The effect of OMT announcements on
sovereign CDS spreads in peripheral
European countries
Anouk Hiensch
ABSTRACT
Since the financial crisis erupted in 2007 credit risk for several European countries increased enormously. Credit default swaps spreads in European peripheral countries have experienced an increasing trend since the eruption of the financial crisis. Nevertheless, around 2012 the CDS spreads have fallen. Between 26 July and 6 September 2012 the European Central Bank announced the OMT program that offered European countries to purchase their short-term bonds in order to bring down their interest rates. A linear regression could not find a significant difference in the CDS spreads in peripheral European countries before and after the OMT announcements, and therefore the fall in the CDS spreads cannot be explained by the announcements of the ECB. The regression did find a significant difference in the CDS spreads for some months during the announcements of the ECB.
Supervisor: Isabelle Salle Universiteit van Amsterdam
BSc Econometrie en Operationele Research June 28, 2015
1 Introduction 2
2 Theoretical part 3
2.1 The EU Debt Crisis 3
2.2 Credit Default Swaps 5
2.3 OMT announcements 6 3 Methodology 6 3.1 Variables 7 3.2 Models 7 3.3 Tests 8 3.4 Hypothesis 9
4 Results and analysis 9
4.1 Test Results 9 4.2 Descriptive statistics 10 4.2.2 Results Italy 16 4.2.3 Results Portugal 18 4.2.4 Results Spain 20 4.3 Analysis 22 5 Conclusion 23 References 24 Appendices 25 Appendix I 25 Appendix II 26 Appendix III 27 Appendix IV 29 Appendix V 31 Appendix VI 33 Appendix VII 36
1 Introduction
The global financial crisis that has erupted in 2007 increased government debts as a result of their attempt to save their banking system from collapsing. This crisis laid the foundation of the Eurozone debt crisis and has increased credit risk enormously for several European countries. Developed countries were considered not to be susceptible to risk, but the financial crisis proved otherwise and showed that also these countries can default on their liabilities. This understanding is observed in sovereign credit default swaps (CDS) spreads in periphery European countries, as they have experienced an increasing trend since 2007 and have been fluctuating ever since. Figure 1 in Appendix I shows the development of the CDS spreads for four periphery countries in Europe, namely Spain, Ireland, Italy and Portugal. CDS spreads are used as a measure of risk assessment of the probability that a country will default, and therefore it is interesting to derive the underlying drivers of these spreads.
Fundamentals such as Debt to GDP ratio and sovereign credit ratings partly explain the movements in European CDS spreads in the period of the financial crisis (Chiarella, Ter Ellen, He, & Wu, 2012). Nevertheless, Chiarella et al. (2012) state that trend following speculation also plays a significant role in explaining the fluctuations of the CDS spreads. De Grauwe & Ji (2013) also present this idea.They find that the movement in CDS spreads is associated with negative self-fulfilling market sentiments. The self-fulfilling aspect in this matter are the investors, who fear default and act in such a way that default is more likely to appear.
The sovereign debt crisis of 2010 had increased the concerns of investors about the survival of the euro. International investors became increasingly anxious about the high national debt that several European countries had developed, and from this moment on intervention activities by the ECB were concentrated at secondary sovereign bond markets.
On July 26, 2012, President Draghi of the ECB stated that they were prepared to do “whatever it takes” in order to prevent the Euro from collapsing. The Governing Councel of the ECB announced on 2 August 2012 that it would undertake outright monetary transactions (OMTs) in the secondary, sovereign bond markets. The program was established in order to safeguard the singleness of the monetary policy and an appropriate monetary transmission process. On 6 September 2012 the technical aspects of the OMT program were announced and it was officially launched.
The announcements by the ECB of policy changes in the period July to September 2012 create groundwork for the investigation on whether country specific news influences CDS spreads. This study aims to explore what the effect of the OMT announcements of the ECB is on sovereign CDS spreads in peripheral European countries.
The study investigates the pricing of CDS contracts on a daily basis over the period from January 2009 until December 2014. Five countries that are mostly affected by the European debt crisis are included in the data, namely Spain, Ireland, Portugal, Italy and Greece. The main aim of this event study is to assess the effect of the OMT announcements on sovereign CDS spread by regressing the credit default swap spreads on event dummies. The regression is conducted in two stages. First, the pre-announcement movement of the CDS spreads, including data from October 2008 until July 2012, is compared to the
post-announcement movement of CDS spreads, which possesses data from August 2012 to May 2015. The second stage of the regression attempts to find if there exists a sizable and significant change in CDS spreads during the OMT announcement, since the OMT announcements covered some period of time.
The first part of section 2 elaborates on the theory behind the European debt crisis, the second part of section 2 gives insight in CDS spreads and drivers of these spreads and the last part of section 2 presents information on OMT announcements and presents existing literature on the effect of OMT announcements on macroeconomic variables and CDS spreads. Section 3 presents the methodology and model concerning the research. Section 4 presents the results of the regression and applied tests and analyses these results. Lastly, section 5 exists of a conclusion that summarizes the main findings of this research.
2 Theoretical part 2.1 The EU Debt Crisis
The Eurozone debt crisis was the trigger for investors to question the survival of the European currency and the financial stability of European countries. In order to further investigate the drivers of sovereign CDS spreads, which are a representative of risk assessment of a country, a fine understanding of the EU debt crisis is necessary.
The financial crisis that occurred in 2007 in Europe was preceded by an increase in private debt as a percentage of GDP for several European countries. The increase in this private borrowing was mainly caused by the introduction of a common currency in Europe,
which allowed banks to raise funds in their own currency, removing the risk of exchange rates (Kolstad, 2013). Lower interest rates where another factor that increased private borrowing. When the financial crisis in the US erupted in September 2008 with the fall of the Lehman Brothers, this quickly spread to Europe and especially affected the European banking sector. High lending to the private sector led to huge liabilities on part of the banking sector and governments stepped in to help these banks with large debts. The bank rescue packages made European governments build up large amounts of debt and all countries in Europe
experienced an increase in debt level. Figure 1 shows that Portugal, Ireland, Italy, Greece and Spain had the most extreme increase in government deficits following the financial crisis.
