Live Performance of Agentic SAA vs Human-Centered SAA

Determine whether agentic strategic asset allocation (SAA) delivers better performance than traditional human-centered strategic asset allocation when evaluated through live, out-of-sample implementation under the same Investment Policy Statement constraints.

Background

The paper proposes an agentic architecture for strategic asset allocation that coordinates approximately 50 specialized agents to produce capital market assumptions, construct portfolios with over 20 methods, and conduct peer review and voting, all governed by an Investment Policy Statement (IPS).

While the framework shows promising design features—such as structured multi-agent deliberation and ensemble selection—the authors emphasize that the ultimate question of comparative performance versus traditional human-led SAA cannot be settled by backtests alone and requires live deployment.