MSc Business Economics: Finance track
Master Thesis
The effect of Emergency Liquidity Assistance on government
bond yields and credit quality of participating countries.
Student:
Bas van Dijk
10868089
Supervisor:
P.F.A. Tuijp MPhil
2
Statement of Originality
This document is written by Student Bas van Dijk who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is 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.
3
Abstract
This thesis investigates two potential consequences of Emergency Liquidity Assistance. Firstly, it is determined that the lower collateral requirements imposed by the ECB have not led to lower bond yields of participating countries. Secondly, the influence of credit quality on government bond yields is overall not decreased by the use of ELA. However, certain Cypriot and Greek bonds do show a decreased influence of credit quality on bond yields.
Key words: Emergency Liquidity Assistance (ELA), European government bonds, yield spreads, bond liquidity, credit quality, collateral
4
Table of Contents
1. Introduction...……… 5
2. Literature review……… 8
2.1 Relationship between government bond quality and its yield…………... 8
2.2 Relationship between government bond liquidity and its yield……… 9
2.3 Emergency Liquidity Assistance and its implications……….. 11
3. Methodology... 13 3.1 Objectives………... 13 3.2 Method of research………...……….... 14 3.2.1 Testing Hypothesis 1……… 14 3.2.2 Testing Hypothesis 2……… 15 4. Data………..……… 17 4.1 Data collection……… 17 4.2 Variable construction……….... 17 4.3 Coefficient proportions……….... 18 4.4 Summary Statistics………. 18 5. Empirical results……… 24 5.1 Results Hypothesis 1………... 24 5.2 Results Hypothesis 2……… 31 6. Robustness check……….. 41 6.1 Methodology……… 41 6.2 Results Hypothesis 3……….. 42 7. Conclusion………... 47 8. References……… 50 9. Appendix……….…. 54
5
1. Introduction
Recent uncertainty over Greece’s future within the Euro area has caused total deposit outflows of 15 billion euro since last December, according to Reuters (2015). Moreover, multiple Greek banks have lost access to financial markets. Currently, their only lifeline is Emergency Liquidity Assistance (ELA) regulated by the European Central Bank. The ECB has agreed to increase the amount of Emergency Liquidity Assistance available for Greek banks to 75.5 Billion Euros last April. According to the ECB, “ELA is an emergency program extended by a national central bank of the Euro area to solvent lenders facing temporary liquidity problems, without such operations being part of the single monetary policy’’. The money extended is provided by the ECB, even though it is extended by national central banks. At first, national central banks only provided ELA against adequate collateral (bonds with a junk status were not accepted) and only to illiquid but solvent credit institutions.
Nevertheless, the ECB has started accepting Greek government bonds, which have a junk status, as collateral for Emergency Liquidity Assistance since December 19th 2012. This decision has as a consequence that theriskiness of the collateral underlying the ECB loans has increased. Furthermore, the lower collateral requirements incentivize banks to invest in government bonds instead of other assets. Following the ECB decision, Greek bond yields where driven down, as is shown in the graph below.
Figure 1: Greek bond yields between 2010 and 2014 (Bloomberg). Daily data on
10 year Greek government bond yields collected from Bloomberg between January 2010 and July 2014. 0% 5% 10% 15% 20% 25% 30% 35%
Jan-10 Jul-10 Jan-11 Jul-11 Jan-12 Jul-12 Jan-13 Jul-13 Jan-14 Jul-14
10 year Greek government bond yield
10 year Greek Gvt bond yield
6
According to Beber et al. (2009), government bond yields can be explained by two core factors: credit quality and bond liquidity, where the higher the credit quality and bond liquidity, the lower the expected bond yield. They find that during normal market periods, credit quality is the main explanatory variable of government bond yields, whereas during market distress, the liquidity factor becomes more and more important. I suspect that since the ECB has started accepting junk bonds as collateral for ELA, the influence of the credit quality factor on government bond yields has significantly decreased for countries participating in the ELA program since no matter the quality of a government bond, it can be used as collateral for the ECB. Moreover, due to the suspected decrease of influence of credit quality on government bond yields, I suspect that bond yields of participating countries have significantly decreased during times when ELA was used.
This master thesis intends to firstly determine whether the Emergency Liquidity Assistance program has contributed to decreasing government bond yields for the participating countries (Belgium, Cyprus, Greece and Ireland). Secondly, I will determine whether ELA has contributed to a decrease of the influence of credit quality on government bond yields. This has led to the following research question:
Has the introduction of ELA decreased government bond yields and the perception of credit quality of participating countries?
The influence of ELA on government bond yields has not been studied before, while I believe that the results can be contributing and relevant, since it creates crucial insights for academics, investors and policy makers. For academics, the results can help understand cross-market dynamics during market distress between liquidity support programs and bond yields. For investors it offers insights in the development of trading strategies. For policy makers, this paper could help to understand what influences their decisions (accepting low quality collateral) have on bond yields of the respective countries.
This thesis is inspired by research conducted by Beber et al. (2009), where they investigate the relation between bond yields and bond liquidity / credit quality in European government bond markets. Their model is expanded by including Emergency Liquidity Assistance variables in order to measure the influence of ELA on bond yields
7
and the perception of credit quality of participating countries (Belgium, Cyprus, Greece and Ireland). In order to achieve this, this research makes use of data collected between January 1st 2008 and April 30th 2015 for eight countries, of which 4 have used ELA (Belgium, Cyprus, France, Germany, Greece, Ireland, Italy, Netherlands). The data is treated as panel data in Stata.
Firstly, the literature review will focus on existing knowledge and is divided in three parts: the relation between bond yields and credit quality, the relation between bond yields and bond liquidity, and the relation between bond yields and Emergency Liquidity Assistance. Secondly, the methodology part presents the model used and the way the hypotheses are tested. Furthermore, the Data section will explain how data is obtained, the creation of variables and the most important summary statistics. Next, the Empirical results section will provide answers to the research question. Finally, additional results are presented in the Robustness Check section.
8
2. Literature review
The literature review is divided in three parts. In the first part, I will elaborate the relationship between government bond quality and its yield. Secondly, I will explain the relationship between government bond liquidity and its yield, and lastly I will elaborate the relationship between emergency liquidity assistance and bond yields of participating countries.
2.1 Relationship between government bond quality and its yield
Government bond quality is most of the times defined as a bond’s credit worthiness, and is mostly measured by a bond rating. But bond ratings are set by rating agencies and might be slow to react to certain events. This is argued by Perraudin and Taylor (2004), who have studied the consistency of the credit-risk orderings implicit in ratings and bond yields. They provide evidence that for significant periods, more than 25% of high quality bonds are rated in a manner that is inconsistent with their pricing. This is why, in this paper, I use another method to define and measure government bond quality: Credit Default Swap spreads. Credit Default Swap spreads provide a representation of how financial markets value credit risk of the underlying asset, in this case government bonds. Fabozzi et al. (2007) investigate the determinants of the pricing of credit default swaps such as the risk-free rate, the sector, credit rating and fundamentals. Their results show that all these variables have a significant effect on the pricing of credit default swaps.
Hull and Predescu (2004) define a Credit Default Swap spread as the cost per annum to be protected against default. Furthermore, they have tested the theoretical relationship between Credit Default Swap spreads and bond ratings. They argue that the relationship holds well, which means that CDS spreads are a proper representation of bond ratings. Finally, they also tested the relationship between bond yields and Credit Default Swap spreads; they argue that CDS spreads can partly explain bond yields.
