关键词: tv-VHAR模型, 联合网络关联度的频率分解, 系统性金融风险, 已实现波动率

Abstract:

In the past decade， China has experienced two critical events - the 2015 stock market disaster and the coronavirus disease 2019 （Covid-19）， which have had a great impact on the financial markets. Through the comparison of the two crises， it is found that the impact of the stock market disaster on financial markets is much stronger and longer than the coronavirus disease 2019， although the financial markets experienced sharp declines in both crises. It matters to both governments and academia to find out the reasons behind the differences in the causes of the two crises to financial risks and further figure out the sources of systemic risks.Taking the high-frequency data of financial stocks as object， a locally stationary non-parametric time-varying vector HAR model （tv-VHAR model） under high dimensions is constructed firstly in this article by assuming that the parameters of the vector Heterogeneous Autoregression Model （HAR model） are functions of time t/T. On this basis， the estimation problem under the Curse of dimensionality is solved by applying the Quasi-Bayesian Local Likelihood methods to the tv-VHAR model. Secondly， the frequency component of the joint connectedness is proposed in this article to increase the measurement accuracy of the systemic financial risks by revising the Baruník and K?ehlík （2018） model. Finally， the systemic financial risks in China are systematically analyzed and is proved to have the following 5 features：（1） From October 2010 to October 2020， the total joint connectedness of the financial system risks in China showed a relatively high value and fluctuated continuously.（2） In normal times， high-frequency components account for a larger proportion of the total joint connectedness， followed by the medium-term components， and finally the long-term components. （3） During the crises， the proportion of the high-frequency components declines rapidly， while that of the medium- and long-term components rises rapidly， which sometimes exceeds the former. （4） It is found that the Covid-19 exerted less influence on investors' mid- to long-term belief changes， and the influence lasts for shorter while analyzing in the perspective of frequency， though the total joint connectedness of the critical event are similar. （5） It is found that large securities companies and joint-stock commercial banks mainly act as risk communicators and occupy a dominant position in financial network risk contagion. However， the four major state-owned commercial banks mainly act as risk receivers， and can play as a stabilizer in the financial system as they have the ability to resist risks. In addition， small securities companies and other financial institutions also act as risk receivers.

Key words: tv-VHAR model, the frequency of joint connectedness, financial systemic risk, realized volatility