The Forecastability-First Principles
A statement of principles for forecasting and for the systems built on it.
The future is not equally knowable everywhere.
A forecast can use only the information available to it, and that information is measurable. It is the ceiling on performance: no model outruns its information.
The ceiling is a theorem, not a benchmark. Derived from information theory rather than from model comparisons, it holds for every method, present or future.
Forecastability is horizon-specific. A series predictable one step ahead may be noise twelve steps ahead, and a forecast without a stated horizon cannot be evaluated.
Measure before you model. Which model is best is the second question; how much of the future is knowable at all is the first.
Accuracy alone is uninterpretable. Baselines show what was easy; the ceiling shows what was possible; a method is judged by how much of the possible it achieved.
Sophistication must earn its keep. Rich information rewards advanced methods; scarce information makes simple ones near-optimal; absent information makes the base rate honest and decision design the real work.
Predictive authority should scale with forecastability. Where the future is weakly knowable a system should advise; only where it is strongly knowable should it act.
SIGN-OFF AND CITATION
Dr Peter Catt, Auckland, 2026.
Cite this page: Catt, P.M. (2026). The forecastability-first principles. The Knowable Future. theknowablefuture.com/principles