Beirne and Fratzscher (2013) have researched the drivers of sovereign risk for 31 emerged and emerging countries during the European sovereign debt crisis. They argue that deterioration in a country’s fundamentals and contagion risk are the main determinants for a rise in sovereign yield spreads and Credit Default Swap (CDS) spreads. Furthermore, they assume that CDS spreads are an appropriate tool to measure bond quality. Maltritz and Molchanov (2013) have conducted the same research with a
9
different model. They use Bayesian Model Averaging to find variables that are most likely to determine credit risk. They find that total debt, history of defaults, currency depreciation and its growth rate are the most important determinants of government bond quality and its yield spreads. Both papers have not compared quieter and distress periods. This is something Wang and Wu (2015) have conducted within corporate bond markets. They argue that investors attribute much more importance to bond quality (and liquidity) in periods of high market distress. Furthermore, they believe credit risk (bond quality) plays a more important role in the explanation of bond yields in both quiet and distress periods.
Furthermore, the research conducted by Beber et al. (2009) was inspired by several other earlier studies. Firstly, Collin-Dufresne et al. (2001) conducted research on the determinants of credit spread changes. Surprisingly, they find that credit spread changes are not driven by credit-risk factors nor liquidity, but by local supply and demand shock that are independent. Another study that has influenced Beber et al. (2009) was a research conducted by Duffie et al. (2003), where the relation between sovereign debt prices and default and restructuring risk is researched. In contrast to Collin-Dufresne et al (2001), Duffie et al. find a significant relation between default risk and Russian government bonds. Furthermore, they find a relatively high explanatory power, which is mostly due to default risk. This could be considered as a relation between credit quality and Russian government bond prices. Lastly, Longstaff, Mithal, and Neis (2005) study corporate credit default swaps in order to measure the influence of default and non-default components in corporate credit default spreads. They find that the majority of the CDS spread is explained by the company’s default risk, and is valid for all bond rating categories.
2.2 Relationship between government bond liquidity and its yield
There is a wide array of papers investigating the influence of bond liquidity on bond yields. Most papers find that (a lack of) liquidity can partly explain their yields.
One of the first articles dedicated to the influence of liquidity on government bond yields was conducted by Amihud and Mendelson (1991). In their study, they examine the effect of illiquidity on the yields of finite maturity securities with identical cash flows: U.S. Treasury bills and notes with maturity less than 6 months; both bonds are equivalent. But their liquidity is different, which is the reason driving the yield
10
difference on both bonds. Since the liquidity on notes is lower, their yields are significantly higher. Their research is based upon their earlier work (Amihud and Mendelson (1986)), where they find that asset returns should be an increasing function of their illiquidity. Thus, the lower the liquidity, the higher the average returns. Boudoukh and Whitelaw (1993) find that almost identical Japanese government bonds trade at large price differentials that could be explained by a lack in liquidity. They argue that this is a way to price bond liquidity.
Ericsson and Renault (2006) link both liquidity and credit risk (or quality) to bond prices and yields. They develop a model to simultaneously capture liquidity and credit risk, by implying that distressed debt is illiquid. As default becomes more likely, their bond yield spreads increase due to increased illiquidity. Furthermore, they find a positive correlation between illiquidity and default risk within the U.S. market. Favero and Pagano (2010) explore determinants of yield differentials between sovereign bonds within Europe. They use a model where they predict that yield differentials across countries should increase in both liquidity and risk, with an interaction between aggregate risk and liquidity. They find that the aggregate risk is consistently priced. When it comes to liquidity, it is only priced for a subset of countries.
Beber et al. (2009) investigate whether bond quality and bond liquidity (partly) determine government bond yields. First of all, they find a unique negative relation between bond liquidity and bond quality for Eurozone government bonds. Most importantly, they find that in normal market conditions, investors attribute more importance to bond quality (measured with credit default swaps on the specific bond) than bond liquidity (measured in 4 different ways). Nevertheless, during periods of high market distress, investors tend to use a “flight-to-liquidity”, and thus prefer liquidity over quality (based on data between 2003 and 2004 for 10 European countries).
Brutti (2011) finds a positive correlation between bond liquidity and credit quality. He investigates the relation between sovereign debt crises liquidity crises. He finds that in emerging markets, when the risk of default for a government increases, the illiquidity on its sovereign debt increases as well. Lastly, De Santis (2014) finds that between 2006 and 2012 within the Eurozone sovereign yields of periphery countries can be explained by the following factors: the traditional credit and liquidity risk, the flight-to-liquidity benefiting the German Bund, and the spillover effect from Greece. Several other studies that study the relationship between bond yields and bond liquidity
11
have been used by Beber et al. (2009). Firstly, the research conducted by Amihud and Mendelson (1986) (which was explained earlier) suggests that asset returns should increase as illiquidity increases (investors get compensated for taking on more risk). Furthermore, Longstaff (2004) researches whether there is a flight-to-liquidity premium in U.S. treasury bonds compared to other government-backed bonds. They find a large liquidity premium in treasury bonds, which is related to consumer confidence. They finally suggest that the popularity of treasury bonds directly influences the value of a U.S. treasury bond. A comparable study has been conducted by Goldreich and Nath (2005). In this case, very liquid U.S. treasury notes and less liquid notes are compared . They find that liquidity of U.S. treasury notes can be predicted (future liquidity), since its liquidity varies very predictably over time. Finally, they argue that the liquidity premium is mostly influenced by the amount of remaining future liquidity.
2.3 Emergency Liquidity Assistance and its implications
The last part of the literature review investigates the impact of Emergency Liquidity Assistance programs on the bond yields of participating countries. Emergency Liquidity Assistance (ELA) is a program created by the ECB in order to supply troubled individual banks with liquidity. ELA provides central bank liquidity solely through the responsible national central bank. Even though ELA is extended by a national central bank, it is still central bank money of the euro area and the ECB’s Governing Council can restrict ELA operations if it considers that these operations interfere with the objectives and tasks of the euro-area system of central banks, or Euro system. The Governing Council takes such decisions with a majority of two-thirds of the votes cast, according to the ECB (2014). However, according to the German Bundesbank (2014), Emergency Liquidity Assistance is not part of a single monetary policy and therefore, resulting costs and risks are to be borne by the national central bank in question. Emergency Liquidity Assistance has been used by multiple countries throughout the European sovereign debt crisis, as is highlighted by Figure 2.
12
Figure 2: ELA outstanding per country between 2009 and 2014. Data is retrieved from
the various balance sheets of the respective National Central Banks for the period between January 2009 and January 2015.
Source: Central Banks of Belgium, Cyprus, Ireland and Greece
Dong (2000) discusses the effects of emergency liquidity support to individual financial institutions under stress. He argues that a properly designed procedure, a clearly laid-out authority and accountability will promote financial stability, reduce moral hazard and protect the lender of last resort from political pressure. Furthermore, Burcu et al. (2013) argue that following Lehman’s failure in 2008, the structural vulnerability of credit markets requires a more central role for central banks as back-stop liquidity providers.
In December 2012 the ECB decided to lower collateral requirements for banks applying for Emergency Liquidity Assistance. Multiple troubled banks ran out of collateral and had mostly Greek government bonds left on their balance sheet, but Greek bonds where having a junk status and where therefore not accepted as ELA collateral. John et al (2003) have conducted research on the effect of collateralized debt on its yields. They find that collateralized debt has lower yields than non- collateralized debt. Koulischer et al (2014) conduct research on whether central banks should lend against low quality collateral. They find that lower quality collateral posted can lead to an
0 20 40 60 80 100 120 140 Ja n-09 A pr -0 9 Jul -09 Oct-09 Ja n-10 A pr -1 0 Jul -10 Oct-10 Ja n-11 A pr -1 1 Jul -11 Oct-11 Ja n-12 A pr -1 2 Jul -12 Oct-12 Ja n-13 A pr -1 3 Jul -13 Oct-13 Ja n-14 A pr -1 4 Jul -14 Oct-14 Ja n-15
Amount of ELA outstanding per country (in billions of euros)
13
increase in interest rates, which in turn leads to a decrease in bond yields. This induces that collateral lowering can lead to decreased bond yields.
