FactorBench: New benchmark pits 9 AI factor-mining methods across 5 markets, finds no winner
PtrPomorski · x · 2026-10-10
A new arXiv paper (Zhuohan Wang, Carmine Ventre) introduces FactorBench, a portfolio-aware benchmark for automated factor mining spanning genetic programming, RL, generative models, and LLM agents. Key points: - Compares 5,000 mined factors from 9 automated methods across 5 equity markets; - A shared data/evaluation contract supports both symbolic expressions and executable Python factors, connecting heterogeneous algorithms to common signal combination and portfolio construction; - Evaluates factor validity, temporal generalization, predictiveness beyond risk/style exposures, pool distinctness (incl. similarity to Alpha101), and after-cost long-only/long-short portfolio performance; - Conclusion: no paradigm consistently dominates in signal quality or portfolio performance.
More from Research
- Eric Topol: 194-year-old land animal and flawed epigenetic clocks — EricTopol · 2026-10-11
- CMU AI Agents course releases Deep Research lecture: building, evaluating and RL-training research agents — AkariAsai · 2026-10-11
- AI health advice carries invisible risks no institution can track, Nature Health paper argues — EricTopol · 2026-10-11
- Hugging Face launches a PyTorch profiling series: from torch.profiler to attention — ariG23498 · 2026-10-11
- Oxford Studies in Philosophy of AI and Computing series to launch, CFP coming soon — sethlazar · 2026-10-11
- Researcher claims new method now beats Shampoo optimizer on batch size too — _arohan_ · 2026-10-11