The impact of central bank asset purchases on liquidity in
bond markets: A study for the Eurozone
Maximilian Oechsner, 11375574
July 1, 2017
University of Amsterdam, Amsterdam Business School
MSc Finance, Banking and Regulation Track
Master Thesis
Abstract
Do large-scale asset purchases by the European Central Bank improve or deteriorate liquidity in bond markets? This thesis examines how the recently launched asset purchase programs in the Eurozone affected liquidity in bond markets. By comparing how relative bid-ask spreads of targeted securities changed compared to not eligible securities, I find that liquidity in covered and government bonds deteriorated, while the liquidity of eligible corporate bonds was actually improved after the European Central Bank began to intervene. These findings are in line with the literature, which is torn over the effect of central bank asset purchases. Moreover, this research finds that the size of purchases is pivotal in affecting the liquidity of targeted securities as a comparison for the most recent intervention in the UK revealed.
Acknowlegdements
I would like to express my gratitude to my supervisor Dr. Tanju Yorulmazer for his useful comments and remarks throughout the process of this master thesis. I especially appreciate that he encouraged me to pursue the presented topic, as well as the discussions we had on the topic. Moreover, I would like to thank Dr. Maurice Bun and Rob Sperna for their unconditional commitment to assist students of this university.
Finally, I would like to thank my family and my girlfriend for their ongoing support during my whole studies. Without them, my course of education would not have been possible.
Statement of Originality
This document is written by student Maximilian Oechsner 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.
List of abbreviations
BOE Bank of England
BOJ Bank of Japan
CBPP3 Covered Bond Purchase Program 3
CDS Credit Default Swap
CSPP Corporate Sector Purchase Program EAPP Expanded Asset Purchase Program
ECB European Central Bank
FED Federal Reserve System
FRBNY Federal Reserve Bank of New York
FTSE Financial Times Stock Exchange Index ISIN International Securities Identification Number
MBS Mortgage-backed securities
LIBOR London Interbank Offered Rate
LTRO Long-term refinancing operations
POMO Permanent Open Market Operations
PSPP Public Sector Purchase Program
QE Quantitative easing
QE2 Quantitative easing 2
SMP Securities Markets Program
TIPS Treasury Inflation Protected Securities
UK United Kingdom
1
Inhaltsverzeichnis
1. Introduction ... 1 2. Literature review ... 3 2.1 Hypotheses... 6 3. Methodology ... 7 3.1 Data ... 7 3.2 Descriptive statistics ... 8 3.3 Empirical model ... 10 4. Results ... 12 4.1 Robustness ... 25 5. Conclusion / Discussion ... 27 6. References ... 30 7. Appendices ... 331
1. Introduction
“The global financial system is awash in liquidity, created by central banks as they have driven short-term interest rates to zero (or even below) and expanded their balance sheets by the equivalent of trillions of dollars. And so the world is swimming in cheap money. At the same time, liquidity is said to be at a low ebb in the financial markets, especially for bonds… As a result, transactions that once didn’t cause prices to budge now send them lurching from trade to trade… And the advice from central bankers on both sides of the Atlantic about this new volatility? Get used to it.”
The above statement is from Nouriel Roubini, a professor at the New York University and former consultant to the US Treasury Department, who warned of deteriorated liquidity in bond markets as a consequence of large-scale asset purchases by central banks from all over the world1. After Japan became the first country to implement quantitative easing (QE) in
2001, the US, the UK and Switzerland have followed suit. Most recently, the European Central Bank (ECB) launched a number of asset purchase programs to stem against lagging economic growth and low inflation in the Eurozone. Since the launch of its third Covered Bond Program (CBPP3) in October 2014, the ECB expanded its purchases to government as well as corporate bonds, and is still purchasing a significant amount of securities every month. Up until now, short- as well as long term rates of securities bought under the different asset purchase programs have fallen significantly. Moreover, the outstanding amount of government bonds with a positive yield from countries in the Eurozone has shrunk remarkably. In light of the aforementioned citation, the question now arises as to how liquidity in bond markets was affected by the recently implemented QE programs in the Eurozone. In particular, is the overall liquidity in bond markets increasing due to the appearance of a committed buyer who accepts a wide range of prices or is one market becoming more liquid while another market is becoming less liquid? This is the gap that this research wants to fill. Earlier studies have already extensively examined the impact of terminated purchase programs in the US as well as the Eurozone on liquidity in bond markets. A genuine discussion evolved on whether liquidity in targeted securities was improved or deteriorated. On the one hand, researchers in favor of an improvement argue that liquidity was enhanced due to the central banks’ appearance as a committed buyer. On the other hand,
1 The full article can be found here:
2
critics argue that a reduction in the outstanding stock of bonds leads to scarcity and thus more illiquidity. However, research on the recently implemented and still ongoing purchase programs in the Eurozone is scarce and the effect on liquidity is not clear. Moreover, a comparison of the effect between programs conducted in different economic areas has not yet to be implemented and provides implications for future asset purchase programs. This study adds to the existing literature in two ways: Firstly, it analyzes whether the ECB improved or deteriorated liquidity in targeted securities and how the liquidity of not eligible groups of bonds evolved in comparison. Second, a comparison to the recent asset purchases in the UK examines whether the size of asset purchase programs enhances their effect.
This study therefore uses a differences-in-differences model to investigate how the liquidity of targeted securities under the different purchase programs was affected by the central bank intervention compared to the liquidity of not-eligible bonds, which are exempt from purchases. I conducted the regressions for different announcements by the ECB over the last three years, starting with the CBPP3 in October 2014. Thereafter, I conducted the same regressions but for the announcement by the Bank of England (BOE) of the purchase of UK government and corporate bonds as a reaction to uncertainties after the UK referendum. Compared to the literature, this approach is novel, since it uses data on individual bonds of countries in the Eurozone and examines how liquidity has changed after the central bank intervention between treated and not treated bonds.
The remainder of this study is structured as follows. In chapter 2, an outlay of the most relevant literature and the ongoing discussion among researchers will be provided. Based on that, hypotheses were derived that will be presented. Chapter 3 describes the methodology and the data used in this study. In addition, summary statistics of the included variables will be provided. In chapter 4, the empirical results as well as additional robustness checks will be introduced. Finally, chapter 5 covers the conclusion drawn from this analysis and provides a discussion on the results and its implications. In the Appendix, additional tables and figures are provided.
3
2. Literature review
In 2001 Japan became the first country to implement QE. Since then, a huge body of research has developed on the impact of asset purchases on financial markets. The studies of Joyce et al. (2011), Gagnon et al. (2011) or of D’Amico and King (2013) are just three out of numerous examples that examine the effects of central bank intervention on asset prices as well as yields. However, research on the effects of QE on bond markets and their liquidity is still scarce, and has not been covered extensively.
