Quantitative easing and the Eurozone crisis : the effect of the European Central Bank’s QE announcements on financial institutions in the Eurozone

(1)

University of Amsterdam

Quantitative Easing and the Eurozone Crisis

The effect of the European Central Bank’s QE announcements on

financial institutions in the Eurozone

Sam Rimmelzwaan

Faculty of Economics & Business Supervisor

Date

Master Thesis

Business Economics: Finance Tanju Yorulmazer

July 7, 2015

Abstract During the latest financial crisis the European Central Bank (ECB) launched several unconventional monetary policy measures to spur economic activity in the Eurozone. Using an event study methodology this paper analyzes the effect of these quantitative easing (QE) announcements on banks within the Eurozone from 2009 to early 2015 and shows that they had a significant impact on the value of Eurozone banks. The paper also shows that banks from PIIGS countries experienced relatively higher abnormal returns than banks outside this region and it provides evidence that suggests that the market anticipated financially weaker banks to have more benefit from QE.

Keywords Quantitative easing ∙ Monetary policy ∙ Eurozone ∙ PIIGS ∙ European Central Bank ∙ Event study

(2)

Statement of Originality

This document is written by Sam Rimmelzwaan who declares to take full responsibility for the contents of this document.

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

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

(3)

1 Introduction

The latest financial crisis has shown that the more conventional instrument of central banks, by using interest rates, to conduct their monetary policy does not always work. Since it is always possible to hold savings as currency instead of in a bank deposit it is not possible to let the short-term nominal interest rate go much below zero (Fawly & Neely, 2013). With short-short-term nominal interest rates reaching their lower bound of zero percent, central banks went looking for more unconventional methods to boost the economy.

A relatively new method used by central banks is called quantitative easing. The bank of Japan introduced QE in March 2001, by raising the ceiling on commercial bank current account balances and increasing its purchases of Japanese Government Bonds, to spur economic activity through a portfolio rebalancing effect (Kimura & Small, 2006; Oda & Ueda, 2005). Amid the financial crisis the United States Federal Reserve (Fed) also started a QE program, with large-scale asset purchases (LSAPs) starting in November 2008 and ending late 2014.1 During this period the Federal Reserve announced that it would credit its balance to purchase large amounts of mortgage backed securities and treasury bonds worth over 3,000 billion euros. The Bank of England (BOE) started with QE in March 2009 and purchased about 500 billion euros worth of government debt until July 2012.2 When it became clear that the Greek debt crisis was becoming unsustainable the ECB followed the same procedure with a Securities Markets Program (SMP), buying large amounts of government debt from secondary markets.3 Together with the introduction of a European Stability Mechanism (ESM) and a European Financial Stability Facility (EFSF) the SMP became part of a comprehensive package of measures to preserve financial stability in Europe (Council of the European Union, 2010). In addition to the SMP the ECB increased its balance through the launch of three-year long-term refinancing operations (LTROs) in December 2011 and February 2012. Given the higher liquidity of banks in the Northern European countries, mainly banks of Southern European countries such as Portugal, Ireland, Italy, Greece, Spain (so-called PIIGS) but also France made use of LTROs.4 Both measures were still relatively small in comparison with those of the BOJ, Fed and BOE. However in July 2012 central bank president Mario Draghi ensured that the ECB was ready to do whatever it takes to preserve the euro. Two and a half years later, on the 22nd of January 2015, the ECB

1_{http://www.federalreserve.gov/newsevents/press/monetary/20141029a.htm} 2_{http://www.bankofengland.co.uk/monetarypolicy/Pages/qe/qe_faqs.aspx} 3

In addition to the Eurosystem’s regular open market operations the ECB launched a set of non-standard monetary policy measures: https://www.ecb.europa.eu/mopo/implement/omo/html/index.en.html

4_{An article published by the Financial Times that shows which banks tapped into the December 2011 489bn euro}

(5)

brought out a press release stating that it would expand its asset purchases with the Public Sector Purchase Program (PSPP), buying 60 billion euros worth of sovereign debt per month until at least September 2016. Although it was expected that the ECB would expand its existing QE program, the new announcement exceeded investors’ expectations greatly.5

As with earlier programs the expanded asset purchase program is associated with monetary expansion of the ECB by increasing its balance sheet. Figure 1 gives an overview of the composition of the ECB’s balance sheet during the financial and sovereign debt crisis. The results of the first four months of sovereign debt purchases are already visible, with the amount of securities of euro area residents showing a sharp increase on the balance sheet.

The effects of QE on the economy have already been investigated through several studies. The results on the action of the Federal Reserve to buy large quantities of government debt suggest that QE affected financial markets significantly. Krishnamurthy and Vissing-Jorgensen (2011) found for example that the Federal Reserve’s first QE program (QE1) and second QE program (QE2) both gave evidence for a declining yield for all bonds in the US. Furthermore Joyce et al. (2011) suggest, although not with full certainty, that the QE program of the Bank of England might have had a small positive effect on UK stock prices. Some studies have been done on the contribution of earlier ECB programs to spur economic activity in the Eurozone, all claiming to have found some evidence but only convincing to some extent.

There is still a lot of discussion about the effects of quantitative easing, which makes it an interesting subject during today’s economic environment. A lot of supporters think this unconventional monetary policy helped to keep the US, UK and Japanese economy from tumbling into an undesired deflationary spiral. On the contrary, opponents find QE highly undesirable for the Eurozone as it might create incentives for EU member countries to increase their debt instead of reforming inefficiencies. They also think that QE would work differently in the Eurozone since it regards a union with strong heterogeneity and different levels of risk in government bonds and therefore could lead to a redistribution of risks between taxpayers in the member countries. 6

Considering that the Federal Reserve ended their large-scale assets purchases very recently and for the US it is labeled as a successful strategy according to most economists and

5_{Several institutions conducted a survey on investor expectations and found that the majority expected a purchasing}

program of around 500 billion euros. Bloomberg for example polled 60 economists asking their expectations about the ECB’s QE program: http://www.bloomberg.com/news/articles/2015-01-19/draghi-s-big-push-seen-delivering-635-billion-with-qe-this-week, while Société Générale polled 176 clients on their views on the ECB and Eurozone quantitative easing: http://www.ft.com/intl/fastft/263742/ecb-set-500bn-qe-plan-socgen-survey

(6)

Fed officials, it is interesting to look into the reaction of the market towards the ECB’s QE announcements. Given the different political policies between Europe’s Monetary Union member states many economists think that QE will affect the European economy in a different way than the United States economy. However, as is suggested in existing literature it is very hard to capture the real effects of QE on the economy as it works through different channels.7 With the real effects on the economy in general being hard to measure it is interesting to look at the fraction of the economy that has been experiencing distress since the start of the crisis, the financial sector. Therefore raising the question: “What has been the effect of QE announcements by the European Central bank on financial institutions in the Eurozone?”

With the sovereign debt crisis still present and some member countries of the Eurozone only showing a fragile recovery it is important for the economy to regain its trust. The ECB’s newest QE program could contribute to this and spur economic activity. However, with existing literature concentrating on the effect of QE on interest rates and mortgage rates, I focus mainly on investor sentiment towards these QE announcements and therefore the effect the announcements have had on stock prices of listed financial institutions in the Eurozone. The results could give an indication if the ECB’s monetary policy of QE has contributed to stabilizing the financial sector and if investors anticipated a positive impact.

