AQuA: Recursively Self-Improving Quantitative Trading Research Agents

2026-08-13

AI Summary

This study investigates recursive self-improvement in quantitative investment research: whether a system can automatically refine subsequent hypotheses and strategies based on prior validation outcomes. We introduce two independent language model–driven systems—one for symbolic factor discovery and another for trainable model development—that operate in parallel within isolated sandboxes, sharing no internal state but iteratively guided solely by preserved validation evidence. This dual-track recursive framework, the first of its kind in quantitative research, integrates multi-agent symbolic mining, hybrid temporal architectures, and configuration-driven optimization, with robustness ensured through information coefficients and sealed backtesting. Empirically, the factor system achieves a portfolio information coefficient of 0.190 on crypto assets, while the model system attains an individual stock information coefficient of +0.0843 on U.S. equities, corresponding to a long–short strategy with an annualized Sharpe ratio up to 2.50 and consistently positive returns from 2021 to 2025.

Abstract

We study recursive self-improvement at the level of quantitative-investment research: whether an autonomous system can use evidence from earlier experiments to improve the hypotheses and candidates proposed in later iterations. We present AQuA, which comprises two separate language-model-driven research systems: one for symbolic factor discovery and one for trainable model development. The two systems do not share agents, memories, candidate spaces, or research state. Instead, each independently closes its own research loop by retaining validated evidence and using it to guide subsequent proposals. In this bounded sense, both systems implement recursive self-improvement at the level of the research process. Each system also uses its own sealed sandbox, which fixes the data splits, feature and label definitions, and evaluator while allowing the model to act only through constrained factor expressions or configuration diffs. The factor system, a manager-mediated multi-agent pipeline, discovers and combines factors into a signal that reaches a combined information coefficient of about $0.190$ on a crypto universe. The model system, a config-driven loop over a hybrid time-series architecture, reaches a per-stock information coefficient of $+0.0843$ on US equities and converts it into a threshold long/short strategy with a held-out Sharpe of up to $+2.50$ at a two-leg cost. The strategy is positive in every year from 2021 to 2025.

Problem

• recursive self-improvement
• quantitative trading
• autonomous research
• hypothesis generation
• evidence-based iteration

Innovation

• recursive self-improvement
• quantitative trading
• language-model-driven research
• factor discovery
• sandboxed evaluation