Most people who follow prediction markets have never opened a weather market, which is odd, because on any given day they are the biggest thing on the board. A scan of open Polymarket events on July 27 counts 117 live temperature markets carrying roughly $2.89 million in 24-hour volume. Hong Kong alone traded $244,000, Seoul $193,000, Chengdu $131,000. That is more than the geopolitics markets people actually talk about. This is a practical guide to what they are, what they settle against, which ones are worth your attention, and the one liquidity trap that catches almost everybody the first time.
The format: an eleven-rung ladder
A city market opens as a ladder of mutually exclusive buckets covering the plausible range of that day's high temperature. Usually eleven of them, in whole degrees, with catch-alls at each end.
A real example, the Milan market for July 27, priced mid-afternoon local time:
- 34°C: 61.5 percent
- 33°C: 30.5 percent
- 35°C: 3.4 percent
- 36°C: 0.4 percent
- 28°C or below, 29, 30, 31, 32, 37, 38 or higher: 0.1 percent each
Each bucket trades between 0 and 100 and reads directly as a probability. Because exactly one of them happens, the ladder should sum to about 100, and watching where the mass sits, and how it slides through the day as observations come in, is a live readout of collective belief. At settlement one bucket pays 100 and every other bucket goes to zero. The next day's ladder opens fresh.
The part most people miss: what it actually settles against
This is the single most important detail and it is buried in the rules text.
A temperature market does not settle against a city average, a forecast, or a weather app. It settles against the highest reading at one named airport weather station, in whole degrees Celsius.
- Seoul → Incheon International (RKSI)
- Tokyo → Haneda (RJTT)
- London → London City Airport (EGLC)
- Milan → Malpensa (LIMC)
- Shanghai → Pudong (ZSPD)
- Guangzhou → Baiyun (ZGGG)
Airports publish routine aviation weather observations called METARs every 20 to 30 minutes, and the temperature field in a METAR is already a whole number of degrees. There is no rounding step to argue about and no proprietary index to subscribe to. It is the same free public stream that pilots read.
Two consequences follow, and both are tradeable.
First, the settlement value assembles in public, in real time. By mid-afternoon local time a large part of the answer already exists. Anyone watching the observation stream is watching the settlement being built.
Second, a single station is not a city. Malpensa sits about 50 km from central Milan. London City is a different microclimate from Heathrow. If your mental model is "how hot was it in town today," you are pricing the wrong variable.
Which cities are actually predictable
Scored against settlement across 189 markets in 20 cities over ten days in July, the market favorite hit the exact degree 44 percent of the time. Against an eleven-bucket ladder, random guessing lands near 9 percent. But that aggregate hides an enormous spread.
| City | Hit rate | Read |
|---|---|---|
| Madrid | 89% | Stable heat dome. Effectively solved. |
| Paris | 78% | Settled continental summer. |
| Shanghai | 70% | Predictable at this time of year. |
| New York | 63% | Moderate. |
| Manila | 60% | Moderate. |
| Moscow / Shenzhen / Singapore / Hong Kong | 50% | Coin flip between two buckets. |
| London / Mexico City | 33% | Genuinely uncertain. |
| Chengdu / Taipei / Tokyo | 30% | Genuinely uncertain. |
| Toronto | 22% | Wide. |
| Beijing / Guangzhou / Seoul | 20% | Monsoon and convective chaos. |
The instinct is to read this as some crowds being sharper than others. It is not. Madrid in late July sits under a stable heat dome where the same number prints day after day. Seoul in late July is monsoon season, where a rain band arriving three hours early moves the daily high by several degrees. The market's accuracy is tracking how legible the underlying physics is, which is exactly what a working price should do.
The practical takeaway inverts the intuition. The accurate cities are the ones with nothing left on the table. An 89 percent favorite in Madrid is correctly priced and boring. If an edge exists anywhere, it is in Seoul, Beijing and Guangzhou, where the market is genuinely unsure and a better read of the afternoon could pay.
The liquidity trap
Here is the mistake that catches people, and it is worth spelling out because the headline volume number actively misleads.
$2.89 million a day sounds deep. It is not evenly spread. Volume concentrates hard in the two or three buckets nearest the expected high, because that is where the real disagreement is. The tail buckets, the ones that look most obviously mispriced, are close to empty.
A concrete example from the Milan ladder on July 27. The 35°C bucket was offered at 4.5 cents. If you thought the day had a real chance of overshooting, that looks like a cheap option. But the entire visible offer at that price was 93 shares. Four dollars and change. You could be completely right about the temperature and still not be able to put on a position that matters.
Three rules follow from that.
- Check depth before you check price. A 20 percent apparent edge on $4 of size is not an opportunity, it is a screenshot.
- The mispricing you can act on is in the middle of the ladder, not the tails, because that is the only place with a book behind it.
- Size to the offer, not to your conviction. Sending an order larger than the visible depth leaves you partially filled at a worse average than you modelled.
The annual markets are priced very differently
Worth knowing that the daily and annual weather markets do not behave alike at all.
The market on where 2026 will rank among the hottest years on record carries $3,210,235, one of the largest climate positions in any public market. It prices first place at 44.5 percent and second at 54.5 percent, with everything else under 3 percent combined. That is 99 percent of the probability mass on the two hottest outcomes in recorded history.
For comparison, Berkeley Earth's statistical estimate gives first place a 19 percent chance, and the World Meteorological Organization puts 2026 at 1.44 degrees above the pre-industrial baseline, near but probably not above the 2024 record of 1.55.
The nearby books carry the same tilt: 91 percent that July 2026 is the hottest July on record, 94.5 percent that some month of 2026 sets a record.
Whether that is anchoring on two consecutive record years, or the market correctly front-running a developing El Nino that seasonal models update too slowly to capture, is not settled by the data available today. What is different from the usual climate argument is that this one resolves, in January, with a number.
How to actually follow them
Three things make these readable rather than just noisy.
- Watch the ladder shape, not one bucket. A ladder with 60 percent on one rung is a confident market. One with 30/30/25 spread across three is telling you the physics is unresolved, and that is information too.
- Check the resolution station, not the city. The rules text names it. That is the thermometer you are betting on.
- Look at positioning, not just price. On markets where the narrative is loud and the book is thin, who is holding matters more than the headline number.
These settle on Polymarket's order book. SmartX is a terminal on top of it that puts the live probability next to what larger positions are actually holding, which is the cut that matters when sentiment and positioning disagree.
Frequently asked
How do weather prediction markets work?
A ladder of roughly eleven mutually exclusive buckets covering the plausible range of the day's high. Prices read as probabilities and sum to about 100. One bucket pays out at settlement, the rest go to zero.
What do they settle against?
The highest reading at one named airport station, in whole degrees Celsius, taken from the public METAR observation stream. Seoul resolves on Incheon, Tokyo on Haneda, Milan on Malpensa.
Which ones are worth trading?
Accuracy tracks local predictability. Madrid at 89 percent is solved. Seoul, Beijing and Guangzhou at around 20 percent are where the genuine uncertainty is, and therefore where any edge lives.
How much liquidity is there really?
About $2.89 million a day across 117 markets, but concentrated in the buckets nearest the expected high. Tail buckets can show large apparent mispricing on only a few dollars of depth. Always check depth before price.
Are they more accurate than a forecast?
They are unbiased, which is a different and arguably more useful property. Across 189 settled markets, 52 resolved hotter than the favorite and 54 cooler, with a mean error of negative 0.04 degrees.