Last but not least, Wolff (2014), commissioned by the European Parliament committee on Economic and Monetary affairs, investigates whether the lowering of collateral requirements for obtaining Emergency Liquidity Assistance was unduly changed. Firstly, it is noted that all Euro system credit operations must be based on adequate collateral. In return for adequate collateral (at market prices), banks are provided with liquidity. As explained earlier, the ECB changed its collateral framework during the credit crisis in order to be able to accept lower rated securities as collateral. According to the author, this change was justified in order to provide sufficient liquidity to banks in the Eurozone (Belgium, Cyprus, Greece and Ireland). Furthermore, these changes were necessary for the European Central Bank to fulfill its treaty based mandate of providing liquidity to solvent but illiquid banks and safeguarding financial stability.
3. Methodology
The methodology section offers insights on the expectations for this research. These expectations can be found in the first part. Furthermore, the method of research is defined.
3.1 Objectives
This paper aims to determine whether the Emergency Liquidity Assistance program installed by the ECB to act as a lender of last resort for troubled Eurozone banks has led to lower government bond yields of participating countries, and whether the perception of credit quality has decreased for participating countries. The European Central Bank always requires collateral for Emergency Liquidity Assistance, but since most of the troubled banks ran out of acceptable collateral, the ECB has decided to accept lower quality collateral, such as Greek government bonds, which have a junk status, according to the ECB (2012). I expect that these market interventions have led to lower bond yields for participating countries, since their government bonds can, no matter their quality, be used as collateral. This has led to the formulation of the first Hypothesis:
14
Hypothesis 1: Emergency Liquidity Assistance has led to a decrease in government bond
yields of participating countries.
Furthermore, according to Beber et al. (2009), bond yields can be explained by two main factors: Credit quality and bond liquidity. I expect that since the acceptance of junk bonds as collateral for Emergency Liquidity Assistance, the credit quality factor has significantly decreased, since collateral quality does not matter anymore to obtain ELA. This has resulted in the formulation of the second hypothesis:
Hypothesis 2: Credit quality of government bonds has become less important since the
introduction of Emergency Liquidity Assistance.
In the following two sections, both hypotheses will be thoroughly explained, such as the model used, the expected relationship and the use of control variables. Lastly, it is explained how the significance and sign of various variables relate to the hypothesis tested.
3.2 Method of research
3.2.1 Testing hypothesis 1
We start by investigating whether bond yields of participating countries have been driven down due to Emergency Liquidity Assistance. For this, a sample of 8 Eurozone countries (Belgium, Cyprus, France, Germany, Greece, Ireland, Italy and the Netherlands) is used. As a time period, I use the period between January 2008 and April 2015. Bonds of different remaining maturities up to 10 years are used with a remaining maturity of 5, 7, and 10 years.
The following model is based on Beber et al. (2009), where government bond yields are defined as a function of credit quality and liquidity. I have added two ELA variables in order to capture the effect of Emergency Liquidity Assistance on bond yields of participating countries:
15 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑏𝑜𝑛𝑑 𝑦𝑖𝑒𝑙𝑑𝑖,𝑡− 𝐸𝑢𝑟𝑜 𝑠𝑤𝑎𝑝 𝑦𝑖𝑒𝑙𝑑𝑡
= 𝛽0+ 𝛽1(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽3𝐷1+ 𝛽4𝐸𝐿𝐴𝑖𝑡 + 𝜀𝑖,𝑡
where 𝐶𝐷𝑆𝑖,𝑡 Is the CDS spread in country i and period t, 𝐶𝐷𝑆𝐴𝑉𝐸 is the cross sectional average, 𝐿𝐼𝑄𝑖,𝑡 is the liquidity measure for country I and period t, 𝐿𝐼𝑄𝐴𝑉𝐸 is the cross sectional average of the 𝐿𝐼𝑄𝑖,𝑡 variables, D1 is a dummy equal to 1 when an amount of ELA is outstanding at time t and country I, and 𝐸𝐿𝐴𝑖𝑡 is the amount of ELA outstanding (in billions of Euros).
We expect to obtain a significant relationship between the sovereign yield spread and Dummy 1 and/or the amount of ELA outstanding. Thus, when ELA is outstanding (for dummy 1), or ELA rises (for ELA variable), I expect the sovereign yield spread to decrease. Furthermore, this model makes use of two control variables, which are (𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) and (𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸). These two variables control for credit quality (CDS) and market liquidity (LIQ), which are the two most important factors known influencing bond yields.
Lastly, I expect to find a significant and positive coefficient for the CDS variable, as in Beber et al. (2009), meaning the higher the CDS price (and thus the lower the credit quality), the higher then bond yield spread. For the liquidity coefficient, I also expect to find a positive and significant value (the higher the illiquidity or bid-ask spread, the higher the government bond yield spread). For the dummy 1 variable, which is equal to 1 when an amount of ELA is outstanding, I expect to obtain a negative and significant coefficient. This would mean that when ELA is outstanding, bond yield spreads of that particular country are decreased. I expect the same results for the ELA variable, which accounts for the amount of ELA outstanding. A negative and significant coefficient would mean that when the amount of ELA is increased, bond yield spreads are decreased, and thus countries would get rewarded for having Emergency Liquidity Assistance.
3.2.2 Testing hypothesis 2
The second hypothesis will test whether credit quality of government bonds has become less important since the introduction of ELA. For this, I will use a sample of participating countries (Cyprus, Greece, Ireland and Belgium) between January 2008 and April 2015. Next, I use the model of Beber et al. (2009) to measure whether credit quality of
16
government bonds has become less important. For this, I use two methods: First, I measure whether credit quality has become less important since the ECB announcement on December 19th 2012. Secondly, I measure whether credit quality has become less important when Emergency Liquidity Assistance is outstanding:
𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑏𝑜𝑛𝑑 𝑦𝑖𝑒𝑙𝑑𝑖,𝑡− 𝐸𝑢𝑟𝑜 𝑠𝑤𝑎𝑝 𝑦𝑖𝑒𝑙𝑑𝑡 = 𝛽0+ 𝛽1(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽3𝐷2+ 𝛽4𝐸𝐿𝐴𝑖𝑡+ 𝛽5𝐷2(𝐶𝐷𝑆𝑖,𝑡 − 𝐶𝐷𝑆𝐴𝑉𝐸) +
𝛽6𝐷2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽7𝐸𝐿𝐴𝑖𝑡(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽8𝐸𝐿𝐴𝑖𝑡(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝜀𝑖,𝑡
where 𝐶𝐷𝑆𝑖,𝑡 Is the CDS spread in country i and period t, 𝐶𝐷𝑆𝐴𝑉𝐸 is the cross sectional average, 𝐿𝐼𝑄𝑖,𝑡 is the liquidity measure for country I and period t, and 𝐿𝐼𝑄𝐴𝑉𝐸 is the cross sectional average of the 𝐿𝐼𝑄𝑖,𝑡 variables. 𝐷1 is a dummy variable equal to 1 after the announcement of the ECB to lower collateral requirements on December 19th 2012, and equal to 0 before this date. 𝐸𝐿𝐴𝑖𝑡 is the amount of Emergency Liquidity assistance outstanding. Again, this model makes use of two control variables, which are (𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) and (𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸). These two variables control for credit quality (CDS) and market liquidity (LIQ), which are the two most important factors known influencing bond yields.