In this thesis I use an event-study approach to estimate the effect of a number of announcements by the ECB, that were part of its unconventional monetary policy, on two portfolios of Eurozone banks. I examine abnormal returns on a portfolio of all Eurozone banks, excluding banks from PIIGS countries and a portfolio of banks that consists only of banks from PIIGS. The timeline runs from the ECB’s announcement of the first Covered Bonds Purchase Program (CBPP) in 2009 through the announcement of the expanded asset purchase program, adding the PSPP policy to its existing asset purchases. The results indicate that while raw bank returns overall were generally positive, both portfolios experienced significant positive a returns around the introduction of the SMP and the ‘whatever it takes’ speech of ECB President Mario Draghi. A remarkable result is that the market significantly outperformed both portfolios on the announcement of the expanded asset purchase program. With both portfolios experiencing significant negative abnormal returns the results suggest that the magnitude of the program, which was higher than investors expected, gave the impression that the Eurozone economy found itself in a worse state than what was thought.

7_{See for example Bowman, Cai, Davies and Kamin (2011), Darracq-Paries and Santis (2013), Joyce, Miles, Scott}

(7)

Next I calculate abnormal and cumulative abnormal returns for 27 Eurozone banks on the announcement date of the SMP, the speech of Mario Draghi and on the announcement of the expanded asset purchase program, where both bank portfolios returned significant abnormal returns. Although not quantifiable, the results present some evidence that banks from PIIGS countries experience higher abnormal returns than banks from non-PIIGS countries.

Finally I conduct a standard cross-sectional event study to examine whether certain bank characteristics had impact on abnormal returns of individual banks. For abnormal returns of the 27 Eurozone banks I estimate the implication of six key indicators that are globally used to measure banks’ financial condition. I find four variables that show a significant relation between bank characteristics and abnormal returns. Although some regression estimators of the different bank characteristics contradict with what is expected, the results to some extent confirm the hypothesis that the market expected that financially weaker banks would benefit more from QE.

The remainder of this thesis is divided into seven sections. Section 2 gives a brief summary of existing literature on QE and lays out some results on the impact of QE in other western economies. It follows with the main hypotheses, which are based on existing literature, and support the research question. Section 3 describes the methodology used to test whether these hypotheses are true and the data that is used and how it is collected. Section 4 concentrates on the results regarding cumulative abnormal returns on both banking portfolios and individual banks and on results from the cross-sectional regression model on bank characteristics. Additional results and robustness checks are given in Section 5. Section 6 provides a brief discussion on the results and section 7 concludes.

2 Literature Review and Hypotheses

2.1 Background information on quantitative easing

The overall view of economists, before the BOJ started with QE, was that when short-term nominal interest rates approached zero there was nothing more a central bank’s monetary policy could do to stimulate the economy. Fawly and Neely (2013) mention that concerns about this zero bound on interest rates date back at least to Keynes (1936). Because savings can be hold as currency, which effectively pays a nominal interest rate of zero, the short-term nominal interest rate cannot be pushed below zero thus creating a bound on the lower-end. Bernanke, Reinhart and Sack (2004) argue that the real short-term rate in such a case may be higher than the rate that is needed to ensure stable prices. This could lead to downward pressure on costs and prices that, in

(8)

turn, raises the real short-term rate even further and will eventually create a deflationary spiral. In other work Bernanke and Reinhart (2004) discuss three other strategies of monetary policy that can be adopted when reaching the zero lower bound. First a central bank could provide insurance that the short-term rate will be kept lower in the future than is currently expected by the market. Second, it could alter the composition of its balance sheet thereby changing the relative supply of securities in the marketplace. Third, it could expand the size of its balance sheet beyond the level that is needed to set short-term rates at zero, which is formally called “quantitative easing”. This expansion of the central banks’ balance sheet can lead to an increase in the money supply through increased bank lending. Romer (1992) argues that the US recovery from the Great Depression is closely related to such an increase in the money supply. Although the increase worked through different channels, like the devaluation of the dollar and increasing instability in Europe, she suggests that conventional aggregate demand stimulus, primarily in the form of monetary expansion, largely contributed to the rapid growth of real output during mid- and late 1930s.

Mishkin (1996) also describes the channels through which different forms of monetary policy work. He concludes that the general conception that monetary policy is impotent when short-term rates reach zero is “demonstrably false”. He argues that increasing liquidity in the economy does not always have to be conducted through the purchase of short-term government securities. In fact he states that monetary expansion helps to revive the economy in a way that it raises general price level expectations and contributes to reflating other asset prices, which stimulate aggregate demand. Usually this expansion of the monetary base includes asset purchases and lending programs to inject reserves into the economy.

Central banks used, since the introduction of QE, a combination of both methods to increase the money supply but the details depended on the particular structure of their economies. With their more bank-centric economies, the ECB and BOJ for example initially injected reserves by generously lending money to banks. The US and UK however, consist traditionally of a more market-centric economy, which led to the Fed and BOE injecting reserves by purchasing bonds (Fawly & Neely, 2013). Despite its bank-centric economy the PSPP that was announced by the ECB on January 22 is much more similar to that of the Fed and BOE.

2.2 Quantitative easing in other western economies

Although QE is a relatively new instrument to spur economic activity there are various publications about the effect it has on the economy. Since the Bank of Japan initiated QE in March 2001 economists were able to study direct and long and short-term effects of the BOJ’s monetary policy on the Japanese economy. For the US, UK and to some extent for the Eurozone

(9)

research on the direct and short-term effects has also been done. The results are good predictors for possible outcomes for this study and combined with the theory based on their conclusions could support the hypotheses.

One of the first empirical studies, that has been done on the effects of non-standard monetary policy, by Bernanke, Reinhart and Sack (2004) already pointed to possible positive outcomes of alternative monetary policy on the economy. Nevertheless, the authors were only able to quantitatively support the success of using communications to affect expectations of the future short-term rate and thus longer-term yields in Japan and the US. It should be noted that their research on QE, here defined as a central bank expanding its balance sheet above the level that is needed to push rates to zero, however, is limited to Japan. The US around that time only deviated from standard policy by changes in the relative supply of Treasury securities to affect long-term rates. The QE program of the BOJ, which was launched in March 2001, however consisted of several “conventional” and “unconventional” expansionary policies that were announced until it was lifted in March 2006. Kobayashi et al. (2006) found that the introduction and expansion announcements of the BOJ’s QE policy led to excess returns in the Japanese banking sector, indicating that the market anticipated a positive impact.

The anticipation of the market on the positive impact of QE studied by Kobayashi et al. (2006) is supported by findings of Bowman et al. (2011) and Kimura and Small (2006). Bowman et al. (2011) use bank-level data from 2000 to 2009 to examine the effectiveness of QE in promoting bank lending and found a positive effect of bank liquidity positions on lending. This suggests that the expansion of the reserves at the BOJ most likely boosted the flow of credit to the economy. Although they are almost certain that the QE policy has been helpful they found that the overall size of the boost was relatively small. This modest effect can be derived from the estimated response of lending to liquidity positions in their regressions, which is also quite small. Furthermore the BOJ’s reserve injections were offset by a decrease in interbank lending which means that overall liquidity rose by less than banks current accounts balances with the BOJ. In addition to the effect on bank liquidity Kimura and Small (2006) found that QE contributed to a decrease in volatility in some asset markets, affecting prices of these assets and lowering returns in these markets. Bowman et al. (2011) and Kimura and Small (2006) therefore both give evidence that the BOJ’s QE program had an intended outcome, however it is argued that the results should be carefully interpreted.

