Faculty of Economics and Business MSc Finance
What determines sovereign spreads in the Eurozone?
A study on debt levels, credit rating and the US Yield.
Sebastiaan C. Pille S3117847 Master Thesis
2018-2019
Supervised by Dr. A.G. Schertler Abstract
2
1. Introduction
The size and importance of sovereign debt markets is increasing rapidly. One of the main reasons for this movement is the current low-interest rate climate driven by, for example, the Quantitative Easing program (QE) of the European Central Bank (ECB). However, extremely high interest rates are also visible. For example, Argentina where the average interest rate on its 7 day notes, the Leliq rate, is 59.38%1.
To analyze the determinants of the sovereign spreads, this paper uses a panel analysis of 14 countries over the period 2008-2017. To estimate the sovereign spread I use sovereign Credit Default Swap (CDS) spread data. Hull (2014) finds that CDS spreads are a good estimator of sovereign spread because the quoted CDS price is the price at which a broker is committed to trade a minimum principal and therefore reflects a real price. Besides that, as CDS quotes are already spreads, it does not need an adjustment. This paper uses three possible determinants. First debt levels. Collin-Dufresne et al. (2001) find that high amounts of debt cause higher credit spreads. This is caused by the fact that extra debt makes it more risky to invest in the bond and investors want a compensation for bearing that extra risk. High debt levels do not have to be a problem for a country as long as its GDP grows at a similar rate. Therefore I use the debt to GDP ratio. Bellas et al. (2010) find that debt to GDP ratios and sovereign spreads are positively related in emerging markets. Second, credit ratings. Credit ratings are often viewed as a summary of the quality of the debt issued by a sovereign. Affonso et al. (2015) find that credit rating changes have a significant effect on sovereign spreads. Aizenman et al. (2013) find similar results and also discovered that riskier bonds react stronger to a change in credit rating than safer bonds do. Last, the US yield. The United States is the largest economy in the world and its bonds are often used by investors as a safe-haven in times of market distress Beber et al. (2009). Longstaff et al. (2011) show that changes in the US market and economy influences sovereign spreads around the world. Moreover, the US yield is also often used as the risk free rate. Since the sovereign spread is the difference between a sovereigns yield and the risk free rate a change in the sovereign spread is expected when the US yield changes.
I will perform two robustness tests. During the financial crisis, Greece, Ireland, Italy, Portugal and Spain suffered from the highest levels of distress (Arnold (2012)). These countries are also called the GIIPS countries. Jones et al. (1998) and Longstaff et al. (2011) find that investors are more sensitive to new information regarding risky investments. However, Ang and Bekaert (2002) and Longstaff et al (2011) find high commonalties in spreads during times of crisis. The sample of this study covers both the financial crisis and the European sovereign debt crisis so these high commonalties might also exist in this sample. To test this I constructed a Principal
3 Component Analysis. After this I will test if GIIPS countries react differently to a change in one of the above mentioned factors. To test this I make a subsample with GIIPS and non-GIIPS countries. This will be the first robustness test. Secondly, I also test whether the speech by the president of the ECB, Mario Draghi, on July 26th 2012 which addressed “the ECB is ready to do whatever it takes to preserve the euro“ had a significant impact on the sovereign spreads. A week after this statement the ECB announced that it would, if needed, buy the bonds of countries in distress. This program was called Outright Monetary Transactions but was eventually never used. Because of this program the debt quality of Euro-zone countries increased because the risk of default became lower. As shown by Pan and Singleton (2008) lower risk means better credit quality and therefore smaller spreads.
To analyze the effect of government debt on the sovereign spreads, this paper uses a panel analysis of 14 countries over the period 2008-2017. This will also contribute to existing literature because the dataset is expanded with 7 extra years compared to the research of for example Longstaff et al. (2011). In their research they only covered the pre-crisis effects.
First, I find a significant positive relationship between government debt to GDP ratios and sovereigns spreads. The positive effect of government debt levels on sovereign spreads is in line with the existing literature. Secondly, I test the effect of credit ratings on the sovereign spread. I find that a credit rating downgrade causes an increase in the sovereign spread. This confirms the findings of Affonso et al. (2015). Last, this study shows that Eurozone sovereign spreads are negatively related with the US yield. This relation is intuitive since the sovereign spread is the difference between a sovereigns interest rate and the risk free rate. The US interest rate is often used as the risk free rate. I also test these factors together and find that they are jointly significant. The results however are puzzling because the sign of the debt to GDP ratio changes to negative. This is caused by high multicollinearity between debt to GDP ratios and credit ratings and therefore these results are not valid.
My robustness test shows that GIIPS countries do not react differently to changes in these factors. A reason for this could be the high commonalities between sovereign spreads during times of crisis. However, I do find that in the period after the Draghi announcement sovereign spreads reacted differently to a change in debt to GDP ratio, credit rating and US yield. Sovereign spreads became less sensitive to changes of the debt to GDP ratio and credit rating changes but more sensitive to a change in the US yield.
4
2. Literature review
This section provides an overview of the existing literature. The literature review covers both empirical and theoretical based articles. The literature review is divided into local factors and in global factors that could both determine sovereign spread.
2.1 Determinants of the sovereign spread
Literature on the determinants of sovereign spread often divide the possible causes into local and global effects (Pan and Singleton, 2008; Longstaff et al., 2011). This section discusses the literature on both the global and the local effects. The former effect consists of credit quality, which is measured by debt level and credit ratings. The latter effects are the US markets, the correlation between different sovereign bond prices and volatility of the financial markets.
