The inefficient pricing of news

May 27, 2026 - 02:46

The biggest barrier to an efficient stock market is not the availability of information, but whether investors can actually read, process, and extract genuinely new data from articles fast enough. A recent study using large language models analyzed over six million publicly available articles on U.S.-listed companies to isolate what the researchers call "pure news" -- the unexpected, truly novel information in each story.

The findings are striking. This unexpected news component turns out to be a very strong predictor of future stock returns. But the market does not react evenly to all types of news. Investors tend to underreact to topics that are dominated by negative tone or dense numerical content. In contrast, they overreact to stories that carry ambiguity or high media attention.

This creates a persistent pricing inefficiency. When a company releases a complex earnings report full of numbers, the market often fails to fully price in the implications. Meanwhile, a vague but heavily hyped story about a new product can cause a temporary spike that later reverses.

As artificial intelligence capabilities expand, the authors argue that financial regulation must evolve. It is no longer enough to ensure equal access to information. Policymakers must also consider differences in processing capacity. If some traders use AI to parse millions of articles in seconds while others read a handful of headlines, the playing field becomes uneven in a new and dangerous way. The future of market fairness may depend on who can compute the news, not just who can read it.