DynamicGreeks

DynamicGreeks is a Python toolkit that recalculates option Greeks after first adjusting implied volatility (IV) with real-time news sentiment generated by a fine-tuned FinBERT model.

---

Requirements

• Python 3.9+
• All packages listed in requirements.txt
  (transformers, numpy, pandas, yfinance, newspaper3k, …)

---

Installation

# 1) Clone the repo
git clone https://github.com/<your-handle>/DynamicGreeks.git
cd DynamicGreeks

# 2) (optional) create and activate a virtual environment
python -m venv venv && source venv/bin/activate      # Windows: venv\Scripts\activate

# 3) Install dependencies
pip install -r requirements.txt

The FinBERT checkpoint (~420 MB) is downloaded automatically from Hugging Face on first use.

---

Quick Start

Command line

python gui_app.py 

Prints the GUI interface where the user inputs the news article and options metrics. Returns the IV adjustment, updated Greeks, and a trade analysis in a separate window.

Repository Layout

BlackScholes.py            # Pricing & Greeks helpers
sentimentMapping.py        # sentiment → IV conversion
finalAnalysis.py           # decision making and analysis 
finetuning.py              # FinBERT fine-tuning script
gui_app.py                 # tkinter interface
dataCleansing.py           # text-preprocessing pipeline
baselineModelEvaluation.py # model benchmarks
requirements.txt           # required packages

BERT Primer

BERT (Bidirectional Encoder Representations from Transformers) learns contextual word embeddings by reading text in both directions.

Pre-training tasks

1. Masked-language modeling – predict 15 % masked tokens
2. Next-sentence prediction – decide if two sentences are consecutive

The resulting 12-layer encoder (for bert-base) captures subtle, context-dependent meanings.

Why BERT for DynamicGreeks?

• Context sensitivity – financial tone hinges on nuanced cues (“guidance missed expectations”).
• Transfer learning – fine-tune quickly on a domain corpus instead of training from scratch.

FinBERT

FinBERT keeps BERT’s architecture but is domain-adapted on ~200 k finance headlines & filings. We fine-tuned it on labelled tweets + articles to classify positive / neutral / negative sentiment with > 90 % F1.

logits → sentiment score *s* ∈ [-1, 1] → ΔIV = α·s
(α is calibrated on historical earnings moves).
The adjusted IV feeds Black-Scholes to update Δ, Γ, Θ, Vega, ρ.

---

Contributing

1. Fork the repo and create a feature branch

   git checkout -b feature/your-feature

2. Write tests and run pytest.
3. Format code with black . before committing.
4. Open a pull request describing your changes.

---

License

MIT