Decomposing Financial Market Dynamics via Mechanism Analysis in an Evolutionary Multi-Agent Simulation

Researchers decompose financial market dynamics by testing four pluggable mechanisms in an evolutionary agent-based model with 120 heterogeneous agents, finding that selection operators control diversity, price microstructure drives realism, and behavioral bias amplifies fragility—but these levers operate largely independently, offering a framework for understanding which market design choices produce which emergent properties.

This research advances agent-based modeling methodology by isolating causal relationships between market mechanisms and emergent outcomes. Traditional ABMs bundle multiple design choices together, making it impossible to determine which mechanism produces which effect. By systematically varying four mechanisms across 3×20 seed runs with statistical rigor, the authors establish that mechanism effects are largely separable and quantifiable.

The findings have direct implications for market design and risk management. The discovery that selection operators significantly influence strategy diversity—with Quality-Diversity approaches generating 0.27 to 1.12 additional bits of entropy compared to truncation selection—suggests evolutionary pressure fundamentally shapes competitive dynamics. More provocatively, the authors show that explicitly optimizing for realism does not produce realistic market behavior, challenging the common assumption that objective alignment guarantees emergent validity.

The separation of microstructure effects reveals that reflexive price feedback (bid-ask dynamics, feedback loops) drives realism while behavioral bias amplifies a fragility proxy. This distinction matters for regulators and exchange designers: improving price discovery mechanisms raises realism without necessarily introducing fragility, while pure behavioral heterogeneity increases systemic fragility. Consensus network topology showed no robust effect, suggesting peer influence structures may matter less than previously theorized.

For cryptocurrency markets, these findings inform mechanism design debates. The research suggests that exchange microstructure choices (matching engines, price formation rules) are orthogonal levers separate from agent diversity and behavioral assumptions. This decomposition enables more targeted market regulation and mechanism optimization without unintended spillovers across market properties.

• →Quality-Diversity selection operators increase strategy entropy by 0.27–1.12 bits over conventional truncation, sustaining more frequent strategy cycling.
• →Reflexive price feedback is the primary driver of market realism, while selection operators and behavioral bias have minimal direct realism impact.
• →Behavioral bias amplification increases systemic fragility proxies by 10.5–14.4 units without affecting price realism, revealing independent control channels.
• →Consensus network topology shows no statistically significant effect on market properties, challenging network-centric market design assumptions.
• →Market mechanisms act as approximately separable control knobs, enabling targeted optimization without unintended cross-mechanism effects.