The airport next door is 1.6C warmer than the one that pays
A temperature market settles on one station, but you can see all of them. The airports around a city report on the same schedule, they are free to read, and when one of them ticks up it is tempting to treat that as a preview of what the settlement station is about to do. We measured how good a preview it actually is, on 1,824 readings across 14 cities, taken in the final six hours before each market closed. The short answer: the neighbours run hot, the gap is bigger than the bucket, and it is different in every city.
The headline number
Take the hottest station in the field around a city, the one a trader would most likely notice, and compare it to what the settlement station finally printed that day.
It averages 1.60C too high. It lands inside the settling bucket 16 percent of the time. On an eleven bucket ladder where each rung is a single degree, a reading that is more than a degree and a half off is not a preview of the answer. It is a different question.
The nearest station does better on average, at 0.31C below settlement, but that average hides the spread. Some cities sit above their neighbours and some sit far below, and averaging them produces a number that describes no city at all.
Every city has its own offset
| City | Readings | Nearest station vs settled | Hottest in field vs settled | Nearest is |
|---|---|---|---|---|
| San Francisco | 44 | -4.95C | +4.61C | 14.7km |
| Amsterdam | 175 | +0.01C | +3.61C | 43km |
| Shanghai | 213 | +1.33C | +3.15C | 45km |
| London | 169 | -0.31C | +2.27C | 21km |
| Beijing | 70 | -0.50C | +2.00C | 68.3km |
| Seoul | 142 | +1.26C | +1.51C | 40km |
| Tokyo | 143 | +0.24C | +1.48C | 25km |
| Milan | 175 | -0.58C | +1.03C | 23km |
| Shenzhen | 138 | -2.48C | +1.02C | 55km |
| New York | 35 | +0.60C | +1.00C | 8.5km |
| Paris | 137 | -0.26C | +0.73C | 16.6km |
| Munich | 210 | -0.69C | +0.63C | 47km |
| Chicago | 35 | -0.63C | +0.37C | 18.1km |
| Taipei | 138 | -1.03C | -1.03C | 33km |
San Francisco is the extreme and the reason is not subtle. The station that settles sits on the bay, in reach of marine air that never gets to the stations inland from it. On the days measured it ran nearly five degrees below its nearest neighbour. If you are reading an inland reading and holding a bucket at SFO, you are not early. You are looking at a different climate.
Taipei is the only city in the set where the neighbours run cooler than the station that pays, and it is the only one where the hottest station in the field is still below settlement. Everywhere else the field overstates the day.
Chicago is the tightest. Its hottest neighbour sits 0.37C above settlement, well inside a single bucket, which makes it the one city in this set where a quick glance at the field is close to honest.
Why the field runs hot
Two reasons, and they stack.
The first is selection. Taking the hottest station out of six or seven is taking a maximum of maxima. Even if every station were an unbiased read of the same city, the hottest of them would sit above the truth simply because you picked it for being high. That alone accounts for part of the gap.
The second is geography, and it is the larger part. Settlement stations are airports, and airports are often placed where cities are not: on coasts, in river valleys, on the windward side of things. Those are the cooler spots. London City sits on the Thames. SFO sits on the bay. The stations further out are frequently drier and warmer for reasons that have nothing to do with today's weather and will still be true next month.
What to do with it
Do not read the field as a leading indicator of level. Read it as a leading indicator of direction. If every station around a city turns over and starts falling, the settlement station is usually not far behind, and that is worth knowing an hour early. If one station is two degrees above the settlement station, that is probably just Tuesday.
If you want a level from the field, subtract the city's offset first, and use the nearest stations rather than the hottest. Restricting the field to stations within 60km cuts the cold bias of the field average from 0.25C to 0.08C in our measurement, which is the difference between a correction that helps and one that quietly drags every read in one direction.
And check which station your market actually settles on before any of this. That is a separate trap and we audited all thirty nine rulebooks for it, because Denver does not settle at Denver International and Dallas does not settle at DFW.
Caveats
The window is the final six hours before each market closes, which is when most of the volume decides. Earlier in the day the gaps are wider and noisier.
New York, Chicago and San Francisco carry 35 to 44 readings each, far fewer than the European and Asian cities, because they were added to our collection later. Treat those three as directional rather than settled, particularly San Francisco, where the effect is dramatic but rests on 44 readings.
These offsets are a few weeks of summer. Marine layers, river cooling and urban heat all shift with the season, and a winter table would not look the same. The sign should hold, since it comes from where the stations are, but the size will move.
Method
Live market snapshots are captured every ten minutes for 14 cities, each carrying the settlement station's observations and the running maximum at every reporting station in that city's field. A day counts once the tape has crossed that city's local midnight, which is when these markets close. The settled value is the highest reading the settlement station posted on its own local calendar day, so readings after midnight are read to confirm the close and then dropped before the maximum is taken.
"Vs settled" is the mean signed difference between that station's running maximum at the time of reading and the settlement station's final maximum. Positive means the neighbour was reading above where the day ended up.