2.1.1 Local factors
The first theories about the effect of debt levels on (sovereign) spreads are done on a company level. Most famous the study of Modigliani and Miller (1958). In their study they discuss the effect of an increase in the debt to equity ratio of a company on the weighted average cost of capital. They find that an increase in a company’s debt level is perceived as riskier and as a result, equity holders demand a higher return, while the weighted cost of capital stays the same.
Collin-Dufresne et al. (2001) show that companies with a high level of debt also have higher credit spreads on their debt. Bellas et al. (2010) wrote a paper on the determinants of emerging market bond spreads and find highly significant results that debt to GDP ratio’s and sovereign spreads are positively correlated.
Yue (2010) explores the relation between sovereign default risk and debt renegotiation during a sovereign debt crisis. This study focusses most on the Argentinian debt crisis where he finds significant empirical evidence that an increase in the debt level of Argentina leads to in an increase in the interest rates paid on Argentinian government bonds.
In a paper for the European Central Bank, Alexopoulou et al. (2009) discuss the determinants of government bond spreads in new European Union countries. The authors find that besides inflation, exchange rates compared to other EU countries and openness to trade, also the debt to GDP ratio has a significant impact on the sovereign spread.
5 The second local effect under discussion is the credit rating of a country, which is one of the key drivers behind the sovereign spread of a country. When the quality of a country’s debt is low, the risk related to the debt is higher. Namely, the credit rating of a country is seen as a summary of the quality of the debt issued by this country. Therefore, this rating is a crucial part in determining the spread of sovereign bonds.
Pan and Singleton (2008) show that credit risk is mainly driven by default risk, but restructuring risk has an effect too. Restructuring of government bonds occurs when a government expects difficulties with repaying the debt. To ensure it can repay the debt, the terms of the bond can be renegotiated which may lead to longer maturities or lower interest rates. Therefore, when the credit quality of a sovereign is low, the risks of this debt will be higher. To be more specific, investors would like to be compensated for the higher risk associated with the credit quality and as a result the spread on the bond is higher.
Under normal market conditions, the importance of credit quality in determining sovereign spread, as described above, holds. However, a large change in market conditions switches the focus of debt investors to different debt characteristics. This becomes evident in the study Beber et al. (2009), which discusses a phenomenon called flight-to-quality and flight-to-liquidity in the Euro-area. They find that in normal market times bond investors indeed chase credit quality. However, in times of market distress, the liquidity of the credit determines the bond spreads. are determined mainly by the liquidity of the credit.
In addition to the aforementioned literature, the influence of credit ratings on both corporate and sovereign bonds is the focus of more studies (Collin-Dufresne et al., 2001; Aizenman et al., 2013). Both studies find that a downgrade of the credit rating causes an increase in the spread. This effect is stronger for bonds with a lower investment grade, because the risks and uncertainties of these bonds are higher. Due to the uncertainties attached to these bonds, news has a larger impact on the movement of the spread The influence of these ratings is so strong that countries or companies that have an S&P AAA rating are nearly risk-free, just as US Treasuries. The main example in this case are German government bonds. Germany has an S&P AAA rating and is therefore also perceived as relatively risk-free. These risk-free rates can be used to determine the sovereign spread (Collin-Dufresne et al. (2001). However, as discussed before this might not be the best method.
2.1.2 Global factors
6 influence of the US markets on the sovereign spread. Second, it elaborates on sovereign spread correlations. Third, it discusses the effect of volatility on the sovereign spread.
The United States is the largest economy in the world and hosts the largest stock market as well. Therefore, a lot of studies exist on the effect of market movements and economic changes in the United States on other economies and markets. The studies of Longstaff et.al. (2011) and Vamvakidis and Arora (2001) show the effect of the US on other economies. The state of the economy in the United States and the movement in its security markets affect the bond prices in other countries. Roll (1988) indicates the strong effect of US shocks on global markets was found by Roll (1988). Roll finds that globally, after the US stock market crash in 1987, 19 out of 23 stock markets declined with more than 20 percent. In addition to that, Longstaff et. al. (2011) find that this strong influence of markets in the United States also is also evident in bond markets. The influence of the United States is likely to remain in the nearby future. With a GDP of 19.4 trillion USD (2017)2 the United States remains the biggest economy in the world. Although China has gone through an enormous economic growth in the past couple of decades, its GDP of 12.2 trillion USD (2017)3 is still far behind the United States.
As becomes clear from aforementioned factors, both local and global determinants have an effect on bond prices. Because they are all driven by the same factors there also is a correlation between bond prices. Taken over all countries the average pairwise correlation of bonds is 62 percent as shown by Longstaff et al. (2011).
Ang and Bekaert (2002) study the effect of correlations during a crisis or market distress. The results of their research show that in times of crisis or market distress correlations tend to go up. Longstaff et al. (2011) confirm this result, because they find that in the period 2000 – 2006 the average CDS spread correlation was 39 percent. In contrast to this, the average correlation during the financial crisis (2007 – 2010) increased to 73 percent. This shows the large difference in correlation
The last global effect discussed in this paper is the effect of volatility, which is subject to many studies. Campbell and Taksler (2003) find that volatility in equity markets has an effect on corporate debt yields. Both Jones et al. (1998) and Bonfim (2003) find evidence of a relation between the price of debt and volatility. The strongest driver of this volatility are macroeconomic announcements. The availability of this new information will have an effect of on the valuation of bonds. This is in line with the efficient market hypothesis (Fama and
2 World Bank data
7 Malkiel, (1970)) which states that all available information will directly be reflected in the price after the macroeconomic announcement.
