Sentiment and Volatility in Financial Markets: A Review of BERT and GARCH Applications during Geopolitical Crises

Battle Count: 81

The paper provides a framework for integrating NLP-derived sentiment signals into volatility forecasting models, which is directly applicable to risk management and position sizing during geopolitical crises. The BERT sentiment scores could serve as alpha signals or risk overlays in trading strategies. However, the daily frequency, low R², and lack of backtesting limit immediate practical trading application. The methodology is more suited for risk management and portfolio monitoring than high-frequency trading.

Requires expertise in both NLP (BERT fine-tuning, tokenization, embedding) and econometrics (GARCH modeling, Student-t distribution, diagnostic testing). The pipeline involves web scraping, text preprocessing (NLTK, BeautifulSoup, regex), BERT inference, and GARCH estimation. Moderate computational requirements for BERT inference on 10,000+ headlines. The integration of NLP outputs into econometric models adds complexity. Python and R scripts are provided, reducing implementation barriers.

The paper provides a GitHub repository with cleaned sentiment dataset, Python and R scripts, and replication instructions. However, the raw news data from Goperigon is under a non-commercial research license and cannot be redistributed. The BERT fine-tuning details (specific pre-trained checkpoint, hyperparameters, training epochs) are not fully specified. The data period (Jan 1 - Jul 17, 2024) and sample size (105 observations, 10,000+ headlines) are clearly stated.