Two opposing views have evolved throughout the existing literature on the impact of asset purchases on liquidity. Studies claim either that central bank intervention improves liquidity, or deteriorates it. The channel through which central bank intervention could thereby improve market liquidity is the liquidity, or market functioning, channel. This channel describes that central banks, acting as committed buyer with deep pockets, can allow dealers and investors to take larger positions in targeted securities since they can be sold to the central bank if needed. Consequently, investors require less liquidity risk premiums which could reduce assets’ yields (Gagnon et al.2011). Moreover, this could spill over to the liquidity of other assets as investors seek higher returns in higher yielding securities, which are not targeted by central banks. The subsequent rise in prices increases investors wealth, which induces them to spend it and thus stimulates trading (Beirne et al. 2011). In support of this theory, Krishnamurthy and Vissing-Jorgensen (2012) have shown that investors indeed value the liquidity of bonds. In their study, they analyzed how particularly a reduction in Treasury supply affects the spread between the Moody’s Baa-rated corporate bond yield and the yield of US Treasury bonds of same maturities. As they expected, a reduction in Treasury supply lowers their yields compared to corporate bond yields, which leads them to conclude that Treasury securities have a convenience yield for investors relative to other assets. Another paper by Christensen et al. (2011) analyzes whether Treasury Inflation Protected Securities (TIPS) purchases under the FED’s second quantitative easing program (QE2) led to a reduction in liquidity premium in TIPS and inflation swaps. By using a counterfactual analysis, they conclude that purchases reduced the liquidity premium for the duration of QE2. However, they find that the examined effect shrank towards the end of the program. Consequently, they argue that purchases indeed had an effect on targeted securities but only as long as purchases were continued, which reveals the programs’ limitations. With regard to the Eurozone, their findings provide implications for policymakers on how to taper the currently ongoing purchase programs in the future.
4
Another study by Pasquariello and Vega (2012) examines the impact of permanent open market operations (POMOs) by the Federal Reserve Bank of New York (FRBNY) on liquidity in U.S. Treasury bonds. They find that bid-ask spreads of on-the-run Treasury notes and bonds decline on days when the central bank intervenes. Moreover, they conclude that the magnitude of decline is greater when Treasury market liquidity is lower and uncertainty about the FRBNY’s policy aims is higher. Further, Gagnon et al. (2011) examine the effects of large scale asset agency debt and agency mortgage-backed securities (MBS) purchases by the US FED. Apart from a meaningful reduction in long term interest rates, they conclude that the FED improved market liquidity in these securities.
For Japan, a recent paper by Iwatsubo and Taishi (2016) studies how asset purchases by the Bank of Japan (BoJ) influence liquidity in government bond markets. Moreover, they try to evaluate whether conditions exist that could prevent a reduction in market liquidity. Their main finding is that a change in the purchasing policy, from April 2013 onwards, improved market liquidity. The changes included, among others, less variability in purchase amounts, which particularly reduced market uncertainty. Their findings can be related to the results obtained in this study, which tries to capture the relevance of the size of asset purchase programs.
Besides studies on QE programs in the US or Japan, more recent studies have proven that intervention by the ECB in the Eurozone, especially in the sovereign debt crisis and its direct aftermath have also improved market liquidity.
Pelizzon et al. (2016) for instance evaluate the interplay between credit risk and market liquidity during the sovereign debt crisis. Instead of asset purchases, their research focuses on the implementation of Long-term refinancing operations (LTRO). Nevertheless, they find that that ECB intervention improved liquidity and thus market functioning significantly. Another study by Beirne et al. (2011) analyzes how the first Covered Bond Program (CBPP1) by the ECB affected primary and secondary markets in the Eurozone. They conclude that the CBPP1 was effective in restoring liquidity in covered bond markets, which had dried up in the sovereign debt crisis. With regard to this study, it will be interesting to see whether the third Covered Bond Program (CBPP3) has a similar positive impact.
De Pooter et al. (2016) shed light on the Securities Markets Program (SMP), which was conducted from 2010 to 2012 by the ECB in order to reactivate liquidity in dysfunctional markets. By extracting the liquidity premium component from sovereign bond yields, they
5
detect that purchases of sovereign bonds lower the embedded liquidity premium. They even find that the effect is partly lasting.
Contrary to the aforementioned papers, which plead for a positive effect of central bank intervention on bond liquidity, Kandrac and Schlusche (2013) find no flow effects in liquidity of Treasury Securities traded by the FED. According to them, this observation reveals that asset purchases by the FED have not affected market functioning in the US. Furthermore, other studies claim that especially large-scale asset purchases even deteriorate the liquidity of targeted securities. The channel through which these interventions could worsen market liquidity is the scarcity channel. The rationale is that a reduction in the stock of securities of specific maturities held by investors, by means of an official purchase program leads to a shortage of those assets. Consequently, the created excess demand puts downward pressure on yields of targeted securities, and makes it especially more difficult to acquire them (D’Amico et al. 2012). This pattern can be of importance, especially in the repo market when a dealer wants to close a short position and requires a specific bond.
Against this background, D’Amico et al. (2015) investigated the scarcity value of US Treasury bonds by analyzing the effect of security-specific demand and supply factors on repo rates. They show that asset purchases affect the special collateral repo rates of securities bought, which supports the existence of a scarcity channel. Specifically, a resulting reduction in available supply of a specific security induces investors to be willing to loan money at lower rates in exchange for the required security. In line with these results, Ferrari et al. (2016) claim that the launch of QE in the Eurozone has an impact on sovereign collateral scarcity premium. However, they also find that this effect is offset by the launch of the ECB securities lending program. Most recently, a paper by Schlepper et al. (2017) analyses the effects of bond purchases by the German Bundesbank on targeted German government bonds. Their approach is novel since they use individual transaction data. One of their main finding is that liquidity and thus market functioning in the German government bond market has deteriorated since the start of the Public Sector Purchase Program (PSPP) in 2015. In addition, they estimate an even greater price impact during periods of low liquidity, which reveals the effects reduced liquidity can have.
6 2.1 Hypotheses
Based on the existing discussion in research, the following hypotheses were developed, which can be tested.
H1: The EAPP by the ECB improves liquidity in targeted securities.
The rationale for this hypothesis is that existing research has already proven that purchase programs in the US improved market functioning. Moreover, studies for the Eurozone have shown that the ECB was successful in intervening in markets after the sovereign debt crisis. Since the recently implemented Expanded Asset Purchase Program (EAPP) is larger in size and still ongoing compared to already terminated purchase programs, it will be interesting to examine whether a comparable effect can be observed.
H2: The effect of asset purchases on the liquidity of bond markets was more pronounced in the Eurozone than in the UK.
The intuition for this hypothesis is that the Bank of England announced in August 2016 the additional purchase of £60bn. of outstanding gilts, and £10 bn. of sterling investment-grade corporate bonds, as a reaction to economic concerns in the aftermath of the UK referendum2.
