Evolving determinants of sovereign yield spreads for core and periphery Eurozone countries

MSc Business Economics, Finance track

Master Thesis

Evolving Determinants of Sovereign Yield Spreads

for Core and Periphery Eurozone Countries

Name: Tong FU

Supervisor: P.F.A. Tuijp, MPhil

Date: August 14, 2015

Abstract

This paper studies the main determinants of sovereign bond yield spreads for 10 EMU countries in different periods from 1990Q1 to 2014Q4, namely pre-Euro period, the convergence period, the crisis period and the divergence period. The countries are divided into the core and the periphery group. I find that firstly, yield spreads decreased substantially after the introduction of the Euro but increased again after the burst of global financial crisis after controlling for credit and liquidity risks. Secondly, credit quality is the main driver of yield spreads especially in stable markets; but during the convergence period, the credit risk is not a good predictor for periphery countries. Thirdly, the liquidity effect is more significant for periphery countries before the introduction of Euro and in the sovereign crisis period.

Table of Contents

Abstract ... 1

I. Introduction ... 3

II. Literature Review ... 5

III. Data ... 8

IV. Empirical Method ... 12

V. Empirical Results ... 16

VI. Robustness Check... 22

VII. Conclusion ... 29

References ... 31

Appendices ... 33

I. Introduction

The introduction of the Euro had a significant impact on the yield curves of government bonds in the European Monetary Union (EMU) (Oliveira et al., 2012). However, despite the apparent convergence of the yields on long-term public debt, the yield differentials did not disappear completely. What’s more, after the 2008– 2009 international financial crisis and the following European sovereign crisis, the credit spread rose rather sharply, which could be due to increased credit risk, increased liquidity risk, or worsened investors’ risk perceptions (Manganelli and Wolswijk, 2009). Therefore, the following question arises: how do the main determinants of yield spreads on government bonds evolve in different stages of the European sovereign debt crisis for core and periphery countries in Eurozone? Understanding the main source of the observed widening of spreads is beneficial to both policymakers who can launch proper policies accordingly, and practitioners who could benefit from sovereign bond investments that have a certain price pattern. For example, if the spread is mainly driven up because of investors’ distrust in the fiscal sustainability of a country, a credible financial restructuring plan becomes rather crucial. Otherwise, if those spread movements are due to changes in the liquidity situation such as a “severe liquidity squeeze”, then it is rather significant for policymakers to boost liquidity supply to markets (Sgherri and Zoli, 2009).

Existing literature that studied the roles of credit risk and liquidity conditions on the yield spread have arrived at quite different conclusions. Firstly, while many studies found evidence for the importance of credit risk factors (e.g.: Beber et al., 2009; Cantor and Packer, 1996), a few studies still question their roles in determining changes in yield spreads (Collin-Dufresne, Goldstein, and Martin, 2001). Secondly, for the liquidity factor, researchers also failed to reach a consensus conclusion. For example, the results of Monfort and Renne (2013) and Bernoth et al. (2012) showed that the yield spreads significantly differ between

countries with different liquidity conditions, at least in some certain periods; however, Oliveira et al. (2012) found an insignificant role for the liquidity risk factor in explaining credit spreads both before and during the crisis. These findings imply that neither the credit risk nor the liquidity risk factor impacts the yield spreads at a constant level and that the determinants of the spread might evolve over time (Beber et al., 2009).

Based on the conclusions of the existing literature, this paper makes its contribution in the following ways. Firstly, my analysis focuses on the changing determination of spreads during different stages of the European debt crisis. Secondly, instead of studying Euro-area countries as whole, I divide them into two groups based on their performance, namely core countries and periphery countries. Previous literature has found evidence that investors’ behaviors are different not only between stable and stressed market conditions, but also between solvent and insolvent countries and hence could influence the yield curves of the bond. For example, Arghyrou and Kontonikas (2012) find that while before the financial crisis, the investors bought periphery government bonds because they expect that their yields would convergence with those of Germany, the market pricing behavior is more driven by macro-fundamentals and international risk after August 2007. Moreover, Beber et al. (2009) found that investor decisions are made more based on liquidity, instead of credit quality, when the market is in stress. However, analysis on the importance of different sovereign yield spread based on combined categorization of this kind with recent data has not been performed before, and this study could shed interesting lights on this topic.

The aim of this paper is to disentangle the changing risk factors in bond prices during the course of debt crisis and provide appropriate policy suggestions about the widened yield spread for different Euro-zone countries.

In order to reach my goal, fixed-effects panel regressions are applied to study the determinants of yield spreads between 10 Eurozone countries and the

benchmark country. I follow the method of Manganelli and Wolswijk (2009) to regress the yield spreads against credit risk and liquidity risk, and then include the fiscal variables (Bernoth et al., 2012) and country-specific macroeconomic fundamentals (Oliveira et al., 2012) into the model. My analysis covers the time period from 1990Q1 to 2014Q4 (quarterly frequency), which includes four sub-periods: the pre-Euro period, the convergence period, the crisis period, and the divergence period. I first run the regression for the whole dataset, and then add group and period dummies to obtain more specific results for the evolving roles of different factors. The main findings are: (i) yield spreads decreased substantially after the introduction of the Euro but increased again after the burst of global financial crisis after controlling for credit and liquidity risks; (ii) credit risk is the main driver of yield spreads especially in stable markets; but during the convergence period, it is not a good predictor for periphery countries, probably due to the bail-out commitment of ECB; (iii) the liquidity effect is more significant for periphery countries before the introduction of Euro and in the sovereign crisis period.

The paper is structured as follows: Section II discusses the related literature. Section III presents the data and some preliminary analyses. Section IV introduces the methodology. Section V reports my empirical results. In Section VI I check the robustness of my results. Section VII concludes.

II. Literature Review

The existing literature typically considers European government bond yields to be determined mainly by two risk variables: credit risk and liquidity risk (Beber et al., 2009; Manganelli and Wolswijk, 2009; Monfort and Renne, 2014).

First of all, the credit risk premium measures the financial compensation investors demand to cover the risk that a government defaults and is expected to explain a considerable portion of Eurozone sovereign credit spreads. It is worth noticing that the long-term sustainability of public finances instead of short-term

budget deficit fluctuations mainly affects the probability of default and therefore accounts for the credit risk premium. A natural proxy for such long-term sustainability could be given by the assessments provided by the leading rating agencies. According to Cantor and Packer (1996), sovereign ratings are “assessments of the relative likelihood that a borrower will default on its obligations”. Although the role of credit risk may seem quite apparent, a few previous analyses still reached quite different conclusions. For instance, by using dealer's quotes and transactions prices on U.S. industrial bonds from 1988 to 1997, Collin-Dufresne, Goldstein, and Martin (2001) found that changes in yield spreads were not associated with natural credit risk factors.

