AI investing agents beat 60/40 in JPMorgan market tests

AI investing agents tested by JPMorgan beat a 60/40 portfolio in two decades of simulations, but the bank says live results remain unproven.

JPMorgan tests allocation agents

Researchers at JPMorgan Chase & Co. built multiple artificial intelligence systems designed to move between stocks and bonds as market conditions changed. The strongest version outpaced a 60% stock and 40% bond portfolio by 0.7 percentage point a year in backtests covering the past two decades, according to strategists led by Thomas Salopek.

The same system also produced lower volatility than the 60/40 mix and beat JPMorgan's rules-based market regime model, the strategists said. That made the experiment more ambitious than using artificial intelligence to summarize research, write code, or screen data.

The bank did not present the results as evidence that AI can reliably beat markets. The simulations were historical tests, not live capital allocation, and the note warned against treating past model performance as proof of future investing skill.

The strategists described the project as JPMorgan's first effort to create an AI system aimed at identifying market regimes. In their words, "The AI agent can be set up with a process to be empowered to make decisions under uncertainty, producing outperformance vs a reasonable benchmark."

The 60/40 hurdle

The comparison matters because the 60/40 portfolio remains a simple reference point for balanced investing. In the framework cited by JPMorgan, 60% of the portfolio is allocated to stocks and 40% to bonds, giving the AI agents a benchmark that reflects both growth exposure and defensive assets.

Market regime models try to identify changes in conditions that may favor one asset class over another. JPMorgan's test used AI agents for that allocation decision, rather than relying only on predetermined rules.

The result puts the experiment at the center of a larger shift inside finance. Banks have spent the past two years adding large language models to research, software development, and internal investing workflows; JPMorgan's test points to a more sensitive question: whether AI should influence portfolio positioning itself.

Backtests face a harder market

The main risk is model overfitting, where a system learns the past too precisely and fails when market relationships change. Backtests can also be shaped by data choices, transaction assumptions, and the way regimes are defined.

For JPMorgan, the immediate value may be less about launching an automated allocation product and more about learning how AI behaves under uncertainty. A system that can explain shifts between stocks and bonds would be more useful to investors than one that produces unexplained trades.

The wider asset-management industry will be watching the governance question as closely as the performance number. If AI systems begin to influence allocation, clients and risk officers will want clearer audit trails, limits on model discretion, and evidence that decisions can survive stressed markets.

Scenarios depend on live results

If live testing preserves even part of the backtested edge, JPMorgan could expand AI agents into research-supported allocation tools. The macro effect would be a gradual increase in machine-guided flows between bonds and equities; the company would gain a stronger case for AI-led investment processes; rivals would face pressure to test similar systems.

If the advantage disappears outside simulations, the outcome would still matter, but for a different reason. JPMorgan would have evidence on where AI allocation fails, the global market effect would be limited, and the industry would likely focus more on controls, transparency, and risk management than on full automation.

The open questions are practical: whether performance survives live trading, whether volatility stays lower after costs, and whether the model can adapt when market regimes shift. Until those points are tested with real money, the experiment is a promising research result rather than a settled investing breakthrough.