See what's coming,
before it's official.
Real public data, screened for genuine links between them, turned into plain-language forecasts — for a country, a city, a business, or a person.
Live prediction — #1
| Indicator | Reading | Direction |
|---|---|---|
| Flash composite PMI (July) | 52.1, up from 49.3 | ↑ |
| Retail sales (July) | −0.5% m/m | ↓ |
| Energy price cap | +13% from 1 July | ↓ |
| GfK consumer confidence (July) | +6pts, biggest jump since Nov 2023 | ↑ |
| Labour market | Unemployment flat at 4.9% | → |
Retail sales dipped, but that reads as payback from a May/June promotional pull-forward rather than new weakness. No analyst consensus was published for this release yet; professional consensus historically misses by ~0.2pp either way.
Method
Correlation isn't the method
Relationships are screened with Granger causality and lag-correlation analysis, then checked out-of-sample — not fit to historical curves after the fact.
Honest about the ceiling
Full prediction of "the fate of an economy" isn't realistic — reflexivity, the Lucas critique, structural breaks and black swans set hard limits. The goal is short-horizon, probabilistic, direction-and-timing forecasts.
A fixed set of cross-checks
Good forecasts weigh a handful of independent indicators against each other. Indicators are capped, not endlessly added — past that point you're fitting a narrative, not improving accuracy.
Stats compute, AI explains
Statistics and ML do the actual forecasting. The language layer narrates and contextualises validated output — it never invents a number or makes the call itself.