Secondly, a liquidity risk premium is another possible explanation for government bond yield differentials across Eurozone members, which measures the extra interest rate an investor requires to be compensated for bearing the risk of having to liquidate the security at a lower price with respect to the benchmark. However, previous literature has reached different conclusions on the role of liquidity. For instance, Monfort and Renne (2013) used an intensity-based model of euro-area sovereign spreads to show that a substantial part of intra-euro spreads’ fluctuations is mainly liquidity-driven for most of the countries in the pre-crisis period. The study of Bernoth et al. (2012) on bond yield differentials among EU government bonds issued between 1993 and 2005 also provided evidence that before the introduction of Euro, the interest rates paid by countries with larger market shares in the DM(Euro) market are significantly lower than those paid by countries with smaller market shares, indicating that liquidity premiums significantly influence yield spreads; however, this effect disappeared after the start of EMU. Oliveira et al. (2012) studied the determinants of government credit spreads in the Euro-area from 2000 to 2010 and found that the coefficient of liquidity risk factor is not significantly different from zero before the financial crisis period, and that it becomes statistically insignificant during the crisis. Not only did earlier studies find different results about the roles of credit risk

and liquidity condition, but they also found that even if investors are driven both by credit quality and liquidity, they probably do so at different times and for different reasons (Beber et al., 2009). For example, according to Favero et al. (2010), while the aggregate risk factor is consistently priced and leads to a common trend in spreads, the role of liquidity is only significant for a certain subset of countries. In their liquidity model, transaction costs have two opposing effects: on the one hand, bond with higher liquidity have lower required yields and higher transaction costs lead to higher yields; on the other hand, transaction costs reduce the trading probability of a bond, which increases the relative bond price and hence lowers the yields. Moreover, the latter effect is proportional to the magnitude of transaction costs. Therefore, the liquidity effect will become more apparent for countries with rather small bid-ask spreads, where the conteractive effect is smaller. Arghyrou and Kontonikas (2012) find that while before the financial crisis, the investors bought periphery government bonds because they expect that their yields would convergence with those of Germany, the market pricing behavior is more driven by macro-fundamentals and international risk after August 2007. Beber et al. (2009) find that sovereign yield spreads is mainly explained by differences in credit quality especially for low credit risk countries and during times of heightened market uncertainty. This, in my study, refers to the core Eurozone countries in the period when the global financial crisis was gradually mutating into a sovereign debt crisis in Europe. However, Beber et al. (2009) also find evidence that when in times of market stress, liquidity explains a substantially greater proportion of spreads especially for the destination of large flow of funds into the bond market. This will also be studied through my investigation into the core countries during the period of European sovereign debt crisis after late 2009.

The disagreement on the significance of the two risk factors’ effects in previous studies also suggest that there may be other factors that could affect credit spread. Firstly, for the variables that have possible influence on credit risk, Geyer et al.

(2004) argue that business cycle variables are directly related to the probability of default. The government’s creditworthiness is likely to improve when it experiences a higher output growth, inasmuch as it reflects a favorable business cycle and anticipates an economy expansion (Oliveira et al., 2012). Secondly, for the variables that link to bond market liquidity condition, Goyenko and Ukhov (2009) find that illiquidity is mostly affected by monetary policy variables, since loose monetary conditions can reduce the costs of margin loan requirements as well as make it easier for dealers to finance their positions, which will increase liquidity and encourage trading.

Studying the main determinants of the widened yield spread is of chief relevance to many aspects. According to Sgherri and Zoli (2009), persistently higher spreads can lead to higher funding costs for Eurozone governments and further weaken the positive effects of low risk-free rate policies. For policymakers, decomposing the spreads into a credit risk and a liquidity risk premium is quite relevant. For example, a large liquidity premium may indicate incomplete bond market integration, pointing to the need for liquidity provision measures; but if the increase in spreads due to the market’s concerns about credit risks, govern-ments should focus on improving the solvency and the sustain-ability of public finance (Codogno et al., 2003; Sgherri and Zoli, 2009). Furthermore, such a decom-position could also provide a lesson for portfolio management. Specifically, bonds with lower prices due to poor liquidity conditions would interest those longer-term investors, since their long-run returns are higher compared to more liquid bonds with the same credit quality (Longstaff, 2009). In addition, the findings of Favero et al. (2010) suggests that liquidity condition (which is reflected by transaction costs) and aggregate risk combined have an impact on the risk sensitivity of the assets being held.

III. Data

This paper uses panel data to analyze the main determinants of EMU members’ 8

government bonds yield spreads in different stages of the European sovereign debt crisis. The empirical analysis will be focused on the following 11 countries in the Eurozone: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal and Spain. The analysis will cover the period of 1990-2014, which is further divided into four sub-periods. The first period, called the “pre-Euro period”, is from 1990 to 1998, when the euro had not been officially introduced. Then after the crisis broke out, I distinguish between the “convergence period” preceding the global financial crisis which burst in the summer of 2007, the “crisis period” during which the global financial crisis in Europe was gradually mutating into a sovereign debt crisis, and the following “divergence period” when the European sovereign debt crisis entered a new phase in late 2009, during which a number of countries reported larger-than-expected increases in deficit/ GDP ratios.

Moreover, in order to get a better understanding about the importance of different determinants, the 10 Eurozone countries above (except for Germany which is regarded as the benchmark country) will be divided into two groups, namely core and periphery countries. The grouping criteria is whether a country’s ten- year bond yield exceeded 7 percent in the divergence period (after Quarter 3, 2009). A 7 percent yield was seen by many in markets as a critical line, since bond yields higher than that level meant that the cost of rolling over sovereign debt in the next few years would be so high that it would become economically unsustainable for the government. The periphery group contains Greece, Ireland and Portugal, the latter two of which were both forced to seek bailouts after their sovereign yields exceeded 7 percent. Those three countries suffered from badly impaired sovereign-debt markets and were also placed under EU-IMF programs (Monfort and Renne, 2013).

The yields of Euro-Area government bonds are generated from the DataStream and IMF dataset, on a quarterly basis. The credit ratings are obtained from Oxford Economics. The liquidity effect is estimated by a liquidity variable. Although bid-

Table 1. Descriptive Statistics

This table contains descriptive statistics for the dependent and independent variables in four sub-periods. The variables are specified as follows: (a) The yield spread between a ten-year government bond issued in Eurozone country i and the benchmark country Germany. (b) Credit ratings from Oxford Economics, where 20 = AAA. (c) The ratio of the debt issued by a government to the total debt of Eurozone countries. (d) Deficit to GDP ratio at the end of a fiscal year averaged on a quarterly basis (differential versus Germany). (e) Six-month interest rate of country i. (f) Average quarterly real GDP growth (differential versus Germany). (g) The level of consumer prices represented by quarterly CPI (differential versus Germany).

Pre-Euro Period Convergence Period

mean sd min max mean sd min max

(a) spread 2.27 3.12 -0.20 17.57 0.18 0.23 -0.10 2.24 (b) rating 17.88 2.49 11.00 20.00 18.81 1.60 12.33 20.00 (c) liquidity 0.08 0.09 0.01 0.32 0.07 0.08 0.01 0.28 (d) deficit 1.39 3.61 -7.69 11.73 -1.10 2.97 -8.10 6.82 (e) interest 4.12 1.84 1.00 9.00 5.73 2.03 1.00 9.00 (f) GDP growth 0.13 1.39 -8.95 8.67 0.34 0.96 -1.86 5.67 (g) inflation 1.39 3.70 -3.50 20.33 1.01 1.04 -2.04 4.84 N 360 340

Crisis Period Divergence Period

mean sd min max mean sd min max

(a) spread 0.58 0.53 0.08 2.65 2.76 3.92 0.20 23.98 (b) rating 18.84 1.62 15.00 20.00 15.74 5.03 1.33 20.00 (c) liquidity 0.08 0.08 0.01 0.27 0.08 0.08 0.01 0.25 (d) deficit 3.15 3.77 -4.80 11.85 4.74 6.63 -1.50 86.02 (e) interest 4.93 1.94 1.00 8.80 5.78 1.31 2.80 9.00 (f) GDP growth 0.05 1.28 -3.82 3.86 -0.51 0.99 -4.84 3.25 (g) inflation 0.24 1.29 -5.87 2.52 0.15 1.21 -6.17 4.42 N 90 200

