How to trade weather markets on Polymarket & Kalshi

Daily-temperature markets look simple — will the high in New York be 87–88°F today? — and that simplicity is exactly why they are tradable. The outcome is set by objective, public data, while the inputs that predict it (weather models) are free. That gap between what the models imply and what the crowd prices is where the edge lives. This guide walks the whole loop, from how a market settles to how to find an edge that survives fees.

What a daily-temperature market actually is

On both Polymarket and Kalshi, a temperature event is a grid of mutually exclusive buckets — “≤84°”, “85–86°”, “87–88°”, and so on — covering the day’s high (or low) for one city. You buy the bucket you think will resolve YES. The single most important thing to internalise: the market does not resolve against the temperature in “the city.” It resolves against one specific weather station, on one specific data source, in one specific time window.

1. The resolution station decides everything

A market that says “New York” may settle on LaGuardia (KLGA), which routinely runs 3–5°F cooler than midtown Manhattan. Trade it off the temperature on your phone’s weather app — which reads a city-center blend — and you are systematically wrong before you start. This is the most common silent mistake in weather markets, and it is worst for international cities, where the resolving gauge is rarely the obvious one.

Before anything else, confirm the exact station and source in the market’s own resolution text. We keep a running reference here: resolution stations, city by city.

2. Turn a forecast into a distribution, not a guess

A single forecast (“high of 88”) is not tradable. What you need is a probability spread across the exact buckets the market offers: maybe 9% on 85–86°, 46% on 87–88°, 31% on 89–90°. That comes from blending several models (GFS, ECMWF, and others), correcting for the station’s known bias, and — critically — modelling the error honestly.

3. Weather doesn’t follow a bell curve

The most expensive rookie error is assuming forecast errors are Gaussian. They are not — they are fat-tailed. A “2-sigma” event a normal distribution says happens 5% of the time shows up closer to 10–12% in real temperature data. The result: trades you think are 90% are really 75–80%, and a long run of “sure things” turns into a brutal losing streak. Calibrate against years of historical forecast-vs-actual error, or you are quietly overpaying for the favourite bucket every day.

4. Find the edge — then check it clears the spread

Edge is your calibrated probability minus the market’s implied price. If your model says 46% and the YES trades at 31¢, that 15-point gap is an edge. But raw edge lies. Two things eat it:

• Spread: in thin buckets the bid/ask can be wide; you pay it on the way in and again on the way out.
• Fees: on a cheap contract, fixed fees are a brutal tax. A $0.01 fee on a $0.05 contract is a 20% hit — you need a huge edge just to break even. Avoid the cheap-tail death zone unless the edge is enormous.

Only trade an edge that is still positive after spread and fees.

5. Time the peak window

A daily-high market is decided in a few afternoon hours. Early in the day the model is your only signal; as observations come in, the live readings dominate and the distribution collapses toward the actual high. Watching the intraday pace — is the temperature tracking, ahead of, or behind the forecast curve? — tells you when a bucket is being confirmed or killed, often an hour or two before the order book catches up.

6. Don’t race the bots — out-position them

The busiest US markets are populated by arbitrage bots that reprice within seconds of each model cycle. You will not win a latency race against them. You can win by being calibrated and correct about the station, by trading the slower, thinner, international cities where the repricing window stays open for hours, and by acting the moment a divergence opens rather than after it closes.

7. Mind cross-platform settlement

Kalshi and Polymarket can settle the same city and day differently — different source, different clock — and a 1°F gap can flip a contract. That is a risk if you’re sloppy and an opportunity if you’re deliberate. See Kalshi vs Polymarket weather for the mechanics.

8. Know if your edge is real

A single losing 65% bet is not a broken model — it’s variance. It takes a few hundred trades to confirm an 8% edge with any confidence. Track every position against the resolved outcome and score your calibration over time; that scoreboard, not any one trade, tells you whether you have an edge or a story.

Where Temprr fits

Every step above is a tab, a spreadsheet formula, or a Python script. Temprr collapses them into one screen: it anchors each market to its exact settlement station, builds a calibrated, fat-tail-aware distribution over the real buckets, overlays the live market odds, surfaces the net-of-cost edge, and shows the evidence and the invalidation rule. No code, no bot, no API key — free, members-only at launch.