever produced.
They built a fund holding roughly $125 billion in assets and over $1.25 trillion in derivatives. In the autumn of 1998 it lost about $4.6 billion in five weeks, and the Federal Reserve Bank of New York put the heads of Wall Street’s largest firms in a room and told them to fix it before markets opened.
Here is the part that should bother you. The risk formula that destroyed them is still running — not in the place most retellings claim, which we will get to, but somewhere closer to you.
The team that shouldn’t have existed
- John Meriwether, the legendary bond trader from Salomon Brothers, assembles a group that reads like a screenwriter’s idea of finance. Myron Scholes and Robert Merton, who in October 1997 would win the Nobel Prize in Economics for the options-pricing formula that supposedly made risk calculable. David Mullins, former Vice Chairman of the Federal Reserve. And the sharpest arbitrage desk Salomon ever assembled.
They called it Long-Term Capital Management.
Wall Street CEOs invested their personal money. Banks queued up to lend on terms they extended to nobody else, in many cases charging zero initial margin, a privilege normally reserved for sovereign governments. The returns justified it: 43% in 1995, 41% in 1996.
One year after the Nobel, the formula destroyed them.
The strategy, in two cans of Coca-Cola
Imagine two identical cans of Coke. Same product, same company, same taste. One costs $1.00 at the corner shop. The other costs $1.02 at the petrol station next door.
You buy the cheap one. You sell the dear one. You pocket two cents.
That is convergence arbitrage, and it is what LTCM did, with government bonds rather than soft drinks. Buy the marginally cheaper bond, short the marginally dearer one, wait for the gap to close as it almost always does. The logic is sound. It is, in fact, close to riskless, in the sense that the two bonds really are nearly identical.
But two cents is nothing. So to make real money, you borrow. Enormously.
LTCM ran, at times, more than 25:1 on its balance sheet — roughly $125 billion of assets — plus over $1.25 trillion of off-balance-sheet derivatives. They bought $125 billion of Coca-Cola. On credit.
The spark, and then the fire
17 August 1998. Russia defaults.
LTCM’s direct exposure to Russia was survivable, and that was never the point. Russia was the spark; what followed was the fire. Global panic, every investor on the planet running for the same exit at the same moment, dumping anything that smelled of risk and buying anything that smelled of safety.
In the Coca-Cola analogy: suddenly nobody will touch the corner-shop can. Everyone wants the petrol-station can, and only that one.
The price gap that was supposed to narrow exploded instead. Their positions were built to profit from convergence, and the market delivered divergence — everywhere, at once.
The number the model said was impossible
The President’s Working Group later recorded the arithmetic. LTCM held $4.1 billion of capital on 31 July 1998 and lost a further $1.8 billion during August alone, taking the equity loss for the year past 50%.
There is a famous claim attached to this, and it is worth handling carefully because it gets repeated in a slightly wrong form. In Roger Lowenstein’s account of the collapse, LTCM’s models implied that it would take a ten-sigma event for the firm to lose all of its capital within one year: a statistical freak with odds of roughly one in ten to the twenty-fourth power.
Put less politely, the model said this could not happen once in the lifetime of the universe. Not once in 13.8 billion years.
It happened in the fund’s fourth year of operation.
That figure comes from a journalist’s reconstruction rather than an official filing, and I would rather say so than let it sit there wearing more authority than it has. The point survives either way, because the official record contains something colder: a fund that lost more than half its equity in eight months while its own risk system reported that it was fine.
And this is where leverage becomes a death sentence
Watch what happens next, because it is the mechanism that kills every leveraged institution and almost nobody explains it.
The assets barely moved. Still roughly $125 billion. The equity underneath collapsed.
Same positions, same balance sheet
Assets stayed at roughly $125 billion throughout. Only the green sliver — the money that was actually theirs — changed.
Nothing was added to the positions. Nobody increased the bet. The leverage went from 30:1 to roughly 250:1 purely because the denominator shrank — and the move needed to erase what was left fell from 3.3% to 0.4% along with it.
Leverage is a ratio. When the denominator evaporates and the numerator does not, the ratio does not drift. It detonates.
By late September, with a few hundred million dollars of capital remaining, LTCM was reportedly running at roughly 250:1 — a level at which a 0.4% move against you erases everything you have. Not a crash. Not a catastrophe. Four tenths of one percent.
