The ECB’s QE and bailout events : the effect on government bond yields in Eurozone countries

(1)

1

Amsterdam Business School

MSc Business Economics: Finance track

Master Thesis

The ECB’s QE and bailout events: the effect on government bond

yields in Eurozone countries

By: Paulius Morkūnas

ID

:

11084030

Date: 2016 July 7

(2)

2 Statement of originality

This document is written by Paulius Morkūnas, 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)

3

This research paper evaluates the effects of the ECB’s Quantitative Easing (QE) and bailout events on the Eurozone countries’ government bond yields. Government bond yields are important for different groups of investors such as central banks, commercial banks and governments. Thus, it is important to analyze the possible factors which affect these yields and to assess the effectiveness of monetary policy. One of the most important QE programme is Public Sector Purchase Programme (PSPP), which was supposed to be finished

in September 20161_{but has been extended up to March 2017}2_{). This programme has had an}

important effect on government bond yields throughout the Eurozone. The reason for this has been a huge money provision through outright purchases of this asset class. As a result, government bond yields have reacted in both stable and non-stable economies throughout the Eurozone. This implies that the programmes implemented by the ECB may have different outcomes to the government bond yields. One could be that government bond yields’ drop below 0, in the more stable economies. Another could be described as a rapid rise in government bond yields, in the less stable economies. However, that would lead to a turmoil in the Eurozone as this union shares the same currency. In this thesis, analysis of the QE and bailout events’ effects on the government bond yields is made. For the extensional contribution to the topic, the Eurozone’s countries have been grouped by economic stability. Furthermore, PSPP programme has been analyzed in detail. This analysis concludes that different stability economies react similarly to bailout and QE events (whether taken together or separately), the only difference is the direction of the change in government bond yields. Also, it shows that the announcement of PSPP has made a higher influence to the government bond yields, than the actual implementation of the programme (which has been approved by two different analyses). As a result, possible PSPP announcements in the future, may have a huge impact on the government bond yields of the Eurozone countries.

1_{European Central Bank press release, 2015. ‘‘ECB announces expanded asset purchase programme’’.} 2_{European Central Bank, 2016. ‘‘Implementation aspects of the public sector purchase programme (PSPP)’’.}

(5)

5

1. Introduction

The recent economic crisis has evoked a lot of financial turmoil. As a result, this has made an emphasis on the poor expertise in situations, when the world is facing high financial distress. In order to make the worst outcomes of the crisis smoother, there has been numerous decisions made and policies applied by governments and central banks. For example, Federal Reserve (FR) in the US and the Bank of England (BoE) in the UK have applied unconventional monetary policies during the crisis, in order to stimulate the

economy and meet economic targets such as the inflation rate5₍_{Rogers et al. (2014),}

Christensen and Rudebusch (2012)).

However, because most of the studies, in terms of QE, have been done about FR, BoE and Bank of Japan (BoJ), the key topic of this research is about the European Central Bank (ECB) (D’Amico et al., (2012), Doh (2010)). The argument for this choice is that there still is a big niche for conducting analysis in a field of unconventional monetary policy in the Eurozone. Also, ECB’s QE schemes are recently conducted (compared to the UK, US or Japan). For example, Altavilla et al. (2015), assess the impact of the APP (Asset Purchase Programme) on asset pricing of some of the Eurozone countries, during the low financial distress. Furthermore, that paper concludes that APP has lowered yields in some of the market segments. Also, some spill-overs have been found over the non-targeted assets. However, this research will talk about the government bond yields of Eurozone countries and the impact from QE and bailout events on them. The combination of these events allows inspecting on possible connection among them. For example, to check whether the effectiveness of the QE events is hampered by bailout events. Furthermore, in order to find more detailed results of QE and bailout events’ impact on government bond yields, Eurozone countries are analyzed as a whole, in groups and separately. Szczerbowicz (2012) distinguishes between three main tools of quantitative easing (QE) by the ECB: exclusive liquidity measures, flexibility in collateral acceptance and assets’ purchasing. This thesis will target the latter one because it has been implemented at a high extent, while conducting the most recent QE programmes in the Eurozone. The easiest way to see why this tool is

5_{Rogers et al. (2014) examines and compares QE events’ effects on bond yields, exchange rates and stock}

prices for Bank of Japan, ECB, Federal Reserve and Bank of England. The paper sums up that QE is effective and concludes that the movement in long-term yields can be predicted after QE announcement in a long-run.

(6)

6

dominating among the other QE initiatives, is to put it into numbers. For example, the

current APP programmes: PSPP, CBPP3, ABSPP and CSPP account for €80 billion per month6

(of which PSPP accounts for €60 billion per month).

Thus, the analysis of PSPP programme is crucial in this thesis. It allows investigating on the differences between PSPP announcement and actual implementation. That is because PSPP’s future is ambiguous. For example, it has recently been extended until 2017 March (even though initially it was supposed to be finished in 2016 September). Thus, new announcements, such as extension or termination, will possibly be reported before 2017 March.

Another objective is broader and allows to inspect changes in the Eurozone countries’ government bond yields in detail. For example, Altavilla et al. (2015) in their approach derive a conclusion that the higher the yield, the more it declines because of the APP’s announcements. In addition to that paper, this thesis looks at the QE announcements and bailout events’ implications on different economies’ yields. The Eurozone countries are divided into two groups according to the government indebtedness (stable vs. non-stable). This measure can be used as a tool for determining country’s stability (Cole and Kehoe (2000)). In the beginning, graphical comparison of both groups of countries has been made (Exhibit 1). It allows observing primary differences in government bond yields of Eurozone countries.

Al in all, the results of these objectives will allow answering two important questions of this thesis:

1) Is there a different effect of QE and bailout events between stable and non-stable Eurozone countries?

2) Was the PSPP announcement effect stronger on government bond yields or the actual implementation?

Answering these questions could be helpful for the investors, central banks and other stakeholders because some of the events/announcements may repeat in the near

(7)

7

future. As a result, it would be possible to make an investment decision before the event has happened. For instance, governments may be interested in this research as they can issue debts cheaper because of the PSPP. If, for example, a new announcement of PSPP programme appears, cheaper funding could be provided after the announcement, as yields are expected to fall because of it. Finally, this research could help for the central banks to see, whether the implementation of the PSPP programme has been effective or not.

