forecasting

Forecasting is the practice of making claims about the future that can be checked once the future arrives, and it is what separates a prediction someone can be held to from an opinion about where things are going. A forecast names a quantity, a date and a degree of confidence, so that a later observation can score it, and a forecaster is judged over many claims, never on one.

Forecasting is the practice of making claims about the future that can be checked once the future arrives, and it is what separates a prediction someone can be held to from an opinion about where things are going. A forecast names a quantity, a date and a degree of confidence, so that a later observation can score it, and a forecaster is judged over many claims, never on one.

A forecast starts from a base rate, how often things of the same kind have happened before, and moves away from it only as far as the specific evidence justifies. This is Bayes' rule used as a working habit, with the belief written down and dated so that how far it moved can be scored afterwards.

Its probabilities are worth something only if they are calibrated: of the claims made at 70%, about seven in ten should come true.

Several sources that estimate one quantity are combined so that their disagreement becomes uncertainty in the answer (random-effects pooling). Sources that describe different worlds are kept apart and weighed (scenario mixture).

The written form of a forecast is the dated bet. What no forecast contains is the black swan, so every serious forecast states the premise that none arrives in its window.

In a system of many parts the pace is usually set by one of them, the rate-limiting step, and finding it is most of the work of forecasting a technology.