Tutorial · forecast band
The forecast band shows the range where the model expects most realized prices and signals honestly when evidence gets weaker. Alvo does not submit orders and does not turn p90 into a price guarantee.
The median price scenario read by the trading plan. It is not a clearing guarantee.
The lower and upper bounds of an approximately 80% uncertainty interval.
The Giacomini-White test must support that the model really beats the naive baseline.
Coverage, PIT/Kupiec/Winkler and pinball show whether the band stays honest.
Do not use the signal when the source is stale, unavailable or the sample is too short.
p50 is the median scenario, useful for planning, but not a guaranteed market outcome.
A wide band means more uncertainty; a narrow band without calibration is not proof.
When significance or coverage is weak, use the band as a warning layer, not as a green light.
Nor is p10 a floor: both bounds describe an uncertainty interval.
A single hour or p50 alone does not say how reliable the forecast is: read it together with data freshness, coverage and caveats.
Alvo is decision support: the trader places and owns the order; Alvo supplies data, band and audit trail.
Field names an integration should treat as uncertainty.
forecast.points[hour] = {
p10: lower uncertainty bound,
p50: median served forecast,
p90: upper uncertainty bound
}
forecast.metrics = {
coverage: share of holdout prices inside p10-p90,
significance: Giacomini-White gate vs naive baseline
}No. p90 is the upper edge of an uncertainty interval, not a cap and not a clearing guarantee.
The Giacomini-White test checks whether the historical model edge over the naive baseline is statistically real, with HAC protection for daily recalibration.
Treat the band as a warning layer: defer the decision, reduce volume or require manual review.
The workspace shows p10/p50/p90 next to OREE prices, cross-border regime and a deterministic decision brief.