This corresponds to 2.90% and 2.30% respectively of outstanding amounts of government and corporate debt securities that are being bought by the BOE. However, national central banks in the Eurozone purchase a larger bulk of outstanding amounts of government and corporate debt. Figures 1 to 3 in the appendix reveal that the recent purchases of government and corporate bonds in the Eurozone were more pronounced in relation to outstanding amounts, and that they were conducted over a longer time period.
2 The full press release can be found here:
7
3. Methodology
This section presents the data and methodology that was used to test the stated hypothesis. First, an overview of the data will be given and, more importantly, how the data was selected. Based on that, summary statistics of the used variables will be provided. Subsequently, an explanation of the empirical model will follow.
3.1 Data
Transaction level data on asset purchase programs is not available which rules out a detailed analysis of the impact of individual transactions. However, I used data on government as well as corporate bonds with remaining time to maturities grouped into short-, medium-, long- and ultra-long term to gauge the effect of central bank intervention on bond market liquidity. Short-term bonds have a remaining time to maturity of 1 to 2 years at the respective date of announcement, medium-term of 3 to 7 years, long-term of 8 to 13 years and ultra-long-term bonds will be redeemed in more than 13 years. The analyzed bonds were selected manually based on different criteria. First, they were classified as being eligible or not according to the eligibility criteria that the ECB and the BOE published within their different purchase programs3. Second, the majority of bonds in the Eurozone were selected based on individual
ISINs, which are purchased and then offered for securities lending by the different national central banks, who conduct the purchases 4. As a third source, I used the bond finding
platforms of the stock exchanges of Stuttgart and of Frankfurt to filter for covered bonds or not-eligible corporate bonds from the UK. To ensure consistency, bonds with option like features, floating rate bonds as well as inflation adjusted bonds, were all excluded. For the chosen bonds, I downloaded bid-, ask-, and mid- prices. In addition, the swap spread as a measure of credit risk was downloaded from Thomson Reuters Datastream. For government and corporate bonds from the UK, I chose a different measure since the prevalent swap curve in the UK is not the same as for the Eurozone and more importantly some bonds were lacking
3 Operational modalities of the expanded asset purchase program (EAPP), published on 22/01/2015:
https://www.ecb.europa.eu/press/pr/date/2015/html/pr150122_1.en.html
Details on the Corporate Sector Purchase Program (CSPP), published on 21/04/2016:
https://www.ecb.europa.eu/press/pr/date/2016/html/pr160421_1.en.html
Information on gilt purchases by the BOE:
http://www.bankofengland.co.uk/markets/Pages/apf/gilts/default.aspx\
Information on the corporate bond purchase scheme by the BOE:
http://www.bankofengland.co.uk/publications/Pages/news/2016/068.aspx
4 Further links to the different ISINs purchased offered for securities lending by each central bank can be found
under the bottom of the following page:
8
data. I therefore used the yield differential between each bond and the three-month British Pound Libor as a measure for credit risk. Further, the initial issue size of each bond was downloaded while the age and the remaining time maturity of each bond was calculated manually as the difference between the respective date of issuance and the maturity date. Table 1 in the Appendix provides a list of bonds used in this study. Data on the VSTOXX, the FTSE Implied Volatility index and country-specific equity indices was also downloaded from Datastream. Moreover, data on the government bonds of some countries were not sufficiently provided, and therefore had to be excluded. These countries were Lithuania, Latvia, Malta, Slovenia, Slovakia and Luxembourg. At last, the dates on which details on the eligibility of asset purchases were released, were manually collected from press releases by the ECB and the BOE.
3.2 Descriptive statistics
Table 2 displays summary statistics for the relative bid-ask spread, which was used as liquidity measure in this study. The bonds were grouped by their remaining time to maturity and bond affiliation. From the presented table, it becomes obvious that the mean is mainly increasing in maturity, which is in line with my expectations. Bonds with a longer time to maturity exhibit a wider relative bid-ask spread, since their prices are more sensitive. As a consequence, dealers charge higher spreads to compensate for the increased risk (Ejsing and Sihvonen 2009). There are however exceptions. The mean of long-term government bonds from the Eurozone (0.21%) is lower than for medium-term government bonds (0.38%). This observation is surprising but could be due to the on-the run or off-the run status of individual bonds. Pasquariello and Vega (2009) state for instance that recently issued US government bonds trade more liquidly than bonds which had been already issued. In addition, medium term covered bonds exhibit a lower relative bid-ask spread (0.25%) than medium term government bonds (0.38%) in the Eurozone. Although covered bonds are collateralized by a pool of assets such as mortgages, the fact that government bonds are significantly larger in issue size should outweigh. Again, this pattern could be due to sample selection. To continue, non-investment grade bonds, which exhibit higher credit risk, trade on higher spreads. The mean of medium term non-investment grade bonds used in the sample (0.83%) is more than twice as high as for eligible investment grade corporate bonds (0.38%). The maximum value of short-term non-investment grade bonds is even 4.35%, compared to 2.02% for investment grade bonds. This pattern is in line with my expectations that high-yield bond exhibit wider bid-ask spreads due to their credit rating (Chakravarty and Sarkar 1999). For the UK, a similar pattern is observable. Nevertheless, it should be mentioned that the mean of sterling
9
investment-grade corporate bonds is decreasing across maturities, contradicting expectations. This could be due to sample selection. In addition, a closer look at the Table 1 indicates that ultra-long term corporate bonds for instance were issued more recently.
Table 3 provides summary statistics for the independent variables that were included as controls, grouped by bond type and remaining time to maturity. Regarding the Credit Spread, which describes the yield differential between a bond’s yield and the prevalent swap curve or the three-month Libor for the UK respectively, it is striking that this measure for government bonds is mostly negative. This observation is in line with a speech by Benoît Cœuré, a Member of the Executive Board of the ECB, who claimed that this pattern arose due to a supply and demand imbalance in government bonds in particular after the announcement of the PSPP in January 2015 5. Prompted by his speech, I created the swap spread for a German
government bond that had a remaining time to maturity of 5-years in 2014, and compared its yield to the prevalent swap curve6. From the chart, it becomes obvious that the swap spread
had been indeed negative since 2014 and decreased sharply after the announcement of the PSPP in January 2015. Even more surprising is that the minimum of short-term non-investment grade corporate bonds from the Eurozone is below zero (-0.05%), which is notable since these securities exhibit a higher default probability. However, these securities exhibit, on average, a higher credit spread than investment-grade corporate bonds. The mean of medium term non-investment grade corporates of the Eurozone (3.29%) is more than three times as high as for investment-grade corporate bonds of comparable maturities (0.94%). Contrary to this, it is surprising that the mean of the credit spread is higher across maturities for government bonds than for bank bonds in the Eurozone. A closer look into the database revealed that this observation could be due to the impact of the government bonds of Cyprus, which suffered a severe banking crisis in 2013. On the variable Age, the only observation that should be mentioned is that the minimum is mostly zero. The rationale is that the selected bonds were issued at different points in time. With regards to the Issue size, government bonds are issued in larger size, in line with expectations that a government’s funding needs exceeds those of corporates. A surprising observation that should be mentioned is that the minimum of medium term government bonds in the Eurozone is only € 4 m. A closer lookninto my data revealed that it is a 5-year Cyprian government bond, which was issued at
5 The full speech can be found here:
https://www.ecb.europa.eu/press/key/date/2017/html/sp170403_1.en.html
10
this size. Further statistics are provided for the change in the Volatility index of the Eurozone and the UK as well as for the Equity returns. It should be noted that the minimum and the maximum value of the VSTOXX and the FTSE VIX index indicate that these indices changed sharply over the sample period
3.3 Empirical model
To test the aforementioned hypotheses, I used a differences in differences regression based on panel data as follows.
