**MSc Economics**

### International Economics and Globalization Amsterdam School of Economics

**Master Thesis **

## The effects of the announcements of quantitative easing on corporate bond spreads in the euro area over the period

## January 2015 to March 2021

*by*

### Annemijn van Iwaarden 13376314

*15-07-2021*

*Supervisor: * *Second reader: *

Kostas Mavromatis Naomi Leefmans

Number of words: 12518

**STATEMENT OF ORIGINALITY **

This document is written by Annemijn van Iwaarden, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

**Abstract **

After the Global Financial Crisis, the European Central Bank started to use unconventional policy measures such as quantitative easing to stimulate the economy in times when conventional policy measures were shown to be relatively ineffective. This thesis analyzes the effects of the announcements of various asset purchase programmes by the ECB focusing solely on corporate bond spreads and with an emphasis placed on the inclusion of the more recently introduced Pandemic Emergency Purchase Programme (PEPP). Event study methodology is used with monthly data spanning January 2015 to March of 2021. Various variables are included representing a selection of transmission channels. The announcements of the asset purchase programmes are represented by a single dummy variable. The thesis adds to existing literature on corporate bond spreads, by presenting evidence in favor of the workings of the portfolio rebalancing channel and the exchange rate channel while including the PEPP. By including various control variables, the results seem to be reasonably robust for the specified event set. However, it needs to be noted that the empirical results could contain some inconsistency and should be interpreted with caution.

**Table of Contents **

**1. INTRODUCTION ... 5**

**2. RELATED LITERATURE ... 7**

2.1EMPIRICAL EVIDENCE OF ASSET PURCHASE PROGRAMME EFFECTS IN THE EURO AREA ... 7

2.2CORPORATE BOND SPREADS AS INDICATOR FOR FUTURE ECONOMIC CONDITIONS ... 7

2.3CHANNELS OF TRANSMISSION ... 8

*Portfolio rebalancing channel ... 9*

*Signaling channel ... 9*

*Exchange rate channel ... 10*

2.4RISKS OF THE PROLONGED USE OF ASSET PURCHASE PROGRAMMES ... 10

**3. DATA DESCRIPTION AND METHODOLOGY ... 11**

3.1EMPIRICAL MODEL ... 11

3.2DATA DESCRIPTION ... 13

*Macro-economic variables ... 14*

*Financial market variables ... 15*

**4. EMPIRICAL RESULTS ... 16**

4.1DIAGNOSTIC TESTS MODEL WITH PEPP ... 16

4.2REGRESSION RESULTS WITH PEPP ... 18

4.3ROBUSTNESS CHECKS ... 20

4.4REGRESSION RESULTS WITHOUT THE INCLUSION OF PEPP ... 24

**5. DISCUSSION ... 25**

*Key findings ... 26*

*Limitations and future research ... 29*

**6. CONCLUDING REMARKS ... 30**

**BIBLIOGRAPHY ... 32**

**APPENDIX ... 36**

A.DESCRIPTIVE STATISTICS ... 36

B.DIAGNOSTIC TESTS ... 38

*1.ADF test ... 38*

*2.VIF ... 39*

*3.kdensity r ... 40*

*4.pnorm ... 40*

*5.qnorm ... 40*

*6.Breusch-Pagan / Cook-Weisberg test for heteroskedasticity ... 41*

*7.Breusch-Godfrey LM test for autocorrelation ... 41*

C.EXTRA REGRESSION MODELS WITH INCLUSION PEPP AND INTERACTION TERMS ... 42

### 1. Introduction

As the onset of the COVID-19 crisis coincided with already low policy rates (close to the effective lower bound), the use of conventional policy measures, to stimulate the economy and thereby avoiding a deep economic collapse, was limited. This required the use of alternative policy measures (FSB, 2020). After the Global Financial Crisis, which was closely followed by the sovereign debt crisis, the European Central Bank (ECB) started with the use of unconventional policy measures such as quantitative easing (QE) to stimulate the economy in times when conventional policy measures were shown to be relatively ineffective.

With QE, the ECB is able to buy bonds both in the primary as well as in the secondary market.

This leads to an increase in the size of the ECB’s balance sheet and to an increase in the price of these bonds. New money is created and injected into the economy. Consequently, interest rates fall and at the same time loans, and therefore borrowing opportunities, become more affordable (European Central Bank, 2021). This results in a boost in consumption and investment which stimulates economic growth and job creation. Similar to the United States (US) and Japan, the ECB launched various asset purchase programmes with the intention of stimulating economic growth and returning inflation back to levels of around 2 percent (European Central Bank, 2021).

The first of these quantitative easing series started in 2015 and consisted of the Asset-Backed Securities Purchase Programme (ABSPP) and the Covered Bond Purchase Programme (CBPP).

These programmes focused on buying long-term European Union (EU) sovereign bonds with the aim of improving economic conditions and stimulating economic growth in the euro area. In March of 2016, the ECB announced the Corporate Sector Purchase Programme (CSPP) as one of the extensions of the other programmes already in place (D'Amico & Kaminska, 2019). With this programme, the ECB was able to purchase corporate bonds both in the primary market as well as in the secondary market (de Santis & Zaghini, 2019). In March of 2020, the ECB announced the Pandemic Emergency Purchase Programme (PEPP) to be able to address the outlook and risks involved with the COVID-19 outbreak. With this programme, the focus will lie on the purchase of public and private sector securities ("Pandemic Emergency Purchase Programme", 2021).

Since the start of QE, literature has debated on the effectiveness of these asset purchase programmes and the resulting positive and negative consequences of the use thereof. Whilst the impact of these asset purchase programmes of the ECB on various financial indicators, such as on Overnight Index Swap (OIS) rates, government bond yields and exchange rates, is already being addressed by a growing body of research, there is yet a need for a narrowing scope to be made on

the impact of these programmes on corporate bond spreads (Altavilla, Brugnolini, Gurkaynak,
Motto & Ragusa, 2019). Various studies show the importance of corporate bond spreads as an
indicator for the future conditions of the economy and risks to the economy as a whole (Dewachter,
Lania, Lemke & Lyrio, 2019; Gilchirst, Yankov & Zakrajšek, 2009). Therefore, further research
into this specific indicator could be relevant for institutions involved with policy making or those
with an interest in addressing risks for the financial system. In this paper, while building upon
previous research, a closer look is directed towards the effects of the announcements of various
asset purchase programmes by the ECB solely focusing on corporate bond spreads and with an
emphasis placed on the inclusion of the more recently introduced PEPP. Hence, the research
**question to be answered in this paper is: How has quantitative easing and the announcements ****thereof affected corporate bond spreads in the euro area during the period January 2015 to ****March 2021? **

To address this research question, event study methodology following research of Neugebauer (2018) is used. The main estimation, with the inclusion of the PEPP, is based on a monthly time series data set spanning January 2015 to March 2021. A second estimation, without the inclusion of PEPP, is added to perceive possible changes due to the inclusion of PEPP and is based on the same monthly data set as the main estimation, excluding March 2020 to March 2021. The focus for both estimations lies on the announcements of the various asset purchase programmes in the euro area, which are represented with a dummy variable equaling one in the month of an announcement and zero otherwise. Various transmission channels are identified throughout literature and a selection thereof is included through the choice of variables in this research.

