SwissFinBERT

Model Description

SwissFinBERT is a fine-tuned BERT model specifically designed for financial sentiment analysis in the context of the Swiss stock market. It has been trained on a comprehensive dataset of Swiss financial news articles to accurately classify the sentiment of financial texts as positive (BUY), neutral (HOLD), or negative (SELL).

Model Details

• Model type: BERT
• Base Model: German FinBERT SC
• Language: German
• Domain: Financial sentiment analysis
• Training data: Approximately 28,000 articles from Swiss financial news sources including Finanz und Wirtschaft, Inside Paradeplatz, finews.ch, Neue Zürcher Zeitung, and SRF News.
• Fine-tuning dataset: Titles of the scraped news articles labeled using OpenAI’s GPT-3.5-turbo-0125.
• Performance: Achieved a validation/test accuracy of 78% and an aggregated accuracy of 87% on the finetuning dataset. 93% Accuracy on the German (translated) financial phrasebank of Scherrmann (2023)

Intended Use

SwissFinBERT is intended for use in financial sentiment analysis tasks, particularly for Swiss financial news articles.

Disclaimer

The outputs of SwissFinBERT are not to be intended as financial advice

How to Use

from transformers import pipeline

# Load the model
model_name = "AlGatone21/SwissFinBERT"
classifier = pipeline("sentiment-analysis", model=model_name)

# Example usage
text = "Die Aktien von UBS stiegen nach der Ankündigung eines neuen CEO."
result = classifier(text)
print(result)

Citation

If you use SwissFinBERT in your research or applications, please cite the following thesis:

Gattoni, A. (2024) Can Large Language Models (LLMs) predict the Swiss Stock Market?, University of St. Gallen, St. Gallen

Contact

For questions or feedback, please contact Alessio Gattoni at alessio.gattoni@gmail.com