FactorMiner: A Self-Evolving Agent with Skills and Experience Memory for Financial Alpha Discovery

TL;DR

FactorMiner addresses the challenge of discovering interpretable, formulaic alpha factors in a vast search space by introducing a memory-augmented, modular agent framework. It combines a composable Factor Mining Skill with structured Experience Memory, enabling a Ralph Loop of retrieve, generate, evaluate, and distill to continually evolve the factor library from a global perspective. Across A-share and Crypto datasets, FactorMiner demonstrates competitive, low-redundancy factor libraries (110 factors) and robust cross-market performance, aided by GPU-accelerated evaluation to scale the search. The work offers a practical, interpretable path to scalable alpha discovery and provides a reproducible artifact for hypothesis-driven market microstructure analysis, with limitations and ethical considerations highlighted for responsible use.

Abstract

Formulaic alpha factor mining is a critical yet challenging task in quantitative investment, characterized by a vast search space and the need for domain-informed, interpretable signals. However, finding novel signals becomes increasingly difficult as the library grows due to high redundancy. We propose FactorMiner, a lightweight and flexible self-evolving agent framework designed to navigate this complex landscape through continuous knowledge accumulation. FactorMiner combines a Modular Skill Architecture that encapsulates systematic financial evaluation into executable tools with a structured Experience Memory that distills historical mining trials into actionable insights (successful patterns and failure constraints). By instantiating the Ralph Loop paradigm -- retrieve, generate, evaluate, and distill -- FactorMiner iteratively uses memory priors to guide exploration, reducing redundant search while focusing on promising directions. Experiments on multiple datasets across different assets and Markets show that FactorMiner constructs a diverse library of high-quality factors with competitive performance, while maintaining low redundancy among factors as the library scales. Overall, FactorMiner provides a practical approach to scalable discovery of interpretable formulaic alpha factors under the "Correlation Red Sea" constraint.

Figures (10)

• Figure 1: FactorMiner System Architecture. The Ralph Loop framework integrates three key components: (1) Experience Memory that stores successful patterns and forbidden regions from past mining sessions; (2) Agent Skill that encapsulates the multi-stage validation pipeline (IC screening, correlation checking, deduplication, and full validation); (3) Factor Library that grows dynamically while maintaining orthogonality constraints. The agent iteratively retrieves memory priors, generates candidates through the skill, and distills outcomes back into memory for improved future exploration.
• Figure 2: Pairwise Spearman correlation heatmap of the released A-share factor library (110 admitted factors), computed from cross-sectionally standardized realized factor signals over the common time--asset panel. The average off-diagonal absolute correlation is Avg $|\rho|$ = 0.203.
• Figure 3: Ablation comparison between Have Memory and No Memory. High-quality candidates are defined as those passing the IC threshold ($|\text{IC}|>0.02$). The bar chart reports the counts (high-quality / rejected / admitted) and the corresponding yield and rejection rates.
• Figure 4: Grouped bar chart of computation time on a log scale for operator-level and factor-level benchmarks. Lower is better; GPU shows consistent order-of-magnitude gains.
• Figure 5: IC time-series analysis for three combination methods. All methods show stable positive IC throughout the evaluation period, with IC-weighted exhibiting slightly higher peaks.