Researchers from Princeton, Ant Group and Stanford Introduce AQuA: A Two-Part Agentic Framework for Autonomous Factor Discovery and Model Development in Quantitative Finance

AQuA is a framework that uses AI agents to automatically discover trading factors and develop predictive models for financial markets while preventing a common research failure: agents learning from their own experimental mistakes instead of genuine discoveries. The system works in two independent parts, one for discovering crypto trading signals and another for building stock return prediction models, with a key design principle that agents can explore freely within their designated tools but cannot change the standards by which their work is evaluated. This matters because quantitative research is vulnerable to subtle methodological errors that produce misleading backtest results, and traditional safeguards like human review fail when both the creator and reviewer share the same blind spots. The framework addresses this by freezing the evaluation criteria and data splits before any iterations begin, ensuring that what improves across iterations is the research process itself rather than the agent's ability to exploit flaws in the evaluation system.