THE EFFECT OF EMU ACCESSION ON BOND YIELD
SPREAD CONVERGENCE FOR NEW EU ECONOMIES
University of Amsterdam
Amsterdam School of Economics
Faculty of Economics and Business
Name: Kasparas Subacius
Student number: 11700815
Contact: kasparas.subacius@gmail.com
Number of words: 11733
Course: MSc Economics
Track: International Economics and Globalization
Supervisor: Dr. Cenkhan Sahin
Second reader: Dr. Dirk Veestraeten
Statement of Originality
This document is written by Student Kasparas Subačius who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract. The main goal of this research paper is to assess whether adoption of the euro in new
EU countries (Cyprus, Latvia, Lithuania, Malta, Slovakia and Slovenia) has led to convergence in long-term government bond yields vis-a-vis Germany. Theory suggests that, in addition to diminished currency risk, the introduction of a common currency could lead to increased bond market integration through another channel - local and global volatility effects lose their relevance while an economy becomes more dependent on regional effects. An empirical analysis is conducted by employing a clustered fixed effects model. This paper finds that, for the sample of new EU economies, among other important determinants, accession to the eurozone leads to a decline in long-term interest rate spreads. Therefore, results suggest that non-EMU countries could potentially gain access to cheaper borrowing by adopting the euro even if they have higher associated risks related to them and macroeconomic fundamentals are weaker.
Keywords: Eurozone, convergence, long-term government bond yields, bond market
Table of Contents
1. Introduction ... 5
2. Overview of relevant literature ... 8
2.1. The role of default risk. ... 8
2.2. Is liquidity risk relevant? ... 9
2.3. The importance of currency risk. ... 9
2.4. Did the creation of EMU lead to deeper bond market integration? ... 10
3. Empirical research ... 14
3.1. Data sample, variables and methodology approach. ... 14
3.1.1 Data sample. ... 14
3.1.2. Variable specification. ... 16
3.1.3. Methodology approach. ... 20
3.2. Panel data analysis. ... 21
3.2.1. Panel diagnostics. ... 21 3.2.2. Heteroskedasticity. ... 21 3.2.3. Autocorrelation. ... 21 3.2.4. Normality of residuals. ... 22 3.3. Results. ... 22 3.4. Robustness checks. ... 25 3.4.1. Within-model transformations. ... 25 3.4.2. Comparison of models. ... 28
4. Policy implications and limitations ... 30
5. Conclusions ... 32
Bibliography ... 33
1. Introduction
The long-term interest rate is a measure which reflects financial markets’ perception towards a country’s economic sustainability and creditworthiness in the long run. Historically, it has been observed that before the official creation of the eurozone in 1999, long-term bond yields of 11 candidate countries have started to converge towards Germany and several other economies that were perceived to be most stable and safe. After common currency was introduced and monetary policies have been synchronized, almost complete convergence of long-term interest rates among these 11 economies has been recorded (see figure 1 below). This essentially implies that all member economies, which inherently are highly heterogeneous, were assumed to carry similar levels of default, currency and liquidity risks, show sustainable performance in economic fundamentals as well as were perceived to be equally stable and creditworthy (Alexopoulou, Bunda & Ferrando, 2009). So, this thesis explores the relationship between accession to the European Economic and Monetary Union (EMU) and the convergence in government bond yields. To be more precise, I empirically assess whether adoption of the euro could lead to increased bond market integration.
It is important to study the phenomena of synchronization of long-term interest rates between EMU member states because economies are not as homogeneous as it might seem. Therefore, during the boom of an economic cycle, riskier countries benefit when the perceived level of risks is poorly reflected by the market as they gain access to cheap international borrowing of capital which can then be used for investment purposes. However, one exogenous shock might reverse the entire integration process as markets reassess default, currency and liquidity risks - this might lead to a sudden jump in long-term interest rates and potentially could make economy’s debt servicing too costly (Bicu & Candelon, 2012; Gupta, Sehgal & Deisting, 2015, Smets, 2013). This scenario of divergence in long-term interest rates has been observed during the European Debt Crisis which started in the end of 2009 (see figure 1 below). As one can see, ever since the creation of EMU in 1999 founding members saw almost complete convergence in long-term interest rates up until the beginning of recession when bond yields severely diverged and such situation persisted for multiple years.
Figure 1. Long-term government bond yield dynamics across original EMU countries
Source: Eurostat
There is an abundant amount of literature that has tried to determine the main factors which influence government bond yields and lead to increased bond market integration. One of the distinguished potential drivers is the accession to EMU that has two main elements to it. First, even the anticipation of monetary union before adoption of common currency might lead to increased levels of bond market integration (European Central Bank, 2003). Secondly, credibility of the European Central Bank (ECB) with respect to its objective of keeping stable inflation after joining the monetary union is also seen as a catalyst which helps financial markets to grow deeper and leads to convergence in long-term government bond yields (Ehrmann, Fratzscher, Gurkaynak & Swanson, 2007; Lane, 2008; Swanson, 2008). Therefore, the main aim of this thesis is to test whether accession to the EMU has any statistically significant effect on integration of the government bond markets. In this research paper, I empirically test if adoption of the euro influences synchronization in government bond yields by assessing provided regression equation for panel data set.
When assessing effects of creation of the EMU on bond markets, most of the research papers are focusing on EU-15 countries and have used either GARCH-based methods, Capital Asset Pricing Model (CAPM) or beta convergence tests (Abad, Chulia & Gomez-Puig, 2009; Adam et al., 2002; Cappiello, Hordahl, Kadareja & Manganelli, 2006; Christiansen, 2007; Ehrmann, Fratzscher, Gurkaynak & Swanson, 2007; Gupta, Sehgal and Deisting, 2015; Pozzi & Wolswijk, 2008). However, much less emphasis has been put on new EU countries and
0 2 4 6 8 10 12 14 19 95M0 1 19 96M0 1 19 97M0 1 19 98M0 1 19 99M0 1 20 00M0 1 20 01M0 1 20 02M0 1 20 03M0 1 20 04M0 1 20 05M0 1 20 06M0 1 20 07M0 1 20 08M0 1 20 09M0 1 20 10M0 1 20 11 M0 1 20 12M0 1 20 13M0 1 20 14M0 1 20 15M0 1 20 16M0 1 20 17M0 1 20 18M0 1 Lon g-te rm gover n m e n t b o n d y ie ld (% ) Belgium Germany Ireland Spain France Italy Luxembourg Netherlands Austria Portugal Finland
literature on them is scarce because most of the papers investigating the effects of EMU on bond markets have been evaluating the impact of creation of EMU but not adoption of the euro. It would be much more challenging to investigate such sample as in this paper as new EU countries all became part of the EMU at different times. In addition, many influential papers have been written even before or during the Great Recession, so there was still a very limited amount of observations for these countries. Since now there is more data available, it is interesting to investigate the relationship between adoption of the euro and bond yield convergence for countries which joined the EU in 2004 and adopted the euro since then - Cyprus, Latvia, Lithuania, Malta, Slovenia and Slovakia. The data set spans from Q1 of 2002 until Q2 of 2017.
