What’s news with you: Price forecasting with global and ESG sentiment scores

We investigate the role of news sentiment, with a particular emphasis on ESG-related news, in forecasting stock prices. By leveraging deep learning models, we assess the predictive power of sentiment for the Dow Jones Industrial Average index. More specifically, we employ RoBERTa to compute sentiment scores from financial news and N-HiTS, a neural network architecture for time series forecasting, to predict next-day closing prices by exploiting sentiment scores and historical prices. Our findings reveal that news sentiment plays a crucial role in stock price forecasting, with models trained on comprehensive sentiment data from all news sources significantly improving predictive accuracy. We further investigate the role of ESG-related sentiment by isolating this subset of news and evaluating its predictive power, particularly in relation to downward price movements.