Weather Market NO-Side Fade Using METAR and Numerical Forecasts

How It Works Polymarket temperature markets (e.g., 'Will high temperature in City X exceed Y degrees on date Z?') price the probability of a weather outcome. This strategy builds a Gaussian probability model for the temperature outcome using two data sources: real-time METAR (Meteorological Aerodrome Reports, official aviation weather observations) and Open-Meteo numerical weather forecasts. When the market's implied probability of NO is lower than the Gaussian model suggests, the bot sells NO shares.

Entry/Exit Conditions Entry is on the NO side when market price implies a lower probability of the event not occurring than the model calculates. The threshold for entry likely incorporates fee costs and model confidence. Exit is at market resolution (binary outcome) or possibly at price convergence if the market corrects before settlement.

Verification Described as a paper-trading bot only. No live PnL or backtested performance metrics are published in the description. The Gaussian model is a reasonable first approximation for temperature distributions but may underfit in extreme weather regimes.

Implementation Notes METAR data provides high-frequency ground truth observations, while Open-Meteo provides forward-looking forecast distributions. Combining both allows estimation of both current conditions and expected settlement-day temperature. The 3193KB repo size is moderate, suggesting meaningful code. Key implementation challenge is matching Polymarket market resolution location/time precisely to METAR station data.

Risks Weather has fat tails and systematic biases that Gaussian models underestimate, particularly during heat waves or cold snaps. METAR station may not match the geographic point used for market resolution. Model edge may be thin and sensitive to fee assumptions. Open-Meteo forecast accuracy degrades beyond 5-7 days.