The Genius and the Abyss
How Long-Term Capital Management turned small pricing gaps into extraordinary returns, and discovered that being right about value is not the same as surviving the journey.
Greenwich, Connecticut, 1994. John Meriwether is building a hedge fund around an idea that has already made him one of the most respected bond traders on Wall Street. Markets often place different prices on instruments that are economically similar. Those differences may persist for a time, but they should eventually narrow. Find enough of them, hedge the broad market exposure and wait for value to reassert itself.
The team assembled around that idea is exceptional.
It includes traders from Meriwether’s former arbitrage group at Salomon Brothers, among them Eric Rosenfeld, Victor Haghani and Lawrence Hilibrand. David Mullins joins after serving as vice-chairman of the Federal Reserve Board. Myron Scholes and Robert Merton bring the intellectual authority of modern financial economics. In 1997, while the fund is still operating, Scholes and Merton will receive the Nobel Memorial Prize in Economic Sciences for their contribution to derivative pricing.
The new firm is called Long-Term Capital Management.
Its early results appear to vindicate the approach. According to the Federal Reserve’s historical account, LTCM earns approximately 20 per cent in 1994, 43 per cent in 1995, 41 per cent in 1996 and 17 per cent in 1997.
Those returns are not produced by forecasting the direction of the stock market or the economy. They come largely from relationships.
A recently issued US Treasury bond may trade at a premium to an older bond with a similar maturity because the newer bond is easier to trade. LTCM might buy the cheaper bond and sell the dearer one, expecting the price difference to narrow. Similar reasoning can be applied to interest-rate swaps, government bonds in different countries, mortgage securities, equity volatility and other instruments.
The price gaps are often tiny. To turn them into attractive returns, the fund uses leverage and applies the same reasoning across a large portfolio.
For several years, the machine works.
Then the market changes the question.
THE EDGE AND THE LEVERAGE
LTCM’s basic insight is sound. Closely related securities can trade at inconsistent prices because investors value liquidity differently, institutions face different constraints, or temporary demand overwhelms patient capital.
If the relationship later normalises, the trade can profit without requiring a broad prediction about whether markets will rise or fall.
But there is a structural problem. A small pricing discrepancy produces a small unleveraged return. The more certain the convergence appears, the more tempting it becomes to borrow against the position.
Suppose two bonds are expected to move one percentage point closer together. An investor using only personal capital may earn a modest return. An investor who controls thirty dollars of assets for every dollar of capital can turn the same movement into something far larger.
The leverage magnifies the gain. It also reduces the distance the trade can move in the wrong direction before the investor is forced to act.
By the end of 1997, LTCM reportedly had about thirty dollars of debt for every dollar of capital. It had also returned capital to investors without reducing the overall scale of its positions. That increased leverage at the same time that profitable opportunities were becoming harder to find.
Leverage did not make the underlying analysis foolish. It changed the time available for the analysis to prove correct.
A bond spread may eventually converge, just as the model predicts. But if it widens sharply first, the fund can face margin calls, collateral demands or withdrawals of financing. A leveraged investor does not merely need to be right about the destination. The investor must remain solvent along the route.
This is the first boundary the model could not remove.
WHAT THE NUMBERS COULD SEE
LTCM did not operate without risk controls. Its partners were deeply familiar with probability, portfolio construction and hedging. They analysed how positions had behaved, how they related to one another and how losses might accumulate.
The difficulty was not the absence of measurement. It was the boundary around what could be measured with confidence.
A risk estimate built from recent market behaviour can describe the range of outcomes that appeared during that period. It can show how frequently spreads moved by a given amount, whether positions tended to offset one another and how much capital might be lost under specified scenarios.
What it cannot establish is that the relationships will remain stable when the reasons for trading change.
In ordinary conditions, one relative-value position may have little connection with another. A spread in Danish mortgages can behave differently from a spread in Italian government bonds. An equity-volatility position may appear to diversify an interest-rate trade.
During a funding crisis, those distinctions can weaken.
Investors stop asking which security offers the better long-term value. They ask what can be sold now, what collateral must be posted today and which position will consume the least scarce balance sheet. Instruments that were unrelated by economics become connected through ownership, leverage and the need for cash.
Diversification measured across normal periods can disappear precisely when it is expected to protect the portfolio.
