How we make a forecast, and how we check it
The steps
- Collect official data (national statistics offices, IMF, OECD, World Bank, Eurostat and others) as soon as it is published.
- Forecast each series with an ensemble of statistical and machine-learning models; ensemble weights are chosen on an earlier period and tested on later years held out from fitting and model choice. Large pre-trained models may have seen some of that history.
- Publish every forecast with an 80% likely range and a reliability grade (high, moderate, experimental) against a "no change" guess.
- For this year's growth, inflation and current account, start from the IMF's latest forecast and correct it monthly with new data (results).
- Stories use only these frozen forecasts and sourced facts; every number carries its status (fact, official estimate, Satoria forecast, scenario) and scope (global or a named country).
What we do not do
We do not forecast shocks, we do not publish a probability we did not measure, and we do not fill a gap with an institution's projection. A series with no forecast of our own has no page here.
All stories · Countries · The map of the future · Why trust Satoria · Engine · About · Satoria vs the IMF · How accurate we are · Method · Indicators · For AI agents