[Tech Review] Fear, Anger, and Finance: Do Emotions in Rumors Actually Move Markets?

Executive Summary

TL;DR: Leveraging the 2016 Indian Demonetization as a case study, this research proves that digital rumors are deeply "colored" by negative emotions like anxiety and anger. While these emotional markers are excellent for identifying misinformation, the study reveals a surprising disconnect: the "emotional temperature" of rumors on Twitter had almost zero impact on the actual movement of the Indian stock market.

Positioning: This work bridges the gap between Psychological Rumor Theory and Behavioral Finance, testing whether the emotional "noise" of social media can break the "Efficient Market Hypothesis" during a localized economic crisis.

Problem & Motivation

Rumors are born from two ingredients: ambiguity and importance. When the Indian government invalidated 86% of the country's cash overnight, both were present in spades.

Existing research often treats rumors as structural phenomena (how they spread through a graph), but ignores the why—the emotional pulse that makes a user click "Retweet." The authors argue that if rumors are psychological, then the text's emotional content should be the ultimate "fingerprint" for detection. Furthermore, if markets are influenced by sentiment, these high-emotion rumors should theoretically tank or buoy stock indices.

Methodology: The Emotional Fingerprint

The researchers moved away from manual labeling and adopted LIWC (Linguistic Inquiry and Word Count) to categorize tweets into four "Emotional Groups":

1. Deception & Doubt: Measured through word count (WC), cognitive processes (cogproc), and negative affects.
2. Overall Tone: The general positivity or negativity.
3. Personal Involvement: Usage of auxiliary verbs and first-person pronouns ("I"), indicating personal stakes.
4. Personal Importance: Keywords related to "Money" or "Death."

The Rumor Identification Model

The study defined a rumor as any post that lacked an explicit source, targeted a specific topic (Demonetization), and carried topical relevance. They then applied a Logistic Regression to see which emotional variables predicted these rumors.

Figure 1: Conceptual framework linking emotional content to rumor classification and subsequent market impact.

Experiments & Results: Detection vs. Impact

1. Emotional Detection of Rumors

The results were striking. Anger, Sadness, and Anxiety were all statistically significant predictors (p < 0.0001) of rumors. Interestingly, "Personal Involvement" (self-references) was a massive indicator, suggesting that rumors are often framed as personal experiences or concerns.

Table 1: Logistic regression results showing which emotions are the strongest predictors of rumor-mongering.

2. The Great Disconnect: Market Impact

Despite the high emotional intensity of these rumors, the correlation with the Nifty and Sensex indices was negligible. Only the FMCG (Fast Moving Consumer Goods) sector showed a slight positive correlation with "Personal Involvement" emotions, likely because demonetization hit daily cash transactions for consumer goods hardest. However, across the board, the emotional volatility of rumors failed to "herd" the financial markets.

Critical Analysis & Conclusion

Takeaway

This paper serves as a reality check for the "Social Media Sentiment" hype. While emotions are powerful tools for identifying rumors (a boon for content moderators), their power to disrupt established financial markets might be overstated in existing literature.

Limitations

• Time Lag: The study used daily averages. Financial markets might respond to rumors in minutes or seconds, a granularity not captured in this daily-level Pearson correlation.
• Context Specificity: Demonetization was a unique "policy shock." In other contexts—like a corporate scandal or a short-squeeze (e.g., GameStop)—the emotional link to the market might be significantly tighter.

Future Outlook

Future research should integrate LLM-based sentiment analysis (which captures nuance better than dictionary-based LIWC) and use high-frequency trading data to see if rumors cause "flash" movements that are smoothed out by the day's end.