Research map / Multi-agent systems and games / LLM agents
MarketSim: Simulating Stock Markets with Large-Scale Generative Agents
Jinghua Piao, zhentao liu, Cheng Huang, Jiarui Huang, Songwei Li, Ranran Wang, Yong Li
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MarketSim, a stock-market simulator built on large-scale LLM-based generative agents, aiming to jointly capture structural and behavioral fidelity of market dynamics.
Problem
Existing stock market simulators face a trade-off between structural fidelity (realistic market mechanics and price formation) and behavioral fidelity (realistic participant behavior among heterogeneous agents).
Approach
Uses large-scale generative (LLM-based) agents to simulate heterogeneous market participants interacting in an evolving information landscape, so that prices emerge from decentralized interactions rather than being imposed.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use empirically grounded agent profiles instead of simplistic archetypes (ablation) | MarketSim · JNJ tariff shock | MAPE: 0.82% to over 2.8% | Improved |
| add self-reflection to manager agents (ablation) | MarketSim · stock simulation ablation | RMSE: increases RMSE by over 10x | Improved |
| add self-reflection to manager agents (ablation) | MarketSim · stylized facts | stylized fact reproduction | Improved |
| use a stronger LLM for the manager (ablation) | MarketSim · stock simulation ablation | price fidelity | Improved |
| allow dynamic strategy allocation by the manager (ablation) | MarketSim · stock simulation ablation | price fidelity | Improved |
| use MarketSim instead of strongest alternative baselines | MarketSim · all stocks and scenarios | RMSE: 11%-41% | Improved |
| use MarketSim instead of strongest alternative baselines | MarketSim · all stocks and scenarios | MAPE: 24%-39% | Improved |
| use MarketSim instead of strongest alternative baselines | MarketSim · all stocks and scenarios | volatility similarity: 15%-43% | Improved |
| use ABM Transformer instead of MarketSim | ABM baseline · JNJ tariff shock | DTW distance: 0.011 -> 0.005 | Hurt |
| integrate predictive models into ABIDES instead of using them standalone | Linear · JNJ tariff shock | Q-Q correlation: 0.892 -> 0.988 | Improved |
| compare agent decisions against matched control decisions | MarketSim agents · human evaluation | market consistency t-test: t = 7.09, p < 0.001 | Improved |
| add 200 momentum-based agents | MarketSim · disruption experiment | market volatility | Mixed |
Closest papers by meaning
- MarS: a Financial Market Simulation Engine Powered by Generative Foundation Model (ICLR 2025 · similarity 0.90)
- TwinMarket: A Scalable Behavioral and Social Simulation for Financial Markets (NeurIPS 2025 · similarity 0.87)
- Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (AAAI 2024 · similarity 0.87)
- Trade in Minutes! Rationality-Driven Agentic System for Quantitative Financial Trading (ICLR 2026 · similarity 0.86)
- R&D-Agent-Quant: A Multi-Agent Framework for Data-Centric Factors and Model Joint Optimization (NeurIPS 2025 · similarity 0.86)
- Towards Financial World Modeling (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
