A Holistic Agent-Based Financial Simulation Platform of the Futures Commodity Market using Self-Learning Speculative Traders

The behavior of traders in financial markets is based on all available data surrounding a financial exchange. In other words, a holistic approach is required to simulate a trading strategy of a trader without making decisions based on incomplete data. In addition, the complexity of the complete market far exceeds the complexity of an individual trader; thus, an agent-based approach is beneficial. This paper describes a holistic agent-based simulation platform. This platform is holistic for three reasons. First, we ground the simulation in the commodity futures market, this makes it possible for agents to make trading decisions based on real-world data, like supply, demand, or external factors. Second, we use scientific literature to identify and group traders with different motivations and model each group independently. For example, the simulation platform contains market makers, portfolio managers, hedgers, etc. In addition, speculative traders are generated using a genetic algorithm on real data, to increase the diversity of agents, and transfer real market behavior into the simulation. Third, the logic of each trader is based on a proven strategy in the real world; either we use well-described methods in the literature, or use machine learning to construct profitable agents on real data. Last, data-driven methodologies are used to validate the simulation with reality. This paper presents the preliminary results of the simulation platform. In future work, this platform will make it possible to revisit existing hypotheses or explore undescribed hypotheses.