Figure 1 Government debt in % of GDP 2003-‐2012
The deficit that Greece had build up, turned out to be a lot more than they previously stated. In October 2009 the Greek government reported a government deficit of 12.7 percent, which was twice as high as reported earlier and this development is evaluated as the building stone of the European debt crisis. The Greek deficit quickly started to increase even more as a result of slow economic growth, decreased tax revenues, less spending by the government and increased financing costs (Kolstad, 2013). It did not take long for other European countries to enrol in the same situation as Greece and generate an increasing government debt. Sovereign risk for several European countries increased enormously and this was displayed by a fast increase in sovereign CDS spreads, as mentioned in the introduction and seen in Figure 1. The five periphery European countries experienced the highest increase in sovereign CDS spreads.
2.2 Credit Default Swaps
CDS spreads represent the risk assessment of the probability that a country will default and create a clear insight in sovereign risk of a country. Therefore, it is of importance to derive theory on CDS spreads and explore factors that influence these spreads.
A credit default swap is a financial contract where the seller of the CDS insures payment of some reference loan if this loan defaults, in exchange for a series of payments made by the buyer of the CDS. This payment is referred to as the CDS spread and represents the price that the buyer is willing to pay for default protection. In case of a sovereign CDS contract the reference loan is known to be the sovereign debt of a country.
A direct proportional relationship exists between CDS spreads and the risk associated with sovereign debt of a country. Heinz and Sun (2014) support that CDS spreads in Central, Eastern and South-eastern Europe countries are influenced by global investor sentiment, economic fundamentals and CDS market liquidity conditions. Big fluctuations in CDS spreads over time are primarily caused by changes in the market perception of
macroeconomic fundamentals (Heinz & Sun, 2014). This idea is also presented in the Global Financial Stability Report 2013 of the IMF, which states that sovereign CDS spreads reflect economic fundamentals.
De Grauwe and Ji (2013) compare the market spreads of European countries to spreads based on fundamentals such as sovereign credit ratings, Debt to GDP ratio and terms of trade volatility. Movement drivers of the market spreads are fluctuations in the underlying fundamentals. On the other hand, De Grauwe and Ji (2013) found that for European
peripheral countries, a big part of the fluctuations in the spreads during 2010-11 could not solely be explained by increases in these economic fundamentals. Moreover, this surge was the result of time dependent negative market sentiments that were at the moment very strong (De Grauwe & Ji, 2013).
Chiarella et al. (2012) support this idea and state that although movements in European CDS markets before and during the European debt crisis can, by degrees, be explained by deteriorating fundamentals, speculative behaviour in the form of following trends has had a significant effect on the movements of CDS spreads.
2.3 OMT announcements
Alongside other interventions initiated to safeguard monetary policy transmission, the ECB introduced the Outright Monetary Transactions Program, which allowed the ECB to purchase government bonds. The idea behind the program is that the ECB would offer to purchase short-term bonds in the secondary market of European countries to bring down their market interest rates. Even though no European country had applied for help under OMT, according to the ECB its existence had greatly calmed down European financial markets.
Altavilla, Giannone and Lenza (2014) find that the OMT announcements have had a
significant effect on credit and on economic growth in Italy and Spain. OMT announcements had an extensive effect on financial markets and led to a decrease of about 200 basis points in the 2-year government bond rates in Italy and Spain (Altavilla, Giannone & Lenza, 2014).
Saka, Fuertes and Kalotychou (2014) use the OMT announcements to further
investigate the hypothesis stated by De Grauwe and Ji (2013) that for members of a monetary union the behaviour of investors intending to avoid default, in the end trigger the actual default they fear. A principal components analysis of Eurozone CDS spreads proposes that the OMT announcements decrease the overall level and the fluctuations of the sovereign CDS spreads for the whole Eurozone (Saka et al., 2014).
The review of empirical literature provides evidence for the fact that OMT announcements have affected the financial market. This paper contributes to the existing literature by investigating whether the OMT announcements by the ECB have had a significant effect on sovereign CDS spreads for European peripheral countries. Based on previously performed studies it is expected that the OMT announcements lower the overall level of the sovereign CDS spreads for the five European peripheral countries.
3 Methodology
To assess the effects of OMT announcements on sovereign CDS spreads in periphery countries, an empirical research is performed with CDS spreads on 10-year sovereign CDS contracts from the period of October 2008 until May 2015 collected from Datastream. The data include CDS spreads on a daily basis for the countries Italy, Spain, Portugal and Ireland. Though Greece belongs to the peripheral European countries, data for this country was not available and Greece is left out.
3.1 Variables
A linear regression, with sovereign CDS spreads as a dependent variable, is performed in two phases. For the specification of the macro economic fundamentals that influence the CDS spreads, the findings of Chiarella et al. (2012) are used. Chiarella et al. base their choice of fundamentals on existing literature. The chosen fundamentals are Credit Ratings, Debt-to-GDP ratio, Terms of trade and Local Stock Market Return. For Portugal the Terms Of Trade data was not available, so this variable is left out in the models for Portugal.
3.2 Models
The first phase of the regression compares sovereign CDS spreads in the period from October 2008 until July 2012 to sovereign CDS spreads in the period from August 2012 to May 2015 for every of the four countries. The regression model contains the abovementioned
explanatory variables and a dummy variable D1. This dummy takes value one for data after
July 2012 and value zero for data before this period. The sovereign CDS spreads for each country are regressed on the dummy and the chosen fundamentals. This first part of the regression will give insight in whether the period after the first OMT announcement on 26 July 2012, provides significant different sovereign CDS spreads than the period before the OMT announcements.
The second phase of the regression focuses on the question whether the OMT announcements have had a significant effect on CDS spreads during the period of the
announcements. For this regression three dummies are created, D2, D3 and D4. D2 takes value
one for the month August 2012, D3 takes value one for the month September 2012 and D4
takes value one for the month October 2012. The sovereign CDS spreads for each country are regressed on these three dummies and the chosen fundamentals. The regression investigates whether and from which moments on the OMT announcements significantly affect the sovereign CDS spreads for the four peripheral countries.
Saka, Fuertes and Kalotychou (2014) have a different method to investigate whether the OMT announcements of the ECB have a significant effect on the sovereign CDS spreads. Their method is called the principal component analysis, where they compare the
pre-announcement period to the post-pre-announcement period based on what percentage of the variation in the CDS spreads can be explained by the first few principal components. The goal of the first part of this research, comparing pre- and post-announcement CDS spreads, is equivalent to the objective of Saka, Fuertes and Kalotychou (2014). Therefore, their results can be used in a comparison to the results of this research. Furthermore, this research
contributes to the findings of Saka, Fuertes and Kalotychou (2014) by means of the second part of the regression. This part investigates whether the CDS spreads in several months are influenced by the announcements of the ECB, which Saka, Fuertes and Kalotychou (2014) did not insert in their research.