Contrary to what one might expect, Bonfim (2003) finds that positive news has a larger influence on the movements in the bond prices than negative news. In his study, unexpected positive news actually causes more volatility of the bond prices which leads to an increase in the volatility of the sovereign spreads. The market shocks caused by these high volatility events forces investors to rebalance their portfolios (for example from high yield to investment grade or from bonds to stocks). These activities could create correlations even though there is no correlation in the fundamentals of these asset classes. Longstaff et al. (2011) and Jones et al. (1998) find that investors respond more sensitive to news of risky debt. These announcements therefore have a stronger effect.
2.2 Hypothesis:
Considering the first hypothesis, I expect that the sovereign spreads of Eurozone countries becomes larger when it’s debt to GDP level increases. This relation is in line with the studies of Alexopoulou et al. (2009) and Bellas et al. (2010) that find a positive correlation between debt to GDP levels and bond spreads.
H1: A change in the debt level of a government has a significant effect on the sovereign spreads in Eurozone countries.
The second hypothesis is based on findings of Collin-Dufresne et al. (2001) and Aizenman et al. (2013). In their study they find that a change in the credit rating of a country has a significant effect on its credit spread. I therefore expect that a downgrade of the credit rating will result in a higher credit spread.
8 The United States is the largest economy in the world and hosts the largest stock market as well. Many research has shown that changes in the US economy also affects other countries around the world (Roll, (1988); Vamvakidis and Arora, (2001); Longstaff et al., (2011)). The US yield is also often regarded as the risk free rate of the world economy. Therefore a change in this rate is expected to have an influence of the sovereign spread since the definition of the sovereign spread is: 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑆𝑝𝑟𝑒𝑎𝑑 = 𝑦𝑖 − 𝑦𝑟𝑓. I therefore expect that an increase in the US yield will lower the sovereign spread.
H3: A change in the US Yield has a significant effect on the sovereign spreads in Eurozone countries.
3. Data and Methodology
This section covers the data and methodology of this thesis. First, it elaborates on the data and the data sources to construct the dependent, independent and shows the descriptive statistics. Second, it shows the different regression equations to test the different hypotheses. Third, it discusses the robustness tests.
Due to data availability on Reuters Datastream, the timeframe of this thesis restricts to the period of January 2008 – December 2017. Also, the GDP and debt data of the studied countries are collected by using the Thomson Reuters Datastream database. I will use quarterly data because this allows me to have sufficient observations and GDP data is published every quarter. To match this frequency I will also use quarterly CDS data. The countries credit rating was retrieved from the Standard and Poors website.
3.1 Variables
3.1.1 CDS Spread
9 The sovereign spread for a country can be measured by different methods. When using yields to determine the sovereign spreads, an assumption about the risk-free rate has to be made Hull et al. (2004) find this way of calculating spread highly questionable. Another method to calculate sovereign spread is using Credit Default Swap (CDS) contract prices.
A CDS is an insurance product that protects the buyer of the contract in case the counterparty defaults on its debt. The “protection buyer” pays a quarterly fee to the “protection seller”. In exchange, the buyer gets notional amount minus the recovered amount of the loan from the seller in case of, for example, default. The International Swaps and Derivatives Association (ISDA) states three important credit events. First, when due, failing to repay a coupon or the principal amount. Missing one coupon can already trigger this credit event but this most often results in a high recovery (Fontana and Scheicher (2016)). Second, restructuring of the debt by, for example, lowering the coupon. Third, a Moratorium or Repudiation, which entails the situation that a government or business (temporarily) stops making payments because it is in financial distress. In the case of a natural disaster, a moratorium temporarily seizes payments so that a country or business can focus on recovering. The price of a CDS is quoted in basepoints (BPS). When two parties enter a Credit Default Swap contract the value of the swap is zero because of the set premium. A CDS contract is terminated at the end of a set maturity or if one of the above credit events occurs.
In the papers of Longstaff et al. (2011), Hull et al. (2004) and Ferrucci (2003), the authors use CDS contract prices as a proxy for the sovereign spread. The price of a CDS contract is a good proxy for the sovereign spread for different reasons. First, the quoted CDS price is the price at which a broker is committed to trade a minimum principal and therefore reflects a real price. Bond yield data is most often an indication made by dealers, but they are not committed to trade at this price (Hull et al. (2004)). The second reason to use CDS spreads is that they are already spreads which does not require an adjustment of the data.
Gilchrist and Zakrajšek (2012) describe the sovereign spread as a strain in the financial system because it increases the price of debt due to the increase in the paid interest. These costs are a premium compared to risk-free debt to offset a sovereign’s additional risk. In the same study, Gilchrist and Zakrajšek find strong evidence that movements in the credit spreads are a good predictor of future economic activities and its risks. Therefore, a smaller credit spread seems to be a good indication of a more creditworthy debt. The reverse would be that a less creditworthy debt means that the credit spread is higher.
10 but the effect is rather small. Moreover, the effects on the results would be negligible. The reason why the effect is so small is mainly because most often CDS contracts require full collateralization. Another reason for this effect is that counterparty risk is not managed by a difference in pricing but via the choice of counterparties (Du et.al. (2018)).
Different countries of the current Eurozone are not in the sample. Due to data availability, this study excludes Luxembourg and Malta, because not enough CDS quotes are available on their debt. Other countries are left out due to their late membership to the Eurozone. Countries that joined the Eurozone after the fourth quarter of 2010 are not in this study. Therefore, Estonia, Latvia and Lithuania are not in this study, because they joined the Euro-zone in 2011, 2014 and 2015 respectively. Although Andorra, Monaco, San Marino and the Vatican have a monetary agreement to use the Euro, they are not in the Eurozone. Therefore, this study excludes these four city-states. Descriptive statistics of the CDS per country can be found in table 1 below. Table 1 provides an overview of the descriptive statistics of the CDS spread per country.
