Dynamic volatility modelling of cryptocurrencies using time-varying transition probability Markov switching models / Tan Chia Yen

Tan , Chia Yen (2021) Dynamic volatility modelling of cryptocurrencies using time-varying transition probability Markov switching models / Tan Chia Yen. Masters thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (170Kb)
	PDF (Thesis M.A) Download (2012Kb)

Abstract

Motivated by the large price fluctuations and excessive volatility observed in cryptocurrency market, this research aims to model and forecast the volatility dynamics of cryptocurrencies. First, we adopt Bai and Perron (2003) multiple change point model by incorporating exogenous variables to determine the number and location of change points in the price series, return series and squared return series of Cryptocurrency Index, Cryptocurrency Index 30, and the top ten cryptocurrencies that are ranked according to market capitalisation. Results show that change points occur very frequently in the price series, followed by squared return series and return series. The change points are consistently observed in the periods from December 2017 to April 2018. Following these findings, we propose to use time-varying transition probability Markov-switching GARCH (TV-MSGARCH) models incorporated with logarithmic trading volume and Google searches series respectively as the exogenous variables to model the volatility dynamics of Bitcoin, Ethereum, Ripple, Bitcoin Cash and EOS. Extensive comparisons are carried out to compare the modelling and forecasting performances of the proposed model with the benchmark volatility models, which are the GARCH, GJRGARCH, TGARCH and MSGARCH. All of the volatility models are incorporated with three different error distributions, namely, normal, Student-t and generalised error. Results reveal that, regardless of error distributions, TV-MSGARCH models always predominate other volatility models for in-sample model fitting which are compared based on Akaike information criteria. Also, the Filardo’s weighted transition probabilities are also computed to assess the marginal contributions of time-varying transition probabilities of TV-MSGARCH models. Furthermore, it has been discovered that TV-MSGARCH model generally offers the best out-of-sample forecast evaluated based on quasi-likelihood loss function and assessed by using Hansen’s model confidence set. Lastly, different levels of long and short positions of value-at-risk for GARCH model, GJRGARCH model, TGARCH model, MSGARCH model and TV-MSGARCH models, all incorporated with Student-t distribution, are calculated and tested using several backtests.

Actions (For repository staff only : Login required)