3.3 Tests
To test for stationarity, an Augmented-Dickey-Fuller test on the explanatory variables and the dependent variables for both models is performed. In this Augmented Dickey-Fuller test the null hypothesis of the variable having a unit root is rejected whenever the Augmented Dickey-Fuller test statistic comes below the test critical value on 5% level. When this null hypothesis cannot be rejected, the first difference of the variable is tested by the Augmented Dickey-Fuller test to check whether this new model does satisfy the condition of stationarity.
One of the assumptions for a correct model is the non-existence of correlation between the error terms. Whenever autocorrelation is present in a regression model, it is said that there is a dependency between the error terms. As a consequence the estimated parameters no longer have minimum variance, thus are not efficient estimates. To test for autocorrelation the autocorrelation function (ACF) and partial autocorrelation function (PACF) are investigated. The Ljung-Box statistic and the associated probability that are calculated belonging to the ACF and PACF will determine whether we reject the null hypothesis that there exists no autocorrelation up to lag k.
Another test applied to test whether there exists correlation between the error terms is the Breusch-Godfrey LM-test with two lags, where the null hypothesis of no serial correlation is rejected whenever the probability of the F-statistic is below 5%. The Breusch-Godfrey test is applied for all four countries and for both models.
To test whether past values of a variable help to explain present values of other variables a Granger-Causality test is carried out. The tests check for every variable, whether this variable Granger causes all the other variables and whether the other variables Granger cause this variable. The F-statistic and belonging probability give information on whether or not to reject the null hypothesis of the one variable not Granger causing the other variable.
In order to investigate whether the explanatory variables are mutually correlated, the correlation matrices for both models for all four countries are given in Appendix V. In case of correlation between the explanatory variables a step-by-step regression is performed to check for the validity of the model. In this step-by-step regression, an explanatory variable is added to the model step by step. In order to check whether the resulting models are valid, the Akaike
info criterion (AIC) and Schwarz criterion (SIC) are analysed. It applies for both the AIC and SIC that the lower these values are, the better the model.
3.4 Hypothesis
Based on the findings of Saka, Fuertes and Kalotychou (2014), who stated that the OMT announcements decreased the overall level and the fluctuations of the sovereign CDS spreads for the whole Eurozone, it is expected that in the first model the dummy variable has a negative coefficient which implies a decrease in the CDS spreads for the period after the OMT announcements. For the second model, it is expected that from the first month of the OMT announcement, there is a significant difference in the CDS spreads.
Table 1 below presents the hypothesis for equation 1. It is expected that D1, dummy 1 which takes the value 1 for data after July 2012, will have a coefficient with a negative sign. Therefore H0 is stated as a minus sign and HA as a positive sign. Table 2 shows the
hypothesis for equation 2. The H0 hypothesises include minus signs and the alternative plus signs as it is expected that from the first month of the announcements the CDS spreads will have significantly decreased.
Model 1 H0 HA
D1 -‐ +
Table 1 Hypothesis equation 1
Model 2 H0 HA
D2 -‐ +
D3 -‐ +
D4 -‐ +
Table 2 Hypothesis equation 2
4 Results and analysis
This section presents the results of the abovementioned tests and linear regression on both equations. It also contains an analysis of the outcomes of the tests and regressions.
4.1 Test Results
In order to check whether the two models are stationary an Augmented Dickey-Fuller test is performed to check whether each explanatory variable and the dependent variable, the CDS spreads, contain a unit root. Appendix II contains four tables that represent the results of the Dickey-Fuller test for Ireland, Italy, Portugal and Spain. The t-statistics of the tests for all countries and all belonging variables are not smaller than the 5% test critical value, so the null hypothesis of having a unit root cannot be rejected. This implies that for all countries and all belonging variables the first differences have to be implied to correct for non-stationarity.
Again, it is checked whether these first differences contain a unit root and the results of the Augmented Dickey-Fuller test on the first differences are displayed in the bottom part of the tables. The test statistics show that for all four countries and all belonging variables the null hypothesis of having a unit root can be rejected. The models, which contain the first
differences of the CDS spreads and the four explanatory variables CREDRAT, DEBTGDP, TOT and LSMR are therefore valid models concerning stationarity.
Another assumption for a correct model is the non-existence of autocorrelation, correlation between the error terms. To check for this the residuals per model per country are investigated using the autocorrelation function (AC) and the partial autocorrelation function (PAC). Appendix II displays the results of the residual diagnostics per model per country. In all the countries, for both models, the null hypothesis of no autocorrelation cannot be rejected based on the Ljung-Box statistic and the belonging probability value.
Additionally, a Breusch-Godfrey LM test with two lags is performed to check whether the model contains autocorrelation. The chosen lag length is based on the expectations for the autocorrelation. Appendix IV contains the results for this test for both equations for all four countries. The tests for all four countries significantly show that the null hypothesis of no serial correlation cannot be rejected. Based on the results from the Ljung-Box statistic and the Breusch-Godfrey LM test we can conclude that there does not exist correlation between the error terms.
4.2 Descriptive statistics
The CDS spreads on 10-year sovereign CDS contracts from the period of October 2008 until May 2015 collected from Datastream are presented below in Figure 2. The graph shows a rise in CDS spreads for all four countries from 2010 to about 2012. Portugal and Ireland seem to have experienced the most extreme rise in their sovereign CDS spreads, Portugal even passes the 1,000 basis points at its highest peak. From 2012 the CDS spreads of the four peripheral countries decrease and from about 2014 they seem to be at the level of the CDS spreads before the rise in 2010.
Figure 2 CDS spreads in basispoints 2008-‐2015
In order to investigate whether the steep fall of the CDS spreads in 2012 seen above in Figure 2 can be explained by the OMT announcements of the ECB two linear regressions are
performed for each country. Two models are given below, where equation 1 investigates whether the period after the OMT announcements in the period of July-September 2012 possesses significantly higher or lower CDS spreads and equation 2 investigates whether during the period of announcements there is a significant difference in the CDS spreads.
