New AI Reads the News to Predict Stock and Energy Prices

A team of researchers has published a new AI system called DualCast that tries to do something investors have wanted for decades: predict where prices are going. The paper, posted on the research site arXiv, focuses on stocks and energy prices. Unlike most forecasting tools, DualCast doesn't just stare at charts. It also reads the news.

Here's how it works, in plain terms. The system is built on top of an existing AI language model — the kind of software behind chatbots — that already understands text. The researchers froze that model (meaning they didn't retrain it) and taught it a small new "vocabulary" for price movements, so it can treat a wiggle in a stock chart like a word in a sentence. That's the fast path: a quick numerical guess. Then an optional second step kicks in. A small add-on — think of it as a plug-in that teaches the AI one new skill — looks at that fast guess plus any news available at the moment, and revises the prediction.

The results are promising. Testing against stocks and energy prices at five-minute, daily and weekly time frames, the revised forecasts were the most accurate of the methods compared in 8 of 12 scenarios — including all of the longest-range ones. Adding news helped in most tests, though how much it helped varied from market to market.

The catch: this is a research paper, not an app you can download. It only tested a limited set of assets and time frames, and accuracy on past data never guarantees future results — especially in markets, where everyone is constantly trying to outsmart everyone else. Still, it points to a clear direction: the forecasting tools of the near future will likely blend numbers and headlines, rather than picking just one.