Coupling News Sentiment with Web Browsing Data Improves Prediction of Intra-Day Price Dynamics

Area of Science:

• Computational social science
• Financial econometrics
• Big data analytics

Background:

• Big data revolution impacts societal understanding and economic forecasting.
• Information heterogeneity challenges predictive accuracy of semantic and statistical methods.
• Financial markets are complex systems with inherent unpredictability.

Purpose of the Study:

• To investigate if aggregated web user behavior can overcome information heterogeneity in financial market prediction.
• To assess the combined predictive power of news sentiment and user browsing activity on stock price movements.
• To explore the
• wisdom-of-the-crowd
• effect in financial forecasting.

Main Methods:

• Sentiment analysis of news articles.
• Tracking user browsing activity on Yahoo! Finance.
• Developing a weighted news signal based on clicks and sentiment.
• Granger causality analysis to test predictive power on stock returns.

Main Results:

• Individual sentiment analysis or browsing activity showed minimal predictive power.
• A combined signal, weighting news sentiment by user clicks, significantly improved forecasting.
• This aggregated signal Granger-caused hourly price returns for nearly 50% of analyzed stocks.
• Demonstrated a "wisdom-of-the-crowd" effect in identifying relevant and surprising news.

Conclusions:

• Aggregated web user behavior, particularly click-weighted news sentiment, enhances financial market forecasting.
• The synergy between user activity and news analysis is crucial for predicting stock price changes.
• This approach offers a novel method to leverage collective intelligence for economic system prediction.