Ordinary Least Squares (OLS) regressions are generated with the use of Stata. Overall, this thesis adds to existing literature by presenting evidence is favor of the workings of two transmission channels, namely the portfolio rebalancing channel and the exchange rate channel. No evidence in favor of the working of the signaling channel is found. By including various control variables, the results seem to be reasonably robust for the specified event set.

The rest of the paper is organised along five more chapters. Chapter 2 summarises the literature on effects of asset purchase programmes in the euro area, asset purchase programme effects on corporate bond spreads, a selection of three different transmission channels and risks of the prolonged use of asset purchase programmes. Chapter 3 introduces the main empirical model and the data used. Chapter 4 presents the results of the empirical estimation and Chapter 5 discusses the results. Chapter 6 finally concludes.

### 2. Related literature

2.1 Empirical evidence of asset purchase programme effects in the euro area

A large body of literature has focused on the effects of quantitative easing in the US, United Kingdom (UK) and Japan. However, over the past few years, interest in studying the effects of quantitative easing in the euro area has grown significantly. Altavilla et al. (2019) find by the use of an event study approach that looking at the source of the surprise, sovereign yields and exchange rates respond to quantitative easing by the ECB in ways that are interpretable. Asset purchase programmes are shown to lower all spreads and yields studied. Furthermore, they find that this effect is long-lasting, to about a year (Altavilla, Brugnolini, Gurkaynak, Motto & Ragusa, 2019).

Falagiarda and Reitz (2015) study the effects of ECB announcements on sovereign spreads of various countries in the euro area between 2008 and 2012. Their empirical results show that the announcements reduce long-term government bond yield spreads in almost all countries studied compared to the German counterpart in those countries. Using an event study approach, Szczerbowicz (2015) compares the effects of asset purchase programme announcements on money-market spreads, covered bond spreads and sovereign bond spreads. She finds that asset purchases have a significant effect on lowering bank and government borrowing costs with the existence of high sovereign risk. Furthermore, she shows that there are spill-over effects of these asset purchases to other market-segments (Szczerbowicz, 2015). Georgiadis and Gräb (2015) estimate the effect of the announcement of the extended asset purchase programme (EAPP) on the exchange rate, equity prices and bond yields. Their results show that the announcement of the EAPP stimulated equity prices both in the euro area and in de rest of the world. Furthermore, they show that the announcement resulted in a depreciation of the euro. Gambetti and Musso (2020), using a standard VAR model, show that EAPP announcements had a positive effect both on real economic growth and Harmonised Index of Consumer Prices (HICP) inflation. Furthermore, they show that the EAPP also had a significant impact on the yield curve, leading to a flattening of the curve in the short-term and a steepening in the medium term. Finally, they notice that the EAPP also seemed to affect other financial variables such as stock prices and exchange rates.

2.2 Corporate bond spreads as indicator for future economic conditions

Corporate bonds and their spreads are used in numerous empirical studies as indicator for the future conditions of the economy. One reason for this is that next to bank lending, corporate bonds are used as one of the main sources of firms’ external financing and therefore are of importance to the

economy. Moreover, from a monetary policy perspective, the corporate bond market is also a relevant indicator for policy decisions as it is a central element in the transmission of monetary policy through financial markets (Dewachter, Lania, Lemke & Lyrio, 2018). Corporate bond spreads are defined as the difference between yields on corporate bonds and yields on government bonds at each point of maturity and reflect the extra compensation investors require for encountering credit risk. They are, next to other credit spreads, widely used in forecasting business cycles on the basis of credit channel theories and financial accelerator type mechanisms. With these theories, emphasize is placed on the role of financial frictions in affecting output and investment decisions of firms, which are reflected in credit spreads. Therefore, these spreads can be seen as signals for real economic activity (Saunders, Spina, Steffen & Streitz, 2021).

Various studies have looked into the information content of indicators for default risk, such as corporate bond spreads, in signaling future economic conditions and risks to the economy as a whole. According to Gilchrist, Yankov and Zakrajšek (2009), credit spreads can rise due to disruptions in the supply of credit. These disruptions can either be the result of a deterioration in quality of corporate balance sheets or the result of a worsening in the health of financial intermediaries that supply this credit. This way, a contraction in credit supply can cause yield spreads to widen before economic downturns. As asset values fall and incentives to default increase, lenders will demand a higher compensation for the increased risk due to an expected increase in defaults (Gilchirst, Yankov & Zakrajšek, 2009).

Focusing on the empirical evidence regarding the effects of asset purchase programmes on corporate bond spreads, different asset purchase programmes and different countries are being studied. A general consensus from these studies is the notion that corporate bond spreads decline directly after the announcement of policy measures on secondary markets, both for eligible and non-eligible bonds (de Santis & Zaghini, 2019). De Santis et al. (2018) show by focusing on the CSPP, that the CSPP accounted for a decline in corporate bond spreads of around 25 basis points for eligible bonds and 20 basis points for non-eligible bonds. Similarly, Zaghini (2019) presents evidence for a decline in corporate bond spreads due to the CSPP and furthermore links this to the effects of the signaling channel and the portfolio rebalancing channel.

2.3 Channels of transmission

Different channels are described in literature through which asset purchase programmes are able to affect financial markets. The three transmission channels expected to have the largest impact on

corporate bond spreads and therefore discussed in this paper are: the portfolio rebalancing channel, the signaling channel and the exchange rate channel.

*Portfolio rebalancing channel *

By purchasing public and private sector assets, the ECB provides investors with additional funds.

These funds can then be invested in other assets. This mechanism can be seen as a rebalancing of the investor’s portfolio and leads to an increase in demand for assets in general. This increase in demand then pushes up prices both for the assets that were targeted by the asset purchase programmes as well as for the assets that were not directly targeted. The resulting decrease in the effective market interest rates reduces the costs for companies looking for financing options in the capital markets. Furthermore, the decrease in yields encourages banks to grant loans to companies and households (European Central Bank, 2019). The increase in the supply of loans leads to lower bank lending rates which improve the general financial conditions.

Support for this transmission channel has been found for instance by Carpenter et al. (2015). Their results show that the rebalancing of investors’ portfolios, the main sellers to the Federal Reserve (such as non-bank financial institutions), was largely towards corporate bonds.

