AQuA: Recursively Self-Improving Quantitative Trading Research Agents
MengdiWang10 · x · 2026-08-15
This post introduces AQuA, a system for recursively self-improving quantitative trading research agents, based on a new paper. AQuA comprises two separate LLM-driven research systems: one for symbolic factor discovery and another for trainable model development. The systems do not share agents, memories, or candidate spaces; instead, each independently closes its own research loop by retaining validated evidence to guide subsequent proposals, achieving recursive self-improvement at the research process level. Experiments show the factor system achieved a combined Information Coefficient (IC) of about 0.190 on a crypto universe, while the model system achieved a per-stock IC of +0.0843 on US equities, converting it into a strategy with a Sharpe ratio of up to +2.50 after costs.
More from coding & agent
- Tutorial: Build Apps 10x Faster with Lovable MCP + Claude — MyCreativeOwls · 2026-08-15
- User Calls Claude Code a Loser, Recommends Codex for Long-Running Jobs — burkov · 2026-08-15
- Codex Desktop app adds 'edit' feature for fine-tuning LaTeX and Markdown — aniketapanjwani · 2026-08-15
- Gemini CLI Fix Preserves Subagent Termination Reason During Recovery — joneba-google · 2026-08-15
- Can AI-Generated Code Scale? Cosmos DB Demo Tests Agent Performance Under Load — adnan_hashmi · 2026-08-15
- Yukon accelerates open research with intelligent agents — gajesh · 2026-08-15