JPMorgan Builds AI Agents That Beat 60/40 Portfolio in Backtests
TL;DR
JPMorgan Chase is testing AI agents that do more than screen stocks: they are meant to decide portfolio weights themselves. In historical backtests, the agents reportedly beat a classic 60/40 split between stocks and bonds. The ambition goes beyond common AI use in research or risk management, because the model is asked to allocate capital directly. The caveat is big: backtests are not live performance, and public details on data, costs, risk and drawdowns remain thin.
Nauti's Take
This is the kind of headline where the calculator should come out before the imagination runs ahead. Beating 60/40 in a backtest sounds strong, but backtests often reward models that map the past very neatly.
The real test is how JPMorgan’s agents handle stress periods, trading costs, liquidity and model error. Until then, this is a serious research signal, not a permission slip for autopilot wealth management.
Briefingshow
If banks move AI agents into asset allocation, AI shifts from analysis support toward decision infrastructure. That matters for investors because a stronger backtest can sound like product proof, but without live performance, cost assumptions and risk data it proves little. The useful signal is the direction of travel on Wall Street: fewer dashboards, more autonomous investment workflows.
Sources