Are AI fund managers better? JPMorgan backtests: outperforms classic investment portfolios on an annualized basis with lower volatility

AI is moving into the most core investment decision-making domain on Wall Street. A team led by JPMorgan strategist Thomas Salopek recently completed an AI investment agent backtesting experiment, for the first time applying an AI system to market mechanism identification. The team built multiple AI agents that can dynamically adjust the equity and bond allocation according to changing market conditions, to explore the feasibility of autonomous investment decision-making.

The backtest results show that the best-performing system achieved an annualized return 0.7 percentage points higher than the traditional 60/40 equity-bond portfolio in a simulated environment over the past twenty years, with lower volatility, while also outperforming JPMorgan’s existing rule-based market mechanism models.

Although Wall Street is accelerating the deployment of AI in analytics, programming, and investment tools, this experiment marks an extension of AI applications into the core decision-making area of capital allocation. But JPMorgan has clearly warned that the result should not be viewed as evidence that AI can consistently outperform the market; the related exploration is still at an early stage.

Strong historical simulations, unverified in live trading

The AI investment agents developed by JPMorgan researchers primarily function to dynamically adjust the proportion of equity and bond allocation based on changes in market conditions. Across historical simulations covering the past twenty years, the best system delivered an annualized excess return of up to 0.7 percentage points, while also achieving lower volatility and outperforming the bank’s existing rule-based market mechanism models.

The strategist team noted in the report that the AI agent was designed to make decisions under uncertainty conditions, enabling better performance than a reasonable benchmark. This is also the first time JPMorgan has publicly shared its research results in the AI-driven capital allocation domain, marking a key step in the bank’s exploration of intelligent investment decision systems.

Despite the positive backtest data, JPMorgan remains cautious in interpreting the related conclusions. The bank explicitly emphasized that all of the above results come from a historical simulation environment and have not been validated through real market trading; therefore, they should not be used to infer that AI inherently has the capability to consistently outperform the market.

The strategist team also warned in the report that market participants should avoid uncritically accepting excessive AI confidence derived from in-sample backtest results. They believe that AI systems based on agents must be built on rigorous and prudent asset allocation processes, rather than simply assuming that the agent itself constitutes a source of expertise.

AI consensus risk is heating up: automated trading moves toward the “deep end of decision-making”

As enthusiasm for AI investment tools continues to rise on Wall Street, academia’s concern about its potential systemic risks is also increasing in parallel. According to Bloomberg, more and more research is focusing on a core question: when many institutions deploy similar AI models to make investment decisions, how will the logic of market operation change?

Researchers pointed out that although AI technology can significantly improve information retrieval efficiency and decision accuracy, it may also create risks such as similar portfolio structures and increased susceptibility of markets to manipulation, especially in stressful scenarios, where many institutions reach similar conclusions simultaneously, further amplifying market volatility. JPMorgan’s strategy team also acknowledged these risks in a recent report.

JPMorgan’s related tests reflect the evolution of AI applications on Wall Street. In the past two years, large banks have widely embedded large language models into supportive scenarios such as generating research reports, writing code, and internal investment tools. And current testing indicates that the industry is now evaluating whether AI systems can progress from assisting employee decision-making to taking on more decisive core functions such as cross-market capital allocation.

Risk warning and disclaimer

        The market is risky; investment requires caution. This article does not constitute personal investment advice, and it does not consider any individual user’s specific investment objectives, financial conditions, or needs. Users should consider whether any opinions, viewpoints, or conclusions in this article align with their own specific circumstances. Investing based on this is at your own risk.