Predictive capacities of social media in the financial market: ARX-GARCH model

Motivation: Social media platforms have emerged as a new data source for social sciences. The data extracted from them, known as big social data, are characterized by their complexity and are described within the ‘V’ big data model. Literature has demonstrated the influence of investor activity, as captured by BSD, on the stock market. Modeling the stock market using Twitter data allows for the capture of real-time market sentiments, potentially enhancing the accuracy of financial forecasts. The value of such models lies in their ability to identify subtle yet significant signals that traditional methods might overlook. Aim: The aim of this study is to explore the predictive capabilities of Twitter sentiment on financial markets, specifically focusing on the application of ARX-GARCH models to analyze the impact of both, negative and positive class of emotions on market volatility. Results: Incorporating sentiment variables into the ARX-GARCH models did not significantly enhance their predictive capabilities. While sentiment variables did not broadly improve model performance, certain variables demonstrated statistical significance at various lag levels. This indicates that some sentiments might have a delayed impact on market returns, though the overall effect size was small. Among the sentiment indicators analyzed, those based on n-gram analysis and the bullish index outperformed others, including the volume of individual emotions like anger, fear, and sadness.