[2Yin-A] Poster 3

[2Yin-A-04] EconTradeAgent: Macroeconomic Simulation based on LLM-powered Agents and International Trade Theory

〇Terufumi Morishita¹ (1. Hitachi)

Keywords:

LLM,economics,agent,simulation,agent based model

International trade is an important phenomenon that can improve living standards across countries. To elucidate the mechanisms of international trade, we propose an LLM-based simulation environment, EconTradeAgent. The environment consists of two components: (1) LLM agents that make human-like decisions, and (2) a dynamic model grounded in international trade theory that governs interactions among agents. Through a 10-year simulation with 100 agents, we demonstrate that EconTradeAgent can reproduce key trade phenomena. We also run a counterfactual scenario, i.e., the outbreak of a “tariff war", to derive implications. Such open-ended and unconstrained scenarios become possible for the first time with LLMs that can act effectively even under previously unseen conditions.