Princeton, Ant, Stanford unveil AQuA, a self-improving quant research agent
jiqizhixin · x · 2026-09-09
Princeton, Ant Group, and Stanford present AQuA, a recursively self-improving quantitative trading research agent. Results: 0.190 Spearman IC on crypto 5-minute data, +0.0843 per-stock IC on US equities 30-minute returns, +2.50 out-of-sample Sharpe after costs, positive Sharpe every year 2021–2025. The recursion never touches model weights — evolution happens in the research process via two parallel tracks (symbolic factors and trainable models), with only verified experiments entering the next round and data pipelines kept outside the loop to prevent leakage. Its memory stores why candidates were proposed, applicable market regimes, and evidence that overturned initial conclusions.
More from Research
- Science isn't declining: reanalysis says crisis claims are statistical artifacts — chris_j_paxton · 2026-09-09
- SREGym benchmark launches: GPT-5.6 leads agents at fixing real production SRE failures — tianyin_xu · 2026-09-09
- Local academic literature pipeline makes fabricated citations structurally impossible — Waste_Public_2985 · 2026-09-09
- LLM-designed antibodies hit 38/80 binders in lab tests, 85% success on RAGE — DeryaTR_ · 2026-09-09
- From 'crackpot' to plausible: AI solving Navier-Stokes flipped in months — EigenGender · 2026-09-09
- Experimenting with central pattern generators to make music from a tiny motion neuron network — Syntetisaattori · 2026-09-09