The spillovers of ECB unconventional measures : an empirical analysis for the UK

(1)

THE SPILLOVERS OF ECB

UNCONVENTIONAL MEASURES:

An empirical analysis for the UK

Master Thesis

Beatriz Macedo Soares Ferreira

Universiteit van Amsterdam Faculty of Economics and Business

Supervised by: Alex Clymo

Master in Economics 2015/2016

Monetary policy, Banking and Regulation Track Word count: 16531

(2)

Statement of Originality:

I hereby declare that this thesis is totally original and by no means was copied or inspired in other thesis or published or unpublished work.

Abstract:

This thesis aims to study the spillover effects of the ECB’s unconventional monetary policy announcements on UK financial assets, from January 2007 to March 2016. The impact of these announcements is analysed through an event-study regression model with daily data. When taking the ECB policies in an aggregate manner, no significant results were found, suggesting a lack of spillovers from EU policy to the UK. However, splitting this variable into several different variables according to the type of policy announced reveals that some policies do have spillovers, while others do not. The CSPP announcement seems to be the most significant in affecting the exchange rate and the sovereign yields . No significant effects were found for the stock market index and the three-month interbank rate. The results confirm that the effects of the ECB unconventional monetary policy announcements differ across the different measures.

Key words: Spillovers, UK, ECB, unconventional monetary policy, event-study.

Acknowledgements:

The author would like to thank her Supervisor, Dr. Alex Clymo, her family and friends.

(3)

List of Tables

TABLE 1: COEFFICIENTS FROM THE VARIABLES UMPTECB , UMPTBOE, UMPTFED AND

TAPTFED REPRESENTED ... 25

TABLE 2: COEFFICIENTS FROM THE SYSTEMIZED VARIABLE UMPTECB REPRESENTED... 28

TABLE 3: SUMMARY STATISTICS FOR THE DEPENDENT VARIABLES... 37

TABLE 4: SUMMARY STATISTICS FOR THE ANNOUNCEMENT VARIABLES AND OTHER CONTROLS .... 38

TABLE 5: ECB UNCONVENTIONAL MONETARY POLICY ANNOUNCEMENTS ... 38

TABLE 6: ECB POLICY RATE AND POLICY RATE CHANGES - DATE OF THE ANNOUNCEMENT FROM THE GOVERNING COUNCIL... 42

TABLE 7: ECB POLICY RATE AND POLICY RATE CHANGES – DATE WHERE THE CHANGE OCCURRED .. 43

TABLE 8: BOE UNCONVENTIONAL MONETARY POLICY ANNOUNCEMENTS ... 43

TABLE 9: BOE’S POLICY RATE AND POLICY RATE VARIATION ... 44

TABLE 10: FED’S UNCONVENTIONAL MONETARY POLICY ANNOUNCEMENTS ... 45

TABLE 11: MACROECONOMIC VARIABLES INCLUDED IN THE EVENT STUDY ANALYSIS ... 45

TABLE 12: COEFFICIENTS FROM THE SYSTEMIZED VARIABLE UMPTECB REPRESENTED (2-DAY WINDOW) ... 47

TABLE 13: COEFFICIENTS FROM THE SYSTEMIZED VARIABLE UMPTECB REPRESENTED (3-DAY WINDOW) ... 48

(6)

1. Introduction

With the financial turmoil arising from the 2008 global financial crisis, the economic and financial collapse in the Eurozone escalated into the sovereign debt crisis in 2010. Markets were under stress, which lead to a reassessment of the solvency of countries with large debts (to GDP ratios). Shortly after the collapse of Lehman Brothers, the spread between the three-month Euribor and the overnight interest rate EONIA rose to unprecedented levels. Liquidity dried up and there was a sudden loss of confidence combined with an increase in uncertainty in the financial markets. Interbank markets were extremely stressed and investors stopped investing. Systemic risk strengthened which led to colossal consequences for the real sector in the economy.

In response to this framework, the European Central Bank (ECB), like other central banks of major economies1_{, aimed at restoring a proper liquidity allocation, by reducing its key}

interest rate in a significant manner. With the short-term nominal interest rates approaching the zero lower bound, as well as a decline in real output and inflation below the desired level, the ECB, and other central banks, implemented a set of different non-standard monetary policy measures, to stimulate economic growth and to contain the European Sovereign debt crisis.2_{However, “while the programs of the ECB and the BoJ reflected the bank-centric}

structure of their financial systems, responding with loans to the banking system, the Fed and the BOE expanded their respective monetary bases by purchasing bonds”. (Fawley and Neely, 2013).

Nevertheless, the spillover effects of these measures are still under study. Most of the literature focuses on the effects of these policies on the domestic economy and financial assets. Another strand of literature approaches the measures undertaken by the US Federal Reserve on other advanced economies and emerging markets. There is growing literature on the effect of the ECB’s policies in other advanced economies, however it gives a lower focus to the effects on the United Kingdom (UK). It is of great interest to study these effects on the UK since the Eurozone is its biggest trading partner and consequently there are close economic ties. In addition, with the recent events in Europe, it is important to know if indeed

1_{The United States Federal Reserve System (Fed), the Bank of Japan (BoJ) and the Bank of England (BOE).} 2_{The main argument to use these policies was that central banks could influence prices and output by increasing} liquidity in the market, for instance by purchasing long-term assets.

(7)

announcements from the ECB can be transmitted to the UK and to shed some light on the magnitude of these effects.

Therefore, my objective is to study the effect that the ECB’s unconventional measures had on financial assets from the UK economy, more specifically, on asset prices, exchange rates, interbank rates and on stock markets. The period of study ranges from 2007 until 2016. Since it is expected that the biggest impact occurs when expectations are formed, I consider the announcements rather than their actual implementation.

My research question therefore is the following:

“What are the Spillovers of ECB’s unconventional policy announcements on UK financial assets?”.

I will study the impact of non-standard monetary policy announcements on five and ten-year CDS spreads, the exchange rate (pound into euro), the five and ten-year government bond yields, the 3-month interbank rate and the stock market index. To answer the research question, an event study methodology will be used, with impulse-dummies, controlling for, among other things, announcements from other major central banks, changes in policy rates and macroeconomic surprises arising from the release of macroeconomic indicators.

When looking at the ECB policy announcements as a whole, there is no significant impact on UK variables. Observing the individual types of announcements, it is possible to conclude that their impact differed across different measures , probably due to their different objectives. The Corporate Sector Purchase Programme (CSPP) announcement seems to be the most significant in affecting the exchange rate and the sovereign yields, while the remaining policies had less significant effects. No significant effects were found on the stock market index and the three-month interbank rate. These results are robust to most of the specifications.

The remaining of this paper is organized as follows: In Section 2 I review the historical background relevant to this topic and in Section 3 I explain the transmission channels that can lead to an impact of these policies on the UK economy. Section 4 provides an overview of all the relevant literature to this subject. Section 5 describes the methodology and data used. Section 6 and 7 reports the results and the robustness checks, while in Section 8 a discussion of some aspects this paper is presented, with future research suggestions. And finally, in Section 9, my main conclusions are provided.

