Core Insights - The article discusses the QuantaAlpha framework, which aims to combine the strengths of genetic programming and large language models (LLMs) for effective alpha factor mining in quantitative finance [1][2][35] - QuantaAlpha introduces a trajectory-based self-evolution paradigm that allows for adaptability in volatile market conditions, marking a significant advancement in AI's role in financial research [2][35] Group 1: Framework Overview - QuantaAlpha's core innovation is viewing the factor mining process as a complete trajectory, enhancing resilience during market shifts [2] - The framework emphasizes trajectory-level evolution rather than single-instance success, focusing on systematic exploration and logical integrity [3] Group 2: Methodology - The framework employs diversified planning initialization to mitigate factor crowding from the outset, generating multiple distinct research paths [5] - It utilizes mutation to target and correct specific decision failures in non-stationary markets, allowing for local refinement while preserving effective components [6][7] - Crossover operations are designed to reuse successful experiences by identifying and recombining high-value segments from different trajectories [8][9] - Structured constraints are implemented to prevent semantic drift and ensure that generated factors maintain economic logic and clarity [10][11] Group 3: Case Study - The evolution of a specific factor, starting from a reversal logic to a nested momentum alignment, demonstrates the framework's ability to adapt and improve performance metrics [17][20] - The final factor, Institutional Momentum Score 20D, integrates insights from both institutional momentum and retail herding, showcasing the framework's capacity for logical synthesis [23][24] Group 4: Performance Metrics - The framework's predictive power is evidenced by an information coefficient (IC) of 0.1501 and an annualized excess return (ARR) of 27.75%, with a maximum drawdown (MDD) of only 7.98% [28][29] - Factors derived from QuantaAlpha have shown strong out-of-distribution (OOD) transferability, generating cumulative excess returns of 160% and 137% on CSI 300 and CSI 500, respectively [31] Group 5: Resilience Testing - QuantaAlpha has maintained high signal strength during market volatility, outperforming traditional factor libraries by adapting to new market dynamics [33]