This thesis does not use GARCH method, CAPM or beta convergence test that are more isolated but instead follows the logic of Ichiue and Shimizu (2012), Min (1998) and Orlowski and Lommatzsch (2005) where various potential determinants of long-term interest rates are evaluated with a panel OLS model. In this case, a dummy variable for adoption of the euro is added among other important determinants of long-term interest rates. This allows to observe whether the adoption of the euro can explain part of the variance in bond yield spreads when compared to other major variables. Fixed effects panel OLS (clustered at a country level) model is estimated and results suggest that, indeed, among other significant control variables, adoption of the euro in new EU countries appears to lead to a higher level of long-term government bond yield convergence.
Thesis is structured in the following way. The first part focuses on the literature review and potential determinants of convergence in long-term government bond yields. The second part describes the variables and data set as well as introduces the empirical research methodology, results and robustness checks. Lastly, policy implications, limitations and conclusions are provided.
2. Overview of relevant literature
Several structural strands of drivers that affect government bond yields in Europe have been identified and investigated on top of macroeconomic variables that reflect the general economic state of a country. Some of the main determinants of long-term interest rates can be split into several broader categories and they are: fundamental risk factors (Pagano & Von Thadden, 2004) and harmonization of national macroeconomic and fiscal policies (Cote & Graham, 2004; Min, 1998; Orlowski & Lommatzsch, 2005). Both risk factors and harmonization of fiscal policies are all closely inter-related since it is difficult to clearly distinguish each type of risk by a separate variable as these risks are not directly observable (Cote & Graham, 2004). Regarding risk factors, several different types of risks have been identified which have been discussed, namely default, liquidity and currency risks.
2.1. The role of default risk.
There is an abundant amount of literature which stresses the importance of default risk, and claims that it reflects the probability of sovereign state defaulting on its foreign debt. Multiple scholars have indeed determined that markets tend to overreact and penalize countries that have loose fiscal positions (Afonso, Arghyrou & Kontonikas, 2015; Costantini, Fragetta & Melina, 2014; Poghosyan, 2012) and concluded that high debt and deficit values, which show increased likelihood of default, are statistically significant. Default risk is distinguished as one of the major drivers that prevents bond yields across the EMU from full convergence and because of it, spreads of long-term interest rates cannot be fully exterminated, therefore, bonds could only be perceived to be close substitutes (Costantini, Fragetta & Melina, 2014; Orlowski & Lommatzsch, 2005; Pagano & Von Thadden, 2004). By using asset pricing model, Bernoth, Von Hagen and Schuknecht (2004) also find that yield spreads of EU countries vis-à-vis Germany show positive default, as well as liquidity, risk premium. What is also interesting, Costantini, Fragetta and Melina (2014) have split their sample countries in two groups – core EMU (countries that have passed the OCA test - inflation differential is not found to be significant) and periphery EMU countries. Empirical findings suggest that for core EMU countries higher debt-to-GDP differential is not the main driver of bond yield spread anymore. This essentially means that, compared to core EMU, peripheral countries are penalized more harshly with skyrocketing borrowing costs when the debt increases. It once again proves the importance that default risk has on long-term debt, especially in less developed economies.
2.2. Is liquidity risk relevant?
In addition, liquidity risk is thought to be linked to long-term government bond yields. Essentially, this risk captures the likelihood of early liquidation and losses of capital due to a reduced number of transactions within the market. Literature investigating the role of liquidity risk on long-term interest rates is rather ambiguous. First of all, it is very hard to capture only this specific type of risk as it could possibly be interacting with default risk (Afonso, Arghyrou & Kontonikas, 2015; Pagano & Von Thadden, 2004). Also, it is complicated to measure liquidity risk alongside the default and currency risks at the same time due to the fact that, for example, default risk represents fiscal position of an economy over a rather longer period of time while liquidity risk is usually expressed as a bid-ask spread and can only be captured by assessing high-frequency data (Pagano & Von Thadden, 2004). Research on the role of liquidity risk on bond yields is not unanimous - several scholars claim that liquidity risk is mainly fueled by a segmented bond market across the eurozone (Danthine, Giavazzi & Von Thadden, 2000; Gomez-Puig, 2006). Costantini, Fragetta and Melina (2004) are in-between mentioning that liquidity risk is perceived to be of smaller importance; nevertheless, it still seems to have an effect in the long-term. However, there is also an opinion that liquidity risk can only explain a small fraction of sovereign spreads and plays a minor role (Codogno, Favero, Missale, Portes & Thum, 2003; Pagano & Von Thadden, 2004; Swanson, 2008). Therefore, since literature on liquidity risk is quite ambiguous, and it is extremely difficult to capture both default and liquidity risks in one data set, it is not reflected in the model of this thesis.
2.3. The importance of currency risk.
A segmented currency market of the EU also brings risks and costs associated with transactions of foreign currencies. When external investors have foreign currency that they are willing to invest locally, higher returns are to be expected due to several reasons. First, as Jappelli and Pagano (2008) stated, throughout the period of investment, exchange rate fluctuations are nearly inevitable if the currency is not pegged, therefore, investors will require a positive premium for holding an asset in local currency. Moreover, transaction costs are to be involved in the conversion of a currency, so, that also brings additional costs for which investors want to be compensated. If a country was to join the common currency area, exchange rate fluctuations and transaction costs – some of the major obstacles to financial integration - are expected to be eliminated completely.
As can be expected, academic literature supports the view that currency risk premium has been significantly diminished due to the creation of a monetary union (Baele, Ferrando, Hordahl, Krylova & Monnet; 2004; Ehrmann, Fratzscher, Gurkaynak & Swanson, 2007; Galati & Tsatsaronis, 2003; Jappelli & Pagano, 2008). In addition, Cote and Graham (2004) find evidence that currency risk has been constantly declining even before the creation of the eurozone - since ratification of the Maastricht Treaty up until creation of EMU in 1999 - when it has practically vanished. Interestingly, Gabrisch and Orlowski (2009) mention that even non-EMU can partly curb the exchange rate risk if they are currency board countries and have their exchange rates pegged to the euro. Baltic states – Latvia and Lithuania have been distinguished as examples that had such policy regime.
Even though it might seem that the currency risk should be eliminated completely once a country adopts the euro and there is no way for it to resurge again, it is not necessarily the case. Arghyrou and Tsoukalas (2011) describe the situation of Greece during the Global Financial Crisis and mention that markets began to be worried about increased likelihood of Greece’s default on debts and voluntary opt-out of the eurozone. So, increased probability of a country’s exit from the eurozone made markets remeasure the currency risk once again, despite the fact, that the country was still using the euro as its currency. Therefore, as also mentioned by Pagano and Von Thadden (2004) elimination of exchange rate fluctuations and transactions costs is not enough to fully extinguish the currency risk as anticipation of exit from the eurozone or even a smallest probability of the collapse of EMU itself could still spark the divergence in long-term interest rates.
2.4. Did the creation of EMU lead to deeper bond market integration?
Even though it has been showed graphically in the introduction that prior to the creation of EMU there was an obvious convergence in long-term interest rates which persisted up until the beginning of the Global Financial Crisis, it is still important to analyze if there is supporting empirical evidence of a statistically significant relationship between adoption of a common currency and further bond market integration. Unfortunately, research on new EMU member countries is very scarce. There are only papers which investigated the effect of creation of EMU in 1999 on convergence in long-term government bond yields for founding member states that became part of the eurozone all at once.
