The influence of financial structure on the effectiveness of quantitative easing

(1)

THE INFLUENCE OF FINANCIAL STRUCTURE ON THE

EFFECTIVENESS OF QUANTITATIVE EASING

Author: Jacob Bakx, 10517138 Supervisor: dhr. prof. dr. A.C.F.J. Houben

Master thesis Monetary Policy & Banking, 15 July 2018 Faculty of Economics and Business, University of Amsterdam

(2)

ABSTRACT

The “Great Recession” has reignited the discussion on the comparative advantages of market-based versus bank-based financial systems. Besides that, major central banks, such as the European Central Bank, resorted to the implementation of quantitative easing. Until now, no research has investigated the influence of financial structure on the ECB’s quantitative easing policy. In light of this unaddressed relationship, this thesis examines to what extent the financial structure of the euro area affected the transmission of quantitative easing. Two fixed effects regression models are estimated over a panel of nineteen euro-area countries. The findings indicate that a bank-based financial structure constrains the transmission of QE to the financial and real economy.

STATEMENT OF ORIGINALITY

This document is written by Jacob Bakx, 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 no sources other than those mentioned in the text and those in 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

Low inflation, high unemployment and low GDP growth led the European Central Bank (ECB) to implement unconventional monetary policy measures, notably including quantitative easing (QE). QE, initiated by the ECB in March 2015, is a policy tool central banks can resort to when the policy interest rate cannot be lowered because the effective lower bound (ELB) has been hit. QE involves a significant expansion of a central bank’s balance sheet and aims to stimulate economic activity by pushing up asset prices, lowering bank and market (long-term) rates and weakening the exchange rate. By affecting asset prices and returns, policymakers try to adjust economic behaviour in ways that will help them to reach their ultimate objective. QE works through several channels, most notably, the portfolio rebalancing channel, the signalling channel, the liquidity channel, the credit channel and the exchange rate channel. These channels impact the primary objective of the ECB, which is to maintain price stability over the medium term, in different ways (Hallet, 2017). It is important to distinguish the different transmission channels to find out how QE affects the real

economy.

To assess the macroeconomic impact of QE, this thesis investigates the effect of QE on two distinctive indicators. The first is an economic indicator, defined as the ten-year interest rate on government bonds. The second is a monetary indicator, defined as total credit lending to the non-financial sector. With QE the ECB aims to push down borrowing rates, thereby encouraging households and businesses to reduce saving and increase borrowing. QE also stimulates financial intermediaries to take on more risk and reduce the cost of borrowing. This results in lower long-term interest rates and more rapid credit growth, thereby

stimulating economic growth and wealth accumulation. So far there is not much research published on the effects of QE by the ECB. This is because the ECB only embarked on QE. While there is initial evidence of a rise in economic activity due to the sharp increase in the ECB’s balance sheet (Burriel & Galesi, 2018; Gambacorta, Hofman & Peersman, 2014; Hohberger et al., 2017), the effects of QE by other major central banks1_{are more clearly}

visible (Gagnon, 2016).

(5)

By using unconventional monetary policy, like QE, a central bank aims to affect financing conditions. For this reason, the financial structure of a country must be taken into consideration, and specifically how the flow of funds is structured (Bini Smaghi, 2009). The financial structure of a country primarily depends on the mix of two stylized forms of financing.

Firstly, bank-based financing; banks generally carry out intermediation on their balance sheets. They take savings as deposits and mainly supply funding in the form of loans through close relationships (Gambacorta, Yang & Tsatsaronis, 2014). By implementing QE, the ECB improves the liquidity of banks and stimulates banks to translate this extra liquidity into new loans. Banks are therefore an important channel for the transmission of monetary policy. Especially, in bank-based financial systems, the effects of QE greatly rely on the banking sector. But given the weak state of European banks in combination with the impact of tighter prudential requirements, the effect of this channel may be moderate (Hausken & Ncube, 2013).

