Cryptocurrency Volatility Forecasting Using Google Trends and GDELT

Bitcoin, Ether, Litecoin and XRP are among the largest cryptocurrencies and together have a market capitalization that constitutes a substantial portion of the digital asset market. At the same time, the cryptocurrency market differs from that of traditional financial assets because it has greater price volatility, which makes constructing more accurate forecasts of daily volatility for these assets especially useful. A more accurate estimate of the daily volatility of a financial asset is an important aspect of trading strategies and arriving at a risk management stance. Volatility forecasts are also a factor in the regulatory policy that financial authorities pursue for the digital asset market. The study collects supplementary non-financial information from the internet and feeds it into an HAR-RV model to improve the accuracy of realized volatility forecasts for the four largest cryptocurrencies. Variables are then derived from GDELT and Google Trends data sources as exogenous indicators of public interest and sentiment. These variables are incorporated into a standard HAR-log-RV model, and various versions of the model for each asset are applied to a moving window from 1 January 2018 to 23 June 2024 to yield more than 2,000 out-of-sample one-day-ahead forecasts on which loss functions are calculated. The set of out-of-sample model errors for each asset is tested using the MCS procedure to select the set of statistically superior forecast models. The data sources proposed in the study and the variables constructed from them significantly improve realized volatility forecasts for the four largest cryptocurrencies.