Long-Term Capital Management was supposed to be the fund that could not lose. It was founded in 1994 by John Meriwether, who had built the legendary bond arbitrage desk at Salomon Brothers, and its partners included Robert Merton and Myron Scholes — who won the Nobel Prize in Economics in 1997 for the model that underpins modern option pricing. Below them sat a floor of PhDs.
In four months during 1998 the fund lost $4.6 billion, about 92% of its capital. Its positions were so large and so entangled with the banks financing them that an uncontrolled liquidation threatened the solvency of Wall Street itself. The Federal Reserve Bank of New York had to convene fourteen banks in a room and organize a $3.6 billion rescue.
The trade the fund was built on
LTCM did convergence arbitrage. The models identified pairs of instruments whose prices had drifted apart by more than history suggested they should, bought the cheap one, sold the expensive one, and waited for the relationship to normalize.
The classic example is the on-the-run versus off-the-run Treasury trade. A newly issued 30-year Treasury is the most actively traded bond in the world and carries a small liquidity premium. A 29-and-a-half-year bond — the previous issue, now slightly stale — is nearly identical in cash flow but yields marginally more. LTCM would buy the older bond and short the newer one. Within months, as the next auction made the current bond stale in turn, the gap closed on its own.
Each trade produced very little: half a percent, perhaps two. The business model was to apply enormous leverage to a very large number of these small, statistically reliable convergences. By mid-1997 the fund had $4.7 billion of capital supporting positions with a notional value above $100 billion, and derivative exposures far larger still.
The results looked like a proof. Roughly 21% in the partial first year, 43% in 1995, 41% in 1996, 17% in 1997, with volatility resembling a bond portfolio rather than a hedge fund. Investors queued to get in, and at the end of 1997 LTCM returned capital because it had more money than opportunities — which, in hindsight, meant the same positions were now supported by less equity.
17 August 1998
Russia defaulted on its domestic debt and declared a moratorium on foreign obligations. The models rated this as effectively impossible: a G8 member with nuclear weapons and a central bank that could print its own currency does not default on debt denominated in that currency.
It did, and what followed was not the loss on Russian exposure — LTCM's direct exposure there was modest. What followed was a global flight to quality. Every investor in the world wanted US Treasuries and wanted out of everything else, simultaneously.
That is the precise inverse of a convergence portfolio. LTCM was, in essence, short liquidity premia everywhere: long the cheap, illiquid, complicated instrument and short the expensive, liquid, simple one, dozens of times over. When the world bid for simplicity and sold complexity, every position moved the wrong way at once.
The spreads did not converge. They widened, then widened further, and the fund's own attempts to reduce risk widened them again because everyone knew what it held.
What actually broke
Swap spreads. LTCM was positioned for the gap between swap rates and Treasury yields to narrow. It had traded in a range of roughly 30 to 70 basis points for years. In September 1998 it reached about 130. Hundreds of millions of dollars evaporated on a spread that had historically never moved that far.
Short equity volatility. The fund had sold index options in such size that dealers called it the central bank of volatility. The bet was that implied volatility near 20% was too high against realized volatility near 15%. During the crisis implied volatility went to 35% and beyond, and a short volatility position loses convexly — the loss accelerates as the move grows.
Convertible arbitrage. Long convertible bonds, short the underlying equity. Convertibles are less liquid, so panicked investors dumped them faster and further than the shares hedged against them. The hedge lost money on both legs.
Emerging markets. Convergence positions across Latin America and Asia, all of which moved together as the Russian shock propagated.
Leverage did the rest. At 25 to 1, a 4% loss on the book erases the capital. The book lost far more than 4%, and as capital fell the ratio rose — the fund became more levered every day without placing a single new trade. By late September, effective leverage was in the hundreds to one.
Losses ran to $1.8 billion by the end of August and $4.6 billion by the end of September.
Why the Fed got involved
An ordinary hedge fund failing is not a public problem. LTCM was not an ordinary hedge fund.