ΔCDSit= α + β1ΔCREDRATit+ β2ΔDEBTGDPit+β3ΔTOTit+ β4ΔLSMRit+β5 D1+εit (1)
ΔCDSit= α + β1ΔCREDRATit+ β2ΔDEBTGDPit+β3ΔTOTit+ β4ΔLSMRit+β5 D2+β6 D3+β7 D4+εit (2)
Equation 1 formulates the linear regression model for the first phase, where CDSit is the CDS
spread of country i in period t. CREDRAT represents the sovereign credit ratings, DEBTGDP represents the debt-‐to-‐GDP ratio, TOT is the terms of trade volatility and LSMR represents the national stock exchanges and D1 is the dummy which takes value 1
for the period after July 2012 and zero otherwise. Equation 2 represents the linear
regression model for the second phase of the regression, which includes the same explanatory variables as equation 1 and furthermore includes dummies D2, D3 and D4 where D2 stands for
The explanatory variables for both models are presented below in figure 3 to 6. Figure 3 displays the development of the debt-to-GDP ratio over the years 2008 to 2015. Based on a comparison between the CDS spreads in figure 2 and the debt-to-GDP ratio the variables do not seem to develop in alike. Figure 4 displays the local stock market return and a similar, though smaller, fall in the variable in the period between 2011 and 2012 is observed just like in the CDS spreads.
Figure 5 represents the credit ratings for all four peripheral countries. Again a fall in the variable is seen in the period between 2011 and 2012 just like for the CDS spreads. Lastly, figure 6 shows the terms of trade and it can be seen that this variable fluctuates a lot over time with no obvious trends.
Figure 3 Debt to GDP Figure 4 Local stock market return
4.2.1 Results Ireland
Table 3 below gives the results of the first model for Ireland. First of all, the F-statistic is evaluated in order to check whether this model makes sense. Whenever the Prob(F-statistic) is lower than the 5% significance level we can reject the null hypothesis that none of the
explanatory variables (apart from the constant term) have any effect on the dependent
variable. The model has a Prob(F-statistic) value of 0.092757 so technically speaking at a 5% level, the null hypothesis cannot be rejected and the explanatory variables seem not to be having any effect on the dependent variable. Though, if we would be taking a significance level of 10%, we can reject the null hypothesis and we assume this model makes sense and that the explanatory variables jointly can influence the dependent variable.
Looking at the fit of this model, the R-squared value has to be interpreted. Model one contains a value of R-squared of 0.128786, indicating a relatively low fit of this model. This result corresponds with the interpretation of the F-statistic of this model, both indicating that this is not a perfect model with a good fit.
When further interpreting the results given above it seems that none of the explanatory variables, CREDRAT, DEBTGDP, LSMR or TOT have a significant effect on the CDS spreads of Ireland since all p-values are higher than 5%. When looking at the dummy created for the period after the OMT announcement, we see that the p-value is 0.7288 indicating that the effect in non significant.
The next phase of the regression is checking whether the CDS spreads of Ireland have changed significantly during the period of the OMT announcements of the ECB. Table 4 below shows the results of the linear regression for the second model, where dummies are created for the period during the OMT-announcement: Augustus, September and October.
The F-statistic of this linear regression has a probability value of 0.003961, meaning we can reject the null hypothesis that none of the explanatory variables have a significant effect on the CDS spreads. The only explanatory variable that seems to be having a significant effect on the CDS spread is the debt-to-GDP ratio. Its corresponding coefficient is 3.612902, meaning that an increase of 1 unit debt-to-GDP ratio, increases the CDS spread by 3.612902 basis points. The R-squared value is 0.266311, which is higher than the R-squared of the first model, indicating that this model has a better fit.
The coefficient of the dummy for the month October 2012 is -126.3216 and it is significant at a 5% level with a p-value of 0.0054. This indicates that in October 2012, the first month after the OMT-announcements of the ECB, the CDS spreads in Ireland have significantly fallen. This is the only one of the three dummies that is significant.
For both models, it could be the case that the explanatory variables are correlated, which would influence the results of the linear regressions. The correlation matrix for the explanatory variables of Ireland is given in Appendix V, Table 13. All explanatory variables experience a mutually significant correlation. The correlation between CREDRAT and DEBTGDP is -0.934220 with a p-value rounded to 0.0000. Since a correlation value is defined as a number between -1 and 1, the correlation between the credit ratings of Ireland
and its debt-to-GDP ratio is relatively high. Other correlation values of explanatory values that step out as a high numbers are the ones between TOT and CREDRAT, which reaches the value of 0.920227 and between TOT and DEBTGDP of -0.868873. The correlations between LSMR and CREDRAT, LSMR and DEBTGDP, TOT and LSMR contain somewhat lower values but are still significantly correlated. We can conclude that all explanatory values are mutually correlated.
Since correlation is present between several variables, a step-by-step regression is performed. Appendix VI table 17 and 18 present the results of the step-by-step regression of Ireland for both equations, where a new variable is added to the model at every step. The Akaike info criterion (AIC) and Schwarz criterion (SIC) of the models are compared in order to check whether adding a variable preserves the validity of the model. In the case of Ireland the AIC and SIC values remain at a value of about 10 for all models in both equations and it can be concluded that the fit of the model is contained while adding explanatory variables. Another aspect of the model we have to check for is Granger-Causality. The pairwise
Granger-cause test tests whether past values of a variable influence present values of another variable. Appendix VII, Table 25 represents the pairwise Granger-Cause test outcomes for Ireland. For every pair of variables the null hypothesis of variable 1 does not granger cause variable 2 and variable 2 does not granger cause variable 1 is tested. The results of this pairwise Granger-cause test indicate that the credit ratings of Ireland Granger cause the CDS spreads of Ireland with a probability value belonging to the F-statistic of 0.0007, and the CDS spreads Granger Cause the credit ratings with a p-value of 0.0361. Debt-to-GDP ratio Granger Causes the CDS spreads of Ireland with a p-value of 0.0113 but the other way around the Granger-Causality does not exists even though this p-value is nearly significant, namely 0.0535. Terms of Trade Granger Causes the CDS spreads, but the CDS spreads do not Granger Cause the Terms of Trade. The credit ratings granger cause the debt-to-GDP ratio, but the debt-to-GDP ratio does not granger cause the credit ratings. The last Granger causality relation is found between credit ratings and terms of trade, where the credit ratings granger cause the terms of trade. An overview of the significant pairwise Granger Cause relations is given below.
significantly Granger Causes CREDRAT ! CDS CDS ! CREDRAT DEBTGDP ! CDS TOT ! CDS CREDRAT ! DEBTGDP CREDRAT ! TOT 4.2.2 Results Italy
Table 5 below shows the results of the linear regression for the first model of Italy. The F-statistic has a corresponding probability value of 0.000001 meaning that the explanatory variables have an effect on the CDS spreads. R-squared for this model is 0.404084, meaning that this model has a relatively good fit.