Firstly, I expect to obtain a significant relationship between the sovereign yield spread and Dummy 2 and/or 𝐸𝐿𝐴𝑖𝑡 and/or 𝐷2(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) and/or 𝐸𝐿𝐴𝑖𝑡(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸). For the dummy variable, which is equal to 1 after December 12th 2012, I expect a negative and significant coefficient. This would induce that after December 12 2012, bond yield spreads were driven down, which I believe would be due to the lower collateral requirements. As explained earlier, I expect the ELA amount coefficient the be significant and negative, inducing the higher the amount of ELA outstanding, the lower the bond yield spread. When it comes to testing hypothesis 2, I expect the 𝐷2(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) coefficient to be significant and negative, meaning that after December 12 2012, credit quality played a smaller role in the determination of bond yield spreads Lastly, I expect the 𝐸𝐿𝐴𝑖𝑡(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) coefficient to be significant and negative as well, which would mean that the more ELA is outstanding, the lower the importance of credit quality of the country using Emergency Liquidity assistance.
(2)
17
4. Data
This section highlights first of all how data has been collected. Next, I offer insights on how various variables and variable proportions have been constructed. Lastly, I present the most important summary statistics.
4.1 Data collection
First of all, I will collect data for eight European countries, both periphery and core member of the Eurozone (Belgium, Cyprus, France, Germany, Greece, Ireland, Italy and the Netherlands) between January 2008 and April 2015. This data will include coupon paying government bonds with remaining maturities of approximately 5,7 and 10 years, in order to calculate the government bond yields. Next, I obtain Euro-swap data for each day in the sample and for each maturity, in order to determine the government bond yield spread. Furthermore, I collect CDS and liquidity data for each maturity, country and day. Next, I have collected data on outstanding amounts of Emergency Liquidity Assistance. This data is available on the balance sheets of the participating countries. For Cyprus and Ireland, ELA was listed under “other items’’ until April 2012. ELA was listed under “sundry items” on the Greek balance sheet until April 2012. All countries moved outstanding ELA to the “Other Claims on Euro Area Institutions” on their respective balance sheets in April 2012. Emergency Liquidity Assistance used by the Belgian government has been extracted from the “Other Claims on Euro Area Institutions” item on the balance sheet of the Central Bank of Belgium. This data can be found in the appendix under Table 2.
4.2 Variable construction
Next, I have constructed the variables necessary for testing the two hypotheses. I start by separating bond prices based on the remaining maturity (5,7 and 10 years) and issuing country. The reason for the choice of these particular maturities this is that both CDS data and yield curve benchmark data are explicitly quoted for these maturities. Bond yields are determined for each maturity, each country and each day, by using the formula of the adjusted current yield.
Adjusted current yield= ([𝐴𝑛𝑛𝑢𝑎𝑙 𝐶𝑜𝑢𝑝𝑜𝑛
𝑀𝑎𝑟𝑘𝑒𝑡 𝑝𝑟𝑖𝑐𝑒 ] × 100) +
(100−𝑚𝑎𝑟𝑘𝑒𝑡 𝑝𝑟𝑖𝑐𝑒) 𝑌𝑒𝑎𝑟𝑠 𝑡𝑜 𝑚𝑎𝑡𝑢𝑟𝑖𝑡𝑦
18
Once this is determined, I set a benchmark in order to create the bond yield spread. For this, I use a method similar to Hull, Predescu, and White (2004), who make use of the Euro-swap yield. The advantages of using this benchmark are that the instrument is very liquid, caries relatively little counterparty risk and provides quotes for 5,7 and 10 year maturities. Our yield spreads are then calculated by subtracting the Euro-swap yield from the sovereign bond yield for each day, each country and each maturity. I will use credit default swaps spreads as a measure for bond quality. In order to measure bond liquidity, I will use the daily average bid-ask spread.
4.3 Coefficient proportions
Coefficient proportions are constructed as soon as the regression results are known. These proportions help explain the contribution that Liquidity, Credit quality and ELA have on sovereign yield spreads of each country. The contributions of all variables are constructed by taking the average value of the variable for each country and maturity and by multiplying this value with the variable coefficient estimate. Next, the proportions can be calculated by dividing the absolute value of the respective contribution by the sum of the absolute value of all variables. These results supply additional information about the relation between yield spreads, credit quality, liquidity and ELA, since the proportion results provide the weight and relative impact of the variables on yield spreads of the different countries in the sample. The formulas for determining the coefficient proportions and contributions can be found in the appendix.
4.4 Summary statistics
The most important summary statistics are presented in Table 2, which are extensively discussed in the paragraph below. Table 11 in the appendix offers an overview of the entire sample, including Finland and Spain. A maximum of 1912 observations per participating country are collected, which equals the number of trading days between January 2008 and April 2015.
Firstly, adjusted bond yield spreads of the eight participating countries are being described. Bond yields of countries having used Emergency Liquidity Assistance are among the highest in Europe (except Belgium). Greek, Cypriot, Spanish and Irish bonds have offered the highest yields over the past 5 years. German, French and Dutch government bonds offered the lowest yields. This is illustrated by the fact that 5 year
19
Greek government bonds have offered a maximum annual adjusted yield spread of 51.79%, while Dutch 5 year government bonds offered a maximum adjusted yield spread of 2.71%. Also, multiple adjusted current bond yields are negative on various days, inducing that investors lose money on the respective bond. Furthermore, for German, Greek and Dutch government bonds, yields are higher for bonds with a shorter maturity (5 years) than bonds with a longer maturity (10 years). For the other countries, government bond yields increase with their maturity. Lastly, as would be expected, standard deviations of countries having used Emergency Liquidity Assistance are among the highest (except Belgium), inducing that returns on these bonds have been very volatile over the past 7 years.
Secondly, It is important to highlight Credit Default Swap spreads. These spreads, which are in fact annual premiums, can be interpreted as the higher the value, the less the quality of the underlying government bond. Again, Greek and Cypriot CDSs’ where among the most expensive in Europe (highest perceived chance of default), followed by Ireland and Italy. German, French and Dutch government bonds have the lowest CDS premiums and thus have the highest perceived bond quality. Furthermore, CDS premiums increase with maturity of the underlying government bond, implying that bonds with longer maturity are riskier than short-term bonds (except for Cypriot government bonds). Lastly, it is important to notice standard deviations of CDS premiums in the past 7 years. Again, Greek, Cypriot and Irish government bonds have the most volatile CDS premiums. Belgium, French, German and Dutch CDS premiums standard deviations were among the lowest in the sample.
Next, liquidity measures in the third Table, which are in fact Bid-Ask spreads, provide information about liquidity of government bonds for each country and maturity over the past 7 year. I firstly remark that liquidity was worst for countries having used Emergency Liquidity Assistance (except Belgian 5 year government bonds), with spreads between 0.21 Eurocents and 1.37 Euro. This implies that it was most costly to trade these bonds. German, French and Dutch government bonds were among the most liquid bonds, with spreads between 0.05 cents and 0.20 cents
Again, countries that have used ELA had highly volatile bid-ask spreads, with a maximum up to 48 euro (equal to 4,8% on a 1000 Euro face value), which can be considered as extremely high. Lastly, it should be noted that several observations have been deleted since they had negative bid-ask spreads, which in practice is not possible.
