Presentation Information
[4Yin-A-19]Spatial Pricing Strategies of LLM Agents under Heterogeneous Customer Distributions
〇Akio Ikegami1,2, Takehiro Takayanagi1,2, Ryuji Hashimoto1,2, Ryosuke Takata1,2, Kiyoshi Izumi1,2 (1. The University of Tokyo, 2. Simulacra Inc.)
Keywords:
Large Language Models,Spatial Competition,Adaptive Pricing Strategy
Recent research applies Large Language Models (LLMs) to economic decision-making and social simulation, yet their adaptation to “spatial elements” such as physical distance to customers and heterogeneous distributions which are essential to real-world market activities, remains underexplored. To address this gap, we construct a spatial multi-agent simulation with LLM-driven firm agents. We investigate whether LLMs can adapt to spatial characteristics and rationally execute critical corporate decisions, such as pricing and inventory management, across heterogeneous customer distributions. Our results show that models with high reasoning capabilities demonstrate flexible strategic adaptability aligned with spatial contexts. Specifically, they maximize profits by exploiting locational monopoly power in an advantageous scenario and lowering prices to capture market share in a disadvantageous scenario. Conversely, some general-purpose models fail to accurately forecast demand, leading to bankruptcy due to overproduction. This study demonstrates that LLM reasoning performance critically influences the ability to recognize spatial competitive environments and execute strategic economic decisions.