ABSTRACT
This paper first evaluates the volatility modeling in the Bitcoin market in terms of its realized volatility, which is considered to be a reliable proxy of its true volatility. Based on the 5-minute return of Bitcoin, the proxy of its true volatility is computed as the sum of the squared intraday returns. To evaluate the performance of volatility modeling, this paper relies on MSE and QLIKE, which are the measures for making the forecast accuracy robust to noise in the imperfect volatility proxy, while different measures are also used for the robustness check. The empirically findings summarized as (1) the asymmetric volatility models such as EGARCH and APARCH have a higher predictability, and (2) the volatility model with normal distribution performs better than the fat-tailed distribution such as skewed t distribution.
Acknowledgments
We thank Ryo Ishida and Toshiaki Watanabe for suggestions. The views expressed in this paper are those of the author and not those of the Ministry of Finance, Japan. All remaining errors are our own. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors except for the database provided by the Ministry of Finance, Japan.
Disclosure statement
No potential conflict of interest was reported by the author.
Notes
1 The intraday data include microstructure noise; thus, the noise could be higher when the intervals are smaller, such as 1 min.
2 We use the skewed t distribution proposed by Hansen (Citation1994).
3 The dummy variable is constructed differently from the usual setup in the GJR-GARCH model because the volatility of Bitcoin has inverse asymmetry.
4 See Hattori and Ishida (Citation2019) for the detail of the market crash.
5 See the online appendix for details.