This study performs sentiment analysis on financial news to assess the impact of sentiments on financial indices using an integrated framework encompassing sentiment analysis and a causal discovery algorithm based on Gaussian process regression to deal with non-linear relationships. Specifically, a news sentiment database was created by extracting financial news articles on various topics sourced from the investing.com web page. The news articles were subsequently inputted into the FinBERT model in order to extract their respective sentiments. The news dataset was combined with financial indicators, specifically the S&P 500 index, dollar index (DXY), and Brent crude oil. Subsequently, the causal discovery model, namely LPCMCI, was employed to conduct causal discovery on the variables of interest. Different from previous works, this study gives a wide perspective on the connections between news and financial behavior, accounting for lag dependencies and the direction of the impacts. The main results indicate a reciprocal relationship between news sentiments and financial variables. In this context, the S&P 500 index can be regarded as a source of Stock Market news together with Commodities news sentiments, where a contemporaneous and lagged effect is observed towards Brent returns. In the opposite direction, news sentiments regarding the Currencies subject appear as a driver of DXY returns. Furthermore, it was observed that various subjects exhibit distinct patterns of interaction with individual financial indicators. The findings of the study clarify what was previously suggested in the literature and provide fresh perspectives that have not been investigated thus far.
Complete tool for submission, evaluation and publication of proceedings for scientific events.