Table 1. Descriptive Statistics CDS spread by country
Country Minimum Average Maximum St. Dev. First obs. Last obs.
Austria 15.94 62.73 183.60 50.11 Q4 2008 Q4 2017 Belgium 14.50 81.31 307.42 73.70 Q1 2008 Q4 2017 Cyprus 16.00 440.38 1,482.05 410.72 Q1 2008 Q4 2017 Finland 11.25 31.03 80.70 17.04 Q2 2008 Q4 2017 France 15.57 64.75 214.86 49.46 Q4 2008 Q4 2017 Germany 5.95 32.50 114.36 25.40 Q1 2008 Q4 2017 Greece 21.90 11,092.57 37,030.49 16,268.94 Q1 2008 Q4 2017 Ireland 25.25 214.34 763.88 230.52 Q2 2008 Q4 2017 Italy 21.25 171.85 485.59 118.20 Q1 2008 Q4 2017 Netherlands 7.50 42.94 115.26 29.76 Q3 2008 Q4 2017 Portugal 20.85 321.79 1,187.74 304.15 Q1 2008 Q4 2017 Slovak R. 8.00 149.87 389.43 111.97 Q1 2008 Q4 2017 Slovenia 13.00 85.60 297.84 65.67 Q1 2008 Q4 2017 Spain 39.15 172.57 505.44 126.81 Q4 2008 Q4 2017 Average 16.87 926.02 3,082.76 1,277.32
Notes: CDS spreads are in basis points.
11
3.1.2 Debt to GDP ratio
The Debt to GDP ratio (Debt/ GDP) is the first explanatory variable in this study This ratio is chosen instead of using nominal debt levels because an increase in debt doesn’t have to be a problem as long as it is accompanied by an increase in production. Namely, when the production of a country increases, the income of that same country increases as well, which improves the ability of a country to repay its debt. Grauwe and Ji (2012) measure the Debt to GDP ratio exactly the same. Table 2 shows the descriptive statistics of the Debt to GDP ratios.
Table 2. Descriptive Statistics Debt to GDP ratio by country
Country Min Average Max St. Dev.
Austria 66.3 80.42 85.6 5.38 Belgium 89.3 104.21 111.3 5.71 Cyprus 44.8 81.65 110.1 24.02 Finland 28.8 51.47 64.2 10.55 France 66 89.04 99.3 9.51 Germany 63.9 73.08 80.9 5.46 Greece 103.7 155.89 180.9 25.17 Ireland 26.4 85.78 124.6 27.39 Italy 101 122.85 135 10.88 Netherlands 43.9 61.18 69.1 6.44 Portugal 67.4 112.13 133 22.99 Slovak R. 26.4 46.24 57.7 9.69 Slovenia 21.8 56.72 83.9 21.81 Spain 34.8 77.92 100.8 23.50 Average 16.87 926.02 3,082.76 1,277.32 3.1.3 Credit Rating
Changes in the Credit Rating have a strong influence on bond prices and therefore they also affect the sovereign spread. Collin-Dufresne et al. (2001) and Aizenman et al. (2013) show this effect in their research. To check for changes in the sovereign spread due to a change in the Credit Rating the variable CreditRating, the regression equations include the variable CreditRating. This variable uses data from Standard and Poor and was obtained from their database. To be able to use the Credit Rating data in my regression it had to be transformed into a numerical value. I gave an AAA rating the value 1, AA+ the value 2, AA the value 3 etc. Therefore, an increase in the value Credit Rating would reflect a downgrade so I expect that the Sovereign Spread us positively related to this variable.
3.1.4 US Yield
12 on markets globally as becomes clear from the literature review. Data for US yield is from the Federal Reserve database. I use the 5 year zero coupon bond and the data is in basis points.
3.2 Regression
This research contains both time series data and cross-sectional data. Therefore the correct way to do this research is by using a panel regression model. The time series data in this research contains the CDS spread from January 2008 to December 2017. The cross-sectional data is the CDS spread per country on a given time.
The following regression will be used to test the first hypothesis: 𝐶𝐷𝑆𝑖,𝑡 = 𝛼𝑡+ 𝛽1𝑖(
𝐷𝐸𝐵𝑇
𝐺𝐷𝑃 )𝑖,𝑡 + 𝜇𝑖+ 𝜐𝑖,𝑡
Formula 1: First debt to GDP regression model.
(1)
To test my second hypothesis the following regression will be used:
𝐶𝐷𝑆𝑖,𝑡 = 𝛼𝑡+ 𝛽1𝑖𝐶𝑟𝑒𝑑𝑖𝑡𝑅𝑎𝑡𝑖𝑛𝑔𝑡+ 𝜇𝑖+ 𝜐𝑖,𝑡
Formula 2: Credit rating model.
(2)
To test my third hypothesis I use the following regression:
𝐶𝐷𝑆𝑖,𝑡 = 𝛼𝑡+ 𝛽1𝑖𝑈𝑆𝑦𝑖𝑒𝑙𝑑𝑡+ 𝜇𝑖 + 𝜐𝑖,𝑡
Formula 3: US Yield model.
(3)
To test whether the above three are also jointly significant I use the fourth and last regression model:
𝐶𝐷𝑆𝑖,𝑡 = 𝛼𝑡+ 𝛽1𝑖(𝐷𝐸𝐵𝑇𝐺𝐷𝑃)
𝑖,𝑡+ 𝛽2𝑖𝑈𝑆𝑦𝑖𝑒𝑙𝑑𝑡+ 𝛽3𝑖𝐶𝑟𝑒𝑑𝑖𝑡𝑅𝑎𝑡𝑖𝑛𝑔𝑡+𝜇𝑖+ 𝜐𝑖,𝑡
Formula 4: Joint test of the explanatory variables.