As with the BOJ’s QE program, research has been done on the effects of the Federal Reserve’s LSAPs as well. Gagnon et al. (2011) and Krishnamurthy and Vissing-Jorgensen (2011)

(10)

found a large and significant impact of QE on mortgage-based security rates through different channels. Especially QE1, which involved large purchases of agency MBS, showed a large decrease in MBS rates in contrast to QE2, which involved buying only Treasuries, where no such effects were found. For QE1 they found a significant effect through the default risk channel by measuring the change in CDS-spreads on corporate bonds. Fuster and Willen (2010) show that this decrease in MBS rates resulted in mortgage lenders quickly reducing their mortgage rates to households. However, Stroebel and Taylor (2009) found, when controlling for the effects of default and prepayment risk, that the impact of the MBS purchasing program of the Federal Reserve on US mortgage spreads has been small and uncertain.

Krishnamurthy and Vissing-Jorgensen (2011) also argue that by purchasing large amounts of long duration assets the central bank can signal that it will not raise short-term rates in the near future. If the central bank raises rates it will take a loss on the long duration assets. Assuming that the central bank takes losses into account for its decision-making, purchasing these assets serves as a credible commitment to keep interest rates low. However, Gagnon et al. (2011) mention that the Fed at the time was in fact deliberately informing investors about the possibility of increasing short-term interest rates and therefore did not use LSAPs as a signal that short-term rates would remain low. They argue that other channels have to be taken into consideration for explaining the decrease in long-term rates and describe the lower risk premiums to be the primary channel through which the LSAPs have worked.

The Bank of England started a program, which looked more similar to QE2 of the Federal Reserve. It consisted of buying both private- and public sector assets, but the majority of purchases would be of UK government securities or “gilts” (Joyce et al., 2011). In their work Joyce et al. (2011) found that medium to long-term gilt yields were about 100 basis points lower than they would otherwise have been as a result of QE. They also suggest, however with considerable uncertainty, that QE most likely had a broader effect on the UK economy, with most other asset prices showing a noticeable recovery through 2009. Their theory is partially confirmed by Kapetanios et al. (2012) who examine the macroeconomic impact of the first round of QE by the BOE. Their estimates suggest some effect on real GDP and on annual CPI inflation and conclude that without QE real GDP would have fallen even more and inflation would be lower or even negative. However, Lyonnet and Werner (2012), who seize the findings of Joyce et al. (2011) to further investigate the transmission of the QE effects to the wider economy, conclude that their empirical analysis indicates that QE had no effect on the UK economy. Together with the mentioning of Bowmen et al. (2011), Joyce et al. (2011), Kimura and Small

(11)

(2006) and Krishnamurty and Vissing-Jorgensen (2011) among others, the findings of Lyonnet and Werner (2012) show that the positive effects of QE on the real economy have still not been identified with full certainty.

An equivalent program to that of the BOE and the Fed is the Securities Markets Program of the ECB, which impact is investigated by Eser and Schwaab (2013). They found that the SMP lowered yields on five-year maturity bonds for PIIGS significantly as well as decreasing bond yield volatility on intervention days. This support program together with other ECB interventions is associated with a significant improvement in economic activity according to the findings of Giannone et al. (2012). They argue that two and a half years after the failure of Lehman Brothers the level of industrial production was about 2% higher, and the unemployment rate 0.6 percentage points lower, than it would have been in the absence of the ECB’s unconventional monetary policy measures.

The three-year LTROs, which can be considered as the next big QE like intervention of the ECB after the announcement of the SMP, had a significant impact according to Darracq-Paries and Santis (2013). They used data from the April 2012 Bank Lending Survey (BLS) and an unpublished ad-hoc questionnaire of the BLS in February 2012 to estimate the macroeconomic impact of the three-year LTROs on the economy of the Eurozone. Their empirical analysis assumes that the main transmission channel of this unconventional monetary policy works through the decrease in liquidity- and funding risks in the banking system of the Eurozone. Darracq-Paries and Santis (2013) argue that the launch of these three-year LTROs not just translate into lower cost of financing but they act more as credit easing due to increases in GDP, loan volume to non-financial corporations and a narrowing of lending rate spreads.

Since most studies aim their conclusion at QE having positive effects on the real economy, although small and not with full certainty, it is likely that the ECB’s QE program will spur economic activity in the Eurozone. Where the existing literature measures these effects through GDP growth, interest rates, and bond- and other asset prices I examine the effect QE has on stabilizing the financial sector. Under the assumption of the efficient market hypothesis that all publicly available information and future expectations are priced into securities the anticipated impact of QE on financial markets should be measurable through bank equity values and their abnormal returns on QE announcement dates (Kobayashi et al., 2006).

(12)

2.3 Hypotheses

Above mentioned theory constitutes to forming the hypotheses that I use to examine the effect of the ECB’s QE announcement on equity values for Eurozone banks. I derive three hypotheses that I elaborate on in the next section and use for the interpretation of the results.

2.3.1 Hypothesis 1

“The QE announcements of the ECB result in positive abnormal returns for Eurozone banks.” Several aforementioned studies point to the possible positive effects of QE on the financial sector and on the economy as a whole (Gagnon et al., 2011; Kapetanios et al., 2012; Krishnamurthy & Vissing-Jorgensen, 2011). Theoretical evidence therefore leads to the expectation of a positive reaction by the market. Furthermore, after a five-year period of QE by the BOJ, which announced at unexpected moments to raise its current account balance and buy Japanese government debt, Kobayashi et al. (2006) studied the anticipated impact on bank values. They found that almost every announcement had significant positive effects on the total bank index, suggesting that the market expected favorable conditions as a result of QE.

“Banks from PIIGS countries experience relatively higher abnormal returns from QE announcements than other Eurozone banks”

Most banks from these countries faced financial constraints due to the dry-up of the credit supply and worsening credibility of their sovereign debt. The worsening condition of these countries ultimately led to the launch of the Securities Markets Program and the launch of the one- and three-year LTROs, that were widely used by banks from PIIGS (Eser & Schwaab, 2013). Because these countries were troubled most during the crisis it is expected that these countries, hence their financial sector should benefit the most.

“It’s anticipated that financially weaker banks benefit more from QE.”

In addition to their results on the positive anticipation of the market on QE announcements, Kobayashi et al. (2006) found cross-sectional evidence that the market anticipated that QE would disproportionally benefit weaker banks. The market could have been anticipating similar effects on the ECB’s QE announcements. Furthermore, assuming that banks from the Southern European countries are financially weaker than those from Northern European countries, the results on the third hypothesis could also contribute to answering the second.

(13)

Kobayashi et al. (2006) use seven indicators of individual bank financial conditions to examine the implications of bank characteristics for observed abnormal returns. They use indicators that are related mainly to liquidity, profitability and deposit growth and control for the size of individual banks by adding a variable that measures scale by the logarithm of total assets. Their model returns significant results for some indicators, which they consider to give evidence for the relation between financial condition and the level of abnormal returns.

In addition to the model of Kobayashi et al. (2006) there are more ways to measure the financial condition of a bank. The different characteristics that are affiliated with for example its credit rating are widely used to measure financial condition. Credit agencies, auditors and bank regulators have traditionally relied on the CAMEL model, which is an acronym of the words Capital, Asset quality, Management, Earnings and Liquidity (Hirtle & Lopez, 1999).8 In research papers the model is widely used, whether or not in a modified or extended version. Pasiouras, Gaganis and Zopounidis (2006), West (1985) and Wheelock and Wilson (2000) among others make use of similar versions of the model. I use indicators from such a model, which enables me to examine cross-sectional differences between banks in the Eurozone and the characteristics that are implicitly associated with their abnormal returns, to investigate if there is a possible relation between a banks’ financial condition and its abnormal returns on QE announcement dates.

