Research ·
Polymarket weather markets: which station settles each city
A Polymarket daily temperature market doesn't settle on "the temperature in the city". It settles on one station's reports, named in the market rules. Pick the wrong station or the wrong kind of report and you get the wrong answer, sometimes by a whole bucket.
What the rules say
On 2026-11-05, 274 of 368 open temperature events cited aviationweather.gov. Their rule: the highest (or lowest) temperature in any routine METAR report from one named airport during the local day. METAR is the standard airport weather report. Special reports (lines starting with SPECI) don't count. Where a report carries a temperature in tenths of a degree, the tenths are used; otherwise the whole-degree value is.
Not every city uses that rule. Hong Kong markets cite the Hong Kong Observatory. A few Chinese cities, Jinan and Zhengzhou among them, still cite Weather Underground. London and Paris moved from Weather Underground to NOAA's page for the same airport stations in late August 2026.
The rules change, so read the current rule on the market itself before relying on any table, including this one.
The station for each city
| City | Station checked | Same bucket as the result | Period |
|---|---|---|---|
| Tokyo | Haneda airport (RJTT) | 233 of 233 days | Mar–Nov 2026 |
| New York | LaGuardia airport (KLGA) | 14 of 14 days | Jul 2026 |
| London | London City airport (EGLC) | 11 of 11 days | Aug 2026 |
| Paris | Le Bourget airport (LFPB) | 11 of 11 days | Aug 2026 |
| Shanghai | Pudong airport (ZSPD) | 14 of 14 days | Jul 2026 |
| Seoul | Incheon airport (RKSI) | 13 of 14 days | Jul 2026 |
| Hong Kong | Airport (VHHH), not the settling station | 2 of 14 days | Jul 2026 |
| Shenzhen | Bao'an airport (ZGSZ), not the settling station | 4 of 14 days | Jul 2026 |
"Same bucket" means the station's highest reading for the local day fell in the bucket that won. For Seoul and Shanghai we tested both airports and kept the one that matched: Incheon, not Gimpo; Pudong, not Hongqiao. The one Seoul miss was a day Incheon read 27.0 °C and the market resolved 26.
Hong Kong and Shenzhen show what happens with the wrong station. Hong Kong settles on the Observatory, not the airport, and the airport matched 2 days in 14. In the public airport archive we used, Hong Kong and Shenzhen each have exactly one station, and neither is the one the market settles on.
Daily climate summaries give the wrong answer
The obvious shortcut is a daily maximum from a climate service. We tested one (Meteostat) against resolved markets in New York, London, Paris, Amsterdam and Madrid: it matched the winning bucket in 234 of 304 markets (77%). The highest routine airport report for the local day matched 329 of 329.
The difference is what gets measured. A climate summary records the true peak, including moments between reports. The market records the highest value among the reports. In New York the summary came out 1 °F higher on 27 of 58 days. That is enough to change the result: Celsius markets use one-degree buckets, and New York's are two degrees Fahrenheit wide.
Reports arrive late, and the price moves first
A METAR is published several minutes after the time printed on it. In one sample on 2026-11-05 the delay was 8.4 minutes for Tokyo Haneda, 5.5 for Shanghai Pudong, 4.5 for Seoul Incheon and 2.3 for Hong Kong's airport.
Faster instruments see the temperature earlier. At LaGuardia on 2026-07-24, the airport's 1-minute sensor record reached the 78–79 °F bucket at 13:06 local time; the METAR showed it at 13:51, 45 minutes later.
In July we also saw prices of buckets the temperature had already passed fall within about 5 minutes of the observation time, before the METAR was published. Some traders read a faster source than METAR.
Tokyo in detail
Tokyo markets settle on Haneda (RJTT). Its routine reports come at :00 and :30 past each hour, in whole degrees Celsius with no tenths. The highest of those reports matched the winning bucket on all 233 resolved Tokyo days we checked, from 2026-03-10.
The Japan Meteorological Agency's AMeDAS station at Haneda reports every 10 minutes in tenths of a degree, about 9 minutes after each reading. Rounded half-up, its value equals the METAR temperature on 102,170 of 102,171 matching reports. Because only the :00 and :30 reports count, a short peak between them can be missed, so the settled high can be lower than the day's true high.
Tokyo markets then go to UMA, the oracle Polymarket uses: a proposed result can be challenged for 15 minutes before it is final. The ones we watched in November were settled about 22 minutes after the Tokyo day ended.
Method and limits
Results are each market's resolution on Polymarket. Airport reports come from the Iowa Environmental Mesonet METAR archive, local day in each city's time zone. Most station checks cover two weeks in July 2026, so a station can match 14 of 14 days and still miss on a day with a peak between reports. This describes how the markets settle; it is not trading advice.