FinBERT and OMXS30: A Study of Nasdaq News and Market Movements

This bachelor thesis investigates the relationship between daily sentiment in financial news articles related to companies listed on the OMXS30 and daily percentage changes in the OMXS30 index itself. Sentiment classification is performed using FinBERT, a transformer-based language modelfine-tuned for financial texts. Articles were collected via web scraping from Nasdaq Stockholm’s official website and analyzed for their positive, neutral, or negative tone. To address FinBERT’s token limit, all long articles was divided into chunks, and a daily average sentiment score (NetScore) was calculated. These scores were then compared with the percentage change in the OMXS30 index. The same analysis was performed for four selected OMXS30 companies (Nordea, Ericsson, Saab AB, and AstraZeneca) for the same day and up to three days ahead, using Pearson correlation. The results show no statistically significant correlation on the index level. At the company level, only Nordea displayed a weak but statistically significant correlation between same-day sentiment and stock return. The study demonstrates the technical feasibility of automated sentiment analysis for financial news, but also indicates that the current method has limited predictive power. Future work may benefit from combining sentiment with other variables such as news volume, volatility, or additional language models.