Understanding the internal decision-making process of NLP models in high-stakes domains such as the financial sector is particularly challenging due to the complexity of domain-specific terminology and the need for transparency and accountability. Although SHAP is a widely used model-agnostic method for attributing model predictions to input features, its standard formulation treats input tokens as independent units, failing to capture the influence of collocations that often carry non-compositional meaning, instead modeled by the current language models. We introduce C-SHAP, an extension of SHAP that incorporates collocational dependencies into the explanation process to account for word combinations in the financial sector. C-SHAP dynamically groups tokens into significant collocations using a financial glossary and computes Shapley values over these structured units. The proposed approach has been evaluated to explain sentiment classification of Federal Reserve Minutes, demonstrating improved alignment with human rationales and better association to model behaviour compared to the standard token-level approach.

This paper examines the relationship between news coverage and the European Central Bank’s (ECB) interest rate decisions. In particular, the hypothesis of a linear relationship between financial news and ECB indications regarding interest rate variations is investigated by leveraging state-of-the-art large language models combined with domain experts and automatically selected keywords. The analysis revealed two key findings related to how news contents can signal the ECB’s decisions to raise or lower interest rates: (1) Sentence Transformer models, when combined with domain-specific keywords, exhibit a higher correlation with ECB decisions than state-of-the-art financial BERT architectures; (2) employing a grid search strategy to select subsets of informative keywords strengthened the relationships between news contents and ECB’s decisions, highlighting how media narratives can anticipate or reflect central bank policy actions.

The purpose of this paper is to extract market signals for the major currencies (EUR, USD, GBP, JPY, CNY) analyzing the Federal Reserve System (FED) minutes and speeches, and, consequently, making suggestions about going long/short or remaining neutral to investors thanks to the causal relationships between FED sentiment and currency exchange rates. To this purpose, we aim to verify the hypothesis that the currency market dynamics follow a trend that is subject to the sentiment of FED minutes and speeches related to specific relevant currencies. The proposed paper has highlighted two main findings: (1) the sentiment expressed in the FED minutes has a strong influence on financial market predictability on major currencies trend and (2) the sentiment over time Granger-causes the exchange rate of currencies not only immediately but also at increasing lags according to a monotonically decreasing impact.