Integrating Reinforcement Learning into Real Business Cycle Models: A Computational Approach to Macroeconomic Dynamics

Baca Ordóñez F.; Ramos Chuquiure J.D.C. · Proceedings of 8th International Conference on Systems Engineering - Cybersecurity and AI: Building a reliable digital future, CIIS 2025 (2025) · DOI: 10.1145/3771678.3771680

Resumen

This paper examines the implementation of Reinforcement Learning (RL) algorithms, specifically Q-learning, to solve Real Business Cycle (RBC) macroeconomic models. Unlike traditional approaches based on dynamic programming or log-linearization around steady states, this work proposes the implementation of an agent that learns optimal consumption and labor decisions through interaction with a stochastic environment that alternates between wage states. It is shown that the RL agent is capable of converging to intertemporal policies consistent with the principles of an RBC model, replicating stylized facts such as procyclical consumption and consumption smoothing. The results demonstrate that RL provides a viable tool for addressing the complexity and nonlinearity of modern macroeconomic models without the need to impose strong functional assumptions or analytically solve equilibrium conditions. This work contributes to the intersection between computational intelligence and macroeconomics by showing that machine learning techniques can facilitate the solution and analysis of sophisticated theoretical models, offering new perspectives for economic policy design.

Instituciones: Universidad del Pacífico