(8)

2. Historical Background

The next section provides an analysis of the different unconventional measures undertaken by the ECB until the end of March 2016. I discuss the different policy announcements, their objectives and their theoretical foundations. Initially, these policies had the objective to alleviate financial market distress but soon they were addressing inflation concerns, stimulating economic activity and containing the European sovereign debt crisis (Fawley and Neely, 2013).

In the recent years the ECB has announced and implemented a set of non-standard monetary policy measures, in response to the global financial crisis. The objective was to address a range of risks, arising from disturbances to the liquidity level of certain asset markets in the euro zone and to improve the transmission mechanism of monetary policy. Accordingly, these unconventional measures have been a topic of focus in the last years.

In the first phase of the crisis, the adopted non-standard measures intended to focus on stopping the deterioration of the interbank market, arising from illiquidity and major funding problems of banks. Following of these events and in order to support the financial markets in the Euro area, the ECB conducted a set of non-standard measures. The fixed rate tender with full allotment programme (FRTFA) was the first measure implemented by the ECB, where it would satisfy all the liquidity demanded by banks at a fixed-rate tender, as long as the banks had collateral, while simultaneously extending the list of eligible collateral. This was different from the ECB’s conventional policies as they provided a fixed allotment of funds at rates determined by the bidding process. The ECB also extended the list of eligible collateral assets for refinancing operations, known as the broader collateral framework (COLL), introduced long-term refinancing operations (LTRO), where the maturity of the long-term refinancing operations was extended and provided liquidity provision in foreign currencies through swap lines with other central banks to enhance banks’ foreign currency funding (FOR) (Fawley and Neely, 2013; Rogers et al., 2014; Falagiarda et al., 2015).

On the 7th_{of May 2009 the Governing Council decided to proceed with the Enhanced}

Credit Support (ECS), to mitigate the impact of collapsing wholesale and interbank markets. This programme focused primarily on banks as they are the main source of credit in the euro area economy. The ECS included the four liquidity providing measures previously mentioned and introduced the first Covered Bond Purchase Program (CBPP1), a program of outright

(9)

purchases of covered bonds, to fight the scarcity of liquidity.

In May 2010, the ECB introduced the Securities Market Programme (SMP), a sterilised purchase of public and private debt securities in the Euro area.3_{Initially purchases were}

limited to Greek, Portuguese and Irish government bonds but on August 2011, the program was extended to Spanish and Italian government bonds. Under the SMP, the ECB intervened by purchasing bonds on a daily basis without any predetermined target in terms of price or quantity and would simply decide based on the observed market conditions (Fratzscher et al. 2014). The objective of this policy was to restore the monetary policy transmission mechanism and to provide liquidity to the markets, thus ensuring price stability in the medium term. However, the ECB did not reveal the total amount or the time span of the purchases, which according to Micossi (2015) reduced their market impact. In October 2011 due to the increased worries about some sovereign markets (Spain and Italy), the ECB announced their second Covered Bond Purchase Program (CBPP2) in which they would buy additional €40 billions of covered bonds.

Following Draghi’s London speech, “whatever it takes”, on the 26th_{July 2012, the}

Outright Monetary Transactions (OMT) was announced to substitute the SMP program, with the objective of repairing the monetary transmission mechanism. The OMT consists of unlimited purchases of government bonds issued by countries under a European Stability Mechanism (ESM) macroeconomic adjustment program. This program, on the contrary of the remaining, has never been implemented, although, just the credible threat of such purchases has been enough to calm the markets (Fratzscher et al., 2014; Rogers et al., 2014).4

Following its meeting on the 4th_{of July 2013 the Governing Council of the ECB}

communicated that it “expects the key ECB interest rates to remain at present or lower levels for an extended period of time”. This policy, known as Forward Guidance entails that by influencing expectations of future short-term interest rates, the central bank can ensure that its policy stance is transmitted to the economy (ECB,2014). In June 2014 the Credit Easing package was announced and its objective was to encourage lending to the real economy and therefore improve the functioning of the monetary transmission mechanism. Specifically, the Governing Council decided to conduct targeted longer-term refinancing operations (TLTROs),

3_{The impact of these interventions is sterilized through specific opera tions to re-absorb the liquidity injected} and thereby ensure that the monetary policy stance is not affected.

(10)

in which the ECB provides financing to credit institutions up to four years. The objective was to improve bank lending in the Eurozone. In October 2014, the asset-backed securities purchase program (ABSPP) was initiated alongside with a new covered bond purchase program (CPBPP3). By securitizing loans, the ABSPP could provide banks with the necessary lending to promote the diversification of funding sources in the banking systems, which eases funding and credit conditions, helping in the monetary policy transmission.

On the 22nd_{of January 2015, the Governing Council decided to launch the expanded}

asset purchase program (APP), which adds a purchase programme for public sector securities (PSPP) to the existing private sector asset purchase programmes (CBPP3 and ABSPP) to address the risks of low inflation. The combined monthly private and government asset purchases amount to €60 billion per month until a sustainable path towards the objective of a rate of inflation which is close but below 2 percent (Falagiarda et al., 2015; ECB, 2015).

Finally, on the 10th_{of March 2016, the Governing Council of the ECB announced that}

they would add a Corporate Sector Purchase Programme (CSPP) to the asset purchase programme and would modify the later one by increasing the monthly purchases from €60 billion to €80 billion, starting in April 2016. Under the new CSPP, euro-denominated bonds issued by non-bank corporations in the Euro area will be included in the list of eligible assets to be purchased monthly. The main aim of the CSPP is to reinforce the transmission from the asset purchases to the financing conditions in the real economy (ECB, 2016).

3. The International transmission of Unconventional Monetary Policy

In this section, I explain the channels through which the transmission of monetary policy from the announcements to the studied variables occurs. I only focus on channels that lead to an international transmission as it is the scope of this thesis. With these channels in mind, it is crucial to distinguish how monetary policy spillovers may vary depending on the specific design of the monetary policy instrument announced by the ECB. These channels of transmission can work in parallel and as a consequence, they can be difficult to identify.

In most of the academic papers that approach this topic, the Portfolio Rebalance

Channel has been considered one of the most important channels in the US (Gagnon et al.

2011) and in the UK (Joyce et al. 2012), being widely referred in the literature. The Portfolio Rebalance channel rests on the assumption that different financial assets are not perfect

(11)

substitutes in investors’ portfolios. Therefore, changes in the net supply of an asset changes its yield incentivizing agents to rebalance their portfolios and to hold other assets with similar characteristics, consequently affecting the yield of those assets. The existence and magnitude of this effect depends on the degree of substitutability between bonds issued by euro area countries and UK sovereign bonds.

