nayelsdk1/boreas-weather-derivatives

Boreas — Weather Derivatives Desk

Interactive pricing app for temperature weather derivatives (HDD/CDD calls), built on stochastic (Ornstein–Uhlenbeck) and machine-learning temperature models.

Companion to the research report Stochastic and Machine-Learning Temperature Models for Weather-Derivative Pricing.

Pages

• —Config — pick a city (7 study cities with full OU/ML artifacts, or any world city via the Open-Meteo archive API) and the study window.
• —Descriptive — climate profile, seasonal distributions, HDD/CDD.
• —Prediction — OU vs ML (Ridge / Random Forest / XGBoost) fan charts.
• —Pricing — HDD call quote under each model's Monte-Carlo trajectories.
• —Sensitivity — price & Greeks vs strike, market-factor sensitivities.

Models bundled

For each of the 7 study cities: OU parameters, Ridge, Random Forest, XGBoost, plus the recursive out-of-sample error pools used by the ML Monte-Carlo generator (Barnor et al., 2026). Random Forest files are tracked with Git LFS.

Run locally

pip install -r requirements.txt
streamlit run app/streamlit_app.py

Self-directed quantitative research project — Nayel Benabdesadok (ENSAE Paris).