Self-Evolving Multi-Agent Symbolic Discovery for Financial Fundamental Analysis

By Kelvin J.L. Koa, Filip Orestav, Shengqiong Wu, Michael J. Wooldridge, Ke-Wei Huang

Rating

1500
Battle Count: 0

Relevance

9/10
Highly relevant as it provides interpretable, data-driven valuation formulas that can be used for stock selection, factor investing, and understanding market regimes. The ability to adapt weights based on market conditions (bull/bear) is directly applicable to dynamic trading strategies.

Implementation Complexity

8/10
High complexity due to the hierarchical multi-agent architecture, integration of LLMs with symbolic regression tools, differential evolution for coefficient fitting, and the need for structured statistical memory management.

Reproducibility

4/5
The paper provides detailed algorithms, implementation specifics (vLLM, Qwen3-4B), and releases discovered equations, distilled learnings, and context weights. However, the specific Bloomberg/FactSet data access is proprietary, though the methodology is fully described.

About this paper

Methodology: mufasa (Multi-Agent Fundamental Analysis with Symbolic Adaptive learning). Problem types: Regression, Time Series Forecasting, Symbolic Discovery, Optimization.

The interactive Everscope explorer (charts, battles, favorites) loads below.