*Signaling channel *

The use of asset purchase programmes can also be seen as a credible commitment by the Central Bank to keep interest rates at a low level for a prolonged time. If the Central Bank would increase interest rates at a later point in time, it would risk taking a loss on the assets purchased with the asset purchase programmes. As this would not be an optimal outcome for the Central Bank and therefore most likely to be avoided, the use of asset purchase programmes will signal to the market that key interest rates will remain low for a considerable amount of time. Furthermore, with this commitment to low interest rates, the ECB will demonstrate its intention to meet its inflation targets (Dunne, Everett & Stuart, 2015). This plays an important role in affecting investment decisions as it lowers the risks and volatility involved with future developments in interest rates (European Central Bank, 2019). As a result of this effect on investment decisions, credit demand increases and at the same time spending is encouraged (Dunne, Everett & Stuart, 2015). The effect through the signaling channel will be larger on shorter and intermediate maturity rates than on longer maturity rates, as the commitment to keep key interest rates low will usually only last until the moment that the economy is able to recover (Krishnamurthy & Vissing-Jorgensen, 2011).

*Exchange rate channel *

According to the exchange rate channel, when the economy is open and the domestic nominal interest rate rises above the foreign counterpart interest rate, the domestic currency tends to appreciate. This is due to the increased attractiveness of the domestic currency as an investment currency. Consequently, this appreciation makes domestically produced goods relatively more expensive than foreign produced goods. Thus, it has an impact on the economy through falling net exports and domestic output, while at the same time leading to a decrease in inflation (Beyer et al., 2017). Vice versa, if QE leads to a drop in the domestic nominal interest below the foreign counterpart, the domestic currency tends to become less attractive, which affects demand for this currency. This will lead to a depreciation of the domestic currency and will increase international competition. In the end, net exports will then increase and output is stimulated (Bua & Dunne, 2017).

2.4 Risks of the prolonged use of asset purchase programmes

The prolonged use of asset purchase programmes is not entirely without risks. Recent literature distinguishes four potential areas of concern: bank profitability, a less distinct dividing line between monetary and fiscal policies, future bubbles, and inequalities in income distribution.

Bank profitability can be negatively affected by asset purchase programmes through the lowering of the interest rate structure and the resulting flattening of the yield curve. This may lead to a fall in the net interest income banks receive on new short- and long-term loans (Benigno, Canofari, Di Bartolomeo & Messori, 2020). However, asset purchase programmes do also positively affect banks’ balance sheets by improving macroeconomic conditions. This, for instance, decreases the insolvency risk of firms that borrow and reduces the exposure of banks to non-performing loans.

According to Neri and Siviero (2019), the positive effect on bank profitability largely offsets the negative effect. The second area of concern mentioned is the notion that asset purchase programmes tend to result in a less distinct dividing line between monetary and fiscal policies.

This could, for instance, encourage various forms of moral hazard in national governments and reduce the need for, and therefore the efforts in, the implementation of structural reforms.

However, Benigno, Canofari, Di Bartolomeo and Messori (2020) point out the lack of strong empirical evidence for this area of concern. The third area of concern focusses on an overvaluation of government bonds and financial assets following the use of asset purchase programmes. This could indicate a future burst of a financial bubble. Meanwhile, Neri and Siviero (2019) mention

results of various studies suggesting that this risk of overvaluation and therefore the risk of a financial bubble is actually very low. The last area of concern is inequalities in income distribution.

Asset purchase programmes could result in increasing real and financial asset prices, which tend to benefit financial asset holders, mostly concentrated in the wealthier part of the population. This is also called a reverse robin hood effect (Benigno, Canofari, Di Bartolomeo & Messori, 2020).

Based on both macro and micro evidence, Casiraghi, Gaiotti, Rodano, and Secchi (2018) find no support for a reverse robin hood effect. Following their reasoning, the opposite effect may occur as the effects through stimulation of employment and other economic activities outweigh the effects of financial variables. This would imply that households at the bottom of the income scale receive larger benefits. A second argument that contradicts this reverse robin hood effect is given by Benigno, Canofari, Di Bartolomeo and Messori (2020). They mention that asset purchase programmes could lead to a decrease of the long-term interest rates which tend to benefit households that have a high debt to income ratio. These households also tend to be the households at the bottom of the income scale.

### 3. Data description and methodology

3.1 Empirical model

Most empirical studies on the effects of asset purchase programmes either use the approach of an event study or the approach of a vector autoregression (VAR). Event studies usually use high frequency data which require the use of variables with the same frequency, such as interest rates or asset prices. VAR models, on the other hand, are also used to study the effects of low frequency data such as inflation. In addition, in contrast to the event study approach, the use of VAR models often makes it more difficult to identify causal relationships (Fiedler, Hanisch, Jannsen & Wolters, 2016). With this in mind and since corporate bond spreads are considered high frequency data, an event study approach is used for answering the research question. Furthermore, the empirical method is based on the notion of efficient market theory. According to Fama (1969), the efficient market theory suggests that stock markets adjust to new information immediately.

Neugebauer (2018) studies the effects of asset purchase programmes on 10-year government bond yields of member states in the euro area. He uses an event study methodology with a daily timeseries dataset. In contrast to, for instance, Falagiarda and Reitz (2015) that use a dummy variable for each separate event, Neugebauer (2018) uses a single dummy variable for all announcements combined. This dummy variable then takes the value of 1 on the day of the announcement and is zero otherwise. By using one dummy variable for all announcements, he is

able to detect a more generalized effect of an ECB announcement. This is more suitable for policy making as the ECB is interested in determining the average effect of comparable future announcements (Neugebauer, 2018). A similar method, as the one used by Neugebauer (2018), is followed by various other studies, such as a study by Urbschat and Watzka (2020) that looks at the development of bond yields and spreads around the asset purchase announcements. This is also the method that is followed in this research. However, in this study a monthly time series data set is used instead and different variables are studied. For the choice of variables and announcements dates, research and suggestions by Altavilla et al. (2019), de Santis (2016) and Elbourne and Duijndam (2018) are followed. The working of the signaling channel is researched by the announcement dummy variable, whereas the workings of the portfolio rebalancing channel and the exchange rate channel are tested by including variables for the aggregate corporate bond holdings in the private sector and the Euro to US Dollar exchange rate. The time-period used in Altavilla et al. (2019) is partly followed in this research and extended to March of 2021 to include the more recently introduced PEPP. Lagged variables are used for the dependent variable and for each independent variable, with the exception of the announcement dummy variable, to control for reversed causality.

The resulting main empirical model used for this research is:

∆SX7Et-1 = β0 +β1At + β2EURIBOR_3Mt-1 + β3HICPt-1 + β4IndusProdt-1 + β5EURUSDt-1 + β6STOXX50Et-1 + β7VSTOXXt-1 + β8CISSt-1 + β9CorpBondt-1 + ε, (1)

The dependent variable, ∆SX7Et-1,represents period-to-period changes in the stock price index comprising banks, denoted in percentages.

The independent variables are respectively:

*• A**t*: announcements of asset purchase programmes denoted by a dummy variable.

*• EURIBOR_3M**t-1*: monthly percentage changes of the 3 months Euribor rate.

*• HICP**t-1*: the working day and seasonally adjusted Harmonised Index of Consumer Prices.

*• IndusProd*t-1: monthly percentage changes in aggregate industrial production including
manufacturing and construction.

