In September 1998, fourteen of the world's largest banks were summoned to the Federal Reserve Bank of New York and asked, in essence, to save a hedge fund almost nobody outside Wall Street had heard of. Not because they liked it — because a fund with roughly \$4.7 billion of capital had built derivatives positions with a notional value estimated above \$1 trillion, scattered across markets on every continent, and its sudden failure risked freezing the entire global financial system at once. The fund was Long-Term Capital Management (LTCM), and its near-collapse is the definitive case study in what extreme leverage does to a strategy that looks, on paper, almost riskless.

This lesson follows on from Barings, but the failure mode is completely different. Barings was one unsupervised trader hiding losses — an operational-control failure. LTCM was run by some of the most sophisticated, well-supervised, mathematically literate people in finance, using strategies that were, in a narrow sense, correct. It failed anyway — from leverage, correlation and liquidity, the very things a risk manager is paid to worry about.

The all-star fund

LTCM was founded in 1994 by John Meriwether, the legendary former head of bond arbitrage at Salomon Brothers. He recruited a roster that read like a finance faculty: senior Salomon traders, a former vice-chairman of the Federal Reserve, and two academics who would jointly win the 1997 Nobel Memorial Prize in Economic Sciences for their work on option pricing — the same Black–Scholes framework this whole course builds on. The fund's pedigree was, by design, unimpeachable: investors were buying access to the smartest quantitative minds on Wall Street.

LTCM's core strategy was convergence (relative-value) arbitrage. Find two securities — or baskets of securities — whose payoffs are nearly identical, whose prices have drifted slightly apart for reasons of liquidity or temporary supply-and-demand rather than fundamentals, and bet that the gap will close. A classic example: a newly issued ("on-the-run") US Treasury bond and an almost-identical, slightly older ("off-the-run") bond of the same maturity. The on-the-run bond is more liquid and trades at a small premium; the off-the-run bond is functionally the same cash flow, cheaper. Short the rich one, buy the cheap one, and wait for the yields to converge — which, historically, they reliably did.

Why leverage was the whole strategy

The trouble with a genuine convergence trade is that the gap it exploits is tiny — a few basis points of yield, a sliver of a percent. Traded at ordinary size, such a trade would barely cover transaction costs. LTCM's answer was leverage: borrow heavily against a thin sliver of equity so that a small percentage move on a huge position becomes a large percentage move on the (much smaller) capital base. At its peak, LTCM's balance-sheet assets were reported near \$130 billion against roughly \$4.7 billion of equity — a leverage ratio near 25-to-1 before counting the derivatives book at all, well above a typical commercial bank's leverage at the time.

Crucially, LTCM did not run one convergence trade — it ran dozens, spread across government bonds, mortgage securities, sovereign debt, interest-rate swaps and equity volatility, in the US, Europe and beyond. The fund's own risk models treated this as diversification: dozens of small, individually low-risk bets, whose combined risk should be far lower than any one trade's risk multiplied by the total leverage. That assumption — that the trades were independent of one another — is exactly what broke in 1998.

Russia defaults, and everything converges the wrong way

In August 1998, Russia devalued the ruble and defaulted on its ruble-denominated government debt. The direct exposure most of LTCM's counterparties had to Russia was modest. What followed was not really about Russia at all — it was a global flight to quality: spooked investors everywhere sold anything remotely illiquid or "risky" and piled into the safest, most liquid instruments they could find, above all the most recently issued US Treasury bonds.

That single behavioural shift, repeated by thousands of investors simultaneously, pushed every one of LTCM's spreads in the same direction at once — the wrong direction. The on-the-run/off-the-run spread widened instead of narrowing, because everyone wanted the liquid bond and nobody wanted the illiquid one. The same happened in mortgage spreads, swap spreads, emerging-market spreads and equity volatility — dozens of trades LTCM's models had scored as independent bets moved together, because they all shared one hidden common factor: a sudden, universal demand for liquidity and safety. The "diversified" portfolio turned out to be a single enormous bet on calm markets, taken dozens of times over.

Because the book was leveraged roughly 25-to-1, a loss of only a few percent on total assets was a catastrophic loss of capital. As losses mounted, LTCM could not simply hold on and wait for convergence — margin calls forced it to start unwinding positions into markets with no buyers, at exactly the moment prices were most unfavourable, which pushed the spreads even wider and deepened the losses further: a self-reinforcing downward spiral. Between May and September 1998, LTCM's equity fell from about \$4.7 billion to under \$1 billion, while its enormous, largely un-unwindable derivatives book still threatened heavy losses for every bank on the other side of those trades.

The Fed-brokered rescue, and the model that had never seen 1998

The New York Fed did not put public money into LTCM. Its concern was systemic: LTCM's counterparties were essentially every major bank on Wall Street, and a disorderly bankruptcy — with creditors racing to dump LTCM's positions onto the market at once — threatened to spread the crisis well beyond the fund itself. So the Fed convened fourteen banks and brokered a private recapitalization: the consortium injected roughly \$3.6 billion in exchange for about 90\% of the fund, in return for orderly control over the unwind. Given time and calmer markets, many of LTCM's individual trades did eventually converge close to where the models had predicted — but by then the original investors had been almost wiped out, and the fund was closed down over the following two years.

The deeper failure was model risk. LTCM's risk management, like most of the industry's at the time, leaned on historical volatilities and correlations — essentially the same variance–covariance style statistics used to estimate Value at Risk. Those statistics were built from a history that had simply never contained a global, simultaneous, multi-market convergence failure of the kind August 1998 produced — so the model didn't underestimate the risk out of carelessness; it had no data at all describing a scenario like the one that actually happened. A model is only ever as good as the scenarios baked into the history it was calibrated on, and 1998 lay well outside that history.