AQuA: Recursively Self-Improving Quantitative Trading Research Agents
A new arXiv paper (2608.12841) introduces AQuA, a system of two independent language-model-driven research agents for quantitative investment. One agent focuses on symbolic factor discovery, the other on trainable model development. They operate in sealed sandboxes with fixed data splits, features, labels, and evaluators, and do not share agents, memories, or candidate spaces. Each agent retains validated evidence from earlier experiments to guide later proposals, implementing recursive self-improvement at the research-process level. The paper is categorized as a cross-type announcement and is available at arxiv.org/abs/2608.12841.
Key facts
- AQuA comprises two separate language-model-driven research systems.
- One system handles symbolic factor discovery; the other handles trainable model development.
- The two systems do not share agents, memories, candidate spaces, or research state.
- Each system independently closes its own research loop by retaining validated evidence.
- Both systems use sealed sandboxes that fix data splits, feature and label definitions, and evaluator.
- The model can act only through constrained factor expressions or configuration diffs.
- The paper is available on arXiv with ID 2608.12841.
- The announcement type is cross.
Entities
Institutions
- arXiv