AI Stock Forecasts Underperform Even Split, ICML Paper Reveals
alejandroll10 · x · 2026-08-01
Researchers at Pusan National University published two papers on AI in finance (accepted by ICML 2026).
- Forecasts Beaten by Even Split: Tested on a 40-stock slice of the S&P 100 from 2020 to 2024, eight deep learning-based AI market forecast architectures were all beaten by a simple "even split" strategy (dividing money equally across all stocks) in return per unit of risk.
- Decision-Trained Model Wins: A model trained to choose portfolios rather than predict prices earned a higher risk-adjusted score (0.6717 vs. 0.5759). Removing the instruction to limit losses caused the biggest drop, falling to 0.5691.
- Five Evaluation Biases: A companion review of 164 finance papers using LLMs found five result-inflating flaws that quietly inflate reported results, none of which were mentioned in more than 28% of the papers.
Related event: LLMs Fail to Beat Equal-Weight Baseline in Stock Trading(2 posts)→
More from Research
- A missing parameter in scaling laws: H, the human effort behind learning signals — lateinteraction · 2026-10-16
- One ancient fold, many jobs: how the immunoglobulin domain evolved into antibodies, MHC and titin — arjunrajlab · 2026-10-16
- Researcher Argues Conferences Should Desk-Reject Far More Papers as Flooding Strains Reviews — boknilev · 2026-10-16
- Universal forgery attack breaks Chinese PQC signature Origami using only the public key — matthew_d_green · 2026-10-16
- How do you attribute agent gains: model training or infrastructure? — jrxna · 2026-10-16
- First-ever paper on multi-agent swarm safety lands NeurIPS spotlight — LauraRuis · 2026-10-16