Using Machine Learning to Compute Constrained Optimal Carbon Tax Rules

21.11.2025 11:15 – 12:15

RESEARCH INSTITUTE FOR STATISTICS AND INFORMATION SCIENCE SEMINARS

ABSTRACT

We develop a computational framework for deriving Pareto-improvingand constrained optimal carbon tax rules in a stochastic overlappinggenerations (OLG) model with climate change. By integrating Deep EquilibriumNetworks for fast policy evaluation and Gaussian process surrogate modelingwith Bayesian active learning, the framework systematically locates optimalcarbon tax schedules for heterogeneous agents exposed to climate risk. We applyour method to a 12-period OLG model in which exogenous shocks affect the carbonintensity of energy production, as well as the damage function. Constrainedoptimal carbon taxes consist of tax rates that are simple functions ofobservables and revenue-sharing rules that guarantee that the introduction ofthe taxes is Pareto improving. This reveals that a straightforward policy ishighly effective: a Pareto-improving linear tax on cumulative emissions aloneyields a 0.42% aggregate welfare gain in consumption-equivalent terms whileadding further complexity to the tax provides only a marginal increase to0.45%. The application demonstrates that the proposed approach producesscalable tools for macro-policy design in complex stochastic settings. Beyondclimate economics, the framework offers a template for systematically analyzingwelfare-improving policies in various heterogeneous-agent problems.

Lieu

Bâtiment: Uni Mail

Boulevard du Pont-d'Arve 40
1205 Geneva

Room M 5220, 5th floor

entrée libre