It is important to mention that, even though the topic is widely prevalent, this thesis contributes to the existing literature in a few ways. Firstly, there is a great deal of attention paid on the PSPP programme. As it is a recently conducted programme, not much about it has been done so far. The division of implementation and announcement effects is exclusive. Secondly, the range of time is unique and the set of events is specific. Combining these facts with a choice of grouping countries, makes the research unique. This study also conducts a country level statistical analysis based on time-series as a robustness check which is unique in the academic literature with regard to this specific topic.

Structure of the rest of this thesis is as follows: in Section 2 the history of monetary policy and QE events are discussed and described. Where the biggest events, related to QE carried by the ECB are described. In Section 3, more arguments about the thesis, its methodology and relevance are argued as a literature review. In addition, hypotheses are provided and explained. Sample selection and data sources are provided in Section 4. All the models used for the thesis are explained in detail under the Section 5 and the outcomes of these are presented in Section 6 as ‘‘Results’’. Additionally, Section 7 includes a detailed discussion about robustness checks (as time-series analysis). Finally, Section 8 draws a conclusion about the whole thesis.

2. History of the monetary policy in Eurozone

2.1 Conventional vs unconventional monetary policies

In simple words, there are two kinds of monetary policies: conventional and unconventional. In order to analyze unconventional policy’s instruments and tools further, it is necessary to distinguish between these two kinds of monetary policy. Conventional monetary policy includes strategies of increasing or decreasing targeted interest rates,

(8)

8

reserve limits of the bank and the money supply in the economy. On the other hand, unconventional monetary policy includes strategies of how to increase demand rapidly and

stimulate economic growth11_{. However, monetary policy’s biggest disadvantage is that the}

implementation takes a lot of time, while decision about the new initiatives can be agreed upon much faster.

Even though the unconventional monetary policy’s practice has become commonly applicable all over the world, some obstacles or mismatches with the general rules, because of its outcomes, may occur. For example, one mismatch is that in general, Eurozone’s central banks are not allowed to print money directly for their country due to the common rules of the ECB. On the other hand, conducting the PSPP programme means increased money supply for each of the Eurozone countries. So, as long as the PSPP takes place, it becomes at odds with the ECB’s policy about money printing. Moreover, unconventional monetary policy provides low costs of funding. As a result, that makes it easier to borrow for the governments and this may result to financial turmoil if the government starts borrowing inappropriately. Also, regulators must set up some new positions, in order to supervise such operations, so that the governments would comply with macro prudential policies. Finally, when borrowing and lending has become extremely intense, some of the assets, used in one or another programme (e.g. SMP – for explanation see the next paragraph) may start being on periphery. Thus, conducting unconventional monetary policy is not only beneficial but also requires intense supervision and analysis of numerous processes.

2.2 Unconventional monetary policy events in Eurozone

As mentioned above, focus of this thesis is on assets purchasing. In order to better understand the QE announcements and programmes, the APP has to be described. It is an asset purchase tool of unconventional monetary policy. To be specific, central banks who implement this programme, stimulate borrowing and private spending by encouraging commercial banks to lend cheaper to their customers. As a result, inflation is expected to

return to the target which is close to 2% in medium-term, in the Eurozone12_{. This tool is}

extremely useful nowadays in the Eurozone, as there prevails a high level of threat in a form

11_{Definitions and comparisons between conventional and unconventional monetary policies have been taken}

from Bini Smaghi, (2009) ‘‘Conventional and unconventional monetary policy’’ speech.

(9)

9

of deflation, which causes a persistent fall in prices. And this is an extreme case for the monetary economy because people stop consuming, companies suspend their investment. As a result, demand decreases and firms have to reduce their expenditures, in order to be able to repay their debts. So, there is less spending and investment in the economy which causes further financial turmoil. And this example could be used as a small part of the puzzle describing the situation in Europe as it has a big amount of debts which require a great deal of struggle for the EU countries. Thus, in order to combat this situation, the ECB has started various schemes, which have led to implementation of new initiatives. These are expected to lead to improved business’ and society’s financial conditions, easier investment opportunities and more liquid funds due to increase in money supply.

The APP is a large purchasing programme, which consists of different subprogrammes. So far, the biggest part of the APP, has been the PSPP programme. However, before launching the PSPP programme, the ECB has launched some other purchase programmes and allotments (Szczerbowicz (2012)). First, in 2009 and 2011 two small programmes were launched targeting covered bonds which were issued by the Eurozone banks. These were respectively CBPP1 accounted for €60 billion and CBPP2 accounted for €40 billion. However, these programmes did not meet the targets, so the new initiatives were needed. Further on, another programme launched by the ECB in 2010 was called Securities Markets Programme (SMP). It was created in order to ensure price stability and restore transmission mechanism through monetary policy (Eser and Schwaab (2016)). After that, in 2012 summer, the Outright Monetary Transaction (OMT) was suggested. Even though, it was evaluated as bringing positive outcomes for solving monetary problems in Eurozone, it had not ever been used. Later on, few more APP programmes were introduced in 2014: CBPP3 and ABSPP. CBPP3 purpose was to improve transmission mechanism of monetary policy, provide positive spillovers to the economy and to ease credit provision. ABSPP was meant to increase the issue of new securities. That should have helped for banks by finding alternatives for funding sources. To sum up, the PSPP was the biggest and most contributory programme of the APP, announced in 2015 (see Exhibit 3). The target of this programme has been set as €60 billion worth monthly purchase of private and public securities. For example, in 2016 March, holdings of the PSPP purchases in the Eurozone totaled €648,022 million while the total APP holdings were €832,654 million. As a result, the

(10)

10

PSPP formed about 78% of the total APP purchases. Moreover, as the PSPP consists of relatively big monthly purchases (to compare with the ABSPP and CBPP3), it can be concluded that over time the PSPP takes bigger part of the APP purchases. Finally, the foreseen purchases until September 2016 have been €1.14 trillion. This is the date up to which the PSPP programme has been supposed to finish (as mentioned before, it has already been extended to 2017 March). So, the PSPP is the highlight of this research because of several reasons: it is the biggest part of the APP programme (and keeps growing), all the Eurozone countries hold strictly defined commitments to it and additional announcements about it are expected to be made in the near future.

Additionally, few more non-standard monetary policy measures are needed to be mentioned as these also take part in the QE events of this research. The first one is the LTRO (long-term refinancing operations with three years of maturity), which purpose is to provide

additional credit to banks in order to enhance liquidity and lending of money in Eurozone13_.

The second one is the TLTRO (targeted longer term refinancing operations), which purpose is the same as the LTRO programme’s. However, this one has longer term to maturity. Also, the TLTROs are named as targeted operations because there is a link between the borrowed

amount of money by the banks and provided loans to households and non-financial

corporations14_.