𝑆𝑝𝑟𝑒𝑎𝑑𝑖𝑡 = 𝛽0+ 𝛽1𝑇𝑟𝑒𝑎𝑡𝑒𝑑𝑖+ 𝛽2𝑇𝑖𝑚𝑒𝑡+ 𝛽3𝐷𝑖𝐷𝑖𝑡 + 𝛽4𝐶𝑟𝑒𝑑𝑖𝑡 𝑆𝑝𝑟𝑒𝑎𝑑𝑖𝑡 + 𝛽5 𝑇𝑖𝑚𝑒𝑡𝑜𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦𝑖𝑡
+ 𝛽6 𝐴𝑔𝑒𝑖𝑡+ 𝛽7 𝐼𝑠𝑠𝑢𝑒 𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽8 𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦 𝐼𝑛𝑑𝑒𝑥𝑖𝑡 + 𝛽9 𝐸𝑞𝑢𝑖𝑡𝑦 𝑅𝑒𝑡𝑢𝑟𝑛𝑠𝑖𝑡+ 𝜀𝑖𝑡
The dependent variable, 𝑆𝑝𝑟𝑒𝑎𝑑𝑖𝑡 , is the relative bid-ask spread of each bond, determined by dividing the difference of the ask and the bid price by the mid-price. Whether a bond belongs to the treatment or the control group is determined by the dummy variable 𝑇𝑟𝑒𝑎𝑡𝑒𝑑𝑖, which takes the value one if security 𝑖 is treated and zero if not.
The variable 𝑇𝑖𝑚𝑒𝑡 takes value one after the respective intervention was announced. Mostly the first announcement of central bank asset purchases was chosen. Exceptions are the announcement date of the CBPP3 and the CSPP. I rather chose the dates on which the ECB revealed further details on eligibility criteria than announced the program for the first time. Table 4 in the appendix provides an overview of all event dates and of the bond groups chosen either as treatment or as control group.
The variable of interest is the 𝐷𝑖𝐷𝑖𝑡 interaction term of the 𝑇𝑖𝑚𝑒𝑡 and the 𝑇𝑟𝑒𝑎𝑡𝑒𝑑𝑖 dummy. In more detail, the coefficient, 𝛽3, of this variable measures the difference in outcome from
before to after the onset of the intervention between the treatment- and the control group. In addition, the difference in the respective means before the treatment is removed, so that it can be regarded as a pure measure of the change among the groups. In light of my hypotheses, I expect the coefficient to be negative, in favor of an improvement in liquidity of purchased securities. The remaining variables are included as controls. The variable 𝐶𝑟𝑒𝑑𝑖𝑡 𝑆𝑝𝑟𝑒𝑎𝑑𝑖𝑡
describes the spread between an individual bond’s yield and the swap rate of the currency in which the bond is denominated. For the UK, the difference between each bond’s yield and the three-month Libor rate was used. The variables 𝑇𝑖𝑚𝑒𝑡𝑜𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦𝑖𝑡 and 𝐴𝑔𝑒𝑖𝑡 display days to maturity of a bond as well as its lifetime since issuance over the sample period. I expect the coefficients of these three variables to be positive. Research has proven that the relative
bid-11
ask spread is higher for bonds of lower credit rating, longer time to maturity and longer life since issuance (Chakravarty and Sarkar, 1999). Moreover, the outstanding amount of a bond reduces over its lifetime because a large part is absorbed in inactive portfolios, which leads to more illiquidity (Hotchkiss and Jostova, 2017). Contrary to that, I expect the coefficient of the variable 𝐼𝑠𝑠𝑢𝑒 𝑆𝑖𝑧𝑒𝑖𝑡 to be negative since earlier studies have shown that in particular the liquidity of corporate bonds increases in issuance size (Bao et al. 2011). Further evidence was provided by Ejsing and Shivonen (2009), who observed that large issues in the German government bond market are traded more frequently. I will take the natural log of this variable. At last, I included the variables 𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦 𝐼𝑛𝑑𝑒𝑥𝑖𝑡 and 𝐸𝑞𝑢𝑖𝑡𝑦 𝑅𝑒𝑡𝑢𝑟𝑛𝑠 𝑖𝑡 of the respective country in which a bond is issued to account for volatility and business sentiment. I expect the former to have a positive impact on the relative bid-ask spread since liquidity and volatility shocks are positively correlated (Chorida et al. 2004). For the latter, I expect a positive relationship as well because research has proven that the equity and the bond market are positively correlated (Goyenko and Ukhov 2009).
This approach will be used to estimate the effect of the purchase program announcements on different asset classes in the Eurozone as well as the UK.
12
4. Results
In this section, I will explain the empirical results for the conducted differences-in-differences regression and whether they confirm or contradict the previously stated hypotheses. First, I will use a simplified graph to show how the relative bid-ask spread of the treatment and the control group has changed from before to after the onset of the respective program in the Eurozone as well as the UK. Second, I will evaluate empirically whether the difference is statistically significant. Thereby, I will proceed chronologically, starting with the CBPP3 in the Eurozone. Table 5 in the appendix presents the results.
Covered Bond Program 3 – Covered bonds vs. Investment-grade corporate bonds
Beginning with the announcement by the ECB to launch its third covered bond program (CBPP3) in October 2014, I analyzed whether this program affected the liquidity of eligible covered bonds, compared to the liquidity of investment-grade corporate bonds, which were not eligible for asset purchases at this time. Hence, the former comprises the treatment group while the latter is the control group. The pre-treatment period extends from the 3rd of March
2014 until the ECB announced operational details of its third covered bond program on the 2nd of October 2014. This marks the onset of the post-treatment period, which extends until
the 1st of May 2015.
For illustration purposes, Figure 5 below shows the mean of the relative bid-ask spread of both groups before and after the intervention.
13
Figure 5: Change in the mean of the relative bid-ask spread from before to after onset of the CBPP3
This chart shows the mean of the relative-bid ask spread of the treatment group and the control group before and after the announcement of the CBPP3. The vertical axis displays the mean of the relative bid-ask spread in percent while the horizontal axis indicates the before and after period. The treatment group (red line) comprise covered bonds while the control group (blue line) are investment-grade corporate bonds. The dashed line presents how the mean of the treatment group would have changed without the intervention and the bracket shows the DiD estimator. The latter is the difference in outcome from before to after the onset of the intervention between the treatment- and the control group.
