Abstract
In this paper we research the hedging error that comes about through synthetic replication of long-term commodity-linked contracts. An extensive data set of close to 300 different commodities has been used in the study. Focusing on the price process, the standard geometric Brownian motion provides the most robust pricing formulas and hedging recipes. Hedging errors are however smaller for a new mean reverting process, supporting the intuition that commodity prices are mean reverting. In this way, hedging tests can act as a complement to purely statistical tests of mean reversion since such tests unfortunately have low power. A stochastic volatility model is also tested but is found to work poorly. Hedging errors are the largest for this model. The issue of choosing an appropriate price process is overshadowed by the problem of parameter estimation. Historic volatility may be very different from the actual volatility and this deviation introduces a substantial error that makes the hedging of long-term commodity-linked contracts very difficult.