3. Literature review

The focus of this literature review is to summarize the main findings from other researches about the unconventional monetary policy and bond yields’ characteristics. After that, some findings about QE conducted by FR, BoE and BoJ are described. Finally, some arguments and statements are made in terms of the sample selection. The review finished with the hypotheses’ sub-section.

3.1 Monetary policy and government bond yields in the Eurozone

First, as mentioned in Section 2, it is important to distinguish between conventional versus unconventional policies. Woodford (2003) explains what is considered to be

13_{European Central Bank Press release (2011): ‘‘ECB announces measures to support bank lending and money}

market activity’’.

(11)

11

conventional monetary policy which is usually related to the short-term interest rates and bunch of rules. On the other hand, unconventional policy is something that includes market operations and a longer term. Talking about unconventional policies, most of the papers written by the researchers support the notion that unconventional monetary policy is an effective strategy (Beirne et al. (2011)) – yields are lowered as well as long-term interest rates. As a result, economies are positively stimulated by these outcomes not only in the developed countries (Joyce et al., 2012) but the emerging markets as well (Bhattarai and Chatterjee, 2015).

Moreover, the effect on the yields is not stable over time. The most recent crisis has proved that during the extreme times, spreads are much more responsive than in the normal times (Haugh et al., 2009). This could be explained by the monetary decisions, implemented in economically bad times. In contrast, before the crisis, spreads have not been as volatile as during the crisis (Bernoth and Erdogan, 2012).

Furthermore, talking about the differences in spreads, there exists some evidence that different countries in the Eurozone have different sensitivity to spreads (D’Agostino and Ehrmann, 2013). However, this thesis will explore further, whether bailout and QE announcements (and events) have had different impact on different countries’ bond yields. In fact, it is interesting to observe whether QE events make a significant impact on government bond yields and how important in this case the indebtedness measure is. There are also other possible explanations of why these yields differ. For instance, that could be due to the bureaucracy as the European Monetary Union countries have their own central banks but they completely rely on the ECB’s control in terms of liquidity (De Grauwe, 2012).

The highlight of the second hypothesis is the APP’s programmes. Altavilla et al. (2015) discusses the possible channels through which government bond yields and prices are affected. Also, it concludes that in the financial distress, effectiveness of unconventional monetary policy is much higher. Moreover, it is worth understanding the possible spill-overs which can occur and how yields react at the zero lower bound. Rogers et al. (2014) concludes, that the term premium can go down even further, however as a flipside, in a long-term it starts increasing again. Lastly, the importance of fiscal deficits and public debt to government bond yields is discussed by Baldacci and Kumar (2010). They conclude that

(12)

12

these measures can have an adverse effect on government bond yields. Thus, it is important to include fiscal deficits in the model.

3.2 Bailouts and their connection to government bond yields

Some literature talks about bailouts and their effects to different measures. Mink and De Haan (2013) analyze the Greek bailout and its effect on bank stock prices. From this, two important remarks can be made. Firstly, news about Greece does not make a significant change in returns. Secondly, news about bailout does lead to abnormal returns. Also, it is interesting that the sovereign bond prices of Portugal, Spain and Ireland respond to both kind of news. As prices and yields of bonds are closely related, it is interesting to investigate on different bailouts and the effect on government bond yields.

Acharya et al. (2014) talk about induced government bailouts by a distressed financial sector. Furthermore, they claim that this leads to an increase in sovereign credit risk. As a result, the higher the sovereign credit risk, the less reliable are government debt guarantees and holdings of bonds. Thus, a conclusion can be drawn that the government bailout causes an increase in sovereign credit risk. As bond yields include risk premium (Bernoth et al. (2004)), that causes an increase in government bond yields of the country which borrows money. In addition, this thesis will analyze the government bond yields of the countries, which lend money.

3.3 Monetary policy and government bond yields: Evidence from other countries

There is an extensive amount of literature about QE and its effectiveness in other countries. A lot of researches about QE and its relevance have been conducted in terms of FR, BoJ and BoE (D’Amico et al., (2012), Doh (2010)). However, the ECB has also become an extreme focus for researchers during and after the crisis. That is, because a lot of different strategies and decisions on the QE programmes have been implemented.

Additionally, there have been analyses made about the reasons of decline in government bond yields because of the announcements made by the BoE and FR (Christensen and Rudebusch, 2012). This paper helps to understand on which factors/events attention has to be paid while conducting the research. Also, it provides further ideas with a description of possible biases, as they usually prevail in studies about ECB’s QE.

(13)

13

Finally, it is important to see a common view about the magnitude of the effect, possible time range and expected outcomes in major countries. Thus, Altavilla et al. (2015) paper is a good way to start with.

3.4 Determinants of government bond yields

There are a lot of different references for sample selection and methods, which could be applied for this study. As a result, there is no common model for this topic, especially for researching on the PSPP programme, which is the most recent one in the Eurozone (if we compare it with the other ECB’s programmes). This argument can be proved by different models used in De Haan et al. (2013) or Altavilla et al. (2015) studies. However, the choice of the model affects the final results of the research. So, that may be the reason

which causes differences in conclusions among the different studies. For example, De Haan

et al., (2013) try to control for contagion and use corresponding measures for that. However, these contagions not necessarily cause big differences in bond spreads (Forbes (2012)). Thus, the model’s suitability can usually be argued. As a result, it is important to carefully choose the most important control variables which affect government bond yields.

First, the factors which make an influence on sovereign bond yields are the ones, which contribute to liquidity, default risk and uncertainty or investors’ preferences. These factors can be divided into international and country-specific (Barrios et al., 2009). One of the most common factors is fiscal deficit (Barrios et al., 2009, Greenwood and Vayanos, 2013). This is expressed in weighted debt to GDP ratio and includes GDP itself, which is a potential factor for measuring liquidity.

Inflation and overnight interest rates can also be treated as components of bond yields. Inflation is a good example of macroeconomic fundamental together with GDP and debt ratio (De Haan et al., 2013). Additionally, the overnight interest rate works as a risk-free component. It is based on the theory of preferred habitat (Modigliani and Shiller, 1973) and works as a benchmark for my methodology. For example, if the agent has a long habitat preference, he/she will choose longer term instruments rather than shorter term ones (Riedel (2004)).