Table 2. Correlation Matrix

This table contains cross-correlation statistics for the dependent and independent variables. All variables are defined in the Appendix at the end of the main article.

spread rating liquidity deficit interest GDP growth inflation spread 1.0000 rating -0.7451 1.0000 liquidity -0.0413 0.0619 1.0000 deficit -0.3930 0.4509 -0.1052 1.0000 interest -0.1305 0.1142 -0.0206 0.0502 1.0000 GDP growth -0.2183 0.1606 -0.0977 0.2483 -0.0988 1.0000 inflation 0.4202 -0.2663 -0.1020 -0.0860 0.0018 -0.0568 1.0000 10

ask spreads which reflect trading costs in trading securities are widely used in earlier studies (Fleming, 2003), I am not able to follow this approach because the dataset of bid-ask spreads is only available from 1999Q1. Therefore, I follow the method employed by Bernoth et al. (2012) to assume that liquidity depends on market size. This method is based on the work of Gravelle (1999), in which the author shows that the correlation between bid-ask spreads and the total supply of debt is significantly negative. Hence, the size of the market for a given security is suggested to have a positive effect on its liquidity. By additionally assuming that all debt issued by a government in a given currency is homogeneous up to maturity, I use the ratio of the debt issued by a government in to the total debt of EU countries as a proxy of the liquidity. Macroeconomics data are taken from World Bank’s dataset.

Table 1 shows the descriptive statistics for both dependent and independent variables. The average yield spread, reported in the ”mean” column of Table 1, varies between 0.28 and 2.76 for the different selected periods, and there is also quite some variation in the spread, as can be seen from the standard deviations, reported in the “sd” column, which vary from 3.92 to 0.23. According to Manganelli and Wolswijk (2009), interest rate spreads of EMU 10-year sovereign bonds against the German benchmark went through a dramatic decline before the introduction of euro, then the yield spreads reached their low points at around 2004-2005 and rebounded after 2007, exceeding the levels in the initial stage of EMU. That trend can already be observed in Table 1: both the mean and the stan-dard deviation take the highest value in the divergence period and the lowest in the convergence period. Table 2 shows the correlation matrix for all the variables. To get a visual impression on the evolvement of the yield spread, the behavior of country-level ten-year bond yields for the selected 15 euro area countries from Quarter 1 1990 through Quarter 3 2014 is depicted in Figure 1. Several particularly featured periods stand out. First, from the year of 1990 to 1998, the government bond yields declined sharply and evidently converged. Second, between 1999 and

Figure 1. Yields on Ten-Year Sovereign Bonds

This graph shows the different country-level ten-year bond yields (percent unit) for 15 euro area countries1 _{from Quarter 1 1990 through Quarter 3 2014. Data are obtained from}

International Financial Statistics (IFS).

2008, the yields continued to decrease but at a rather slow speed and the trend of convergence fluctuated to some extent. Then after the burst of global financial crisis, credit spreads in the Eurozone started to increase again, but it was not quite apparent until Q3 2009, when the spreads came across a rather sharp rise. Within this period, the Greek yield was the first to diverge from other countries and Greece soon required a bailout in May 2010. Ireland was the next to join the EU-IMF programs in November 2010, followed by Portugal in May 2011 (Lane, 2012), which behaved as a comovement between their yields in Figure 1. This was then followed by Italy and Spain whose spreads were at an intermediate level between

1 _{Austria, Belgium, Cyprus, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Malta,}

Netherlands, Portugal, Slovak Republic, Spain

0 5 10 15 20 25 30 1990 1993 1996 1999 2002 2005 2008 2011 2014 Austria Belgium Cyprus Finland France Germany Greece Ireland Italy Luxembourg Malta Netherlands Portugal Slovak Republic 12

the bailed out (periphery) countries and the core countries (such as Germany and France). The yield spreads peaked at mid-2012, and dropped again thereafter. Most euro-area countries have reached a new low since 2014 except for Greece and Cyprus.

IV. Empirical Method

From the previous summary statistics, we can see that the mean of the yield spread drops from 2.27 in the pre-Euro period to 0.18 in the convergence period. Based on that observation, my first hypothesis is made as follows.

H1: When controlling for credit and liquidity risk, the yield spreads tend to decrease after the introduction of euro.

Beber et al. (2009) find that sovereign yield spreads is mainly explained by differences in credit quality especially for low credit risk countries and during times of heightened market uncertainty. This, in my study, refers to the core Eurozone countries in the period when the global financial crisis broke out in 2007Q3. Therefore, my second hypothesis is stated as:

H2: The spreads of core countries are more driven by the credit risk factor during the crisis compared to the convergence period before 2007.

In addition, Beber et al. (2009) also find evidence that when there are large flows into or out of the bond market, liquidity explains a substantially greater proportion of spreads for the destination market (countries with lower credit risks). This can be examined indirectly using my third hypothesis.

H3: The effect of the liquidity risk factor on bond yield spreads will increase especially for the core countries after late 2009.

In order to test these hypotheses, I first develop a basic regression model which contains only the main explanatory variables:

𝑟𝑟𝑡𝑡𝑖𝑖− 𝑟𝑟𝑡𝑡𝐺𝐺 = 𝛼𝛼 + 𝛽𝛽′𝑋𝑋𝑖𝑖𝑡𝑡+ 𝛾𝛾′𝑌𝑌𝑖𝑖𝑡𝑡+ 𝛿𝛿′𝑊𝑊𝑖𝑖𝑡𝑡+ 𝜇𝜇𝑖𝑖 + 𝜀𝜀𝑖𝑖𝑡𝑡 (1) The dependent variable is the yield spread between a ten-year government bond issued in EMU country i and the benchmark country Germany, both denominated in the same currency. 𝑋𝑋𝑖𝑖𝑡𝑡 is a vector of explanatory variables which include two risk factors, namely the bond market liquidity conditions and the credit risk factor. The Ratings from Oxford Economics are used as a proxy of credit risks. The value range is from 1.33 to 20, where 20 represents the credit level of AAA. For the liquidity risk factor, I employ the ratio of the debt issued by a government in to the total debt of EU countries as explained in the previous sector.

For the fiscal variables 𝑌𝑌𝑖𝑖𝑡𝑡 , which reflect the government's quality as a borrower, I use the commonly used deficit/GDP ratio (Bernoth et al., 2012), measured as the difference relative to the benchmark country, Germany. Sgherri and Zoli (2009) found that markets pay much more attention to the potential fiscal implications and sovereign debt situations after August 2008. In addition, short-term interest rates are included, with an expected positive coefficient, since Manganelli and Wolswijk (2009) found that an increase in interest rates is associated with a widening of spreads and conversely a tightening of monetary policy with a reduction in spreads.

Finally, 𝑊𝑊𝑖𝑖𝑡𝑡 are country-specific macroeconomic fundamentals that are selected based on and based on existing economic literature and empirical reasoning in order to measure the economic climate. Oliveira et al. (2012) find that during the crisis, the market behavior of the sovereign credit spreads seems to be strongly determined by those factors. I followed their method to use the output growth as a proxy for the business cycle and include the level of consumer prices to take into account the effect of inflation (differential versus Germany). Geyer et al. (2004) find a negative relationship between GDP growth and the spread, since a higher output indicates a positive economic situation. Oliveira et al. (2012) point out that when the inflation level is high, it means that the real value of the government debt

is reduced and hence may reflect structural financial problems.