Unconventional monetary policy announcements may also spread through the expectations of future economic conditions and policy actions, from economic agents. This channel is often mentioned as the Signalling/Expectations Channel and operates by the fact that with the unconventional monetary policy announcement agents create expectations of a lower path of future short-term policy rates, which by the expectation hypothesis of the term structure results in a reduction of the long-term interest rate.5_{Consequently, the}

interest rate differential leads to spillovers to international economies. According to Bauer and Neely (2013), signalling effects are expected to be stronger between countries where interest rates are responsive to monetary policy surprises.

With the policy announcements the ECB can lead to a change in general confidence from agents which affects uncertainty in the Eurozone and internationally. According to Falagiarda et al. (2015), this can influence financial assets both in positive and negative ways as it impacts investors’ risk appetite and their portfolio decisions. This channel is known as the Confidence Channel.

Fratzscher, Lo Duca and Straub (2014) test for some of these channels and find that the portfolio rebalancing channel was not an important channel of transmission for ECB policies, which stands in contrast to the portfolio rebalancing observed in response to the Fed’s policies. However, the authors stress that these discrepancies might arise from differences in the instruments and the size of operations. In addition, they find evidence of a confidence channel in the Eurozone, characterized by a decrease in risk aversion succeeding the announcements.

5 _{Neely (2010) explains this channel referring to the following equation: y𝑡,𝑡 +𝑛}_{= 𝑦̅𝑡 ,𝑡+𝑛}_{+ 𝑇𝑃}_{𝑡 ,𝑛 where} 𝑦𝑡 ,𝑡 +𝑛 is the yield at time t on an n-year bond, 𝑦̅𝑡 ,𝑡+𝑛 is the average expected overnight rate over n years at time t and 𝑇𝑃𝑡 ,𝑛 is the term premium on an n-year bond at time t. If forward guidance or asset purchase announcements reduce expectations of future short-term interest rates than the average expected overnight rate (𝑦̅𝑡 ,𝑡+𝑛) is reduced which causes a reduction on the long-term interest rates (𝑦𝑡 ,𝑡+𝑛).

(12)

4. Literature Review

Unconventional monetary policy has been a discussed topic in the last few years since the onset of the crisis. Most of the literature starts by studying the impact of these policies on their own economy and domestic assets (Joyce et al., 2011; Gagnon et al., 2011) or even within the Eurozone (Altavilla, Giannone and Lenza, 2014; Falagiarda and Reitz, 2015). Later on, the focus of this topic shifted towards its spillovers to other economies.

4.1. Spillovers from the Fed

This thesis relates to a strand of empirical studies analysing the spillovers of Central Bank policies to international financial assets. Most of the literature that approaches this topic covers policies announced by the Fed.

An example of this is Neely (2010), who studies the impact of the Fed’s long-term asset purchases (LSAPs) and forward guidance announcements, between 2008 and 2009, on the exchange rate and international yields of advanced economies6_{. According to his results, the}

announcements led to a decrease in the value of these variables, which is consistent with a portfolio balance effect away from the US. The author stresses that an event study is the most suitable methodology to deal with variables that react very fast to news, as long as all the changes in expectations are caught within the event windows.

Fratzscher, Lo Duca and Straub (2013) perform a detailed analysis on the effects of US unconventional monetary policy announcements and portfolio flows. During the first phase of QE (QE1) in 2008-2009, the policies led mainly to a rebalancing of portfolios shifting from emerging market economies (EMEs) and advanced economies (AE) into US equity and bond funds, thus decreasing sovereign yields, raising equity prices in the US, therefore appreciating the dollar. The reason for this appreciation is that US unconventional policies seem to have led investors to hold US assets during liquidity shortages. The effect of Fed’s measures since 2010 (QE2) on yields was negligible although it increased equity prices worldwide. Specifically, the purchases led to a large portfolio reallocation across assets (from bonds into equities) and across countries, from global markets mainly into EME markets.

(13)

The conclusions regarding QE1 differ slightly from previous literature, as for instance, Neely (2010) finds that the LSAPs depreciated the US dollar compared with other advanced economies in the sample. However, Neely does not take into account the evolution of portfolio flows in his analysis. There might be other factors, such as higher confidence or lower risk, that may lead investors to prefer to hold US assets in these situations, as referred by Fratzscher et al. (2013).

Interestingly, the authors try to determine why the international spillovers from Fed’s activities are so heterogeneous between countries, that is, why EMEs are much more affected than AEs. Firstly, they attribute this discrepancy in part to the policy response that countries have when trying to protect from the spillovers (e.g. increasing foreign exchange reserve holdings or using monetary policy). According to results, a greater use of monetary policy and a stronger presence of a central bank helps countries insulate themselves from the shocks. Secondly, a higher quality of institutions and domestic policies contributes to a better shielding and less exposure.

It should be noticed that endogeneity is likely to be a problem since Fed’s interventions can be endogenous to market conditions, meaning that the interventions can be a measure to improve the economic framework. To overcome this issue authors usually use lags or a two stage approach. Using daily data also decreases the chance for simultaneity to persist.

Rogers et al. (2014) examines the impact of unconventional measures from the Fed, BoE, ECB and BoJ on bond yields, stock market prices and exchange rates. They find that both conventional and unconventional monetary policy affect the term premium but in different ways: the later has larger effects on the long-term interest rates whereas conventional monetary policy impacts more on short-term rates.7_{Cross-country spillovers from}

non-standard measures were found to be asymmetric among the economies. Specifically, monetary policy shocks by the Fed greatly affected European and British yields, whereas the opposite did not happen. As a result, monetary policy easing by the Fed usually eases economic conditions abroad; currencies generally appreciate vis -à-vis the dollar and the long-term interest rates are reduced. Their methodology makes use of an event study regression to measure the immediate effect of the announcements on asset prices. In addition, instead

(14)

of detecting the monetary policy announcements and measuring their impact, the authors use the intraday change in government bond yields around the announcements and regress this on other asset returns. This methodology only measures the pass -through from sovereign yields to other assets and not the impact of the announcements on bond yields itself. The innovation relies on the type of data used - intraday data. High frequency data ensures that, within the event window, the number of other factors influencing the change in yields is minimized. However, it is argued that international spillovers have longer transmission lags which implies that the effect is not totally captured in the chosen event-window. Rogers et al. (2014) do not comment on the differences between using daily and intraday data.

In conclusion, results seem to change in a significant manner when the data differ: the use of fund level data appears to provide further insights on the direction of the international portfolio rebalancing from investors (Fratzscher et al. 2013), showing that under similar programs, the economic consequences might be completely different. However, data used does not account for the presence of micro and macroprudential measures, which probably played a role in the transmission of Fed’s policies (Fratzscher et al. 2013). Although event studies are the most common, authors follow different approaches. Fratszcher et al. (2013) uses lagged variables reflecting market conditions, although they only affect the magnitude of the estimates. The remaining do not include controls to account for economic conditions and the framework in the countries analysed.

4.2. Spillovers from the ECB

The present thesis relates mainly with empirical studies analysing the international spillovers from the ECB’s non-standard policy announcements on financial assets.