The only explanatory variable that seems to have a significant effect on the CDS spreads in Italy in this first model is the local stock market return (LSMR). With a p-value rounded to 0.0000 its coefficient is -0.018105, meaning that the local stock market return a significant, but small, negative effect on the sovereign CDS spreads in Italy. The dummy for the period after the OMT-announcements has a corresponding p-value of 0.3238 indicating that the effect is not significant.
The second model for Italy is represented below in table 6, the corresponding Prob(F-statistic) indicates that the explanatory values have effect on the CDS spreads and the R-squared of 0.494336 provides us with the information that the model has a relatively good fit.
When evaluating the coefficients of the explanatory variables, it seems, just like the first model, that only the local stock market return has a significant effect of -0.018649 on the CDS spreads in Italy. For the first month during the OMT-announcements, there is no significant difference detected in the CDS spreads. The second month of the announcement period, September, however does imply a significant difference in the CDS spread. Its
corresponding coefficient is -89.29429, meaning that during this month the CDS spreads have significantly fallen. In the first month after the announcements there is no significant
difference found in the CDS spreads.
The results of both models can be influenced when there exists collinearity between the explanatory variables. Appendix V, table 14 presents the correlation matrix for Italy. The most extreme number of collinearity is found between debt-to-GDP ratio and credit ratings, which has a correlation value of -0.899442. Furthermore, a significant correlation is found between local stock market return and credit ratings and between local stock market return and terms of trade. All other variables do not seem to be significantly correlated.
The present correlation induces a step-by-step regression. Appendix VI table 19 and 20 present the results of the step-by-step regression of Italy for both equations. For Italy the
AIC and SIC values decrease when more explanatory variables are added, indicating that the model becomes more valid when adding variables.
The pairwise Granger-Cause tests for Italy indicate that credit ratings Granger Cause debt-to-GPD ratio, local stock market return Granger Causes credit ratings, credit ratings Granger Cause terms of trade, CDS Granger Causes credit ratings and terms of trader Granger Causes local stock market return. An overview of the Granger Causality relations between the variables is given below and the complete results for the Granger-Cause test for Italy are given in Appendix VII table 26.
significantly Granger Causes
CREDRAT ! DEBTGDP LSMR ! CREDRAT CREDRAT ! TOT CDS ! CREDRAT TOT ! LSMR 4.2.3 Results Portugal
Table 7 below presents the results of the linear regression of the first model for Portugal. The corresponding Prob(F-statistic) indicates that this model makes sense, as the explanatory variables have influence on the CDS spreads. On the other hand, the R-squared value is relatively low, indicating that this model does not have a perfect fit.
The only significant variable in this model is the local stock market return with a coefficient of -0.105586 and a p-value of 0.0000. The local stock market return seems to have a
significant, negative but small effect on the CDS spread in Portugal.
The dummy variable has a p-value of 0.4439 indicating that there is no significant difference found in the CDS spreads before and after the OMT announcements.
Table 8 below represents the results of the linear regression for model 2 for Portugal. The probability value corresponding to the F-statistic shows that the explanatory variables jointly influence the CDS spread in Portugal. Just like in the first model, the R-squared of this model is relatively low indicating that this model does not have a perfect fit. Local stock market return is the only variable that significantly influences the CDS spread in Portugal. When evaluating the coefficients of the dummies created for the months during the OMT-announcements, it seems that in none of the months the CDS spreads are significantly different.
Collinearity between the explanatory variables could influence the result of the models. The correlation matrix of Portugal is displayed in Appendix V, Table 15. It seems that all 3 variables are mutually correlated, with a correlation of -0.955847 for debt-to-GDP and credit ratings, 0.669018 for local stock market return and credit ratings and -0.570698 for local stock market return and debt-to-GDP ratio.
The results for the step-by-step regression done as a result of correlation are presented in Appendix VI table 21 and 22. The AIC and SIC values are fluctuating around 11 indicating that the validity of the model remains about the same when adding more variables.
Evaluating the results from the Pairwise Granger Causality tests, it can be concluded that the CDS spreads Granger Cause the credit ratings, the local stock market return Granger Causes the CDS spreads, the CDS spreads Granger Cause the local stock market return and the credit ratings Granger Cause the local stock market return. An overview is given below and complete results are given in Appendix VII table 27.
significantly Granger Causes
CDS ! CREDRAT LSMR ! CDS
CDS ! LSMR
CREDRAT ! LSMR
4.2.4 Results Spain
Table 9 below presents the results for the first linear regression of Spain. The F-statistic and its corresponding probability value indicate that the variables jointly influence the CDS spread. The R-squared value of 0.590955 indicates that this model has a quite good fit. Credit ratings, local stock market return and terms of trade all three have a significant effect on the CDS spreads in Spain. Credit ratings seem to have a positive effect on the sovereign CDS spreads, while the coefficients of the local stock market return and the terms of trade indicate a negative effect on the CDS spreads. The p-value of the variable debt-to-GDP ratio is 0.0931, which indicates that the effect of debt-to-GDP on CDS spreads is not significant.
Table 10 below displays the results for the linear regression of the second model for Spain. Just like in the first regression model for Spain the F-statistic indicates a good model, with explanatory variables that have an effect on the CDS spreads and a R-squared stating that this model has a quite good fit. For this second model there are two variables with a significant effect on the CDS spread of Spain, namely local stock market return and terms of trade. Both effects are negative. The only month during the OMT-announcements that seems to have significant different CDS spreads is September 2012, where the CDS spreads are lower for this second month of the spreads.
To check for collinearity between the variables, the correlation matrix of Spain in Appendix V, Table 16 is evaluated. The most extreme number of correlation is found between credit ratings and debt-to-GDP ratio, namely -0.930535. All other variables, except local stock market return with debt-to-GDP ratio, seem to possess significant correlation.
