QuantAgents:

Towards Multi-agent Financial System via Simulated Trading

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.

Overall Framework

QuantAgents comprises four specialized agents, each contributing to different aspects of fund management:

  • Otto, the Manager: Responsible for executing decisions.
  • Bob, the Simulated Trading Analyst: In charge of testing strategies.
  • Dave, the Risk Control Analyst: Evaluates risks associated with various investment decisions.
  • Emily, the Market News Analyst: Provides comprehensive market reports.

These agents collaborate by participating in various meetings to assist manager Otto in decision-making. Among these meetings, market analysis meetings are held weekly to produce comprehensive market reports, while strategy analysis meetings also occur weekly, focusing on enhancing investment strategies through simulated trading. Additionally, risk alert meetings are convened as needed.

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).

Leaderboard on NASDAQ-100 Financial Dataset

Leaderboard for evaluating various methods on 9 metrics in the NASDAQ-100 Financial Dataset.

# Models Categories Source ARR(%) TR(%) SR CR SoR MDD(%) Vol(%) ENT ENB
1 QuantAgents 🥇 LLM-based 👑️ Link 58.68 299.55 3.11 11.38 66.94 16.86 1.43 2.97 1.49
2 FinAgent 🥈 LLM-based 👑️ Link 45.31 206.83 2.25 6.98 47.66 38.48 2.92 2.71 1.38
3 FinMem 🥉 LLM-based 👑️ Link 37.73 161.25 1.89 6.16 43.02 40.19 2.82 2.25 1.24
4 FinGPT LLM-based 👑️ Link 36.71 155.52 1.66 6.34 42.31 37.99 2.83 1.94 1.21
5 AlphaMix+ RL-Based 🎮 Link 32.51 132.72 1.49 5.76 30.66 40.71 2.85 2.76 1.36
6 DeepTrader RL-based 🎮 Link 32.06 130.29 1.27 7.16 30.31 29.16 2.81 1.88 1.19
7 SAC RL-based 🎮 Link 22.14 82.23 0.84 2.99 23.63 40.13 2.85 1.49 1.11
8 MV Rule-Based 🛠 Link 11.30 37.87 0.72 3.27 22.05 64.15 5.79 1.01 1.02
9 TSM Rule-Based 🛠️ Link 5.68 18.02 0.64 3.11 17.27 58.36 5.65 1.03 1.07
10 ZMR Rule-Based 🛠 Link 4.19 13.1 0.63 2.52 18.43 72.89 5.82 1.43 1.09

🚨Evaluation Metrics: ARR: Annual Return Rate, TR: Total Return, SR: Sharpe Ratio, CR: Calmar Ratio, SoR: Sortino Ratio, MDD: Maximum Drawdown, Vol: Volatility, ENT: Entrop, ENB: Effect Number of Bets.

🚨 The Multi-Asset Financial Dataset comprising Bitcoin, foreign exchange, and the Dow Jones component stocks. These data were sourced from reputable financial databases, namely Yahoo Finance and the Alpaca News API. The dataset spans from January 1, 2015, to December 31, 2023, encompassing daily data points such as open, high, low, and close prices, as well as volume and adjusted close prices. Additionally, daily news updates and 60 standard technical analysis indicators are included for each asset .


Cumulative Returns Comparison of our QuantAgents and all baselines.

QuantAgents Meeting Mechanisms Video Demonstrations

To thoroughly examine our framework's cognitive processes during execution, we have employed visualizations to elucidate multi-agent collaborative processes. This section presents three video demonstrations highlighting the collaborative nature of key meetings within QuantAgents.

Market Analysis Meeting: This video illustrates the collaborative process where agents analyze market trends, industry dynamics, and individual stock performances to inform investment decisions.

aaa


Strategy Development Meeting: In this video, the agents work together to develop and refine trading strategies. The meeting includes evaluations through simulated trading to ensure that the strategies are both effective and resilient.

aaa


Risk Alert Meeting:This video demonstrates the team’s collaborative efforts to monitor portfolio risk and evaluate strategies based on risk forecasts. The goal is to ensure the system's resilience and adaptability to market fluctuations.

aaa

Algorithms


The following content outlines the core algorithms that constitute the decision-making fabric of the QuantAgents Coordination Framework.Below is provided a detailed examination of each algorithm through pseudocode, elucidating the procedural steps and decision-making criteria that govern the collaborative interactions among agents within the QuantAgents system.

Agent

The following content provides an exhaustive outline of the QuantAgents team member profiles, which are integral to our investment decision-making simulation. The profiles are articulated using XML, chosen for its flexibility and robustness in structuring and representing complex data. XML's self-descriptive nature and the ability to define custom tags make it an ideal choice for encoding the intricate details of each agent's profile. It allows for a high degree of customization and scalability, which is essential for simulating a dynamic and complex environment such as a hedge fund's investment strategy.