A new study challenges the hype around AI-powered stock trading, finding that large-language models struggle to outperform basic buy-and-hold strategies over extended periods. Researchers backtested LLM-based trading approaches across two decades of market data, including the 2008 financial crisis and the pandemic crash.

UCLA mathematics professor Mihai Cucuringu led the research, which tested models across diverse market environments and incorporated delisted stocks to eliminate survivorship bias. The findings reveal that LLMs missed bull market gains by being overly cautious and suffered heavy losses in downturns through aggressive trading.

Earlier studies suggesting AI trading advantages typically focused on limited stock selections over brief timeframes. This comprehensive analysis shows those benefits largely disappear when evaluated across broader market conditions and longer horizons.

Investors increasingly rely on AI for earnings analysis and market data crunching, but this research suggests caution before trusting algorithms with actual trading decisions. The study demonstrates that despite immunity to fear and greed, AI lacks the adaptability that human judgment provides during market turbulence.