Eurobond Correlation Network: fearing the Grexit

Eurobond Correlation Network: fearing the Grexit

Over the last five years, fears over a Greek debt and a financial contagion spreading to PIGS countries have dominated Euro Bond markets. Looking into 10 year bond price correlations across the same period, we are able to observe that, in this time, the Greek markets have already decoupled from those of other EU states. Greek debt is no longer significantly correlated with any other country’s with the exception of Spain, Ireland and Portugal.

Recently, advances made in the study of correlation networks lead to new methods of analysing the current dynamics in the European financial market. Correlations between assets carry a great deal of information but the data can be difficult to understand. Representing assets correlations as networks and applying techniques like a Minimum Spanning Tree can be used to reduce the data dimension and to differentiate meaningful signals from the noise. Using correlation network analyses and visualization techniques provided by FNA, we will examine the change in relationship between European bond markets in the lead up to last weekend’s Greek referendum.


GreeceBondReturns_cropped2In the figure above, the length of links corresponds to the strength of correlations and statistically insignificant correlations are shown as dashed lines.

Ever since the major possibility of default arose in 2010, Greek debt has shown a cascade of negative returns and increasing volatility.

The figure to the left shows that bond prices dropped for two consecutive years after 2010, and displays an almost quadruple increase in volatility with returns outside of confidence level of normal distribution, four to six times more frequent than expected.

In the autumn of 2014, Greek yield spreads increased again.

The 2015 elections and subsequent difficult negotiations between Greece and the creditors dominated the news until the referendum on July 5th. The morning following the vote, European private bank shares went down.




International stock markets reacted strongly on Monday, July 6th, showing renewed fear of Greek exit from the Eurozone. However, worries concerning the Chinese equity markets impacted emerging markets equity indices far more.


Motivated by the many twists and turns during these negotiations and the immediate market reactions, Peter Schwendner and his colleagues were looking for a method to analyse the correlation dynamics in short intervals.

Due to the market fluctuations, it seems necessary to filter for the statistically most relevant correlations between sovereign bonds. The universe of Eurozone bonds is not a “closed system”, as the risk perception depends on the external news flow. Therefore, modelling causality is not adequate.

They implemented a correlation influence network as the correlation influence looks at simultaneous yield changes, but compensates the correlation between two markets for correlations inflicted from a third market of that universe, using partial correlations. (Schwendner, P., Schüle, M., Ott, T. and Hillebrand, M. European Government Bond Dynamics and Stability Policies: Taming Contagion Risks (May 24, 2015). Available at SSRN:

The resulting noise-filtered network shows blue arrows for positive correlations that dominate the correlations between the market at the arrowhead and all other markets, and red arrows for negative correlations. The colour saturation and arrow thickness symbolise the influence strength. We interpret the blue arrows as reinforcing movements, and the red arrows as shearing forces within the Eurozone. However, the correlation influences do not show a causal relationship, but a sensitivity of the relative bond yield movements with respect to external shocks.


The time span from June 8th, 2015 to July 3rd is depicted.

The absolute levels of the bond yields did not yet move much in these four weeks, but the correlations already show large structural breaks. If we look at the spreads between Italy, Portugal and Germany they are not showing much movement up until the last days of June. However the dynamics of correlation influences predicts the widening of the spreads a couple of weeks in advance.


Each of the four graphs below shows hourly correlations in a one-week time window. Only influences with absolute values above three standard deviations of their statistical noise are displayed.

The time evolution in those four weeks shows the decreasing trust of investors in a successful outcome of the negotiations between the Troika and Greece. It is interesting to note that Greece itself is not part of the network, as the fluctuations there are always stronger than the absolute value of the influences.

Week 1: June 8th –June 12th, 2015 <p><span><br><br>On June 4th Greece asks the IMF to postpone the installment due June 5th until the end of the month.</span> <span><br><br>The following week only positive correlations are visible, as the negative correlations were weaker than their noise level.</span> Week 1: June 8th –June 12th, 2015On June 4th Greece asks the IMF to postpone the installments due June 5th until the end of the month.The following week only positive correlations are visible, as the negative correlations were weaker than their noise level.
Week 2: June 15th - June 19th, 2015 Week 2: June 15th – June 19th, 2015
In the next week negative correlation influences between Germany and the periphery countries Portugal and Spain develop.
Week 3: June 22nd – June 26th, 2015 Week 3: June 22nd – June 26th, 2015In the third week, the negative influences spread to Italy.
Week 4: June 29th-July 3rd, 2015 Week 4: June 29th-July 3rd, 2015Following the eventful weekend when on June 27th Prime Minister Tsipras announced a referendum on a bailout agreement, to be held on July 5th and on June 28th the Greek parliament approved the referendum we see the for the next week as we await referendum results negative influences that have developed from the core countries to periphery persist.On July 5th, the Greek referendum further sharpened the political situation.Increased volatility is only to be expected.


pschwender Peter Schwendner
Guest writer Peter Schwendner is a Senior Lecturer at the Institute for Wealth and Asset Management at ZHAW School of Management and Law, Zurich, Switzerland. His research interests are financial markets, asset management and network analytics. Peter received a PhD in Physics in 1998 for his research at Max Planck Institute in Goettingen. He has 15 years’ work experience in the financial industry as a head of quantitative research at Sal. Oppenheim and as a partner at Fortinbras Asset Management, developing investment products.


Eugene Neduv

Eugene Neduv

Risk Researcher at Centre for Risk Studies
Eugene Neduv is an expert in quantitative methods in finance and a risk management professional. His interests include network analytics, volatility trading and portfolio optimisation. Eugene graduated from Columbia University in 2002 with a PhD degree in Mathematics and continued as a postdoctoral researcher at the Brazil Institute for Pure and Applied Mathematics and Humboldt University in Germany. Eugene has 10 years' work experience in financial analytics and risk management software and as an independent consultant for several hedge funds in New York and Sao Paulo.

Would you like to comment?

This site uses Akismet to reduce spam. Learn how your comment data is processed.

We are using cookies on our website

Are you happy to accept our analytics cookies, which help us learn about our website visitors and their use of this site? Learn how to disable all cookies.