13 To estimate this panel regression I will use a fixed effects model. Fixed effect models allow the cross-sectional intercept to differ but not over time. One problem with using a fixed effects model is that variables that do not vary over time will be canceled out of the regression. To check if fixed effects are indeed correct I will perform a Hausman (1978) test. This tests if the unique errors and independent variable are correlated. The null hypothesis of this test is no correlation between these errors and the independent variable. If you can’t reject the null hypothesis, random effects are preferred.
The type of data used in my research (historical time series) is inherent with heteroscedasticity and autocorrelation. This could yield incorrect results if not corrected for. A commonly used correction for heteroscedasticity and autocorrelation is the use of robust standard errors.
3.3 Robustness tests
As a first robustness check I will test if GIIPS countries react different to a change in one of the above mentioned factors. I test this by using the GIIPS dummy variable which takes a value of 1 for each GIIPS country. This dummy variable is used to create an interaction term between the dummy and the three independent variables. The GIIPS countries are Greece, Italy, Ireland, Portugal and Spain. During the financial crisis, these countries suffered from the highest levels of distress (Arnold, (2012); Acharya et al., (2018)). Jones et al. (1998) and Longstaff et al. (2011) find that investors are more sensitive to new information regarding risky investments. However, as mentioned before, another important driver behind sovereign spreads is correlation. Ang and Bekaert (2002) and Longstaff et al.(2011) find that during crisis periods correlations between sovereigns spreads are higher. My sample covers both the financial crisis and the European sovereign debt crisis so high correlation might exist. To test for commonalities in these spreads I conduct a Principal Component Analysis.
14
4. Results
This section of the thesis covers the results of my Principal Component (PC) analysis and the empirical results of the panel regressions.
4.1 Principal Component Analysis
To construct the PC analysis I first compute the correlation matrix of the quarterly CDS spread changes. This method is similar to the method used by Longstaff et al. (2011). The constructed correlation matrix is shown in the appendix. I find many large pairwise correlations in my sample. The correlation between the quarterly changes of the sovereign CDS spreads often exceeds 90 percent. For example, the pairwise correlation between Finland and Germany is 96 percent and the pairwise correlation between Austria and the Netherlands is 99 percent. An example of low pairwise correlation is between Slovenia and Spain, 62 percent. There are no negative correlations. The average pairwise correlation is 81 percent.
In times of economic distress or crisis correlations in the financial markets have the tendency to increase. Ang and Bakaert found evidence of the phenomenon in 2002 and Longstaff et al. confirmed these results in 2011 with their research. I use two subsamples to test if this effect also persists in the Euro-zone during (2008-2010) the financial crisis and after (2011-2017) the financial crisis. After computing the pairwise correlation of the CDS changes for the crisis period I find that the pairwise correlation of that period is 88 percent. Especially the correlations with Greece increased sharply. For example, the average pairwise correlation of the CDS spread change between Greece and Belgium increased from 83 percent to 95 percent. The average pairwise correlation of the CDS spread during a non-crisis period is 69 percent. These findings are in line with the findings by Ang and Bakaert (2002) and Longstaff et al. (2011).
15
Table 3. Results Principal Components Analysis
2008-2017 2008-2010 2011-2017 Principal component Proportion explained Cumulative Proportion explained Cumulative Proportion explained Cumulative Comp1 83.53 83.53 89.12 89.12 75.16 75.16 Comp 2 7.40 90.93 8.50 97.62 12.08 87.24 Comp 3 3.44 94.37 0.97 98.59 4.98 92.23 Comp 4 1.84 96.20 0.82 99.42 3.76 95.99 Comp 5 1.44 97.64 0.40 99.81 1.53 97.52
Notes: Values are in percentages and based on the correlation matrix of the quarterly changes of the
sovereign CDS spreads. All 14 countries in my dataset are used in this analysis.
Compared to Longstaff et al. (2011) the results of my Principal Component Analysis are even higher. In my sample, the first component explains about 84 percent of the variation in the sovereign CDS spread during the 2008-2017 full sample period compared to 84 percentage Longstaff et al. (2011) found in their research. The cause of this difference most likely has to do with the fact that the Longstaff et al. (2011) research focused on 25 sovereigns around the world compared to this research which only focusses on the Euro-zone only. The first component loadings are similar for most of the countries in the sample. Because the weighing vectors are similar one could conclude that this PC is a parallel shift factor of sovereign CDS spreads (Longstaff et al. (2011). An overview of the loading of the vectors can be found in table 4. The second Principle Component puts a negative weight on the Euro-zone countries that could be perceived as safer like Germany and the Netherlands. High positive weights are put on countries that were in distress during the financial crisis like Portugal, Spain and Greece.