Source: Own calculations
From the chart, it can be seen that the mean of the relative bid-ask spread of the treatment goup is obviously higher after the onset of the intervention than before, while the mean of the control group is slightly lower. The dashed line indicates how the mean of the treatment group would have changed without the intervention. In addition, the bracket indicates the DiD estimator, which is the difference in outcome from before to after the intervention between the treatment and the control group. The graph therefore suggests that liquidity in covered bonds would have improved without the intervention.
To examine whether this observation is statistically significant, Table 5 column 1 presents the regression results. 0.00% 0.01% 0.02% 0.03% 0.04% 0.05% 0.06% 0.07% 0.08% 0.09% Before After
Covered Bond Porgram 3
Covered bonds Investment-grade corporate bonds DiD
14
First, the coefficient of the variable Time is the expected mean change in outcome from before to after the announcement within the control group. It is negative (-0.000426) and highly significant at a 1% level. Hence, the mean of the relative bid-ask spread of the control group in the period after is lower, compared to the period before.
Next, the coefficient for the variable Treated describes the difference in the mean of the relative bid-ask spread between the treatment and the control group prior to the intervention. A negative relationship persisted, as the highly significant (1% level) coefficient of -0.0679 indicates. As expected, covered bonds’ relative bid-ask spreads were on average 0.0679% narrower than those of corporate bonds before the launch of the CBPP3. For interest, the coefficient is the differences-in-differences estimator, DiD, which declares whether the mean change in outcome from before to after the announcement was different between the treatment and the control group. Against expectations, the estimated coefficient is positive (0.0336), and highly significant at a 1% level. This observation in particular reveals that the relative bid-ask spread of covered bonds increased compared to the same liquidity measure of investment-grade corporate bonds after the ECB started to intervene. In light of the aforementioned hypotheses, this finding contradicts the idea that the launch of the CBPP3 improved liquidity in eligible covered bonds (Hypothesis 1). Even further, it is conceivable that the liquidity in these securities deteriorated through the scarcity channel, as Ferrari et al. (2016) claimed.
Apart from the aforementioned variables, the estimated coefficients for the variables Credit
Spread, Time to maturity and Age are all highly significant at a 1% level, and in line with
expectations. An increase in the Credit Spread by one percentage point leads to an increase in the relative bid-ask spread by 0.1344 percentage points as the estimated coefficient indicates. However, the coefficient should be interpreted with caution since Table 6 in the appendix reveals that the Relative bid-ask spread and the Credit Spread are positively correlated (0.5807) and therefore vary in the same direction. The coefficients for Time to maturity and
Age are small in magnitude (0.0000617 and 0.00000561) and indicate that an additional day
closer to maturity or further in time since issuance leads to an increase in the relative bid-ask spread by 0.0000617% and 0.00000561% respectively. The estimated coefficient for the variable Issue size (-0.0102), is highly significant at a1% level and states that a 1% increase in the issue size leads to a decrease in the relative bid-ask spread. Further, the coefficients for the variables Volatility index (0.000107) and Equity returns (0.00129) point towards a positive impact on the used liquidity measure, but the coefficients are not statistically significant.
15
The estimated R-squared (0.3376) reveals that 33.76% of the variance in the dependent variable could be explained by the included independent variables.
Public sector purchase program – Government bonds vs. Investment-grade corporate bonds
In January 2015, the ECB expanded its asset purchase programs to government bonds across the Eurozone by launching the PSPP. To gauge the effect of this program on eligible government bonds, I conducted a similar regression as before. In this setup, government bonds comprise the treatment group, while investment-grade corporate bonds are again chosen as the control group. The period before the announcement of the program extends from the 5th of May 2014 until one day before the announcement in January 2015, while the
period after starts with the announcement on the 22nd of January 2015 and lasts until the 5th of
September 2015. Figure 6 below displays the mean of the relative bid-ask spreads of the treatment group and the control group before and after the launch of the PSPP in January.
Figure 6: Change in the mean of the relative bid-ask spread from before to after the onset of the PSPP
This chart shows the mean of the relative-bid ask spread of the treatment group and the control group before and after the announcement of the PSPP. The vertical axis displays the mean of the relative bid-ask spread in percent while the horizontal axis indicates the before and after period. The treatment group (red line) comprise government bonds while the control group (blue line) are investment-grade corporate bonds. The dashed line presents how the mean of the treatment group would have changed without the intervention and the bracket shows the DiD estimator. The latter is the difference in outcome from before to after the onset of the intervention between the treatment- and the control group.
Source: Own calculations
0.50% 0.55% 0.60% 0.65% 0.70% 0.75% 0.80% Before After
Public Sector Purchase Program
Investment-grade corporate bonds Government bonds DiD
16
The presented figure shows that the mean of the relative bid-ask spread of government bonds was higher after the intervention than before. On the contrary, the mean of investment-grade corporate bonds seems to be lower in comparison to the pre-treatment period. The dashed line presents how the spread of government bonds would have developed without the intervention. From the visible difference in outcome, it could be assumed that the relative bid-ask spread of government bonds increased, and these securities became thus more illiquid, compared to the control group.
Column 2 of Table 5 presents the statistical evidence. In line with the aforementioned results, the estimated coefficient for the Time variable is negative (-0.01227) and statistically significant. Again, I can therefore assume that investment-grade corporate bonds became more liquid. The estimated coefficient for the Treated dummy is negative and of a greater magnitude (-0.13892). Moreover, the coefficient is highly significant at a 1% level. This finding is in line with expectations that government bonds are considerably more liquid than corporate bonds. As could be found for covered bonds, the estimated coefficient for the DiD variable reveals that government bonds became less liquid compared to investment-grade corporate bonds. The coefficient is positive (0.0424) and highly significant at a 1% level. This finding once again contradicts again my hypothesis that the ECB improved liquidity in targeted securities with its asset purchase programs. The PSPP therefore did not affect the liquidity of government bonds through the market functioning channel, but rather through the scarcity channel.
The remaining control variables are in line with expectations. In particular, the variables Time
to maturity and Age have a positive impact on the relative bid-ask spread, while a higher Issue Size leads to narrower spreads. All coefficients are statistically significant at a 1% level.
The included Volatility Index, as well as the variable Equity returns, have a positive impact, although their coefficients are not significant.
Public sector purchase program - Government bonds vs. Bank bonds
The previously mentioned regressions have shown that covered and government bonds issued in the Eurozone became less liquid due to the intervention by the ECB, compared to investment-grade corporate bonds. The focus in this regression is on not eligible bonds issued by banks and how their liquidity evolved compared to eligible government bonds. These bonds therefore comprise the control group, while government bonds exhibited the treatment.