Fratzscher, Lo Duca and Straub (2014) perform a study on the impact of ECB policies on equity, exchange rate returns, yields, risk measures and capital flows in the Eurozone and abroad using daily data. Thirty-eight advanced economies and other emerging economies are considered.8_{In the baseline specification, the controls are constituted by surprises from the}

release of macroeconomic indicators in the Euro area and the US, country fixed-effects monetary policy decisions, unconventional policy announcements in the US, among others.

8_{From the advanced economies outside the Eurozone, the ones considered in the analysis were Australia,} Canada, Denmark, Japan, New Zealand, Norway, Sweden, Switzerland, UK and US.

(15)

Spillovers to global economies outside the Eurozone are found to have a positive impact on global equity markets and a negative one on risk aversion and credit risk leading to an increase in the general confidence. The individual effects, however, differed in terms of the type of measure used and its objective. This paper relates with the present thesis because it studies, among other things, the international spillovers from the ECB policy to AEs, which include the UK. There are two main differences. Firstly, they also consider the implementation of the policy. Secondly, authors only study the effect of OMT and SMP announcements.

Fratzscher et al. (2014) find that the OMT announcements lead to an increase in sovereign yields in AEs and to an appreciation of the euro; the SMP announcements did not affect the sovereign yields but caused a depreciation of the euro vis -à-vis AEs. In comparison, this thesis does not find evidence of the effect of OMT announcements while the SMP announcements affected the short-term sovereign yields. The divergence in results could arise from the fact that in this thesis I consider the effect on the UK alone while in Fratzscher et al. (2014) the AEs are considered as a whole. 9

Their findings also suggest that the international portfolio rebalancing channel was not an important channel of transmission, opposing to the Fed’s case (Fratzscher et al. 2013). This may reflect the use of distinct instruments and operations or even the structure of the financial market in Europe. The study shows that different measures, although similar in some sense (e.g. asset purchases), can lead to very different effects in the same group of countries. Nevertheless, the authors do not provide any reason for this discrepancy. Authors believe simultaneity is unlikely to raise concerns due to the high frequency data. The idea is that changes in market conditions, such as government bonds or equity, hardly affect policies on a daily basis, as policy-makers cannot react that fast.

Finally, Falagiarda, McQuade and Tirpák (2015), find evidence of strong spillovers to four Central and Eastern European countries (CEE), in particular on sovereign yields and exchange rates, recurring to an event study analysis.10_{Several controls are present such as}

the unconventional policy announcements from the Fed and the domestic central bank, both the policy rate of the ECB and the domestic central bank, the surprise arising from the release of macroeconomic indicators in the country and volatility in the Euro area. In particular, the announcements were associated with a decline in sovereign bond yields in Czech Republic,

9_{A discussion of these results comparing to the results from this thesis is included in the results section.} 10_{These are Hungary, Czech Republic, Poland and Romania .}

(16)

Poland and Romania. The same did not happen with Hungary, which might be related with the higher perceived riskiness of the country. No spillovers from the announcements to C DS spreads or stock market indexes were found. However, according to their analysis, the spillovers varied significantly between the different measures implemented by the ECB: the SMP announcements seem to have impacted substantially most of the financial variables (specifically bond yields and CDS spreads) while the OMT announcements had a reduced impact and no statistically significant coefficient was found for the PSPP announcement.

Following Falagiarda et al. (2015), a country’s perceived riskiness seems to be an important factor in determining the spillovers, as it can be a determinant of asset substitution. Also, the proximity to Europe in terms of financial integration is another important caveat as it facilitates the transmission of the policies. For instance, if the presence of banking groups belonging to the Eurozone in these countries is substantial, changes in the Euro area framework can lead to changes in the demand for funds in these banks, causing the interbank rates to differ. The results from this paper coincide with the results from Fratzscher et al. (2014) for the emerging EU countries included in their sample11_{. This thesis follows the same}

methodology as this paper but applied to the UK.

In conclusion, spillovers from ECB policies differ according to the country analysed due to several factors, as for example, the country’s rating or the level of financial market integration. Since related literature is rather new, there are not yet enough different methodologies to infer on how could these affect the results. However, notice that Fratzscher et al. (2014) use several controls in their analysis, including some used by Falagiarda et al. (2015) and their estimates remain significant. The problem related with endogeneity of the estimates seem to be accounted for, by using daily data, since the decision to engage in the policies cannot depend on daily changes in financial conditions.

5. Data and Empirical approach

5.1. Empirical approach

In order to study the spillover effects of unconventional monetary policy announcements from the ECB on UK financial assets I follow the methodology described in Falagiarda,

(17)

McQuade and Tirpák (2015). The event study regression has been widely used in this type of study12_{and Falagiarda, McQuade and Tirpák (2015) used this methodology to study the}

spillover effects of ECB policies on four Central and Eastern European Countries (CEE)13_{, that}

are closely integrated with the Eurozone, in an economic and financial manner. The United Kingdom is affected by Eurozone events especially through the trade channel, as the Eurozone is its main trading partner. In addition, this methodology relies on publicly available data. Only announcements are to be considered, and not the implementation itself since usually expectations are already formed at the time of the announcement.14

To analyse the spillover effects from the ECB’s announcements on the UK economy I proceed with an event study approach using an ordinary least squares (OLS) estimation. The regression is estimated using daily data from 01/01/2007 until 31/03/2016. The use of daily data in contrast with intra-daily data is justified by the fact that international spillovers are likely to have longer transmission trails compared to domestic policies (Falagiarda, McQuade and Tirpák, 2015). Therefore, the baseline scenario is the following:

∆𝑌_𝑡 = 𝛼 + 𝛽₁∆𝑉𝐼𝑋_𝑡+ 𝛽₂∆𝑃𝑅_𝑡𝐵𝑂𝐸+ 𝛽₃∆𝑃𝑅_𝑡𝐸𝐶𝐵+ 𝛽₄𝑈𝑀𝑃_𝑡𝐵𝑂𝐸+ 𝛾₁𝑈𝑀𝑃_𝑡𝐹𝐸𝐷+ 𝛾₂𝑇𝐴𝑃_𝑡𝐹𝐸𝐷+ 𝛿𝑈𝑀𝑃_𝑡𝐸𝐶𝐵_{+ 𝜆𝑁𝑒𝑤𝑠}

𝑡 + 𝛽5∆𝑃𝑅𝑡𝐵𝑂𝐸 𝑥 ∆𝑃𝑅𝑡𝐸𝐶𝐵+ 𝛽6𝑈𝑀𝑃𝑡𝐵𝑂𝐸 𝑥 𝑈𝑀𝑃𝑡𝐸𝐶𝐵+ 𝑢𝑡 (1)

where the dependent variable 𝑌_𝑡, represents different variables from the UK. Therefore, the dependent variable is alternatively the exchange rate (Sterling into Euro), the stock market index, the 3-month interbank rate, five and ten-year CDS spreads and the five and ten-year sovereign bond yields both denominated in domestic currency. To control for periods of elevated volatility in the Euro area financial markets, which may affect financial markets in the UK, a volatility index for the Eurozone is included in the regression, 𝑉𝐼𝑋𝑡.

