Your climate normals are stale: introducing Volt Normals

Ask anyone in a weather market what "normal" means and they'll cite a 30-year climatology — usually WMO 1991–2020. Here's the problem: in a warming climate, a trailing 30-year average is a forecast of the past. Its center of mass sits around 2005, two decades of warming ago. At Brussels Airport, our trend-adjusted normal for today's date runs +0.7°C above the stale climatology. If your "degrees vs normal" anchor is biased by nearly a degree, every HDD/CDD estimate, every "5 degrees above normal" headline, and every contract priced off the anomaly inherits the bias — always in the same direction.

Volt Normals are our answer: per-station, per-day-of-year trend-adjusted quantile normals, evaluated at the current year, free to browse at voltcast.com/normals and to query by API.

How they're built

For each settlement station (the named airport gauges that CME weather contracts and prediction markets resolve on) and each day of the year, we fit quantile regressions of daily Tmax and Tmin against calendar year, using a ±10-day window around the day-of-year across the historical record. The P50 fit evaluated at 2026 is the normal; the P90 fit gives the hot tail, similarly trend-adjusted. The regression slope itself is published — the warming trend per decade, per station, per season — because the slope is often the most interesting number: warming isn't uniform across the calendar, and the P90's slope frequently exceeds the P50's. Extremes are warming faster than means, and you can see it station by station.

Nothing is hand-tuned. The pipeline refits monthly as new observations land, the evaluation year rolls automatically, and the method is versioned (this is VN-1) so any number you quote is reproducible.

The stale-bias table

The normals page shows, for every station: today's Volt Normal, the 1991–2020-style climatology for the same date, and the difference — the stale bias. Summer biases of +0.5 to +1.0°C are typical across western Europe; the CSV export has every station and every day of the year, free.

Why traders should care

Degree-day settles. A monthly HDD/CAT settle is an accumulation of daily anomalies. Feed it a cold-biased normal and your projected settle drifts warm systematically — the worst kind of model error, because it never averages out. Our Desk's Degree-Day panel projects final monthly settles using Volt Normals, and the switch moved projections by amounts that matter at contract scale.

"Above normal" trades. Prediction-market temperature questions and seasonal positioning both hinge on the anchor. A +1°C anomaly against a stale normal may be a +0.3°C anomaly against the true current-climate normal — a completely different bet.

Energy demand models. Population-weighted HDD/CDD built on stale normals over-forecast heating and under-forecast cooling, structurally. If your load model's residuals trend with the season, this is a candidate reason.

Honest limitations

A linear trend per day-of-year is deliberately simple: robust, explainable, hard to overfit — and blind to acceleration. Where warming is accelerating, even Volt Normals will lag slightly (much less than a 30-year average does). Station moves and instrument changes are handled by the underlying homogenized records, not by us. And normals describe climate, not weather: they're the anchor, never the forecast — for the distribution of what tomorrow will actually do, that's our station forecast distributions, scored in public like everything else.

Use them

Browse: voltcast.com/normals. Query: GET /v1/normals?icao=EBBR returns the normal, the stale climatology, the bias and the per-decade trends for any station, with a CSV a click away. Free tier covers it. If you're pricing anything that settles on a thermometer, stop anchoring to 2005's climate.