The hedge fund with two Nobel laureates blew up in four months

Long-Term Capital Management opened for business in February 1994 with the most impressive roster any hedge fund has ever assembled. John Meriwether, who had run arbitrage at Salomon Brothers, was the managing partner. Myron Scholes and Robert Merton, who would share the 1997 Nobel Prize in Economics for their options-pricing work, sat on the investment committee. The firm had hired most of the best quantitative traders out of Salomon. Starting capital was 1.25 billion dollars, the largest initial raise in hedge fund history at the time.

For four years the numbers justified the pedigree. After fees, LTCM returned 21 percent in 1994, 43 percent in 1995, 41 percent in 1996, and 17 percent in 1997. The strategy was called relative-value arbitrage. LTCM would find two bonds or securities whose prices had drifted apart in ways its models said should not last, short the expensive one, buy the cheap one, and wait for the gap to close. Each trade made a fraction of a percent. The profits came from running thousands of trades at enormous scale.

The leverage problem

To make relative-value arbitrage profitable at scale, you need leverage. At the peak, LTCM had about 5 billion dollars of its own capital and total positions over 100 billion, roughly 25 to 1. Off-balance-sheet derivatives added another layer on top, bringing total notional exposure closer to 1.25 trillion. Every major Wall Street bank had lent to LTCM, in part because the fund's returns were so reliable that its counterparties competed to finance it.

The model assumed something like normal market behavior. Relative-value trades had long historical track records. Over long enough windows, spreads converged. A spread that widened unexpectedly would, in the model, narrow again within weeks or months. LTCM sized its positions based on the volatility of those spreads. The bigger the divergence looked, the more capital it committed to the convergence trade.

August 1998

On August 17, 1998, the Russian government defaulted on its ruble-denominated debt and devalued the ruble. The default was a specific event in one country, but it triggered a global flight to safety. Investors everywhere sold anything that looked risky and bought US Treasuries and German Bunds. Every single relative-value spread LTCM had been betting on widened at once.

This was the scenario the model had said was vanishingly unlikely. Different markets, different countries, different asset classes, the model treated their spreads as largely independent. In the panic after the Russian default, they all moved together. In the language Charles Kindleberger had used to describe every historic crash, correlations went to one.

By early September, LTCM had lost 1.85 billion dollars, about 45 percent of capital. Its banks started asking for more collateral. Pulling collateral out of positions meant selling into weakness, which widened spreads further. By September 21, the fund had lost 4.6 billion dollars, about 90 percent of its capital, in four months.

The rescue that was not a bailout

William McDonough, president of the New York Federal Reserve, believed an LTCM collapse would force enough forced selling to take one or more Wall Street banks with it. He invited the CEOs of fourteen major banks to the New York Fed on September 23, 1998. Over the course of a day they agreed to a 3.625 billion dollar recapitalization of LTCM in exchange for 90 percent of its equity. No Fed money was used directly. The Fed provided the room and the convening authority.

The crisis passed within weeks. The rescued LTCM positions were wound down through 1999 without major further losses, and the consortium recovered most of its capital. The banks had protected themselves against the systemic damage they might have caused by lending so aggressively to a single fund.

What it is usually misremembered as

LTCM is often remembered as a case of academic arrogance, two Nobel laureates who did not understand that markets are not like coin flips. This version is mostly wrong. Scholes and Merton were not running the fund day to day, and the LTCM risk model accounted for fat-tailed distributions. What it did not account for was the second-order effect of its own scale. When a single fund is large enough that its forced selling moves global markets, the statistical independence of its trades stops being a property of history and becomes a claim about the future.

This is the lesson that propagated forward into every subsequent crisis. Archegos in 2021. The UK gilt crisis in September 2022. The collapse of FTX in November 2022. In each case, the same pattern: a large, leveraged position whose unwinding is correlated with the conditions that forced it to unwind. The specific asset class changes. The specific model changes. The geometry does not.

Roger Lowenstein's 'When Genius Failed' came out in 2000 and remains the best book on the episode. The sections on the individual trades read like a thriller. Everything in it about how position sizing interacts with market depth is still textbook material for anyone running a large book, and the shape it describes is the same shape Newton lived through in 1720.