Princeton, ant group and Stanford researchers present aqua: a framework for autonomous factor discovery and model development in finance

Researchers from Princeton University, Ant Group, and Stanford University introduced AQuA, a two-part agentic framework for autonomous factor discovery and model development in quantitative finance. AQuA addresses methodological errors and non-reproducible backtests by constraining agent actions to prevent overfitting and leakage. The framework includes a six-agent pipeline for discovering symbolic alpha factors in cryptocurrency and a hybrid model for predicting stock returns in US equities. It achieved a Spearman IC of 0.190 on a crypto dataset, outperforming baselines, and demonstrated a Sharpe ratio of +2.50 with a causal volatility-targeting overlay. The research team emphasized structured research and evaluation over individual factor expressions. AQuA’s performance was tested from 2021 to 2025, showing positive returns annually, including during the 2022 drawdown. The paper noted that the two parts use different conventions and should not be directly compared. No single price-volume feature carried the strongest signal, with a ridge combination reaching +0.25. The framework’s effectiveness stems from constrained agent actions and rigorous evaluation protocols.