- JPMorgan Chase researchers built an array of AI-powered investing agents that shift between stocks and bonds based on changing market regimes; in backtests spanning the past two decades, the best-performing system topped a traditional 60/40 portfolio by 0.7 percentage point per year with lower volatility, and all eight AI agents tested outperformed the 60/40 benchmark on a risk-adjusted basis — they also beat JPMorgan’s own rules-based market regime model, suggesting the AI improved on an existing institutional framework rather than just beating a passive benchmark.
- The system, described by strategists led by Thomas Salopek as JPMorgan’s first attempt to build AI for market regime identification, uses models from OpenAI and Anthropic to classify the market into four regimes based on growth and inflation: Goldilocks (strong growth, low inflation), reflation, stagflation, and risk-off — the agents then allocate across asset classes based on the identified regime, favoring equities in strong-growth environments and shifting to fixed income as the outlook deteriorates.
- JPMorgan explicitly warns against overinterpreting the results: “We strongly caution against uncritically accepting what amounts to in-sample, overly confident answers of AI,” the strategists wrote, noting the results are based on historical simulations rather than live investing and that “agentic AI needs to be grounded in a well thought-out asset allocation process, rather than naively assuming the agent can be the source of the domain knowledge” — the caveat reflects a broader industry concern that AI trained on historical data will encounter regime changes it has never seen.
- The experiment reflects Wall Street’s next phase of AI adoption: after two years of embedding LLMs into research, coding, and internal tools, banks are now testing whether AI can make the industry’s most consequential decisions — capital allocation — autonomously; researchers have warned that widespread AI adoption in investing could produce more crowded trades, make markets easier to manipulate, and amplify volatility if too many firms’ models reach similar conclusions simultaneously.
What Happened?
JPMorgan published research Thursday describing its first AI-powered market regime identification and asset allocation system. Eight AI agents, powered by models from OpenAI and Anthropic, were tested against two decades of historical data. All eight outperformed a 60/40 portfolio on a risk-adjusted basis; the best beat it by 0.7 percentage point annually with lower volatility. The agents also outperformed JPMorgan’s existing rules-based market regime model. JPMorgan strategists led by Thomas Salopek described the results as promising but cautioned strongly against treating backtests as proof of future outperformance.
Why It Matters?
The 60/40 portfolio is the benchmark for passive multi-asset investing — beating it with lower volatility is a meaningful result, even in backtests. The finding that the AI agents also beat JPMorgan’s own institutional regime model is arguably more significant, because it suggests AI can identify market regime transitions with at least as much accuracy as expert-designed rule systems. The systemic risk concern raised in the paper is equally important: if the largest asset managers all adopt similar AI allocation systems trained on the same historical data, the resulting correlation in portfolio positioning could amplify market moves during stress periods — the opposite of the diversification benefit the 60/40 model is designed to provide.
What’s Next?
JPMorgan says it is “enthusiastic about the possibilities of agentic AI, even as we are wary to hand off asset allocation decision-making to an agent” — the implication is that live deployment, with human oversight, is the next step under evaluation. Watch for whether JPMorgan or other major banks announce AI-assisted or AI-directed allocation products for institutional clients. The regulatory question is equally live: the SEC and CFTC have been examining AI use in trading and have signaled interest in understanding whether AI-driven allocation systems create new systemic risks that existing market structure rules do not address.
Source: Bloomberg
