MASS: Multi-Agent Simulation Scaling for Portfolio Construction (2505.10278v1)

Abstract: LLM-based multi-agent has gained significant attention for their potential in simulation and enhancing performance. However, existing works are limited to pure simulations or are constrained by predefined workflows, restricting their applicability and effectiveness. In this paper, we introduce the Multi-Agent Scaling Simulation (MASS) for portfolio construction. MASS achieves stable and continuous excess returns by progressively increasing the number of agents for large-scale simulations to gain a superior understanding of the market and optimizing agent distribution end-to-end through a reverse optimization process, rather than relying on a fixed workflow. We demonstrate its superiority through performance experiments, ablation studies, backtesting experiments, experiments on updated data and stock pools, scaling experiments, parameter sensitivity experiments, and visualization experiments, conducted in comparison with 6 state-of-the-art baselines on 3 challenging A-share stock pools. We expect the paradigm established by MASS to expand to other tasks with similar characteristics. The implementation of MASS has been open-sourced at https://github.com/gta0804/MASS.

An Expert Analysis of Multi-Agent Scaling Simulation for Portfolio Construction

The paper introduces a novel Multi-Agent Scaling Simulation (MASS) framework aimed at enhancing portfolio construction through LLM-based multi-agent systems. The primary focus of MASS is to address the limitations of existing simulation models which are often restricted to pre-defined workflows or purely theoretical constructs. By leveraging the scalability of LLMs, MASS provides dynamic and comprehensive market analysis aimed at optimizing portfolio returns.

The paper outlines several critical components and procedures within the MASS framework. The process begins with LLM-based initialization of multiple agents assigned diverse investment styles. This diversity allows the agents to simulate a range of market participant behaviors, contributing to a robust decision-making engine. The agents generate investment strategies tailored by macroeconomic data and specific market insights, further refined through a process called backward optimization. This optimization seeks to continuously adjust agent distributions, aligning them with optimal strategies derived from recent market data.

Experimental validation is a highlight of this research. MASS outperformed six state-of-the-art baselines across three challenging A-share stock pools: SSE 50, CSI 300, and ChiNext 100. Notably, MASS achieved IC values of 5.90% with an ICIR of 33.43 for SSE50, marking a considerable lead over other methods. The scalability of the multi-agent system is demonstrated by expanding the agent count to 512, yielding a linear growth in performance metrics such as the RankIC.

An intriguing component of MASS is its foundation in the market disagreement hypothesis, which posits that investor diversity can predict security returns. This theoretical underpinning, supported by empirical results, lends credibility to the agents' simulated decision-making. By aggregating scores from the agent population, MASS formulates a portfolio signal that balances high consensus with low investment disagreement, aiming for superior returns.

The authors also present a well-rounded set of complementary experiments to illustrate the robustness and adaptability of MASS. For instance, backtesting in the CSI 300 context showed MASS's exceptional capacity to achieve both higher cumulative returns and lower drawdowns compared to the index itself. Furthermore, ablation studies emphasize that the inclusion of candidate stock pools, macroeconomic data provision, and backward optimization are indispensable to the system's effectiveness.

The implications of MASS extend beyond portfolio management. By introducing a scaling law for multi-agent systems, this research opens avenues for applying such simulations to areas like supply chain optimization and climate modeling, where dynamic system understanding is crucial.

In the context of future AI developments, MASS exemplifies the potential of harnessing multiple LLM-driven agents working collaboratively. The transition from pre-defined workflows to more flexible, data-driven strategies signifies an advancement in AI's ability to navigate complex, uncertain environments. As MASS is open-source, further research and iterations can fine-tune the framework, enhancing its utility and expanding its application scope.

In summary, this paper presents MASS as a substantial advance in LLM-based multi-agent simulation, offering a potent tool for portfolio construction while establishing a new baseline for future investigations into agent collaboration and strategy optimization across multiple domains.