EconGrowthAgent: Economic Growth Simulation based on LLM Agents and Growth Theory
Its authors withdrew this paper, so there is no full text to read or summarize.
Lists
Only you see this
Abstract
Economics has long sought to understand economic phenomena to enrich human society, yet it has been constrained by being "a discipline where experiments cannot be conducted'' due to ethical and practical limitations. While Agent-Based Models (ABM) offer a computational alternative, modeling complex human decision-making remains fundamentally challenging. Recent large Language Models (LLMs) provide new capabilities for this modeling. We present $\textbf{EconGrowthAgent}$, the first LLM-based ABM that simulates economic growth---the most essential phenomenon in economics. Our approach decomposes macroeconomic growth theory into a micro-level dynamic model, where decisions of LLM-based economic agents interact and evolve. Through 25-year simulations with 100 agents, we demonstrate that EconGrowthAgent reproduces economic growth and related key macro phenomena while enabling seamless micro-to-macro analysis. We further demonstrate its value by simulating a counterfactual scenario difficult to explore in reality -- an approaching civilization-ending asteroid. As a "laboratory for economic experiments," EconGrowthAgent advances our understanding of economic phenomena and propels economics forward.
BibTeX
@misc{morishita2026econgrowthagent,
author = {Terufumi Morishita and Atsuki Yamaguchi and Masaya Tsunokake and Gaku Morio and Yasuhiro Sogawa},
title = {{EconGrowthAgent: Economic Growth Simulation based on LLM Agents and Growth Theory}},
year = {2026},
url = {https://openreview.net/forum?id=mMdnPMzeN2},
howpublished = {OpenReview},
note = {Submitted to ICLR 2026},
}