17
The pre- and post-treatment period in this setup are set to be the same as for the previous regression. Figure 7 below once again describes the mean of the relative bid-ask spread of treated government bonds, as well as of bonds issued by banks before and after the onset of the purchase program.
Figure 7: Change in the mean of the relative bid-ask spread from before to after the announcement of the PSPP
This chart shows the mean of the relative-bid ask spread of the treatment group and the control group before and after the announcement of the PSPP. The vertical axis displays the mean of the relative bid-ask spread in percent while the horizontal axis indicates the before and after period. The treatment group (red line) comprise government bonds while the control group (blue line) are bank bonds. The dashed line presents how the mean of the treatment group would have changed without the intervention and the bracket shows the DiD estimator. The latter is the difference in outcome from before to after the onset of the intervention between the treatment- and the control group.
Source: Own calculations
The presented graph shows a similar pattern as described before. The mean of the relative bid-ask spread of government bonds is higher after the intervention while it is lower for bank bonds. It is again conceivable to assume that government bonds became less liquid, as the dashed line indicates. Column 2 in table 5 therefore presents the results for the conducted regression. The estimated coefficient of the variable Time is negative (-0.00172) but not statistically significant. Secondly, in the sample included, government bonds exhibited narrower bid-ask spreads compared to bank bonds, as the negative coefficient for the variable
0.40% 0.45% 0.50% 0.55% 0.60% 0.65% 0.70% Before After
Public Sector Purchase Program
Government bonds Bank bonds
18
Treated (-0.167) implies. The latter is highly statistically significant (at a 1% level), and in
line with the prior findings. Similar to the previously explained findings, the estimated coefficient for the DiD interaction term is positive (0.0064), and significant at a 5% level. With regard to the hypotheses of this research, this finding vindicates the prior finding that the ECB did not improve liquidity in government bonds by purchasing them in large size. (Hypothesis 1). The control variables Credit Spread, Time to Maturity, Age and Issue Size have a positive impact on the relative bid-ask spread before, while also being highly statistically significant at a 1% level. Beyond this, the estimated coefficients for the variables
Volatility index and Equity returns are not significant.
Corporate sector purchase program – IG vs. Non-IG corporate bonds
Although the CSPP was already announced on the 10th of March 2016, I chose the release of
the eligibility criteria by the ECB on the 21st of April 2016 as the crucial event for this
research. The chosen pre-treatment period therefore extends from the 1st of June 2015 to the
1st of April 2016, while the period from the announcement date until the 31st of March 2017
represents the post-treatment period. Securities in the treatment group are investment-grade corporate bonds that are denominated in Euro and fulfill the requirements of the purchase program. The control group comprises non-investment grade corporate bonds that have a credit rating below investment grade, and are therefore not eligible. As before, Figure 8 below shows the mean of the relative bid-ask spread in both groups before and after the intervention.
19
Figure 8: Change in the mean of the relative bid-ask spread from before to after the onset of the CSPP
This chart shows the mean of the relative-bid ask spread of the treatment group and the control group before and after the press release on eligibility criteria of the CSPP. The vertical axis displays the mean of the relative bid-ask spread in percent while the horizontal axis indicates the before and after period. The treatment group (red line) comprise investment-grade corporate bonds while the control group (blue line) are non investment-grade corporate bonds. The dashed line presents how the mean of the treatment group would have changed without the intervention and the bracket shows the DiD estimator. The latter is the difference in outcome from before to after the onset of the intervention between the treatment- and the control group.
Source: Own calculations
From the graph, it can be assumed that the means of both groups increased, and thus became more illiquid.
The dashed line indicates however that the mean of the treatment group increased less. Column 4 in Table 5 underpins this observation. Non investment-grade corporate bonds did
indeed become less liquid, as the highly significant (1% level) estimated coefficient (0.0632) for the Time dummy reveals. Next, the coefficient for the Treated dummy (-0.171) is significant at a 1% level as well, and states that the relative bid-ask spread among the treatment group was on average 0.171% narrower than among the control group before the intervention. As expected, corporate bonds with a higher credit rating traded more liquidly. The estimated coefficient for the DiD interaction term is negative (-0.0165), and significant at a 5% level. The liquidity of investment-grade corporate bonds therefore improved compared to that of non-investment grade corporate bonds. This finding supports hypothesis 1 that the
1.00% 1.05% 1.10% 1.15% 1.20% 1.25% 1.30% 1.35% 1.40% Before After
Corporate Sector Purchase Program
Non investment-grade corporate bonds Investment-grade corporate bonds DiD
20
intervention by the ECB improves liquidity in targeted securities. Apart from that, the coefficients for the control variables are in line with the prior findings. It should be noted that the estimated coefficient for Equity returns (-0.0046) is now significant at a 5% level compared to the two regressions conducted before, in which the equity market had no significant impact on the dependent variable.
UK – Gilt purchases
In the wake of the UK referendum in June 2016, the Bank of England announced on the 4th of
August 2016 that it will expand its asset purchase facility by £60 bn. in UK government bonds (gilts), and £10 bn. of sterling investment grade corporate bonds. To examine whether these securities became more liquid, compared to not eligible securities and whether the effect is comparable to the Eurozone, I conducted the same regressions as presented before. Therefore, I used the described announcement as the event date to examine at first whether UK gilts (treatment group) became more liquid, compared to not-eligible corporate bonds (control group). The latter comprise a sample of bonds issued by banks, building societies and insurance companies which are exempt from the program according to the BOE 7. The period
before the onset of the intervention extends from the 1st of June 2015 to August 2014,
followed by the post-treatment period from the event date onwards to the 1st of April 2017.
Figure 9 below displays the means of the relative bid-ask spreads of both groups before and after the announcement.
7 Further information on the eligibility criteria can be found under the following link:
21
Figure 9: Change in the mean of the relative bid-ask spread from before to after the announcement by the BOE
This chart shows the mean of the relative-bid ask spread of the treatment group and the control group before and after the announcement by the BOE. The vertical axis displays the mean of the relative bid-ask spread in percent while the horizontal axis indicates the before and after period. The treatment group (red line) comprise UK government bonds (gilts) while the control group (blue line) are bonds issued by banks, insurances and building societies (bank bonds). The dashed line presents how the mean of the treatment group would have changed without the intervention and the bracket shows the DiD estimator. The latter is the difference in outcome from before to after the onset of the intervention between the treatment- and the control group.
Source: Own calculations
The presented chart suggests that the mean for UK gilts was slightly higher after the intervention than before. In addition, the dashed line indicates that government bonds would have been more liquid without the intervention. On the contrary, the red line (control group) indicates that ineligible corporate bonds became more liquid. Another observation that can be drawn from the graph above is that the bonds included in the treatment group traded on wider bid-ask spreads before and after the intervention compared to the control group. This pattern is surprising, since I expected government bonds to exhibit more liquidity than bonds issued by banks, but this could be due to sample selection and the on-the-run or off-the-run status of these securities.
