---
title: FactorBench: A Portfolio-Aware Benchmark for Automated Factor Mining
url: https://www.ml-quant.com/papers/arxiv/2610.06947/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-10-09
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2610.06947
source_url: https://arxiv.org/abs/2610.06947
featured: 2026-10-09
citations: unknown
topic: Asset Pricing & Factors
---


# FactorBench: A Portfolio-Aware Benchmark for Automated Factor Mining

A portfolio-aware benchmark comparing five thousand factors from nine automated mining methods across five equity markets finds that no discovery paradigm consistently dominates in signal quality or portfolio performance.

- Source: https://arxiv.org/abs/2610.06947
- Identifier: arXiv:2610.06947
- Released: 2026-10-07
- First featured: Quant Letter No. 134 (2026-10-09): https://www.ml-quant.com/issues/2026-10-09/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Asset Pricing & Factors
- Authors: Zhuohan Wang, Carmine Ventre

## Abstract (arXiv, CC0)

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.

## Related

- [AlphaPADI: Formulaic Alpha Discovery via Pool-Aware Hierarchical Discrete Diffusion](https://www.ml-quant.com/papers/arxiv/2610.04959/): Introduces a hierarchical discrete diffusion framework that generates pools of formulaic alphas by reconstructing complete candidate pools under current context and maximizing joint predictive performance and inner diversity.
- [An Auditable Public-Data U.S. Equity Research Pipeline: Point-in-Time Factors, Asset Pricing, Portfolio Construction, and Walk-Forward Machine Learning](https://www.ml-quant.com/papers/ssrn/7528419/): A reproducible pipeline for factor research finds no model reliably beats simple approaches after accounting for transaction costs in point-in-time factor tests.
- [Corporate bond pricing in the AI era](https://www.ml-quant.com/papers/repec/bdi-opques-qef-1057-26/): Using ChatGPT's launch as a natural experiment, the paper finds that hyperscalers saw borrowing costs decline while software firms faced worse terms as debt markets repriced AI winners and losers.
- [US Tariff Policy: Chaos Order](https://www.ml-quant.com/papers/ssrn/5232092/): Chaos Order: The unpredictability of U.S. tariff policy under President Trump's second administration aligns with momentum-style investing and reinforcement learning, suggesting a dynamic and reactive policy process.
- [A multi-factor market-neutral investment strategy for New York Stock Exchange equities](https://www.ml-quant.com/papers/arxiv/2412.12350/): The report outlines a market-neutral investment strategy for NYSE equities, demonstrating superior performance with risk parity.
- [European Asset Pricing with Generative AI](https://www.ml-quant.com/papers/ssrn/4715376/): The study questions traditional factor models in the European stock market, proposing an AI asset pricing model that considers a wider range of factors, indicating a more intricate risk-sharing mechanism.