Price modeling and portfolio optimization in commodity markets

This thesis analyzes the commodity market and existing stochastic price models for the valuation of commodity futures contracts in two parts. The first part proposes a machine learning-based model for state prediction of agricultural commodity prices. Motivated by the strong dependence of these prices on external factors, the application of a clustering and classification algorithm allows the inclusion of these factors in the price determining process. Combined with a stochastic price model for agricultural commodity futures, the proposed model allows the generation of state-dependent price scenarios via Monte Carlo simulation. The second part solves an investor's portfolio optimization problem in a market with investment possibilities in the money market account and in commodity futures contracts. The price of the futures contract is derived from a one-factor stochastic price model that considers a stochastic market price of risk. Using stochastic control methods, a stochastic optimal portfolio strategy is derived. Two simulation studies follow - among others the optimal portfolio strategy is compared with deterministic strategies that are more practicable in their application.