HugoxBaca/bis-xtd-deriv

BIS Exchange-Traded Derivatives (XTD) Project

This repository contains a complete pipeline to fetch, parse, store, model, and visualise the BIS WS_XTD_DERIV data set.

Data source

Bank for International Settlements (2026), Exchange-traded derivatives statistics, BIS WS_XTD_DERIV 1.0 (data set), https://data.bis.org/topics/XTD_DER/data (accessed on 23 October 2026).

Project layout

exchange_traded_derivatives/
├── data/
│   └── raw/
│       └── bis_xtd_deriv.parquet          # tidy table produced by src/fetch_and_parse.py
├── src/
│   ├── fetch_and_parse.py                 # downloads SDMX‑JSON → parquet
│   └── build_dataset.py                   # builds HF Dataset and pushes to the Hub
├── models/
│   └── train_baseline.py                  # quick Prophet baseline (optional)
├── spaces/
│   └── xtd_dashboard/
│       ├── app.py                         # Gradio dashboard
│       └── requirements.txt
├── requirements.txt                        # root dependencies
└── README.md                               # this file

Quick start

# 1. Install dependencies
pip install -r requirements.txt

# 2. Fetch & parse the BIS data
python src/fetch_and_parse.py

# 3. Create & push the Hugging Face Dataset (requires `hf auth login`)
python src/build_dataset.py

# 4. (Optional) Run a simple Prophet baseline
python models/train_baseline.py

# 5. Launch the Gradio dashboard locally
cd spaces/xtd_dashboard
pip install -r requirements.txt
python app.py

Hugging Face Hub artefacts

• Dataset – your-username/bis-xtd-deriv (train / test split)
• Model – your-username/bis-xtd-forecaster (replace with your own model)
• Space – your-username/bis-xtd-dashboard (interactive Plotly chart)

Replace your-username with your actual Hugging Face user / org name before pushing.

Extending the project

• Add more MEASURE, INSTRUMENT, or MATURITY codes to the request URL.
• Pull a longer history by moving startPeriod earlier.
• Join external macro series (policy rates, FX spot, CPI) for richer conditioning.
• Replace the Prophet baseline with a 🤗 Transformers TimeSeriesTransformerForPrediction for probabilistic multi‑series forecasting.