DEMFI Demo

DEMFI aggregates hourly temperature forecasts from 14 numerical weather prediction (NWP) models, including ECMWF, GFS, ICON, JMA, and more. Each line represents one model's forecast trajectory. Ensemble models also show their spread (p10-p90 bands). The black line shows actual observations. Where models converge, confidence is high; where they diverge, uncertainty is greater.

Using all forecast models, DEMFI computes the probability that tomorrow's maximum temperature will fall in each temperature bucket. The tallest bar is the most likely outcome. A Shapiro-Wilk normality test determines whether to use a Gaussian fit or kernel density estimation (KDE) for the distribution curve.

The Edge Finder compares DEMFI's model-driven probabilities against live Polymarket prices for each temperature bucket. When DEMFI sees a higher probability than the market price implies, there is a "BUY YES" edge. When the market overprices an outcome, there is a "BUY NO" edge. Position sizes are calculated using the Kelly Criterion (quarter-Kelly, capped at 25%).

Long Shots identify temperature buckets priced at 2 cents or less by Polymarket — the market thinks they are near-impossible — but where DEMFI models estimate a 12%+ probability. These are high-risk, high-reward opportunities: a $1 bet can return $50+ if the bucket hits.

Safe Bets identify buckets where DEMFI estimates a 90%+ probability that the outcome will NOT occur. By buying NO shares at a discount, you collect a small but high-probability return — the "bond-like" positions in your weather portfolio.