*• EURUSD**t-1*: monthly percentage changes in the closing price of the exchange rate Euro to
US Dollar.

*• STOXX50E**t-1*: monthly percentage changes in the stock market index.

*• VSTOXX**t-1*: monthly percentage changes in the volatility index based on EURO STOXX
50.

*• CISS**t-1*: Composite Indicator of Systemic Stress.

*• CorpBond**t-1*: monthly percentage changes in the aggregate corporate bond holdings of the
private sector.

Research by Valiante (2015) presented evidence in favor of the announcement effect (which represents the main part of the signaling channel). He found that the announcement effect led to a drop in interest rates for Eurozone government bonds, specifically the ones that are part of the asset purchase programmes. Moreover, he found evidence for a decrease in the differential between AAA-rated bonds and other government bonds in the Euro Area. Therefore, also in accordance with research by Zaghini (2019) that found evidence for the workings of both the signaling channel and the portfolio rebalancing channel while focusing on CSPP, the expectation for this thesis is to find evidence as well in favor of both of these channels. This means that the coefficients of the announcement dummy variable (A) and the aggregate corporate bond holdings for the private sector (Corpbond) are expected to have a significant effect on the stock price index comprising banks (SX7E). Furthermore, expectations on the exchange rate channel are present following research by Demertzis and Wolff (2016), that show that QE weakened the Euro to US Dollar exchange rate significantly through the working of the exchange rate channel. Therefore, the expectation for this thesis is also to possibly find evidence in favor of the working of the exchange rate channel on corporate bonds spreads. This means that the coefficient of the Euro to US Dollar exchange rate (EURUSD) is expected to have a significant effect on the stock price index comprising banks.

Thus, based on the literature mentioned, the hypothesis is stated as follows: the announcements of the asset purchase programmes, with the inclusion of the PEPP, have a significant impact on the stock price index comprising banks over the period January 2015 to March 2021 through the workings of the portfolio rebalancing channel, the signaling channel and the exchange rate channel.

3.2 Data description

To estimate the effect of the asset purchase programmes with the inclusion of PEPP on corporate bond spreads in the euro area, monthly data on the euro area (19 countries that use the Euro)

between 2015:M1 and 2021:M3 are used.^{1} Table 5 and Table 6 in appendix A give an overview
of the descriptive statistics of the variables used for both the estimation with the inclusion of PEPP
and without the inclusion of PEPP.

The dependent variable, the stock price index comprising banks (SX7E),is used as a proxy for corporate bond spreads and the data are retrieved from Investing.com. As the main independent variable, denoted by A, the announcements of the various asset purchase programmes are used. A large body of literature found that most of the impact of the asset purchase programmes arises around the announcements of the programmes rather than the actual implementations, hence the decision for the use of the announcement dates as independent variable (Altavilla, Carboni &

Motto, 2015). The data on the announcements are retrieved from the Euro Area Monetary Policy event study Database (EA-MPD) (Altavilla, Brugnolini, Gurkaynak, Motto & Ragusa, 2019). The EA-MPD is split into three different subcategories: press release window, press conference window and monetary event window. For each announcement in the dataset, the press release (window) starts at 13:45h and this is followed by the press conference (window) that starts at 14:30h and ends at 15:30h. The combination of these two windows compromises the monetary event window. This last combination window is the window of interest for this research. The announcement dummy variable equals one for each month that includes an announcement date and is zero otherwise.

There are various other variables included as variables of interest for the transmission channels and as control variables to limit the risk of endogeneity. These variables are divided into two categories: (aggregate) macro-economic variables and financial market variables.

*Macro-economic variables *

The included macro-economic variables are: EURIBOR_3M, HICP, IndusProd and EURUSD.

The variable EURIBOR_3M, the 3 months Euribor rate, is included as a control variable. It is included as a proxy for the OIS 3-month rate and therefore as a proxy for the risk-free rate.

According to de Santis (2016), an increase in the risk-free rate leads to a lowering of the credit spreads of all bonds and therefore to a lowering of corporate bond spreads. This results from the

1 The use of an event study with daily data should reduce the risk of obtaining endogeneity. According to Haitsma et al. (2015), one-day windows are less likely to result in contamination of the data by other news than the quantitative easing announcements. However, not all variables of importance were available in daily data, therefore monthly data are used, which could result in less precise results.

notion that an increase in the risk-free rate stimulates the drift of firms’ assets to risk-neutral assets, which lowers the probability of a default. Data for this variable are retrieved from the ECB Statistical Data Warehouse. The variable HICP, Harmonised Index of Consumer Prices, is used to measure inflation and is also included as a control variable. It is denoted in period-to-period percentage changes and retrieved from the ECB Statistical Data Warehouse. According to de Santis (2016), inflation is positively related to corporate bond spreads. IndusProd, the growth in aggregate industrial production including manufacturing and construction, is used as proxy for real GDP growth. Similar to HICP, it is included as a control variable and denoted in period-to-period percentage change and is retrieved from the ECB Statistical Data Warehouse. Finally, the variable EURUSD, the exchange rate Euro to US Dollar, is included as one of the main independent variables to possibly observe the effects of the exchange rate transmission channel and is retrieved from Investing.com.

*Financial market variables *

The financial market variables included are: STOXX50E, VSTOXX,CISS and CorpBond. The variable STOXX50E is a stock market index that includes the stock prices of 50 large, both financial and non-financial, corporations located in the euro area. It is included as a control variable. Data for this variable are denoted in period-to-period percentage changes and are retrieved from Investing.com. VSTOXX is a volatility index that is based on EURO STOXX 50 and is included as measure to reflect the market expectations of near-term up to long-term volatility. It is included as a control variable and retrieved from Investing.com. The variable CISS is a Composite Indicator of Systemic Stress of which the aim is to measure the current state of instability (Hollo, Kremer & Lo Duca, 2012). It is retrieved from the ECB Statistical Data Warehouse and incorporates a total of 15 financial stress measures. According to Elbourne and Duijndam (2018), the euro area monetary policy responds systematically to financial system shocks. Furthermore, the CISS index is suggested to capture relevant international effects which reach the euro area as a whole (Elbourne & Duijndam, 2018). The CISS index contains some elements related to the volatility index (VSTOXX), however they are not perfectly aligned and convey different information (Dewachter, Iania, Lemke & Lyrio, 2019). Consequently, it is important to include CISS as a control variable. Lastly, CorpBond represents the aggregate corporate bond holdings of the private sector in the euro area. It is reported in period-to-period percentage changes and retrieved from the ECB Statistical Data Warehouse. It is included as one

of the main independent variables to see the possible effects of the portfolio rebalancing channel on corporate bond spreads.

### 4. Empirical results

In this chapter the empirical results are presented. First, the results of diagnostic tests are mentioned, which are needed to check important assumptions before running OLS regressions.

Next, the main regression results with the inclusion of PEPP are discussed. Next, two robustness checks are included to test if the results of the main regression models can indeed be interpreted as valid and robust. Furthermore, some variations on the main empirical model, with the inclusion of newly generated interaction terms, are discussed. Finally, diagnostic tests and results for the estimation without the inclusion of PEPP are considered.