∆𝑃𝑅_𝑡𝐷𝑂𝑀_{𝑎𝑛𝑑 ∆𝑃𝑅}

𝑡𝐸𝐶𝐵 stand for the BOE’s policy rate variation and the ECB’s policy rate

variation respectively, and the objective is to control for the variation in these rates, as they can affect the dependent variables. These changes can be consulted in Tables 9 and 6, respectively. 𝑈𝑀𝑃_𝑡𝐷𝑂𝑀 _{is an event-dummy which tracks the announcements of}

unconventional monetary policy performed by the BOE. Similarly, the variable 𝑈𝑀𝑃_𝑡𝐹𝐸𝐷 is an

12_{See Section 4 for a discussion.}

13_{The study is performed for Hungary, Romania, Czech Republic and Poland.}

(18)

event-dummy that tracks the expansionary unconventional monetary policy measures announced in the US by the Fed and their purpose is to control for spillovers from these announcements in the UK. Since expansionary announcements have to be coded differently than contractionary announcements, the variable 𝑇𝐴𝑃𝑡𝐹𝐸𝐷 measures the tapering

announcements from the Fed.15_𝑈𝑀𝑃

𝑡𝐸𝐶𝐵 contains the event-dummies related to ECB

announcements of non-standard policy measures. I perform my analysis with two different specifications of the latter variable and therefore running two different regressions: firstly, I aggregate all ECB announcements regarding various nonstandard measures, meaning that it takes the value one in the days where the announcement took place and zero otherwise. Secondly, I build several different dummy variables to separate each type of announcement, according with the different ECB measures in line with the classification in Table 5 in the appendix. This is performed in order to account for the fact that the announcements of the programs had different objectives which can imply different international spillovers.

In addition, another control is introduced, to control for the surprise element in the release of macroeconomic variables, which can easily affect the variables of interest in this analysis. The objective is to account for possible movements in the dependent variable due to unexpected changes in macroeconomic variables. The data used to compute this variable is collected via Bloomberg and consists of expectations of market participants about all available macroeconomic variables released in the UK. The surprise element of these variables can be calculated by computing the difference between the actual value on the day of the release and its expected value. The expected values are median (consensus) forecasts collected up to one day before the data release (Altavilla and Gianonne, 2014; Falagiarda, McQuade and Tirpák, 2015).

Finally, the previous regression includes two interaction terms, one for BOE and ECB key policy rate changes and another one for BOE and ECB non-standard monetary policy measures. The objective is to account for the fact that in days that there are announcements from both the BOE and the ECB or both the policy rate changes, the dependent variables are affected differently.

One question that can be raised regarding the use of this methodology is the issue that endogeneity may be present. In Falagiarda, McQuade and Tirpák (2015) the ECB’s

(19)

monetary policy is not influenced by the monetary policy used in the countries where the analysis is being performed. However, in this case the country where the spillovers are taking place sets its monetary policy having in account the ECB’s policy and the ECB sets its own policy also considering the monetary policy decisions in the UK. Therefore, endogeneity may be present as the dependent variable is affected by the ECB announcements and simultaneously, these announcements can be affected by the dependent variable. However, Fratzscher et al. (2014) argues against this fact asserting that monetary policy decisions cannot depend on daily changes in market conditions, depending solely on the “big picture”, which in turn has relevant implications on a daily basis. Therefore, the use of daily data should reduce the risk of this issue arising. Another way of trying to account for this risk is by lagging the dependent variable. This hypothesis is assessed in the robustness section.

5.2. Data

In this analysis I use data in a daily basis, from the 1st_{of January 2007 until the 31}st_{of March}

2016. Most of the data is taken from Datastream.16_{Due to data availability problems the CDS}

spreads are only available for a smaller sample which means that their analysis refers to a smaller time span.17_{The descriptive statistics of the dependent variables and some of the}

independent variables used are summarized in the Table 3 and 4, respectively.

Data related to the UK like the exchange rate vis-à-vis the Euro and the five and ten-year gilts are downloaded from the website of the Bank of England. To account for the differences in the days that the variables are published (i.e. variables are not published on a public holiday in the UK) I drop these observations, since it is not possible to obtain their value on this day.

In addition, information about the policy rate of the Bank of England and the main refinancing rate of the ECB are taken from their respective websites. Regarding the policy rate of the ECB, the information on the main refinancing operations rate is taken from the minutes of the Governing Council that are posted every month on the website. This is an important detail because this date is different than the one stated on the website: In the ECB website

16_{Corresponding codes available in appendix A.}

17_{Specifically, the five-year CDS spreads will be analysed from 07/11/2008 onwards and the ten -year CDS} spreads from 12/11/2008 until the end of the sample.

(20)

the days that is dated a change in the policy rate correspond to the days that the change actually occurred. Since the expectations of the agents are very relevant in this study, I use the minutes from the governing council between 2007 and 2016 to obtain this information. However, in the robustness section a test using the date where the change in the policy rate actually occurred is performed.

Regarding the announcements of the ECB’s, the BoE’s and the Fed’s unconventional policies, I take them from their respective websites by reading press releases, speeches and minutes done by the Governing Council, the Monetary Policy Committee and the Federal Open Market Operations Committee, respectively. All the announcements are registered respectively in Tables 5, 8 and 10 in the appendix. For the ECB and the Fed, several announcements are taken from Falagiarda et al. (2015), while the remaining are constructed by me.18_{Moreover, all the announcements from the UK are compiled by me.}19_{All the}

announcements are confirmed several times in order to make sure that they are correctly noted.20

Finally, I also use the surprise stemming from macroeconomic releases in the UK as an extra control. To construct this variable, it is necessary to obtain the forecasted value for a determined macroeconomic index and the actual released value, at the date of the release. All the indexes that were published between 01/01/2007 until 31/03/2016 are considered. Table 11 reports all the indexes used in this analysis. To construct this variable, two different values from these indexes in the day of their release are required: the expected value and the actual value. Following Altavilla, Giannone and Lenza (2014), the expected values are median consensus forecasts collected until one day before the official date release and the actual values are the values of the variable on the day of its release. To construct the variable that measures the surprise arising from this release, I compute the difference between the actual value and the expected value. This variable measures the surprise content of each relevant release of economic data. Therefore, if the forecast is accurate and there is no surprise stemming from the release, then the variable will take the value zero meaning that the release

18_{Most of the announcements from the Fed and the ECB from 2007 until 22 -02-2015 were taken from Falagiarda} et al. (2015).

19_{To construct the UK announcements, I als o consulted related literature as Hosono and Isobe, (2014) and Joyce} et al. (2012).

20_{As different authors may use different coding to account for the announcement days, I attempted to follow} the same coding strategy during the announcements collection.

(21)

will not have an effect on other assets. If the variable takes a different value, posi tive or negative, then there is a surprise associated to this release and it may have an impact in other asset prices.

