A new Twitter based credit rating model methodology.

Area of Science:

• Financial markets
• Computational linguistics
• Data science

Background:

• Traditional credit rating models often lack comprehensive predictive power.
• The emergence of social media platforms like Twitter presents novel data sources for financial analysis.

Purpose of the Study:

• To investigate the utility of Twitter data for predicting corporate credit ratings.
• To develop and compare methods for extracting sentiment scores and linguistic features from tweets for credit rating prediction.
• To assess the impact of incorporating Twitter-derived information on model performance.

Main Methods:

• Collected and analyzed tweets from/about NASDAQ and NYSE listed companies (2011-2019).
• Developed two distinct sentiment scoring methods using alternative word lists.
• Extracted linguistic features from tweets.
• Integrated sentiment scores and linguistic features into credit rating prediction models and compared their performance against baseline models.

Main Results:

• Models incorporating Twitter data demonstrated superior predictive performance compared to models without this information.
• Both sentiment scores and linguistic features derived from Twitter contributed positively to prediction accuracy.
• Different sentiment analysis approaches yielded varying degrees of predictive improvement.

Conclusions:

• Twitter data, encompassing sentiment and linguistic features, offers valuable insights for enhancing corporate credit rating predictions.
• The integration of alternative data sources like social media can significantly improve the robustness and accuracy of financial risk assessment models.
• Future research can explore more sophisticated natural language processing techniques for deeper insights from social media data.