Text data for trading

Machine learning in finance: forecasting and usage of large language models

This repository contains the final project of the lab conducted with Ammagamma at UniMi. The project aims to leverage the use of large language models to perform a sentiment analysis of financial news to enrich a model for predicting exchange values in the forex market.

• DistilBert_FinancialNews.ipynb contains the fine tuning of DistilBERT on a labeled dateset of financial news on the task of sentiment analysis. TFDistilBertForSequenceClassification from the HuggingFace🤗Transformers library was exploited to produce a model to classify news as positive, neutral or negative.

• News_scraping.ipynb contains the scraping of news from the forexrate.co.uk website.

• Sentiment_analysis.ipynb exploits the classifier produced to extract the sentiment for the collected news and then aggregate it for each date.

• Forecast.ipynb contains different forecast approaches of the EURUSD exchange value expoiting the Prophet model. The data retrieved with the sentiment analysis were used to enrich the model.