option

Download option data using yfinance

IV Prediction Model (iv_prod_xgb.json)

Recent improvements were made to the Implied Volatility (IV) prediction model (train_prod_model.py and pricing.py):

• Asset Classes: Added support for specific tickers (aapl, gold, sp500, etc.) and general asset classes (Stock, Index, Commodity).
• Data Filtering: Excluded extreme IVs (outside 1% to 150%) and near-expiration options (DTE < 1.1).
• Target Transformation: Predicting log(IV) to stabilize variance for better performance across different volatility regimes.
• Feature Engineering: Added features like inv_dte, vix_x_log_moneyness, VIX-based atm_iv_proxy, and explicit buckets for Moneyness and DTE.
• Training Strategy: Parallel loading of historical CSV data with 80/20 train/val split and Early Stopping.
• Model Tuning: Integrated Optuna for hyperparameter optimization across XGBoost, LightGBM, and CatBoost.

Usage

Data are stored in Parquet format, in the options_data and spread directories. You can see some examples in DATA.md

Ensure you are in the virtual environment:

source venv/bin/activate

1. Training the Model

Train one or more models (XGBoost, LightGBM, CatBoost, YDF). The best model for general prediction is saved as iv_prod_xgb.json.

# Train XGBoost with Optuna tuning
python train_prod_model.py --models xgb --tune --tune-trials 50

# Train all supported models
python train_prod_model.py --models all

2. Pricing a Specific Option (pricing.py)

Predicts IV, Price, and Greeks (Delta, Gamma, Vega, Theta, Rho) for a single strike.

Syntax:

python pricing.py <underlying> <strike> <days> <vix> [rate] [-t ticker] [-m model]

Example:

python pricing.py 500 505 30 18.5 -t sp500 -m cb

3. Pricing an Option Chain (pricing-form.py)

Generates a beautiful console-rendered option chain (16 strikes above/below ATM) with IV skew and Greeks for both Calls and Puts.

Syntax:

python pricing-form.py <underlying> <days> <vix> [rate] [-t ticker] [-m model]

Example:

python pricing-form.py 150.5 45 22.1 -t aapl -m xgb

---

Note: Supported tickers for the -t flag include: gold, silver, longterm, aapl, amzn, goog, sp500, nq100, dowjones. Use stock, index, or commodity for general asset class defaults. Supported models for the -m flag include: cb (CatBoost, default), xgb (XGBoost), lgb (LightGBM), ydf (YDF).