Agent-Based Models

Agent-based models (ABMs) are computational simulations in which autonomous agents — individuals, firms, organisms, or nations — interact according to local rules, and the macroscopic behavior of the system emerges from these micro-level interactions rather than being imposed by a global equation. ABMs are the methodological complement to equation-based modeling: where differential equations describe aggregate flows, ABMs describe the discrete decisions that produce those flows.

The approach is particularly powerful when heterogeneity matters. If all agents are identical and interact uniformly, an equation-based model is usually more efficient and equally accurate. But when agents differ in their rules, their information, or their network position — when the system is structurally diverse — ABMs can capture dynamics that aggregate equations flatten into misleading averages. Financial markets with heterogeneous trading strategies, ecosystems with diverse foraging behaviors, and societies with stratified social networks all exhibit phenomena that emerge from agent-level differences rather than from population-level parameters.

ABMs are also the natural experimental framework for testing theories of emergence: by varying agent rules and interaction topologies, researchers can identify which micro-level mechanisms produce which macro-level patterns. The methodology has been criticized for its parameter flexibility — with enough tuning, an ABM can be made to produce almost any aggregate behavior — but this flexibility is also its strength: it forces explicit formalization of the mechanisms that other approaches assume away.