QuantAgents: Towards Multi-agent Financial System via Simulated Trading

TL;DR

QuantAgents addresses the gap between current LLM-based financial agents and real-world fund management by embedding simulated trading into a four-agent, meeting-driven framework coordinated by a manager. It introduces a dual reward mechanism that blends real-market performance with predictive accuracy in simulated trading, driving forward-looking investment decisions. Empirical results demonstrate superior profitability and risk management across nine financial metrics, with a cumulative return near 300% over three years and strong live-trading performance in diverse markets. The work provides a practical, reproducible approach with detailed methodology and datasets to advance forward-looking, multi-agent quantitative investing.

Abstract

In this paper, our objective is to develop a multi-agent financial system that incorporates simulated trading, a technique extensively utilized by financial professionals. While current LLM-based agent models demonstrate competitive performance, they still exhibit significant deviations from real-world fund companies. A critical distinction lies in the agents' reliance on ``post-reflection'', particularly in response to adverse outcomes, but lack a distinctly human capability: long-term prediction of future trends. Therefore, we introduce QuantAgents, a multi-agent system integrating simulated trading, to comprehensively evaluate various investment strategies and market scenarios without assuming actual risks. Specifically, QuantAgents comprises four agents: a simulated trading analyst, a risk control analyst, a market news analyst, and a manager, who collaborate through several meetings. Moreover, our system incentivizes agents to receive feedback on two fronts: performance in real-world markets and predictive accuracy in simulated trading. Extensive experiments demonstrate that our framework excels across all metrics, yielding an overall return of nearly 300% over the three years (https://quantagents.github.io/).

Figures (11)

• Figure 1: Our method has surpassed all baselines on the PRUDEX sun2023prudexcompass benchmark.
• Figure 2: The workflow of QuantAgents, which is equipped with 26 tools, 3 types of memory to execute 10 actions. Furthermore, three meetings, i.e., market analysis, strategy development, risk alert meeting will assist in decision-making (e.g., buy).
• Figure 3: Our framework has been optimized to obtain rewards from both simulated and real-world trading.
• Figure 4: Cumulative Returns Comparison.
• Figure 5: Ablation analysis on several LLM backbones, from open-source to closed-source models. The numbers presented in the figure have been normalized and converted into percentage values.