Several studies have applied the method of measuring conditional correlations between long-term interest rates in different countries. They try to observe changes in correlations over two periods – before and after adoption of the euro – and assess the impact of the euro on specific variable across a time dimension as well as investigate the time-varying volatility in bond yields (Cappiello, Hordahl, Kadareja & Manganelli, 2006; Christiansen, 2007; Ehrmann, Fratzscher, Gurkaynak & Swanson, 2007; Fratzscher, 2001; Gabrisch & Orlowski, 2009; Kim, Lucey & Wu, 2005; Skintzi & Refenes, 2006).
Cappiello, Hordahl, Kadareja and Manganelli (2006) employed a time-varying GARCH-type correlation and a regression quantile-based codependence measure. GARCH method helps to observe evolution of the correlation in the short run - for the sample of 11 founding members of EMU they find that correlation within the 10-year government bond market has almost reached a value of 1, meaning that after the introduction of the euro, bond markets across EMU became much more integrated. Nevertheless, regression quantile-based codependence measure showed that, even though the bond markets became much more integrated, the dynamics has not changed significantly, and previous major determinants of long-term interest rate spread continue to play the primary role (Cappiello, Hordahl, Kadareja & Manganelli, 2006). Similar conclusions have been drawn by Kim, Lucey and Wu (2005) and Skintzi and Referenes (2006) where also by employing GARCH model, they find lower conditional correlations between EMU and non-EMU bond markets on a regional level, suggesting that economies of the eurozone are more integrated.
In addition, Ehrmann, Fratzscher, Gurkaynak and Swanson (2007) have also measured conditional correlations to identify whether monetary union within the EU has led to increased integration of financial markets. Even though, prior to the creation of EMU, economies of France, Germany, Italy and Spain have been particularly heterogeneous, with Italy having its general debt-to-GDP ratio at 120 percent while Germany and France’s were 60 and 78 percent respectively in 1999 just before the introduction of EMU (OECD, 2018), there was an obvious tendency of convergence in bond yields. Conditional and unconditional correlation tests confirm that convergence and decreasing heterogeneity between France, Germany, Italy and Spain is statistically significant and strong, (Ehrmann, Fratzscher, Gurkaynak & Swanson, 2007). In addition, UK was used as a “control” country in this analysis, and it was found that convergence is exclusively more intense for the euro area countries. Such findings strongly suggest the idea that convergence is observed due to EMU creation instead of a general global tendency towards synchronization of advanced economies.
Lastly, Christiansen (2007) applied GARCH-type model for the US and European bond markets (both EMU and non-EMU) and found that there are three types of volatility effects: global, regional and local. This paper is very important for the thesis as it suggests a potential channel through which adoption of the euro could partly lead to the convergence in bond yields – EMU countries were found to be less affected both by global and local effects, while regional effect appeared to be the dominant one (Christiansen, 2007). For non-EMU countries, on the other hand, local effects have been found to be in the primary position. Therefore, it is possible to argue that after adoption of the euro, EMU member states become much more synchronized with other fellow EMU countries and local volatility as well as global effect lose their importance. Theoretically, that could be the focal point when the convergence in bond yields is observed, as then the eurozone economies are considered to carry comparable levels of risks because they face similar volatility effects.
This thesis does not employ the method of conditional correlations since sample countries in the data set have joined the eurozone at different times (there is no single point where all countries became part of the EMU) and are relatively small economies, so, it would be hard to compare time-varying correlations properly.
Multiple studies have also been conducted where a different approach was taken - bond market integration under EMU has been investigated by performing Capital Asset Pricing Model (CAPM).
Abad, Chulia and Gomez-Puig (2009) found that the introduction of the euro appears to be significant in the integration process of the European government bond market. Just like Christiansen (2007), they also confirm that countries which are part of the EMU are less vulnerable to global risk factors while vulnerability to internal EMU risk becomes more prevalent. Non-EMU countries, on the other hand, are significantly more sensitive to world risk factors. Paper by Pozzi and Wolswijk (2008) has also employed the CAPM and concluded that idiosyncratic components have converged to zero, so it appears that government bond markets of Belgium, France, Italy and the Netherlands have become more integrated after the introduction of the common currency.
Lastly, conclusions about the effect of adoption of EMU reached by previously mentioned papers have been partly confirmed in studies of Adam et al. (2002) and Gupta, Sehgal and Deisting (2015) where beta-convergence test has been conducted to show the speed of financial market integration after adoption of the euro. Both papers confirm that a negative value for
beta coefficient is present, which means that there is a mean reversion of returns – convergence is indeed present and empirically observable. In normal times, the value of beta coefficient usually has a larger value in absolute terms, which is between -1 and -2, while during the crisis period, beta is still negative, but has a lower value which is between 0 and -1 (Gupta, Sehgal & Deisting, 2015). It means that the process of convergence is much slower during recessions and periods of uncertainty.
Adam et al. (2002) also provide evidence that there is convergence in the 10-year government bond market. They distinguish two time periods – before and after 1999. Interestingly enough, they found that convergence has been most intense from the end of 1997 up until 1999 - just before the creation of EMU (Adam et al., 2002). Overall, beta-convergence tests confirm that EMU member states see more intense process of financial market integration than non-EMU countries
On the other hand, one must still acknowledge the fact that government bonds issued by different economies are not seen as perfect substitutes and bond markets remain segmented due to incomplete elimination of default, market liquidity and currency risks even though EMU countries are perceived to experience a larger degree of convergence than non-EMU countries (Abad, Chulia & Gomez-Puig, 2009; Ehrmann, Fratzscher, Gurkaynak & Swanson, 2007; Pagano & Von Thadden, 2004). There is even a paper by Canova, Ciccarelli and Ortega (2007) where, after employing a panel VAR model for G-7 countries, they claim that business cycles after the creation of EMU have not become more synchronized so, it might be possible that convergence among member states of EMU could have been milder or not even present at all. From the overview of the relevant literature, it appears that there are multiple important determinants of long-term bond yields. Adoption of the euro is also distinguished among them and appears to play an important role as well – at least partly leads to deeper integration of financial markets and convergence in long-term interest rates. Therefore, empirical research part of the thesis will test the hypothesis which is provided below:
H1: Adoption of a common currency leads to increased convergence in long-term government
3. Empirical research
In this part of the thesis, I test empirically whether adoption of the euro, among other important variables, has an impact on bond market integration. First of all, I describe the data set and variables that are used in the regression. Then methodology is explained and empirical regression analysis is conducted before summarizing the results. I finish by discussing the limitations of this study and provide recommendations for future research.
3.1. Data sample, variables and methodology approach.
3.1.1 Data sample. Data sample chosen for this empirical study consists of European
economies that have joined the EU during the fifth enlargement in 2004 and eventually adopted the euro. There are 6 countries that have introduced the euro after becoming part of the EU – Cyprus, Latvia, Lithuania, Malta, Slovakia and Slovenia.
Such a sample was selected due to several reasons. First of all, above mentioned countries are less developed than 11 original economies that created the EMU in 1999 (e.g. Germany, France, Netherlands, Austria). According to the Solow Growth Model, poorer countries are likely to catch up with the incomes of more developed countries if they manage to increase their savings rate – that can be done more easily if investments are to be increased. However, less developed countries find it more difficult to borrow capital in the international financial markets for low interest rates due to higher risks. Therefore, it is interesting to study this sample, as economies within the data set would be affected more and are likely to benefit, compared with economies that are in the steady state, from the possibility of access to cheap financing as this capital could then be used for productive investments.