(14)

14

Further on, for the further discussion, it is possible to adjust the chosen model of this thesis in order to broaden the discussion about the topic. For example, Greenwood and Vayanos (2013) explores analysis about the impact on different maturity bond yields through increased supply of bonds. So, this indicates that it could be taken into account that the increased supply of bonds is a possible control for measuring the variables’ effect on government bond yields. In order to be able to use this adjustment in the research, the model should be adjusted and availability of the data should be checked. Glick and Leduc (2011) is another source which could improve the discussion about this topic. This paper describes announcements made on huge asset purchasing strategies by the BoE and FR. In addition to the QE announcements’ influence on long-term interest rates, the research by Glick and Leduc (2011) provides analyses about the influence on the fluctuations in currencies, exchange rates and commodity prices. Finally, it claims that expectations about the effectiveness of monetary policy are crucial. This could be also analyzed in terms of the Eurozone and differences between stable and non-stable economies. However, this thesis analyzes in detail Eurozone’s government bond yields. Nevertheless, the similarities of the QE’s effect on government bond yields can be indicated between the US, UK and the Eurozone countries.

Finally, robustness checks and possible endogeneity problems matter. That could appear in different forms such as simultaneity or omitted variable bias. For example, the choice of monetary strategy has been caused by the financial turmoil. As a result, government bond yields have changed. So, the problem is to choose exogenous variables and events. However, yields are responsive to a lot of different economic aspects. So, choosing the right variables can be a challenge. For example, Rigobon and Sack (2002) try to estimate responses of interest rates and asset prices to implementation of monetary policy decisions. Additionally, for robustness checks, further analysis could be made by choosing a sample on a daily basis. Ghysels et al. (2014) concentrate on ECB Securities Markets Program which is a part of unconventional monetary policy implemented by ECB. This study follows a similar approach by assuming an exogenous PSPP policy measure and by further investigating government bond yield dynamics on a daily basis by country.

(15)

15 3.5 Hypotheses

In this section, thesis’ hypotheses and objectives are explained in detail. These are summarized into two different sub-sections representing the first and the second hypotheses, which are tested later in this research.

3.5.1 The effect of QE and bailout events on government bond yields: stable vs. non-stable economies

This hypothesis allows inspecting on bailout and QE events’ relation to government bond yields of the Eurozone’s countries. Additionally, it helps to identify significance of these events to the different groups of Eurozone’s countries.

1) The effect of QE and bailout events have a different effect on stable and less stable economies.

There is an expectation of changes in bond yields caused by the large-scale asset purchasing programmes and announcements. These changes appear through different channels such as scarcity and duration. For instance, analysis, of the possible channels, has been made by D’Amico et al. (2012). Moreover, expectations about the government bond yields can be formed from the graphs (Exhibits 1 and 2). Some changes are expected to appear due to the announcements of QE and bailout events. These changes are analyzed from the perspective of the two groups of countries. The first group consists of the more stable countries (Austria, Belgium, France, Germany, Luxembourg, The Netherlands, Malta, Slovakia, Latvia and Lithuania). The second group consists of the less stable (more indebted) countries (Finland, Ireland, Italy, Portugal, Spain, Greece, Slovenia and Cyprus). Thus, these two exhibits indicates that the higher the debt of a country, the more fluctuating government bond yields are. An extreme emphasis for each of the groups can be made on 2015 January, when the announcement of the PSPP was made. On this date, we can observe a huge drop in yields. Moreover, scatter diagrams indicate the downward trend in fitted values of government bond yields. Even though there is a downward scatter trend in most of the graphs, the most indebted countries seem to have upward fitted values slope (Portugal, Greece and Cyprus) or almost flat line (Ireland, Italy, Slovenia and Spain). Thus, these observations indicate that the yields of the most indebted economies’ group may be

(16)

16

biased upwards due to the high fluctuations during the crisis. Due to that reason, investors can switch to the flight to quality. A situation when investors seek to sell the risky assets and buy the safe ones

Finally, in order to test this hypothesis, panel data method has been chosen. Also, the monthly data has been chosen because some of the control variables are not measured on a daily basis.

3.5.2 The effect of the PSPP announcement and implementation on government bond yields

Second, in terms of the QE announcements, it should be interesting for the investors and the central banks whether government bonds yields are affected more by the announcements or actual beginning of the programs. The results of this hypothesis provides information about the possible investment strategies in anticipation of an announcement. For example, if government bond yields react more/less to the particular announcement rather than actual beginning of the programme. The focus of this hypothesis is on the PSPP programme. As discussed in Section 1, the chosen event is one of the biggest QE schemes and the biggest of expanded asset purchasing programme (Exhibit 3). So, attention is paid to it, in order to test the following hypothesis:

2) Government bonds reacted more to the announcement (2015 January) of the PSPP programme rather than to the actual beginning (March 2015) of the PSPP.

Relatively higher response by government bond yields is expected on the PSPP announcement rather than the actual implementation. For example, Altavilla et al. (2015) claim that for FR and the BoE, purchase programmes’ announcements have had higher effects than the actual implementation. In other words, that is expected because after the announcement has been made, everyone knows the exact date, when the programme is supposed to start. Thus, the actual implementation is not an unexpected event to compare with the first announcement. Thus, it should not lead to the higher changes in government bond yields than the announcement. Additionally, this hypothesis is relevant as the PSPP is a recent programme which supposed to be terminated in September 2016. However, it has

(17)

17

already been extended until 2017 March. Thus, there is an expectation of further announcements made for this programme, whether it is finished on March 2017 or not.

Finally, the method used for this hypothesis is panel data. Such choice has been made because this method is widely used in studies about the government bond yields (for example De Haan et al., 2013; Bhattarai and Chatterjee, 2015).

4. Sample selection and data sources

In this section, data sources are presented together with the reasoning behind the choice of the sample.

4.1 Sample selection

First of all, the sample size consists of the 18 Eurozone countries. However, in the Eurozone, there are 19 countries and the choice to exclude Estonia has been made. The option to exclude Estonia has been taken into account because the Estonian government has not issued any government bonds. As a result, Eesti Pank (the Bank of Estonia) is buying

bonds from international institutions in Europe18_{. Thus, there is no data on Estonian}

government bond yields in the database.

The time range is from 2004 January to 2015 December. The beginning of the sample has been chosen in order to cover the most important ECB’s QE implementations and announcements. The focus of this research is on the biggest QE events announced by the ECB, which have occurred during and after the most recent financial crisis. As a result, 2004 has been chosen as a threshold for observing yields. That is because 2004 was far before the crisis. 2015 December has been chosen as a last date for the research because there is no data yet on some of the variables used in this research, after 2015 December. The chosen frequency of time is monthly data. This choice has been made, in order to preserve the effect of some of the control variables on the government bond yields. For example, GDP

and government net debt as % of GDP data (𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑑𝑒𝑏𝑡

𝐺𝐷𝑃 ) also used by De Haan et al.

