Does media tone lead stock returns?

90 large companies · GDELT news tone vs sector-relative returns · 2024-09-01 → 2026-08-31 · updated 2026-09-25

The idea came from mrsindex.com, which ranks brands by how their press coverage reads. If good coverage builds reputation, does it also show up in the stock price, and does it show up before the price moves or after?

Which comes first?

Each bar is the average, across companies, of the correlation between a day's tone shock (tone versus its own last 60 days) and the stock's return relative to its sector peers some days earlier or later. The grey band is where 300 time-scrambled placebos land, so a bar outside it is unlikely to be chance.

x-axis: days tone leads the return (negative = the return came first). Shaded right-hand zone = +2 days and later, the first lag you could trade: GDELT's day ends at 20:00 New York, so day +1 already reflects after-hours news.

By sector

Key numbers

Blue = outside the placebo band. "Trailing month → next month" is the original question as asked: average tone shock over 21 trading days against the next 21 days' sector-relative return.

Could you trade it?

A deliberately naive test: every day, buy the fifth of companies with the most positive tone shock, short the most negative fifth, and hold for the day that starts two days later. Sector-neutral, before costs; at daily turnover, costs would take a large bite of anything shown here.

Company by company

Sorted by the tradable lag (+2 days). One company's correlation needs to pass roughly ±0.09 to stand out on its own, and among 90 companies a handful will do that by luck.

How it was built

• Tone: GDELT's daily average tone for English-language articles that name each company as an organization or person, read from the raw 15-minute Global Knowledge Graph files (about 86,000 of them). Weekend and holiday news rolls onto the next trading day, weighted by article count, and each day is scored against the company's own previous 60 trading days so only the surprise counts.
• Returns: Yahoo Finance adjusted closes. Each stock's return minus the average of its sector peers, so a market-wide or sector-wide day doesn't count as news. Foreign companies trade here as US ADRs.
• Significance: every company's tone series is rotated in time by the same random offset 300 times. That keeps each series' shape and the links between companies but breaks the timing, which gives an honest band for "what chance looks like".
• Holdout: the second year (from 2025-09-01) is reported separately so a pattern found in year one has to show up again.
• Known gaps: stock-market stories aren't filtered out, which pushes the "price first" side up. GDELT tone is a dictionary score, not a modern language model. No intraday timing, earnings-day controls or trading costs.
• Left out: BMW (no price history), Bank of America (median 1 articles/day), Nordstrom (no price history), McDonald's (median 0 articles/day), Walgreens (no price history); plus Fidelity (private), Samsung, Hyundai, Kia and Volvo (no clean US listing).

Research note, not investment advice. Code: macro-engine src/sentiment/. Built by Phill with Jasper.