Its counterparties included essentially every major dealer on Wall Street and in Europe, on positions with a notional value above a trillion dollars once derivatives were counted. Nobody, including the banks themselves, had a complete picture of the aggregate exposure, because each dealer saw only its own slice and LTCM had deliberately spread its financing to prevent any single lender from understanding the whole.
If the fund had been forced into a fire sale, it would have been selling exactly the illiquid instruments that everyone else already held and was already marking down. Prices would have fallen further, generating mark-to-market losses at institutions that had nothing to do with LTCM. Several of them were themselves fragile.
On 23 September 1998 William McDonough, president of the New York Fed, brought the heads of the major banks together. After a day of negotiation, fourteen firms put up about $3.6 billion for 90% of the fund and installed a committee to unwind the positions in an orderly way over the following months.
It is worth being precise about what this was. No taxpayer money was used; the capital came from private banks, several of which were LTCM's own creditors and were choosing between a controlled loss and an uncontrolled one. But the Fed convened the meeting and made clear it wanted a solution, which is a form of intervention even when no public funds move.
Whether that was correct is still argued. The case against is moral hazard: the market learned that a sufficiently interconnected fund would be assisted. The case for is that the alternative was a liquidity spiral in the middle of an emerging market crisis. The same argument returned, with higher stakes and a different answer, when Lehman Brothers failed ten years later.
The four lessons
Models describe the past, and the past is a small sample. LTCM's risk estimates were built on historical data that contained no sovereign default by a major power and no global liquidity freeze. Financial returns have fat tails: extreme moves happen far more often than a normal distribution allows. A model that assigns a probability of once in ten thousand years to an event that has occurred three times this century is not conservative, it is wrong.
Leverage converts a mistake into a failure. The positions LTCM held were not stupid. Almost all of them eventually converged exactly as predicted — after the fund had been liquidated. Being right was irrelevant, because 25 to 1 leverage meant there was no capital left to hold the position with. Leverage does not increase your edge. It shortens the time you are allowed to be wrong.
Liquidity is an assumption, not a property. Every risk model assumed positions could be exited at quoted prices. In September 1998, for large blocks of the instruments LTCM held, there was no bid at any sensible price. Worse, the market knew what the fund owned and traded against it — a large position in a stressed market is not an asset, it is a public advertisement.
Diversification requires independent risks, not different tickers. LTCM held dozens of trades across countries, asset classes and instruments, and considered itself diversified. Every one of those trades was a version of the same bet: that spreads revert to normal. When that single assumption failed, the diversification counted for nothing. Correlation between strategies is what matters, and it is almost always higher under stress than in the sample used to estimate it.
Afterwards
Meriwether launched JWM Associates in 1999 and raised over $3 billion, running a similar strategy at lower leverage. It lost 44% in the 2008 crisis and closed in 2009. A third venture followed and did not last.
Merton and Scholes returned largely to academic and advisory work. Both maintained, reasonably, that the models were not the problem — the leverage applied to them was. Critics, Nassim Taleb most loudly, argue the models were the problem, because a model that misprices tail risk invites exactly that leverage.
The regulatory legacy was slower. LTCM demonstrated that a non-bank could create systemic risk, but meaningful disclosure requirements for large leveraged funds arrived only after 2008, and how much they capture remains debated.
Why it matters to a retail trader
The instinct is to file LTCM under institutional history. That is a mistake, because the failure modes scale down without changing shape.
A retail account running high leverage on a strategy that works most of the time is the same structure. So is a portfolio of "different" trades that are all long the same underlying condition — carry, momentum, mean reversion in a range. So is a stop placed in a market where the liquidity to fill it will not be there during the move that triggers it.
The line usually attributed to Keynes covers it: the market can stay irrational longer than you can stay solvent. LTCM is the strongest proof of it ever assembled. The fund's analysis was largely correct. The spreads did converge. The people were among the most capable in the industry, and they had the arithmetic on their side.
They were simply unable to survive the interval between being right and being paid, and there is no model that fixes that. Only position size does.