The model also incorporates country-specific fixed effects 𝜇𝜇𝑖𝑖, time-specific fixed effects 𝜂𝜂𝑡𝑡, and an error term 𝜀𝜀𝑖𝑖𝑡𝑡.

Then I extend regression (1) to compare among different groups and periods. The first extended form includes three period dummies as follows:

𝑟𝑟𝑡𝑡𝑖𝑖− 𝑟𝑟𝑡𝑡𝐺𝐺 = 𝛼𝛼 + 𝛽𝛽1𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡+ 𝛽𝛽2𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡+ 𝛾𝛾′𝑌𝑌𝑖𝑖𝑡𝑡+ 𝛿𝛿′𝑊𝑊𝑖𝑖𝑡𝑡+ � 𝜌𝜌1𝑘𝑘𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜌𝜌2𝑘𝑘𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡,× 𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜌𝜌3𝑘𝑘′𝑌𝑌𝑖𝑖𝑡𝑡× 𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜌𝜌4𝑘𝑘′𝑊𝑊𝑖𝑖𝑡𝑡× 𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜌𝜌5𝑘𝑘× 𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + 𝜇𝜇𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑡𝑡 (2)

where 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 represents the liquidity factor and 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 is the credit risk factor. 𝐷𝐷𝑘𝑘 (𝑘𝑘 = 1, 2, 3) represents the period of Jan. 1999- June 2007 (Convergence Period), July 2007- Sept. 2009 (Crisis Period) and Oct. 2009- Dec. 2014 (Divergence Period) respectively, while Jan. 1990 – Dec. 1999 (Pre-Euro Period) is denoted as Period 0. By adding these dummies with their interactions with other explanatory variables, I will be able to test whether there is a significant difference among the coefficients of determinants before the introduction of euro and in different periods of the crisis.

Next, I repeat specification (2) but divide the sample into core countries and periphery countries. This set of regressions allow me to study whether the deter-minants are different between the two groups. Table A1 in the Appendix at the end of the main article lists all the variables used in the models and their descriptions. A few expectations could be made based on my hypotheses. First, according to H1 that the yield spreads tend to decrease after the introduction of euro, 𝛼𝛼 should be smaller than zero. Secondly, according to H2 that the yield spread is more affected by credit risk factor after the global financial crisis, the coefficient for the credit ratings in the crisis period should be larger than that in the

convergence period, i.e. 𝜌𝜌2,2 < 𝜌𝜌2,3. Thirdly, according to H3 that for the core countries liquidity risk effect will increase after the sovereign debt crisis, the coefficient for liquidity (i.e. 𝛽𝛽1) should be significantly larger in the regression of periphery countries than that of core countries.

V. Empirical Results

The first set of regressions are based on specification (1) which includes both time and country fixed effects. I begin with a simple fixed effects OLS regression, and the results are reported in Table 3. The full specification (column 3 in Table 3) shows that rating, deficit to GDP, interest rate and inflation are statistically significant 5% level of significance, while liquidity and GDP growth are not. Yield spreads are negatively related to credit ratings as we expected. The positive coefficient of short-term interest rates suggests that lower interest rates are associated with lower government bond spreads, probably because funding liquidity as well as the incentives of investment managers to take risk are increased. The deficit to GDP also has an expected positive sign, indicating that the fiscal performance does have an impact on sovereign debt markets. A higher GDP growth relative to the benchmark government, on the other hand, reduces the credit spread, which could be explained as the government’s creditworthiness improves when its economy is in expansion. The R-squared increases as more control variables are added into the model, and takes the value of 0.779 in column (3), showing that nearly 78% of variation in the spread across time and country is explained by the model of specification (1).

However, this set of regressions won’t allow us to see how the different effects evolve over time. Therefore I run another regression using specification (2), with the three period dummies introduced in the previous section included (Table 4). Column 1 in Table 4 only includes the two risk factors 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 and 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡. Column 2

adds the fiscal variables 𝑌𝑌𝑖𝑖𝑡𝑡 . Column 3 further adds country-specific

macroeconomic fundamentals 𝑊𝑊𝑖𝑖𝑡𝑡.

Table 3. The Main Determinants of Sovereign Bond Yield Spreads (Baseline Estimation)

This table presents results of panel regressions examining the main determinants of sovereign bond yield spreads (specification (1)). The dependent variable is 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮, the yield spread

between a ten-year government bond issued in Eurozone country i and the benchmark country Germany at time t. Column 1 regresses the spread on the two risk variables, namely the credit risk (𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡, credit ratings from Oxford Economics, where 20 = AAA) and liquidity

factor (𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡, the ratio of the debt issued by a government to the total debt of Eurozone

countries). Column 2 adds the fiscal variables 𝑌𝑌𝑖𝑖𝑡𝑡 (Deficit to GDP ratio at the end of a fiscal

year between country i and Germany averaged on a quarterly basis; six-month interest rate of country i), based on column 1; column 3 further adds the macroeconomic variables 𝑊𝑊𝑖𝑖𝑡𝑡

(average quarterly real GDP growth and quarterly CPI, both are differential versus Germany). All variables are defined in the Appendix at the end of the main article. The sample period is 1990Q1 through 2014Q4. Standard errors clustered by countries are reported in parentheses. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively.

(1) (2) (3)

spread spread spread

𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 -0.7533*** -0.7450*** -0.8617*** (0.2096) (0.2060) (0.2015) 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 -5.8605 -8.9560 -0.0323 (25.7490) (26.2979) (16.0458) Deficit-to-GDP 0.0258 0.0405** (0.0161) (0.0132) Interest rate 0.2488* _0.1887* (0.1219) (0.0901) GDP growth -0.1085** (0.0386) inflation 0.6500*** (0.1292)

Time fixed effects Yes Yes Yes

Country fixed effects Yes Yes Yes

Constant 17.1303*** _12.7845** _17.1499***

(4.9813) (4.2127) (4.1020)

N 962 962 962

R2 _{0.621 0.635 0.779}

Table 4. The Main Determinants of EMU Sovereign Bond Yield Spreads in Different Periods

This table presents results of panel regressions examining the main determinants of sovereign bond yield spreads during different periods (specification (2)). The four time periods are as follows: Pre-Euro Period (Jan. 1990 - Dec. 1998), Convergence Period (Jan. 1999- June 2007), Crisis Period (July 2007- Sept. 2009) and Divergence Period (Oct. 2009- Dec. 2014). The dependent variable is 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮, the yield spread between a ten-year government bond issued in

Eurozone country i and the benchmark country Germany at time t. Column 1 regresses the spread on the two risk variables, namely the credit risk (𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡, credit ratings from Oxford

Economics, where 20 = AAA) and liquidity factor (𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡, the ratio of the debt issued by a

government to the total debt of Eurozone countries). Column 2 adds the fiscal variables 𝑌𝑌𝑖𝑖𝑡𝑡 (Deficit to GDP ratio at the end of a fiscal year between country i and Germany averaged on

a quarterly basis; six-month interest rate of country i), based on column 1; column 3 further adds the macroeconomic variables 𝑊𝑊𝑖𝑖𝑡𝑡 (average quarterly real GDP growth and quarterly CPI,

both are differential versus Germany). Standard errors clustered by countries are reported in parentheses. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively.

