How to Trade Weather Markets
A practical guide to trading weather prediction markets on Polymarket and Kalshi. Learn how to read the odds, compare them to real forecasts, and use ensemble models plus AI probabilities to find consistent edges.
Traders new to prediction markets who want a forecast-driven edge.
How bucket markets work, where value hides, and how to size positions.
Medium. Weather markets resolve to a single verifiable outcome.
1. What are weather prediction markets?
Prediction markets let traders buy and sell contracts that pay out based on real-world events. In weather markets, the event is usually a specific measurement of temperature, rainfall, or wind at an official station — most commonly the maximum temperature (tmax) recorded at a city's primary airport or weather station during a defined 24-hour window.
Two of the largest venues are Polymarket and Kalshi. Polymarket runs crypto-settled markets; Kalshi is a regulated U.S. exchange. Both list contracts like:
- "Will the high temperature in Chicago on July 6 be 80°F or higher?"
- "Will New York's tmax on August 15 fall in the 88–89°F bucket?"
- "Will it rain more than 0.1 inches in Miami today?"
2. How bucket markets work
Most weather markets use bucket or range contracts. Each bucket represents a narrow band of outcomes, such as 78–79°F, 80–81°F, or 36°C. At market close, every bucket except the one that matches the final observation expires worthless.
Each bucket trades as a YES/NO token. If you buy YES on the 80–81°F bucket and the final tmax is 80.3°F, your YES token settles at $1.00. If the final tmax is 79.8°F, the YES token settles at $0.00 and the NO token settles at $1.00. Because both sides trade freely before expiration, the current price reflects the market's live estimate of probability.
Example
The Dallas 100–101°F bucket is offered at YES = $0.42. That implies the crowd thinks there is a 42% chance the final tmax will land inside that bucket. If your own forecast says the chance is only 20%, buying NO at $0.58 is positive expected value.
3. Why forecasts beat the crowd
Most participants in weather markets trade on intuition, news headlines, or a quick glance at a phone app. They do not systematically compare multiple model runs, station microclimates, or observation timing. That creates repeatable edges for anyone who can source better data.
The ERNA engine processes several independent forecast layers:
- Standard forecast — global numerical weather models, updated several times per day.
- Ensemble forecast — multiple perturbed model runs combined into a probabilistic distribution.
- Erna Pro — our in-house AI model trained to blend and bias-correct the inputs for city-level tmax.
- METAR observations — real airport readings used to pace the forecast during the event day.
4. Using ensemble forecasts to find an edge
A single weather model can be wrong by several degrees. An ensemble aggregates dozens of model runs and tells you not just the most likely outcome but the full shape of the uncertainty. The two numbers that matter most for bucket trading are the mean and the spread.
Mean forecast
The average predicted tmax across all ensemble members. Use it to pick the central bucket.
Spread / variance
Wide spread means the outcome is uncertain. Tight spread means the central bucket is more reliable.
When the ensemble mean sits near the top of one bucket but the market price is still low, you have a value entry. When the mean sits between two buckets and the market is pricing one side as a heavy favorite, the opposite side often contains the edge.
5. Reading AI model probabilities
Raw model output is a temperature number. To turn it into a tradable probability, the model must be mapped to the exact bucket structure of the market. This means handling unit conversion, bucket floors, and city-specific biases. A forecast of 35.9°C in one city may map to the 35°C bucket; in another it may map to 36°C depending on how Polymarket labels the contract.
The ERNA dashboard shows the model-implied probability next to the market price for each bucket. The gap between the two is the edge. If the model thinks a YES token has a 70% chance and the market sells it at $0.55, the expected value is positive over time.
6. A simple trading workflow
- Check the market structure — confirm the exact city, station source, resolution time, and bucket labels.
- Compare forecasts — look at Standard, Ensemble, and Erna Pro tmax for the same date.
- Find the largest probability gap — identify the bucket where market price diverges most from model probability.
- Consider the timing — markets move as METAR observations arrive during the event day. Earlier entries carry more risk but more edge.
- Size the trade — never risk more than a small fixed fraction of your bankroll on a single bucket.
- Monitor or exit — weather markets can swing fast as new observations confirm or contradict the forecast.
7. Risk management for weather markets
Even the best forecast is not perfect. Weather markets resolve to a single station reading, and microclimates, instrument errors, or timing quirks can move the final number by a degree or two. Manage that risk with three rules:
- Flat stake sizing — bet the same dollar amount on every trade so no single loss wipes out a run of wins.
- Price filters — avoid tokens priced below $0.05 or above $0.95; transaction and slippage costs eat thin edges.
- Daily cost caps — set a maximum amount you are willing to deploy across all open weather positions in one day.
8. Start trading with ERNA
ERNA is not a trading bot. It is a decision-support engine: it reads the market, runs the models, and surfaces the gaps. You decide whether to trade, how much to stake, and when to exit.