Half of every temperature market is already decided by 1pm
A daily temperature market looks like a forecasting question, and for the first few hours it is. Then at some point during the afternoon it quietly stops being a forecast and becomes a fact that not everyone has noticed yet. Knowing roughly when that happens, for a specific city, is the single most useful thing you can know about these markets. So we measured it: 13,185 city-days of airport observations across 33 cities over 400 days. Here is the curve.
The headline curve
For each hour of local time, the share of days on which the highest reading so far already equals the eventual daily high. Once that is true, the market has an answer even though it has not settled yet.
| Local time | Already decided | Midsummer | Deep winter | Clear | Rain |
|---|---|---|---|---|---|
| 10:00 | 14.6% | 10.0% | 18.9% | 8.9% | 23.2% |
| 11:00 | 23.8% | 19.2% | 28.2% | 17.8% | 32.8% |
| 12:00 | 36.5% | 32.2% | 40.9% | 30.8% | 44.9% |
| 13:00 | 53.7% | 47.8% | 59.7% | 49.5% | 60.0% |
| 14:00 | 71.6% | 64.4% | 78.3% | 70.2% | 73.8% |
| 15:00 | 85.7% | 79.7% | 90.5% | 85.9% | 85.3% |
| 16:00 | 93.5% | 90.7% | 95.3% | 94.1% | 92.6% |
| 17:00 | 97.0% | 96.8% | 96.5% | 97.6% | 96.0% |
One o'clock local is the coin-flip line. Four o'clock is where it is effectively over. Between those two hours the market is resolving in real time, in public, in front of anyone reading the observation feed.
Rain caps the day early, and it is the strongest single effect
Look at the two right-hand columns in the morning. At 10am the high is already in on 23.2 percent of rainy days against 8.9 percent of clear ones. That is a factor of 2.6.
The mechanism is simple. Cloud and precipitation put a ceiling on daytime heating, so a wet morning often means the day peaked early and will not climb again. A clear morning means the sun still has hours of work to do.
What is interesting is that the effect inverts by late afternoon. By 3pm the clear and rainy curves have crossed and converged, and by 5pm clear days are marginally more settled. Rain front-loads the decision. It does not change the endpoint.
Anyone modelling these markets who is not reading present-weather codes is missing the largest available variable, and missing it exactly where it does the most work: in the morning, when the market is still uncertain and the price still has room to move.
Season moves the whole curve by an hour
At 1pm local the high is already set on 59.7 percent of deep-winter days but only 47.8 percent of midsummer days. That gap holds across the afternoon and only closes near sunset.
Longer days and a higher sun angle keep pushing heat into the afternoon. In January the sun is already losing at 1pm. In July it is still winning at 3pm.
This matters more than it sounds for anyone building a model on recent data. A relationship fit on summer observations will systematically underestimate how settled a winter market is, and vice versa. The seasonal shift is roughly one hour of curve, which is enough to flip a lot of decisions.
When the peak actually lands
The hour that first records the day's high, across all 13,185 city-days:
- 14:00, 18.0 percent (the mode)
- 13:00, 17.2 percent
- 15:00, 14.0 percent
- 12:00, 12.7 percent
- 11:00, 9.2 percent
- 16:00, 7.8 percent
Roughly two thirds of daily highs land between 11am and 4pm, centred on 2pm. There is also a stubborn 6 percent that peak in the midnight hour, which are the days a warm front arrives overnight and the temperature falls all day from there. Those are the days a market anchored on "it will warm up this afternoon" is quietly already wrong.
The city spread is four hours wide
The aggregate curve is a blend of very different places. Here is the local hour at which each city crosses 50 percent decided, and then 90 percent.
| City | 50% decided | 90% decided |
|---|---|---|
| Los Angeles | 11:00 | 13:00 |
| Wellington | 11:00 | 15:00 |
| Taipei | 11:00 | 14:00 |
| Miami / Tel Aviv / Istanbul / Shanghai | 12:00 | 13:00 to 15:00 |
| London / NYC / Tokyo / Seoul / Hong Kong / Singapore | 13:00 | 15:00 to 16:00 |
| Toronto / Munich / Warsaw / Milan / Moscow / Beijing | 14:00 | 16:00 to 17:00 |
| Paris | 15:00 | 18:00 |
| Chengdu | 15:00 | 16:00 |
Los Angeles is done before lunch. Paris is still live at teatime. That is a four-hour spread on the same question, and it is entirely a function of local geography: marine layer cities cap early, continental cities keep heating.
Which means the phrase "these markets resolve in the afternoon" is nearly useless. The afternoon is a different thing in Los Angeles than it is in Paris, and the difference is bigger than most of the edges people are chasing.
What this is actually good for
Three practical uses, in rough order of how much they matter.
1. Knowing when the question stops being a question. If it is 4pm in Milan, the market is 93.5 percent settled whether or not the price reflects that. Anything still trading at a wide spread at that point is a spread on a known answer, not on a forecast.
2. Knowing when your information is worth something. The inverse is more useful. The hours where a better read pays are the ones where the market is genuinely undecided, which is late morning through early afternoon. At 11am the day is only 23.8 percent settled. That is where forecasting skill converts into price.
3. Knowing which cities to bother with. If you want a window in which to work, Paris gives you until 3pm and Los Angeles gives you until 11am. Continental cities with long heating curves leave more time for the market to be wrong in.
A caveat worth stating plainly
This measures when the temperature is decided, not when the market is decided. Those are different, and the gap between them is exactly where the trading happens. A market can stay mispriced well after the physics is settled, and often does, because not everyone is watching the observation feed.
It also says nothing about whether you can act on it. Depth in these books concentrates hard in the two or three buckets nearest the expected high, and the tails that look most obviously wrong are frequently backed by a few dollars of size. Being right and being able to size are separate problems.
Method
Source data is the routine METAR observation stream from 33 airport stations, retrieved from the Iowa State ASOS archive, covering June 22, 2025 to July 26, 2026. These are the same stations that daily temperature markets settle against: Incheon for Seoul, Haneda for Tokyo, London City for London, Malpensa for Milan, and so on.
Temperatures are handled as whole degrees Celsius throughout, matching how METAR reports them and how the markets resolve. Days are local calendar days at the station, with daylight saving handled per hemisphere. Days with fewer than 18 observations are excluded. Final sample: 13,185 city-days.
Every number above is a description of past observations, not a forecast. Seasonal coverage is a little over one full year, which is enough to see a seasonal cycle once and not enough to say it repeats.
Frequently asked
When is a daily temperature market effectively decided?
By 1pm local in 53.7 percent of cases, by 3pm in 85.7 percent, by 4pm in 93.5 percent. City by city the 50 percent line ranges from about 11am in Los Angeles to about 3pm in Paris.
Does rain change it?
Sharply, and early. At 10am the high is already in on 23.2 percent of rainy days versus 8.9 percent of clear ones. The two converge by mid-afternoon.
Does the season matter?
Yes. At 1pm, 59.7 percent of deep-winter days are settled against 47.8 percent of midsummer days. The whole curve shifts by roughly an hour.
Where does the day's high usually land?
The mode is 2pm at 18.0 percent, with about two thirds of all highs falling between 11am and 4pm.
Where to read these markets
Daily temperature markets settle on Polymarket's order book. SmartX is a terminal on top of it that shows the live probability alongside what larger positions are holding, which is the useful cut on a market that is resolving in real time while the price catches up.