New benchmark reveals how adaptive policies fool standard regulation simulations

Agent-based models (ABMs) are widely used for regulatory simulation, but most treat policy as a fixed parameter, ignoring how agents or policies adapt. Roberto Garrone’s new paper on arXiv provides a much-needed methodological benchmark: a configurable emissions-regulation ABM that systematically compares four regime combinations—constant policy/constant agents, constant policy/adaptive agents, adaptive policy/constant agents, and adaptive policy/adaptive agents. The benchmark tests three adaptive controllers (setpoint, safety-margin, and one-sided control) alongside naive and tracking-aware fixed policies.

The results reveal that policies producing similar average outcomes can behave very differently under the hood. Setpoint control tracks the cap but causes frequent boundary crossings; safety-margin reduces violations through conservatism; one-sided control can ratchet toward over-conservatism when combined with adaptive agents. The paper argues that regulatory conclusions drawn from static scenarios may mislead real-world implementation. To avoid this, researchers should evaluate regimes via “structural distinguishability” —using scalar indicators, symbolic diagnostics, and trajectory motifs—rather than relying solely on average performance.