Long-Term Capital Management was, by resume, the most intelligent trading firm ever assembled. Founded in 1994 by John Meriwether, Salomon Brothers' legendary bond-arbitrage chief, its partners included Myron Scholes and Robert Merton — who, mid-way through the fund's life, won the Nobel Prize in Economics for the option-pricing theory half of finance runs on.
For three years it looked like the resume was destiny: returns around 40% in 1995 and 1996 with barely a losing month. In 1998 the fund lost $4.6 billion in a few weeks, required a $3.65 billion rescue organised by the Federal Reserve to prevent a chain reaction through the banking system, and was gone shortly after.
The trade: picking up "certainties"
LTCM's core business was convergence arbitrage: find two nearly identical instruments trading at slightly different prices — an old 30-year Treasury versus a newly issued one, say — buy the cheap, short the rich, and wait for the gap to close. The edge per trade was tiny and looked near-riskless, so the returns came from leverage: roughly $4.7 billion of capital controlling over $120 billion of positions — 25 to 1 and more — plus a derivatives book with a notional value over a trillion dollars.
The models said the combined portfolio was safe: the positions were diversified across markets and, historically, their spreads didn't blow out together. The daily-loss estimates were computed to scientific precision from years of data.
August 1998: correlation goes to one
Then Russia defaulted on its domestic debt. Investors everywhere fled anything risky or illiquid and bought the same few safe assets. Every one of LTCM's "uncorrelated" spread trades was, in truth, the same single trade — short panic — and panic was suddenly the only thing anyone owned. Spreads that history said should converge widened violently, everywhere, at once.
Leverage did the rest. Losses ate capital; falling capital triggered collateral calls; meeting them required selling positions into markets where every other levered fund was selling the same things. The firm lost 44% in August alone. By late September its capital was nearly gone, and fourteen banks — herded into a room by the New York Fed, each terrified of what LTCM's fire-sale would do to their own books — bought the carcass to unwind it slowly.
The moral is not "models are useless"
The models were merely incomplete — they knew the statistics of normal weather and were sized as if storms were impossible. The fatal step wasn't the math; it was the decision, encouraged by years of smooth profits, to run storm-irrelevant leverage on fair-weather estimates. Any trader who has computed a "safe" position size from a calm backtest and then met a news gap has lived the same failure at retail scale.
The durable defenses are unglamorous: assume your worst case is worse than your data's worst case; size so that the storm is survivable; and treat rising correlation — everything moving together — as the loudest risk signal there is. In futures terms: your five "different" trades on dollar pairs are one trade the moment the dollar decides to move.
What this story teaches
- Risk models know the past, not the future. Size for the storm your data has never seen, because that's the one that ends accounts.
- Diversification is a fair-weather friend. In stress, correlated positions become one giant position. Count your true exposure by theme, not by ticker.
- Leverage is a clock. Unlevered, being early is inconvenient. At 25:1, being early is identical to being wrong.
- Intelligence doesn't exempt you from arithmetic. Two Nobel laureates could not out-think a margin call. Neither can you or I — so we size as if we can't.