Impact of Google searches and social media on digital assets' volatility

The increasing digitalization and the use of financial technology have highlighted the significance of data mining and sentiment analysis. This study focuses on the influence of Google search activity on the volatility of digital assets, a topic of growing importance given the substantial price fluctuations in these markets. The study is motivated by observations of a strong correlation between Google searches and Bitcoin volatility, particularly before market crashes. Previous research has established a link between user sentiment on platforms like Twitter and digital asset market movements. However, this study goes further by examining the combined impact of Google search activity across various platforms (web, news, YouTube) and Twitter sentiment on the volatility of six major digital assets: Bitcoin, Bitcoin Cash, Ethereum, Ethereum Classic, Litecoin, and Ripple. These assets collectively represented 71% of the market capitalization in 2019, making them a representative sample of the broader market. The study's findings are expected to be valuable to both investors seeking to develop effective investment strategies and policymakers aiming to understand and manage the volatility of digital asset markets. The specific focus on the combined effect of Google search activity across different platforms, alongside Twitter sentiment, differentiates this study from previous research, offering a more comprehensive understanding of market dynamics.

The study reviews existing literature on behavioral finance theory, highlighting the role of investor psychology and external factors in influencing investment decisions and market volatility. Researchers such as Kahneman and Tversky (1979) and Thaler (1980) are cited for their contributions to understanding irrational investor behavior. The Efficient Market Hypothesis (EMH) theory, which posits that market prices accurately reflect all available information, is contrasted with behavioral finance theories that emphasize the role of sentiment and psychological factors. The literature review also examines prior studies exploring the relationship between Google search activity, social media sentiment (especially from Twitter), and digital asset prices. While some studies have found significant correlations between Google Trends and Bitcoin prices, others have focused on Twitter sentiment. A key gap identified is the lack of research examining the combined influence of various Google search sources (web, news, YouTube) on digital asset volatility. This study aims to fill this gap by analyzing these different search types alongside Twitter sentiment data.

The study employed a mixed-methods approach combining quantitative analysis with qualitative data collection. Data were collected from three main sources: Coinmarketcap (for digital asset prices), the Twitter API (for user tweets), and Google Trends (for web, news, and YouTube search data). Data were collected daily from September 1, 2019, to January 31, 2020. The study used the following steps: 1. **Data Collection:** Collected daily data on digital asset prices, Twitter sentiment (using the VADER lexicon), and Google Trends data for the six selected digital assets. 2. **Volatility Calculation and Tweet Cleaning:** Calculated daily returns for each asset using GARCH to model volatility. Tweets were preprocessed to remove noise and irrelevant information. Sentiment analysis was done using VADER, classifying tweets into positive, negative, and neutral categories based on a compound score. 3. **Sentiment Analysis:** Analyzed the cleaned tweets using the VADER lexicon to compute sentiment scores (positive, negative, neutral). 4. **Data Organization and Normalization:** The datasets were organized, and Z-transformation was applied to normalize the sentiment and Google Trends data, ensuring equal scale and variance for all variables in the analysis. The Augmented Dickey-Fuller (ADF) test was used to confirm the stationarity of the time series data. 5. **VAR Analysis:** A Vector Autoregression (VAR) model was implemented to analyze the relationships between Google search variables (web, news, YouTube), Twitter sentiment variables (positive, negative, neutral), and digital asset volatility. Lag selection was based on the Schwarz Criterion (SC), Akaike Information Criterion (AIC), and Hannan-Quinn (HQ) Criterion. The Granger causality test and Impulse Response Function (IRF) analysis were used to determine the causal relationships and the impact of shocks on volatility. The study used RStudio and Python software for data analysis and processing.

The key findings of the study are summarized below: * **Google Search Impact:** The VAR estimation revealed that Google search variables (web, news, and YouTube searches) significantly influenced the volatility of Bitcoin, Ethereum, Litecoin, and Ripple. The Granger causality test supported these findings, showing a causal link between search activity and volatility for these assets. The impulse response functions illustrated that shocks in web and news searches led to a significant increase in volatility for Bitcoin. For Ethereum and Litecoin, web searches were the primary drivers of volatility. For Ripple, both web and YouTube searches affected volatility. * **Twitter Sentiment Impact:** Twitter sentiment had a significant impact on the volatility of Ethereum Classic and Litecoin. Positive and negative sentiment exhibited different relationships with volatility, highlighting the complexity of sentiment's influence. * **Asset-Specific Responses:** The study found significant differences in how each digital asset responded to Google search activity and Twitter sentiment. For example, Bitcoin Cash's volatility was primarily influenced by its own past shocks and news searches, while other assets showed varying sensitivities to different types of search data and sentiment. * **VAR Model Diagnostics:** Diagnostic tests confirmed the stability of the VAR model and indicated no serial correlation in the residuals.

The findings highlight the importance of Google search activity as a significant driver of volatility in several major digital assets. This surpasses the influence of Twitter sentiment alone, implying that information seeking and access play a dominant role in shaping market dynamics. This aligns with behavioral finance theories, which emphasizes the non-rational aspects of investor behavior. The significant impact of Google searches across web, news, and YouTube platforms underscores the need for investors and policymakers to pay attention to online information consumption patterns as a gauge of market sentiment and risk. The asset-specific differences in the responses observed highlight the nuanced nature of these relationships. This suggests that understanding the unique characteristics and market dynamics of each digital asset is crucial for informed investment decisions and risk management. The overall conclusion is that Google search data provide valuable insights for predicting and mitigating volatility, complementing traditional financial indicators.

This study contributes to the understanding of digital asset market volatility by demonstrating the significant impact of Google search activity, surpassing the influence of Twitter sentiment in some cases. The findings highlight the importance of considering various information sources (web, news, YouTube) to gain a holistic perspective of market dynamics. Future research could explore the impact of other social media platforms or incorporate additional financial variables to provide a more comprehensive analysis. The results offer valuable insights for investors and policymakers to manage and mitigate volatility effectively.

The study is limited by the time period of the analysis (September 1, 2019 – January 31, 2020), which might not fully capture the complexity of long-term market trends. The analysis is focused on six major digital assets, and the results might not generalize to all cryptocurrencies. Moreover, the sentiment analysis relied on a rule-based lexicon (VADER) and did not employ more sophisticated natural language processing techniques that might better capture the nuances of online sentiment.