Intervalprognoser

Intervalprognoser are forecasting methods that output an interval [lower, upper] for a target variable, rather than a single point estimate. The interval is intended to contain the true value with a specified probability, such as 90%.

They can be produced by several approaches. Quantile regression estimates lower and upper quantiles to form prediction intervals. Bayesian methods yield posterior predictive intervals. Bootstrap and conformal prediction provide distribution-free or minimal-assumption intervals. Ensemble methods can use the spread of multiple forecasts to derive an interval.

Interval forecasts are used in finance, energy planning, weather forecasting, and supply chain management to quantify uncertainty and support risk-aware decisions.

Evaluation focuses on coverage probability (the stated confidence level), interval width (sharpness), and calibration. Proper scoring rules and interval scores assess the trade-off between accuracy and informativeness.

In English-language statistics, the term is often called a prediction interval or interval forecast; intervalprognoser reflects the same concept in Danish or related languages.