An LLM-Driven Agent-Based Framework for Cross-Regional Construction Policy Evaluation in the Greater Bay Area

Abstract: Cross-jurisdiction practice policies in the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) seek to reduce institutional frictions and enable cross-regional delivery of construction professional services, but ex ante evaluation is difficult because outcomes emerge from heterogeneous actors operating under strict project-process gates and evolving collaboration networks. We propose an LLM-driven agent-based simulation that integrates an executable, stage-gated project lifecycle, role-consistent stakeholder agents with capacity limits, and a lightweight macro environment capturing time-varying demand, price pressure, and regulatory intensity. Cross-jurisdiction rules are encoded as a policy parameter vector—entry friction, ongoing compliance burden, and scope constraints—and injected into eligibility/waiting times, cost–risk functions, and event/transition mechanisms. Stylized-facts tests reproduce increasing concentration with head-firm dominance and distance decay with strong local bias. Counterfactual experiments compare policies via network stratification, diffusion paths, and directional collaboration asymmetries, providing a reproducible testbed for mechanism-based policy analysis.