(1) (2) (3)

spread spread spread Pre-Euro Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 2.56 -5.29 -0.46 (3.97) (3.95) (3.39) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 -0.75*** -0.71*** -0.51*** (0.058) (0.058) (0.054) Deficit-to-GDP 0.18*** _0.19*** (0.029) (0.025) Interest rate 0.17*** _0.018 (0.047) (0.041) Inflation 0.64*** (0.037) intercept 15.5*** _14.4*** _10.6*** (1.08) (1.08) (0.99) Convergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷1 -1.98 -0.20 -1.43 (1.45) (1.45) (1.26) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡,× 𝐷𝐷1 1.08*** 0.97*** 0.37*** (0.070) (0.085) (0.082) Deficit-to-GDP 0.19*** _0.13*** (0.049) (0.043) Interest rate -0.19*** _-0.041 (0.064) (0.055) Inflation -0.56*** (0.083) 𝐷𝐷1 -21.5*** -18.7*** -7.56*** (1.35) (1.53) (1.50) 18

Crisis Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷2 -2.02 -0.44 -1.88 (2.23) (2.17) (1.86) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷2 1.02*** 0.96*** 0.30** (0.11) (0.13) (0.12) Deficit-to-GDP 0.15** _0.13** (0.060) (0.053) Interest rate -0.33*** _-0.13 (0.095) (0.085) Inflation -0.74*** (0.11) 𝐷𝐷2 -20.1*** -17.4*** -5.42** (2.15) (2.63) (2.38) Divergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷3 -6.37*** -4.50** -3.56** (1.82) (1.77) (1.51) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷3 0.22*** 0.16*** -0.19*** (0.048) (0.048) (0.048) Deficit-to-GDP 0.13*** _0.15*** (0.033) (0.029) Interest rate 0.30*** _0.39*** (0.094) (0.084) Inflation -0.10 (0.090) 𝐷𝐷3 -4.16*** -5.37*** 0.80 (0.86) (1.01) (0.95)

Additional controls No No Yes

Country fixed effects Yes Yes Yes

N 962 962 962

R2 _{0.593 0.632 0.734}

From Table 4 we can observe that the differential intercept coefficient of the convergence period (𝐷𝐷1) has a significant negative sign, which suggests that the

intercept for yield spreads from 1999Q1 to 2007Q2 is smaller than the pre-euro period. This result provides evidence for our first hypothesis that when controlling for credit risk and liquidity risk, the spreads tend to decrease after the introduction of the Euro. The differential intercept coefficient of the crisis period is also negative but relatively smaller than that of D1, possibly indicating that the spread level increases again.

Another noticeable result is that contrary to the findings of Beber et al. (2009), the magnitude of the differential coefficient for credit rating increases from the

pre-euro period to the convergence period, but drops in the crisis period, which could be a sign that credit quality explains more of the sovereign yield spreads when markets are more stable and that spreads are less driven by credit risk in times of stress. Another possible explanation is the so-called ‘convergence-trade’ model in which the market believes in the “credible EMU commitment to the bail-out of its member states” (Arghyrou and Kontonikas, 2012; Oliveira et al., 2012). This means that during that period, even if there are times when the credit situa-tions of some periphery countries are worsened, investors still believe that these countries would be bailed out due to the existence of the EMU although they are faced with the risk of bankruptcy. This view is further supported by the fact that the coefficient of the deficit-to-GDP ratio decreases after the introduction of Euro and its differential coefficient remains negative ever since, which is in line with Manganelli and Wolswijk’s (2009) findings that the years 2003-2005 have seen rising deficits but declining spreads. This probably suggests that investors bought the bonds of peripheral European governments in the hope that their yields would convergence with those of Germany and thus the borrowing costs for the governments are lowered even in the presence of deteriorating fundamentals.

Moreover, we can see that coefficients for the liquidity factor are only significant in the divergence period. This suggests that during most of the time periods, the credit risk has a major impact in determining sovereign yield spreads. However, the role of liquidity risks becomes more important when the market is quite unstable.

The coefficient of inflation is also interesting, which is significant in most periods except for the divergence period. In the pre-Euro period, it has an expected positive sign, in line with the findings of Oliveira et al. (2012). However, the sign becomes negative in the crisis period. This might indicate that investors perceive a higher consumer index as a sign of a better economic performance in the time of distress. This is contrast to the argument of Manasse et al. (2003) that a high inflation rate means that the economy is more likely to remain in distress.

Table 5. The Main Determinants of Sovereign Bond Yield Spreads in Different Periods for Core and Periphery Countries

This table presents results of panel regressions examining the main determinants of sovereign bond yield spreads for different groups of countries during the four periods. Countries of which the ten-year bond yield exceeded 7 percent after Q3, 2010 are grouped as periphery countries, the rest core countries. The dependent variable is 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮, the yield spread between

a ten-year government bond issued in Eurozone country i and the benchmark country Germany at time t. The sample period is 1990Q1 through 2014Q4. The explanatory variables are the two risk variables (𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡, credit ratings from Oxford Economics where 20 = AAA

and 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡, the ratio of the debt issued by a government to the total debt of Eurozone countries),

the fiscal variables 𝑌𝑌𝑖𝑖𝑡𝑡 (Deficit to GDP ratio at the end of a fiscal year between country i and

Germany averaged on a quarterly basis; six-month interest rate of country i), and the macro-economic variables 𝑊𝑊𝑖𝑖𝑡𝑡 (average quarterly real GDP growth and quarterly CPI, both are

differential versus Germany). Column 1 and column 2 report the results for core countries and periphery countries. Standard errors are reported in parentheses. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively.

(1) (2)

Core Countries Periphery Countries Pre-Euro Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 0.28 -726.1** (9.16) (109.2) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 -0.37 -1.54** (0.24) (0.20) Deficit-to-GDP 0.14* _0.14 (0.069) (0.19) intercept 7.94 38.7** (5.54) (4.79) Convergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷1 -3.03 484.8** (2.28) (97.2) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡,× 𝐷𝐷1 0.44 1.37** (0.24) (0.31) Deficit-to-GDP 0.042 -0.20 (0.042) (0.20) 𝐷𝐷1 -8.47 -30.6** (5.33) (6.55) Crisis Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷2 -2.24 443.1** (3.02) (45.1) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷2 0.31 0.48 (0.28) (0.48) Deficit-to-GDP 0.083 0.15 (0.096) (0.16) 𝐷𝐷2 -6.01 -10.4 (6.03) (9.83) Divergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷3 0.41 324.1* (1.92) (110.8) 21

𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷3 0.13 0.49 (0.24) (0.28) Deficit-to-GDP -0.0016 0.11 (0.097) (0.21) 𝐷𝐷3 -3.10 -16.9 (5.37) (6.21)

Additional controls 𝑊𝑊𝑖𝑖𝑡𝑡 Yes Yes

Country fixed effects Yes Yes

N 679 283

R2 _{0.576 0.864}

Nevertheless, since the liquidity factor here seems not to have significant effects in most periods, I divide the sample into core and periphery countries and run regression (2) again, in order to see the evolvement of the liquidity effect (Table 5).

Table 5 reports the major results of the third specification. Column 1 reports the results for core countries and column 2 for periphery countries. A noticeable fact is that the results for core countries are insignificant for all coefficients. For the periphery group, the results show that the differential coefficient for 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑖𝑖 is

actually positive in the convergence period. This could be the reason why we previously observed an unexpected decrease in the absolute value of the coefficient for credit ratings from the pre-euro period to the convergence period in Table 4, and further supports the explanation that investors’ perception of convergence which resulted in lower-than-expected yield spreads for periphery countries during that period is the reason of this phenomenon.