Secondly, as mentioned in the previous chapter, effects of creation of the EMU on long term interest rates have already been analyzed, however, this has only been done for the original member states of the EMU that adopted the euro in 1999. Unfortunately, but the effects of adoption of the euro in my sample countries have not been as extensively analyzed, therefore, I decided that it would be interesting to see if adoption of the euro in the countries that joined the EU in 2004 had any effect on their bond market integration. To add, there is not much literature overall, that analyzed long-term interest rates and their determinants in countries that I am analyzing, so this thesis also checks the validity of previous research whether determinants of long-term interest rates are actually significant in this sample.
The data range spans from the 1st quarter of 2002 up to the 2nd quarter of 2017. This date range is selected because the beginning of 2002 is the point at which data on long-term interest rates began to be provided for countries in the data set. Middle of 2017 is the last period of my sample due to the availability of observations on some of the control variables at the time of writing this thesis – such data range allowed to build a strongly balanced data set.
A graph in the following page illustrates the evolution of long-term government bond yield spreads for the sample countries. It is visible, that after Cyprus, Latvia, Lithuania, Malta, Slovakia and Slovenia became part of the EU in 2004, up until the Global Financial Crisis (GFC), spreads converged to minimal values. After the collapse of Lehman Brothers on September 15th, 2008 crisis deepened and spread to Europe – Lithuania and Latvia were hit the hardest (with their long-term interest rate spreads peaking at 11.25 and 10.48 percent respectively during the year of 2009). Cyprus, Malta and Slovenia have joined the EMU just before the crisis and did not seem to be hit that severely, just as Slovakia which joined the monetary union already during the recession (in the beginning of 2009). However, in the end of 2009, Global Financial Crisis evolved into the European Sovereign Debt Crisis (EDC) which hit majority of EU countries and Cyprus, among other economies, saw a rather sharp increase in interest rate spread which persisted for several years. The graph below offers some interesting observations – if only the GFC is considered, it might appear that belonging to EMU leads to lower spreads - Lithuania and Latvia were the only countries in the sample that did not belong to the eurozone during the Global Financial Crisis, but they were the ones out of the entire sample that got hit most severely. On the other hand, Baltic countries quickly managed to lower their yield spreads and when the EDC began, it was now economies that belonged to the eurozone, such as Cyprus and Slovenia, that had the highest long-term interest rate spreads within sample countries.
Figure 2. Evolution of long-term government bond yield spreads for new EU countries
Source: Eurostat
3.1.2. Variable specification. For empirical analysis, I consider an equation that takes the
spread of 10-year government bond yields over Germany (𝑙𝑡𝑖𝑖𝑡) as a dependent variable. German bond yield is used as a risk-free rate towards which the foreign bond yields should converge once an economy adopts the euro - such practice of taking Germany as a benchmark has been widely used by several authors within the literature on determinants of long-term government bond yields (Adam et al., 2002; Costantini, Fragetta & Melina, 2014; Kim, Lucey & Wu, 2005; Orlowski & Lommatzsch, 2005; Poghosyan, 2012). To add, quarterly data on long-term government bond yields has been extracted from Thomson Reuters Datastream. Independent variable of interest – adoption of the euro – is represented with a dummy variable (𝑑𝑢𝑚𝑚𝑦𝐸𝑀𝑈𝑖,𝑡) which has a value of 0 for all periods before an economy adopted the euro and switches to the value of 1 once a common currency has been introduced.
In addition, multiple control variables are added to the empirical analysis part that could potentially help determine the spread of long-term government bond yields and lead to a deeper bond market integration. Control variables can be split into two broad categories. First category includes variables that are part of the Maastricht convergence criteria – government budget surplus, gross government debt, real effective exchange rate and inflation (HICP). Second category includes demographic changes, real labor productivity, log of potential growth of GDP and openness of an economy – macroeconomic variables that reflect the economic state of a country. In addition, short-term interest rates, common risk factor and crisis variables have
-2 0 2 4 6 8 10 12 Q 1 2002 Q 3 2002 Q 1 2003 Q 3 2003 Q 1 2004 Q 3 2 00 4 Q 1 2005 Q 3 2005 Q 1 2006 Q 3 2006 Q 1 2007 Q 3 2007 Q 1 2008 Q 3 2008 Q 1 2009 Q 3 2009 Q 1 2010 Q 3 2010 Q 1 2011 Q 3 2011 Q 1 2012 Q 3 2012 Q 1 2 01 3 Q 3 2013 Q 1 2014 Q 3 2014 Q 1 2015 Q 3 2015 Q 1 2016 Q 3 2016 Q 1 2017 Q 3 2 01 7 Q 1 2018 Sp re ad in lo n g-term b o n d y ie ld s, %
been added to the model as, according to the literature, they are found to directly influence the dependent variable. All of the control variables and rationale behind them are shortly described below.
Budget balance and gross government debt. Within the literature on determinants of
long-term government bond yields, multiple scholars have investigated the effect of government budget balance as well as public debt and found that fiscal imbalances are significant in explaining part of the variance of bond yield (Afonso, Arghyrou & Kontonikas, 2015; Alexopoulou, Bunda & Ferrando, 2009; Ciocyte, Muns & Lever, 2016; Costantini, Fragetta & Melina, 2014; Cote & Graham, 2004; Ichiue & Shimizu, 2012; Min, 1998). In general, weak fiscal fundamentals of a country are expected to reflect a higher probability of default on sovereign debts. So, countries that are exposed to higher default risk, face positive risk premium. Therefore, it is expected that lower budget surplus and higher government debt should indeed have negative values for beta coefficients. In addition to the first channel, Poghosyan (2012) also claims that fiscal expansion can lead to crowding out of investment and higher marginal product of capital what would eventually cause higher real interest rates – once again a negative value for the coefficient is to be expected. Data on net government budget balance as a percentage of GDP (𝑔𝑏𝑎𝑙𝑖,𝑡) and gross government debt as a percentage of GDP (𝑔𝑑𝑒𝑏𝑡𝑖,𝑡) has been extracted from Eurostat.
Inflation. Inflation has been analyzed as another potential variable which can have an impact
on long-term interest rates, however, literature on the effect of inflation rate is not fully unanimous. Ichiue and Shimizu (2012), Ciocyte, Muns and Lever (2016) and Costantini, Fragetta and Melina (2014) find that higher inflation has a positive relationship to nominal interest rates as increased levels of inflation raise uncertainty towards inflation expectations. That consequently increases the risk premium as well as nominal bond rate. In addition, inflation targeting after becoming part of the monetary union is executed by a more credible and less dependent central bank, therefore, inflation expectations are likely to be lower and so is risk premium (Swanson, 2008). Alexopoulou, Bunda and Ferrando (2009), on the other hand, argue that it is also possible for inflation to have a negative beta coefficient value if a rise in inflation is seen by the market as having structural rather than transitory source. Nevertheless, positive beta coefficient is expected and data on monthly frequency observations for HICP index (ℎ𝑖𝑐𝑝𝑖,𝑡) has been extracted from Eurostat with quarterly observations being interpolated.
