Amaranth, 2006

$6.6bn lost in a single month on natural gas spreads. Not a wrong view — a position so large relative to the market that exiting it was the loss.

What happened

  • 2005 — Amaranth Advisors, a multi-strategy fund, earns very large profits in natural gas after Hurricanes Katrina and Rita disrupt supply. One trader is responsible for most of the firm's returns.
  • 2006 — the energy book grows to dominate the fund. The core position is a calendar spread: long March contracts, short April contracts, betting the winter-to-spring price gap would widen.
  • The logic — March is the last winter month and April the first spring month. A cold winter drains storage and the March–April spread blows out. It had done exactly that before.
  • By mid-2006 — the fund holds positions reported at around half of the entire open interest in some contracts. It is not a participant in the market; it is a large share of it.
  • September 2006 — a mild hurricane season and comfortable storage send the spread the other way. It moves against the position steadily.
  • Mid-September 2006 — margin calls force liquidation. There is no one to sell to at anything near the marked price. Losses reach roughly $6.6bn — about 70% of the fund — in a matter of weeks.
  • Late September 2006 — the remaining book is transferred at a steep discount to JPMorgan and Citadel. The fund closes.

The mechanism

  • The position was too large for its own market. A calendar spread is normally a modest, low-volatility trade. At this size it stopped being a spread trade and became the market's price-setting mechanism — an outcome no risk model produced by measuring the spread's history could anticipate.
  • Marks were not exit prices. The book was valued at screen prices for a size those screens could never absorb. The difference between a valuation and an achievable exit is the entire loss.
  • Liquidation is self-punishing at scale. Every contract sold moved the spread further against the remaining position. The exit created the loss it was trying to escape — the same reflexivity that destroyed the short-volatility ETPs and the LDI funds.
  • Standard risk measures missed it entirely. A VaR model calibrated on the spread's historical volatility said the position was moderate. VaR has no term for "you own half the market", because market impact is not a distributional property.
  • Diversification was nominal. Amaranth was marketed as multi-strategy. In practice one book determined the outcome — a concentration visible in the risk contribution of the portfolio long before it was visible in the weights.

What it teaches

  • Position size relative to market capacity is a risk factor in its own right. Ask what fraction of daily volume the position represents and how many days a full exit would take. If the answer is weeks, the marks are fiction.
  • Liquidity risk is not volatility risk. They are measured differently, they behave differently, and only one of them shows up in the standard reports. The cost of trading at ordinary size tells you nothing about the cost at extraordinary size.
  • "Multi-strategy" is a description of intent, not of risk. Measure where the risk actually sits, not where the capital is allocated.
  • Last year's genius is this year's concentration. The trader who produced enormous returns in 2005 was given more capital and less resistance in 2006. Success systematically weakens the controls that constrain it — the identical dynamic as Barings.
  • No taxpayer money and no systemic crisis. Unlike LTCM, Amaranth failed without a rescue and markets absorbed it. That is what a hedge fund failing correctly looks like, and it is a useful contrast: the problem is not that funds lose money, it is when their losses have to be socialised.