Deep Reinforcement Learning For Nash Equilibria in Non-Renewable Resource Differential Games

Characterizing Nash equilibria in oligopolistic non-renewable resource markets raises important challenges for computational economics, as traditional iterative methods suffer from scalability issues due to the curse of dimensionality. In this work, we implement and evaluate a machine learning approach for computing these equilibria based on a reinforcement learning algorithm. The proposed method is compared with an iterative baseline to assess both performance and computational efficiency. We conduct experiments in monopoly, duopoly, and multi-player settings, evaluating reward accuracy and scalability. Our results indicate that, while iterative schemes achieve good accuracy in low-dimensional problems, reinforcement learning scales more effectively to three- and four-player games, yielding a substantial reduction in computation time.