Anecast°

Weather worth knowing

◐

The market vs. the forecasters

Every evening, three calls go on record for tomorrow's high in the 14 cities where Kalshi runs a temperature market: the National Weather Service's, ours, and the traders' (the expected high implied by the bracket prices). Next morning the station each market settles on grades all three, the same way.

Standings · Oct 10 – Oct 20 · 142 city-days

On the next day's high, the market has missed by 1.1° on average, Anecast by 1.9°, and NWS by 2.1°.

Tuesday, graded · the calls from the evening before

Bold is the closest call (ties within half a degree share it). The number under each call is its miss. ☾ marks a maximum set before dawn: the official high is the calendar day's, so after a night-time front it can be the previous evening's warmth; NWS forecasts the daytime high, so on those days it is scored against a number it never forecast, while the market prices the calendar day and tends to get it.

Today's high · the three calls now, widest split first

Market: traders' expected high and the favorite bracket with its odds, from the latest Kalshi snapshot. Models (wider screens): the middle 80% of the weather-model runs. Split: the gap between the highest and lowest of the three calls.

By city · average miss since Oct 10

How it's graded

• The truth is the settlement station. Each Kalshi market names an NWS climate-report station (Chicago is Midway, Dallas is DFW, New York is Central Park); every call here is graded against that station's official high. Kalshi settles on the value as The Weather Company reports it, which can occasionally differ by rounding.
• Same moment, same yardstick. All three calls are as they stood at 00:00 UTC the evening before (8 PM Eastern, 5 PM Pacific), the last moment before the day begins; Kalshi keeps trading through the day itself, watching the thermometer, so a later cut wouldn't be a forecast. The market's price is read after our forecast is on record, never before it. (On days graded before Oct 19 the market was read about five minutes before our forecast, in the same pass; no model run lands in that gap.) The calls: NWS's forecast, the Anecast forecast as published, and the market's expected high from the bracket prices. Each is scored by its miss in degrees. Our likely range isn't used here, because a range against a single number wouldn't be a fair fight.
• Published, never recomputed. The Anecast number is the one that was on the page. A bug left the forecasts we published Oct 13–16 equal to NWS's in every city, so on the days they graded (Oct 14–17) our column matches NWS's. They stay in the standings as published.
• A day counts only when all three called it. Closest call: the smallest miss, with ties within half a degree shared.
• The calendar day, not the daytime. The official high is the highest reading between midnight and midnight. After a front that passes overnight it is set at 12:01 AM, the previous evening's warmth; NWS's forecast is for the daytime, so on those days NWS is scored against a number it never forecast (Boston, Oct 17: 62° at 12:04 AM, 50° in the afternoon, NWS said upper 40s). The market prices the calendar day — that is part of its edge, and it is the number the contract settles on, so the grading stands; the ☾ marks those days so no one mistakes them for a daytime bust.
• Why the market tends to win, in fairness. Traders read the NWS forecast and the models before they price, so the market is an aggregator with an information edge, not an independent forecaster. Its number is a probability-weighted expected value, which naturally misses by less on average than any single integer call; it can also never be exactly right the way an integer can. And this is a small, lumpy sample — one nor'easter bust along the East Coast moves it. What matters is whether the market's edge holds as the sample grows, and this page will keep showing that either way.

Not trading advice. Prices from Kalshi's public data; forecasts from the National Weather Service and Anecast.