16
Table 4. Eigenfactors
Comp1 Comp2 Comp3 Comp4 Comp5
Austria 0.28 -0.27 0.14 0.04 -0.10 Belgium 0.27 0.29 0.09 -0.11 -0.38 Cyprus 0.28 -0.07 -0.25 0.18 0.15 Finland 0.29 -0.18 -0.01 -0.08 0.09 France 0.28 0.10 -0.11 -0.28 -0.16 Germany 0.28 -0.12 -0.16 -0.44 -0.08 Greece 0.26 0.13 -0.28 0.72 -0.04 Ireland 0.25 -0.01 0.69 0.14 0.16 Italy 0.28 -0.01 -0.23 -0.21 0.04 Netherlands 0.28 -0.26 0.16 -0.13 -0.02 Portugal 0.21 0.60 0.11 -0.14 0.66 Slovak Rep. 0.27 -0.13 -0.39 0.06 0.25 Slovenia 0.27 -0.34 0.25 0.21 0.04 Spain 0.24 0.456 0.08 0.11 -0.50
4.2 Regression Analysis results
Table 5 displays the results of the estimation of equations 1, 2, 3 and 4. The results of equation one shows that it is evident that the relation between the Debt to GDP ratio is significant at the 95 percent confidence level. Furthermore, the relation is positive, which is in line with existing literature which also find that the Debt to GDP ratio of a sovereign increases the spread of its bonds (Alexopoulou et al., 2009; Bellas et al., 2010). These results support my first hypothesis.
Looking at the results of equation two we find that the coefficient of the Credit Rating is positive and significant at a 99 percent confidence interval. This means that, on average, for every Standard & Poor’s Credit Rating downgrade the spread increases by 53.88 basis points. These findings are in line with previous studies (Collin-Dufresne et al., 2001; Aizenman et al. ,2013). These results confirm my second hypothesis.
Considering the third equation, the yield of United States bonds affects the sovereign spread negatively, which is significant at the 99 percent confidence level. A one basis point increase of the US interest rate leads, on average, to a 1.37 basis point decrease of the sovereign spread. This is intuitive since the sovereign spread is the difference between the country’s yield on debt and the risk-free rate, which often is the US interest rate. Besides that, as discussed before, the US economy influences markets worldwide as confirmed by for example Vamvakidis and Arora, (2001) and Longstaff et al. (2011).
17 I found in equation three and is again significant at a 99% confidence level and negatively related to the sovereign spread. However, the value of the debt to GDP ratio coefficient is puzzling. In equation one I find a significant positive effect. Equation four shows that the coefficient remains significant (at a 99% confidence level), but with a negative sign. This is a surprising and counter intuitive result. Besides that there is also no literature supporting this finding. A reason for this sign change could be multicollinearity (Lipovetsky and Conklin (2003). Multicollinearity exists when two independent variables are highly correlated. To test for multicollinearity Variance Inflation Factors (VIF) are often used (Mansfield and Helms (1982)). Stata cannot calculate VIFs for panel data so to calculate I estimate a simple regression and add a dummy for every country in the sample. After this I calculate the VIFs and find that they are very high for the Debt_GDP (13.27) and Credit_Rating (12.41) variable. This confirms the existence of multicollinearity and is most likely the explanation of the change of sign.
To check if fixed effects are suitable for this data a Hausman test is performed. Table 6 contains the results of the Hausman test. This test developed by Hausman in 1978 tests if the unique errors and independent variable are correlated. The null hypothesis of this test is no correlation between these errors and the independent variable. In this case, I can reject the null hypothesis at a 99 percent confidence level and therefore assume that it’s right to use fixed effects in this model.
Table 5. Regression results equation 1, 2 ,3 and 4
(1) (2) (3) (4) Debt_GDP 4.07** (2.04) -8.39*** (-3.88) Credit_Rating 53.88*** (3.31) 92.62*** (3.90) US_Yield -1.37*** (-3.35) -1.02*** (-3.91) Constant -167.75 (-0.99) -95.05 (-1.16) 408.78*** (5.90) 591.22*** (3.48) F-statistic 12.67*** Observations 534 534 534 534 R2 0.21 0.49 0.06 0.30
Notes: Table 5 shows the results of the equation 1,2,3 and 4 regressions. The coefficients
are in Basis Points. This regression uses robust standard errors and Cross sectional fixed effects.
18
Table 6. Hausman test
Chi Squared Statistic P-value
Cross-section random effects 78.65 0.0000
4.3 Robustness tests
4.3.1 GIIPS versus non-GIIPS
Table 7 shows the result of the results of the first robustness test and compares the GIIPS (Greece, Ireland, Italy Portugal and Spain) countries with the other countries in the sample. To do this an interaction term with the GIIPS dummy is added to the equation.
In column one I find that the debt to GDP coefficient is not significant at the 90, 95 or 99 percent confidence interval. The interaction term GIIPS_dummy*DEBT_to_GDP is also insignificant and these confidence levels. This is not in line with existing literature by Jones et al. (1998) and Longstaff et al. (2011) who find that risky investments are more sensitive to new information than safer investments.
The second column shows that credit ratings do affect the sovereign spread. On average a downgrade of a sovereigns credit rating results in a 46.23 increase in the sovereign spread. This is similar to the results I found before and is also in line with literature by for example Collin-Dufresne et al.(2001) and Aizenman et al. (2013). I find no evidence that GIIPS countries are affected more by a change in its credit rating compared to non-GIIPS countries. As mentioned before existing literature suggests that this would be the case because riskier investment are more sensitive to the availability of new information (Jones et al., (1998); Longstaff et al., (2011)).
Considering the third column I find similar results as I found before. The US yield has a significant and negative relation with the sovereign spread. The results are significant at a 90 percent confidence level. I find no evidence that this relation is different for the GIIPS countries, compared to the non-GIIPS countries in the sample.
