A Study Of Crisis Contagion Based On Markov Switching Dynamic Conditional Correlation Analysis

It has erupte several economic (or financial) crises in the world since the nineties of twentieth century. And the research on crisis contagion is becoming a popular and difficult topic for two reasons. On the one hand, it is not only unknowable in advance but also not easy to judge the crisis impacts afterwards. On the other hand, there is no good approach to identify when to take rescue implementation and take what implementation. At what time the crisis of one country transmitted to the other countries and the duration of crisis contagion are the primary problems for the academics. It was blind for follow-up implementations if we do not solve those problems at first. There is close relationship between the investigation of market volatilities and the identification of crisis, and the analysis to contagion depends on correlation structures. So I investigate the relationship between crisis contagion, volatilities and correlations based on the relationship investigation between volatilities and correlations which are neglected in the research about second-order moment. However, it has important applications in assets pricing, portfolio selection, hedging, risk management and so on.Although the empirical tests for univariate autoregressive conditional heteroscedasticity (ARCH) model and generalized ARCH (GARCH) model on financial market volatilities have achieved a great success, it has encountered many problems for multivariate GARCH models including "curse of the dimension", positive definition to covariance matrix. Dynamic conditional correlation (DCC) model (Engle,2002) is not only convenient to estimate the parameters but also proper to investigate dynamic correlations between markets. In this study, I introduce hidden Markov chain to the framework of DCC and construct a direct way to investigate relations between volatility and correlation. And it can avoid the arbitrary division to the sample.Using Markov switching univariate GARCH model to investigate volatilities in different periods for the five major stock markets in chapter two, we have found those intervals dominated by volatility regimes. And the situation that all of those stock markets' volatilities switch to the highest level occurs only in periods of American subprime mortgage crisis in nearly two decades. I develop Markov independent switching DCC (MIS-DCC) multivariate GARCH model to investigate non-continuous change of the correlations in chapter three based on those conclusions of univariate GARCH and find that there is contagion to the other markets not only for American subprime mortgage crisis and European debt crisis but also for the other crises originated in America. Then, I further introduce one Markov chain driving both volatility and correlation to MIS-DCC in chapter four and find markets showing high volatility and high correlation in periods of American subprime mortgage crisis and European debt crisis. The method in chapter four provides a direct way to express comovement. To further analyze the appropriateness of one Markov chain driving both volatility and correlation and investigate the relationship between volatility and correlation, I introduce two Markov chains corresponding to volatility and correlation respectively in the DCC model in chapter five. Comovement is a special case of the joint regime between volatility and correlation. I find that high volatility is consistent with high correlation during American subprime mortgage crisis, but not the same for the other periods. The regime of high volatility with low correlation has almost not presented in stock markets. However, increase of volatilities will not always result in the increase of correlations, and increase of correlation may emerge before the increase of volatility sometimes. Markets need to gradually identify and filter the complex information in the early stage of the crisis, which results in the frequent switches in different volatility and correlation regimes. Therefore, decision of reactions to crisis should follow certain rules. It is necessary for countries to cooperate with each other in the crisis, since both the American subprime mortgage crisis and European sovereign debt crisis is systemic. The method to investigate the second moments developed in this study can be widely used in fields of asset pricing, portfolio selection, risk management and so on. In chapter six, based on investigation to impacts from external shocks in the previous chapters, I provide a preliminary identification for changes of the correlation structure caused by internal and external factors so as to further separate the impacts of the market's efficiency improvement from correlations.