Failure modes · Process
LTCM: Two Nobel Laureates and No Pre-Committed Deleveraging Trigger
The models assumed spreads would converge. They did not include the fund’s own size as a risk, nor a written rule for what to do when every position moved against them at once. Intelligence did not substitute for the missing commitment.
Long-Term Capital Management began 1998 with roughly 25-to-1 balance-sheet leverage. By the end of August, after the Russian default, leverage had risen toward 50-to-1 as equity fell. By September it reached extremes near 250-to-1 on some measures as losses compounded. Assets dropped from about $129 billion to a fraction of that. Equity fell from roughly $4.7 billion to under $600 million in weeks. A Fed-brokered rescue followed.
The partners included Myron Scholes and Robert Merton. The strategies were market-neutral arbitrage. The VaR models treated positions across asset classes as imperfectly correlated. When Russia defaulted, correlations went to one. Almost every spread trade moved against the fund at the same time. There was no position-level stop and no aggregate drawdown limit that would have forced deleveraging while the fund still had equity.
The missing cold-state question
The failure was not a lack of intelligence. It was the absence of a pre-committed answer to the question: what do we do if every position moves against us simultaneously? That question is easy to ask when volatility is low and the models are confirming. It is expensive to answer for the first time when cortisol is elevated and counterparties are stepping back.
Size itself became the risk. The fund was large enough that its own attempts to exit moved prices further against it. The models did not include the fund. This is the practical limit of any risk system that treats the portfolio as a price-taker in all scenarios.
What transfers to smaller capital
Most investors will never run 25-to-1 leverage. The transferable lesson is smaller. If your process relies on mean-reversion, convergence or low historical correlation, write the aggregate loss level at which you reduce exposure regardless of whether the models still say the trades are attractive. Write it when the account is calm. The LTCM partners were not short of analytical horsepower. They were short of a rule that survived contact with a correlated tail.
What to do with this
If you run any strategy that assumes positions will not all move against you at once, write the portfolio-level drawdown that triggers a mandatory reduction in gross exposure. Make the number small enough that you can still act. Review it once a year in a cold state. The existence of the number is the commitment. Its absence is the LTCM error at retail scale.