Researchers led by Zhi Yang (a leading scientist in financial AI) introduced QuantaAlpha, an evolutionary framework that treats the entire process of finding profitable trading signals as a "trajectory." Instead of starting from scratch every time, this Agentic AI system uses mutation and crossover operations—methods inspired by biological evolution—to tweak and combine successful search paths to find even better results.

One of the biggest hurdles in financial forecasting is ensuring that the model's logic matches its code. QuantaAlpha enforces "semantic consistency," meaning the underlying hypothesis, the mathematical formula (factor expression), and the actual executable code must all align perfectly. This prevents the system from generating "garbage" code that looks good in a simulation but fails in the real world. It also limits factor complexity to avoid "crowding," a phenomenon where too many investors use the same simple strategy, causing its profits to vanish.

In testing on the CSI 300 index, the system powered by GPT-5.2 delivered an impressive 27.75% Annualized Rate of Return (ARR). This performance was measured using the Information Coefficient, which tracks how accurately the model predicts stock movements. Perhaps more importantly, the signals discovered on one index proved highly effective when applied to others, like the S&P 500. This robustness suggests that QuantaAlpha is discovering fundamental economic truths that remain valid even when global market conditions shift.

Researchers led by Zhi Yang (a leading scientist in financial AI) introduced QuantaAlpha, an evolutionary framework that treats the entire process of finding profitable trading signals as a "trajectory." Instead of starting from scratch every time, this Agentic AI system uses mutation and crossover operations—methods inspired by biological evolution—to tweak and combine successful search paths to find even better results.

One of the biggest hurdles in financial forecasting is ensuring that the model's logic matches its code. QuantaAlpha enforces "semantic consistency," meaning the underlying hypothesis, the mathematical formula (factor expression), and the actual executable code must all align perfectly. This prevents the system from generating "garbage" code that looks good in a simulation but fails in the real world. It also limits factor complexity to avoid "crowding," a phenomenon where too many investors use the same simple strategy, causing its profits to vanish.

In testing on the CSI 300 index, the system powered by GPT-5.2 delivered an impressive 27.75% Annualized Rate of Return (ARR). This performance was measured using the Information Coefficient, which tracks how accurately the model predicts stock movements. Perhaps more importantly, the signals discovered on one index proved highly effective when applied to others, like the S&P 500. This robustness suggests that QuantaAlpha is discovering fundamental economic truths that remain valid even when global market conditions shift.