Weather’s Edge

People claimed automated systems could make easy money trading weather prediction markets. I wanted to find out whether the claim survived contact with a working system.

Weather’s Edge combined forecasts from several models, corrected them against the weather stations used for settlement, estimated probabilities, compared those probabilities with market prices, and decided when to trade.

Paper trading produced returns that made the original claim look plausible. They also made the simulation suspect. It assumed fills were available at the observed price, with no meaningful spread, slippage, failed execution, or market impact.

A forecast can be accurate without creating a trade another person can execute. The backtester now separates forecast skill from illustrative profit and loss, and states the cost assumptions behind the latter.

Real trades made liquidity, fill quality, spread, fees, and capital constraints impossible to treat as footnotes. The system could contain careful weather modelling and still lack a tradable edge.

Two data sources that appeared to refer to the same Dallas/Fort Worth location were actually several miles apart. Under some weather patterns, that distance can change the observed daily high enough to change settlement.

I incorporated station-faithful resolution and bias correction, compared model output with actual settlements, and examined blockchain trading history. I also looked at successful large traders to understand whether their advantage came from forecasting, execution, market access, or capital.

Weather’s Edge was complete when it had answered the question it could answer at sensible cost.

Testing the remaining market-structure questions properly would require materially more capital. I retired the bot, published the loss, documented the caveats, and cleaned the code into a reference implementation. The forecast pipeline, paper trader, backtest harness, settlement logic, and risk controls remain available without pretending they form a profitable track record.