A study across 100+ stocks finds most AI-driven trading models underperform simple buy-and-hold investing over two decades.
A multi-university study backtested AI trading strategies across more than 100 stocks over 20 years and found they largely failed to outperform buy-and-hold investing. Researchers from the University of Edinburgh, UCLA, and Sungkyunkwan University attributed earlier perceived advantages of AI to flawed testing methods, such as short timeframes and limited stock samples.
Previous studies often tested AI models on small groups of high-performing stocks over periods as brief as three to six months. The new research suggests such narrow testing created an illusion of effectiveness, as strategies appeared successful only in isolated, favorable conditions.
The findings, set to be presented at the ACM SIGKDD Conference in August, challenge the notion that AI can consistently outsmart traditional passive investing. The study has not yet undergone peer review.