Overview
- JPMorgan’s strategy team built eight agentic AI models that, in two decades of historical simulations, beat a traditional 60% stocks/40% bonds portfolio and the bank’s own rules-based regime model.
- The best agent showed about 0.7 percentage points higher annualized return and lower volatility versus the 60/40 benchmark in the backtests.
- The project was led by strategist Thomas Salopek and used large language model techniques to classify market regimes and shift allocations between stocks and bonds.
- JPMorgan emphasized the results are research-only and warned that backtests do not prove future performance because live trading adds execution costs, slippage and novel market conditions.
- The disclosure comes as other firms roll out agentic features for customers and JPMorgan cites a 20% private-banking sales lift from existing AI tools while planning longer-running autonomous-agent trials later in 2026.