3 Methodology

This section discusses how the data is structured and the methodology that is used in the regression analysis. The next paragraph gives an outline of the data used in the research, the sources used to extract this data and the selection criteria. The second paragraph provides some descriptive statistics. Thereafter I provide an overview of all the regression models and related variables applied in this paper. First I concentrate on the impact of decisions made by the ECB, that are part of its unconventional monetary policy, on two banking portfolios during the period January 2009 through February 2015. Appendix A provides a list of unconventional monetary policy decisions announced by the ECB during this period. For three events that amounted in significant abnormal returns on the two banking portfolios I conduct an event study on cross-sectional differences in cumulative abnormal returns (CARs) between different banks.

8

A sixth component was added to the model in 1996, which measured Sensitivity to market risk, changing the acronym to CAMELS. This last component measures the extent to which a banks’ earnings and the value of its assets and liabilities are sensitive to changes in interest rates under various scenarios and stress environments (Sahajwala and van den Berg, 2000).

(14)

3.1 Data

The event dates in the sample are hand-collected from ECB press releases and the ECB financial crisis timeline. I selected those events that were part of the ECB’s non-standard monetary policy and were aimed at easing the whole financial system of the Eurozone. To address the research question I used the EMU EX PIIGS and PIIGS indices constructed by Datastream and calculated their log returns to examine the difference between banks from Northern European (EMU EX PIIGS) and weaker Southern European (PIIGS) countries. Furthermore I collected historical stock prices of the largest banks from the five largest economies within the Eurozone, excluding Spain and Italy, and the five countries from the PIIGS region (see Appendix B for an overview of the sample). Stock prices and index returns are all collected through Datastream and used to calculate log returns for bank equity values as well as the STOXX Europe 600 index (STOXX 600). Finally I selected seven indicators of financial conditions for individual banks to investigate cross-sectional differences between Eurozone banks. Annual data on individual banks is collected through Datastream and missing values are hand-collected from banks’ annual reports.

3.2 Descriptive statistics

A short summary on the data for this regression is given in Table 1. The highest return in the dataset for the STOXX 600, EMU EX PIIGS and PIIGS portfolios is 6.9%, 16.4% and 18.6% respectively. These returns were all earned on May 10, 2010 on the announcement date of the SMP, ESM and EFSF. These exceptionally high returns already suggest that investors reacted positively on the announcement of this set of policy measures.

Table 2 presents descriptive statistics for the dependent and independent variables used in the regression on cross-sectional differences between banks. A distinction is made between the data that is used for each of the three event dates. The mean (cumulative) abnormal return is fairly positive for the first two events and negative for the third event. The liquidity ratio, measured by dividing liquid assets by customer deposits and short-term funding lies on average close to one, indicating that the average bank was capable of meeting sudden withdrawals. The mean tier 1 capital ratio has risen from 0.104 to 0.124 since 2010, with not a single bank reporting a ratio of below 0.105 in 2014. Loan provisions to net income ratio, defined as Loanperf, shows loan provisions are on average almost 50% of net interest revenue for 2010 and 2014. In 2012 these loan provisions were on average relatively almost twice as high as in other years. Profitability, measured by dividing net income by total assets was on average positive in 2010 and shows a negative result for 2012 and 2014. The efficiency of managing expenses that is measured by dividing OPEX by net interest income seems to report higher values for 2012. With

(15)

a relatively higher mean and higher minimum and maximum values it seems that banks on average were less cost efficient in 2012. Finally the average size of the sample of banks, measured as a logarithm of total assets, shows a small increase over time.

3.3 Bank portfolio return on ECB announcements

To estimate the effect of the ECB’s unconventional decisions on the Eurozone banking sector I pursue an event study approach to examine both the overall return of the Eurozone banking sector as well as its abnormal returns. From the announcement of the first covered bonds program in May 2009 till the latest announcement of the expanded asset purchase program in January 2015 I identified 10 event dates associated with unconventional monetary policy announcements of the ECB. To compare the difference between the weaker and stronger economies of the Eurozone a distinction is made between a portfolio of banks that consists only of banks from PIIGS countries (PIIGS) and a portfolio of all Eurozone banks but excluding those from PIIGS countries (EMU EX PIIGS). I calculate abnormal returns using the market model approach, following from theory of Mackinlay (1997). Normal returns are predicted during an estimation window twenty days prior to an announcement of the ECB. Comparing the prediction of these normal returns with the actual returns on the event horizon results in the abnormal returns for the both portfolios. The standard market model is used to estimate the daily expected returns for both portfolios, which looks as follows:

𝑅𝑅𝑖𝑖𝑖𝑖 = 𝛼𝛼𝑖𝑖 + 𝛽𝛽𝑖𝑖𝑅𝑅𝑚𝑚𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑖𝑖 (1)

𝐸𝐸(𝜀𝜀𝑖𝑖𝑖𝑖 = 0) 𝑣𝑣𝑣𝑣𝑣𝑣(𝜀𝜀𝑖𝑖𝑖𝑖) = 𝜎𝜎𝜀𝜀2𝑖𝑖

where 𝑅𝑅_{𝑖𝑖𝑖𝑖} and 𝑅𝑅_{𝑚𝑚𝑖𝑖} are the rates of return of portfolio i and the market at date t, respectively. The error term 𝜀𝜀_{𝑖𝑖𝑖𝑖} is assumed to have zero mean, to be independent of 𝑅𝑅_{𝑚𝑚𝑖𝑖}, and to be uncorrelated across firms. 𝛼𝛼�_𝑖𝑖, 𝛽𝛽̂_𝑖𝑖, and 𝜎𝜎_𝜀𝜀2_𝑖𝑖 are the parameters of the market model.

Abnormal returns are calculated as the difference between the realized return and the estimated return from an Ordinary Least Squares (OLS) regression that is used to estimate 𝛼𝛼�𝑖𝑖 and 𝛽𝛽̂𝑖𝑖. The sample abnormal return is therefore defined as:

𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖} = 𝑅𝑅_{𝑖𝑖𝑖𝑖}− 𝛼𝛼�_𝑖𝑖 − 𝛽𝛽̂_𝑖𝑖𝑅𝑅_{𝑚𝑚𝑖𝑖} (2) The parameters of the market model are estimated using a T = 100 trading day window, beginning 120 days before each event date. As it may take some time for the market to evaluate the news, although avoiding the risk of the estimated reaction being driven by other news events, a two-day event window (0,+1) is used for nine out of ten events. Since the ESM and EFSF were

(16)

introduced on Sunday, May 9, 2010 and the SMP followed a day later that morning, only a one-day window is used.

To determine whether abnormal returns are significantly different from zero, each abnormal return is standardized using the Mikkelson and Partch (1986) test statistic:

𝑆𝑆𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖} =𝐴𝐴𝐴𝐴𝑖𝑖𝑖𝑖 𝑆𝑆𝑖𝑖𝑖𝑖 ∼ 𝑡𝑡(𝑇𝑇 − 2), (3) Where 𝑆𝑆𝑖𝑖𝑖𝑖 = 𝜎𝜎𝑖𝑖�1 +₁₀₀1 + �𝐴𝐴𝑚𝑚𝑖𝑖−𝐴𝐴𝑚𝑚𝑖𝑖� 2 ∑−21𝑚𝑚=−120�𝐴𝐴𝑚𝑚𝑚𝑚−𝐴𝐴𝑚𝑚𝑖𝑖�2 (4)

𝜎𝜎𝑖𝑖 is the residual standard error of portfolio i’s market model regression, 𝑅𝑅𝑚𝑚𝑖𝑖 is the market return

on day t, and 𝑅𝑅_{𝑚𝑚𝑖𝑖} is the average market return in the estimation period.

