JPMorgan Builds AI Agents That Beat 60/40 Portfolio in Backtests
TL;DR
JPMorgan Chase & Co. is testing AI agents that go beyond stock picking or risk analysis and try to allocate capital across asset classes themselves, Bloomberg reports. In backtests, the agents reportedly beat a classic 60/40 stock-and-bond portfolio. That is a useful signal, but not proof that the system works in live markets. The shift is the role of AI: from research assistant to allocation engine. That moves the model closer to real portfolio decisions.
Nauti's Take
Backtests are every strategy’s favorite playground because the future is already labeled and on the shelf. JPMorgan’s work still matters because it shows where institutional AI is heading: out of the chat window and into decisions with money, liability and audit trails attached.
The sober test is whether anyone would trust the agent in March 2020, during a bond crash or after a model update. Without that answer, this is closer to a strong research demo than an investable product.
Briefingshow
When a major bank tests AI agents for asset allocation, the topic moves beyond polished analyst summaries. The real question becomes whether models should be allowed to reshape portfolios over time. For investors and finance teams, the backtest win matters less than the control setup: who stops the agent when regimes, liquidity or correlations break?
Sources