4.1 Diagnostic tests model with PEPP

Before running the OLS regressions with Stata, various steps are required. First, the data are
*identified as time series data with the tsset command. The variable Date is generated, representing *
the month and year of each of the observations. Next, some assumptions need to be checked.

One of the most important assumptions when working with time series regression is that the data
are stationary. This means that the data’s properties (mean, variance and autocorrelation structure)
do not depend on time. If the data are not stationary, there is a risk of a spurious regression and the
empirical results cannot be used to make a meaningful interpretation. An Augmented Dickey-
Fuller (ADF) test is used to test this specific assumption. On each variable the ADF test is
performed three times: with constant, constant and trend, and finally with constant and drift (and
for the residual also with no constant). To determine the right amount of lags the minimum
*Schwarz Information Criterion (SIC) value is calculated (by using the command dfgls in Stata). *

For most variables one lag is used, except for the variable EURIBOR_3M and the residual (r) for which two lags are used. All variables, with the exception of EURIBOR_3M, are shown to be stationary. This means that the null hypothesis that a variable contains a unit root is rejected. The variable EURIBOR_3M does contain a trend. By applying first differentiation, variable EURIBOR3M_d1 is generated, which by performing the ADF test is shown to be stationary. The ADF test results are presented in Table 7 in Appendix B.

A second assumption is that the residuals are not serially correlated. This could otherwise result in biased variances of the estimated coefficients which could ultimately lead to unreliable results.

The latter is checked with a Breusch-Godfrey test with 12 lags as monthly data are used. The null hypothesis cannot be rejected, meaning that the error term is not serially correlated. Therefore, no correction is necessary. The results for the Breusch-Godfrey test are presented in section 7 of Appendix B.

The third assumption that is checked is the assumption of no perfect multicollinearity between the
*variables. The vif command in Stata is used for this. The VIF values should all be lower than 10. *

As shown in Appendix B (Table 9), all variables have VIF values below 10, with a mean VIF of 1.49. Thus, there is no need for corrections with respect to perfect multicollinearity.

A fourth assumption is that the variance of the residuals is homoscedastic. This means that there
should not be any pattern between the predicted values and the residuals. However, if such a pattern
were to exist, the residuals would be heteroskedastic, and a correction would be needed. A
*Breusch-Pagan test (with Stata command estat hettest) is used to test for heteroskedasticity. As *
shown in Appendix B (section 6), the null hypothesis of homoscedastic variance of the residuals
cannot be rejected, so no correction is necessary.

Lastly, the normality of residuals is checked. This is necessary to ensure that the tests (p-value, F- test, t-test) are valid and therefore usable inferences can be drawn from the regression. This is checked by looking at three different graphs in Stata: kdensity, pnorm and qnorm. The Shapiro- Wilk test for normality could also be used. The results in Appendix B (Graph 1, Graph 3 and Graph 5) show that the residuals are normally distributed with a mean of roughly zero.

4.2 Regression results with PEPP

(1)

VARIABLES LagSX7E

A -0.00260

(0.0128)

LagSTOXX50E 1.980***

(0.149)

LagVSTOXX 0.0858***

(0.0173)

LagCISS -0.0581

(0.161)

LagEURUSD 0.625**

(0.257)

LagCorpbond -0.0206***

(0.00636)

LagInduspro 0.00108

(0.00126)

LagHICP -0.0115

(0.0226) LagEURIBOR3M_d1 0.149

(0.160)

Constant 0.00525

(0.0228)

Observations 73

R-squared 0.808

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

*Table 1: Stock price index comprising banks model with robust standard errors *

The results of the regression analysis regarding the main estimation with the inclusion of PEPP are illustrated in Table 1 (Model 1).

Model 1 includes all the financial market variables and the macro-economic variables and has 73 observations. The regression results of this model show an R-squared of 0.808, meaning that 80.8 percent of the variation in the stock price index comprising banks can be explained by variations in the independent variables. In this model, the announcement dummy variable (A), representing the signaling channel, does not have a significant effect on the stock price index comprising banks (SX7E) at the 10 percent, 5 percent or 1 percent significance level (p=0.841). The variables for the stock market index (STOXX50E) and the volatility index (VSTOXX) are both significant at the 1 percent significance level (with p=0.000 and p=0.000). The variable for the Composite Indicator

of Systemic Stress (CISS) does not have a significant effect on the stock price index comprising
banks (p=0.719). Moreover, the Euro to US Dollar exchange rate variable (EURUSD),
representing the exchange rate channel, has a significant effect on the stock price index comprising
banks at the 5 percent significance level (p=0.018). The aggregate corporate bonds variable
(Corpbond), representing the portfolio rebalancing channel, has a significant effect on the stock
price index comprising banks at the 1 percent significance level (p =0.002). Finally, the variables
for the growth in aggregate industrial production (IndusProd), Harmonised Index of Consumer
Prices (HICP)*and the 3-months Euribor rate (EURIBOR_3M) do not have a significant effect on *
the stock price index comprising banks (with p=0.396, p=0.613 and p=0.354).

4.3 Robustness checks

Two robustness checks are included to test if the results of the main regression models can indeed be interpreted as valid and robust. Furthermore, some variations on Model 1, with the inclusion of newly generated interaction terms, are included as an extra regression output in Appendix C.

(1) (2)

VARIABLES LagSX7E LagSX7E

A -0.00260 0.00228

(0.0128) (0.0119)

LagSTOXX50E 1.980*** 1.941***

(0.149) (0.140)

LagVSTOXX 0.0858*** 0.0809***

(0.0173) (0.0179)

LagCISS -0.0581 -0.106

(0.161) (0.149)

LagEURUSD 0.625** 0.824***

(0.257) (0.277) LagCorpbond -0.0206*** -0.0184***

(0.00636) (0.00546)

LagInduspro 0.00108

(0.00126)

LagHICP -0.0115

(0.0226) LagEURIBOR3M_d1 0.149

(0.160)

Constant 0.00525 0.00326

(0.0228) (0.0199)

Observations 73 74

R-squared 0.808 0.793

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

*Table 2: Stock price index comprising banks models for robustness check *

The first robustness check is comparing the main estimation model (Model 1), including all financial market and macro-economic variables, with Model 2 only including the financial market variables and the most important macro-economic variables (the ones representing the various transmission channels) and thereby leaving some macro variables out of the regression. For the results to be robust, the models should be similar in terms of signs of coefficients, sizes of coefficients and the levels of significance for the variables included in both models. Looking at

Table 2, the regression results of Model 2 show an R-squared of 0.793. This means that 79.3 percent of the variation in the stock price index comprising banks (SX7E) can be explained by variations in the independent variables. This is slightly lower than the R-squared of Model 1 (79.3 percent versus 80.8 percent). Similar to Model 1, the announcement dummy variable (A) does not have a significant effect on the stock price index comprising banks (p=0.849) for Model 2. The variables for the stock market index (STOXX50E) and the volatility index (VSTOXX) both have a significant effect on the stock price index comprising banks at the 1 percent significance level (with p=0.000 and p=0.000). The variable for the Composite Indicator of Systemic Stress (CISS) does not have a significant effect on the stock price index comprising banks, the p-value is greater than the critical value of the alpha of 5 percent (p=0.479). Furthermore, the Euro to US Dollar exchange rate variable in this model has a significant effect on the stock price index comprising banks at the 1 percent significance level (p=0.004). Lastly, the aggregate corporate bonds variable (Corpbond) also has a significant effect on the stock price index comprising banks at the 1 percent significance level (p =0.001). Overall, the results are very similar for both models in terms of signs of the coefficients and the variables that show significance, meaning that the results are reasonably robust. The main difference is the switch of the sign of the coefficient of the announcement dummy variable, which is negative in Model 1 and positive in Model 2.

