Published April 5, 2026 | Version v1
Publication Open

AQAA: An Always-On Autonomous Quant AI Agent for Continuous Alpha Discovery, Self-Optimization, and Live Trading

Authors/Creators

Description

Autonomous AI agents for quantitative trading have advanced rapidly on individual components — alpha mining, portfolio
optimization, and execution — yet no system integrates these into a continuously operating, fully autonomous closed loop. We introduce
AQAA (Always-On Quant AI Agent), an end-to-end autonomous system with five tightly coupled modules: a Hypothesis Engine for
continuous LLM-driven alpha discovery; a Code Synthesis and Backtesting Agent with a Temporal Integrity Framework (TIF) that
eliminates LLM lookahead bias; an Adaptive Portfolio Optimizer with regime-aware Kelly sizing; a Live Execution and Monitoring
Agent with Bayesian IC decay detection and autonomous retirement; and a Feedback Loop Orchestrator using multi-armed bandit
compute scheduling. Over a 5-year backtest (2020–2025) and a 12-month live deployment (2025), AQAA achieves 42.7% annualized
return and Sharpe 1.89 — a 36.9% relative improvement over the strongest baseline. Our TIF prevents 14.3 percentage points of
spurious ARR inflation from LLM memorization bias, a finding with broad reproducibility implications.

Files

AQAA_KDD2026_v4.pdf

Files (1.7 MB)

Name Size Download all
md5:5e7b44e868407f12e16b8c2e8f395f4c
1.7 MB Preview Download

Additional details

Software

Repository URL
https://github.com/Thanh-Van-2001/An-Always-On-Autonomous-Quant-AI-Agent-for-Continuous-Alpha-Discovery-Self-Optimization
Programming language
Python
Development Status
Active

References

  • Auer, P., Cesa-Bianchi, N., and Fischer, P. 2002. Finite-time analysis of the multiarmed bandit problem. Machine Learning 47, 2, 235–256.
  • Cui, C., et al. 2021. AlphaEvolve: A learning framework to discover novel alphas in quantitative investment. In Proc. ACM SIGMOD, 2208–2216.
  • Page, E. S. 1954. Continuous inspection schemes. Biometrika 41, 1–2, 100–115.
  • Glasserman, P. and Lin, C. 2023. Assessing look-ahead bias in stock return predictions generated by GPT sentiment analysis. arXiv:2309.10959
  • Gu, S., Kelly, B., and Xiu, D. 2020. Empirical asset pricing via machine learning. Review of Financial Studies 33, 5, 2223–2273.
  • Killick, R., Fearnhead, P., and Eckley, I. A. 2012. Optimal detection of changepoints with linear cost. J. American Statistical Association 107, 500, 1590–1598.
  • Ruan, Y., et al. 2025. Look-Ahead-Bench: A standardized benchmark for look-ahead bias in financial LLMs. arXiv preprint (2025).
  • Lopez-Lira, A., Tang, Y., and Zhu, M. 2025. LLM memorization and its implications for financial return prediction. Working Paper.
  • Li, Y., et al. 2025. FinMem: A performance-enhanced LLM trading agent with layered memory. IEEE Transactions on Big Data.