QuantaAlpha: LLM-Driven Alpha Mining with Evolutionary Framework
A new framework called QuantaAlpha has been proposed for alpha mining in financial markets, addressing noise and non-stationarity. It treats mining runs as trajectories and applies mutation and crossover operations to improve factors. The framework localizes suboptimal steps for targeted revision and recombines high-reward segments to reuse effective patterns. It enforces semantic consistency across hypothesis, factor expression, and executable code while constraining complexity and redundancy. The approach enables structured exploration and refinement across iterations, aiming to overcome limitations of existing agentic frameworks that lack controllable multi-round search and reliable experience reuse.
Key facts
- QuantaAlpha is an evolutionary alpha mining framework for financial markets.
- It addresses noise and non-stationarity in financial data.
- Each mining run is treated as a trajectory.
- It uses mutation and crossover operations to improve factors.
- Suboptimal steps in trajectories are localized for targeted revision.
- Complementary high-reward segments are recombined to reuse effective patterns.
- Semantic consistency is enforced across hypothesis, factor expression, and code.
- The framework constrains complexity and redundancy in factor generation.
Entities
—