The past thirty years the European banking system relative to GDP has grown considerably. However, there are still significant differences in the size and composition of euro-area countries’ financial systems (Langfield & Pagano, 2016). In turn, these cause differences in the transmission of monetary policy across the eurozone (Cecchetti, 1999). Langfield and Pagano (2016) also show that different financial structures have a diverse influence on economic growth. Therefore, the effects of the ECB’s QE programme are

expected to vary from country to country. This divergence provides an opportunity to identify the links between financial structure and quantitative easing.

The second form of financing is market-based financing. Markets serve as a platform where equity and debt securities are issued and traded (Gambacorta et al., 2014). By changing values of bonds and equities, monetary policy can be transmitted through markets. QE

directly pushes prices of purchased assets up and puts downward pressure on yields. As a result, investors rebalance their portfolios in search of higher returns, indirectly increasing credit demand and prices of other assets (Demertzis & Wolff, 2016). Eventually, this stimulates aggregate demand and increases inflation. Furthermore, when capital markets develop, borrowers find it easier to issue bonds and to attract alternative financing from non-bank sources. In case the amount of non-non-bank finance compared to non-bank finance increases, the relative importance of monetary policy transmission via bank lending declines and that

through capital markets increases (IMF, 2016). However, after the financial crisis, there was both a limited availability of bank credit as well as a lack of alternative financing possibilities

(6)

in the European bank-based financial systems. This might be the reason of the gradual transmission of QE and the subsequent slow economic recovery.

How financial structures and quantitative easing affect economic growth is much debated. According to the credit channel view, banks are especially important for the transmission of monetary policy. However, based on recent empirical evidence, economists seem to favour market-based financial systems over bank-based financial systems in terms of economic growth and systemic risk (Bats & Houben, 2017; Gambacorta et al., 2014;

Langfield & Pagano, 2016). Concisely, whether banks dampened or amplified the

transmission of quantitative easing is not determined. Though, knowing this is important for the future conduct of monetary policy.

In this context, this thesis empirically studies the effects of quantitative easing from the viewpoint of financial structure. The research question is: what is the influence of financial structure on the effectiveness of quantitative easing? In particular, how does the structure of a financial system influence the effect of quantitative easing on long-term interest rates and total credit lending?

Two fixed effects regression models are estimated over a panel of nineteen euro-area countries. The models use five distinctive quantitative easing indicators. Every QE indicator is used as independent variable in separate regressions. Additionally, the models include an independent variable, which represents financial structure, in interaction with a QE indicator. The results provide strong evidence for the fact that the more bank-based a financial system is, the less powerful QE is in reducing long-term interest rates and increasing total credit lending.

The remainder of this paper is organised as follows. Section 2 reviews the different concepts of quantitative easing. Section 3 introduces the two archetypes for financial systems. Section 4 describes the link between quantitative easing and financial systems. The

methodology is set out in section 5 and section 6 describes the data. Thereafter, section 7 presents and analyses the results that are discussed in section 8. Section 9 concludes.

(7)

2. QUANTITATIVE EASING

The Bank of Japan, the Federal Reserve and the Bank of England all conducted non-standard monetary policy measures before the ECB. For this reason, more research on the effectiveness of QE by these central banks has been published. Consequently, after reviewing the quantitative easing programme implemented by the ECB, it is also useful to describe QE programmes executed by other major central banks.

This section starts with a general definition of QE followed by a comprehensive explanation of QE by the ECB and its transmissions channels. Thereafter, a short overview of QE by the BoJ, Fed and BoE, respectively, is given. To conclude, a summary of empirical evidence on the effects of the unconventional monetary policies is presented.