(2013), are only available monthly or yearly. Thus, regressing on a daily government bond

(18)

18

yield data would make these control variables meaningless and cause distortions for the regression results.

4.2 Overview of data sources

First, for the government bond yields and control variables (GDP, government net debt as % of GDP, NIIP as % of GDP, inflation as CPI and Euro overnight index swap rate as risk-free rate) Datastream source has been used. In detail, the particular sources, chosen

from Datastream database are: Eurostat (for government bond yields, 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑑𝑒𝑏𝑡

𝐺𝐷𝑃 , 𝑁𝐼𝐼𝑃

𝐺𝐷𝑃), national statistics departments (GDP), the World Bank (CPI) and Thomson Reuters

(Euro overnight index swap rate as risk-free rate). Finally, some crucial QE events have been

chosen from the ECB’s webpage19_{of the press releases on monetary policy in line with}

Rogers et al. (2014).

Summary of the chosen variables can be seen in the table below. The column ‘‘Effect on government bond yields’’ means that the particular variable has an expected positive/negative (+/- respectively) effect on the government bond yields in the regression results of both hypotheses:

(19)

19 Table 1: Summary of the chosen variables

Variable Description Definition Sources

Effect on government bond yields g Long-term government bond yield Monthly

average Datastream (Eurostat) N/A

gf Euro swap rate (overnight

index)

Monthly average

Datastream (Thomson

Reuters) +

GDP Real GDP, current prices,

seasonally adjusted Monthly, quarterly averaged Datastream (National statistics departments) -

ND Government net debt as %

of GDP Monthly, yearly averaged Datastream (Eurostat) + NIIP NIIP as % of GDP Monthly, quarterly averaged Datastream (Eurostat) +

CPI Inflation as CPI Monthly

average

Datastream (World

Bank) +

D Dummy for QE events Month of

the event

ECB's webpage, Rogers

et al. (2014) +/-

Bailout Dummy for bailout events Month of

the event

ECB's webpage, BBC

News +/-

Risk free rate (gf) is positively related to the government bond yields. The higher is the risk-free rate, the higher is yield. GDP is a measure which provides information about the economic growth of a country and its stability. So, the higher GDP, the more stable is the country. So, the risk premium for the country decreases. Debt to GDP reflects the default risk of a country. As a result, the higher debt to GDP is expected to be positively related to yields since investors require a higher yield when they face riskier investments. NIIP indicates the external assets and liabilities of the country. For example, the higher government or private debts are (in foreign), the riskier the country is. Again, investors want to be compensated for risks and government bond yield increases. Finally, CPI indicates inflation. That is, the higher is inflation in the country, the higher is the yield, as they have to be compensated for the inflation.

Furthermore, descriptive statistics table is presented below. Yield represents government bond yields, GDP is gross domestic product, Debtratio is government net debt as % of GDP, NIIPratio is NIIP as % of GDP, CPI is inflation and EOI is Euro overnight index swap rate as risk-free rate. On the horizontal line, S.D. stands for the standard deviation and

(20)

20

N stands for the number of observations. The table shows statistics for the period of 2004 – 2015 and chosen parameters’ distribution in terms of the 18 chosen Eurozone countries.

Table 2: Descriptive statistics

Variable Mean Median S.D. N

Yield 4.18 4.03 2.50 2592 GDP 90947.57 19813.05 142498.30 2592 Debt ratio 67.20 65.20 35.37 2592 NIIP ratio -32.88 -26.00 49.98 2592 CPI 2.08 1.96 2.17 2592 EOI 1.70 1.33 1.50 2592

From the table, we can observe that government bond yields are biased upwards, as 4.18% for the government bond yields is a high return. Also, from the provided standard deviation column, we can claim that Eurozone consists of different types of countries and there are possible outliers in the data. Another important measure is Debt ratio. From the table, conclusion can be drawn that the debt ratio deviates from the mean a lot. So, in addition to other arguments, the chosen strategy to divide countries into groups by this measure makes the strategy sensible.

4.3 Summary of the chosen events: QE and bailout

The QE events20_{have been chosen in a way, so that they would represent the}

biggest implementations of unconventional monetary policy made during and after the crisis. Moreover, attention is mostly paid on the events which are related to asset

purchasing due to the reasons explained under the Section 2. The events21_{, that have been}

chosen, are shown in the table below:

20_{Chosen events and their abbreviations have been explained in detail in Section 2.}

21 The strategy of how these events are implemented into the model is described below in the methodology section.

(21)

21 Table 3: QE events

Date QE events

200708 Supplementary LTRO & allotment

200803 New LTRO introduced (6-m)

200906 CBPP announcement

201005* SMP announcement

201209 OMT program announcements

201406 TLTROs announcement

201410 CBPP and ABS operational details announcements

201501 PSPP announcement 201503 PSPP beginning

*Date 201005* includes both: QE event as stated above as well as bailout event of 1st_{Greece’s bailout.}

Most of the QE events are announcements about new programmes in the Eurozone. Again, an emphasis in this research will be made on the PSPP events which can be seen in the bold font, in the table above. In general, the expectation is that these events should decrease the yields as funds become cheaper and borrowing/lending intensity increases.

Furthermore, 7 macroeconomic events have been chosen from the BBC News and

the ECB’s timelines22_{. The chosen bailout events are focused on bailouts of the Eurozone}

countries. Such choice of events has been made as these events have been evoked by the most recent financial crisis and required a particular reaction from the ECB. As a result, these events have had an impact on the whole EU economy and financial institutions such as banks. Thus, government bond yields were affected by these events through different channels such as risk and liquidity, which are represented by the control variables. Additionally, this research helps to summarize how significant these changes in government bond yields have been. So, the bailout events, that have been chosen are:

(22)

22 Table 4: Bailout events

In general, bailouts should increase government bond yields of the countries. For instance, if a country needs bailout, it means that country government bonds are riskier and the risk premium should increase in order to compensate the investor. From a perspective of lending country, it is ambiguous situation. On the one hand, it means that the country is stable and has funds to lend. On the other hand, it may suffer from huge loses if the borrower fails to repay.

However, one event from each of the events’ group is omitted from the regressions in order to avoid multicollinearity problem. To avoid misses of the important schemes, some less significant events have been chosen as a benchmark. It is worth mentioning that they have also been important for the Eurozone countries but not as much as the others. That is because these events have not made much turmoil, at least in the media. From QE events it is 200708 – ‘‘Supplementary LTRO & allotment’’ and from the bailout events it is 201102 – ‘‘Set permanent bailout mechanism’’ (which is also an important agreement but leaves some ambiguity behind and does not incline into category of bailouts).