The results for the step-by-step regression done as a result of correlation are presented in Appendix VI table 23 and 24. The AIC and SIC values are decreasing when adding more variables, indicating that the models validity increases when adding variables.
Appendix VII table 28 presents the pairwise Granger Cause tests for the variables for Spain. Credit ratings Granger Cause the CDS spreads, CDS spreads Granger Cause the credit ratings, CDS spreads Granger Cause the terms of trade, credit ratings Granger Cause the debt-to-GDP ratio, the credit ratings Granger Cause the local stock market return, the terms of
trade Granger Cause the credit ratings and the terms of trade Granger Cause the local stock market return. An overview is given below.
significantly Granger Causes CREDRAT ! CDS CDS ! CREDRAT CDS ! TOT CREDRAT ! DEBTGDP CREDRAT ! LSMR TOT ! CREDRAT TOT ! LSMR 4.3 Analysis
All four peripheral countries did not show a significant difference in the CDS spreads after the OMT-announcements in the first regression model. Comparing this to the hypothesis stated before, the null hypothesis of a significant decrease in the CDS spreads due to the OMT announcements can be rejected.
For Ireland these findings could possibly be ascribed to the fact that all explanatory variables are mutually correlated, with 3 pairs of these variables containing a relatively extreme correlation value. The other three countries also contained at least 2 pairwise high correlation values between the variables. Additionally, for each country at least 4 Granger Cause relations were found, indicating that for at least 4 variables their past values can influence future values of other variables. These Granger Causality relations influence the results of the linear regression.
For the second regression model, Italy, Spain and Ireland possess a significant difference in the CDS spread for one or two of the months during the OMT-announcements. In the case of Italy, September 2012 had significantly lower CDS spreads, while August and October showed no such significant difference. For Spain, it was also September 2012 that had significantly lower CDS spreads and August and October did not contain a significant difference. In Ireland, it was the first month after the OMT-announcements that showed significantly lower CDS spreads, while August and September did not. Table 11 below
represents the significant results for the dummies, where D3 stands for the month October and D2 for the month September. Comparing this to the hypothesises stated before, for the month September in Spain and Italy and for the month October in Ireland the null hypothesis cannot
be rejected while for the other months in all countries the null hypothesis of a significant decrease in the CDS spreads due to the OMT announcements can be rejected.
Table 11 Significant results equation 2
5 Conclusion
Chiarella et al. (2012) state that fundamentals such as debt-to-GDP ratio and sovereign credit ratings do not solely explain the movements in European CDS spreads. Trend following speculation also plays a significant role in explaining the fluctuations of these CDS spreads. The announcements by the ECB of policy changes in the period July to September 2012 create a good groundwork for the investigation on whether country specific news influences CDS spreads.
Even though Saka, Fuertes and Kalotychou (2014) found a significant decrease in the overall level and the fluctuations of the sovereign CDS spreads for the whole Eurozone, the results of this paper are not sufficient to confirm this. For none of the four peripheral
countries, Ireland, Italy, Portugal and Spain, a significant decrease can be found in the level of the sovereign CDS spreads. For Italy, Spain and Ireland a significant difference in the CDS spread for one or two of the months during the OMT-announcements was found. It can be concluded that overall the OMT-announcements of the ECB do not have influence on the level of the CDS spreads for the whole period after the announcements. Nevertheless, in Italy, Portugal and Spain the level of the CDS spreads did decrease for some period during the OMT-announcement. Significant Results Ireland D3 -‐ 126.3216 Italy D2 -‐ 89.29429 Spain D2 -‐ 55.25556
References
Altavilla, C., Giannone, D., & Lenza, M. (2014). The financial and macroeconomic effects of OMT announcements (ECB Working Paper, 1707). Retrieved from https://www.ecb.europa.eu/pub/pdf/scpwps/ecbwp1707.pdf
Caceres, C., Guzzo, V., & Segoviano, M. (2010). Sovereign Spreads: Global Risk Aversion, Contagion or Fundamentals? (IMF Working Paper, 10/120). Monetary and Capital Markets Department. Retrieved from
https://www.imf.org/external/pubs/ft/wp/2010/wp10120.pdf
Chiarella, C., Ellen, ter, S., He, T., & Wu, E. (2012). Fear or fundamentals? Speculative behaviour in the European CDS market. Retrieved from http://editorialexpress.com
De Grauwe, P., & Ji, Y. (2013). Self-fulfilling crises in the Eurozone: An empirical test. Journal of International Money and Finance, 34, 15-36.
http://dx.doi.org/10.1016/j.jimonfin.2012.11.003
Heinz, F.F., & Sun, Y. (2014). Sovereign spreads in Europe – The Role of Global Risk Aversion, Economic Fundamentals, Liquidity, and Spillovers. (IMF Working Paper, 14/17), European Department. Retrieved from
http://www.imf.org/external/pubs/ft/wp/2014/wp1417.pdf
Kolstad, M. (2013). An analysis of Eurozone sovereign credit default swaps. Unpublished master’s thesis, Copenhagen Business School, Copenhagen, Denmark.
Saka, O., Fuertes, A., & Kalotychou, E. (2014). ECB Policy and Eurozone Fragility: Was De Grauwe Right? Retrieved from City University London, Cass Business School http://www.cassknowledge.com/sites/default/files/article-
Appendices Appendix I
Figure 1. CDS Spreads for four European ‘periphery’countries: Spain (ES), Ireland (IE), Italy (IT) and Portugal (PT). Greece is not displayed for scaling reasons. Adapted from “Fear or fundamentals? Speculative behaviour in the European CDS market,” by C. Chiarella, S. ter Ellen, T. He and E. Wu, 2012.