The room, and who refused to pay
23 September 1998. The Federal Reserve Bank of New York convenes the heads of Wall Street’s largest institutions. The following day, fourteen firms put in $3.6 billion.
Bear Stearns, LTCM’s own clearing broker, refused to contribute a cent. Ten years later Bear Stearns would need rescuing itself, which is either irony or arithmetic depending on how much you believe about the way these institutions were run.
And here is the detail that gets misreported constantly. Not one dollar of taxpayer money was used. The banks paid, with their own capital, and not out of altruism: a disorderly liquidation of $125 billion of positions into a panicking market would have destroyed their balance sheets too. The Fed organised the room and applied the pressure. That is all it did, and Alan Greenspan defended exactly that framing before Congress a week later.
Now the formula, because it is still in your life
Forget the fund. This is the part that matters.
Value at Risk is a weather app
Imagine a weather app that only looks at the last fortnight. Fourteen sunny days, so it tells you there is a 95% chance of sunshine tomorrow. It does not tell you what happens in the other 5%, and it does not mention that the 5% might be a hurricane.
Here is the genuinely dangerous part. If everybody uses the same app, and the app says sunny, then everybody leaves the umbrella at home. When the storm arrives, everyone gets soaked simultaneously.
That app is called Value at Risk. It takes one or two years of market data, runs it through a statistical model, and produces a single number: the most you should lose on a normal day. It breaks in three places, and the three of them compound.
Break one: markets are not a bell curve
VaR assumes a normal distribution, in which small moves are common and extreme moves are nearly impossible. Markets do not read statistics textbooks. Extreme moves happen far more often than the bell curve permits, which is how a fund run by the people who invented modern options pricing managed to experience an event their own model had priced at somewhere between improbable and impossible.
The maths was not wrong. The assumption underneath the maths was wrong.
Break two: it makes crises worse
This is the one that should genuinely alarm you, and it is the reason the model is not merely inaccurate but actively harmful.
The loop the model creates
Each step is rational on its own. Together they form a circle with no exit.
Calm markets
Volatility is low, so the model reports low risk. There is room on the balance sheet, and the model says so in writing.
Everyone borrows more
Positions get bigger because the measured risk is small. Nobody is being reckless. Each firm is following its own risk system.
A shock lands
Volatility jumps overnight. The same model that said all clear yesterday now reports far more risk on positions that have not changed.
Everyone sells at once
Every institution runs the same kind of model, so they all get the same instruction in the same hour. Prices fall further, volatility rises again — and the loop starts over, louder.
The uncomfortable part is step 4. The model is not failing there — it is working exactly as specified. The damage comes from everyone owning the same instrument and reading it at the same moment.
The doom loop is not a failure of the model. It is the model working, in every institution, at the same time. And regulators know it.
Break three: it assumes someone will buy
VaR assumes you can liquidate at today’s price. That assumption holds right up until the day it matters, which is the day it stops holding. In a panic there is no bid at $125 billion of size, and the number on the screen is quietly describing a market that has temporarily ceased to exist.
What the regulators actually did about it
Most retellings of this story end with “and the same formula is still used everywhere”, which is the satisfying ending and is no longer quite true. The correction is worth making, because the real situation is more interesting than the myth.
The Basel Committee moved bank trading books off value-at-risk. The revised market risk framework replaces VaR in the internal models approach with expected shortfall, a measure that asks what the average loss looks like beyond the cut-off rather than stopping at it. The standard was issued in January 2016, with supervisory implementation from January 2019. The reason given is precisely the first break above: VaR says nothing about the shape of the tail.
So the largest banks’ trading books are, in principle, measured with a better instrument now.
Your margin account is not. Neither, in most cases, is the clearinghouse standing behind your broker, or the volatility-targeting strategy inside a risk-managed fund. VaR-type models, with all three breaks intact, are still what sets the number that can force you to sell. The formula did not disappear. It moved downstream, to the part of the system with the least capital and the least ability to argue.
Two things people get wrong about LTCM
“It was a black swan. Nobody could have seen it coming.” No. When you run 250:1 leverage on correlated positions and never stress-test for a liquidity shock, a wipeout is not an anomaly. It is arithmetic waiting for a date.