Since it is not possible to compare between the coefficients of the two groups due to the insignificant results, I am not able to test Hypothesis 2 at this stage. Hypothesis 3 cannot be tested directly as well, but we can still observe that the absolute value for the coefficient of 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 drops quite a lot for periphery countries

in the convergence period, but then increase slightly in the crisis period and becomes insignificant at the 5% level during the divergence period. This indicates that the effect of liquidity decreases after the introduction of Euro. A possible explanation could be increased integration of the Euro bond market (Bernoth,

2012). According to them, this result could be due to the fact that the introduction of Euro largely enlarged the size of Euro-denominated debt market for all countries by converting all existing government debt of Eurozone countries into a single currency.

In order to confirm the abovementioned results, I run another regression based on specification (2) but add a dummy for the periphery group. In this way, the whole sample is included in a single regression and the coefficients are now directly comparable. The equation writes as follows:

𝑟𝑟𝑡𝑡𝑖𝑖− 𝑟𝑟𝑡𝑡𝐺𝐺 = 𝛼𝛼 + 𝛽𝛽′𝑋𝑋𝑖𝑖𝑡𝑡 + 𝛾𝛾′𝑌𝑌𝑖𝑖𝑡𝑡+ 𝛿𝛿′𝑊𝑊𝑖𝑖𝑡𝑡+ � 𝜌𝜌1𝑘𝑘′𝑋𝑋𝑖𝑖𝑡𝑡𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜌𝜌2𝑘𝑘′𝑌𝑌𝑖𝑖𝑡𝑡𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜌𝜌3𝑘𝑘′𝑊𝑊𝑖𝑖𝑡𝑡𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜌𝜌4𝑘𝑘𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + 𝐷𝐷𝑝𝑝𝑝𝑝𝑝𝑝𝑖𝑖× (� 𝜃𝜃1𝑘𝑘′𝑋𝑋𝑖𝑖𝑡𝑡𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜃𝜃2𝑘𝑘′𝑌𝑌𝑖𝑖𝑡𝑡𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜃𝜃3𝑘𝑘𝑊𝑊𝑖𝑖𝑡𝑡𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 + � 𝜃𝜃4𝑘𝑘𝐷𝐷𝑘𝑘 3 𝑘𝑘=1 ) + 𝜇𝜇𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑡𝑡 (3)

where 𝑋𝑋𝑖𝑖𝑡𝑡 is a vector consist the two risk factors 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 and 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 , 𝑌𝑌𝑖𝑖𝑡𝑡 are the fiscal variables, 𝑊𝑊𝑖𝑖𝑡𝑡 are the macroeconomic variables. 𝐷𝐷𝑘𝑘 represents the four different sub-periods. 𝐷𝐷𝑝𝑝𝑝𝑝𝑝𝑝𝑖𝑖 is a dummy that equals 1 for periphery countries and 0 for core countries. The results of specification (3) are presented in Table 6.

From Table 6 we can observe that the results are quite similar to those reported in Table 5, with minor differences in the magnitude of coefficients. Results for core countries are still insignificant. Hence, it is still no possible to test Hypotheses 2 and 3 at this stage. I explore the possible reasons for those insignificant results in the next section, where a number of robustness checks are conducted.

VI. Robustness Check

Since the panel data used in this paper is likely to be autocorrelated and heteroskedastic, for the robustness check I run the three set of regressions in the previous sector again using Newey-West standard errors which allows for an error

Table 6. The Main Determinants of Sovereign Bond Yield Spreads in Different Periods for Core and Periphery Countries (Single Regression)

This table presents results of panel regressions examining the main determinants of sovereign bond yield spreads for different groups of countries during the four periods (specification (3)). Countries of which the ten-year bond yield exceeded 7 percent after Q3, 2010 are grouped as periphery countries, the rest core countries. The dependent variable is 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮, the yield spread between a ten-year government bond issued in Eurozone country i and the benchmark

country Germany at time t. The sample period is 1990Q1 through 2014Q4. The explanatory variables are the two risk variables (𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡, credit ratings from

Oxford Economics where 20 = AAA and 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡, the ratio of the debt issued by a government to the total debt of Eurozone countries), the fiscal variables

𝑌𝑌𝑖𝑖𝑡𝑡 (Deficit to GDP ratio at the end of a fiscal year between country i and Germany averaged on a quarterly basis; six-month interest rate of country i), and the

macro-economic variables 𝑊𝑊𝑖𝑖𝑡𝑡 (average quarterly real GDP growth and quarterly CPI, both are differential versus Germany). Standard errors are reported in

parentheses. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively. Core countries

Pre-Euro Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 Deficit-to-GDP Intercept

0.28 -0.37 0.14* _17.0***

(9.01) (0.24) (0.068) (4.03)

Convergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷1 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷1 Deficit-to-GDP 𝐷𝐷1

-3.03 0.44* _{0.042 -8.47}

(2.25) (0.24) (0.041) (5.25)

Crisis Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷2 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷2 Deficit-to-GDP 𝐷𝐷2

-2.24 0.31 0.083 -6.01

(2.97) (0.28) (0.094) (5.94)

0.41 0.13 -0.0016 -3.10

(1.89) (0.24) (0.095) (5.29)

Periphery countries

Pre-Euro Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 Deficit-to-GDP 𝐿𝐿𝐼𝐼𝑖𝑖𝑝𝑝𝑟𝑟𝐼𝐼𝑝𝑝𝑝𝑝𝑖𝑖× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖

-726.3*** _-1.18*** _{-0.0047 (omitted)}

(92.4) (0.29) (0.18)

Convergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷1× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷1× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 Deficit-to-GDP 𝐷𝐷1× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖

487.8*** _0.93** _{-0.25 -22.1}**

(81.9) (0.35) (0.17) (7.61)

Crisis Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷2× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷2× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 Deficit-to-GDP 𝐷𝐷2× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖

445.3*** _{0.17 0.071 -4.38}

(38.1) (0.49) (0.16) (10.2)

Divergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷3× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷3× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖 Deficit-to-GDP 𝐷𝐷3× 𝐷𝐷𝑝𝑝𝑝𝑝𝑟𝑟𝑖𝑖

323.7*** _{0.35 0.12 -13.8}*

(93.4) (0.33) (0.20) (7.44)

Additional controls 𝑊𝑊𝑖𝑖𝑡𝑡 Country fixed effects N R2

Yes Yes 962 0.860

Table 7. The Main Determinants of Sovereign Bond Yield Spreads in Different Periods for Core and Periphery Countries

(with Newey-West Standard Errors)

This table presents results of panel regressions examining the main determinants of sovereign bond yield spreads for different groups of countries during the four periods. Countries of which the ten-year bond yield exceeded 7 percent after Q3, 2010 are grouped as periphery countries, the rest core countries. The dependent variable is 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮, the yield spread between

a ten-year government bond issued in Eurozone country i and the benchmark country Germany at time t. The sample period is 1990Q1 through 2014Q4. The explanatory variables are the two risk variables (𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡, credit ratings from Oxford Economics where 20 = AAA

and 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡, the ratio of the debt issued by a government to the total debt of Eurozone countries),

the fiscal variables 𝑌𝑌𝑖𝑖𝑡𝑡 (Deficit to GDP ratio at the end of a fiscal year between country i and

Germany averaged on a quarterly basis; six-month interest rate of country i), and the macro-economic variables 𝑊𝑊𝑖𝑖𝑡𝑡 (average quarterly real GDP growth and quarterly CPI, both are

differential versus Germany). Column 1 and column 2 report the results for core countries and periphery countries. The estimations correct the error structure for heteroskedasticity and autocorrelation by using Newey-West standard errors. Standard errors are reported in paren-theses. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively.