Real exchange rate. Likewise, depreciation of a currency works as a signal that economy
could be more volatile, is not able to maintain the exchange rate within specific band and is losing its competitiveness (Afonso, Arghyrou & Kontonikas, 2015; Alexopoulou, Bunda & Ferrando, 2009; Min, 1998; Orlowski & Lommatzsch, 2005). According to the IMF‘s knowledge base (n. d.): “An increase in real effective exchange rates implies that exports become more expensive and imports become cheaper; therefore, an increase indicates a loss in trade competitiveness“ – positive relationship between an increase in REER and long-term government bond yield spread is expected. Quarterly data on real effective exchange rates compared to 28 trading partners (𝑟𝑒𝑒𝑟𝑖,𝑡) was extracted from Thomson Reuters Datastream.
Short-term interest rates. To add on variables from Maastricht convergence criteria,
Alexopoulou, Bunda and Ferrando (2009) have also explained that short-term interest rates are important determinants because expectations theory says that long-term interest rates are a reflection of current and expected short-term interest rates. Therefore, Eurostat database was used to extract quarterly data on 3-month money market rates (𝑠𝑡𝑖𝑖,𝑡) and one could expect
higher short-term interest rates to have positive impact on long-term yields, so, positive beta coefficient is predicted.
Common risk factor. Variable that reflects the level of general volatility and risk, that is
common across EMU, was stressed to be extremely important as well (Afonso, Arghyrou & Kontonikas, 2015; Alexopoulou, Bunda & Ferrando, 2009; Beber, Brandt & Kavajecz, 2009; Bicu & Candelon, 2012; Pozzi & Wolswijk, 2008). By applying a method of principal component analysis, Smets (2013) also states that 88 percent of the overall variance in long-term bond yields can be explained by a global factor. A sudden increase in perceived risk across all of the economies can be expected to cause a capital flight to safety. This would inevitably mean that EMU economies that are perceived to be more risky would bleed the capital while economies that are considered to be the safest ones, such as Germany or the Netherlands, are expected to attract more capital, leading to unbalanced outcome - increased interest rates for risky and lower interest rates for safe economies. Therefore, a positive relationship is predicted between increased global risk factor and long-term interest rate spread. Datastream provided quarterly data on a composite indicator of systemic stress across the eurozone (𝑣𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑡) which reflects the volatility of the European bond markets.
Crises. As common risk factor reflects shorter term volatility of the markets, dummy variables
present. In the empirical literature, effects of crisis have been distinguished and it was found that economic downturn has a negative impact on bond yield convergence (Gupta, Sehgal & Deisting, 2015; Koukouritakis, 2017). This thesis follows the logic of Gupta, Sehgal and Deisting (2015) and introduces two dummy variables – one for the Global Financial Crisis (GFC) from the 3rd quarter of 2007 until the 3rd quarter of 2009 (𝑑𝑢𝑚𝑚𝑦𝐺𝐹𝐶𝑡) and one for the European Debt Crisis (EDC) from the 4th quarter of 2009 onwards (𝑑𝑢𝑚𝑚𝑦𝐸𝐷𝐶𝑡).
Demographic changes. Population ageing has also been distinguished as a potential
macroeconomic determinant of government bond yield (Ciocyte, Muns & Lever, 2016; Ichiue & Shimizu, 2012). Demographic changes are becoming more and more relevant for developed economies and as population ages, demand for safe assets increases, leading to lower yields on government bonds. Likewise, as labor supply diminishes and marginal productivity of capital goes down, population ageing is expected to have a negative effect on long-term interest rates. Ciocyte, Muns and Lever (2016) have employed a counter-variable – share of people aged 20-39 years in relation to the total population (𝑑𝑒𝑚𝑐ℎ𝑎𝑛𝑔𝑒𝑠𝑖,𝑡) – as an indicator for demographic variable, and so higher proportion of young labor force is expected to have a positive impact on long-term bond yields. Yearly observations have been obtained from Eurostat for the latter variable and quarterly data has been linearly interpolated.
Labor productivity. In addition to population ageing, labor productivity has also been
mentioned among significant macroeconomic fundamentals as it is seen to have a positive impact on long-term interest rates (Ichiue & Shimizu, 2012). Therefore, quarterly data on real labor productivity per hour worked (𝑙𝑎𝑏𝑝𝑟𝑜𝑑𝑖,𝑡) has been obtained from Eurostat and is expected to have a positive beta coefficient value.
Gross domestic product. Gross domestic product is one of the major macroeconomic
indicators which signals country‘s economic health and is sometimes used as an indicator to reflect the catching-up process in terms of standards of living (Alexopoulou, Bunda & Ferrando, 2009; Orlowski & Lommatzsch, 2005). Real GDP growth filtered of cyclical fluctuations was found to be significant (Ciocyte, Muns & Lever, 2016; Poghosyan, 2012). I chose the latter form of GDP variable (expressed in millions of euros) and obtained data for it from Eurostat database. Since there is very large variation in GDP figures, I used the spread of logarithms in order to normalize the variable (𝑙𝑜𝑔𝐺𝐷𝑃𝑖,𝑡). Lastly, negative beta coefficient is
expected as economically strong country is more likely to cope better with crises and risks, so should be facing lower long term interest rates.
Openness. Last macroeconomic variable that is found to be important - openness of an
economy (Alexopoulou, Bunda & Ferrando, 2009). Usually a higher degree of openness signals country‘s sound solvency and ability to generate surplus when it needs to finance existing debts, so negative beta coefficient is expected. This variable is expressed as a sum of imports and exports in relation to GDP (𝑜𝑝𝑒𝑛𝑛𝑒𝑠𝑠𝑖,𝑡) and has been constructed using quarterly data from Eurostat database.
3.1.3. Methodology approach. In this thesis I would like to explore, if, put together with other
main determinants distinguished within the literature, country’s admittance to the EMU could be statistically significant and has potential to explain part of the variance of bond yield spread. Literature suggests two main methodologies that can be used. The first strand of literature suggests using error correction models to explain what are the main determinants behind government bond yields and their convergence (Alexopoulou, Bunda & Ferrando, 2009; Ciocyte, Muns & Lever, 2016; Cote & Graham, 2004; Koukouritakis, 2017; Poghosyan, 2012). However, since my thesis takes accession to the EMU as a dummy variable, which is, of course, an exogenous variable, it is no longer feasible to use error correction models. On top of that, if one was to perform panel VECM, according to the theory, before going further structural break would have to be identified (Koukouritakis, 2017). Unfortunately, since I am analyzing countries that have joined the European Economic and Monetary Union at different dates spanning from 2007 to 2015, it would not be possible to identify a clear structural break for panel VECM.
On the other hand, several researchers have used panel OLS models to extract the main determinants of long-term government bond yields (Ichiue & Shimizu, 2012; Min, 1998; Orlowski & Lommatzsch, 2005) and their results were not contrary to other findings. Likewise, Lane (2005) was testing if EMU member states have higher levels of external bond holdings by estimating a robust pooled OLS model and, most importantly, his equation specification had a dummy variable for membership in EMU. General specification of Lane (2005) had a pair-wise dummy for the euro area countries together with various control variables. Therefore, it might be appropriate to carry on with a panel OLS model in order to replicate the real-world simulation within the sample and test if dummy variable for adoption of the euro, together with many other important indicators, can appear to be significant in determining the convergence in bond yield spreads.
Below is provided the regression equation which is estimated using OLS. Within the regression equation, subscript i denotes spreads of individual countries vis-a-vis Germany while subscript
t represents different time periods.
