A forecast comparison of volatility models using realized volatility: evidence from the Bitcoin market

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.