(1) (3)

VARIABLES LagSX7E LagSX7E

A -0.00260 0.000465

(0.0128) (0.0123)

LagSTOXX50E 1.980*** 1.988***

(0.149) (0.162)

LagVSTOXX 0.0858***

(0.0173)

LagCISS -0.0581

(0.161)

LagEURUSD 0.625** 0.605**

(0.257) (0.247) LagCorpbond -0.0206*** -0.0178***

(0.00636) (0.00642)

LagInduspro 0.00108 0.00154

(0.00126) (0.00134)

LagHICP -0.0115 -0.00977

(0.0226) (0.0193)

LagEURIBOR3M_d1 0.149 0.164

(0.160) (0.151)

LagVIX 0.0786***

(0.0187)

Constant 0.00525 -0.00432

(0.0228) (0.0106)

Observations 73 73

R-squared 0.808 0.813

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

*Table 3: Stock price index comprising banks models with VSTOXX/ CISS replaced by VIX *

As a second robustness check, the results of Model 1 are analyzed where the indicators for volatility CISS and VSTOXX are replaced by another indicator for volatility, namely VIX, leaving everything else unchanged. The VIX index is calculated by the Chicago Board Options Exchange (CBOE) and is often used as an indicator of uncertainty for financial markets. It is denoted in period-to-period percentage changes and is retrieved from Investing.com. It shows the volatility of S&P500 index options (Lamoen, Mattheussens & Dröes, 2017). Table 3 shows the results of these changes in variables, where Model 1 is the same model as used in Table 1 and Model 3 is the new model with indicator VIX. The results are very similar in terms of coefficients, signs of coefficients and significance levels. The announcement dummy variable does not have a significant effect on the stock price index comprising banks for both Model 1 and Model 3,

whereas the variables for the stock market index, Euro to US Dollar exchange rate and aggregate corporate bond holdings, show similar levels of significance for both models. The R-squared has improved somewhat, from 0.808 to 0.813, implying that Model 3 is slightly better in describing the variation of the stock price index comprising banks. One difference between the two models, similar to the first robustness check, is the switch of the sign of the coefficient of the announcement dummy variable. It is negative for Model 1 and positive for Model 3. This could be an indication of reversed causality or omitted variable bias.

Various variations on Model 1, with the inclusion of newly generated interaction terms, are included as an extra regression output in Appendix C. The interaction terms included are interactions of the announcement dummy variable with the second lag of the stock price index comprising banks (Alag2SX7E), the first lag of the Euro to US Dollar exchange rate (AlagEURUSD), the first lag of the volatility index (AlagVSTOXX), and the first lag of the VIX index (AlagVIX). The variable with the second lag of the dependent variable (stock price index comprising banks) is included to check if there is a possible effect on corporate bond spreads present for more than one period (month), whereas the second lag is used to avoid a high level of multicollinearity with the dependent variable (which is the first lag of the stock price index comprising banks). Other interaction terms are included to check whether they result in a better fit of the models. Therefore, four new models (Model 5, Model 6, Model 7 and Model 8) are created one each for the inclusion of each interaction term. As shown in Appendix C in Table 11, the interaction terms with the second lag of the dependent variable (Alag2SX7E) and the first lag of the Euro to US Dollar exchange rate (AlagEURUSD) both do not have a significant effect on the stock price index comprising banks (with p=0.116 and p=0.956). However, the interaction terms with the first lag of the volatility index (AlagVSTOXX) and the first lag of the VIX index (AlagVIX) do have a significant effect on the stock price index comprising banks at the 5% level for the volatility index and at the 10% level for the VIX index (with p=0.037 and p=0.068).

Furthermore, the signs of coefficients of these last two interaction terms are negative, -0.112 for the volatility index interaction term and -0.0881 for the VIX index interaction term. In contrast, both the volatility index and the VIX index, both significant at the 1 percent significance level, have positive coefficients with 0.180 for the volatility index and 0.143 for the VIX index.

4.4 Regression results without the inclusion of PEPP

(4)

VARIABLES LagSX7E

A -0.00472

(0.0109)

LagSTOXX50E 1.584***

(0.293)

LagVSTOXX 0.0314

(0.0426)

LagCISS_d1 -0.392

(0.505)

LagEURUSD 0.437*

(0.228)

LagCorpbond -0.0168*

(0.00877)

LagInduspro -0.00881*

(0.00509)

LagHICP 0.00363

(0.0555) LagEURIBOR3M_d1 0.350

(0.324)

Constant 0.000302

(0.0109)

Observations 60

R-squared 0.736

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

*Table 4: Stock price index comprising banks models without PEPP *

Table 4 shows the results without the inclusion of PEPP. PEPP was first announced in March 2020 (Aguilar, Arce, Hurtado, Martínez-Martín, Nuño & Thomas, 2020). Therefore, to check whether the results are different without the inclusion of the recently introduced PEPP, data spanning January 2015 to February 2020 are used for the OLS regressions. The same steps and assumptions are followed and checked as in section 4.1.

On each variable included, the ADF test is performed three times: with constant, constant and
trend, and finally with constant and drift. To determine the right amount of lags, the minimum SIC
*value is calculated (by using the command dfgls in Stata). For most variables one lag is used, *

except for the variable Induspro and the residual (r) for which two lags are used. All variables, with the exception of EURIBOR_3M and CISS, are shown to be stationary. This means that the null hypothesis that a variable contains a unit root is rejected. The variables EURIBOR_3M and CISS do contain a trend. By applying first differentiation, variables EURIBOR3M_d1 and CISS_d1 are generated, which by performing the ADF test are shown to be stationary. The ADF test results are presented in Table 8 in Appendix B.

