Doyne Farmer Proposes $100M Global Economic Simulator to Guide Climate and Financial Decisions

Doyne Farmer, an Oxford complexity scientist with a long history in applied computation and markets, is advancing a concrete plan to build a comprehensive, data-rich model of the global economy aimed at improving policy and corporate decision-making. The proposal asks for roughly $100 million to assemble and process operational data across tens of thousands of firms and hundreds of thousands of physical assets, and to run agent-based simulations that let economic actors follow simple, evolving rules rather than assuming perfect rationality. Farmer frames the project as an answer to the persistent mismatch between traditional economic models and observed outcomes — models that often rely on equilibrium assumptions or abstract representative agents and therefore miss cascading dynamics and emergent crises. His approach replaces those assumptions with many interacting digital agents that learn and adapt, producing booms, busts, and transitions internally in simulated economies rather than requiring exogenous shocks. The initial work targets the energy sector: building a detailed 25-year operational dataset covering roughly 30,000 companies and about 160,000 assets, to test how different policies and business strategies change real deployment and costs. If calibrated successfully, the platform would allow policymakers and firms to compare pathways for decarbonization while quantifying trade-offs in price, supply, and investment timing. Farmer underscores the platform’s potential to reduce extreme tail losses, citing the 2008 global shock as an example of avoidable systemic damage that better foresight might have mitigated. He also emphasizes computational advances and richer data sources as enablers: modern compute power and machine learning permit scale and adaptation that were impossible decades ago. The design intentionally simplifies individual decision rules so the system can scale to millions of agents while preserving realistic heterogeneity in behavior. That simplifying strategy also addresses the problem that actors change behavior when the environment changes, by embedding learning algorithms that allow agent strategies to evolve. Practical challenges remain substantial: access to reliable, standardized enterprise data; institutional buy-in from governments and corporations; validation of counterfactuals; and governance around model transparency and misuse. Farmer suggests a multi-stakeholder funding model and a decade-scale timeline with early sectoral demonstrations; he frames the ask as modest relative to the potential avoided losses and economic gains. If delivered, the tool would not eliminate uncertainty but could materially shift the information set available to decision-makers, making policy more anticipatory and less reactive. The proposal reframes economic modeling as an engineering problem—build the large, realistic simulation and let it reveal levers for smoother transitions and risk reduction.