There is a second boundary. A quoted market price assumes that someone is willing to trade there. A large portfolio may appear liquid while conditions are calm, yet become difficult to reduce when many participants want the same exit.
The risk is not merely that prices will move. It is that the fund’s own attempt to sell will push them further away.
They had eliminated every risk they could model. The market specialises in risks that cannot be modelled.
The line is deliberately provocative. No model can literally eliminate risk. What LTCM had done was hedge many exposures that could be identified and estimated. The danger lay in treating the remaining exposure as smaller and more stable than it proved to be.
RUSSIA CHANGES THE QUESTION
The crisis does not begin with a single trade at LTCM.
Financial stress has been moving through Asia since 1997. By the northern summer of 1998, confidence is already fragile. On 17 August, Russia devalues the rouble, restructures rouble-denominated government debt and imposes a temporary moratorium on some private foreign obligations.
The decision shocks markets.
Investors move towards the safest and most liquid assets they can find. US Treasury securities rise as demand for safety increases. Credit spreads widen. Emerging-market debt falls. Positions that depend on the normal relationship between liquid and less-liquid instruments begin moving together.
LTCM is not ruined simply because it owns Russian bonds. Its direct Russian exposure is not the centre of the story.
The default changes behaviour across markets. The fund is often positioned to benefit when a cheaper or less-liquid instrument moves towards a dearer or more-liquid counterpart. During the flight to quality, investors demand the liquid instrument regardless of relative value. The expensive side becomes more expensive, while the cheap side becomes cheaper.
The spreads widen instead of narrowing.
In August alone, LTCM loses 44 per cent of its value. By the end of that month, the partners report that the fund is down 52 per cent for the year and seek new capital.
The losses create their own momentum. Falling capital increases effective leverage. Lenders and counterparties become more cautious. The fund needs capital at the moment potential investors can see that its positions are under pressure. Other market participants begin anticipating what LTCM might have to sell.
Value has not disappeared. Time has.
FIVE WEEKS
Between Russia’s announcement on 17 August and the private recapitalisation on 23 September, LTCM moves from severe distress to the edge of disorderly failure.
The speed matters.
A relative-value portfolio can contain offsetting positions that look modest after they are netted together. Yet every position still has to be financed, margined and closed. Gross exposures determine how many counterparties are involved and how much trading would be required if the portfolio had to be unwound.
LTCM was active across government bonds, mortgage-backed securities, equities and a large range of derivatives, including swaps, forwards and options. More than seventy-five counterparties had exposure to the fund.
This network created a problem larger than the loss of the partners’ capital.
If LTCM defaulted, each counterparty would try to protect itself. Positions would be closed, collateral sold and replacement hedges established. Those actions might be rational for each firm individually. Taken together in already fragile markets, they could drive prices sharply against everyone attempting the same escape.
The fund had become too large for liquidation to be treated as a private event with neatly contained consequences.
This does not mean every position was bad or every model was wrong. Many trades might have converged if they could have been held long enough. The failure was that the portfolio required financing and liquidity to remain available until that happened.
Being right eventually is not enough when survival is settled daily.
THE NEW YORK FED MEETING
On 20 September, a team from the Federal Reserve Bank of New York visits LTCM’s offices. The scale of the positions and the difficulty of reducing them in thin markets become clearer.
The concern is not that LTCM’s owners should be protected from loss. It is that an abrupt close-out could force hundreds of billions of dollars of transactions into markets already struggling to absorb sales.
The New York Fed brings major counterparties together to consider a private solution.
On 23 September, fourteen banks and securities firms agree to contribute approximately $3.625 billion in exchange for 90 per cent of the fund. LTCM’s partners and existing investors retain only 10 per cent and suffer substantial losses. The portfolio is then reduced in a more orderly manner.
No Federal Reserve money is invested. No government guarantee is provided. The New York Fed facilitates the negotiations, but the counterparties finance the recapitalisation because they have the most to lose from a disorderly collapse.
This distinction is important. Calling the transaction a bailout can suggest that public money restored the partners. It did not. The intervention was a privately funded attempt to prevent a fire sale from spreading losses through already stressed markets.
The episode still raised a serious question.
How had institutions with sophisticated risk systems allowed one counterparty to build positions large enough that its failure could threaten market functioning?