“They had Nobel Prize winners. The models must have been sound.” The models were computed correctly. They were built on the assumption that markets behave like physics, and markets are not physics. Markets are people running for the same exit at the same time.
One further detail from the official post-mortem ought to be taught in every risk course. LTCM had 75 lenders and over 50 derivatives counterparties, and not one of them knew the fund’s total exposure — there were no covenants limiting aggregate leverage, and no mechanism anywhere in the system that could have added the pieces together even if somebody had thought to ask for them.
They switched off their own alarm system, because they believed they were lending to gods.
What this means for your portfolio
CHECK — if you use margin, you are inside this machine
Open your broker app and find maintenance margin or portfolio margin. Those numbers are produced by VaR-type models. In a volatility spike they rise automatically, and if you fall below the threshold your positions can be liquidated at the worst possible moment, at the worst possible price, without anyone calling you.
The same logic sits inside risk-managed robo-advisers. A strategy that cuts exposure when a volatility threshold is breached is, by construction, selling into falling markets. That is not a scandal; it is what it promised to do. But you should know that it is what it does.
AVOID — two things
Daily-reset leveraged ETFs, unless you can explain volatility drag to another person without looking it up. They rebalance every day, which makes them procyclical by design and can grind you down in a choppy market even when your directional call was right.
And avoid believing your diversified portfolio is diversified. LTCM was spread across eight countries, which is textbook global diversification on paper. In reality every position was the same bet: short liquidity, long credit risk. When the panic came, correlations went to one and the diversification evaporated at exactly the moment it was needed.
Diversification across countries is not diversification across risk factors.
BUILD — a liquidity buffer, and be honest about why
Cash, or short-dated government bonds. Unleveraged. Boring. Slightly embarrassing to hold in a bull market.
It is not dead money. It is a call option on the next crisis. When the leveraged investor gets a margin call and is forced to sell at a fire-sale price, somebody is on the other side of that trade at a price they will talk about for years. Which side you are on gets decided long before the day itself, in a boring account you were slightly embarrassed to be holding.
And if you think this is history
Right now, hedge funds are running the same convergence trade — cash Treasury bonds against Treasury futures — funded overnight in the repo market, where the funding can vanish in a day.
In a FEDS Note published on 22 June 2026, Federal Reserve staff sized the basis trade at roughly $830 billion of hedge funds’ Treasury long positioning, about double its previous peak in early 2020, with hedge fund net repo borrowing near $1.8 trillion at the end of 2025.
Same structure. Same funding fragility. In the market that prices every other asset on earth.
Two Nobel laureates, the smartest traders alive, and a formula that said the disaster was impossible. The formula was right, mathematically. But markets are not mathematics. They are human beings running for the same exit at the same time.
The banks have since been made to measure this differently. The margin system on your phone has not.
Primary sources
- 01Hedge Funds, Leverage, and the Lessons of Long-Term Capital Management (April 1999) — President's Working Group on Financial Markets (US Treasury, Fed, SEC, CFTC)
- 02Near Failure of Long-Term Capital Management — Federal Reserve History
- 03Testimony of Chairman Alan Greenspan on the private-sector refinancing of Long-Term Capital Management (1 October 1998) — Board of Governors of the Federal Reserve System
- 04Decomposing Hedge Funds' U.S. Treasury Exposures (22 June 2026) — Board of Governors of the Federal Reserve System (FEDS Notes)
- 05Revised market risk framework, executive summary — records the shift from value-at-risk in the internal models approach to an expected shortfall measure of risk under stress; standard issued January 2016, supervisory implementation from January 2019 — Bank for International Settlements, Financial Stability Institute (Basel Committee on Banking Supervision framework)
- 06When Genius Failed: The Rise and Fall of Long-Term Capital Management — source of the widely repeated claim that LTCM's models implied it would take a ten-sigma event for the firm to lose all of its capital within one year — Roger Lowenstein (2000); review by the Federal Reserve Bank of Minneapolis
Questions people actually ask
What was Long-Term Capital Management?
A hedge fund founded in 1993 by John Meriwether, the most celebrated bond trader of his era, staffed with a team that would be implausible in fiction: Myron Scholes and Robert Merton, who won the 1997 Nobel Prize in Economics for the options-pricing formula that bears their names, a former Vice Chairman of the Federal Reserve, and the sharpest arbitrage traders Salomon Brothers had produced. It returned 43% in 1995 and 41% in 1996. In the autumn of 1998 it nearly took the global financial system down with it.
