University of the Aegean
Empirical analysis of contagious financial crisis: the key role of multivariate dynamic conditional correlations, copula functions, markov regime switching models and machine learning
Abstract
dc:descriptionThis doctoral thesis studies the volatility spillover effects from shock events of financial crisis. To achieve this goal, individual research problems are empirically analyzed, aiming at a better understanding of the subject. The empirical analysis is divided into five parts/researches which cover fields of financial contagion in the global financial system focusing on the most recent crisis events. The first part investigates the volatility spillover effects from South to North Eurozone during the Sovereign Debt Crisis. I propose the Dynamic Conditional Correlation (DCC) model and the BEKK model to identify possible linkages during the period 2005-2015. These two models are the most appropriate in quantifying the correlations and the variance-covariance matrices between asset markets. The findings showed that both models behave perfectly and are flexible in presenting the spillover effects. However, when it comes to figure illustration of conditional correlations, the ADCC model seems to fit better. Additionally, Spain and Italy are those countries which can produce significant damage on all Northern strong economies while Greece’s negative shocks are capable of co-moving the French index. France, on the other hand, is the most correlated country with the South Eurozone. The findings support significant interesting contribution to the literate of contagion in capital markets.The second part studies the spread of the Subprime Crisis and the European Sovereign Debt Crisis from Eurozone countries to the real economy by examining ten sectors in major developed and emerging stock markets. First, I employ Cappielo's et al. (2006) model and copula functions to detect and cross-check the correlations and the contagion thereafter. Second, I uncover evidence of correlation behavior between policy uncertainty indexes and stock market returns. The results demonstrate that no country and sector was immune to spillover effects, highlighting the limited effectiveness of policy makers for both the Subprime Crisis and the European Sovereign Debt Crisis. The empirical application provides evidence of significant volatility and tail dependence from the financial sector to many real sectors in the U.S. economy. Additionally, there is clear evidence that certain sectors, particularly Healthcare, Telecommunications, Utilities and Technology, were less severely affected by the crisis, as observed by Baur (2011).The third part applies a dynamic conditional correlation DCC model to investigate the volatility spillovers and the interdependence between the Greek Debt crisis and the Cypriot financial crisis. The subprime mortgage crisis created large shocks to most major economies. Shortly after, the new economic framework obliged Greece to decrease its high deficit and public debt. Subsequently, the Cypriot financial crisis occurred after the credit event in Greece. Possible contagion channels were created after the Cypriot government's decision to impose a bank deposit levy in return for the bailout. The findings support the existence of contagion during the period 2008-2013. By 2015, the financial environment in both countries was quite different and this is evident in correlations in the last two years. Observing the behaviour of the correlations, I conclude that both economies are being “treated like lab rats” to test for austerity measures in order for the rest of the systemic countries to be secured from a possible transfer of the crisis from Greece to their own state.The fourth part studies the effects of the June 2016 United Kingdom European Union membership referendum and the subsequently triggered article 50 on 43 major developed and emerging stock markets. I detect which countries are vulnerable to the transmission of the shock and which others have immunity during the period of turmoil. Specifically, on a bivariate basis, I use dependence dynamics through copulas with regime switching of Silva Filho et al. (2012) using intraday data returns to identify contagion among stock markets. The empirical results add significant evidence to the literature on the financial contagion from the Brexit to other countries for a very large sample thus far. Evidence shows that the methodology identified immediate financial contagion produced from the referendum results. However, the contagion was not sufficiently significant given the short duration. I suppose that the negative reaction in the markets was overall small and held only for a short period. In general, results showed instant financial contagion due to the shock and increased uncertainty from the referendum results; however, the shock and uncertainty were very limited, because a few days after the polling day, most stock exchange markets had fully recovered their losses. Additionally, no significant contagion produced from the trigger of article 50. The approach provides significant information not only to policymakers but also to investors about the stock market’s reaction to the expected Brexit. Lastly, the fifth part of the research studies on ‘early warning systems’ (EWS) by investigating whether measures of contagion risk, which are based on modeling the global financial system as a network, can serve as early warning indicators and improve the performance of standard crisis prediction models. In doing so, I combine network analysis and machine learning algorithms to create an accurate model for predicting the vulnerable periods of contagion during shock events and crisis periods in stock exchange markets. The empirical results add significant evidence to the literature since few prior studies have focused on the network topologic metrics in the financial networks. Regarding the financial networks, they are interpreted by a significant percentage of the actual geographic location of the markets. In addition, the volatility of the correlations largely follows the volatility of the centralities where significant shocks in the correlations trigger huge volatility in all centralities. Based on this evidence I use hypothesis testing to determine the possibility of contagion risk inside the network. The results verify the presence of contagion risk on the dates where I observe a significant increase in the correlations and centralities. Regarding the empirical results of the machine learning approach to predict and forecast the contagion risk inside the financial network, the accuracy of the quadratic Support Vector Machine reached 98.8%, making the predictions fairly accurate. The model provides substantial information not only to policymakers (institutions) but also to investors about employing the financial market network as a useful device to improve the portfolio selection process by targeting a group of assets based on their centrality.
Degree
thesis:*- Grantor dc:publisher
- University of the Aegean
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kampouris, Elias
- Καμπούρης, Ηλίας
Subjects
dc:subject × 22- Χρηματοποιστωτική κρίση
- Μεταδοτικότητα
- Υπόδειγμα DCC
- Υποδειγμα asymmetric BEKK
- Εξισώσεις πολυμετάβλητων κατανομών
- Υποδείγματα μεταγωγής της κατάστασης
- Ανάλυση κοινωνικών δικτύων
- Προβλέψεις
- Μηχανές εκμάθησης
- Financial crisis
- Contagion
- DCC
- Asymmetric BEKK
- Copula functions
- Regime - switching models
- Social network analysis
- Forecasting
- Machine learning
- Κοινωνικές Επιστήμες
- Οικονομικά και Επιχειρήσεις
- Social Sciences
- Economics and Business
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44880