Is AI Good at Stock-Market Timing? A New Study Casts Doubt

- A new academic study challenges the narrative that AI can reliably time the stock market (i.e., predict when to buy or sell for maximum profit). The paper tests large language models (LLMs) like GPT on financial prediction tasks and finds they perform no better than simple baselines, despite earlier hype. - "Stock market timing" refers to the strategy of making buy/sell decisions based on predictions of future price movements, as opposed to long-term holding. It's notoriously difficult even for human experts. - The study adds to a growing debate in finance: Wall Street firms (e.g., JPMorgan, Goldman Sachs) and hedge funds have been rushing to deploy AI for trading, fueled by claims that LLMs can analyze news and sentiment faster than humans. Skeptics argue that financial markets are too noisy and efficient for pattern-based AI to consistently beat the market. - Key context: Earlier research (e.g., from the University of Chicago) had suggested LLMs could predict earnings surprises better than analysts. The new study specifically tests market *timing* rather than fundamental analysis, and finds much weaker results.