How Often Are Financial News Sources Right?
We check the tone of every headline against the later price move — and compare each source with chance.
Of 18 sources with at least 300 checks, none is measurably better than chance.
Chance: how often the same up, down and sideways calls would be right if shuffled against the same price moves within each market, so knowing which markets move little or a lot earns nothing.
The interval is the range the edge would most likely fall in if the measurement were repeated (95 %). The verdict already corrects for testing many sources at once (Holm correction): one interval missing zero does not by itself make that source distinguishable from chance.
All sources, from the first tradable price
37,235 checks of the 13 sources that have a rate at every horizon: did the market move the way the article pointed, measured from the first price a reader could have traded? Sources with fewer checks are left out here so that no withheld rate can be worked out from these totals.
| Horizon | Checked reports | Days | Hit rate | Chance | Vs. chance (95 % interval) | Detectable from | Verdict |
|---|---|---|---|---|---|---|---|
| 1 d | 12,594 | 286 | 37.5% | 36.5% | +1.0 pts(−0.5 to +2.4) | ±2.1 pts | Not distinguishable from chance |
| 3 d | 12,383 | 283 | 36.7% | 37.0% | −0.3 pts(−2.6 to +2.0) | ±3.3 pts | Not distinguishable from chance |
| 7 d | 12,258 | 279 | 36.7% | 36.7% | 0.0 pts(−1.7 to +1.7) | ±2.4 pts | Not distinguishable from chance |
By source (all horizons)
16 more sources have fewer than 300 checks; no rate is published for them yet.
By kind of publisher and by language
The same numbers for groups of sources, so a pattern that no single source shows clearly can still be seen.
| Horizon | Checked reports | Days | Hit rate | Chance | Vs. chance (95 % interval) | Detectable from | Verdict |
|---|---|---|---|---|---|---|---|
| Crypto editorial (6 sources), 1 d | 3,480 | 258 | 38.5% | 38.4% | +0.1 pts(−2.8 to +3.1) | ±4.2 pts | Not distinguishable from chance |
| Crypto editorial (6 sources), 3 d | 3,458 | 256 | 37.7% | 38.9% | −1.2 pts(−4.8 to +2.4) | ±5.1 pts | Not distinguishable from chance |
| Crypto editorial (6 sources), 7 d | 3,419 | 252 | 38.3% | 39.1% | −0.8 pts(−4.1 to +2.6) | ±4.8 pts | Not distinguishable from chance |
| Editorial news (5 sources), 1 d | 4,885 | 276 | 37.4% | 36.0% | +1.5 pts(−0.6 to +3.5) | ±2.9 pts | Not distinguishable from chance |
| Editorial news (5 sources), 3 d | 4,721 | 272 | 35.5% | 36.3% | −0.9 pts(−3.1 to +1.3) | ±3.2 pts | Not distinguishable from chance |
| Editorial news (5 sources), 7 d | 4,660 | 267 | 36.6% | 36.2% | +0.3 pts(−1.9 to +2.6) | ±3.2 pts | Not distinguishable from chance |
| Specialist press (1 sources), 1 d | 3,812 | 151 | 36.5% | 35.8% | +0.7 pts(−1.1 to +2.4) | ±2.4 pts | Not distinguishable from chance |
| Specialist press (1 sources), 3 d | 3,800 | 151 | 37.2% | 36.1% | +1.1 pts(−2.1 to +4.4) | ±4.6 pts | Not distinguishable from chance |
| Specialist press (1 sources), 7 d | 3,800 | 151 | 35.5% | 35.3% | +0.2 pts(−1.2 to +1.6) | ±2.0 pts | Not distinguishable from chance |
| Other (under review) (1 sources), 1 d | 417 | 73 | 37.7% | 37.2% | +0.4 pts(−4.9 to +5.8) | ±7.7 pts | Not distinguishable from chance |
| Other (under review) (1 sources), 3 d | 404 | 71 | 36.6% | 38.4% | −1.8 pts(−7.1 to +3.6) | ±7.6 pts | Not distinguishable from chance |
| Other (under review) (1 sources), 7 d | 379 | 66 | 36.7% | 40.1% | −3.4 pts(−9.6 to +2.8) | ±8.9 pts | Not distinguishable from chance |
| Deutsch (3 sources), 1 d | 1,868 | 247 | 36.4% | 35.1% | +1.3 pts(−1.3 to +4.0) | ±3.8 pts | Not distinguishable from chance |
| Deutsch (3 sources), 3 d | 1,828 | 243 | 33.9% | 35.1% | −1.2 pts(−3.7 to +1.2) | ±3.5 pts | Not distinguishable from chance |
| Deutsch (3 sources), 7 d | 1,789 | 238 | 34.2% | 35.8% | −1.6 pts(−5.2 to +2.0) | ±5.2 pts | Not distinguishable from chance |
| English (10 sources), 1 d | 10,726 | 283 | 37.6% | 36.7% | +0.9 pts(−0.7 to +2.5) | ±2.3 pts | Not distinguishable from chance |
| English (10 sources), 3 d | 10,555 | 280 | 37.1% | 37.3% | −0.2 pts(−2.6 to +2.3) | ±3.5 pts | Not distinguishable from chance |
| English (10 sources), 7 d | 10,469 | 276 | 37.2% | 37.0% | +0.2 pts(−1.4 to +1.8) | ±2.3 pts | Not distinguishable from chance |
Our Methodology
AI classifies each article's tone as up, down or sideways for the market it covers. From the first price a reader could trade after publication we check the move after 1, 3 and 7 days. A source's hit rate is compared with chance for the same calls and outcomes within each market, so covering calm or busy markets earns nothing. Each difference comes with a 95 % interval; checks close in time are resampled together in blocks of 7 or 14 days, because they share market moves. A verdict needs at least 300 checks and must hold after correcting for the number of sources tested (Holm). This answers "did the market move that way?": a hit rate, not a return. AIamond makes no forecasts of its own.
Method source-record-v2, registered on November 3, 2026 and never edited since; this run was computed Nov 12, 2026, 09:30 AM UTC. Read the method
SHA-256 2df8da59ae4dd7bf3c46fafdcb26cbb56d9712a270b647d6b186bc9d427faa81
Are you a named source and think a number is wrong, or want to reply? Write to info@aiamond.com; corrections are listed in the change log.
This run’s totals are a snapshot as of Nov 12, 2026, 09:30 AM UTC and are limited to admitted sources — they can be smaller than a live, still-running count shown elsewhere on the site.
Changes to how these numbers are measured, and mistakes we found in the data, are listed in the change log.
Use this data: the full record is open under CC BY 4.0, with an API and an MCP server for agents.
Use this data: API · MCP · CSV