Next, serial correlation of the residuals is checked with a Breusch-Godfrey test with 12 lags as
monthly data are used. The null hypothesis cannot be rejected, meaning that the error term is not
serially correlated. Furthermore, as shown in Appendix B (Table 10), all variables have VIF values
below 10, with a mean VIF of 1.78. Thus, there is no need for corrections with respect to perfect
*multicollinearity. A Breusch-Pagan test (with Stata command estat hettest) is used to test for *
heteroskedasticity. The null hypothesis of homoscedastic variance of the residuals cannot be
rejected, so no correction is necessary. Lastly, the normality of residuals is checked. This is
checked by looking at three different graphs in Stata: kdensity, pnorm and qnorm. The results can
be found in Appendix B (Graph 2, Graph 4 and Graph 6) and show that the residuals are normally
distributed with a mean of close to zero.

Model 4 includes, similar to Model 1, all financial market variables and macro-economic variables and has 60 observations. The regression results of Model 4 show an R-squared of 0.737, meaning that 73.7 percent of the variation in the stock price index comprising banks (SX7E) can be explained by variations in the independent variables. Like Model 1, the announcement dummy variable (A), representing the signaling channel, does not have a significant effect on the stock price index comprising banks at the 10 percent, 5 percent or 1 percent significance level (p=0.666).

Moreover, the stock market index (STOXX50E) has a significant effect on the stock price index comprising banks at the 1 percent significance level (with p=0.000). However, in contrast to Model 1, the volatility index (VSTOXX) does not have a significant effect on the stock price index comprising banks (p=0.465) and aggregate corporate bond holdings (Corpbond) is only significant at the 10 percent level of significance (p=0.062). Furthermore, the variable for aggregate industrial production (IndusProd) is significant at the 10 percent level of significance (p=0.090).

### 5. Discussion

In this thesis, a closer look is taken towards the effects of quantitative easing on corporate bond spreads in the euro area during the period January 2015 to March 2021. More specifically, the effect of the announcements of the various asset purchase programmes on corporate bond spreads

is studied. The stock price index comprising banks is used as a proxy for corporate bond spreads, where the stock price index comprising banks is negatively correlated with corporate bond spreads.

If the stock price index comprising banks decreases, corporate bond spreads will increase and vice versa. This section discusses the key findings for the stock price index comprising banks models both with and without the inclusion of PEPP. Thereafter, limitations of this research and ideas for future research are mentioned.

*Key findings *

As mentioned in the results section, there are four variables that have a significant effect on the banks stock index in the main estimation model including the PEPP (section 4.2), namely: the stock market index (STOXX50E), the volatility index (VSTOXX), the Euro to US Dollar exchange rate (EURUSD) and the aggregate corporate bond holdings (Corpbond). With the exception of the volatility index, these variables all have coefficients with signs that are in accordance with existing literature.

First, the empirical results for Model 1 with the inclusion of PEPP, show that the stock market

index has a significant effect on the stock price index comprising banks. The coefficient is positive, 1.980 for Model 1, indicating that a one percentage point increase in the stock market index increases the stock price index comprising banks by 1.980 percentage points when everything else is held constant. The broad euro area stock market index was included in this research as a control variable as it is assumed to also react to announcements of asset purchase programmes (Bats, Giuliodori & Houben, 2020). However, as it was included as a control variable, it is not a main variable of interest for this study.

Second, the empirical results also show a positive correlation between the Euro to US Dollar exchange rate and the stock price index comprising banks. This means that if the Euro to US Dollar exchange rate rises, the stock price index comprising banks also rises. The coefficient of the Euro to US Dollar exchange rate is positive, 0.625 for Model 1. Keeping everything else constant, this indicates that a one percentage point increase in the Euro to US Dollar exchange rate increases the stock price index comprising banks by respectively 0.625 percentage points. Due to the negative correlation between the stock price index comprising banks and corporate bond spreads, this results in a decrease in corporate bond spreads. This result presents evidence in favor of the working of the exchange rate transmission channel. Depreciation of the exchange rate is triggered by capital outflows. Upon the announcements (and implementation) of the asset purchase

programmes, foreign investors liquidate their QE eligible assets and bonds holdings. These investors then rebalance their portfolio towards assets and bonds with similar maturities outside of the euro area. This depreciates the Euro against the US Dollar as demand for the Euro (as currency) goes down. The latter increases the stock price index comprising banks and lowers corporate bond spreads. Due to increased international competition, net exports increase and output is stimulated (Bua & Dunne, 2017). This is in line expectations based on Demertzis and Wolff (2016), that show that QE weakened the Euro to US Dollar exchange rate significantly through the working of the exchange rate channel.

Third, the results present a negative correlation between the stock price index comprising banks and the aggregate corporate bond holdings. This can be seen when looking at the negative coefficient of aggregate corporate bond holdings of the private sector. The coefficient is -0.0206 for Model 1. While keeping everything else constant, this indicates that a one percentage point increase in the aggregate corporate bond holdings decreases the stock price index comprising banks by 0.0206 percentage points. This presents evidence in favor of the portfolio rebalancing transmission channel. As people sell their position of corporate bonds (long-term assets) and rebalance towards short-term assets such as stocks, the aggregate holdings of corporate bonds in the private sector are reduced and the stock price index comprising banks increases due to lower returns. In the end, due to the negative correlation between stock price index comprising banks and corporate bond spreads, this leads to a decrease in corporate bond spreads. This rebalancing results in a decrease in the effective market interest rates which reduces the costs for companies looking for financing options in the capital markets. Furthermore, the decrease in yields and spreads encourages banks to grant loans to companies and households, thus increasing the supply of loans and reducing bank lending rates and improving financial conditions (European Central Bank, 2019). This finding of the portfolio rebalancing channel is in accordance with expectations and existing literature. While focusing solely on the CSPP, Zaghini (2019) presents evidence for both the portfolio rebalancing channel and the signaling channel and furthermore finds evidence for a decline in corporate bond spreads due to the use of CSPP.

Not all results are fully in line with existing literature. The results for the volatility index are in contradiction with expectations based on a growing body of literature. The empirical results for Model 1 with the inclusion of PEPP, show that the volatility index has a significant effect on the stock price index comprising banks. The coefficient of VSTOXX is positive, 0.858 for Model 1, indicating that a one percentage point increase in market volatility increases the stock price index comprising banks by respectively 0.858 percentage points when everything else is kept constant.

Due to the negative correlation between the stock price index comprising banks and corporate bond spreads, this means that according to these results corporate bond spreads are expected to be lowered, or in other words that they will narrow. This is not in line with economic theory.

Corporations with more volatile assets have a higher probability of reaching the conditions of default, as it is more likely that, at some point in time, the value of these corporations will be lower than the value of their outstanding debt. Therefore, when there is more market volatility and hence more uncertainty, bond investors will require a higher level of compensation which is expected to result into a widening of / rise in corporate bond spreads (European Central Bank, 2005). However, looking at Model 7 in Appendix C with the inclusion of the interaction term of the announcement dummy variable with the first lag of the volatility index, the coefficient of this interaction term is -0.112, thus negative. One explanation for this difference in signs of coefficients between the interaction term and the volatility index variable is that the volatility index variable presents the average effect on the stock price index compromising banks, whereas the interaction term gives the effect of the volatility index on the stock price index compromising banks only in the months of an announcement. This could well be different. If the pattern is that more volatility leads to increases in the stock price index comprising banks over the years, then if there is one or more events, that are the opposite, it cannot necessarily be expected that these events dominate. The results for the interaction term are significant and in accordance with economic theory.

