Re-examining Bitcoin Volatility: A CAViaR-based Approach

ABSTRACT

The article aims to explore the heterogeneous feature in the determination of Bitcoin volatility using a Markov regime-switching model and test its forecasting ability. The forecasting methodology of the risk measurement of Bitcoin’s returns is based on the Conditional Autoregressive Value at Risk models (CAViaR) approach. Our results show that Bitcoin’s volatility is significantly related to the volatility of the crypto-asset’s return and the main determinants of volatility are speculation, investor attention, market interoperability and the interaction between speculation and market interoperability. In addition, we present evidence that investors’ attention is the main source of volatility. Speculation and the interaction term are related in a “U-shaped” form, whereas investor attention and market interoperability show a linear trend on the volatility of Bitcoin.

KEYWORDS:

• Bitcoin
• heterogeneous
• CAViar
• Markov regime-switching regression model

JEL:

• C44
• C51
• G17

Acknowledgments

The work was supported by Guangdong Natural Science Foundation (No. 2018A030313115).

Notes

1. Llorente et al. (2002) detrend turnover using t-200 instead of t-50. However, following Blau (2017), given our data limitations, we chose to only use 50days when detrending turnover.

2. We note that we add a small constant (0.00000255) to volume to account for days without trading volume. This constant is further shown to normalize the distribution of trading volume in Llorente et al. (2002) and Covrig and Ng (2004).

3. Most of VFAs trading around the world originate from China and the United States markets. Thus, the difference between the market prices of China and United States is the main determinant causing the fluctuation of VFAs return risk as well as the exchange rate between RMB and U.S. dollar.

4. See Engle and Manganelli (2004). In principle, the parameter G itself could be estimated; however, this would go against the spirit of this model, which is simplicity. We test different values of G, like 5, 15, 20, and get same result as G=10.

Additional information

Funding

This work was supported by the Guangdong Natural Science Foundation [2018A030313115].