Home › Accuracy
How accurate are our predictions?
We grade ourselves against reality. Every figure below comes from 23,513 predictions checked against the actual ex-dividend dates that followed.
Across the 19,002 predictions we publish (high & medium confidence), graded Jul 2023–Jul 2026.
Confirmed in real time
We also grade ourselves the moment a company officially declares a dividend — weeks before it goes ex. Of the 394 predictions officially confirmed since we started tracking (2026-07-22), our accuracy vs the declared date:
Accuracy by confidence rating
Every prediction carries a confidence rating. We publish the high and medium ones and abstain on the rest — and the track record shows why:
| Confidence | Predictions | Median error | Within 3d | Within 7d |
|---|---|---|---|---|
| high | 16,369 | 2d | 78% | 97% |
| medium | 2,633 | 3d | 60% | 85% |
| low | 4,511 | 4d | 48% | 62% |
How we measure this — honestly
It would be easy to cheat at this: re-run today's model over past dates and "predict" dates it has already seen. We don't. For every past ex-dividend date, we rebuild the prediction using only the data available before that date — the model never sees the answer it's being graded on. That's the same point-in-time discipline as a proper backtest, and it's why these numbers reflect what you'd actually have got at the time, not hindsight.
Predictions are still statistical estimates — a company can change or cancel a dividend without warning, which no method can foresee. This is educational information, not financial advice. See how the predictions work or browse the upcoming calendar.