(1) (2)

Core Countries Periphery Countries Pre-Euro Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 0.28 -726.1*** (4.11) (87.9) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 -0.37** -1.54*** (0.17) (0.25) Deficit-to-GDP -0.14*** _-0.14** (0.034) (0.065) Convergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷1 -3.03*** 484.8*** (1.06) (83.9) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡,× 𝐷𝐷1 0.44*** 1.37*** (0.12) (0.24) Deficit-to-GDP 0.042 -0.20*** (0.042) (0.068) 𝐷𝐷1 -8.47*** -30.6*** (2.26) (5.28) Crisis Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷2 -2.24* 443.1*** (1.21) (106.9) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷2 0.31** 0.48 (0.15) (0.44) Deficit-to-GDP 0.083** _0.15 (0.041) (0.11) 𝐷𝐷2 -6.01** -10.4 (2.79) (9.36)

Divergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷3 0.41 324.1*** (1.27) (107.2) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷3 0.13 0.49* (0.15) (0.29) Deficit-to-GDP -0.0016 0.11 (0.050) (0.072) 𝐷𝐷3 -3.10 -16.9** (2.92) (6.75)

Additional controls 𝑊𝑊𝑖𝑖𝑡𝑡 Yes Yes

Country fixed effects Yes Yes

N 679 283

R2 _{0.576 0.864}

structure with heteroskedasticity and possible serial correlation up to some lag. The results of specification (1) and (2) are reported in Table A2 and Table A3 respectively, in the Appendix at the end of this paper. Those two tables show that the regression results are quite robust, since the significance level for most independent variables remain the same.

The results of specification (2) with core and periphery groups are reported in Table 7 below. In Table 7 we can see that many of those insignificant results for core countries in Table 5 in the previous sector become significant here. For instance, the differential coefficient for credit risks in core countries drops from the convergence period to the crisis period, which is contrast to the finding of Beber et al. (2009) that sovereign yield spreads is more driven by differences in credit quality especially for low credit risk countries and during times of heightened market uncertainty, therefore rejecting my second hypothesis. In addition, it is still quite obvious that the liquidity factor plays a more important role for periphery countries, and it becomes insignificant in the crisis and divergence period for core countries. This observation rejects my hypothesis 3 that the effect of the liquidity risk factor will increase especially for the core countries after late 2009.

Same with the previous section, I then run specification (3) with Newey-West standard errors. The results are reported in Table A4 in the Appendix and show that they are quite robust for periphery countries but not for core countries.

Table 8. The Main Determinants of Sovereign Bond Yield Spreads in Different Periods for Core and Periphery Countries

(Bid-ask Spreads)

This table presents results of panel regressions examining the main determinants of sovereign bond yield spreads for different groups of countries during the four periods, but uses the minus bid-ask spread as the liquidity measure, and only covers the period of 1999Q1-2014Q4. Countries of which the ten-year bond yield exceeded 7 percent in Q3, 2010 are grouped as periphery countries, the rest core countries. The dependent variable is 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮, the yield spread

between a ten-year government bond issued in Eurozone country i and the benchmark country Germany at time t. The explanatory variables are the two risk variables (𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡, credit

ratings from Oxford Economics where 20 = AAA and 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡, minus bid-ask spread), the fiscal

variables 𝑌𝑌𝑖𝑖𝑡𝑡 (Deficit to GDP ratio at the end of a fiscal year between country i and Germany

averaged on a quarterly basis; six-month interest rate of country i), and the macro-economic variables 𝑊𝑊𝑖𝑖𝑡𝑡 (average quarterly real GDP growth and quarterly CPI, both are differential

versus Germany). The divergence period is used as the base period. Column 1 and column 2 report the results for core countries and periphery countries. Standard errors clustered by countries are reported in parentheses. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively.

(1) (2)

Core Countries Periphery Countries Convergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷1 9.88 -6.90* (-BAS) (5.81) (2.13) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡,× 𝐷𝐷1 0.31*** 0.65** (0.059) (0.11) Deficit-to-GDP -0.084** _0.055 (0.024) (0.045) 𝐷𝐷1 -5.77*** -3.36 (1.23) (3.13) Crisis Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡× 𝐷𝐷2 4.22 0.39 (-BAS) (4.91) (6.92) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡× 𝐷𝐷2 0.24*** 0.51*** (0.032) (0.016) Deficit-to-GDP -0.071** _-0.020 (0.027) (0.039) 𝐷𝐷2 0.76 3.67 (0.46) (1.96) Divergence Period 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 -12.2** -0.53 (-BAS) (4.52) (0.76) 𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 -0.27*** -0.98* (0.034) (0.29) Deficit-to-GDP 0.088** _0.045 (0.031) (0.016) Intercept 3.67** _5.41 (1.01) (2.33) 28

Additional Controls Yes Yes

Country fixed effects Yes Yes

N 443 188

R2 _{0.852 0.838}

Next, I repeat the last set of regressions but replace the liquidity measure with the bid-ask spread. Compared to liquidity proxy (the ratio of the debt issued by a government to the total debt of EU countries) used in previous sectors, this measurement is more direct and reliable, for it timely reflects market reactions. However, since the dataset for the bid-ask spread is only available from 1999Q1 to 2014Q4, the regression only includes the convergence period, the crisis period and the divergence period. In addition, because bid-ask spreads actually represent the illiquidity, I take the negative value of the BAS to make the results easier to compare with the previous tables. In these regressions, the divergence period is used as the base period. The convergence period and the crisis period are denoted as 𝐷𝐷1 and 𝐷𝐷2 respectively. The results are reported in Table 8.

When we look at the coefficients of the new liquidity factor in Table 8, we can see that the coefficient for 𝐷𝐷1 is negative and significant in the convergence period for core countries, but becomes positive in the divergence period. This is in line with our conclusion in the previous sector. In addition, the credit ratings still have a positive sign in the convergence period, and the magnitude is quite large for periphery countries, which is similar to the results in Table 4. The fact that for core countries, the credit effect is smaller in the crisis period than in the convergence period also supports my conclusion from the previous Newey-West estimations for specification (2). However, we can observe that the liquidity factor loses its significance, indicating that my results might not be very robust regarding the measure of liquidity risk.

VII. Conclusion

This paper explores the evolving determinants of sovereign yield spreads for 10 Eurozone countries from 1990Q1 to 2014Q4 (quarterly frequency). In order to

analyze the development of yield spreads under different economic situations, the abovementioned time period is further divided into four sub-periods, namely the pre-Euro period, the convergence period, the crisis period, and the divergence period. The countries are categorized into core and periphery groups based on their yield performance in the Eurozone sovereign crisis. I regress the yield spreads against credit risk and liquidity risk, and then add the fiscal variables and country-specific macroeconomic fundamentals to the model. Fixed-effects panel regressions with period and group dummies are applied in order to obtain more specific results for the evolving roles of different factors.

The main findings of the paper can be summarized as follows: (i) yield spreads decreased substantially after the introduction of the Euro but increased again after the burst of global financial crisis after controlling for credit and liquidity risks; (ii) credit risk is the main driver of yield spreads especially in stable markets; but during the convergence period, it is not a good predictor for periphery countries, probably due to the bail-out commitment of ECB; (iii) the liquidity effect is more significant for periphery countries before the introduction of Euro and in the sovereign crisis period.

