Heterogeneous agent beliefs and clustered volatility in commodity futures market

We propose a multi-agent-based model of a futures market to analyze the characters of commodity futures price stylized facts. The model includes hedger agents and speculative agents. We use the Brenner's "stochastic belief learning model" to describe the speculative agents' beliefs learning process, which is a social learning process with local information. In our model, the bounded rational speculative agents have different learning capability and risk aversion degree, and they can learn both from individual and others' experiences. New market price is generated though a sealed-bid auction clearance mechanism. The simulation can reproduce the important observed stylized facts in futures markets price time series, including fat tails, clustered volatility, and long memory in returns distribution. Our results show clustered volatility in returns depends on the level of speculators' imitation behaviors. Social learning process leads to imitation behaviors and futures price volatility has close relation with large speculators' trading.