FactorBench: A Portfolio-Aware Benchmark for Automated Factor Mining

Paper: arXiv 2610.06947

Authors: Zhuohan Wang, Carmine Ventre

Abstract

Factor mining seeks to discover signals from financial data that predict future asset returns and guide portfolio construction. Automated factor mining now spans genetic programming, reinforcement learning, generative models, and large language model agents. Yet it remains unclear whether advances across these paradigms yield more generalizable, distinct, and economically useful financial signals. We introduce FactorBench, a portfolio-aware benchmark comparing roughly five thousand mined factors from nine automated mining methods across five equity markets. A shared data and evaluation contract supports both symbolic expressions and executable Python factors, connecting heterogeneous discovery algorithms to common signal combination and portfolio construction procedures. FactorBench traces the outputs of mining systems across three levels: factor validity, temporal generalization, and predictiveness beyond measured risk and style exposures; within- and across-method pool distinctness, including similarity to the benchmark Alpha101; and composite-signal quality and after-cost long-only and long–short portfolio performance. After systematically assessing whether advances in factor mining translate into signal quality and portfolio performance, FactorBench finds that no paradigm consistently dominates.

Complexity vs Empirical Score

• Math Complexity: 6.0/10
• Empirical Rigor: 9.0/10
• Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper introduces a highly rigorous and comprehensive benchmark for automated factor mining, addressing a critical gap in the field. While the underlying math of the factor mining methods is diverse, the benchmark itself involves significant statistical and evaluation complexity. The empirical setup across multiple markets and methods is exceptionally strong.

Research Flowchart

  flowchart TD
    A[Research Goal: Assess Automated Factor Mining Advances] --> B{Methodology: FactorBench};
    B -- Inputs --> C[Data: 5 Equity Markets, 9 Automated Mining Methods, ~5000 Factors];
    C -- Processing --> D[Computation: Factor Validity, Temporal Generalization, Predictiveness, Distinctness, Portfolio Performance];
    D -- Outcomes --> E[Key Finding: No Paradigm Consistently Dominates];

Cited References (26)

1. 101 formulaic alphas (2016)
2. Generating synergistic formulaic alpha collections via reinforcement learning — → 2306.12964
3. Alpha Discovery via Grammar-Guided Learning and Search — → 2601.22119
4. AlphaQCM: Alpha Discovery in Finance with Distributional Reinforcement Learning (2025)
5. Alphaforge: A framework to mine and dynamically combine formulaic alpha factors (2025)
6. Alphasage: Structure-aware alpha mining via gflownets for robust exploration (2026)
7. R&D-Agent-Quant: a multi-agent framework for data-centric factors and model joint optimization (2026)
8. Alphaagent: Llm-driven alpha mining with regularized exploration to counteract alpha decay (2025)

4 of 26 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.