However, these variables have some observations missing. Firstly, I complete the missing values by searching in other sources.21_{Overall, out of 42 indexes, 3 of them were not used in}

the analysis since their consensus value was not released22_{(signalled in the appendix). In}

some other indexes some consensus values are missing in a few releases which allows me to continue my analysis, however with some assumptions:

 Assumption 1: When the survey value is missing I assume rational expectations: the survey value of today is equal to the actual value of yesterday. This is an a dequate assumption as yesterday’s value should be a good guide of people’s expectations.  Assumption 2: In the case where there is no actual value preceding the missing survey

value, in the baseline specification, I assume that there is not any surprise steaming from the release of these variables.

 Assumption 3: If there are not more releases from a variable after a certain date, I assume that the variable does not carry any more surprises with its release and therefore it takes the value zero in the rest of the analysis.

Moreover, some surprises are omitted from the baseline specification as they occur in a day that does not enter in the sample (i.e. a public holiday). Therefore, as robustness test, I place the surprise in the first day considered in the analysis after the release, since it is the day that the new values are released and therefore the changes should be detected in those values. The results are robust to all these different specifications.

6. Results

6.1. Aggregate measure of ECB announcements

This section reports the results from the regression in equation (1) using the independent variable of interest in an aggregate manner, with just one dummy variable measuring all the unconventional monetary policy announcements by the ECB.23_{The results obtained for the}

21_{The main sources were http://uk.investing.com/economic-calendar/ and www.fxstreet.com.} 22 _{The three indexes that will not be used are signalled in the appendix in Table 11 with *.} 23_{Stationarity was taken into account to perform this analysis, using the dickey-fuller test.}

(22)

full sample are reported in Table 1. The standard errors presented in the table are HAC standard errors.24_{The first column shows the coefficient of interest, which is the coefficient}

of the variable 𝑈𝑀𝑃𝑡𝐸𝐶𝐵. The interpretation of this coefficient is that, when an unconventional

monetary policy announcement takes place, the dependent variable changes by δ percentage points, holding everything else constant.

As is observable, no significant results were found on the baseline specification for any of the financial assets analysed. There are several reasons that could lead to a negligible effect of the unconventional monetary policy announcements on asset prices. For instance, since sovereign yields were not affected by the announcements, the effect on other assets might be small since this effect arises in part from the pass through from sovereign yields to other assets (Rogers et al., 2014). Moreover, the fact that the UK may be well shielded from European imbalances may explain these results, as it allows them to better insulate from shocks in the Eurozone. As referred in the literature previously, the presence of a strong central bank, higher quality of domestic policies and institutions helps countries to reduce their exposure. Lastly, the international channels of transmission for unconventional monetary policy might be negligible between the UK and the Eurozone. Fratzscher et al. (2014) find that the international portfolio rebalancing was not an important channel of transmission of ECB policies to economies outside the Eurozone. In contrast, these policies led to a significant decrease in uncertainty and risk aversion in the Euro area. This is consistent with their statement that ECB policies had little or no impact on international government bond yields, affecting mainly global equity markets. To have further knowledge on why the effects are negligible, a test for the international transmission channels would be necessary, which could allow to understand which channels were actually active. However, this is beyond the scope of this thesis.

Comparing with literature, Rogers et al. (2014) found spillovers from ECB unconventional policies to the UK, specifically, an increase in gilt yields and a slight but significant depreciation of the pound relative to the dollar. However, their methodology is quite different from the one followed in this thesis: the authors regress the Italian-German

24_{The regression was tested for serial correlation in the error terms using the Breusch-Godfrey test, with several} lags. The null hypothesis (no serial correlation) was rejected. In addition, comparing the regression with and without HAC standard errors, it is noticed that the significance of the Fed’s coefficients increases, however this is not very relevant. The coefficient of the ECB’s policies remains insignificant when taking the aggregate measure of the ECB announcements. A

(23)

ten-year yield spread for the Eurozone on the yield changes, thus controlling for the surprise component. Therefore, the announcements are not accounted for with an impulse-dummy. The remaining columns in Table 1, display the coefficients of the variables 𝑈𝑀𝑃𝑡𝐵𝑂𝐸,

𝑈𝑀𝑃_𝑡𝐹𝐸𝐷_and_𝑇𝐴𝑃

𝑡𝐹𝐸𝐷. These coefficients represent the change in the dependent variables

when an unconventional monetary policy announcement is performed by the BOE, by the Fed or a Tapering announcement (by the Fed), respectively.

As is observable from Table 1, policy announcements from the BOE are not significant according to this specification which seems very unusual since they are expected to have a sizeable impact. In order to seek answers, I analyse other papers that study the impact of QE from the BOE on UK sovereign yields and financial assets. Joyce et al. (2011) perform their analysis for the period 2009-2010 using an event-study strategy with daily data. The authors control for expectations by using the Reuters poll of economists, in order to obtain the market participants’ expectations of QE. Then, the variable measuring the announcements’ expectations is calculated by the difference between these expectations and the QE amount released after the MPC’s decision. Using this control, the results show a decrease in sovereign yields. However, similar results are obtained without this control. In general, there is a slight depreciation of the pound and the stock market index is not affected.

Another example is Hosono and Isobe (2014) who study the effects of unconventional policy announcements on sovereign yields and other financial assets, since January 2007 until April 2013. They also use an event study regression with several controls. Firstly, they control for the bandwagon effect.25_{In addition, the mean reversion effect}26_{and the lag of the}

dependent variable are also taken into account. Although the latter is used in the robustness tests, the remaining were not used in the methodology followed in this thesis. The variable measuring the announcements is also presented with two different specifications: one of them is the normal impulse dummy while the other one uses a surprise component captured by the change in government bond futures prices around policy announcements. The main results are that these announcements lead to a decrease in the short run yields, the corporate bond yields and the interbank loan spread.

25_{The bandwagon effect is the tendency from economic agents to follow what other agents have already done} mostly due to conformism and the fact that their information is derived from these agents.

(24)

With this analysis, it is observable that the main differences between this paper and the mentioned literature are present in the controls used, the surprise effect of the announcements and the time span of the analysis. However, Hosono and Isobe (2014) state that they control for the market expectations in a different way as Joyce et al. (2011) but that the results are broadly consistent. In this thesis I use the surprise component in the macroeconomic releases to control for the surprise in the markets , however I do not use the surprise effect of the announcements, which is left for further research.

One reason that may contribute to the insignificance of this coefficient is the fact that during all the considered years, the BOE unconventional policy announcements were concentrated within a few years inside the sample. Running the baseline regression during a smaller time span where the BOE announcements are more recurrent, from 2009 until 2012, the significance of this coefficient increases substantially. Increasing continuously the window leads to a decrease in the significance of the estimate. Therefore, the fact that the analysis is performed during nine years, but the announcements only took place in two of them (approximately) may explain the unexpected low significance level.

