Polymarket lists daily “highest temperature in city X” markets for about fifty cities, split into one-degree buckets. Automated traders reportedly made large profits on them in 2024–25. The idea: price each bucket from public forecasts — as they were issued, at the exact airport station each market resolves on — and trade the gaps.
| Decision time | Trades | Return per trade |
|---|---|---|
| Day before, edge ≥ 5¢ | 6,590 | −15% |
| Day before, edge ≥ 20¢ | 1,159 | −7% |
| Same morning, edge ≥ 5¢ | 7,289 | −28% |
| Same morning, edge ≥ 20¢ | 1,813 | −17% |
When the model disagreed with the price, the market was usually the one that knew something: newer model runs, local forecasters and, by the morning, actual readings. The later the decision, the worse the result.
A second try: settled buckets
Once the day’s high has passed a bucket, that bucket cannot win. On paper, such buckets still traded at 3–30¢, and selling them returned +11% per trade. A live check of 211 settled buckets told the truth: every one was offered at 0.1¢ with no buyers. The historical prices were mid-points of empty order books.
Both versions rejected. Public data loses to a market that already uses it, and historical prices in thin books are not tradable prices.