When Machines Learn to Mine Alpha: Inside the Framework That Just Beat Wall Street's Quants at Their Own Game

A new evolutionary AI system delivered a 27.75% annualized return on Chinese equities and transferred its winning factors across markets without retraining.

Alpha mining has been the closely guarded craft of quantitative researchers who spend years hunting for signals buried in market noise. The work is grueling, the failure rate brutal, and the half life of any discovered edge shockingly short. Now a research team spanning Shanghai University of Finance and Economics, Stanford, Peking University, and a handful of other institutions has published something that should make every quant pay attention. Their framework, called QuantaAlpha, treats factor discovery as an evolutionary process driven by large language models, and the numbers it produced are not the kind that get politely ignored.

The Problem Nobody Solved Cleanly

To understand why this matters, consider what factor mining actually involves.