In contrast to Zaghini (2019), no evidence is found in this thesis in favor of the signaling channel as both the announcement dummy variable and the interaction term of the second lag of the dependent variable with the announcement dummy variable are found not to have a significant effect on the stock price index comprising banks. This is also not in line with expectations based on Valiante (2015), that found that the announcement effect (which represents the main part of the signaling channel), led to a drop in interest rates for Eurozone government bonds, specifically the ones targeted by the asset purchase programme. Moreover, he found evidence for a decrease in the differential between AAA-rated bonds and other government bonds in the Euro Area. So, despite the notion that the signaling channel is known to be a strong channel for assets and bonds issued by governments, it seems to be weak in working for corporate bonds in this research. Limitations of this research could be an explanation for this result. However, this could also be an indication that the working of the signaling channel are less present for corporate bonds compared to the portfolio rebalancing channel and the exchange rate channel. Due to the limited amount of research available on the effect of this specific transmission channel on corporate bond spreads, this is still open for question.

Looking at the empirical results for Model 4 without the inclusion of PEPP, there are some differences compared to the models with the inclusion of PEPP. Some variables, such as the volatility index, are not significant anymore. Others, for instance the aggregate corporate bond holdings, are only significant at a lower significance level (10 percent). There are various explanations possible. It could be an indication of endogeneity, or, for example, the result of less precise results due to the reduced number of observations available. It could also be an indication of variances in the PEPP programme compared to the other asset purchase programmes. Further research would be necessary in order to draw meaningful interpretations on this asset purchase programme.

*Limitations and future research *

For this thesis, an event study with monthly time series data is used to approach the research question as daily data were not available (accessible) for all variables included in this research.

Event studies rely on the hypothesis that announcements are not anticipated and moreover, that stock markets adjust to new information immediately (Fama, 1969). They rely on certain market conditions on the day of the announcement and therefore the choice of events is very important as it can have strong effects on the study itself (Evanoff, Kaufmann & Malliaris, 2019).

Consequently, when using event studies on quantitative easing, daily data are often used in order to get as precise results as possible and to limit the amount of contamination of the data by other news than the quantitative easing announcements (Haitsma, Unalmis, & de Haan, 2015). In this thesis, by using monthly data, the risk of contamination is higher due to for instance the higher probability of included months that have overlap in announcement dates and actual implementation dates. For that reason, the specific market conditions on the day of the announcement cannot be as precisely measured as might be needed and might be blurred with market conditions from other days than the announcement day in the same month. This could have impacted and perhaps even contaminated the results and makes the identification of causal interpretations more difficult.

Furthermore, throughout all models, lagged values were used to control for reversed causality and to solve possible problems of endogeneity. However, there is still a possibility that the problem of endogeneity is not fully resolved and controlled for. This could result in inconsistent and biased coefficients and might explain why the sign of coefficient of the volatility index is not in line with

existing literature and why the sign of coefficient of the announcement dummy variable changed in various models (Lamoen, Mattheussens & Dröes, 2017).

The findings presented in this paper add to a growing body of literature on the effects and impact of asset purchase programmes on corporate bond spreads in the euro area. However, taking into account the above stated limitations, there are areas within this research that can still be improved.

Future research could focus on the use of daily data when using the methodology of an event study, like Neugebauer (2018) or Urbschat and Watzka (2020), to limit contamination and to get more precise results. This could perhaps help in finding more evidence in favor of or against the weak workings of the signaling channel found in this thesis. Furthermore, the model without the inclusion of the PEPP presented quite some variation in the results. Whether this is attributable to either the reduced number of observations and the limitations of the model which resulted in less precise or biased results, or to the specific results of the PEPP, is still open for question. Therefore, this leaves an important area for a possible future extension of the current analysis. For instance, it could be of interest to postpone research on the PEPP to a later point in time until more data on the PEPP has become available and can be gathered for research, to be able to see the effects on the PEPP more clearly. Additionally or alternatively, one could focus only on the announcements of the PEPP and the effects thereof on corporate bond spreads.

### 6. Concluding remarks

The objective of this thesis, while building upon previous research, was to determine whether the announcements of various asset purchase programmes by the ECB have an effect on corporate bond spreads over the period January 2015 to March 2021. Event study methodology following research of Neugebauer (2018) was used, where announcements of the various asset purchase programmes in the euro area were represented with a dummy variable equaling one in the month of an announcement and zero otherwise. Variables representing a selection of the transmission channels were also included. Additionally, control variables were added following suggestions of existing literature to limit the risk of endogeneity. In the end, OLS regressions were generated with the use of Stata.

The key finding of this research is that, while focusing on the effects of the announcements of asset purchase programmes on corporate bond spreads, evidence is found in favor of the workings of two transmission channels: the portfolio rebalancing channel and the exchange rate channel.

This is represented by the significance and (signs of) coefficients of the aggregate corporate bond holdings of the private sector (variable) and the Euro to US Dollar exchange rate (variable). This is in accordance with expectations and a growing body of literature on asset purchase programmes in the euro area. No evidence is found in favor of the signaling channel as the announcement dummy variable and the interaction term of the announcement dummy variable with the second lag of the stock price index comprising banks variable, did not have a significant effect on the stock price index comprising banks. Though the latter is known to be a strong channel for assets and bonds issued by the governments, it seems to be weak in working for corporate bonds in this research. Limitations of this research, by for instance using monthly data instead of daily data, could have had led to this interpretation. However, this could also be an indication that the working of the signaling channel are less present for corporate bonds compared to the workings of the portfolio rebalancing channel and the exchange rate channel. Similarly, results for the model without the inclusion of PEPP presented quite some variation in the results. Whether this is attributable to either the reduced number of observations and the limitations of the models or to the specific results of the PEPP, is also still open for question. Therefore, future research could focus on one, or both, of these areas.

In view of the importance of corporate bond spreads as indicator for future economic conditions and as only a limited amount of research is available on the topic of this thesis, the results and the suggestions for future research discussed in this paper may well be of interest for various organizations that are concerned with policy decisions and/ or are dealing with the aftermath of the COVID-19 crisis. For instance, the ECB and various national central banks will be interested in checking and confirming whether the responses to the COVID-19 crisis in the euro area have had a stabilizing effect on financial markets and to improve the responses for future use where necessary (Aguilar, Arce, Hurtado, Martínez-Martín, Nuño & Thomas, 2020). This requires extensive research into the effects and impact of, among others, asset purchase programmes on various financial indicators as studied in this paper. Obviously, in view of the limitations of this research, certain caution in applying these results should be exercised due to some highlighted contradictions and inconsistencies as presented in the results of this thesis. Nevertheless, the main conclusions could be very useful as a starting point for further research in the effectiveness of applied programmes by central and national banks.