CHAD: A Scalable Turn-Based Simulator for Complex Heterogeneous Agent Dynamics

Patrick Liston, Artem Lensky, Charles Gretton

Keywords: Agent-based simulation, multi-agent systems, market microstructure, emergent dynamics, scalability, heterogeneous agents

Abstract: Simulating large populations of heterogeneous, interacting agents We present CHAD (Complex Heterogeneous Agent Dynamics), an agent-based market simulator implemented in Julia that addresses scalability and reproducibility limitations of existing frameworks. CHAD introduces two principal design contributions: (i) a pre-scheduled, turn-based architecture that precomputes all agent activation times, enabling deterministic execution and near-linear computational scaling to millions of agents; and (ii)~a dual-book system comprising a public Limit Order Book and a private Stop-Loss Order Book that endogenously produces cascading liquidity dynamics. Through a combinatorial grid search over 784 agent population configurations, we demonstrate that---within the tested parameter regime---macroscopic market properties including fat-tailed return distributions, volatility clustering, and rapid cascading liquidity events are primarily driven by agent population composition. Validation against empirical BTC/USDT data confirms that the simulator reproduces a broad range of stylized facts. Computationally, the combination of a pre-scheduled architecture and Julia's JIT compilation enables stable execution at one million agents, a scale at which the event-driven ABIDES framework could not complete. We conclude with open questions on calibration methodology and the generalisability of these ecological dynamics beyond financial markets.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 27