QuantaAlpha: An Evolutionary Framework for LLM-Driven Alpha Mining

QuantaAlpha is an evolutionary alpha mining framework that treats each end-to-end mining run as a trajectory and improves factors via trajectory-level mutation and crossover.

It addresses limitations in existing agentic systems, such as fragile controllability, limited trustworthiness, and constrained exploration, by localizing suboptimal steps for targeted revision and recombining complementary high-reward segments.

The framework enforces semantic consistency across hypothesis, factor expression, and executable code, and constrains complexity and redundancy to mitigate crowding.

Experiments on CSI 300 show consistent gains over baselines, achieving an IC of 0.0472, ARR of 4.68%, and MDD of 11.8% with GPT-5.2.

Factors mined on CSI 300 transfer effectively to CSI 500 and S&P 500, delivering about 40.28% and 19.1% cumulative excess return over four years, respectively, indicating robustness under market distribution shifts.

Ablation studies confirm the importance of diversified planning initialization, mutation, crossover, and consistency/complexity/redundancy controls.

The framework maintains performance through regime shifts, such as the 2023 A-share market transition, by discovering structural factors like Mean-Reverting Range Deviation and Overnight Gap Structure.

QuantaAlpha also shows sample efficiency, with performance stabilizing after early iterations and sustaining improvement up to around 15 iterations before diminishing returns.