5. Methodology

Under this section, methodologies of the thesis are explained in detail. First, the benchmark model is presented (see De Haan et al. (2013)). After that, methods for testing each of the two hypotheses are explained.

5.1 Benchmark model: components of government bond yields

The method of the research can be simplified into the benchmark model, based on

the assumption of Leo de Haan et al. (2013) that government bond yields (gi,t) consist of

Date Bailout events

201011 Ireland's bailout

201102 Set permanent bailout mechanism

201105 Portugal's bailout

201107 Agreement on 2nd Greece bailout

201203 2nd Greece's bailout

201206 Spain's bailout

201303 Cyprus' bailout

(23)

23

three main components: risk-free rate (gfi,t), risk premium component (gpi,t), which is

elaborated below and the residual term (Ɛi,t):

(1) gi,t = gfi,t + gpi,t + Ɛi,t

In equation (1), i stands for the country and t for the time period. Risk premium term (gpi,t)

compensates for different factors and risks such as inflation and credit risk. As mentioned above, risk premium factor from the benchmark model is elaborated and split into different

control variables for the main regression. The risk-free rate (gfi,t) is taken as the Euro

overnight index swap rate. Such strategy is chosen because the overnight index swap rates are very close to the risk-free rates (J. Hull and A. White (2012)). Moreover, the benchmark model idea has been taken from Leo de Haan et al. (2013), who also have switched the risk-free rate to overnight index swap rate and used nominal government bond yields.

5.2 The impact of QE and bailout events on government bond yields

The first method of the research is panel data with interaction variables. As mentioned before, 18 Eurozone’s countries have been divided into two groups. Both of these groups have been examined for each of the chosen QE and bailout events. Where the bailout events consist of the bailouts of the Eurozone countries. An assumption is made that

these bailouts are similar and do not require further specifications. However, for the 2nd

regression, each of the bailout is taken separately for each of the country groups. The reason of choosing this model is because of checking the first hypothesis requires regression results for each of the two groups of the Eurozone countries. Thus, panel data perfectly suits the model as it allows for country and time fixed effects and interaction terms. These terms will illustrate how much government bond yields would change for the specific country group (allocated in terms of the economic stability), if one or another event takes place. As a result, it helps to draw a conclusion about different type of countries (in terms of economic stability) and their sensitivity to bailout and QE events separately and together.

In this model, variable gpt, from the benchmark model, is split into different

macroeconomic measures, which reflect factors of countries such as inflation, earnings and attractiveness to foreign investments and credit worthiness. To be specific, the model includes measures such as gross domestic product, government net debt as percentage of

(24)

24

GDP, net international investment position as percentage of GDP, inflation and Euro overnight index swap rate which stands for the risk-free rate. Reasoning of this choice has been explained above, in the ‘‘Literature review’’ section. It has to be mentioned, that this hypothesis is tested in few different ways. Firstly, it is conducted without interaction variables and groups, assuming the effect is the same for all countries, then taking all the events as a whole variable and splitting between each other. As a result, the following regression is implied:

(2) gi,t = β1GDPi,t + β2NDi,t + β3NIIPi,t + β4CPIi,t + β5gfi,t + φ1Groupk * Event1 + … +

φ15Groupk * Event15 + γ1Event1 + … + γ15Event15 + θkGroupk + Φi + λt + Ɛi,t

Where gi,t is government bond yields at time t of country group i. Groupk * Eventi represents

the interaction variable which indicates how much government bond yields would change

for the specific country (k can be 1 or 2 – indicator of group), if QE or bailout event (Eventx)

takes place (where x is from 1 to 15 as there are 15 events in total). Included control

variables are: GDPi,t as the GDP of a country in billions of euros at time t, NDi,t as the net

government debt as % of GDP (𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑑𝑒𝑏𝑡

𝐺𝐷𝑃 ) at time t, NIIPi,t as the net international

investment position as % of GDP (𝑁𝐼𝐼𝑃_𝐺𝐷𝑃) at time t, CPIi,t as the consumer price index at time t

and gfi,t as the the Euro overnight index swap rate at time t. Φi and λt are country and time

fixed effects respectively. φy is the coefficient of interaction variable, βy is are the

coefficients of control variables (where y indicates the number of coefficient), θk is

coefficient for ‘‘Group’’ dummy and γi are coefficients for Event dummies. Ɛt is the error

term.

From the second perspective of this model, bailout and QE events have been taken as a whole. As a result, a bit modified regression is used:

(3) gi,t = β1GDPi,t + β2NDi,t + β3NIIPi,t + β4CPIi,t + β5gfi,t + φkGroupk * Events + γ1Events +

θkGroupk + Φi + λt + Ɛi,t

All the variables are the same as for the regression (2) above. However, the interaction variable changes a bit here. The part ‘‘Events’’ represents all the 15 events chosen in the sample. To be specific, this interaction variable explains how much government bond yields

(25)

25

would change for the specific country group (allocated in terms of economic stability), if we assume that all the events have the same effect on government bond yields.

Finally, the choice of such model specification allows not only inspecting on different groups of the Eurozone’s countries but also provides the importance to the chosen bailout and QE events to government bond yields all over the Eurozone. Thus, this analysis provides a clear answer of which group is the most sensitive in terms of the government bond yield changes. The more detailed analysis (time series for each of the country) is described under the ‘‘Robustness check’’ section below.

5.3 PSPP announcement versus implementation

To test the second hypothesis, panel data model is used with variations in time, country fixed effects and clustered errors. A different model (from the first one) has been chosen, as for the second hypothesis, there is no need to include interaction variables with groups since this is not the focus in the second hypothesis. That is because dummies have been created and inspected. Focus of this hypothesis is on the PSPP programme. As mentioned before, the results of this hypothesis will allow illustrating the importance of the announcement of the programme versus the actual implementation. Thus, panel data fits here, as it allows controlling for the time and country fixed effects (indices i and t in the regression below). As a result, the main regression is (4). Thus, regressions (4) and (5) allow inspecting on their different specifications. So, the essential model for testing the hypothesis on PSPP programme, is as follows:

(4) gi,t = β1GDPi,t + β2NDi,t + β3NIIPi,t + β4CPIi,t + β4gfi,t + τ1Ann + ϱ1Imp1 + ϱ2Imp2 + ϱ3Imp3

+ ϱ4Imp4 + ϱ5Imp5 + Φi + λt + Ɛi,t

Where gi,t stands for government bond yields for country i at time t. τ1 is the coefficient on

the PSPP announcement dummy (Ann). ϱx are the coefficients on PSPP implementation

dummies (Imp). Φi stands for the country fixed effects and λt stands for the time fixed

effects. All other variables and coefficients are the same as in the regression (2) above, however coefficients of the regression (4) include indices i, which represent different countries.

