Optimal Hedging Under Liquidity Constraints: The Metallgesellschaft Lesson

Executive Insight

In late 1993, Metallgesellschaft AG’s U.S. subsidiary MGRM accumulated $1.3 billion in losses on energy derivatives positions—losses large enough to threaten the survival of one of Germany’s largest industrial conglomerates. The losses did not arise from speculation: MGRM had entered into long-term fixed-price oil delivery contracts with customers and hedged them with short-dated rolling futures. The hedging strategy was theoretically sound but operationally catastrophic.

This paper formalizes the Metallgesellschaft failure as a constrained variance-minimization problem. Using the framework of Glasserman (Columbia) and the optimal solution of Albrecher & Leobacher, we derive the minimum possible maximum variance achievable by any hedging schedule and show that the standard rolling-stack hedge is suboptimal by a factor of 3.8×. The optimal schedule reduces peak variance from 0.148 to 0.0389—a result with direct implications for any institution running maturity-mismatched derivatives books.

The Metallgesellschaft failure was not a failure of hedging theory. It was a failure to incorporate funding path constraints into the hedging objective. The optimal hedge ratio is not 1.0—it is a function of the variance-of-the-path, not just the variance of the terminal payoff.

The MGRM Business Model

MGRM offered customers long-term fixed-price contracts for heating oil and gasoline delivery at prices above the prevailing spot rate. A typical contract promised delivery of oil at $a$ dollars per barrel for up to 10 years. To hedge the resulting short position in oil, MGRM purchased short-dated futures contracts and rolled them forward each month—the so-called stack-and-roll strategy.

The economics appeared straightforward: the long-term delivery premium ($a - c$, where $c$ is the expected long-run price level) was positive and contractually locked in. The hedge via short-dated futures eliminated price risk at each maturity. But the strategy ignored a critical constraint: the path of cumulative margin flows between inception and final delivery.

When oil prices dropped sharply in 1993, the short-dated long futures positions lost money month after month, generating cumulative margin calls that reached $1.3 billion. The fixed-price delivery contracts were simultaneously gaining in value—but those gains were unrealized and illiquid. The board of Metallgesellschaft, facing a liquidity crisis, unwound both sides of the position at the worst possible time: just before oil prices reversed.

Formal Problem Specification

The source study models the oil price as a random walk with drift:

For the illustrative case: $c = 50$ (long-run price), $\sigma = 1$. MGRM delivers quantity $q$ at fixed price $a > c$ per period, for $N$ periods. The unhedged cumulative profit through period $k$ is:

The first term is the locked-in delivery premium; the second term is pure noise driven by the cumulative sum of random shocks. The hedged profit, using a futures position $h_n$ in period $n$, becomes:

where $g_n$ is the cumulative hedge ratio through period $n$. The risk measure is maximum variance over the life of the contract:

In the continuous-time (Itô) formulation, the hedged residual risk process is $\tilde{R}(t) = \int_0^t (t - s - g(s))\,dW(s)$, and the variance is $V(t) = \int_0^t (t - s - g(s))^2\,ds$.

The Rolling-Stack Hedge

The rolling-stack strategy sets $g(s) = 0$ for all $s$—i.e., the firm holds a full one-for-one futures position in the nearest contract and rolls it forward at each maturity. This is the strategy MGRM actually employed.

Under this strategy, the variance function is $V_{\text{stack}}(t) = \int_0^t (t-s)^2\,ds = t^3/3$. The maximum over $[0,1]$ is:

If MGRM had perfectly rolled a one-for-one stack hedge (the textbook solution), the maximum variance drops to 0.148—a 56% reduction from the unhedged case. But as we show below, this is far from optimal.

Glasserman’s Improved Strategies

Glasserman (Columbia) proposed two refined strategies that reduce maximum variance further by optimizing the hedge schedule:

Glasserman Strategy I (Optimal Horizon): Hedge only over the subinterval $[0, \tau^*]$ with a full position, then hold no position from $\tau^*$ to maturity. The optimal cutoff is $\tau^* \approx 0.733$, yielding maximum variance 0.0583.

