Identification of Bitcoin Volatility Drivers Using Statistical and Machine Learning Methods

The data start from August 2, 2017 and end on March 31, 2023.It includes logarithms of realized variance for BTC/USD (Bitcoin), logarithms of first lags for daily, weekly, and monthly realized variances and exogenous variables which are categorized into three main groups: (1) BTC-specific factors, (2) financial markets, and (3) market and policy uncertainty. The first group includes factors such as trading volume, the number of transactions per day, average block size per day, market capitalization, average hash rate per day, the number of unique addresses per day, and Google searches for Bitcoin. The financial markets group is the most extensive and includes stock indices (S&P 500, Nasdaq Composite, Euro STOXX 50, FTSE 100, Nikkei 225, Shanghai Composite), currency exchange rates (EUR/USD, JPY/USD, CNY/USD), the nominal broad U.S. dollar index, U.S. 2-year and 10-year bond yields, and commodities (NYMEX light sweet WTI crude oil, gold spot, NYMEX Henry Hub natural gas, and Bloomberg Commodity Index). The third group, market and policy uncertainty, includes the CBOE Volatility Index, EURO STOXX 50 Volatility Index, CBOE Gold ETF Volatility Index, an economic policy uncertainty index based on newspapers, a Twitter-based market uncertainty index, a Twitter-based economic uncertainty index, a geopolitical risk index based on newspapers, a risk aversion index based on financial variables, an uncertainty index based on financial variables, and an infectious disease equity market volatility tracker based on newspapers. The data comes from various sources, which are detailed in the paper in Table 2. For the variables in the first and third groups, we apply logarithms and first differences of logarithms. For the financial series in the second group, two measures are calculated: logarithmic returns and the Parkinson estimator of variance. We use the following symbols to denote these transformations: ln, delta, r, and v for logarithm, first difference of logarithms, logarithmic return, and the Parkinson estimator of variance, respectively.