Forecasting

Structured estimation of how conditions may evolve, stated as explicit assumptions.

A forecast widening from a record of observations Observations run into a single origin point, marked by a vertical rule for the present. From that point four nested bands spread to the right and widen as they go: outermost everything possible, then everything plausible, then everything probable, with a narrower band above the others marking what would be preferable. A few individual estimates sit inside the bands.

At a glance

Produces
Conditional estimates of future values, reported as ranges rather than single points.
Depends on
A reference base that genuinely resembles the thing being estimated, and assumptions that can be stated and attacked.
Fails when
The structure of the situation changes faster than the record used to estimate from it.

Forecasting is the systematic estimation of future conditions, values or events under stated assumptions. It does not claim certainty. It produces conditional statements about what is likely given what is currently known, and it is only as good as the assumptions it makes visible.

How we approach it

We treat a forecast as an argument rather than an answer: a claim that can be inspected, challenged and revised.

  • Every estimate is conditional. Assumptions, data sources and method travel with the number, so that a reader can disagree with the reasoning rather than only with the conclusion.
  • Quantitative estimation and structured judgement are kept separable. Where a model and an expert disagree, we report the disagreement instead of averaging it away.
  • Ranges and probabilities are preferred to single points. Narrow precision is treated as a claim requiring justification, not a sign of rigour.
  • Forecasts are revisited and scored. Being wrong in a documented way is more useful than being vague in a safe way.