Citadel Weather Arbitrage

Weather traders are turning public forecasts into automated bets on Kalshi and Polymarket, with open-source code now showing how the pipeline works. (github.com) The basic trade is simple: prediction markets price a temperature or rainfall outcome in cents, and a bot compares that price with forecast odds from weather models. If the model says a contract should trade higher or lower than the market price, the bot flags an edge. (help.kalshi.com) One GitHub project published in 2026 says it scans 11 cities, blends forecasts from five providers — National Oceanic and Atmospheric Administration, Global Forecast System, ICON, European Centre for Medium-Range Weather Forecasts, and GEM — and sizes positions with a quarter-Kelly formula. (github.com) A companion dashboard repo says it tracks live positions, profit and loss, backtests, win rates, edge calibration, provider accuracy, and paper-trading results through a React and Django interface. (github.com) That setup targets a market structure Kalshi itself advertises: daily highs, rain, snow, and climate contracts tied to official weather readings in specific cities. Kalshi says those contracts settle from final climate reports, and most markets settle within a few hours after the outcome is known. (kalshi.com) (help.kalshi.com) Other public repos use the same playbook on Polymarket: pull weather forecasts, translate them into “true” probabilities, compare those with market-implied odds, and either recommend trades or execute them automatically. One Polymarket-focused bot says it uses Open-Meteo forecasts and calculates expected value for single bets and “blanket” positions across multiple buckets. (github.com) Another repo frames the edge less as meteorology than timing. It says the trade is to buy after a new National Oceanic and Atmospheric Administration model run appears and before retail traders reprice the contract, then exit once the market catches up. (github.com) That timing matters because weather contracts are unusually mechanical. A New York City temperature market, for example, does not settle on a vague citywide average; it settles on a specific official reading under written market rules, which is why traders obsess over station choice, model update times, and forecast bias. (help.kalshi.com) The social post behind this latest burst of attention claims a replication of a Citadel-linked weather-desk idea turned a small seed account into a larger one with an 84.2% win rate, but those performance figures were not independently verified in the public materials reviewed here. The code that is public does show that weather-market automation, paper trading, and live edge monitoring are now being packaged for anyone with Python, API keys, and a tolerance for forecast error. (github.com 1) (github.com 2)