𝐶𝐶𝐴𝐴𝑅𝑅𝑖𝑖(𝑡𝑡1,𝑡𝑡2) is defined as the sample cumulative abnormal return from 𝑡𝑡1 to 𝑡𝑡2 (the event

window) to determine the abnormal returns for a period where 𝑡𝑡₁ ≤ 𝑡𝑡₂. The CAR from 𝑡𝑡₁ to 𝑡𝑡₂ is the sum of the included abnormal returns,

𝐶𝐶𝐴𝐴𝑅𝑅_𝑖𝑖�𝑡𝑡_1,𝑡𝑡₂� = ∑𝑖𝑖_{𝑖𝑖=𝑖𝑖}2 𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖}

1 (5)

where the standardized cumulative abnormal returns are given as: 𝑆𝑆𝐶𝐶𝐴𝐴𝑅𝑅𝑖𝑖𝑖𝑖 =

∑𝑖𝑖2_{𝑖𝑖=𝑖𝑖1}𝑆𝑆𝐴𝐴𝐴𝐴𝑖𝑖𝑖𝑖

�𝑖𝑖2−𝑖𝑖1+1 ∼ 𝑡𝑡(𝑇𝑇 − 2) (6)

The effect of QE announcements of the ECB on the two portfolios can either be negative, positive or zero. Although I expect to find results that are similar to the existing literature and support the hypothesis that the QE announcements of the ECB result in positive abnormal returns for Eurozone banks it could be possible that an announcement results in negative abnormal returns. This could happen for example when investors believe that the announced monetary policy will not result in a desired outcome or when it reveals something about the state of the economy that the market was not yet informed about.

3.4 Cross-sectional evidence on bank characteristics

Given the results of abnormal returns on both banking portfolios on QE announcement dates I examine cross-sectional differences in abnormal returns within a sample of 27 large banks in the Eurozone. I look at three announcement dates that had a significant impact on bank equity values for both portfolios. The first date is May 10, 2010, where the ECB introduced the launch of the SMP. Together with the announcement of the ESM and the EFSF the announcement of the SMP

(17)

led to a large increase on the overall banking portfolio as well as the market9. The second date is at the time of the speech of Mario Draghi. On July 26, 2012, Draghi (2012) stated the following: “Within our mandate, the ECB is ready to do whatever it takes to preserve the euro. And believe me, it will be enough”. Although no real action was taken on that day, the speech of the ECB president gave a strong signal of the measures the ECB was willing to take to stimulate the economy of the Eurozone. The market anticipated on the possibility of an asset purchasing program similar to that of the FED and the BOE, which led to a substantial increase of the STOXX Europe 600 index and a large drop in Italian and Spanish bond yields.10 Draghi’s ‘whatever it takes’ speech was finally proven on January 22, 2015 when the ECB announced that it would buy 60 billion euro worth of sovereign debt per month until at least September 2016. Although it was almost certain that the ECB would take action, it was the size of the program that greatly exceeded investors’ expectations.

To determine abnormal returns of individual banks I conduct an event study on the three aforementioned event dates, similar to the one that is used to calculate abnormal returns for both banking portfolios. As with the event study on the two portfolios, a two-day event-window interval is used for the events on July 26, 2012 and January 22, 2015. For the event on May 10, 2010 a one-day interval is used.

To find if financially weaker banks are more affected by QE announcements I look at the implication of seven bank specific characteristics for observed cumulative abnormal returns. I look at bank liquidity, capital strength, asset quality, profitability, efficiency in expenses management, size and bank type using the following OLS regression model:

𝐶𝐶𝐴𝐴𝑅𝑅𝑖𝑖𝑖𝑖 = 𝛽𝛽0+ 𝛽𝛽1𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑖𝑖+ 𝛽𝛽2𝐸𝐸𝐿𝐿𝐸𝐸𝐿𝐿𝑡𝑡𝐸𝐸𝑖𝑖𝑖𝑖+ 𝛽𝛽3𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿𝑖𝑖𝑖𝑖+ 𝛽𝛽4𝑃𝑃𝑣𝑣𝐿𝐿𝐿𝐿𝑖𝑖𝑖𝑖+ 𝛽𝛽5𝐶𝐶𝐿𝐿𝐶𝐶𝑡𝑡𝑖𝑖𝑖𝑖+

𝛽𝛽6𝑆𝑆𝐿𝐿𝑆𝑆𝐿𝐿𝑖𝑖𝑖𝑖+ 𝛽𝛽7𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝐷𝐷𝑆𝑆𝑖𝑖𝑖𝑖+ 𝜀𝜀𝑖𝑖𝑖𝑖 (7)

where 𝐿𝐿𝐿𝐿𝐿𝐿_{𝑖𝑖𝑖𝑖}, 𝐸𝐸𝐿𝐿𝐸𝐸𝐿𝐿𝑡𝑡𝐸𝐸_{𝑖𝑖𝑖𝑖}, 𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿_{𝑖𝑖𝑖𝑖}, 𝑃𝑃𝑣𝑣𝐿𝐿𝐿𝐿_{𝑖𝑖𝑖𝑖}, 𝐶𝐶𝐿𝐿𝐶𝐶𝑡𝑡_{𝑖𝑖𝑖𝑖}, 𝑆𝑆𝐿𝐿𝑆𝑆𝐿𝐿_{𝑖𝑖𝑖𝑖} are the independent variables based on the CAMEL model and theory from Pasiouras et al. (2006) explained in the next paragraph. 𝐷𝐷𝑃𝑃𝐷𝐷𝐷𝐷𝐷𝐷𝑆𝑆𝑖𝑖𝑖𝑖 represents a dummy variable that takes value one if a bank is located within one of the

countries from the PIIGS region and zero otherwise.

9_{On May10, 2010 the returns on the STOXX Europe 600, PIIGS and EMU EX PIIGS indices amounted around}

6,9%, 18,6% and 16,4% respectively.

(18)

3.5 Description of the variables

𝐿𝐿𝐿𝐿𝐿𝐿𝑖𝑖𝑖𝑖 is defined as liquidity of bank i at time t, which is measured as the liquid assets to customer

deposits & short term funding ratio.11 This ratio indicates the percentage of customer deposits and short term funding that could be withdrawn suddenly. The higher this ratio the more liquid the bank is and the better its financial condition. Based on the hypothesis I therefore expect 𝐿𝐿𝐿𝐿𝐿𝐿_{𝑖𝑖𝑖𝑖} to have a negative relation with the dependent variable 𝐶𝐶𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖} (𝛽𝛽₁ < 0).

Pasiouras et al. (2006) describe that capital strength can be defined by various measures. In their research they use a measure of dividing equity by total assets. They mention however that the Tier 1 ratio probably is a better measure but they suffer a huge number of missing data in their dataset. The Tier 1 ratio represents the ratio of Tier 1 Capital to total risk-weighted assets. Tier 1 capital is defined by common shareholders’ equity and qualifying preferred stock, less goodwill and other adjustments (Basel Committee, 1988; Basel Committee, 2004). 𝐸𝐸𝐿𝐿𝐸𝐸𝐿𝐿𝑡𝑡𝐸𝐸_{𝑖𝑖𝑖𝑖} is therefore defined as the Tier 1 capital ratio of bank i at time t. As with the liquidity ratio, the higher the Tier 1 ratio the better a bank’s financial condition and therefore I expect a negative relation between 𝐸𝐸𝐿𝐿𝐸𝐸𝐿𝐿𝑡𝑡𝐸𝐸_{𝑖𝑖𝑖𝑖} and 𝐶𝐶𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖} (𝛽𝛽₂ < 0).

