JPMorgan's AI agents beat the 60/40 by 70 basis points in twenty-year backtests

verifiedFinance

Friday, July 10, 2026

Confidence

Medium · — Bloomberg primary + JPMorgan researcher attribution; no live-money results, overfitting risk material

Evidence

Bloomberg exclusive citing JPMorgan strategists

JPMorgan researchers built AI agents that allocate between stocks and bonds by market regime — and in backtests the best system beat a traditional 60/40 by 0.7 percentage points a year with lower volatility, Bloomberg reported.

The details

Agents powered by OpenAI and Anthropic models classified markets into four regimes — Goldilocks, reflation, stagflation, risk-off — and all eight systems tested beat the 60/40 on a risk-adjusted basis .
The results are historical simulations, not live investing, and JPMorgan warns against treating them as proof AI can consistently outperform markets .

What it means

Hold onto 0.7 percentage points, not the headline. That sits inside the range a well-tuned rules-based overlay produces; JPMorgan flagged it too. The signal: a G-SIB will publish agentic-allocation results at all. The question moves from "can it work" to "can we defend it to a regulator." Assume peers are running the same test and disclosing only winners.

Agents powered by OpenAI and Anthropic models classified markets into four regimes — Goldilocks, reflation, stagflation, risk-off — and all eight systems tested beat the 60/40 on a risk-adjusted basis .
The results are historical simulations, not live investing, and JPMorgan warns against treating them as proof AI can consistently outperform markets .

Sources

and in backtests the best system beat a traditional 60/40 by 0.7 percentage points a year with lower volatility ↗

Apply this today

The hands-on layer the brief points to: a workflow you can run, a tool to test, and today’s 60-second video.

Workflow of the day

Turn a Raw Webinar Transcript into a 5-Email Nurture Sequence with Distinct Angles

Sales & Marketing · tested · saves 3-4 hours per webinar

Full walkthrough →

Tool watch

GPT-Live

GPT-Live for sales reps: does full-duplex voice actually change your next call?

Read the breakdown →

In this issue

← Back to brief

Get the next one in your inbox

One daily brief. Every story gets a hype verdict.

No spam. Unsubscribe anytime.

On this page

The details
What it means

In this issue

← Back to brief

Stat of the Day

$2.5B▲

Microsoft Frontier Company

Microsoft's commitment to a 6,000-person forward-deployed AI unit — the largest single-vendor push into AI implementation services yet announced. Source

Today’s Take

Thursday was the day OpenAI stopped acting like an API vendor. A persistent workplace agent, a three-tier frontier model family, and preferred-model status inside the world's most-deployed enterprise Copilot arrived together, and Microsoft's own $2.5B professional-services build-out from the week prior reads as the distribution partner reinforcing the moat around itself. The FT export-controls story is the other side of that coin: US labs are now enterprise incumbents big enough that their sales-channel decisions are foreign policy. Watch what does not happen next — whether the government-tuned frontier lab that just cleared a security review starts to look less like a company and more like critical infrastructure, and whether Anthropic's stricter China posture becomes the regulatory floor rather than a competitive stance. Considered and passed: the IKS Health–TruBridge $557M deal (a rural-hospital consolidation with a thin AI angle) and the Deutsche Telekom OpenAI case study (lab-authored blueprint, no independent corroboration).

— Agentic desk

Never miss a signal

Join professionals getting the daily AI brief every weekday morning.

No spam · Unsubscribe anytime

No sponsored verdicts
Sources linked
Corrections public