20
Next, I analyze average outstanding Emergency Liquidity amounts. For Belgium, an average amount of 2.17 billion euros has been outstanding between 2008 and 2015, which is relatively low compared to its peers. Furthermore, it should be noted that Belgium has only used ELA until September 2013, due to problems with Dexia and Fortis. Cyprus had an average amount of 4.64 billion euros outstanding, but it should be noted that the GDP of Cyprus is much lower than its peers. Cyprus only started in October 2011 using ELA, but is still using the support program. As to Greece, an average amount of 24.43 billion euros has been outstanding, which is the highest of all participants. Furthermore, it should be noted that Greece has increased its Emergency Liquidity Assistance since January 2015. Lastly, Ireland has had an average amount of 21.16 billion euros outstanding, but Ireland terminated the support program in February 2013.
Lastly, I analyze correlations between the most important variables of this study. First of all, yield spreads are positively correlated with CDS premiums, Bid-Ask spreads and the amounts of ELA outstanding. Furthermore, opposed to findings obtained by Beber et al. (2009), credit quality and bond liquidity are positively correlated, with a correlation coefficient between 0.33 and 0.55. Lastly, Emergency Liquidity Assistance is positively correlated with credit quality and bond liquidity.
21
Panel A: Correlation Table for 5 year government bonds
Yield spreads CDS diff LIQ diff ELA Yield spreads 1
CDS diff 0.8883 1
LIQ diff 0.2860 0.3311 1
ELA 0.7316 0.6425 0.2023 1
Panel B: Correlation Table for 7 year government bond
Yield spreads CDS diff LIQ diff ELA Yield spreads 1
CDS diff 0.8897 1
LIQ diff 0.4165 0.4412 1
ELA 0.7487 0.6606 0.3368 1
Panel C: Correlation Table for 10 year government bond
spreads Yield CDS diff LIQ diff ELA Yield spreads 1
CDS diff 0.9118 1
LIQ diff 0.5062 0.5455 1
ELA 0.7580 0.6845 0.4218 1
Table 1: Correlation Tables for 5, 7 and 10 year maturities. The correlations are
computed by Stata, for the period between January 2008 and April 2015. The highlighted variables are the government bond yield spreads, computed by subtracting the Euro swap yield from the various government bond yields, the difference between the various CDS spreads and the cross-sectional average and the difference between the various bid-ask spreads and the cross-sectional average
22
Table2: Summary statistics (adjusted bond yields spreads, bid-ask spreads ,CDS premiums and Emergency Liquidity Assistance
respectively). The summary statistics are presented in 4 panels where the various maturities have been separated. Data is retrieved from Datastream en computed by Stata for the period between January 2008 and April 2015.
Panel A: Government bond yield spreads
5 year 7 year 10 year
Country Observations Mean Std. Dev. Min Max Mean Std. Dev. Min Max Mean Std. Dev. Min Max Belgium 1912 1.5230 2.1277 -3.2301 5.7497 1.5145 2.1919 -3.4879 5.7087 1.7163 1.5295 -2.3398 5.1590 Cyprus 1902 7.4379 5.2084 1.9235 21.0449 7.6599 4.0165 2.9102 18.9343 7.5440 4.8744 2.2750 19.6309 France 1912 1.3344 1.8524 -2.6128 4.9444 0.8601 2.2723 -5.3890 4.7837 1.9189 1.7230 -2.1662 4.9897 Germany 1912 0.9393 1.8638 -3.2597 4.9080 0.8829 1.9786 -3.3015 4.8591 0.4779 1.7889 -3.5080 3.6327 Greece 1486 18.8595 13.1067 2.2505 51.7874 15.8231 11.8161 1.5656 49.3937 14.7137 10.5678 4.1044 42.1763 Ireland 1500 2.2907 4.9232 -7.0127 19.6013 3.3415 4.2354 -3.9788 16.7913 4.7631 3.3183 -1.7055 14.3738 Italy 1496 2.2899 1.7915 -0.8467 8.2428 2.5250 2.0788 -2.1938 7.8851 3.2151 1.7816 -1.3728 7.2907 Netherlands 1420 0.2954 1.3632 -3.0441 2.7147 0.1003 1.7207 -4.1159 2.9001 0.0209 1.4723 -3.5214 2.6552 Panel B: Bid-Ask spreads (liquidity measure)
5 year 7 year 10 year
Country Observations Mean Std. Dev. Min Max Mean Std. Dev. Min Max Mean Std. Dev. Min Max Belgium 1912 0.1273 0.1266 0.0195 2.7200 0.2067 0.1420 0.0200 1.0032 1.3647 0.7731 0.2650 4.9819 Cyprus 1902 1.1328 0.3625 0.3750 2.0000 1.2059 0.3523 0.7500 3.0000 1.1100 0.5401 0.4400 4.0000 France 1912 0.0924 0.0597 0.0233 0.3494 0.1039 0.0636 0.0213 0.4006 0.1407 0.0823 0.0447 0.5727 Germany 1912 0.0513 0.0283 0.0174 0.1838 0.0391 0.0227 0.0147 0.2576 0.1851 0.0648 0.0600 0.4477 Greece 1486 0.8035 1.3425 0.0288 35.2800 0.7341 0.3374 0.1180 1.0000 2.6945 1.1637 0.7500 5.0000 Ireland 1500 0.7893 0.4162 0.0187 2.0000 0.8259 1.4477 0.0200 48.0000 1.0964 0.9083 0.0400 10.0000 Italy 1496 0.1612 0.1154 0.0010 0.8000 0.1647 0.1213 0.0300 2.0000 0.1974 0.1350 0.0500 1.7680 Netherlands 1420 0.0649 0.0339 0.0249 0.2310 0.0856 0.0413 0.0100 0.2396 0.2396 0.0914 0.0700 0.5567
23 Panel C: CDS premiums (credit quality measure)
5 year 7 year 10 year
Country Observations Mean Std. Dev. Min Max Mean Std. Dev. Min Max Mean Std. Dev. Min Max Belgium 1912 77.5447 63.2506 14.5000 341.9800 83.2519 59.7626 16.6000 333.7100 88.8155 56.5002 19.7500 324.5398 Cyprus 1902 498.4819 441.3570 17.0000 1674.2200 480.8723 413.4923 19.8000 1590.0600 454.3305 373.7533 24.0000 1461.8300 France 1912 50.6739 33.9904 6.0000 171.5600 59.9015 33.3890 8.6250 176.0300 66.9384 33.9428 11.0250 181.3600 Germany 1912 26.0588 16.9028 5.2000 92.5000 31.1879 16.6473 6.4000 92.2416 35.6784 16.7158 8.7500 91.9800 Greece 1486 960.3223 587.9042 145.0000 2458.6400 1006.6190 620.7091 139.0000 2557.9800 1054.2680 661.7234 135.0000 2647.6600 Ireland 1500 260.6384 187.4091 67.0500 1019.8390 271.8683 217.1012 51.1200 1114.2750 275.9762 244.0525 35.7200 1191.1580 Italy 1496 184.4432 108.2781 48.0000 498.6599 193.6071 101.4039 49.2000 480.6599 198.3941 93.4215 51.0000 468.1858 Netherlands 1420 34.5929 21.7238 8.7000 122.9100 51.9884 39.0440 14.7500 193.4600 48.4400 21.2495 20.8900 140.6200
Panel D: Emergency Liquidity Assistance amounts (in billions of euros)
Country Observations Mean Std. Dev. Min Max Belgium 1912 2.2204 3.8409 0.0000 23.2050 Cyprus 1902 4.0089 4.4308 0.0000 11.4004 France 1912 0.0000 0.0000 0.0000 0.0000 Germany 1912 0.0000 0.0000 0.0000 0.0000 Greece 1486 21.1982 35.4860 0.0000 124.0847 Ireland 1500 18.9253 20.7164 0.0000 70.0680 Italy 1496 0.0000 0.0000 0.0000 0.0000 Netherlands 1420 0.0000 0.0000 0.0000 0.0000
24
5 Empirical Results
This section presents results obtained from Hypothesis 1 and 2. Both results are thoroughly discussed in the two sub-sections.
