We present a dynamic stochastic programming model to manage future contracts in Chinese commodity future markets. We simulate the uncertainty in asset prices by means of the scenario trees that approximate the empirical joint distributions implied by historical market data. We firstly propose an improved algorithm that generates a discrete joint distribution consistent with the first four marginal moments and correlation matrix of random variables. We point out that algebra modeling language and our moment matching algorithm make it possible for any non-specialist users to build models in order to solve complex sequential decision making problems. Secondly, we propose a stochastic programming model for managing the futures contract. Empirical results based on the Chinese commodity futures contracts compare the performance of our proposed model with those of other popular trading strategies presented in finance literature. We also show that the multi-stage model outperforms single-stage models in terms of the stability of return series generated.