19
Table 7. GIIPS Robustness test
(1) (2) (3) Debt_GDP 1.67 (1.37) GIIPS_Dummy*Debt_GDP 3.93 (1.24) Credit_Rating 46.23*** (3.87) GIIPS_Dummy*Credit_Rating 12.09 (0.44) US_Yield -0.93* (-1.97) GIIPS_Dummy*US_Yield -1.23 (-1.51) Constant -110.16 (-0.89) -89.65 (-1.23) 410.87*** (6.25) Observations 534 534 534 R2 0.21 0.50 0.01
Notes: Table 7 shows the results of the first robustness test. The coefficients are in Basis Points. This
regression uses robust standard errors and Cross sectional fixed effects. The GIIPS_dummy takes the value 1for the countries Greece, Italy, Ireland, Portugal and Spain. The t-statistics are in parentheses. *** Significant at a 99 percent confidence level
** Significant at a 95 percent confidence level * Significant at a 90 percent confidence level
4.3.2 The effect of the Draghi announcement on July 26th 2012
To test whether spreads reacted different after the announcement of ECB president Mario Draghi on July 26th 2012 the Draghi_Dummy is introduced to the equation. This dummy is used in the interaction terms in table 8.
The first column of table 8 shows that there is a significant and positive relation between the debt to GDP level of a country and it’s sovereign spreads. This is similar to the results I found in table 5. On average when the debt to GDP ratio goes up by one point the spread increases by 7.50 basis points. This result is significant at a 99 percent confidence level and is in line with existing literature by Jones et al. (1998) and Longstaff et al. (2011). In column one I also observe a significant interaction term. The coefficient of the Draghi_Dummy*debt_GDP term is -1.54 and significant at a 95 percent level. This indicates that after the announcement of Draghi a change in the debt to GDP level lead to a 1.54 basis points smaller change than before. This is in line with research done by Van De Heijden et al. (2018) where they find that investors had more trust in Eurozone bonds.
20 of Draghi. Calm markets effect bonds as shown by Jones et al. (1998) and Bonfim (2003) who find that bonds are affected by volatility. The second column of table 8 shows that a downgrade of a sovereigns credit rating results in an, on average, 108.66 significant (99%) basepoints increase in the sovereign spread. The negative sign of the Draghi_Dummy*Credit_Rating indicates that after the Draghi announcement the sovereign spreads were less affected by a change of the sovereigns Standard and Poor’s credit rating than before. The interaction term is significant at a 99% confidence level.
Similar as before the US Yield is significantly at a 99 confidence interval. The coefficient is again negatively correlated with the sovereign spread of Euro-zone countries and has a value of -1.49. The Draghi_Dummy*US_Yield interaction term has a coefficient of -0.47. So after the announcement by Draghi the sovereign spread of Eurozone countries became sensitive to a change in the US yield. Van De Heijden et al. (2018) find that markets calmed down after the announcement of Draghi. So a possible explanation of this find could be that after the announcement investors started to focus more on factors outside of the Eurozone because they were more confident about the state of the Eurozone economy.
Table 8. Draghi announcement Robustness test
(1) (2) (3) Debt_GDP 7.50*** (3.36) Draghi_Dummy*Debt_GDP -1.54** (-2.85) Credit_Rating 108.66*** (6.98) Draghi_Dummy*Credit_Rating -38.11*** (-3.52) US_Yield -1.49*** (-3.67) Draghi_Dummy*US_Yield -0.47* (-2.02) Constant -377.20* (-2.08) -242.92*** (3.60) 467.26*** (6.65) Observations 534 534 534 R2 0.23 0.56 0.08
Notes: Table 8 shows the results of the second robustness test. The coefficients are in Basis Points.
This regression uses robust standard errors and Cross sectional fixed effects. The Draghi_Dummy
takes value 1 for all quarters after the announcement of Draghi on July 26th 2012. The t-statistics are
in parentheses.
21
5. Conclusion
This study explores the effect of the debt level, credit rating and US yield changes on the sovereign spreads in Eurozone between 2008 and 2017. I use Credit Default Swap data as a proxy for the sovereign spread. This section summarizes the main findings of this study and discusses its limitations.
First, a relation between the debt level of a country and its sovereign spread exists, it is a positive relation, this is in line with what current literature predicts. An increase of the debt to GDP ratio by one, increases the sovereign spread by, on average, 4.07 basis points. I also find that the sovereign spread of GIIPS (Greece, Italy, Ireland, Portugal and Spain) countries do not significantly react different to changes in a countries debt to GDP ratio compared to the other Eurozone countries. This is not in line with current literature which suggests that riskier investments are more sensitive to new information than safer investments (Jones et al. (1998); Longstaff et al. (2011)). However I do find that in the period after the announcement of European Central Bank president, Mario Draghi, sovereign spreads we’re less exposed to changes in the debt to GDP ratio than before.
Second, this paper also explores whether a relation between the sovereign spread of Eurozone countries and a countries credit rating by Standard and Poor’s exists. This relation exists. When Standard and Poor’s downgrades a sovereigns credit rating the spread on the sovereigns debt on average increases with 53.88 basis points. My research indicates that this effect does not differ between GIIPS and non-GIIPS countries. I did find highly significant results when testing the effect of the announcement of ECB president Mario Draghi. My results indicate that in the period after the announcement a downgrade of a countries credit rating caused a 38.11 basis point smaller increase in its sovereign spread than before.
22 Besides the relation between the sovereign spread and debt level, credit rating and US yield I also tested the commonality in sovereign spreads. I find that the sovereign spreads tend to have high pairwise correlations between countries. A Principal Component Analysis showed that about 84 percent of the variation in the sovereign CDS spread is explained by the first component. This is high and the reason for this is that my sample covers the financial crisis of 2008 and the EU sovereign debt crisis. During periods of crisis correlations increase as shown by Ang and Bekaert (2002) and Longstaff et al. (2011). These high correlations and commonalities could be an explanation for why I don’t find any difference between the GIIPS and non-GIIPS countries.