5.2 Results Hypothesis 1
We start by investigating whether bond yields of participating countries have been driven down due to Emergency Liquidity Assistance. The results can be found in Table 2, where the general regression can be found (1), a regression without countries having used ELA (2) and a regression with only ELA countries (3). For this, I used the next regression:
𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑏𝑜𝑛𝑑 𝑦𝑖𝑒𝑙𝑑𝑖,𝑡− 𝐸𝑢𝑟𝑜 𝑠𝑤𝑎𝑝 𝑦𝑖𝑒𝑙𝑑𝑡 = 𝛽0+ 𝛽1(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽3𝐷1+ 𝛽4𝐸𝐿𝐴𝑖𝑡+ 𝜀𝑖,𝑡
Firstly, I analyze the results obtained with all countries in the sample, for which 13,540 observations are collected. Standard errors are presented in parentheses, and all but one coefficient are significant at a 5% significance level. The results reveal a value for R-squared of around 82%, which is considered high, inducing that the model has a very high explanatory power. Furthermore, as would be expected according to previous research such as Beber et al. (2009), the credit differential has a significant positive influence on bond yield spreads, which implies that a lower credit quality increases government bond yield spreads. The magnitude of the credit quality coefficient induces that a 100 basis points credit differential above the average results in an increase in the bond yield spreads of between 76 and 99 basis points.
The second coefficient I analyze is the liquidity differential. For 5 and 10 year government bonds, I find a positive significant relation between the bid-ask spread differential and bond yield spreads, inducing the higher the bid-ask spread, the higher the bond yield spread. These results are in line with previous research. In practice, these results would imply that a 1 Euro increase in Bid-Ask spread leads to a 143 and 809 basis points increase for respectively 5 and 10 year government bond yield spreads. Next, it is important to highlight the influence of Emergency Liquidity Assistance on bond yield spreads of countries in the sample. For both the ELA dummy (equal to 1 when a an amount of ELA is outstanding) and the ELA amount, I find a significant
25
positive relation between ELA and bond yields spreads, implying that if a country uses Emergency Liquidity Assistance, its yield spreads increase significantly. This result contradicts Hypothesis 1, which states that ELA decreases bond yields due to decreased collateral requirements. Lastly, the sign of the constant may seem surprising since it induces that sovereign yields may be below the Euro-Swap yield curve, but this is the consequence of having two different benchmarks on either side of the equation.
Results from the second regression (2), which only includes countries not having used ELA (France, Germany, Italy and the Netherlands), can also be found in Table 2. All coefficients are significant at a 5% significance level. Furthermore, R-squared varies between 0.58 and 0.74, which is very high, but lower than the entire sample including ELA countries. Again, credit quality differential coefficients are all significant at 1% and positive. The magnitude of the credit quality coefficient induces that a 100 basis points credit differential above the average results in an increase in the bond yield spreads of between 82 and 131 basis points, slightly higher than the entire sample. Liquidity differentials are again positively correlated with bond yield spreads. In practice, these results would imply that a 1 Euro increase in Bid-Ask spread leads to an increase up to 2769 basis points government bond yield spreads. As these countries have not used Emergency Liquidity Assistance, the ELA dummy and ELA amount coefficients cannot be used. Lastly, constant coefficients are again negative, as was explained in the previous section.
Results from the third regression (3), which only include observations from countries having used ELA, have again very high explanatory power (around 0.80), which is higher than the sample without ELA countries. The reason for this can be found in Table 4, where the contributions and proportions of each coefficient are presented. In this Table, I find that the amount of ELA outstanding counts for an average contribution of 20% of R-squared, implying a very high contribution. Credit differentials are again positive and significant. However, liquidity differentials are both positive and negative. Again, ELA dummy and ELA amount coefficients are all significant at a 5% significance level and positive, thus inducing that ELA leads to higher bond yield spreads. Constant variables are again all negative and significant.
Table 3 offers more specific details on regression (3), since all countries having used ELA are separated. R-squared values are relatively high for all countries, but the regression has the least explanatory power for Belgium (with R-squared between 0.40
26
and 0.79). This is probably due to the fact that the amount of ELA outstanding has less influence on Belgium’s yield spread than on the yield spreads of other participating countries (see Table 4). Again, as would be expected, credit differential coefficients are all positive. The magnitude of the credit quality coefficient induces that a 100 basis points credit differential above the average results in an increase in the bond yield spreads of between 75 and 123 basis points. The second coefficient I analyze is the liquidity differential. For most government bonds, I find a positive significant relation between the bid-ask spread differential and bond yield spreads, inducing the higher the bid-ask spread, the higher the bond yield spread. Several individual bonds show a negative relation. Furthermore, both the ELA dummy and the ELA amount coefficient positively influence yield spreads from Belgium, Greece and Ireland. Nevertheless, 5 year Cypriot government bonds yield spreads have been influenced negatively by the ELA dummy. This induces that several Cypriot government bonds had their yields decreased due to ELA. It is interesting to note that Cypriot government bonds were reduced to junk status in January 2012, but their bonds where thus eligible for ELA since December 2012.
Lastly, it is important to highlight the coefficient contributions and proportions. Proportions help explain the contribution that Liquidity, Credit quality and ELA have on sovereign yield spreads of each country. Credit differential proportions are usually around 50 to 70%, but this depends on the country. Some countries that do not use ELA have contributions around 95%, such as Germany and the Netherlands. The size of the credit differentials is comparable to results obtained by Beber et al. (2009). However, liquidity differential proportions are on average much smaller than the proportions obtained by Beber et al. (2009). This research obtains liquidity differentials that usually vary between 1 and 10%. Furthermore, countries having used Emergency Liquidity Assistance tend to have lower liquidity differential proportions than other countries. This can be explained by the fact that countries having used ELA have two other variables that help explain the bond yield spreads. The ELA dummy equal to 1 when any amount of ELA is outstanding is not contributing very much to the yield spreads, whereas the ELA amount variable accounts to a proportion of 20% on average.
Overall, I can thus conclude that ELA has not led to a decrease in bond yields. Nevertheless, 5 year Cypriot government bonds yield spreads were driven down by ELA.
27
The fact that bond yields were driven upwards instead of downwards could eventually be explained by the fact that Emergency Liquidity Assistance is not part of a single monetary policy and potential costs and risks should be borne by the national central bank in question. Another explanation could be that Emergency Liquidity Assistance is perhaps a proxy for the banking sector being at risk. Through contagion risk, it might pose a systematic threat to the country, which could justify the yield increase.
28
Table 3: Regression results for hypothesis 1 testing whether ELA has decreased bond yields of participating countries.
Regression (1) is the general regression with the selected sample, regression (2) includes only countries that have not used Emergency Liquidity Assistance and regression (3) only includes countries having used ELA. The credit differential is the difference between the CDS value at time t for country i and maturity T and the cross sectional average. The Liquidity differential is the difference between the the bid-ask spread at time t for country i and maturity T and the cross sectional average. The ELA dummy equals 1 when an amount of ELA is outstanding and 0 otherwise. The ELA amount variable is the amount of ELA outstanding at time t. *** stands for significance at a 1% level, ** for significance at a 5% level and % for significance at a 10% level. Standard errors are reported between brackets.
𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑏𝑜𝑛𝑑 𝑦𝑖𝑒𝑙𝑑𝑖,𝑡− 𝐸𝑢𝑟𝑜 𝑠𝑤𝑎𝑝 𝑦𝑖𝑒𝑙𝑑𝑡= 𝛽0+ 𝛽1(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽3𝐷1+ 𝛽4𝐸𝐿𝐴𝑖𝑡+ 𝜀𝑖,𝑡
(1) (1) (1) (2) (2) (2) (3) (3) (3)
Maturity 5 year 7 year 10 year 5 year 7 year 10 year 5 year 7 year 10 year Constant -1.2253*** -1.5266*** -1.6535*** -0.4774*** -1.4687*** -1.2868*** -1.1583*** -1.3261*** -0.8998*** (0.03428) (0.0320) (0.0276) (0.0257) (0.0295) (0.0195) (0.0799) (0.0701) (0.0624) Credit differential 0.0099*** 0.0076*** 0.0082*** 0.0082*** 0.0124*** 0.0131*** 0.0098*** 0.0075*** 0.0080*** (0.0001) (0.0001) (0.0001) (0.0002) (0.0002) (0.0002) (0.0002) (0.0001) (0.0001) Liquidity differential 0.0143** -0.0186 0.0809*** 2.7689*** 1.8062*** 1.4855*** 0.0097 0.1516** -1.1959*** (0.0048) (0.0412) (0.0246) (0.0696) (0.0640) (0.0271) (0.0684) (0.0572) (0.503) ELA dummy 0.2254** 1.3077*** 1.4389*** - - - 0.2898* 1.3604*** 1.7858*** (0.0847) (0.0730) (0.0688) - - - (0.1164) (0.0975) (0.0937) ELA amount 0.1624*** 0.1431*** 0.1109*** - - - 0.1619*** 0.1450*** 0.1113*** (0.0019) (0.0018) (0.0015) - - - (0.0026) (0.0024) (0.0021) Adjusted R-squared 0.8172 0.8162 0.8463 0.5758 0.5531 0.7450 0.7919 0.7848 0.8166 Observations 13540 13540 13540 6740 6740 6740 6800 6800 6800
29
Table 4: Regression results for hypothesis 1. The following Table contains regression results for separated countries (Belgium, Cyprus, Greece
and Ireland) having made use of Emergency Liquidity Assistance somewhere between 2 1st 2008 and April 30th 2015. The credit differential is the
difference between the CDS value at time t for country i and maturity T and the cross sectional average. The Liquidity differential is the difference between the the bid-ask spread at time t for country i and maturity T and the cross sectional average. The ELA dummy equals 1 when an amount of ELA is
outstanding and 0 otherwise. The ELA amount variable is the amount of ELA outstanding at time t. *** stands for significance at a 1% level, ** for significance at a 5% level and % for significance at a 10% level. Standard errors are reported between brackets.
𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑏𝑜𝑛𝑑 𝑦𝑖𝑒𝑙𝑑𝑖,𝑡− 𝐸𝑢𝑟𝑜 𝑠𝑤𝑎𝑝 𝑦𝑖𝑒𝑙𝑑𝑡= 𝛽0+ 𝛽1(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽3𝐷1+ 𝛽4𝐸𝐿𝐴𝑖𝑡+ 𝜀𝑖,𝑡
Belgium Belgium Belgium Cyprus Cyprus Cyprus Greece Greece Greece Ireland Ireland Ireland Maturity 5 years 7 years 10 years 5 year 7 years 10 years 5 years 7 years 10 years 5 years 7 years 10 years Constant -1.2274*** -2.6710*** -1.8640*** -1.5550*** -0.0116 -1.3535*** 2.7935*** 0.2976 3.8852*** -3.2692*** -3.0691*** -0.5557*** (0.0706) (0.0432) (0.0367) (0.1384) (0.0999) (0.0796) (0.3627) (0.3595) (0.2972) (0.1317) (0.0575) (0.0390) Credit differential 0.0059*** 0.0072*** 0.0082*** 0.0099*** 0.0074*** 0.0105*** 0.0076*** 0.0048*** 0.0049*** 0.0189*** 0.0117*** 0.0069*** (0.0003) (0.0002) (0.0003) (0.0001) (0.0001) (0.0001) (0.0003) (0.0003) (0.0002) (0.0007) (0.0003) (0.0002) Liquidity differential 2.1064*** 0.6421*** -0.0782** 1.3066*** 0.2141 0.01744 0.1905 3.5747*** -2.2389*** -4.4694*** 0.1418*** 0.2042*** (0.1466) (0.0852) (0.0275) (0.2198) (0.1341) (0.1257) (0.1281) (0.5427) (0.1393) (0.1802) (0.0244) (0.0375) ELA dummy 0.7996*** 1.6563*** 0.1812*** -1.4252*** 3.6584*** -0.1203 2.1059*** 4.2361*** 4.8691*** 0.2987 1.6941*** 0.8252*** (0.0568) (0.0409) (0.0412) (0.2870) (0.2277) (0.2987) (0.3939) (0.4250) (0.3731) (0.2196) (0.1034) (0.0647) ELA amount 0.0240*** 0.0331*** 0.0279*** 0.4055*** -0.0495* 0.3701*** 0.2086*** 0.1784*** 0.1518*** 0.0352*** 0.0096** 0.0253*** (0.0059) (0.0044) (0.0046) (0.0293) (0.0226) (0.0286) (0.0058) (0.0057) (0.0043) (0.0076) (0.0036) (0.0022) Adjusted R-squared 0.5221 0.7394 0.3955 0.8717 0.8814 0.8593 0.8059 0.7742 0.8486 0.7026 0.8955 0.9290 Observations 1912 1912 1912 1902 1902 1902 1486 1486 1486 1500 1500 1500
30
Table 5: Contribution and proportion results for all variables used in Hypothesis 1. Contributions are calculated by multiplying the regression coefficient estimate by the average variable value. Proportions are then calculated by dividing the absolute contribution by the sum of absolute contributions. The credit differential is the difference between the CDS value at time t for country i and maturity T and the cross sectional average. The Liquidity differential is the difference between the bid-ask spread at time t for country i and maturity T and the cross sectional average. The ELA dummy equals 1 when an amount of ELA is outstanding and 0 otherwise. The ELA amount variable is the amount of ELA outstanding at time t
5 year 7 year 10 year Country Credit Liquidity D (ELA>0)
ELA
amount Credit Liquidity D
(ELA>0) ELA amount Credit Liquidity D (ELA>0) ELA amount Belgium Contribution -1.82 -0.04 0.17 0.36 -1.44 0.00 1.01 0.32 -1.55 -0.39 1.11 0.25 Proportion 0.76 0.02 0.07 0.15 0.52 0.00 0.36 0.11 0.47 0.12 0.34 0.07 Cyprus Contribution 2.35 0.10 0.11 0.65 1.58 -0.01 0.64 0.57 1.45 -0.19 0.70 0.44 Proportion 0.73 0.03 0.03 0.20 0.56 0.01 0.23 0.20 0.52 0.07 0.25 0.16 France Contribution -2.09 -0.04 0.00 0.00 -1.62 0.01 0.00 0.00 -1.73 0.60 0.00 0.00 Proportion 0.98 0.02 0.00 0.00 1.00 0.00 0.00 0.00 0.74 0.26 0.00 0.00 Germany Contribution -2.33 -0.05 0.00 0.00 -1.83 0.01 0.00 0.00 -1.99 0.56 0.00 0.00 Proportion 0.98 0.02 0.00 0.00 1.00 0.00 0.00 0.00 0.78 0.22 0.00 0.00 Greece Contribution 6.92 0.06 0.12 3.44 5.58 -0.01 0.68 3.03 6.37 -1.47 0.75 2.35 Proportion 0.66 0.01 0.01 0.33 0.60 0.00 0.07 0.33 0.58 0.13 0.07 0.21 Ireland Contribution -0.01 0.06 0.16 3.07 0.00 -0.01 0.91 2.71 -0.02 -0.18 1.00 2.10 Proportion 0.00 0.02 0.05 0.93 0.00 0.00 0.25 0.75 0.00 0.05 0.30 0.64 Italy Contribution -0.76 -0.03 0.00 0.00 -0.60 0.00 0.00 0.00 -0.65 0.55 0.00 0.00 Proportion 0.96 0.04 0.00 0.00 0.99 0.01 0.00 0.00 0.54 0.46 0.00 0.00 Netherlands Contribution -2.25 -0.05 0.00 0.00 -1.68 0.01 0.00 0.00 -1.88 0.52 0.00 0.00 Proportion 0.98 0.02 0.00 0.00 1.00 0.00 0.00 0.00 0.78 0.22 0.00 0.00
31
5.3 Results Hypothesis 2
I start by investigating whether credit quality perception of countries having used ELA countries has been driven down due to Emergency Liquidity Assistance. The results can be found in Table 5, where the general regression can be found (1), a regression without countries having used ELA (2) and a regression with only ELA countries (3). Table 6 elaborates the ELA countries separately. The proportions and contributions of each variable can be found in Tables 7, 8 and 9. I test hypothesis 2 with the following regression.
𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑏𝑜𝑛𝑑 𝑦𝑖𝑒𝑙𝑑𝑖,𝑡− 𝐸𝑢𝑟𝑜 𝑠𝑤𝑎𝑝 𝑦𝑖𝑒𝑙𝑑𝑡= 𝛽0+ 𝛽1(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽3𝐷2+ 𝛽4𝐸𝐿𝐴𝑖𝑡+ 𝛽5𝐷2(𝐶𝐷𝑆𝑖,𝑡 − 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽6𝐷2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝛽7𝐸𝐿𝐴𝑖𝑡(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) + 𝛽8𝐸𝐿𝐴𝑖𝑡(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸) + 𝜀𝑖,𝑡
Firstly, I analyze the results obtained with all countries in the sample, for which 13540 observations are collected. Standard errors are presented in parentheses, and all but 3 coefficients are significant at a 5% significance level. It is important to firstly highlight the relatively high explanatory power of the model with all countries included, since the value of R-squared is equal to 0.85 to 0.89.
As earlier highlighted for Hypothesis 1 and in line with Beber et al. (2009), the credit differential coefficient is significant and positive for all three maturities. These results are in line with our expectations, and induce that the higher the quality of a particular government bond, the lower its yield spread. The liquidity differential, also as earlier seen in hypothesis 1, has a significant and positive significant relationship with government bond yield spreads, which is also in line with Beber et al. (2009). The meaning of this sign is that the higher the bid-ask spread (or the higher the illiquidity), the higher the bond yield spread. Next, I analyze the 19 December 2012 dummy. The coefficient is significant at a 5% level and negative for all maturities. This means that after December 12 2012, European government bond yield spreads have significantly decreased. This is probably (mostly) due to the calmer economic climate after 2012 instead of the lower collateral requirements for ELA. The ELA amount variable significant and positive for all maturities, as earlier seen in hypothesis 1, which leads to
32
the conclusion that the more Emergency Liquidity Assistance a country is using, the higher its yield spreads. The interaction variable 𝐷2(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) measures whether credit quality has played a less important role in the determination of yield spreads after December 12 2012. The coefficient is significant at a 5% level and positive for all maturities. Furthermore, in Table 7, 8 and 9, the proportions of this variable are shown, which vary between 10 and 20%. The economic meaning of this coefficient is that instead of becoming less important, credit differentials played a more important role in the determination of bond yield spreads after December 12 2012. It thus seems that since the introduction of lower collateral requirements for obtaining Emergency Liquidity Assistance, financial markets have put more emphasis on credit quality. Another interaction variable, 𝐷2(𝐿𝐼𝑄𝑖,𝑡− 𝐿𝐼𝑄𝐴𝑉𝐸), which measures liquidity differences before and after December 2012, does not provide unanimous answers. This is not surprising, as I did not expect liquidity to have heavily changed. Next, I highlight the 𝐸𝐿𝐴𝑖𝑡(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) variable, which measures whether the amount of ELA has influenced the importance of credit quality in the determination of bond yield spreads. The coefficient found is significant and positive for all three maturities, which is in contradiction with the expectations. The meaning of this coefficient sign is that the more Emergency Liquidity Assistance is outstanding, the more important credit quality is in the determination of bond yield spreads. However, according to Table 7,8 and 9, it should be noted that the contribution of this variable is very limited (0 to 13% for the 4 countries having used ELA). Lastly, I observe whether liquidity was influenced by the amount of ELA outstanding. Again, I do not find a unanimous answer, which is in line with expectations.
Results from the second regression (2), which only includes countries not having used ELA (France, Germany, Italy and the Netherlands), can also be found in Table 5. Furthermore, R-squared varies between 0.57 and 0.81, which is very high, but lower than the entire sample including ELA countries. Only the important regression coefficients will be discussed. Firstly, the 19 December 2012 dummy coefficient is again significant and negative (except for 5 year bonds). This leads to think that bond yields of all countries, and not only countries having ELA, have decreased after December 2012. Next, the interaction variable 𝐷2(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) does not provide a unanimous sign.
33
For bonds with a 5 year maturity, the sign is significant and negative, which is line with our expectations (credit quality is less important in the determination of yield spreads since December 19 2012). However, for bonds with a 7 and 10 year maturity, I find a significant and positive sign, inducing that credit quality has only become more important.
Results from the third regression (3), which only include observations from countries having used ELA, have again a very high explanatory powers (around 0.85), which is higher than the sample without ELA countries. The reason for this can be found in Tables 7, 8 and 9, where the contributions and proportions of each coefficient are presented. In these Tables, the importance of ELA in the explanation of bond yield spreads is highlighted, since the 3 variables including ELA explain a relatively high proportion of bond yield spreads. Again, as seen earlier and as expected, credit and liquidity differentials coefficients are all positive and significant (except the 10 year liquidity differential coefficient). For the 19 December dummy, I find a negative coefficient for 5 and 7 year bonds, but a positive coefficient for 10 year bonds, meaning that short term bonds have had decreasing yield spreads after December 2012, and long term bonds have had increasing yield spreads. This could be due to uncertainty for the long-term. Next, the amount of ELA is again positively influencing bond yield spreads for all maturities, which means countries have increased bond yields for taking on Emergency Liquidity Assistance. Furthermore, the interaction variable 𝐷2(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) is again significant and positive for all three maturities, which is contradicting to hypothesis 2. The results induce that credit quality has only become more important since December 19 2012. Lastly, the interaction variable 𝐸𝐿𝐴𝑖𝑡(𝐶𝐷𝑆𝑖,𝑡− 𝐶𝐷𝑆𝐴𝑉𝐸) is observed. As in regression (1), the coefficient is significant and positive. However, its contribution is relatively small (even equal to 0% for Ireland). Nonetheless, these results are also contradicting hypothesis 2, since the results imply that the importance of credit quality in the determination of bond yield spreads, have increased when ELA increased.
Next, regression results from Table 6, which provide details on separated ELA countries, are discussed. These results offer more specific details on regression (3), since all countries having used ELA are separated. R-squared values are again relatively high, and varying between 0.49 and 0.85. Almost all (except 3) credit and liquidity differentials are significant at a 5% level and positive. Next, the December 12 dummy is
