• Cournot oligopoly
• q-learning
• reinforcement learning

This thesis explores learning dynamics within economic games through reinforcement learning, specifically Q-learning. The strategic interactions analyzed refer to a Cournot oligopoly in a stochastic environment. This approach, attempts to give an innovative solution in order to simulate market behavior and evaluate potential collusion and coordination between competing firms.
The Cournot model examined is traditionally analyzed through Nash equilibria, in this thesis the model is revisited under the lens of Q-learning and different results from different learning scenarios are proposed.
Two methodologies are mainly used: Central Q-Learning (CQL) which aims at an optimization of the collective strategy of firms and Independent Q-Learning (IQL) in which each enterprise is decentralized from the others and the learning process is completely autonomous. These two methodologies are used in order to understand their effectiveness in fostering collusive behavior in the absence of explicit communication between enterprises.
The simulations show convergence to a collusive equilibrium, and this effect is more significant for the centralized model. In any case, both models show a decrease in cooperation with respect to the number of firms. The result is significant because, aside from Q-learning, there are almost no models of the learning behavior of individual economic agents that predict collusive behavior in Cournot games without relying on trigger strategies.