Headlines vs Body Text as Signals
Headlines are compressed, fast, and biased toward drama, while body text is slower to arrive but carries the nuance and caveats that determine whether a story actually matters for a trade.
Prerequisites: Building a News Sentiment Pipeline End to End
A news signal built purely from headlines and one built from full article text behave differently, and a pipeline that only reads headlines is implicitly trading a different thing than one that reads the whole story. Headlines are written to be attention-grabbing and are available first, often seconds before the body text, which matters for latency-sensitive strategies. But they are also compressed and prone to overstating certainty — "Company Misses Earnings" can hide a body paragraph explaining the miss was due to a one-off charge that markets shouldn't care about.
Body text arrives later but carries qualifiers, numbers, and context that change or reverse the apparent direction of a headline. A signal that reacts only to headline sentiment will systematically overreact to stories the body text would have tempered, and a signal that waits for the full body sacrifices the speed advantage that made news trading attractive in the first place.
Headlines are fast but noisy and directionally overconfident; body text is slower but corrects that overconfidence — most production pipelines use headline sentiment as an early, provisional signal and body-text sentiment to confirm or override it.
Worked example
A headline reads "Retailer Slashes Guidance." A pure headline-sentiment model scores this strongly negative and a strategy sells immediately. The body text, arriving 40 seconds later, clarifies the cut was driven entirely by a currency translation effect from a stronger dollar, not falling demand — informed traders who read past the headline don't sell, and the price partially recovers. A pipeline combining both signals would flag the initial negative headline reaction as low-confidence until the body-text signal confirms or contradicts it.
Further reading
- Tetlock, 'Giving Content to Investor Sentiment' (Journal of Finance)