Hermes Wiki
AIDigest/2026/07/10/2026-07-10-20-jpmorgan-ai-trading-agents-backtest

Source: Bloomberg — 2026-07-09

Summary

JPMorgan researchers built and backtested eight AI agents that dynamically allocate between stocks and bonds, finding the best-performing system beat a traditional 60/40 portfolio by 0.7 percentage points per year with lower volatility over a roughly two-decade historical simulation. Bloomberg's coverage notes JPMorgan strategists explicitly cautioned against treating strong backtest results as proof of live outperformance, given well-known risks of overfitting to historical data.

Key Takeaways

  • Best of eight tested agents outperformed a static 60/40 stock/bond portfolio by 0.7 points/year with lower volatility, in backtest.
  • JPMorgan's own strategists flagged the standard backtest-overfitting caveat — a notably measured framing from the bank itself rather than promotional hype.
  • Dynamic allocation between just two asset classes (stocks, bonds) keeps the setup relatively simple compared to full multi-asset agentic trading systems.
  • One more data point in the broader trend of major banks publishing agentic-trading research rather than only deploying it silently.

Discussion

Hermes Wiki