What did LTCM actually do?
Convergence arbitrage. Two nearly identical government bonds sometimes trade at slightly different prices; LTCM bought the cheaper one, shorted the dearer one, and waited for the gap to close. The logic was sound and the profit per trade was minuscule — a few basis points. To turn a few basis points into a real return, you have to do it with an enormous amount of borrowed money. That is the whole story: the strategy was not reckless, the leverage applied to it was.
How much leverage did LTCM have?
On its balance sheet, at times more than 25:1 — roughly $125bn of assets against a few billion of capital — plus over $1.25 trillion in off-balance-sheet derivatives. But the number that matters is what happened as the equity evaporated: the assets barely moved while the capital underneath collapsed, so the ratio exploded. By late September 1998 the fund was reportedly running at around 250:1. At 250:1, a 0.4% move against you erases everything.
What triggered the collapse?
Russia defaulted on its domestic debt on 17 August 1998. LTCM's direct Russian losses were manageable — the default was the spark, not the fire. What followed was a global flight to safety: every investor on earth ran for the same exit at the same moment, dumping anything risky and buying anything safe. The price gaps LTCM was betting would narrow did the opposite. They exploded.
How much did LTCM lose, and how fast?
The President's Working Group records that the fund held $4.1bn of capital on 31 July 1998 and lost a further $1.8bn during August alone, taking the equity loss for the year past 50%. By late September the capital was down to a few hundred million. In total, roughly $4.6bn evaporated in about five weeks — a fund that had been one of the most admired institutions in finance.
Did taxpayers bail out LTCM?
No, and this matters. On 23 September 1998 the Federal Reserve Bank of New York convened the heads of the major Wall Street firms and applied pressure. Fourteen of them put in $3.6 billion of their own capital — to protect their own balance sheets from the fallout of a disorderly liquidation. Bear Stearns, LTCM's own clearing broker, refused to contribute. Not one dollar of public money was used. The Fed organised the room; the banks paid.
What is Value at Risk (VaR)?
A single number that answers the question: on a normal day, what is the most I should lose? A model takes the last one or two years of market data, feeds it through a statistical distribution, and outputs, say, '95% confident the daily loss will not exceed X'. Nearly every major bank, broker and clearinghouse uses some version of it. It is the weather app of finance — and like a weather app that has only ever seen two weeks of sunshine, it has nothing useful to say about the hurricane.
Why does Value at Risk fail?
Three ways, and they compound. Fat tails: VaR assumes a bell curve, but extreme moves happen far more often than the bell curve allows — LTCM's own model classified its August 1998 loss as an event that should not occur once in the lifetime of the universe. Procyclicality: in calm markets VaR reports low risk, so institutions lever up; when a shock arrives it demands selling from everyone simultaneously. Liquidity blindness: it assumes you can sell instantly at today's price, which is precisely what stops being true in the moment you need it to be.
What is the doom loop?
The self-reinforcing spiral that VaR creates. A shock hits. Volatility spikes. Every institution's VaR model — and they are all built the same way — simultaneously reports a limit breach and demands position cuts. They all sell into the same market at the same moment. Prices fall further. Volatility rises again. VaR rises again. More forced selling. The model designed to contain the crisis is the thing propagating it.
Does this still affect me today?
Directly, if you use margin. Your broker's maintenance and portfolio margin requirements are set by VaR-type models. When volatility spikes, those requirements rise automatically — and if you fall below the threshold, your positions can be liquidated at the worst possible price, without a phone call. The same logic sits inside risk-managed robo-advisers, which by construction sell into falling markets. You do not need to run a hedge fund to be inside this machine.
Could an LTCM happen again?
It is happening, at a larger scale, in the most important market in the world. Hedge funds run a convergence trade between US Treasury cash bonds and Treasury futures, funded overnight in the repo market. In a FEDS Note published on 22 June 2026, Federal Reserve staff put the basis trade at roughly $830 billion of hedge funds' Treasury long positioning — about double its previous peak in early 2020 — with hedge fund net repo borrowing near $1.8 trillion at end-2025. The structure is LTCM's. The scale is not.