𝑙𝑡𝑖𝑖,𝑡 = 𝛼 + 𝛽1𝑑𝑢𝑚𝑚𝑦𝐸𝑀𝑈𝑖,𝑡 + 𝛽2𝑔𝑏𝑎𝑙𝑖,𝑡+ 𝛽3𝑔𝑑𝑒𝑏𝑡𝑖,𝑡+ 𝛽4ℎ𝑖𝑐𝑝𝑖,𝑡 + 𝛽5𝑟𝑒𝑒𝑟𝑖,𝑡+ 𝛽6𝑠𝑡𝑖𝑖,𝑡+ 𝛽7𝑣𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦𝑡+ 𝛽8𝑑𝑒𝑚𝑐ℎ𝑎𝑛𝑔𝑒𝑠𝑖,𝑡+ 𝛽9𝑙𝑎𝑏𝑝𝑟𝑜𝑑𝑖,𝑡+ 𝛽10𝑙𝑜𝑔𝐺𝐷𝑃𝑖,𝑡+ 𝛽11𝑜𝑝𝑒𝑛𝑛𝑒𝑠𝑠𝑖,𝑡+ 𝛽12𝑑𝑢𝑚𝑚𝑦𝐺𝐹𝐶𝑡+ 𝛽13𝑑𝑢𝑚𝑚𝑦𝐸𝐷𝐶𝑡+ 𝜀𝑖,𝑡, i = 1, …, N; t = 1, …, T.
3.2. Panel data analysis.
3.2.1. Panel diagnostics. First of all, I construct theoretically appropriate panel OLS model in
Stata software and run panel diagnostics to determine which specification would be more appropriate and reliable to use (see appendix A). After performing the Hausman test, miniscule p-value of 0 suggests that the null hypothesis of random effects model being consistent must be rejected and fixed effects model should be used instead. Therefore, based on panel diagnostics, fixed effects model seems more plausible and suitable for empirical analysis.
3.2.2. Heteroskedasticity. Since panel diagnostics suggest using the fixed effects model for
the regression, possible danger of heteroskedasticity has to be investigated as well (see appendix B). Therefore, modified Wald test for groupwise heteroskedasticity is performed on the fixed effects model and it yields a p-value of 0. This result means that the null hypothesis, which says that residuals are homoskedastic, must be rejected and there is a threat of heteroskedasticity. This issue has to be kept in mind as it will later impact the choice of most suitable specification for the model.
3.2.3. Autocorrelation. Additionally, the model has to be tested for a serial correlation and
that is done by performing Wooldridge‘s test for autocorrelation (see appendix C). Wooldridge‘s method uses the residuals from a regression in first differences and tests the null hypothesis of no serial correlation (Drukker, 2003). After performing the test of autocorrelation for my model, p-value of 0.0005 is observed, meaning that the null hypothesis has to be rejected even at 99% confidence interval, and, indeed, first-order autocorrelation is present.
Therefore, in order to deal with the problem of heteroscedasticity and serial correlation, two different models can be approached. Fixed effects model can still be used, but it has to be clustered at a country level. Such model specification is considered to be consistent and bear the danger of heteroscedasticity and autocorrelation in mind when providing the output.
3.2.4. Normality of residuals. Lastly, test statistics for normality of residuals for clustered
fixed effects model indicates that residuals are not normally distributed (see appendix D), as p-value for the model is merely equal to 2.1*10-6 – the null hypothesis of residual normality must be rejected. Histogram, plotting residuals, provides visual evidence that residuals are not normally distributed (see appendix E). Therefore, when assessing empirical findings, one must be careful and acknowledge the fact that residuals are not distributed normally, so, findings might be biased.
3.3. Results.
In this sub-section I provide and describe the results of a fixed effects model which was recommended by panel diagnostics but clusters at a country level are used to account for autocorrelation and heteroskedasticity issues. Table 1 below presents the final clustered fixed effects model. Already from the first sight, it can be observed that regression results support the hypothesis of this thesis which suggests that adoption of the euro leads to long-term government bond rate convergence. Assessment of the significance of the coefficients and explanation of the relationships that independent variables have with the dependent variable (long-term government bond yield spread) are provided as well.
Table 1. Final model – clustered fixed effects. Dependent variable – long-term government bond yield spread; Independent variable of interest – dummy for EMU
FE clustered Dummy EMU -0.869* (-2.92) Surplus -0.0298 (-1.95) Debt 0.00107 (0.50) Inflation 0.0389* (2.65) REER 0.0970** (6.36)
Short-term interest rate 0.565**
(4.23) Volatility 7.647** (4.31) 20-39 population share 0.572*** (7.58) Labor productivity -0.0315 (-1.15) GDP -7.422*** (-20.54) Trade openness -0.833*** (-7.58) Dummy GFC 0.571 (2.20) Dummy EDC 1.172** (4.12) Constant -35.70*** (-20.29) N Adj. R2 372 0.733
Note: t-statistics in parentheses; * indicates significance at the 95 percent level; ** indicates
significance at the 99 percent level; *** indicates significance at the 99.9 percent level.
Dummy EMU. Model indicates that the dummy variable for EMU (Dummy EMU) is
significant at 95% confidence level with a t-value of -2.92. A negative beta coefficient of -0.87 follows the logic of the hypothesis that once a country adopts the euro, it should see a reduction
of 0.87% in long-term government bond yield spread. One must also keep in mind that data set contains almost all most important determinants of long-term government bond interest rates that are almost unanimously thought to be significant within the literature. Therefore, the results of this model are very favourable in a sense that a dummy variable seems to be significant among all strong determinants and it appears that adoption of the euro could work as a channel which partly impacts long-term interest rates. Findings of this research are in line with the empirical literature on this matter.
Most significant variables. Demographic changes (20-39 population share), GDP and
openness (Trade openness) - all of these variables are significant even at 99.9% significance level with absolute t-values above 7 (7.58, -20.54 and -7.58 respectively). 0.57 coefficient for demographic changes shows that a 1% increase in spread of population aged between 20-39 years can lead to increased interest rate spread by 0.57% meaning that the younger population of the country does not have such high demand for safe securities and that does not drive down interest rates for them. On the other hand, a logarithm value of GDP spread has a negative coefficient, meaning that when an economy is economically catching up with Germany by 1%, the spread in interest rates goes down by 7.42% as a proportion. The same coefficient sign is observed for openness, which is expressed as a sum of imports and exports relative to GDP and has a negative coefficient of -0.83, suggesting that a 1% increase in relative trade volumes vis-a-vis Germany leads to a 0.83% decrease in interest rate spread.
The following control variables are significant at least at 95% confidence interval – short-term interest rate, inflation, real exchange rate (REER), common risk factor (Volatility) and dummy variable for the European Debt Crisis (Dummy EDC) (with t-values of 4.23, 2.65, 6.36 and 4.12 respectively). All variables within this category have positive coefficients. A coefficient of 0.57 for short-term interest rates shows that a 1% increase in spread leads to an increase of long-term interest spread by 0.57%. Volatility measure has a beta coefficient equal to 7.65, and it is important to note that the common risk factor which is expressed as volatility of the euro area bond market as a whole, has the single biggest coefficient – this finding supports the claim in theory that markets tend to miscalculate associated risks during the economic boom but in times of uncertainty, markets become more volatile and sudden spikes in bond yield spreads are observed (Bicu & Candelon, 2012; Gupta, Sehgal & Deisting, 2015, Smets, 2013).
