Weather Market NO-Side Overpricing

How It Works The bot pulls live METAR surface observations and Open-Meteo numerical weather forecast data for the location and date in question. It then fits a Gaussian probability model to the forecast distribution to estimate the true probability of the temperature threshold being met. When the Polymarket YES price implies a higher probability than the model's estimate, the bot sells the YES (equivalently buys the NO) to capture the mispricing.

Entry / Exit Conditions Entry: Polymarket implied probability > Gaussian model probability by a configurable margin (edge threshold). Exit: position is held to resolution since binary weather markets resolve on the event date; early exit could be triggered if the Polymarket price converges to model fair value before expiry.

Verified Results The repository is described as a paper-trading bot, meaning no real capital results are confirmed. The existence of a live data pipeline and probabilistic model provides a credible testing framework.

Implementation Notes Two data sources (METAR for realized conditions and Open-Meteo for forecasts) cross-validate each other. The Gaussian model is a well-understood distributional assumption for temperature, though fat tails and model error remain. Repo size is modest (2.2 MB), suggesting a lean, focused codebase.

Risks Weather models carry forecast uncertainty that widens near the event threshold. METAR observations are station-specific and may not match the market's exact measurement location or method. Liquidity in Polymarket temperature markets may be thin, causing slippage that erodes the edge.