When Genius Failed

Roger Lowenstein · 2000

Thesis

Long-Term Capital Management, a hedge fund run by the most celebrated traders and economists of its day, collapsed in 1998 because its partners trusted mathematical models built on recent history, assumed that extreme market moves were nearly impossible, and borrowed so heavily that a single crisis could destroy them, and its fall nearly took the financial system with it.

The main ideas

Brilliance and credentials are no protection against a flawed model of risk.
LTCM's trades bet that small price gaps between related securities would close. Each trade earned little, so leverage made the returns large.
Its models assumed that price moves follow a normal distribution and that past correlations would hold. In a crisis, both assumptions failed at once.
Leverage turns small errors into ruin. At about 25 to 1, a 4 percent loss on assets wipes out the equity.
In a panic, liquidity disappears just when a leveraged fund needs it most.
A firm big enough and connected enough can threaten the whole system, forcing a rescue that raises its own questions.

The dream team

Roger Lowenstein's account, published in 2000, begins with John Meriwether, who had run Salomon Brothers' bond arbitrage group, one of the most profitable trading operations on Wall Street. In 1994 he founded Long-Term Capital Management with former Salomon traders and recruited two economists whose work underpinned modern finance, Robert Merton and Myron Scholes, who shared the Nobel Prize in economics in 1997, and a former vice chairman of the Federal Reserve, David Mullins.

The firm charged fees of 2 percent of assets and 25 percent of profits, above the industry norm, and investors lined up anyway. Its returns after fees were about 20 percent in 1994, 43 percent in 1995 and 41 percent in 1996.

Lowenstein's theme is hubris. The partners believed their models and their talent let them see risk more clearly than anyone else, and their early success seemed to prove it.

Implication: credentials and a strong record are not evidence that a strategy is safe. Ask what conditions the record was earned in and whether they could change.

Convergence and leverage

LTCM's core strategy was convergence trading: finding pairs of nearly identical securities whose prices had drifted apart, and betting that the gap would close. A classic example was the spread between the most recently issued Treasury bond, which trades at a premium because it is more liquid, and a slightly older one with almost the same terms. LTCM would sell the expensive one and buy the cheap one, and profit as the prices converged.

Each such trade earned very little. To make large returns, LTCM borrowed heavily. At the end of 1997 it held about $129 billion of assets on about $4.7 billion of equity, leverage of about 28 to 1, and its derivative positions had a notional value of more than $1 trillion. Late in 1997, the fund returned about $2.7 billion to outside investors without reducing its positions, which raised its leverage further.

As the obvious opportunities were competed away, the fund moved into less familiar trades, including selling long-term options on stock indexes, a bet that markets would stay calm.

Implication: when a strategy depends on leverage to make small edges profitable, its survival depends on the edges never turning sharply against it.

The models

The partners' risk models estimated how much the fund could lose, based on how its positions had moved in the past. They treated large price moves as extraordinarily rare, as they would be if prices followed a normal distribution, and assumed that the fund's many positions, in different markets and countries, were largely independent of one another.

Lowenstein argues that both assumptions were wrong in the conditions that mattered. Market prices have fat tails: extreme moves happen far more often than a normal distribution predicts. And in a panic, correlations jump, because frightened investors sell everything risky at once. Positions that looked diversified turned out to be versions of the same bet, on calm markets and narrowing spreads.

The mechanism is that a model built on recent, calm history measures the risk of normal times. The losses that destroy a leveraged fund come from abnormal times, which the model treats as nearly impossible.

Implication: when a risk model says a large loss is a once-in-many-lifetimes event, ask whether the model's assumptions would survive a crisis.

August 1998

On August 17, 1998, Russia devalued the ruble and stopped paying its debts. Investors around the world fled to the safest, most liquid assets. Instead of converging, the spreads LTCM had bet on widened, in market after market at once.

The fund lost 44 percent of its value in August alone, nearly $2 billion. Meriwether wrote to investors on September 2 describing the results as very disappointing and seeking new capital. By mid-September its equity had fallen below $1 billion while it still carried positions of well over $100 billion.

The mechanism Lowenstein traces is liquidity. LTCM's positions were enormous, and other traders knew it. To reduce its exposure, the fund had to sell into markets where buyers had disappeared, and the attempt to sell would itself push prices further against it.

Implication: the ability to sell is not a constant. Size and leverage that are manageable in calm markets can become impossible to unwind in a panic.

The rescue

By September, LTCM's failure threatened the banks that had lent to it and traded with it, and possibly markets as a whole, since a forced liquidation of its positions could move prices violently. Officials of the Federal Reserve Bank of New York, led by its president, William McDonough, and Peter Fisher, examined the fund's books and convened its major creditors.

A last-minute offer from Warren Buffett's Berkshire Hathaway, together with AIG and Goldman Sachs, to buy the fund's portfolio for $250 million failed. On September 23, 1998, fourteen banks and brokerages agreed to inject $3.625 billion in exchange for 90 percent of the fund. Bear Stearns, which cleared LTCM's trades, declined to join. No public money was used; the Fed chairman, Alan Greenspan, told Congress the aim was to avoid a disorderly fire sale. The partners' own stake was nearly wiped out. The consortium recovered its money, and the fund was wound down by early 2000.

Implication: when an institution becomes large and interconnected enough, its failure becomes everyone's problem, which is a reason to limit leverage before it reaches that point.

Limits

It treats the collapse mainly as a story of flawed models. The finance scholar Philippe Jorion argued that the fund also suffered from concentration, using the same estimates both to choose trades and to measure their risk, and the sociologist Donald MacKenzie argued that imitators holding the same positions created a crowded superportfolio whose forced selling turned losses into a collapse.
It is an unauthorized account. Lowenstein drew on interviews and internal documents, but the story is told largely from outside the firm.
The rescue's lessons remain disputed. Critics argued that the Fed's role encouraged large firms to expect help in a crisis, while defenders noted that only private money was used and the partners lost most of their stake.

The Black Swan + Antifragile + Skin in the Game + The Bed of Procrustes — the theory of fat tails and fragility that LTCM's collapse illustrates.
Devil Take the Hindmost — the history of speculation, which ends with LTCM.
The Big Short — the next great crisis of leverage and flawed models, a decade later.