To prove statistical significance, Column 5 in table 5 provides the prevalent results. The estimated coefficient for the Time dummy is negative (-0.0269), but not statistically significant. I can therefore not assume that the bonds in the control group became more liquid
3.50% 4.00% 4.50% 5.00% 5.50% 6.00% Before After
UK gilt purchases
Government bonds Bank bonds
22
over the sample period, as Figure 9 suggested. Next, the coefficient for the Treated dummy is positive (1.36), and highly significant at a 1% level, which confirms that the chosen government bonds exhibited higher relative bid-ask spreads than ineligible bonds before the announcement.
The main coefficient of interest, DiD, is positive (0.03), but not significant. This leads me to conclude that the announcement by the BOE of the purchase of an additional amount of £ 60bn. in UK gilts neither improved nor deteriorated the liquidity of these securities, contradicting hypothesis 1. Moreover, hypothesis 2 can be verified, since the effect of the asset purchases under PSPP on the liquidity of targeted government bonds in the Eurozone was highly significant. The estimated coefficients of the remaining variables are in line with prior findings and expectations. Only the coefficients for the Volatility Index and Equity
Returns are not significant.
UK – Corporate Bond Purchase scheme
After the above stated results show that the liquidity of government bonds was not affected significantly by the BOE purchases, I examined further whether liquidity in eligible corporate bonds was affected by the same announcement over the sample period. Since I found in column 4 of Table 5 that the launch of the CSPP in the Eurozone improved liquidity in targeted securities, it will be interesting to analyze whether a comparable effect in significance and magnitude can be observed.
Therefore, Figure 10 below displays the mean of the relative bid-ask spread of the treatment and the control group before and after the announcement on the 4th of August 2016.
23
Figure 10: Change in the mean of the relative bid-ask spread from before to after the announcement by the BOE
This chart shows the mean of the relative-bid ask spread of the treatment group and the control group before and after the announcement by the BOE. The vertical axis displays the mean of the relative bid-ask spread in percent while the horizontal axis indicates the before and after period. The treatment group (red line) comprise eligible corporate bonds from the UK while the control group (blue line) are bonds issued by banks, insurances and building societies (bank bonds). The dashed line presents how the mean of the treatment group would have changed without the intervention and the bracket shows the DiD estimator. The latter is the difference in outcome from before to after the onset of the intervention between the treatment- and the control group.
Source: Own calculations
From the chart, it can be assumed that the mean of the relative bid-ask spread of the control group was slightly higher for investment-grade corporate bonds (treatment group) after the announcement, which would mean that these securities became less liquid. The dashed line suggests that the intervention by the BOE deteriorated liquidity in eligible corporate bonds. At the same time, the mean of not eligible bonds issued by banks or building societies (control group) seems to have decreased, in favor of a liquidity improvement in these securities.
Subsequently, column 4 of Table 5 presents the results for the conducted regression. The estimated coefficient for the Time dummy is negative (-0.0025) but not significant. Therefore, I cannot conclude that the mean of the relative bid-ask spread of the control group decreased over the sample period. On the contrary, the coefficient for the Treated dummy is negative
(-3.50% 3.60% 3.70% 3.80% 3.90% 4.00% 4.10% 4.20% 4.30% 4.40% 4.50% Before After
UK corporate bond purchases
Investment-grade corporate bonds Bank bonds DiD
24
0.280) and highly significant at a 1% level. Investment-grade corporate bonds therefore exhibited lower relative bid-ask spreads in the pre-treatment period than ineligible bonds. The coefficient for the variable of interest, DiD, is positive (0.0549) but not statistically significant. Similar to the regression results presented before, I can neither infer that eligible corporate bonds became less nor more liquid compared to ineligible corporate bonds. In light of the hypotheses, this finding contradicts once again hypothesis 1 that the BOE improved liquidity in targeted corporate bonds. In comparison to the Eurozone, the findings support again hypothesis 2 that the effect on eligible corporate bonds was more pronounced in the Eurozone. I therefore assume that the size and the length of the CSPP in the Eurozone are crucial in stimulating liquidity in eligible securities. Apart from the described variables, the coefficients for the included control variables are in line with expectations and the previous findings. The only exception is the Volatility Index, which exhibits a positive coefficient but is however not significant.
25 4.1 Robustness
In this section, I will explain the additional results of the robustness checks that I performed to ensure internal validity of the model.
Firstly, I verified graphically that the treatment and the control group had the same trend before the respective interventions took place. As an illustration, Figure 11 in the appendix reveals that the means of the relative bid-ask spread of government bonds and of investment-grade corporate bonds in the Eurozone were decreasing before January 2015. It also becomes obvious that the relative bid-ask spread of government bonds shot up after the announcement of the PSPP, in line with the presented results in Table 5 columns 2 and 3.
Secondly, I conducted the same regressions but with a shorter pre-treatment period to gauge whether the effect in fact took place after the respective announcements. At last, I regressed the relative bid-ask spread on the Time dummy, but only for the control group in order to verify that these securities were not affected by the treatment. In light of the asset purchase programs, this was one of my concerns, since the ECB launched its LTRO program already in 2014, which gave room for speculation on further asset purchases.
Table 7 in the appendix presents the obtained results of the regressions with shorter pre-treatment periods. Column 1 comprises the results around the onset of the CBPP3 in October 2014. Eligible covered bonds from the Eurozone account for the treatment group, while the control group comprises investment-grade corporate bonds. The period before the treatment was set to start from the 7th of July 2014. From the results, it becomes obvious that the
estimated coefficients are of similar magnitude and significance. In line with the results presented in table 5, covered bonds in the treatment group became less liquid compared to investment-grade corporate bonds as the highly significant coefficient (1% level) of the DiD variable states. For the second and third regressions (columns 2 and 3), which analyze whether government bonds became more liquid compared to corporate bonds and bank bonds after the launch of the PSPP in January 2015, the pre-treatment period was shortened to start from the 3rd of November 2014. Again, the magnitude and significance of the estimated
coefficients are in line with the prior presented results, which verifies the model. Column 4 presents the results for the conducted regression around the launch of the CSPP. The pre-treatment period was shortened to start from the 4th of January 2016. Finally, Column 5 and 6
present the results for the analysis of the BOE announcement in August 2016. The pre-treatment period was shortened to start from the 7th of March 2016 and the estimated
26
As a second test, I conducted an OLS regression but only for the control group. The regression equation therefore looks as follows:
𝑆𝑝𝑟𝑒𝑎𝑑𝑖𝑡𝑐 = 𝛽0+ 𝛽1𝑇𝑖𝑚𝑒𝑡+ 𝛽2𝑆𝑤𝑎𝑝𝑖𝑡 + 𝛽3𝑇𝑖𝑚𝑒𝑡𝑜𝑀𝑎𝑡𝑢𝑟𝑖𝑡𝑦 𝑖𝑡
+ 𝛽4𝐴𝑔𝑒𝑖𝑡+ 𝛽5 𝐼𝑠𝑠𝑢𝑒 𝑆𝑖𝑧𝑒𝑖𝑡+ 𝛽8𝑉𝑜𝑙𝑎𝑡𝑖𝑙𝑖𝑡𝑦 𝐼𝑛𝑑𝑒𝑥𝑡+ 𝛽9 𝐸𝑞𝑢𝑖𝑡𝑦 𝑅𝑒𝑡𝑢𝑟𝑛𝑡 + 𝜀𝑖𝑡
The dependent variable is again the relative bid-ask spread while the variable of interest is the
Time dummy. This dummy takes the value zero in the period before the intervention and one
in the period after. With this test, I want to check that my control groups were not affected by the intervention. In addition, the remaining control variables are the same as for the regressions conducted before. Table 8 in the appendix presents the results. Again, column 1 displays the results for the conducted regressions around the onset of the CBPP3, while column 3 and 4 present the results around the announcement of the PSPP. Columns 4 displays the results for non-investment grade bonds and columns 5 and 6 correspond to the intervention in the UK. In line with my expectations, the estimated coefficients exhibit the similar coefficients but are not significant. Moreover, the coefficients for the variables Time
to maturity, Age and Issue Size have the same impact on the relative bid-ask spread of the
control group as for the whole sample, and are all significant. I can therefore infer that the control groups were well selected and are not affected by the treatment in form of asset purchases.