Finally, in contrast, policies from the US seem to have been very significant, as it is observable from Table 1. Specifically, the Fed’s tapering policies had significant negative effects on the exchange rate and positive on the stock market index while the Fed’s expansionary unconventional measures had significant negative effects on the stock market index and the sovereign yields. Indeed, Rogers et al. (2014) also finds that the Fed’s LSAP’s affect gilt yields, the exchange rate and stock market prices. In addition, Rogers et al. (2014) find that relevant cross-country spillovers from unconventional monetary policy measures were found among these advanced economies. They appeared to be asymmetric, as the effects of monetary policy shocks in the US on asset prices in the other economies are greater than the opposite. Moreover, according to the table, the effects from the Fed’s policies seem to be non-linear since the contractionary announcements have a more significant coefficient than expansionary announcements (Falagiarda et al., 2015).

However, it is essential to bear in mind that in this specification several factors from the US are not being controlled for as some from the Eurozone were, like the volatility in the Eurozone and changes in policy rates from the central bank. Therefore, these results may not be accurate as their significance and sign is very likely to change once all these conditions are

(25)

controlled for. For instance, the fact that both the expansionary and contractionary policies from the Fed have a negative effect on the stock market index is a good indicator.

Table 1: Coefficients from the variables 𝑼𝑴𝑷𝒕𝑬𝑪𝑩 , 𝑼𝑴𝑷𝒕𝑩𝑶𝑬, 𝑼𝑴𝑷𝒕𝑭𝑬𝑫 and

𝑻𝑨𝑷_𝒕𝑭𝑬𝑫 represented

Note: *** p<0,01, **p<0,05, *p<0,1. Newey-west standard errors are presented in parenthesis below.

𝐔𝐌𝐏_𝐭𝐄𝐂𝐁 _𝐔𝐌𝐏 𝐭𝐁𝐎𝐄 𝐔𝐌𝐏𝐭𝐅𝐄𝐃 𝐓𝐀𝐏𝐭𝐅𝐄𝐃 EXCHANGE RATE 0.0008 (0.07) 0.0004 (0.0019) 0.0027 (0.002) -0.0052*** (0.0013) SOVEREIGN YIELD 10-YEAR 0.001 (0.0024) 0.001 (0.007) -0.019** (0.0085) 0.0157 (0.0177) SOVEREIGN YIELD 5-YEAR 0.003 (0.0045) -0.0035 (0.0101) -0.0259* (0.0156) 0.0445 (0.0326)

STOCK MARKET INDEX -0.0019 (0.0014) 0.0012 (0.0015) -0.0069** (0.0033) -0.0036*** (0.0011) 3-MONTH INTERBANK RATE 0.0034 (0.0028) 0.0044 (0.0039) 0.0025 (0.004) 0.0077 (0.0058) CDS SPREADS 5-YEAR -0.0038 (0.0078) 0.01283 (0.0119) 0.01345 (0.0101) 0.0502* (0.0289) CDS SPREADS 10-YEAR -0.0037 (0.0079) 0.01172 (0.0117) 0.0103 (0.0089) 0.0415 (0.0264)

6.2. Systematic measures of ECB announcements

In this sub-section I present a different specification of the independent variable with the aim of illustrating the different objectives and impacts that each type of policy may have. The classification of policies followed is the one presented in Table 5 which is in accordance with the description of policies in Section 2. Therefore, twelve different dummy variables were created to further distinguish between each policy. Although the impact of the announcements is not significant in the previous specification, with a closer look at each policy, the outcome might differ. The results are reported in Table 2.

Comparing these results with the case with normal standard errors, there are small differences regarding the significance of the coefficients, however not very significant.27_In

the case of the exchange rate, the FRTPFA programme is positive and significant at the five percent level with normal standard errors and not significant with HAC standard errors. This is in line with economic theory since the FRTPFA programme may increase liquidity in the Euro area which can lead to a shift from other assets to Eurozone assets. If this would be the

(26)

case, then it is possible that the demand for Euros compared to pounds would increase, causing a relative depreciation in the pound. The discussion here is whether this depreciation should be significant, however it is something that is not possible to evaluate since it would depend on the efficacy of the programme and if indeed there was an increase in the demand for euros in detriment of the pound. For the stock market index, the COLL policy has a significant and negative impact without HAC standard errors while it is not significant with HAC standard errors. Allowing to increase the list of collateral in refinancing operations may lead also to a higher level of confidence in the Eurozone, shifting portfolios to the latter, consequently depressing the stock market index. Although there is no certainty regarding the significance of these variables, I proceed with the analysis using the HAC standard errors, in an effort to account for autocorrelation in the error terms.

According to the results exhibited in Table 2, each individual policy produces different effects on the financial assets. The CSPP seems to heavily influence the exchange rate and the sovereign yields. This goes in line with economic theory. For instance, if agents expect good future economic conditions in Europe, then they will demand Eurozone assets, in detriment of UK assets, which causes an appreciation of the euro comparing with the pound (and therefore a positive effect on the exchange rate) and an increase in UK gilt yields. In addition, this might mean that the announcement of the CSPP was very important since for the first time the ECB will buy directly corporate assets, which is considered unprecedented.

The ten-year yield was mostly influenced by asset purchase programmes like the CBPP and the CSPP. One explanation for this positive effect might be through the increased liquidity in the Eurozone market, leading to an increase in the demand inside the Eurozone, which lowered the pressure on the UK gilts consequently increasing the yields, as explained previously.

The stock market index was affected by the announcements mainly in a negative manner, although the estimates are not very significant. The estimates for the 3-month interbank rate are almost negligible.28_{Since the banking system in the UK is constituted by mainly British}

banks, there should not be an obvious impact in the Libor rate as the banking system in the UK would not be directly affected. However, there could be other ways for the Libor to be

28_{Actually, the 3-month interbank rate according to the estimates, is affected by forward guidance and Public} Sector Purchase Programme (PSPP) announcements. However, only at the 10% signifi cance level which may not be very robust.

(27)

affected, as the financial system is very open and interconnected nowadays, it is very easy to influence the financing credit conditions abroad.

The five and ten-year CDS spreads were significantly affected by announcements of FOR and the ABSPP. The results seem to be contradictive as the purchase programmes affect the CDS spreads with different signs. Higher liquidity in the market shouldn’t necessarily affect CDS spreads, only if agents fear that banks are in danger due to liquidity problems. Therefore, there might be an effect from the announcements on the CDS spreads, however the sign of the effect is unpredictable, since the announcements may signal that the economic framework is changing in a positive way or they may screen that liquidity problems are coming. In addition, is important to be aware of the fact that the regression for the CDS spreads are run for a smaller time window (approximately one year less than the other variables), which may affect the estimates, making the impact non-comparable with the other variables.

These results seem to suggest that the announcements led to a shift in portfolios from the UK to the Eurozone, probably due to increased liquidity and confidence, promoting financial stability in the Eurozone market. However, this question remains to be answered.

Clearly, the most significant policy was the CSPP. However, it is worth mentioning that although the announcements of the CSPP were only performed once in the full sample, a separate dummy variable was created to account for this program, in order to avoid inflating other variables that are announced in the same day.

