Research On Risk Of Bitcoin Based On Realized Volatility

Cryptocurrencies have become a world-class phenomenon,with governments,companies and investors facing enormous chal enges and opportunities.Bitcoin is the most widely known as a stable currency in cryptocurrencies.Compared to the payment function of bitcoin,people are more inclined to use bitcoin as an asset for investment and hedging risk.Therefore,it is very important to study the risk of bitcoin.Many scholars have used the daily data of bitcoin to construct different GARCH models to analyze their price volatility,and the research of high frequency data of bitcoin is also gradually being carried out in the past two years.The purpose of this paper is to study the models suitable for bitcoin volatility fitting,prediction and VaR calculation.In the theoretical part,the paper first introduces the yield form adopted in this paper and the four distributions that may be applied.The basic contents of the four types of GARCH family models are briefly introduced.The realized volatility and the realized range-based volatility in detail.The models of the realized volatility sequence has been introduced.Finally,the structure and parameter estimation methods of the realized GARCH model are described.Based on the theoretical introduction of the time series model and the model of predicting realized volatility,the paper mainly considers constructing different models from the following aspects in the empirical part.First,the GARCH model and the EGARCH model are constructed by using low-frequency data,and the HAR model and ARFIMA model are constructed by using high-frequency data,and the realized GARCH model is constructed by using lowfrequency data and high-frequency data at the same time.Second,considering the characteristics of the peak and thick tail of the yield,when constructing the GARCH model,the residual obeys five different distributions.Third,the HAR model,the ARFIMA model and the implemented GARCH model are constructed using different realized volatility sequences.The empirical results show that the EGARCH model has the best simulation effect in the GARCH family,but the realized GARCH model can be adapted to the rapid fluctuation of the volatility.The variation of the residual distribution can affect the simulation effect of the GARCH model.In the model for predicting the realized volatility,the fitting effect of the HAR model is better than that of the ARFIMA model,and the fitting effect of the realized rangebased volatility is better than that of the realized volatility,and the logarithmic form of the realized volatility sequence is fitted better than that of the general form of the realized volatilit y.In models for volatility forecasting,the ARFIMA-lnRRV model has the best predictive effect on future volatility.In models of VaR prediction,all models pass the backtest,which means all of the models can accurately predict VaR.The VaR predicted by the realized GARCH model has a relatively small overall,but the number of failures is large.For the VaR predicted by the realized volatility models,the failure times is small,but the value of part of them is large.The best predictive effect is the ARFIMA model using realized range-based volatility in logarit hmic form.