A banks’ asset quality can be estimated by the quality of its loans. Loan loss provisions divided by net interest revenue explains the relationship between provisions for bad loans in the profit and loss account and the interest income it has earned over the same period. Because loan provision depends on the probability of a loan becoming non-performing, a higher provision usually indicates higher probability of non-performing loans and therefore implies lower asset quality. The asset quality of bank i at time t is therefore defined through a loan provision ratio using the variable 𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿_{𝑖𝑖𝑖𝑖}. A higher value of 𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿𝐿𝐿𝐿𝐿𝑣𝑣𝐿𝐿_{𝑖𝑖𝑖𝑖} therefore indicates a lower financial condition, hence it is expected to be positively related to 𝐶𝐶𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖} (𝛽𝛽₃ > 0).

A widely used measure for profitability in the banking industry is the Return on Assets (ROA) ratio (Kocagil et al., 2002). Generally the year average of total assets is used to calculate this ratio. Therefore the independent variable 𝑃𝑃𝑣𝑣𝐿𝐿𝐿𝐿_{𝑖𝑖𝑖𝑖} represents net income divided by the year average of total assets for bank i at time t. The higher its value the better the financial condition of a bank, therefore it is expected to have a negative relation with 𝐶𝐶𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖}(𝛽𝛽₄ < 0).

11_{Pasiouras et al. (2006) define liquid assets as “short-term assets that can be easily converted into cash, such as cash}

itself, deposits with the central bank, treasury bills, other government securities and interbank deposits among others”.

(19)

A measure for efficiency in expenses management is the cost to income ratio used by Pasiouras et al. (2006). 12 The independent variable 𝐶𝐶𝐿𝐿𝐶𝐶𝑡𝑡_{𝑖𝑖𝑖𝑖} therefore represents operating expenses (OPEX) divided by net interest revenue for bank i at time t. A high value of 𝐶𝐶𝐿𝐿𝐶𝐶𝑡𝑡_{𝑖𝑖𝑖𝑖} represents a lower financial condition and therefore a positive relation with 𝐶𝐶𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖} is expected (𝛽𝛽5 > 0).

Size is found to be a possible determinant of default according to Pasiouras et al. (2006) and could therefore be used as a measure for financial condition. Additionally this assumption is supported by Lennox (1999), who states that large companies are less likely to encounter credit constraints and work of Falkenstein et al. (2000), who explain that smaller size implies less diversification and therefore higher sensitivity to idiosyncratic risk. To compare bank size I define 𝑆𝑆𝐿𝐿𝑆𝑆𝐿𝐿_{𝑖𝑖𝑖𝑖} as the logarithm of total assets for bank i at time t. The higher this value the bigger the bank and therefore the stronger its financial condition, hence a negative relation is expected with 𝐶𝐶𝐴𝐴𝑅𝑅_{𝑖𝑖𝑖𝑖}(𝛽𝛽₆ < 0).

3.6 Pearson correlation matrix

The Pearson correlation matrix in Table 3 displays the correlation, and its direction, of the different variables that I explained in the previous section. To measure the overall correlation of the variables I combined the data from the three events. The matrix shows an unexpected positive but insignificant correlation for abnormal returns with liquidity and size. The correlation between abnormal returns and the other variables seem to be corresponding to the hypothesis that financially weaker banks have more benefit from QE. As expected, there is a significant negative correlation between abnormal returns and capital strength. Although not significant, the positive correlation with loan performance and cost to income and the negative correlation with profitability meet expectations as well.

When using the different independent variables as a proxy for financial condition it is expected that liquidity, equity, profitability and size are positively correlated with each other and negatively correlated with loan performance and cost to income. Furthermore, assuming that banks from PIIGS are financially weaker this variable should be negatively correlated with liquidity, equity, profitability and size and positively with loan performance and cost to income. The Pearson correlation matrix shows that both assumptions are true for almost all variables. All variables that should be positively correlated with financial condition are positively correlated

12_{Cost are defined as overheads that are the expenses for running business according to Pasiouras et al. (2006). This}

(20)

with each other. Correlation of liquidity with loan performance, size and PIIGS seems to meet expectations with reasonable significance. This goes also for the correlation of equity with loan performance, profitability and PIIGS and for the correlation of loan performance with profitability, size and PIIGS. Lastly, the PIIGS dummy seems to be significantly correlated in a way that is in line with the assumption that banks from these countries are financially weaker, for all variables except for cost to income ratio.

4 Empirical Results

This section gives a description of the main results. First I discuss the results of the event-study on the effect of the ECB announcements on the EMU EX PIIGS and PIIGS bank portfolios. Second, a brief summary on the abnormal and cumulative abnormal returns of different Eurozone banks on three announcement days is given. Lastly, the results on the implication of bank characteristics on abnormal and cumulative abnormal returns for different Eurozone banks are discussed to examine if the market anticipated that QE would disproportionally benefit financially weaker banks.

4.1 Results of announcement effect on bank portfolios

Using a market model approach I investigate the reaction of a portfolio of Eurozone banks on unconventional monetary policy announcements of the ECB from January 2009 through February 2015. The findings are presented in Table 4, which contains abnormal returns of both bank portfolios on the STOXX 600 and raw returns for all three indices. Looking only at raw returns it should be noted that returns are positive on 9 out of 10 event dates for the EMU EX PIIGS portfolio and 8 out of 10 event dates for the portfolio of PIIGS banks. This corresponds to the reaction of the overall market as the STOXX 600 shows a positive reaction on the same dates.

As would be expected from Joyce et al. (2011) and Kobayashi et al. (2006), the results show that the market anticipated a positive effect of QE for almost all event days that report significant abnormal returns. The introduction of the SMP, ESM and EFSF and the speech of ECB president Mario Draghi both give positive abnormal returns for both portfolios and are significant at least at the 5% confidence level. Furthermore the portfolio of PIIGS reports positive abnormal returns that are significant at the 5% confidence level on the announcement of the targeted LTROs (TLTROs). The EMU EX PIIGS portfolio on the other hand shows a significant and positive abnormal return on the announcement of the new sovereign debt buying program on September 6, 2012. These results give evidence that investors expected these unconventional

(21)

monetary policy decisions to benefit financial markets significantly. Nevertheless, it should be noted that the TLTROs announcements were accompanied by a reduction of the three main interest rates, which went even as far as below zero. Therefore it is possible that the reduction of the three main interest rates creates bias in the regression and that the real reaction on the TLTROs announcements cannot be identified with certainty.

Also notable are the negative abnormal returns that are reported on both portfolios as a result of the announcement of the expanded asset purchase program. On January 22, 2015 the EMU EX PIIGS and PIIGS bank portfolios lagged behind on the STOXX 600 with a raw return of only 1% and 0.2% respectively against a return of 1.6% of the market. Both portfolios even lost around 4% the day after, while the market gain about 1.7%. This comes down on a point estimate cumulative abnormal return of -5.0% on the EMU EX PIIGS portfolio and abnormal returns of -5.8% on the PIIGS portfolio, both significant at the 1% level. In contrast to what is expected from theory of Bowman et al. (2011), Joyce et al. (2011) and Krishnamurthy and Vissing-Jorgensen (2011) among others, these negative abnormal returns suggest that investors anticipated that the expanded asset purchase program would have no or less benefit on banks than it would have on the market. Another explanation could be that the magnitude of the program, which was almost twice as high as expected, gave the impression that the economy found itself in a worse state than what was originally thought.

