Do prediction markets forecast rain better?
An initial September 2026 analysis reported the figures below. They are exploratory results, not a validated claim that prediction markets outperform weather forecasts.
8 a.m. local · same day being forecast
394 matched city-days · September 15–29, 2026
Kalshi's Brier score was approximately 48% lower. Lower is better: the score rewards probabilities that are both directionally correct and appropriately confident.
What we compared
For every city, we took Kalshi's live probability at 8 a.m. local and the newest archived NOAA National Blend of Models forecast that would have been available by then. Both predicted measurable rain during that same calendar day and were scored against the market's eventual Yes/No settlement.
Why 8 a.m.
This analysis used the maximum hourly NOAA precipitation probability from 8 a.m. through midnight as a proxy. That is not the same as the probability of rain at any time during the full day. Later in the day, a conventional weather forecast usually describes rain still to come, while the market asks whether rain occurred at any point in the entire day. A fair later-day test must also incorporate precipitation already observed.
Limitations
The original aggregate report covered 394 observations; its underlying matched rows were not saved with the app, so these totals have not been independently reproduced here. The hourly-to-daily comparison also needs validation. NOAA NBM is a strong public forecast baseline but is not a literal archive of every consumer weather app screen. Market probabilities can also be affected by liquidity and bid–ask spreads.