Interestingly, no significant results were found for the effect of OMT announcements, which contrasts with Fratzscher et al. (2014) where these announcements boosted equity prices and sovereign yields, while appreciating the Euro. In addition, the baseline results for the SMP announcements differ clearly from the results in Fratzscher et al. (2014); while in this specification only the five-year yield was affected, in the referred publication only the exchange rate was affected, in a positive way.

(28)

Table 2: Coefficients from the systemized variable 𝑼𝑴𝑷𝒕𝑬𝑪𝑩 represented

Note: *** p<0,01, **p<0,05, *p<0,1. Newey-west standard errors are in parenthesis below.

LTRO FRTPFA COLL CBPP FOR FWG EXCHANGE RATE -0.0021 (0.0015) 0.00416 (0.00284) 0.00134 (0.0009) 0.0019 (0.0041) 0.0005 (0.0012) 0.0028 (0.00404) 10-YEAR SOVEREIGN YIELD -0.0075** (0.0038) 0.0068 (0.0073) 0.0054 (0.0071) 0.0165*** (0.0063) 0.0058* (0.0034) 0.0049 (0.0096) 5-YEAR SOVEREIGN YIELD -0.01109 (0.00747) 0.00848 (0.01565) 0.0123 (0.01541) 0.0198* (0.0105) 0.0117* (0.0063) -0.0024 (0.0221) STOCK MARKET INDEX -0.0031 (0.0028) -0.0002 (0.003) -0.0105 (0.0083) -0.0027 (0.0035) -0.0003 (0.0021) 0.0131* (0.0067) 3-MONTH INTERBANK RATE 0.0057 (0.0097) -0.0019 (0.0169) -0.0031 (0.0027) 0.0003 (0.0042) 0.0017 (0.0061) 0.0033* (0.0019) 5-YEAR CDS SPREADS -0.0177 (0.0189) 0.0177 (0.0189) 0.0158 (0.0151) 0.0332** (0.0148) -0.0292** (0.0121) 0.1526** (0.0764) 10-YEAR CDS SPREADS -0.0214 (0.0178) 0.0159 (0.0179) 0.0121 (0.0136) 0.0239* (0.0134) -0.0264** (0.0119) 0.133* (0.0751)

7. Robustness tests

In this section the robustness tests of the results found in the previous chapter are presented. I perform three main types of robustness tests: The first one tries to account for the missing data problem encountered with the data to compute the variable “news”. The second sensitivity test is a check of the event-window. The third specification uses a different date of the policy rate change from the ECB. Finally, the last robustness test attempts to account for the possible simultaneity problem that arises in the baseline specification.

ABSPP TLTRO PSPP OMT SMP CSPP EXCHANGE RATE 0.0016 (0.0041) -0.0055* (0.0029) 0.00167 (0.003) -0.0004 (0.0019) -0.0029 (0.003) 0.0153*** (0.0051) 10-YEAR SOVEREIGN YIELD -0.0011 (0.0081) -0.0024 (0.0073) 0.0011 (0.0124) -0.0081 (0.0108) -0.009 (0.0079) 0.0349** (0.0171) 5-YEAR SOVEREIGN YIELD -0.011 (0.0162) -0.0052 (0.0138) -0.0004 (0.0199) -0.0045 (0.0171) -0.0249** (0.0125) 0.1083*** (0.0319) STOCK MARKET INDEX 0.0015 (0.0049) -0.0006 (0.0026) -0.0064** (0.0027) -0.00002 (0.0034) 0.0026 (0.0088) -0.0114* (0.0069) 3-MONTH INTERBANK RATE 0.0007 (0.0028) 0.0019 (0.002) 0.0023* (0.0014) 0.0027 (0.0025) -0.0006 (0.0031) -0.0104 (0.0209) 5-YEAR CDS SPREADS -0.0617*** (0.0201) 0.0289 (0.0354) 0.0187 (0.0186) -0.0157 (0.0129) -0.048 (0.0337) 0.0603 (0.0474) 10-YEAR CDS SPREADS -0.0308** (0.0152) 0.0231 (0.0474) 0.0194 (0.0248) -0.01 (0.0091) -0.0496 (0.0346) 0.0729 (0.0624)

(29)

7.1. Missing data

First of all, regarding the “macroeconomic news” variable, the missing observations that were coded as zero meaning that there was no surprise, are going to be dropped. Secondly, the surprises that were dropped are going to be accounted for in the day after, so that it is possible to account for their effect. Lastly, I drop the variables where I assumed rational expectations to observe on how much does the impact differs. Nevertheless, the results are robust to missing data with these three different assumptions, changing slightly the magnitude of the coefficients, although not in a significant manner.29

7.2. Event Window

One of the biggest judgements in an event study is to define the event-window. If this is too short, one may be missing the full market reaction but if it is too large, the event may be contaminated with “noise” from other surrounding events. Therefore, as further robustness test I test the range of the event-window, repeating the analysis for a window of two and three days. Note that in the baseline specification a one-day event-window was used. The objective is to allow for a lag in the response of the markets and to observe if the res ults change in a significant manner.

With a two-day event-window, and analysing the announcements as a whole, the sign and significance of the estimates is the same, only changing their magnitude. With the separated policies the significance changes substantially while the differences in the sign or magnitude are negligible. The estimates of the regression are presented in the Table 12. Overall, the results are robust to the enlargement of the event-window, although with some exceptions. The spillovers seem to be somewhat stronger than comparing with the baseline scenario for the sovereign yields, which suggests a slight delayed response from the financial markets. In contrast, the CDS spreads lost some significance.

With a third-day event-window, the significance and the sign of the coefficients change substantially with the systemized policies which suggests that the baseline estimation is no longer robust. However, this can also arise from the fact that the window may be too big and may be contaminated by effects from other events. This is the drawback of a too broad

(30)

window. The estimation results of this robustness test are represented in Table 13 in the Appendix.

7.3. ECB’s Policy rate

As referred in Section 5.2, there are two possible specifications of the policy rate from the central bank that can be used in this study. The difference between them is that the variations are accounted for in different dates: in the baseline specification the date of the announcement by the Governing Council was taken into account while in this section the date at which the policy rate change takes place is considered. The latter is represented in Table 7. Nonetheless, the results remain robust to this specification. 30

7.4. Lagging the dependent variable

Lastly, I include in the regression a lag of the dependent variable to attempt to account for the possible simultaneity effect that may be captured in the regression coefficients. If one assumes that the policy only responds to the lagged value of the dependent variable and not today’s value, including the lag should help controlling for these influences on the dependent variable, which possibly contributes to solving this problem. The results remain the same although with small changes in the systemized regression, but not significant.31

8. Discussion and further research suggestions

In this section, critiques to this work are provided. Suggestions for further research follow the critiques, stressing the most interesting options.

8.1. Discussion

In this thesis the innovation on the methodology relies on the use of daily data. Falagiarda et al. (2015) argues that to measure international spillovers the use of daily data is preferred since there are longer transmission lags. However, the use of intraday data has been regarded

30_{Regression output not included for brevity.} 31_{Regression output not included for brevity.}

The spillovers of ECB unconventional measures : an empirical analysis for the UK