27
5. Conclusion / Discussion
At the outset, I positioned this thesis in existing literature, which is in dispute as to whether large-scale central bank asset purchases improve or deteriorate liquidity in bond markets. Furthermore, since a huge body of literature has examined the effects of already terminated programs in the US or Japan, my research sheds light on the more recent period of QE in the Eurozone. Therefore, I proposed two main hypotheses: (1) The EAPP by the ECB improves liquidity in targeted bond markets, (2) The effect of asset purchases on liquidity in bond markets was more pronounced in the Eurozone than in the UK. To test these hypotheses, I conducted six differences-in-differences regressions, with the relative bid-ask spread as a measure of liquidity to gauge the effect of purchase programs on eligible and thus treated securities compared to ineligible securities. Firstly, I analyzed how the liquidity measure of covered and government bonds in the Eurozone changed after the onset of the respective programs, compared to investment-grade corporate bonds. The relative bid-ask spread of the latter decreased over the sample periods, while covered and government bonds exhibited a higher spread after the intervention. The estimated coefficients were highly significant and robust with regards to the tests conducted in the corresponding section. I can therefore conclude that asset purchases by the ECB did not improve liquidity in covered and government bonds, which contradicts hypothesis 1. Moreover, these results clearly support the findings of Schlepper et al. (2017) who found that asset purchases by the German Bundesbank, under the PSPP, impact liquidity in German government bonds adversely. Even further, I can therefore assume that liquidity in these securities deteriorated, and scarcity may play a role. In contrast to that, I found that the intervention by the ECB in the corporate bond market did indeed affect eligible investment-grade corporate bonds positively. The latter became more liquid than non-investment grade corporate bonds. Hence, I can infer that the CSPP actually improved liquidity in targeted securities through the liquidity channel, in line with hypothesis 1. With regard to the existing literature, this finding is in line with various papers, such as Pasquariello and Vega (2012), who found that large scale asset purchases by the US FED improved liquidity in agency debt and mortgage-backed securities. Apart from the Eurozone, I also analyzed whether the latest announcement by the BOE of the expansion of its asset purchase facility by an additional amount in UK gilts and corporate bonds had a similar effect. In line with my expectations and hypothesis 2, the effect was more pronounced in the Eurozone due to the larger scale of the purchase programs. Even further, the effect on
28
UK gilts and eligible corporate bonds was not significant. Based on the aforementioned findings, the main implications of this thesis are that policy makers should assess the monthly size and pace of their asset purchases. Furthermore, the findings suggest that purchase programs of larger scale are more effective, as can be seen from the comparison to the UK.
However, the scope of this thesis was limited due to data availability. Although Schestag et al. (2016) found that the relative bid-ask spread is a valid measure for liquidity in bond markets, transaction level data could be used to create other liquidity measures such as Roll’s estimator or the Amihud illiquidity measure. MTS Markets, a fee-based inter-dealer transaction data base, provides an overview of individual transactions and bid-ask prices of MTS market participants, consisting mainly of banks and bond dealer desks. Earlier research has argued that data from the inter-dealer market can be used to measure liquidity in dealer-customer markets, since spillover effects will occur if liquidity in the former deteriorates (Schlepper et al. 2017). Based on that, it would be interesting to evaluate the effect of individual transactions on the liquidity of bonds, either affected ones or near substitutes. The recently published study by Schlepper et al. (2017) uses individual transaction data from the German Bundesbank as well as MTS markets data. They found that asset purchases by the Bundesbank have a direct impact on bond prices in the inter-dealer market and that liquidity has deteriorated following the onset of the PSPP. Since their study analyses the impact of purchases on German government bonds only, an expansion to a Eurozone level, as well as to corporate bonds under CSPP, would add meaningful value to research and for policymakers. In light of the findings of this research, further studies could focus on whether scarcity in targeted securities can be verified. It is conceivable to assume that there is a timing effect. I could imagine that central bank intervention affects the liquidity of targeted securities positively in the short run, while scarcity effects could occur over the long run. D’Amico et al. (2015) analyzed already the impact of the US FED’s asset purchase programs on special collateral repo rates in targeted securities. Moreover, a recent study of Ferrari et al. (2016) finds that the launch of the PSPP in January 2015 had a meaningful impact on scarcity collateral premia in sovereign bond markets. Hence, it would be interesting to see whether similar results can be obtained for eligible bonds included in the CSPP, and more importantly how the securities lending program of the ECB can counteract. An equivalent study would require data on special collateral repo rates of bonds as well as on individual transactions in the repo market. Apart from that, the question whether corporate bond issuance has picked up could be analyzed since this study has shown that investment-grade corporate bonds became
29
more liquid. At last, an analysis and comparison of earlier terminated asset purchase programs is needed. Quantitative easing in the US, for instance, comprised a meaningful intervention in the agency and mortgage-backed securities market, which exhibited turmoil during the financial crisis. In addition, I expect a significant impact from large-scale UK gilt purchases by the Bank of England in 2009. The reason for this is that the BOE purchased in a short period, from March 2009 to January 2010 a large amount of nearly £200 bn. of outstanding assets, consisting mainly of medium- and long term gilts. According to Joyce et al. (2011a), these purchases made up to 30% of the outstanding amount of gilts. Compared to the Eurozone, in which the outstanding amount of government debt securities is larger, a more significant effect seems reasonable. Unfortunately, I was not able to access data on securities dating back to 2009.
Hence, there are still promising avenues for further research, especially since asset purchases in the Eurozone are still ongoing.