(26)

26

Finally, the main focus is on the two dummies: Annt (Announcement) and ImpZt

(Implementation). Annt is a dummy representing the event of the PSPP announcement at

the month it happened (January 2015). ImpZt are the dummies which represent the actual

beginning of the PSPP and its impact on the evolution of government bond yields after the implementation has been made. Reasoning behind such choice is because after implementation has been made, the impact of it is valid all the time onwards. Thus, implementation dummies are chosen in a way so that they would represent this impact of implementation on the government bond yields throughout the time (up until December

2015 - the last month of the chosen sample). Specifically, each of the ImpZt dummies (Z

represents the number of dummy in the regression, e.g. 1, 2, 3, 4) represent two consecutive months, starting from March 2015, when the PSPP was implemented. In detail,

Imp1t is equal to 1 for two months - March and April 2015; Imp2t is equal to 1 for the other

two consecutive months - May and June of 2015; Imp3t is equal to 1 for the other two

consecutive months - July and August of 2015; Imp4t is equal to 1 for the other two

consecutive months – September and October of 2015; and finally, Imp5t is equal to 1 for

the other last two consecutive months in the sample – November and December of 2015.

Also, additional modifications of the regression (4) will be conducted in order to compare and better understand what causes differences in the outcomes. In total, five regressions have been conducted under this hypothesis. For the first three regressions, see specification of regression (4) with the following conditions:

 First one only includes country fixed effects;

 Second one includes both – country fixed effects and time fixed effects;  Third regression includes both – time and country fixed effects. Additionally

it involves clustered standard errors.

The fourth and fifth regressions slightly differ in their specifications from the regression (4): instead of five ‘‘Imp’’ variables (described in this section before), these include only one ‘‘Implementation’’ variable which stands as a dummy variable. This dummy is equal to 1 after the implementation of the PSPP (starting from 2015 March) has been proceeded and equal to 0 before the implementation of PSPP (before 2015 March). Such

(27)

27

method has also been used by Beirne and Fratzscher (2013) who analyzed the effects of crisis on price of sovereign risks:

(5) gi,t = β1GDPi,t + β2NDi,t + β3NIIPi,t + β4CPIi,t + β4gfi,t + τ1Ann + ϱ1Implementation + Φi +

λt + Ɛi,t

 Fourth regression is a specification of (5) model, excluding time fixed effects and clustered standard errors;

 Fifth regression is the same as (5) model (including time fixed effects and clustered standard errors).

The reason behind separating clustered errors from some models is to inspect whether the correction of autocorrelation makes significant impact on the results or not.

Clustered standard errors help to correct for autocorrelation which appears in a way that Ɛi,t

(omitted factors) might be correlated over time for a given country. Furthermore, making separate analysis without the time fixed effects, indicates whether these are important or not for the Eurozone countries. That is, whether changes over time, but not entities, make significant impact on the chosen model, as ECB’s policy measures are common to all of the Eurozone countries.

Finally, the choice of these specifications will allow to not only distinguish between the PSPP’s announcement and implementation impact on government bond yields but also on changes throughout the time, that the government bond yields have suffered due to the PSPP implementation. However, it must be mentioned that this model needs an assumption of ceteris paribus.

6. Results

Under this section, results of both hypotheses will be presented. Thus, it includes two separate sections: one for the analysis and results’ description of the first hypothesis and another one for the analysis and results’ description of the second hypothesis.

(28)

28

6.1 The effect of QE and bailout events on government bond yields: stable vs non-stable economies

Results for the first hypothesis include two perspectives: one is that every single event is interacted with each of the groups of countries separately (and with all the countries together). The other, wider perspective, is that all the events together are interacted with each of the group of countries.

6.1.1 The effect of QE and bailout events on government bond yields: group 1 countries interaction with events

First of all, the more detailed analysis of interaction variables is discussed. The main purpose of this section and analysis is to see, what has been the reaction of each event for a different group of countries. Also, to see whether any of the events have been statistically significant to the whole group of countries. The outcomes, which are discussed further, can be seen in the Appendix (Exhibit 4). Where the first table (2a) represents results without defining any group, assuming that the effect is the same for all countries. The second table (2b) of Exhibit 4 indicates outcomes for group 1 economies (Austria, Belgium, France, Germany, Luxembourg, The Netherlands, Malta, Slovakia, Latvia and Lithuania).

To begin with the provided output, it can be seen that few of the events have been significant for one of the groups and more events have not. This can also be proved by the F test results. According to the F statistics, interaction variables are jointly statistically significant in this case. The result is that F statistics = 5.23, which is higher than the critical value for the F distribution under 1% significance level with 5 degrees of freedom (which is

equal to 2.04)23_{. Also, from the table 2a, it can be noticed, that there are no significant}

events, if we take them separately. On the other hand, including interaction terms (table 2b), provides with some significant results. From the interaction variables’ perspective, CBPP and OMT programmes seem to be significant. In addition to that, Greece, Portugal and Spain bailouts seem to be meaningful. What is more, the coefficients on Greece2 and Spain bailouts become significant when interaction variables are added as well. Finally, even though F statistics seem significant, most of the results are not significant even at 10% significance level.

(29)

29

Finally, GDP and Debt/GDP ratio seem to be significant in both of the cases. Thus, the increase in GDP usually decreases government bond yields. Also, an increase in government debt causes default risk and the premium for bond holders has to be paid. To conclude, these results confirm the findings of Acharya et al. (2014) that the bailouts causes risks which has to be compensated.

6.1.2 Joint QE and bailout events’ effect on government bond yields

Now, the more abstract analysis of interaction variables is discussed. The main purpose of this section and analysis is to see, what has been the reaction of all the events as a whole to the different group of countries. Also, to see whether these interaction variables have been statistically significant when events are taken together for the group of countries. The outcomes, which are discussed further, can be seen in the Appendix (Exhibit 5).