Dummy variable for the European Debt Crisis, unlike GFC, is significant at 95% confidence level and has a coefficient value of 1.17 meaning that since the beginning of the Sovereign
Debt Crisis long-term interest rate spreads have structurally increased. Even though inflation and real exchange rate were found to be strongly significant, their coefficients are of miniscule size (0.04 and 0.01 respectively).
Budget surplus and dummy variable for GFC. The budget surplus (Surplus) is only
significant at an 80% confidence interval with a p-value of 0.11. A negative beta coefficient of -0.03 is nearly negligible but suggests that a 1% decrease in budget deficit (an increase in surplus) vis-a-vis Germany leads to a lower long-term rate spread by 0.03%. Dummy variable for GFC (Dummy GFC) is only significant at a 90% confidence level and indicates that period of Global Financial Crisis has caused an increase in long-term interest rate spreads by 0.57%.
Insignificant variables. On the other hand, as can be seen in table 1, gross government debt
and labor productivity variables are least informative (with p-values of 0.64 and 0.30 respectively), so can be discarded as insignificant.
3.4. Robustness checks.
In order to justify the findings of empirical analysis and make sure that results are consistent, robustness checks need to be performed. First of all, I perform within-model transformations to see if results of the original model are consistent. I have decided that it would be beneficial to run the regression equation using different sets of variables and check whether coefficient signs are consistent. Lastly, the findings of the main model are compared with other models – regular fixed effects and generalized least squares (GLS).
3.4.1. Within-model transformations. As a first robustness check, clustered fixed effects
model is ran using only the dummy variable for EMU as an independent variable in the regression equation. The following regression equation is therefore evaluated.
𝑙𝑡𝑖𝑖,𝑡 = 𝛼 + 𝛽1𝑑𝑢𝑚𝑚𝑦𝐸𝑀𝑈𝑖,𝑡+ 𝜀𝑖,𝑡, i = 1, …, N; t = 1, …, T.
Then, several more control variables are added to the regression equation. These variables have been outlined by the European Union as Maastricht convergence criteria and they are: government budget surplus (𝑔𝑏𝑎𝑙𝑖,𝑡), gross government debt (𝑔𝑑𝑒𝑏𝑡𝑖,𝑡), inflation rate (ℎ𝑖𝑐𝑝𝑖,𝑡) and real effective exchange rate (𝑟𝑒𝑒𝑟𝑖,𝑡) (European Monetary Institute, 1996).
𝑙𝑡𝑖𝑖,𝑡 = 𝛼 + 𝛽1𝑑𝑢𝑚𝑚𝑦𝐸𝑀𝑈𝑖,𝑡+ 𝛽2𝑔𝑏𝑎𝑙𝑖,𝑡+ 𝛽3𝑔𝑑𝑒𝑏𝑡𝑖,𝑡+ 𝛽4ℎ𝑖𝑐𝑝𝑖,𝑡+ 𝛽5𝑟𝑒𝑒𝑟𝑖,𝑡+ 𝜀𝑖,𝑡,
Lastly, remaining variables, that reflect macroeconomic conditions of the economies as well as market volatility and crises, are added and the original regression equation which is then evaluated by using fixed effects model clustered at a country level. Table 2 below compares outputs of these 3 models with different specifications.
Table 2. Models with different equation specifications. Dependent variable – long-term government bond yield spread; Independent variable of interest – dummy for EMU
Model 1 Model 2 Model 3
Dummy EMU 0.289 -0.562 -0.869* (0.49) (-0.65) (-2.92) Surplus -0.0812** -0.0298 (-4.11) (-1.95) Debt 0.0155 0.00107 (1.03) (0.50) Inflation 0.0984** 0.0389* (4.70) (2.65) REER 0.00765 0.0970** (0.24) (6.36)
Short-term interest rate 0.565**
(4.23) Volatility 7.647** (4.31) 20-39 population share 0.572*** (7.58) Labor productivity -0.0315 (-1.15) GDP -7.422*** (-20.54) Trade openness -0.833*** (-7.58) Dummy GFC 0.571 (2.20) Dummy EDC 1.172** (4.12) Constant 1.631** (5.87) 2.603** (5.47) -35.70*** (-20.29) N Adj. R2 AIC 372 0.003 1499.3 372 0.181 1429.9 372 0.733 1005.0
Note: t-statistics in parentheses; * indicates significance at the 95 percent level; ** indicates
As can be observed from the table, when dummy variable for EMU is taken as the only independent variable which explains all variance in long-term interest rates (model 1), a positive coefficient of 0.29 provides counter-intuitive results suggesting that adoption of the euro led to increased spreads in long-term government bond yields. The same regression specification has been used to run additional models for each country individually in order to investigate this matter further (see appendix F).
From the results, it appears that Latvia and Lithuania, which were the only non-EMU countries when GFC and EDC started, have negative beta coefficients (with t-values of -2.47 and -1.65 respectively). Remaining 4 countries, that became part of the EMU just before the Global Financial Crisis over the period from January of 2007 to January of 2009, seem to have positive coefficients. Cyprus has by far the most significant and largest coefficient (coefficient value is 2.35 while t-value is 6.20) out of 4 economies that became part of EMU just before or already during the crisis – it led to strong upward pressure on the coefficient sign. However, results from Cyprus cannot be considered to be reliable because from graph 2 in section 3.1.1, it can be seen that once the European Sovereign Debt Crisis started, Cyprus experienced an upward structural shift in its long-term government bond yield spread and this tendency persisted for an extended period of time. Therefore, this model with only one explanatory variable is clearly vulnerable to omitted variable bias as it does not consider multiple important indicators such as market volatility or crisis variable. After running the same model with Cyprus being excluded from it, results suggest that the EMU dummy variable is insignificant but now it has a negative beta coefficient of -0.19 (see appendix G) confirming the hypothesis that Cyprus was the entity that caused discrepancies.
When comparing the final clustered fixed effects model (model 3) with the model that only considers Maastricht convergence criteria variables (model 2), results are quite satisfactory because all included variables have the same coefficient signs. Despite that, it appears that the model with only four control variables is still subject to omitted variable bias because of two reasons. First, empirical literature provides several more variables that are found to be important determinants of long-term interest rates. Also, it is still visible that the significance of variables differs across models, so, there is a possibility that additional variables might have to be included.
3.4.2. Comparison of models. Table 3 is below compares the results of main clustered fixed
effects model to regular fixed effects and GLS models in order to check if beta coefficients are consistent.