Appendix II
Table 1. Augmented Dickey-Fuller test for unit root for Ireland
Ireland CDS Credrat TOT LSMR Debt/GDP
Level t-‐Statistic -‐1.124 -‐2.0847 -‐1.694 1.1096 -‐2.5707
Prob. 0.7026 0.2514 0.4297 0.9973 0.1036
Test critical value 5% -‐2.899 -‐2.899 -‐2.901 -‐2.899 -‐2.9012 First difference t-‐Statistic -‐8.531 -‐8.884 -‐11.63 -‐8.209 0.9164
Prob. 0.0000 0.0000 0.0001 0.0000 0.9951
Test critical value 5% -‐2.899 -‐2.899 -‐2.9017 -‐2.899 -‐2.9092 Table 2. Augmented Dickey-Fuller test for unit root for Italy
Italy CDS Credrat TOT LSMR Debt/GDP
Level t-‐Statistic -‐1.898 -‐0.7410 -‐1.037 -‐1.809 -‐1.6804
Prob. 0.3315 0.8293 0.7360 0.3736 0.4369
Test critical value 5% -‐2.899 -‐2.9006 -‐2.899 -‐2.899 -‐2.9012 First difference t-‐Statistic -‐8.831 -‐3.1921 -‐8.827 -‐7.949 -‐9.4527
Prob. 0.0000 0.0243 0.0000 0.0000 0.0000
Test critical value 5% -‐2.899 -‐2.900 -‐2.900 -‐2.899 -‐2.90177
Table 3. Augmented Dickey-Fuller test for unit root for Portugal
Portugal CDS Credrat LSMR Debt/GDP
Level t-‐Statistic -‐1.482 -‐2.0624 -‐1.471 -‐1.6772
Prob. 0.5375 0.2603 0.5428 0.4385
Test critical value 5% -‐2.899 -‐2.9023 -‐2.899 -‐2.9012 First difference t-‐Statistic -‐10.75 -‐1.2684 -‐7.180 -‐9.7509
Prob. 0.0001 0.6400 0.0000 0.000
Test critical value 5% -‐2.899 -‐2.9023 -‐2.899 -‐2.90177
Table 4. Augmented Dickey-Fuller test for unit root for Spain
Spain CDS Credrat TOT LSMR Debt/GDP
Level t-‐Statistic -‐1.827 -‐1.2839 -‐2.120 -‐1.559 -‐1.1742
Prob. 0.3647 0.6331 0.2372 0.4985 0.6813
Test critical value 5% -‐2.899 -‐2.900 -‐2.900 -‐2.899 -‐2.9029 First difference t-‐Statistic -‐9.675 -‐2.676 -‐13.77 -‐8.129 -‐3.1803
Prob. 0.0000 0.0829 0.0001 0.0000 0.0253
Appendix III
Figure 2. Correlogram Spain equation 1 Figure 3. Correlogram Spain equation 2
Figure 6. Correlogram Italy equation 1 Figure 7. Correlogram Italy equation 2
Appendix IV
Table 5. Breusch-Godfrey serial correlation LM test Ireland – equation 1 Breusch-Godfrey Serial Correlation LM Test:
F-statistic 0.156392 Prob. F(2,65) 0.8555
Obs*R-squared 0.349597 Prob. Chi-Square(2) 0.8396
Table 6. Breusch-Godfrey serial correlation LM test Ireland – equation 2 Breusch-Godfrey Serial Correlation LM Test:
F-statistic 1.096961 Prob. F(2,63) 0.3402
Obs*R-squared 2.456615 Prob. Chi-Square(2) 0.2928
Table 7. Breusch-Godfrey serial correlation LM test Italy – equation 1 Breusch-Godfrey Serial Correlation LM Test:
F-statistic 0.893019 Prob. F(2,66) 0.4143
Obs*R-squared 1.949764 Prob. Chi-Square(2) 0.3772
Table 8. Breusch-Godfrey serial correlation LM test Italy – equation 2 Breusch-Godfrey Serial Correlation LM Test:
F-statistic 1.535852 Prob. F(2,64) 0.2231
Obs*R-squared 3.389001 Prob. Chi-Square(2) 0.1837
Table 9. Breusch-Godfrey serial correlation LM test Portugal – equation 1 Breusch-Godfrey Serial Correlation LM Test:
F-statistic 7.195505 Prob. F(2,68) 0.0015
Obs*R-squared 13.10004 Prob. Chi-Square(2) 0.0014
Table 10. Breusch-Godfrey serial correlation LM test Portugal – equation 2 Breusch-Godfrey Serial Correlation LM Test:
F-statistic 1.829531 Prob. F(2,65) 0.1687
Table 11. Breusch-Godfrey serial correlation LM test Spain – equation 1
Breusch-Godfrey Serial Correlation LM Test:
F-statistic 2.103012 Prob. F(2,66) 0.1302
Obs*R-squared 4.433320 Prob. Chi-Square(2) 0.1090
Table 12. Breusch-Godfrey serial correlation LM test Spain – equation 2 Breusch-Godfrey Serial Correlation LM Test:
F-statistic 1.726983 Prob. F(2,64) 0.1860
Obs*R-squared 3.789154 Prob. Chi-Square(2) 0.1504
Appendix V
Table 13. Correlation Matrix Ireland
Table 15. Correlation Matrix Portugal
Appendix VI
Table 17. Step-‐by-‐step regression Ireland Equation 1: Coefficient (p-‐value)
Table 18. Step-‐by-‐step regression Ireland Equation 2: Coefficient (p-‐value)
Model 1 Model 2 Model 3 Model 4
c 2.0937 (0.695) -‐2.6606 (0.630) 0.214386(0.9700) 0.0195 (0.997) Δcredrat -‐5.597(0.6128) -‐7.2091 (0.502) -‐6.447 (0.5430) -‐9.7071 (0.373) Δdebtgdp 4.1278 (0.013) 3.31066 (0.0512) 3.6129 (0.035) Δlsmr -‐0.05246 (0.0904) -‐0.0544 (0.082) Δtot 5.1117 (0.183) D2 -‐53.483(0.247) -‐73.909 (0.106) -‐72.243 (0.1097) -‐75.953(0.094) D3 -‐63.728 (0.168) -‐58.974 (0.188) -‐55.6857 (0.2083) -‐50.654 (0.254) D4 -‐128.30 (0.006) -‐123.55 (0.007) -‐127.99 (0.0048) -‐126.32(0.005) AIC 10.53756 10.48713 10.4710 10.486 SIC 10.68863 10.67395 10.6890 10.737
Table 19. Step-‐by-‐step regression Italy Equation 1: Coefficient (p-‐value)
Model 1 Model 2 Model 3 Model 4
c 7.948 (0.1609) 6.1463 (0.316) 3.4948 (0.479) 3.5125 (0.481) Δcredrat 25.996 (0.112) 24.350 (0.149) 15.184 (0.265) 15.301 (0.269) Δdebtgdp 2.5598 (0.484) 0.9856 (0.738) 0.9627 (0.748) Δlsmr -‐0.0181 (0.000) -‐0.0181 (0.000) Δtot 0.1934 (0.950) D1 -‐14.217 (0.094) -‐14.164 (0.116) -‐7.2755(0.319) -‐7.3438 (0.323) AIC 10.0646 10.132 9.710 9.7373 SIC 10.1552 10.2566 9.866 9.9242
Model 1 Model 2 Model 3 Model 4
c 5.41715(0.4670) -‐3.177(0.6968) -‐1.2914(0.874) -‐1.4621(0.858) Δcredrat -‐2.69 (0.8217) -‐6.71(0.5738) -‐6.4629 (0.584) -‐9.8185 (0.419) Δdebtgdp 3.57 (0.0503) 2.9006 (0.118) 3.177 (0.0909) Δlsmr -‐0.0488(0.140) -‐0.0513 (0.125) Δtot 5.2773 (0.198) D1 -‐14.77 (0.1946) -‐5.926 (0.6275) -‐4.2527 (0.726) -‐4.2577(0.728) AIC 10.60271 10.59173 10.58710 10.603 SIC 10.69336 10.71627 10.74278 10.7913
Table 20. Step-‐by-‐step regression Italy Equation 2: Coefficient (p-‐value)
Model 1 Model 2 Model 3 Model 4
c 3.8040(0.3715) 2.500225 (0.5988) 2.4004 (0.5125) 2.40748 (0.5156) Δcredrat 24.8012(0.1185) 23.6038 (0.150) 15.2913 (0.2278) 15.2552 (0.2354) Δdebtgdp 2.3027 (0.521) 0.3453 (0.9010) 0.3561 (0.899) Δlsmr -‐0.01864 (0.000) -‐0.0186 (0.0000) Δtot -‐0.0756 (0.9799) D2 -‐3.9342 (0.9127) -‐5.8542 (0.8745) 19.5333 (0.4983) 19.3158 (0.5241) D3 -‐90.558(0.0134) -‐89.2551 (0.0179) -‐89.241 (0.0025) -‐89.294 (0.0027) D4 -‐57.373 (0.1128) -‐56.07012 (0.132) -‐47.695 (0.0974) -‐47.661 (0.1006) AIC 10.03858 10.10907 9.6002 9.6272 SIC 10.18955 10.2958 9.81816 9.8763
Table 21. Step-‐by-‐step regression Portugal Equation 1: Coefficient (p-‐value)
Model 1 Model 2 Model 3
c 6.6027 (0.5981) 2.4564 (0.8636) 1.2791 (0.9199) Δcredrat -‐27.1012 (0.2370) -‐29.6937 (0.2150) -‐26.2617 (0.2180) Δdebtgdp 3.2798 (0.4793) 1.4511 (0.7259) Δlsmr -‐0.1055 (0.0000) D1 -‐20.1655 (0.2816) -‐17.2015 (0.3986) -‐13.8843 (0.4439) AIC 11.5976 11.6714 11.44963 SIC 11.6882 11.7959 11.60531
Table 22. Step-‐by-‐step regression Portugal Equation 2: Coefficient (p-‐value)
Model 1 Model 2 Model 3
c -‐0.19203 (0.9838) -‐3.5254 (0.7416) -‐4.1178 (0.6673) Δcredrat -‐32.1029 (0.1507) -‐34.0699 (0.1406) -‐30.2345 (0.1445) Δdebtgdp 4.46157 (0.3304) 2.3987 (0.5608) Δlsmr -‐0.10300 (0.0001) D2 -‐87.4278 (0.2741) -‐96.1407 (0.2448) -‐57.9682 (0.4357) D3 -‐84.4109 (0.2909) -‐81.0775 (0.3239) -‐59.4833 (0.4195) D4 -‐3.6458 (0.9635) -‐0.31248 (0.9970) 16.0596 (0.8269) AIC 11.63324 11.70188 11.4928 SIC 11.7843 11.88869 11.71078
Table 23. Step-‐by-‐step regression Spain Equation 1: Coefficient (p-‐value)
Model 1 Model 2 Model 3 Model 4
c 7.9851 (0.1396) 1.8946 (0.7521) 3.8414 (0.3852) 3.3656 (0.4211) Δcredrat 11.3262 (0.2154) 8.3374 (0.3634) 14.4173 (0.0359) 14.590 (0.0251) Δdebtgdp 6.4824 (0.0246 2.9729 (0.1656) 3.4244 (0.0931) Δlsmr -‐0.0424 (0.0000) -‐0.0420 (0.0034) Δtot -‐3.3896 (0.0034) D1 -‐15.8251 (0.0522) -‐15.4396 (0.0657) -‐9.7576 (0.1148) -‐8.6996 (0.1373) AIC 9.940623 9.943611 9.33883 9.2388 SIC 10.03127 10.06816 9.494514 9.4256
Table 24. Step-‐by-‐step regression Spain Equation 2: Coefficient (p-‐value)
Model 1 Model 2 Model 3 Model 4
c 2.3694 (0.5569) -‐3.10955 (0.5152) 0.9385 (0.7836) 0.8993 (0.7829) Δcredrat 6.3172 (0.4807) 3.7977 (0.6733) 11.4666 (0.0788) 11.683 (0.0612) Δdebtgdp 6.3045 (0.0289) 2.4871 (0.2302) 3.0216 (0.1301) Δlsmr -‐0.0435 (0.0000) -‐0.0427 (0.0000) Δtot -‐3.0673 (0.0080) D2 0.22039 (0.9949) -‐3.7574 (0.9129) 27.6600 (0.2641) 13.455 (0.5773) D3 -‐77.6294 (0.0264) -‐72.1504 (0.0389) -‐63.6472 (0.0109) -‐55.255 (0.0213) D4 -‐45.1995 (0.1958) -‐41.4011 (0.2354) -‐34.4768 (0.1645) -‐36.550 (0.1236) AIC 9.9535 9.9644 9.28832 9.207951 SIC 10.1046 10.15126 9.50628 9.457039
Appendix VII
Table 25. Pairwise Granger causality tests Ireland
Table 27. Pairwise Granger causality tests Portugal
Table 28. Pairwise Granger causality tests Spain