5.1 Limitations
One of the main limitations of my research is its time frame. I could not use data before 2008 because pre-2008 CDS quotes aren’t available in the databases provided by the University. It would have been interesting to study the pre-financial crisis effects in the Euro-zone so they could be compared to the post-financial crisis effects.
23
Bibliography
Acharya, V. V., Eisert, T., Eufinger, C., Hirsch, C. W. (2018). Whatever it takes: The real effects of unconventional monetary policy.
Afonso, A., Arghyrou, M.G., Kontonikas, A. (2015), The determinants of sovereign bond yield spreads in the EMU. European Central Bank working paper.
Aizenman, J., Binici, and M., Hutchison, M. (2013). Credit ratings and the pricing of sovereign debt during the euro crisis
Alexopoulou, I., Bunda, I., Ferrando, A. (2009). Determinants of government bond spreads in new EU countries. European Central Bank working paper.
Ang, A., Bekaert, G. (2002). International asset allocation with regime shifts. Review of Financial Studies, 15(4), 1137-1187.
Arnold, I. (2012). Sovereign debt exposures and banking risks in the current eu financial crisis. Journal of Policy Modeling, 34(6), 906-920.
Arora, N., Gandhi, P., Longstaff, F. A. (2012). Counterparty credit risk and the credit default swap market. Journal of Financial Economics, 103(2), 280-293.
Beber, A. (2009). Flight-to-quality or flight-to-liquidity? evidence from the euro-area bond market. Review of Financial Studies, 22(3), 925-957.
Bellas, D., Papaioannou, M. G., and Petrova, I. (2010). Determinants of emerging market sovereign bond spreads.
Campbell, J. Y., Taksler, G. B. (2003). Equity volatility and corporate bond yields. The Journal of Finance, 58(6), 2321-2350.
Collin-Dufresne, P., Goldstein, R. S., and Martin, J. S. (2001). The determinants of credit spread changes. The Journal of Finance, 56(6), 2177–2207.
24 Dieckmann, S. and Plank, T. (2012). Default risk of advanced economies: An empirical analysis of credit default swaps during the financial crisis. Review of Finance, 16(4):903–934.
Du, W., Gadgil, S., Gordy, M. B., Vega, C. (2018). Counterparty risk and counterparty choice in the credit default swap market.
Ferrucci, G. (2003). Empirical determinants of emerging market economies’ sovereign bond spreads. Quarterly Bulletin- Bank of England, 43, 457-457.
Fontana, A., Scheicher, M. (2016). An analysis of euro area sovereign cds and their relation with government bonds. Journal of Banking and Finance, 62, 126-140.
Hausman, J. (1978). Specification Tests in Econometrics. Econometrica, 46(6), 1251-1271
Hull, J., Predescu, M., White, A. (2004). The relationship between credit default swap spreads,bond yields, and credit rating announcements. Journal of Banking & Finance, 28(11):2789–2811.
Jones, C. M., Lamont, O., Lumsdaine, R. L. (1998). Macroeconomic news and bond market volatility. Journal of Financial Economics, 47(3):315–337.
Lipovetsky, S., Conklin, W. M. (2003). Classroom note: A model for considering multicollinearity. International Journal of Mathematical Education in Science and Technology, 34(5), 771-777.
Longstaff, F. A., Pan, J., Pedersen, L. H., Singleton, K. J. (2011). How sovereign is sovereign credit risk? American Economic Journal: Macroeconomics, 3(2):75–103.
Malkiel, B. G., Fama, E. F. (1970). Efficient capital markets: A review of theory and empirical work. The journal of Finance, 25(2), 383-417.
Mansfield, E., Helms, B. (1982). Detecting Multicollinearity. The American Statistician, 36(3), 158-160.
Modigliani, F., Miller, M. (1958). The Cost of Capital, Corporation Finance and the Theory of Investment. The American Economic Review, 48(3), 261-297.
25 Roll, R. (1988). “The International Crash of October 1987.” Financial Analysts
Journal, 44(5): 19-35.
Pan, J., Singleton, K. J. (2008). Default and recovery implicit in the term structure of sovereign cds spreads. The Journal of Finance, 63(5):2345–2384.
Vamvakidis, M. A., Arora, M. V. B. (2001). The Impact of US Economic Growthon the Rest of the World: How Much Does it Matter?
Van Der Heijden, M., Beetsma, R., Romp, W. (2018). ‘whatever it takes’ and the role of eurozone news. Applied Economics Letters, 25(16), 1166-1169.
26
Appendix
Appendix 1: Correlation Matrix of the Quarterly Sovereign CDS spread changes
Austria Belgium Cyprus Finland France Germany Greece
Austria 1.00 Belgium 0.81 1.00 Cyprus 0.89 0.84 1.00 Finland 0.98 0.84 0.94 1.00 France 0.88 0.94 0.89 0.92 1.00 Germany 0.93 0.85 0.90 0.96 0.94 1.00 Greece 0.80 0.83 0.88 0.83 0.85 0.78 1.00 Ireland 0.86 0.81 0.75 0.83 0.78 0.75 0.69 Italy 0.91 0.87 0.93 0.95 0.95 0.96 0.84 Netherlands 0.98 0.81 0.88 0.98 0.90 0.94 0.77 Portugal 0.52 0.80 0.63 0.61 0.75 0.61 0.68 Slovak Rep. 0.89 0.79 0.93 0.92 0.88 0.90 0.84 Slovenia 0.98 0.73 0.85 0.94 0.81 0.85 0.75 Spain 0.67 0.93 0.74 0.72 0.84 0.71 0.80 Appendix 1: Continued
Ireland Italy Netherlands Portugal
Slovak
Rep. Slovenia Spain
