JPMorgan builds AI agents that outperform traditional portfolios in two decades of backtesting

JPMorgan Chase has developed AI-powered agents that dynamically shift allocations between stocks and bonds based on market conditions, and the results from backtesting are turning heads.

Over two decades of historical simulations, the best-performing AI agent outpaced the traditional 60/40 portfolio by 0.7 percentage points per year. In English: if a standard balanced portfolio returned 8% annually, JPMorgan’s AI would have delivered 8.7%.

What the AI actually does

Led by JPMorgan strategist Thomas Salopek, the project uses AI agents that read market conditions and decide when to tilt toward stocks or retreat into bonds.

The AI agents didn’t just generate higher returns. They did it with lower volatility than the 60/40 benchmark. The AI strategy also beat JPMorgan’s own existing rules-based market regime model.

Still in the lab, not in your brokerage account

As of now, these AI agents exist entirely in the research and backtesting phase. JPMorgan has not launched any live trading systems or client-facing products based on this technology.

That said, JPMorgan has reported a 20% increase in gross sales within its private banking division attributed to the deployment of existing AI tools. The bank has also signaled plans to deploy longer-running autonomous agents later in 2026.

What this means for investors and the broader market

The entire active management industry has spent decades struggling to beat passive benchmarks by any amount after fees. A systematic, scalable approach that consistently adds 70 basis points per year would represent a genuine edge.

The risk for investors to consider is straightforward. Backtested outperformance is a necessary condition for a viable strategy, but it is nowhere near sufficient. Until these agents face market conditions they haven’t been trained on, with real capital on the line and execution costs eating into that 0.7 percentage point edge, JPMorgan’s AI agents remain an impressive proof of concept, not a proven product.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.