The app for independent voices
27.75% annual return. 8% max drawdown. Factors that transfer across continents.
A new paper introduces QuantaAlpha — an evolutionary framework that uses LLMs to mine alpha factors through trajectory-level mutation and crossover, the way genetic algorithms evolve solutions.
Each factor mining run is treated as a trajectory — hypothesis, factor construction, backtest — then evolved iteratively.
Mutation localizes failure points and rewrites only the broken step. Crossover recombines validated segments from high-performing parents.
Constraint gates enforce semantic consistency, complexity limits, and redundancy filtering via AST subtree matching.
On CSI 300: IC of 0.15, ARR of 27.75%, MDD of 7.98% — beating RD-Agent and AlphaAgent across every metric.
Zero-shot transfer to CSI 500 and S&P 500 delivers 160% and 137% cumulative excess return over four years.
The most interesting part? QuantaAlpha's factors survived the 2023 China regime shift that crushed every baseline — by discovering structural microstructure signals instead of momentum proxies.
Paper:
Code:
