Stock Price Trend Prediction Based on News Headlines

Abstract

Behavioral finance shows market sentiment affects market trends. And news headlines, as information carriers with emotional elements. Analyzing the day’s trending news headlines enables the prediction of the day’s market trends. This study focuses on using news headlines to forecast stock price trends, aiming to capture sentiment’s impact on overall stock price changes macroscopically. It uses 25 most viewed Yahoo Finance news headlines (2000–2016) as text data, combined with corresponding Dow Jones Industrial Average (DJIA) fluctuations. Algorithms like Random Forest, XGBoost, Logistic Regression, Naive Bayes, and Support Vector Machine (SVM) were used to learn the mapping between headline features and stock price movements for prediction. Among these models, Logistic Regression performed best: highest accuracy (85.71%), most balanced predictive ability for rises/falls (0.01 F1-score difference), and fast training speed (1.26 seconds). And all models achieved over 83% accuracy, verifying news headlines’ sentiment value for stock price prediction and providing a reference for further research.