Firstly, all the controls but NIIP/GDP ratio are more or less statistically significant. Also, the significance level of the particular control variables is the same for both of the groups of countries. While in the perspective before (sub-section 6.1.1), significance levels of coefficients varied over the groups of countries. Secondly, the interesting fact is that the both of the interaction variables are statistically significant at 10% level. It may be the case that if more groups are created, the result would be even more significant. Because now, some countries’ responses overlap with other countries’ responses in the same group. Furthermore, another fact complies with the first perspective of this hypothesis: that the group’s 2 countries government bond yields have increased and group’s 1 yields have decreased. So, the directions of government bond yield changes have been the same as in the first perspective of this hypothesis. However, none of the groups have suffered from significant changes in government bond yields due to the chosen events.

To sum up, these results have proved two major things: first is that they comply with the argument that the riskier countries have higher government bond yields. In case of the bailout events and QE events, these yields react accordingly to the stability of the economies in the Eurozone. Another important fact is that, in terms of magnitude of the effect on government bond yields, both groups have been equally responsive to the chosen events.

(30)

30

6.2 PSPP announcement versus implementation results

As described earlier, this hypothesis’ focus has been on the PSPP programme. From Section 5, we know that, five different regressions have been conducted and thus, five outcomes are summarized in the table in the Appendix (Exhibit 6).

The main purpose of this hypothesis is to make a comparison between the differences in government bond yields between the PSPP announcement and the PSPP implementation with the ongoing time effects. From the Exhibit 6, it can be seen that if there are only country fixed effects, all of the coefficients on dummies are significantly different from 0 at the 1% significance level (see column ‘‘Regression 1’’). Thus, the announcement and implementation of the PSPP, using this specific methodology, have been significant in the Eurozone. However, if time fixed effects are added, the significance levels diminish and results are not significant any more (see column ‘‘Regression 2’’). However, the most interesting results which comply with the first hypothesis are obtained from the third regression (see column ‘‘Regression 3’’) with country and time fixed effects, including clustered standard errors. This column indicates that under the given model, the announcement of the PSPP programme has been significant at 5% significance level, while the implementation of the programme throughout the time has not been significant even at 10% significance level. Thus, according to this model, the hypothesis that government bonds reacted more to the announcement (2015 January) of the PSPP policy rather than to the actual beginning (March 2015) of the PSPP is valid. Moreover, the R squared is 16% higher with the inclusion of time fixed effects. The final remark on this specification is that, even though the coefficients on dummies in Regression 3 are not significant separately, F-statistic test indicates that all together they are significant at 1% significance level. F statistics = 5.70 which is higher than the critical value for the F distribution under 1% significance level with

5 degrees of freedom (which is equal to 3.02)27_.

Furthermore, the same results can be based on slightly different methodology described under Section 5 (specifications (5) and (4)). Coefficients on both - the announcement and implementations of the PSPP programme are significant when time fixed effects are not included in the model (see column ‘‘Regression 4’’). However, inclusion

(31)

31

of time fixed effects and clustered standard errors gives the same outcome as before – only the coefficient on the announcement of the PSPP becomes significant at 5% significance level. Additionally, R squared is the same as for the previous models specifications: 27% and 43% respectively.

Final remarks of this model are on the coefficients themselves. Under the columns without time fixed effects (see columns ‘‘Regression 1’’ and ‘‘Regression 4’’), the coefficients seem to be reasonable as they are negative; when the PSPP was announced (2015 January), government bond yields decreased (the same can be seen from the Exhibit 1). However, addition of time fixed effects (year dummies) makes it look like the government bond yields have increased slightly due to the announcement of the PSPP. It must be said that the dummies used for the time fixed effects have all been negative and the decreasing effect in yields has been incorporated into the coefficients of these dummies. As a result, the dummy on the PSPP announcement has become positive in the statistics table. This is clearly indicated in the ‘‘Regression 1’’ column, where all the coefficients on dummies are negative.

7. Robustness check

This section includes the additional inspection on the first hypothesis and possible alternatives to the chosen model.

7.1 Alternative model to the first hypothesis: a time series analysis by country

As an alternative to the panel data model, time series analysis has been chosen. Such a choice helps to make a detailed analysis about each of the Eurozone’s countries separately. As a result, it provides with the information about separate countries’ reaction to each of the chosen sample events. In the end, comparison between the first hypothesis’ results and these robustness checks is made. A conclusion can be drawn on whether the chosen countries could be mixed in another way, according to their bond yields’ responsiveness to the chosen bailout and QE events.

In order to supplement the test results of the first hypothesis, an alternative method (time-series analysis) for the robustness check has been chosen. Such a choice requires repeating the regression, in the chosen sample, for each of the country separately. Again, the reason of choosing this model is because the more specific check of the first hypothesis

(32)

32

needs regression results for each of the country in the sample. In this regression, variable

gpt from the benchmark model (see the Section ‘‘Methodology’’) is split into different

macroeconomic measures, which reflect factors of countries such as inflation, earnings and attractiveness for foreign investments and credit worthiness. To be specific, the model includes the same control variables as these, chosen in the models above: a gross domestic product, government net debt as percentage of GDP, net international investment position as percentage of GDP, inflation and Euro overnight index swap rate which stands for the risk-free rate. Thus, the regression is then as follows:

(6) gt = α + β1GDPt + β2NDt + β3NIIPt + β4CPIt + β4gft + δ1D1t + … + δ8D8t + θ1Bailout1t +…

+ θ9Bailout7t + Ɛt

Where gt is government bond yields at time t, αis an intercept, GDP is GDP of a country in

billions of euros at time t, ND is net government debt as % of GDP (𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛 𝑑𝑒𝑏𝑡

𝐺𝐷𝑃 ) at time

t, NIIP is net international investment position as % of GDP (𝑁𝐼𝐼𝑃

𝐺𝐷𝑃) at time t, CPI is consumer

price index at time t and gft is the Euro overnight index swap rate at time t. DX are the

dummies for each of the QE events in the chosen sample (if D = 1, the X event happened at the given time t or if D = 0, the X event did not happen at the given time t), BailoutY are the dummies for each of the bailout events in the chosen sample (if Bailout = 1, the Y event happened at the given time t or if Bailout = 0, the Y event did not happen at the given time

t), Ɛt is the error term and βx, δx, θx are the coefficients of variables and dummies.

In addition to the regression, a comparison of each country’s regression coefficients on these chosen QE and bailout events is made. There are few possible ways to conduct these comparisons:

 One is aggregate – sum of the absolute values of QE events’ coefficients on dummies (D) relatively to sum of absolute values of bailout events’ coefficients on dummies (Bailout) for each country.

 Another is absolute – it images the total effect on yields due to these events during the chosen time period. Thus, it shows the actual change in yields in terms of