Lastly, as opposed to the theory of Krishnamurthy and Vissing-Jorgensen (2011) and what Kobayahsi et al. (2006) suggest did the commitment of the ECB, to remain key interest rates at present or lower levels, have no significant effect on bank values in the Eurozone. The announcement on July 4, 2013 gives very insignificant abnormal returns for both portfolios. This suggests that within the Eurozone the announcement probably did not work through a ‘signaling channel’ to ease financial markets.

Overall the PIIGS portfolio showed a stronger reaction, with relatively higher abnormal returns for 9 out of 10 events, which confirms the hypothesis that banks from PIIGS experienced higher abnormal returns. When the abnormal returns that returned strong significant results on both portfolios are compared, the portfolio of PIIGS experienced a stronger reaction for all three events. For the second event on May 10, 2010 the abnormal returns for PIIGS (7.1%) are evidently higher than for EMU EX PIIGS (5.1%). The statement of Mario Draghi in his speech on July 26, 2012 even results in estimated abnormal returns that are more than twice as high for PIIGS (9.1% against 4.4%). Lastly, with the announcement of the expanded asset purchase

(22)

program the negative abnormal returns are relatively higher for PIIGS, with an estimate of -5.8% against -5.0% for EMU EX PIIGS.

4.2 Cross-sectional results on bank characteristics

To investigate cross-sectional differences between 27 Eurozone banks I calculated abnormal and cumulative abnormal returns on individual banks during three event dates that returned significant abnormal returns for both portfolios against the STOXX 600. Using an OLS regression I estimated the implication of bank characteristics on these abnormal returns. The first section describes results on the abnormal and cumulative abnormal returns of all banks. The second section reports results on the regression of these returns on bank characteristics.

4.2.1 Abnormal and cumulative abnormal returns of individual banks

Table 6 gives an overview of abnormal returns of each bank on the first event, and two-day window (0,+1) cumulative abnormal returns of each bank during the other two event dates. Table 6A reports these abnormal returns of 10 banks from the Northern European economies and Table 6B reports the abnormal returns of 17 banks from PIIGS countries.

The results show strong positive abnormal returns on the announcement of the securities markets program on May 10, 2010 for a large group of banks (14 out of 27 are significant at the 1% or 5% level). This positive effect on bank values is probably enhanced by the introduction of the ESM and the EFSF. Together with the abnormal returns of the bank portfolios the results give evidence that the market anticipated a positive impact of these announcements, benefitting financial markets in the process. The strong positive returns suggest that investors were highly confident about this unconventional monetary policy measure and this perception is ultimately supported by the work of Eser and Schwaab (2013) and Giannone et al. (2012).

It looks that banks from France, Italy, Spain and Belgium have had the most benefit from the ECB announcement on the SMP, ESM and EFSF since almost all of them experienced abnormal returns of more than 6%. Assuming that banks from PIIGS countries are financially weaker, the high abnormal returns from Italian and Spanish banks could provide some evidence that weaker banks have had more benefit more from QE announcements. The banks from France and Belgium that also experienced high abnormal returns are BNP Paribas (7.7%), Société Générale (9.8%) and KBC Bank (7.7%). While they are clearly not from the PIIGS region there

(23)

could be other factors that caused these high abnormal returns. Therefore, further investigation on the financial condition of the banks is needed and will be provided in the next section.13

Table 5 shows that the speech of ECB president Mario Draghi, where he stated to do whatever it takes to preserve the euro, had a positive effect on the share price of almost all banks. This manifested into positive cumulative abnormal returns, during the event window of July 26-27, for a large group of banks, which are reported in Table 6A and 6B. Unlike the first event the cumulative abnormal returns are lower in the second event and 4 out of 27 are reported to be significant at the 1% or 5% level. Three Spanish banks, BBVA (10.5%), Banco Santander (11.5%) and Banco Popular (9.7%) experienced strong positive and cumulative abnormal returns that are significant at the 1% level. The results show a significant positive cumulative abnormal return for BNP Paribas (5.7%) as well and to some extent, on a 10% significance level, for Deutsche bank (3.8%), Société Générale (6.7%) and Caixabank (5.6%). Three banks experienced negative cumulative abnormal returns, but not one is found to be significant. With the majority of banks experiencing positive cumulative abnormal returns after the speech of the ECB president, it seems that this event positively affected investors’ expectations, and could therefore have had an easing effect on financial markets.

The last column in Table 6A and 6B reports abnormal returns that the 27 banks experienced after the announcement of the expanded asset purchase program, during the event window January 22-23, 2015. 10 out of 27 banks report abnormal returns that are significant at the 1% or 5% level. As expected from the results on the banking portfolio abnormal returns, almost all banks experienced negative abnormal returns. Only Allied Irish Banks and all Greek banks returned positive abnormal returns, although only Alpha Bank (8.3%) is significant and at a 10% level. Although they are not exceptionally significant they do stand out between the negative abnormal returns for almost all other banks during the event window. Particularly the large positive raw returns on Greek bank equities presented in Table 5 and the resulting positive cumulative abnormal returns presented in Table 6B are remarkable since Greek government bonds were not eligible for the PSPP.14 It is therefore possible that the magnitude of the expanded asset purchase program suggested a worse condition of the Eurozone economy as a whole, which

13_{Cross-holdings between banks could also be a factor, but research on this subject goes beyond the scope of this}

thesis. In future research one could investigate if these banks had significant exposure to PIIGS countries during the announcement.

14

The Public Sector Purchase Program (PSPP) is structured in such a way that the ECB can only purchase 33% of an issuer’s outstanding securities and only 25% of each issue (European Central Bank, 2015). Since the ECB already owns more than this percentage because of earlier programs, such as the SMP, Greek debt will not be eligible until part of it, which is already owned by the ECB, matures.

(24)

led to the negative abnormal returns after the announcement, but that the bad state of the Greek economy was already known, resulting in positive abnormal returns for these banks.

The banks that experienced strong negative cumulative abnormal returns are mainly from PIIGS countries and regard mostly smaller banks.15 Raiffeissen bank (-8.0%) is the only non-PIIGS bank that experienced strong negative cumulative returns, significant at the 5% level. Within the PIIGS area, UBI Banca (-7.9%), Bankia (-7.9%), Caixabank (-8.8%), Banco Popular (-8.9%) and Bank of Ireland (-8.6%) experienced strong negative abnormal returns, all significant at the 1% level.16 These results are consistent with the hypothesis that banks from PIIGS countries experienced higher abnormal returns from QE announcements.

Contrary to what is expected from theory and results from Eser and Schwaab (2013), Joyce et al. (2011), Kimura and Small (2006) and Krishnamurty and Vissing-Jorgensen (2011) among others did the expanded asset purchase program not lead to positive abnormal returns on Eurozone bank values. As mentioned earlier it is possible that a QE announcement has led to negative abnormal returns if for example the content of the unconventional monetary policy goes against investors’ expectations. It is therefore possible that investors were more or less skeptical about the impact of the expanded asset purchasing program on Eurozone banks on the on hand or that the size of the program suggested that the state of the economy turned out to be even worse than investors expected on the other hand.