The answer did not sit inside LTCM alone. Its counterparties had extended credit, written derivatives and accepted collateral while often lacking a complete view of the fund’s exposures elsewhere. Each firm could see its own relationship. Few could see the network.
The system had measured the pieces and missed the whole.
WHAT ACTUALLY FAILED
It is tempting to say that LTCM proved quantitative finance was an illusion. That conclusion is too easy.
The fund found real discrepancies. Its early profits were real. Hedging related instruments can reduce unwanted exposure. Models can reveal relationships that intuition misses.
The failure came from treating several conditional truths as though they were permanent.
Spreads usually converge, but not on a schedule chosen by the investor. Historical relationships often provide useful information, but they can change when leverage and funding pressures dominate. A position may be hedged against a small movement in normal conditions while remaining exposed to a gap, a liquidity withdrawal or a common rush for safety.
Most importantly, the risk of a leveraged portfolio is not contained in its price history.
It also lives in its financing, collateral terms, counterparties, position size and capacity to exit. These are features of the market architecture, not merely properties of the securities.
LTCM’s losses did not show that intelligence was dangerous. They showed that intelligence applied to a narrow problem can create confidence beyond the boundary of the solution.
The model described relative value.
It could not guarantee the time, funding or liquidity required to realise it.
THE OUTLIER HUNTER'S DIFFERENCE
An Outlier Hunter begins from a different problem.
The aim is not to identify a small pricing discrepancy and magnify it until it produces an attractive return. It is to participate broadly, begin with small exposure and remain available for the rare sustained move that cannot be identified in advance.
That does not make trend following immune to leverage, crowding or illiquidity. A trend follower can still be too large, trade markets that cannot absorb an exit or mistake scale matching for risk control.
The difference lies in the intended payoff structure.
Convergence trading expects a relationship to return towards normal. Its gains are often limited to the closing of the spread, while losses can expand if the relationship moves further away. Trend following accepts repeated small losses in exchange for the possibility that an unusual movement will continue much further than expected.
During the 1998 flight to quality, government bond prices rose sharply as investors sought safety. A trend follower did not need to predict the Russian default or calculate its probability. Rules could respond to the movement as it developed, including through long positions in strengthening government bond markets where signals and portfolio construction allowed.
The practical defence remains structural: small initial positions, broad diversification, predefined exits and no reliance on one estimate of how markets should behave.
LTCM had extraordinary knowledge of the relationships it traded.
What it lacked was enough room for those relationships to be wrong, or merely early, at the same time.
The genius was real.
So was the abyss beneath it.
Next: Article 6, The River That Was Not on the Map
New York, 2006.
Mortgage risk is being sliced, recombined and sold in structures so complicated that the distance between the borrower and the final investor has almost disappeared from view.
Ratings depend on assumptions about default rates, recovery values and the extent to which mortgage losses in different regions will occur together. Recent history appears reassuring. National house prices have not suffered the kind of broad decline that would challenge the architecture.
The models can calculate what happens inside the range they were given.
The housing market is about to leave it.
Richard Brennan writes on systematic trading, complex adaptive markets, and the philosophical foundations of trend following at atstradingsolutions.com. His books include The Fractals of Finance, Complex Adaptive Markets, Carved by Impossibility and The Aussie Turtles Trend Following Guide.
Want to explore why structure exists at all?
Carved by Impossibility: What Remains When Everything Else Is Eliminated
The book explores the architecture of constraint, emergence, and reality itself, and what it means for how we understand markets, life, and the universe.
Available now on Amazon in paperback, hardcover, and Kindle.
Want the theoretical foundation for why trend following works?
The Fractals of Finance: Determinism, Adaptation and the Geometry of Markets
The book explores the full architecture of feedback, fat tails, and fractal structure in financial markets, and what it means for how we trade, invest, and understand risk.
Available now on Amazon in paperback, hardcover, and Kindle.
Want a practical field manual for trading trends and capturing outliers?
The Aussie Turtles Trend Following Guide: A Field Manual for Hunting Outliers adapts the timeless principles of the original Turtle traders into a systematic, rules-based approach for modern markets. Co-authored with Adam Havryliv.
Available now on Amazon in paperback, hardcover, and Kindle.