Learning by Playing: A Stock Market Simulation Game With Deep Reinforcement Learning-powered NPCs

Abstract

The research explored Deep Reinforcement Learning (DRL) to enhance educational simulation games for business education, focusing on the stock market. The key challenge was integrating real-world data into an engaging learning experience. A playable game simulation mirroring the real stock market was developed. This safe and controlled environment empowered players to experiment with investment strategies and gain market insights. The research pursued two objectives: (1) creating engaging experiences with DRL-powered Non-Player Characters (NPC) and (2) designing progressively challenging scenarios. The first objective leveraged the Octalysis Framework to seamlessly integrate market data into gameplay through NPC design and behavior. Core Octalysis drives fostered connections between real-world data and game mechanics, creating engaging learning experiences. The second objective explored the DRL model's potential for continuously improving NPC decision-making. The focus was on the DRL model's ability to learn and adapt, resulting in a progressively more intelligent NPC system that reflects market dynamics. The DRL training explored suitable reward and environment designs for training DRL models in an investment landscape. Back-testing results showed the potential of the trained DRL model to reveal financial indicators for real-world investment analysis. This fosters a deeper understanding of financial concepts and risk management. The research findings suggested future exploration of advanced algorithms and data integration. Incorporating data sources like news sentiment analysis or social media trends could enable NPCs to gain a more holistic understanding of market forces, ultimately equipping players with a more comprehensive perspective on market dynamics.