4.2.2 Abnormal returns and bank characteristics

In this section I present results from the regression on the implication of bank characteristics on abnormal returns for 27 Eurozone banks. Table 7 gives an overview of the results from an OLS regression on each of the three main events. As mentioned in the previous section I use abnormal returns on May 10 as the dependent variable, where I use cumulative abnormal returns for the other two events. All three events use bank characteristics as the independent variables. First I report the results only using the six factors that are derived from the CAMELS rating system to measure the financial condition of banks. The dummy variable that specifies whether a bank is located in one of the PIIGS countries is added in the second column of each event.

It appears that the first two events return significant results on some of the variables. Although they are not all significant the estimators for both events show the same sign on each estimator except on the effect of liquidity and cost efficiency. The positive sign of the estimator

15

Size based on average of total assets between 2013 and 2014.

16_{News about Spanish macro-economic factors such as unemployment rate was relatively positive around this event}

date. Also no other news about Bankia, Caixabank or Banco Popular can explain the large negative abnormal returns for these Spanish banks.

(25)

on liquidity seems to be similar to the results of Kobayashi et al. (2006), although in both studies this was not expected.

For the May 10 and July 26-27 event windows, 2 out of 4 variables that report an estimator that is significant at least at the 5% confidence level are in line with the expectation of a negative relation between financial condition and abnormal return. The variable that measures capital strength is significant at the 1% level for the second event, suggesting that the market expected that banks, which are less capital constrained, would benefit less from QE. The variable for loan provision shows an expected positive relation on both events, which is significant at the 5% level for the event on May 10, 2010. This positive relation suggests that banks with a higher ratio of loan loss provision to net interest revenue are expected to gain more from QE, implying that the market anticipated that the program would disproportionately benefit banks that are more financially constrained. Kobayashi et al. (2006) to some extent found similar results on their measure of bad loan ratio, which contributed to the same conclusion that financially weaker banks had more benefit from QE.

The two variables for profitability and size both show an unexpected result for almost all events. The estimator on the variable profitability that measures the implication of ROA on abnormal returns seems to suggest that more profitable banks have earned higher abnormal returns during the three announcements. For the profitability variable the results on the May 10 event date appear to be significant at the 5% level, but loses its significance strength in the other two events. Nonetheless, when using profitability as a measure for financial strength, the results indicate that financially stronger banks were expected to benefit more from QE announcements. The same unexpected result is returned by the variable size, which shows a positive relation on two announcement dates that is significant at least at the 5% confidence level. As with the characteristic of ROA, the CAMEL rating system considers bigger size to be related with a better financial condition (Pasiouras et al., 2006). The positive relation between the variable Size and abnormal returns thus suggests that the market expected bigger and therefore “stronger” banks to benefit more from the ECB’s unconventional monetary policy. Work of Baele et al. (2007) and Castrén et al. (2006) however, suggests that an explanation for this positive relation could be that bigger banks are more diversified, but at the cost of an increase in systemic risk. They state that diversification increases the systemic risk making diversified banks more prone to market-wide news. This could explain why results on the Size variable are contrary to the hypothesis.

(26)

The output on the variable Cost shows contradicting results on the different event dates, however no estimator is significant. They seem not to reveal a clear relation between a banks’ management in cost efficiency and the magnitude of the abnormal returns it experienced.

Lastly, the inclusion of the dummy variable PIIGS does not appear to have a qualitative effect on the results. However, for the July 26-27 event window the dummy variable reports a positive relation, which is significant at the 10% level, suggesting that banks from PIIGS have experienced higher abnormal returns after the speech of the ECB president. Furthermore, adding the dummy seems to have almost no impact on the sign of the regression estimators of the other variables and only changes the significance strength for some variables to some extent. The variable Size loses its significance at the 5% level, although it remains significant at the 10% level for the May 10 event date. The estimator on the capital strength variable becomes less significant on both the first and second event dates.

Overall it looks that capital strength, loan provision, profitability and size are good estimators for the abnormal returns that banks experienced around QE announcements of the ECB. Although size returns the most consistent estimator its unexpected positive relation with abnormal returns could also be related to an increase in systemic risk instead of financial condition. With capital strength and loan provision being more related to financial condition, as is emphasized by the Basel III Accord of the Basel Committee (2011), it seems that financially weaker banks are expected to disproportionately benefit more from QE announcements.

5 Additional results and robustness checks

To examine the robustness of the results described in the previous section I run some additional test, which I present in this section. First, I control for noise in the estimation window of the market models that are used to calculate abnormal returns for the EMU EX PIIGS and PIIGS portfolios. Second, I verify robustness on the cross-sectional results by using a three-day event-window (-1,+1) to calculate cumulative abnormal returns. Finally, I run both cross-sectional models for the January 22 event, in which all four Greek banks are excluded from the sample.

5.1 Controlling for noise in the estimation windows

With the calculation of the abnormal returns of the two banking portfolios there is a possibility of noise in the estimation windows that are used for estimating the beta of the market model. Although I used a short estimation window of 100 trading days, some of the events take place during one of the estimation windows for another event, which could lead to a shift in the market

(27)

beta after a QE announcement took place. Four out of the ten events that I selected occur during an estimation period of another event. Although I do not expect that this has a significant effect on the results, an additional regression with a control dummy is run for each of the overlapping estimation periods as a robustness check. To control for a shift in beta a control dummy is added to the standard market model resulting in the following equation:

𝑅𝑅𝑖𝑖𝑖𝑖 = 𝛼𝛼𝑖𝑖 + 𝛽𝛽𝑖𝑖𝑅𝑅𝑚𝑚𝑖𝑖+ 𝛽𝛽𝑒𝑒(𝑅𝑅𝑚𝑚𝑖𝑖× 𝐷𝐷𝑒𝑒𝑖𝑖)

+𝜀𝜀𝑖𝑖𝑖𝑖 (8)

with 𝐷𝐷_{𝑒𝑒𝑖𝑖} being a dummy variable that takes on value 0 prior to the date of event e and 1 afterwards, and 𝛽𝛽_𝑒𝑒 representing the coefficient on the beta shift for event date e.

The abnormal returns are presented in Table 8. As expected, adding the dummy to control for a shift in the market beta does not have a substantial effect on the results. The dummy is significant at the 1% level for the December 8, 2011 and adding it leads to a change in sign for the cumulative abnormal return of the EMU EX PIIGS portfolio, although it stays insignificant. For the September 6, 2012 event adding the dummy to the market model causes abnormal returns to decrease for both portfolios but they remain positive. For the other two events the dummy failed to become significant, suggesting that the market beta did not shift. The impact it has on the results seems negligible and confirms that the events that fell in one of the estimation periods did not have a significant impact on the interpretation of the main results.

5.2 Three-day event-window interval

In his work on event study methodology Mackinlay (1997) recommends a (-1,+1) interval as most accurate since it allows for spillover effects in days surrounding the event, but does not weaken the power of the regression. Therefore, as a robustness check I run an additional regression of three-day interval cumulative abnormal returns on bank characteristics for the three events.

Table 9 gives an overview of the regression with three-day event-window cumulative abnormal returns. The interval contains abnormal returns of one day before and one day after each event. It seems that using a longer event-window only has a minor effect on the outcome of the regression of bank characteristics on abnormal returns. The significance of the four variables that measure capital strength, loan provision ratio, profitability and size turns out to become even stronger for the May 10 event, while the measures for liquidity and cost to income ratio stay rather insignificant. This is different for the July 26 event where Profitability becomes significant at the 5% level when not controlling for bank location, while the estimator for Size fails to be significant for this event. The capital strength variable Equity that is measured by Tier 1 capital

Quantitative easing and the Eurozone crisis : the effect of the European Central Bank’s QE announcements on financial institutions in the Eurozone

University of Amsterdam