One of the limitations of my paper lies in the fact that due to the limited time-span of the accessible data, I didn’t use the conventional liquidity factor (the bid-ask spread) in the main regressions, but instead followed the method employed by Bernoth et al. (2012) to use the ratio of the debt issued by a government in to the total debt of EU countries as a proxy of the liquidity premium. This approach assumes that all debt issued by a government in a given currency is homogeneous up to maturity, which is probably not the case with Eurozone countries. As a result, my conclusions on the liquidity effect of yield spreads may not be precise. In fact, the results in the section of robustness check suggest that the liquidity factor indeed affects the yield spread in most periods.

In terms of policy implications, my conclusion that credit rating plays a more important role than liquidity risk imply that governments should attach more

importance to their debt solvency as well as the sustainability of public finances. Moreover, since the liquidity effect is more apparent in the period of crisis especially for periphery countries, measures that could provide more liquidity to the market as well as prevent severe liquidity squeeze are therefore envisaged.

References

Arghyrou, M. G., and Kontonikas, A. (2012). The EMU sovereign-debt crisis: Fundamentals, expectations and contagion. Journal of International Financial Markets, Institutions and

Money, 22(4), 658-677.

Beber, A., Brandt, M. W., & Kavajecz, K. A. (2009). Flight-to-quality or flight-to-liquidity? Evidence from the euro-area bond market. Review of Financial Studies, 22(3), 925-957.

Bernoth, K., Von Hagen, J., and Schuknecht, L. (2012). Sovereign risk premiums in the European government bond market. Journal of International Money and Finance, 31(5), 975-995.

Cantor, R., and Packer, F. (1996). Determinants and impact of sovereign credit ratings.

Economic Policy Review, 2(2).

Codogno, L., Favero, C., and Missale, A. (2003). Yield spreads on EMU government bonds. Economic Policy, 18(37), 503-532.

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.

Favero, C., Pagano, M., and Von Thadden, E. L. (2010). How does liquidity affect government bond yields?. Journal of Financial and Quantitative Analysis, 45(01), 107-134.

Geyer, A., Kossmeier, S., and Pichler, S. (2004). Measuring systematic risk in EMU government yield spreads. Review of Finance, 8(2), 171-197.

Goyenko, R. Y., and Ukhov, A. D. (2009). Stock and bond market liquidity: A long-run empirical analysis. Journal of Financial and Quantitative Analysis, 44(01), 189-212.

Gravelle, T. (1999). “Liquidity of the Government of Canada Securities Market: Stylized Facts and Some Market Microstructure Comparisons to the United States Treasury Market”, Bank of

Canada Working Paper. 99-11.

Lane, P. R. (2012). The European sovereign debt crisis. The Journal of Economic

Perspectives, 26(3), 49-67.

Longstaff, F. A. (2009). Portfolio claustrophobia: Asset pricing in markets with illiquid assets. The American Economic Review, 1119-1144.

Manasse, P., Roubini, N., Schimmelpfennig, A. (2003). Predicting sovereign debt crises. IMF

Working Paper, No. 03/221.

Manganelli, S., and Wolswijk, G. (2009). What drives spreads in the euro area government bond market?. Economic Policy, 24(58), 191-240.

Monfort, A., and Renne, J. P. (2014). Decomposing euro-area sovereign spreads: credit and liquidity risks. Review of Finance, 18(6), 2103-2151.

Oliveira, L., Curto, J. D., and Nunes, J. P. (2012). The determinants of sovereign credit spread changes in the Euro-zone. Journal of International Financial Markets, Institutions and

Money, 22(2), 278-304.

Sgherri, S., and Zoli, E. (2009). Euro area sovereign risk during the crisis. IMF Working Papers, 1-23.

Appendices

Table A1. Descriptions of Variables

This table lists all the variables used in the models and their sources and descriptions.

Variable Name Definition Unit of

Mea-surement Dependent variable

Spread 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮 The yield spread between a ten-year government

bond issued in Eurozone country i and the benchmark country Germany.

Source: DataStream, IMF

Percent

Risk variables

Credit risk 𝑪𝑪𝑪𝑪𝑪𝑪𝒊𝒊𝒕𝒕 Represented by the credit ratings from Oxford

Economics. 20 = AAA

Source: Datastream

Index

Liquidity 𝑳𝑳𝑳𝑳𝑳𝑳𝒊𝒊𝒕𝒕 The ratio of the debt issued by a government to the

total debt of Eurozone countries.

Source: Datastream

Percent

Fiscal variables 𝒀𝒀𝒊𝒊𝒕𝒕

Deficit Difference of deficit to GDP at the end of the fiscal year

between the issuer country and the benchmark country averaged on a quarterly basis. Deficit is calculated as the negative of government balance (Maastricht Definition).

Source: Datastream

Percent

Interest rate Short-term (6-month) interest rate of country i

Source: WES Database

Percent

Macroeconomic Variables 𝑾𝑾𝒊𝒊𝒕𝒕

GDP growth Average quarterly real GDP growth (differential

versus Germany).

Source: Datastream, OECD Statistics

Percent

Inflation The level of consumer prices represented by quarterly

CPI (differential versus Germany).

Source: World Bank

Percent

Dummy Variables

𝑫𝑫𝟏𝟏 Equals 1 if the data belonging to the period of 1999Q1 –2007Q2, otherwise 0

𝑫𝑫𝟐𝟐 Equals 1 if the data belonging to the period of 2007Q3 – 2009Q3, otherwise 0

𝑫𝑫𝟑𝟑 Equals 1 if the data belonging to the period of 2009Q4 – 2014Q4, otherwise 0

𝑫𝑫𝒑𝒑𝒑𝒑𝒓𝒓𝒊𝒊 Equals 1 for periphery countries, otherwise 0

Table A2. The Main Determinants of Sovereign Bond Yield Spreads (with Newey-West Standard Errors)

This table presents results of panel regressions examining the main determinants of sovereign bond yield spreads (specification (1)). The dependent variable is 𝒓𝒓𝒕𝒕𝒊𝒊− 𝒓𝒓𝒕𝒕𝑮𝑮, the yield spread

between a ten-year government bond issued in Eurozone country i and the benchmark country Germany at time t. Column 1 regresses the spread on the two risk variables, namely the credit risk (𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡, credit ratings from Oxford Economics, where 20 = AAA) and liquidity

factor (𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡, the ratio of the debt issued by a government to the total debt of Eurozone

countries). Column 2 adds the fiscal variables 𝑌𝑌𝑖𝑖𝑡𝑡 (Deficit to GDP ratio at the end of a fiscal

year between country i and Germany averaged on a quarterly basis; six-month interest rate of country i), based on column 1; column 3 further adds the macroeconomic variables 𝑊𝑊𝑖𝑖𝑡𝑡

(average quarterly real GDP growth and quarterly CPI, both are differential versus Germany). All variables are defined in the Appendix at the end of the main article. The sample period is 1990Q1 through 2014Q4. The estimations correct the error structure for heteroskedasticity and autocorrelation by using Newey-West standard errors. Standard errors clustered by countries are reported in parentheses. ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively.

(1) (2) (3)

spread spread spread

𝐶𝐶𝐶𝐶𝐶𝐶𝑖𝑖𝑡𝑡 -0.75*** -0.74*** -0.86*** (0.057) (0.055) (0.054) 𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑡𝑡 -5.86 -8.96* -0.032 (4.52) (4.68) (3.31) Deficit-to-GDP -0.026** _-0.041** (0.013) (0.018) Interest rate 0.25*** _0.19*** (0.058) (0.043) GDP growth -0.11** (0.054) inflation 0.65*** (0.047)

Time fixed effects Yes Yes Yes

Country fixed effects Yes Yes Yes

N 962 962 962

R2 _{0.621 0.635 0.779}