Dear INFORMS Colleagues,
I would like to share our peer-reviewed computational model recently published in the CoMSES Computational Model Library:
Integrating Reinforcement Learning in Agent-Based Modeling for Dynamic Investment Decisions
https://doi.org/10.25937/644j-cv09
The model explores the integration of Reinforcement Learning (RL) and Agent-Based Modeling (ABM) to study dynamic investment behavior in complex and uncertain environments. Specifically, investor agents employ Proximal Policy Optimization (PPO) to learn and adapt their investment strategies over time rather than relying on fixed behavioral rules.
The model extends the classic investment ABM introduced by Volker Grimm and Steven F. Railsback and serves as a proof-of-concept for incorporating modern AI methods into agent-based decision systems.
One particularly interesting observation was that learning agents not only improved their own performance but also contributed to healthier system-level outcomes. Compared with earlier model variants, PPO-enabled agents achieved higher wealth accumulation, improved utility, lower risk exposure, fewer business failures, and a greater number of flourishing investment opportunities.
This work is part of a broader research agenda investigating how reinforcement learning can enhance adaptive decision-making within agent-based systems. A companion study currently under review compares Q-Learning, Deep Q Networks (DQN), and PPO using extensive NetLogo BehaviorSpace experiments, mathematical modeling, and ANN-based recommendation capabilities.
The model is openly available, and I would welcome feedback from researchers working in simulation, analytics, reinforcement learning, decision intelligence, complex adaptive systems, and related areas.
I am particularly interested in connecting with researchers exploring the integration of AI and machine learning techniques within agent-based models and simulation-based decision support systems.
Many thanks to my co-authors, Haider Ali and Hafiz Mohammad, whose contributions were instrumental to this work.
Best regards,
------------------------------
Muhammad Khurram Ali
Associate Professor
University of Engineering and Technology, Taxila, Pakistan
Taxila
LinkedIn: www.linkedin.com/in/muhammad-khurram-ali-ied
------------------------------