Table 3. Comparison of models in levels. Dependent variable – long-term government bond interest rate spread; Independent variable of interest – dummy for EMU
FE clustered FE GLS Dummy EMU -0.869* -0.869*** -1.133*** (-2.92) (-4.59) (-5.65) Surplus -0.0298 -0.0298** -0.0410*** (-1.95) (-2.84) (-3.31) Debt 0.00107 0.00107 0.00904* (0.50) (0.19) (2.18) Inflation 0.0389* 0.0389* -0.0238 (2.65) (2.09) (-1.17) REER 0.0970** 0.0970*** 0.0211 (6.36) (6.37) (1.60)
Short-term interest rate 0.565** 0.565*** 0.590***
(4.23) (16.77) (15.28) Volatility 7.647** 7.647** 11.59*** (4.31) (3.31) (4.37) 20-39 population share 0.572*** 0.572*** 0.237*** (7.58) (10.86) (6.19) Labor productivity -0.0315 -0.0315 -0.0335 (-1.15) (-1.76) (-1.58) GDP -7.422*** -7.422*** -0.761*** (-20.54) (-10.67) (-5.62) Trade openness -0.833*** -0.833** -0.853*** (-7.58) (-2.62) (-6.56) Dummy GFC 0.571 0.571* 0.434 (2.20) (1.99) (1.51) Dummy EDC 1.172** 1.172*** 2.010*** (4.12) (3.62) (7.83) Constant -35.70*** -35.70*** -3.433*** (-20.29) (-10.69) (-4.72) N Adj. R2 AIC 372 0.733 1005.0 372 0.729 1023.0 372 1173.7
Note: t-statistics in parentheses; * indicates significance at the 95 percent level; ** indicates
First of all, as economic theory suggests, dummy variable for EMU has the negative coefficient sign in all models what supports the null hypothesis that adoption of the euro leads to long-term interest rate convergence. Dummy variable is significant in the clustered fixed effects model with an absolute t-value of negative 2.92, meaning that adoption of the euro is statistically significant at 95% level. In addition, both regular fixed effects and GLS models provide the same conclusion, showing that dummy variable is statistically significant even at a 99% confidence interval with an absolute t-value ranging from negative 5.65 to negative 4.59. From the table 3 it can also be implied that labor productivity is insignificant not only in the fixed effects model clustered at country level (t-value is equal to -1.15) but also in other models. Therefore, the latter variable could be discarded as an insignificant one which does not have any impact on bond market integration, and does not explain any part of variation in the dependent variable.
In addition, even though fiscal balance fundamentals (government surplus and gross government debt) have the predicted coefficient signs in all models and are found to be significant in at least one of the three models, they are not found to be significant at 95% level in the recommended clustered fixed effects model.
Lastly, when considering the crisis variables, it appears that the European Debt Crisis (EDC) has taken a heavier toll on European economies than Global Financial Crisis (GFC). That is the case because EDC‘s beta coefficient is twice as large as GFC‘s (1.17 and 0.57 respectively) and also EDC was found to be significant at 99% confidence level while GFC is considered to be irrelevant with a t-value of 2.20 in the chosen model of this study (and is only significant at 90% level). These findings support the reasoning that adoption of a common currency leads to a lower importance of global and local volatility effects, meaning that regional volatility becomes more relevant, so, EMU economies become more synchronized leading to convergence.
For the rest of control variables, they all have expected coefficient signs that are in line with theory and have all been found to be statistically significant at least at a 95% confidence interval.
Therefore, it is possible to claim that the clustered fixed effects model reflects reliable relationships, as both simple fixed effects and generalized least squares model both show familiar coefficients and similar levels of significance of the variables.
4. Policy implications and limitations
The results of empirical analysis in the thesis suggest that adoption of a common currency can lead to lower long-term government bond yields in new EU economies. Therefore, countries that are candidates for EMU should keep in mind that accession to the eurozone could potentially lead to lower long-term government bond yields. They may take advantage of this situation by gaining access to cheaper borrowing of the capital which could then be used for productive investments. Therefore, by being less uncertain about the value of debt in the future, converging countries can then more easily catch up with the standards of living in more developed economies that are already in a steady state.
However, policy makers should recognize the fact that new member states of the EU are highly heterogeneous and those that adopt the euro can gain access to cheaper borrowing even if they have slightly weaker macroeconomic or fiscal fundamentals than highly developed economies. This situation could pose potential obstacles in the future, as has been seen during the Global Financial Crisis and the European Sovereign Debt Crisis. During the times of economic prosperity markets tend to miscalculate the risks associated to sovereign states and once an exogenous shock is observed, sudden spikes in borrowing rates might be caused.
One should be careful when considering the results as this study has some limitations regarding the model and empirical methodology.
First of all, convergence in long-term government bond yields has also been one of the 5 Maastricht criteria – long-term interest rates could not be higher more than 2% when compared to 3 EU member states which had the lowest inflation rate (European Monetary Institute, 1996). Therefore, convergence in long-term interest rates is supposed to and was historically observed prior to the creation of EMU. In the future research it would be interesting to somehow isolate the anticipation of creation of the common currency area in order to see the level of convergence purely due to adoption of the euro.
In addition, the fact that the main independent variable of interest is a dummy variable and new EU countries joined the EMU at different times, it was impossible to apply error correction model. VECM model is a dynamic model which allows to see both short and long-term effects of independent variables. However, since due to data set limitations it was not employed - fixed effects model clustered at a country level was used. The latter model is not as advanced as dynamic models, therefore, results ought to be evaluated carefully considering that more
primitive model might present less reliable results. It might be beneficial to explore the possibilities of employing different more advanced models that would allow to see both short-term and long-short-term dynamics.
Moreover, as mentioned in the literature review, some researchers have used GARCH-based methods to observe conditional correlations between different bond markets over time. Since new EU countries joined the EMU at different times, it would be difficult but may be possible to expand the correlation matrix of countries every time a new economy joins the eurozone. However, several new EU countries have become members of EMU just before the crisis (Cyprus, Malta, Slovakia and Slovenia), therefore, a crisis effect has to be properly isolated, otherwise it might seem that after adoption of a common currency, conditional correlations went downwards.
Lastly, this thesis did not evaluate the liquidity risk because then the data would have had to be obtained at a much higher frequency (even observations at a weekly frequency would not be enough) in order to see appropriate bid-ask spreads of government bonds. However, then the default risk could not be reflected properly as data on many macroeconomic indicators is usually not reported more often than on a quarterly basis. Therefore, in the future, it would be interesting to replicate the model used in this thesis by using more frequent observations and properly reflecting the liquidity risk.
5. Conclusions
The main aim of this thesis was to test empirically if there exists a relationship between adoption of the euro in new EU countries and convergence in long-term government bond yields.
A review of the academic literature in the second chapter described the main determinants that drive long-term government bond yields and explain part of the variation in spreads. Then, focus was put on the effects of EMU. On top of lower currency risk, diminishing importance of local and regional volatility effects due to the creation of a common currency area has been suggested as a potential channel through which adoption of the euro could lead to further bond market integration. This theory has been confirmed by the literature - mainly by applying GARCH-based methods, scholars have concluded that founding member states of EMU have seen convergence in long-term interest rates after the introduction of the euro.
Based on the literature, I established a hypothesis which predicted that adoption of a common currency by new EU countries should lead to increased bond market integration and convergence in long-term interest rates towards Germany. Hypothesis assumes that after accession to EMU, regional volatility effects gain more importance leading to higher integration within the monetary union.
In the empirical analysis part of the third chapter, hypothesis was put to a test where I ran a regression equation taking the spread of long-term government bond yields as a dependent variable and dummy variable for EMU as the main independent variable of interest. Data set, consisting of 6 new EU countries that have adopted the euro, has been analyzed with a time frame spaning from the beginning of 2002 up to the second quarter of 2017 which means that the crisis period has been captured in the data as well. Fixed effects model clustered at a country level was selected as the most consistent one and empirical tests showed that adoption of the euro has a negative relationship with a spread of long-term interest rates. These results support the main hypothesis of this thesis and are in line with the findings of other researchers who investigated the effects of EMU on bond markets. Results which were obtained in the empirical analysis part are also supported by the robustness tests, therefore, I suggest that indeed adoption of the euro could lead to convergence in bond yield spreads.
