Brokers, Anlysts Face Doom as AI Stock Pickers Show Promise, Says Study

In Summary

  • University of Florida professor tests ChatGPT, Grok and DeepSeek for stock picks
  • ChatGPT-4 simulation yielded over 650% return from late 2021 to 2023
  • AI models now used in real trades via Autopilot investment app
  • Analysts warned AI could automate many trading functions soon


Catenaa, Monday, June 09, 2025- Stockbrokers and analysts may soon find themselves outpaced by artificial intelligence  or AI Chatbots, a finance professor testing the real-world trading abilities of leading AI chatbots says.1
The study paper titled “Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models” can be found here.
Alejandro Lopez-Lira of the University of Florida has spent years evaluating large language models like OpenAI’s ChatGPT, DeepSeek and Elon Musk-backed Grok in academic and real-world simulations of equity market investing. His findings suggest these models can outperform many traditional strategies, with ChatGPT-4 alone yielding a simulated return exceeding 650% over two years.

In early tests, Lopez-Lira used more than 134,000 headlines from over 4,000 companies to assess AI’s ability to determine whether news was positive, negative or neutral for a stock.

The models then simulated trades based on those interpretations, producing significant returns in historical back tests.

The professor transitioned to real-world trials in late 2023, partnering with investment app Autopilot. Since then, ChatGPT, Grok and DeepSeek have been providing monthly portfolios based on company data, macroeconomic conditions and financial reports.

Real brokerage accounts now mirror these AI-generated picks, while human oversight remains key to preparing the input data.

Lopez-Lira noted that while today’s AI tools aren’t flawless and lack real-time data awareness, they show signs of grasping economic principles and generating credible investment theses. His experiments now allow the models more autonomy, including flexible portfolio composition and asset weightings.

He warned that while AI cannot yet replace human oversight, its expanding capabilities could disrupt finance industry norms faster than expected.

The first version was released April 2023. While the latest publicly availabe paper is dated as September 2023.

Sources
  1. https://www.marketwatch.com/story/a-professor-testing-chatgpts-deepseeks-and-groks-stock-picking-skills-suggests-stockbrokers-should-worry-f54d583a[↩]
Protected by Copyscape