Glasserman Strategy II (Optimal Fraction): Hold a constant fractional position $\gamma^*$ throughout $[0,1]$. The optimal fraction is $\gamma^* \approx 0.630$, yielding maximum variance 0.0456. This corresponds to hedging 63% of exposure throughout the contract life.

The Albrecher–Leobacher Optimal Solution

Albrecher and Leobacher derive the globally optimal hedge schedule by solving the minimax problem exactly. The solution is a piecewise function with three regimes:

• Early phase ($s < s_1$): Full hedge — $h(s) = 1$ (comparable to rolling-stack)
• Middle phase ($s_1 \le s \le s_2$): Declining hedge ratio — $h(s)$ decreases smoothly
• Late phase ($s > s_2$): No hedge — $h(s) = 0$ (position is unwound)

The resulting minimum possible maximum variance is:

This is the theoretical lower bound: no hedging schedule can achieve a lower maximum variance under the random walk model. It reduces maximum variance by 88% relative to no hedge and by 74% relative to the rolling-stack strategy MGRM actually used.

The intuition is profound: the optimal hedge is not one-for-one. It starts fully hedged, then gradually reduces the position as the remaining delivery period shortens—optimally trading off terminal hedge effectiveness against cumulative variance (i.e., margin call risk). This is precisely the trade-off that MGRM’s treasury failed to recognize.

Practical Implications for Commodity Desks

The Metallgesellschaft case remains the canonical example of how a theoretically sound hedge can destroy a firm. The quantitative framework above provides actionable guidance:

For Trading Desks

• Hedge ratio calibration: The optimal hedge ratio for a 10-year rolling commodity position is approximately 0.63, not 1.00. Full hedging maximizes terminal P&L protection but generates path-of-margin variance that can trigger forced liquidation.
• Dynamic schedule: The Albrecher–Leobacher schedule provides a concrete template: full hedge in the first third of the contract, declining hedge in the middle third, and no hedge in the final third. This reduces peak margin calls by 74% relative to a full stack-and-roll.

For Risk Managers

• Path variance monitoring: VaR and CVaR metrics computed at maturity are insufficient for maturity-mismatched books. Risk systems must compute and report the maximum variance over the life of the position—not just the terminal distribution.
• Liquidity stress scenarios: Hedge programs for long-dated delivery contracts must be stress-tested against multi-year adverse price paths, with explicit liquidity buffers sized to the cumulative margin requirement under the $2\sigma$ scenario.

For Boards and Audit Committees

• Any long-term delivery or supply contract hedged with short-dated instruments creates an implicit funding commitment equal to the maximum cumulative mark-to-market loss over the contract life. This commitment must appear in the firm’s liquidity planning as a contingent liability. MGRM’s board did not understand this commitment; they unwound the position at a realized loss of $1.3B, just before oil prices reversed and the fixed-price contracts would have become profitable.

Methodology & Academic Foundation

The analysis draws on the following academic and institutional sources:

• Glasserman, P. (Columbia University): minimax variance framework for maturity-mismatched hedging; optimal horizon and optimal fraction strategies
• Albrecher, H. & Leobacher, G.: exact solution of the optimal hedging schedule under random-walk spot dynamics; closed-form expression for minimum maximum variance involving $e^{-\pi/(6\sqrt{3})}$
• Culp, C. L. & Miller, M. H. (1995, Journal of Applied Corporate Finance): defense of MGRM’s hedging strategy and analysis of the board’s decision to liquidate
• Edwards, F. R. & Canter, M. S. (1995): institutional analysis of the funding crisis and margin mechanics
• Pirrong, S. C. (1997): critique of stack-and-roll hedging for long-dated delivery commitments
• Continuous-time formulation: Itô integral representation of the residual risk process $\tilde{R}(t)$ enables exact variance computation via the isometry $\operatorname{Var}[\tilde{R}(t)] = \int_0^t (t-s-g(s))^2\,ds$

All numerical results are computed from closed-form expressions; no Monte Carlo simulation is required. The illustrative parameters ($c=50$, $\sigma=1$, $N=1$) normalize the problem to unit interval and unit variance, making the variance ratios universally applicable to any commodity hedging context.