Predicting U.S. Stock Sectors Using AI-Driven News Sentiment Analysis

This paper explores the predictability of U.S stock sectors using an AI-driven sentimentanalysis model. This is a logistic regression model that analyzes news headlinesand predicts stock fluctuations. Using approximately 11,000 GDELT news headlinesfrom 2017 to 2025, the model generates trading signals, which are then backtested bysector. The results of the model show the Sharpe ratios, a reliable measure of an investment’sreturn relative to its risk, for each sector. The model has both an unbalancedand a balanced mode: the unbalanced mode predicts each sector based on its frequencyin the news, whilst the balanced mode predicts each sector with an equal number oftrades per sector. Our results show that the technology and financial sectors appear tobe consistent and highly correlated (Sharpe ratios of 0.6 to 1.1) with news headlines.Conversely, the healthcare and real estate sectors have negative Sharpe ratios, indicatingpoor tradability based on news headlines. In summary, AI-driven news sentimentanalysis may be dependent on sectors and their frequency in news headlines.