Abstract

Fluctuations in the financial markets are fundamentally driven by information, and financial news articles constitute a primary mechanism through which such information is distributed to market participants. Whereas previous research on applying Natural Language Processing (NLP) to link financial news to market movements mainly focuses on predictive tasks, this thesis takes an explanatory approach by considering a price movement and related articles after it occurred. This thesis examined the capability of a Large Language Model (LLM) to generate a narrative explanation of significant market movements of Swedish listed stocks using financial news articles from Dagens Industri (DI), published between the 2nd of January 2026 and the 20th of April 2026. The methodological approach implemented an algorithm in three steps. Firstly, significant market movements were identified using a volatility-normalised z-score. Secondly, financial news articles were linked to Swedish listed stocks through Named Entity Linking (NEL) performed by Claude Haiku 4.5. Thirdly, Claude Sonnet 4.6 was prompted with the significant market movement and its matched articles and (i) score the articles relevance, (ii) generate a narrative explanation based on the relevant articles, and (iii) assign a verbalised confidence label to signal the quality of the explanation. The results were lastly evaluated through a structured form by financial professionals. The results demonstrated that the quality of the narrative explanations generated by the general-purpose LLM to a high degree aligned with the opinions of the financial professionals. However, the LLM tended to overestimate the relevance of the matched set of articles to the market movement.