About WeatherCaster

WeatherCaster is an automated trading system for Polymarket daily temperature markets. It uses a four-model meteorological ensemble to compute bucket probabilities, compares them to live market prices, and trades the gap with Kelly-sized positions.

The problem we're solving

Polymarket daily temperature markets are among the most technically transparent prediction markets available. The outcome is objective, the resolution data is public, and the pricing mechanism is well-understood. Yet they remain structurally mispriced — most participants rely on point forecasts rather than calibrated probability distributions, creating persistent edge for traders who do the quantitative work correctly.

Doing it correctly requires ingesting multiple numerical weather models, converting their ensemble output into bucket-level probabilities, applying station-specific bias corrections, and executing at the right time in the 6-hour model update cycle. That pipeline is non-trivial to build and requires continuous maintenance as market conditions evolve.

WeatherCaster does this automatically, across all active Polymarket weather cities, 24 hours a day.

The technical approach

The core of the system is a four-model ensemble: ECMWF IFS, GEFS, UKMO, and NWS. Each model run produces a probability distribution over the relevant temperature buckets for a given city and date. WeatherCaster weights these distributions using a calibration history that measures each model's performance by city, season, and lead time.

Before any position is entered, the system applies an 8% minimum edge filter — the gap between the ensemble-weighted probability and the current market price must be at least 8 percentage points after accounting for the spread. Positions are sized using fractional Kelly (0.25×) with hard per-position and per-market dollar caps. Circuit breakers halt trading automatically if daily loss limits are reached.

The system monitors live METAR feeds during the final hours before market close, which allows dynamic probability updates as real observations come in and other models have not yet repriced.

Transparency and limits

We are straightforward about what this system can and cannot do. Weather markets have real capacity constraints — the total capital that can profitably enter any single market is bounded by liquidity, and edge compresses as more algorithmic participants enter. We publish a “Known limitations” section on the main site because credibility comes from honesty, not hype.

WeatherCaster is a tool for technically capable traders who want institutional-grade forecast infrastructure without building it themselves. It is not a guaranteed profit machine.

Contact

Questions, partnership enquiries, and early-access requests: [email protected].

For background on the markets and strategy, the blog covers the mechanics, models, and trading approaches in detail.