2.1 Definition

To begin with, it is important to make a distinction between conventional and unconventional monetary policy. Normally, a central bank conducts monetary policy by setting a goal for the short-term nominal interest rate and adjusting the supply of central bank money through open market operations. By doing so, the central bank controls liquidity conditions in the money market and this ultimately impacts the level of inflation. The reaction function of central banks has been summarized formally in the Taylor rule, which requires nominal interest rates to go up if the projected inflation exceeds the central bank’s inflation target (two percent for the ECB, Fed, BoE and BoJ) and if GDP rises above potential (Williamson, 2017). By contrast, this rule requires interest rates to go down if expected inflation is below target and if GDP falls below potential. But, in case the nominal interest rate hits the effective lower bound, central banks can no longer resort to conventional

monetary policy measures to steer inflation and economic growth. The related problem is that, when the nominal interest rate is close to or below zero, the real interest rate may be higher than what is necessary to ensure balanced prices and full utilization of resources (Bernanke et al., 2004). In times of stable inflation with low unemployment and (moderate) GDP growth this poses no immediate threat to financial stability. However, in the macroeconomic environment of the last ten (for Japan twenty) years, particularly when deflation risks have emerged, the use of conventional monetary policy to achieve a central bank’s objective has shown to be insufficient. Therefore, central banks have sought other ways to stimulate

(8)

economic activity. Accordingly, central banks implemented unconventional monetary policy measures like QE.

Bernanke and Reinhart (2004) discuss three forms of unconventional monetary policy central banks can resort to when the effective lower bound is hit. In addition, the ECB

introduced a fourth type of unconventional monetary policy. The first form is forward

guidance, using communication as a tool to form public expectations about the future path of interest rates. In doing so, a central bank aims to reduce long-term interest rates and interest rate volatility, and ultimately reduce risk premia. The second is adjusting the composition of the central bank’s balance sheet. Central banks generally hold a variety of assets in terms of risk, liquidity and yields. By shifting its composition, suppose, from low-yield securities to high-yield securities, it changes the relative demand of these securities. As a result, it affects overall yields and term premia. In turn, this stimulates economic activity. The third manner of unconventional monetary policy is increasing the size of the balance sheet by buying large-scale securities, commonly known as quantitative easing. The fourth is negative policy rates introduced by the ECB. In 2014, the ECB was the first major central bank that lowered its deposit facility rate to below zero. Thereafter, the assumption of a “zero lower bound” nuanced to the “effective lower bound”. Negative interest rates stimulate the economy through the regular channels. They lower real yields and consequently encourage borrowing and economic activity. This thesis investigates the effect of the third manner, namely, quantitative easing.

The Bank of Japan was the first to use the term quantitative easing in 2001 when they had to combat deflation risks. Generally, QE programmes are defined as large-scale security purchase programmes with the aim of lowering long-term yields and increasing prices of assets and bonds. They involve an ample expansion of a central banks’ balance sheet accompanied by an increase in the supply of money. Ultimately, these policies intend to stimulate economic activity and foster price stability. Because the way QE is implemented differs across central banks, this is explained in greater detail in later paragraphs of this section. However, the focus lies on the Expanded Asset Purchasing Programme launched by the ECB since this is the subject matter.

(9)

2.2 QE by the European Central Bank

This paragraph discusses the non-standard monetary policy measures implemented by the ECB and the channels through which they have effect on financial markets and the wider economy.

The years prior to the crisis are known as the years of “Great Moderation”. A period of robust productivity growth and low and stable inflation. On top of that, policy interest rates were unusually low while financial assets increased in value. This led to the underestimation of risk by banks and investors. All these developments contributed to the rapid expansion of credit and caused heavily indebted governments, firms and households. Developments that eventually had catastrophic consequences when the financial crisis hit in 2008. The aftermath of the “Great Recession” left debtors overleveraged and unable to pay their debt. At the same time, creditors had to deleverage. This troubled economic situation with already low (real) interest rates asked for an unconventional response by the ECB (Praet, 2017). Most notably, the provision of liquidity on demand at fixed-rate full allotment2, the use of negative interest rates, the use of forward guidance and quantitative easing measures by implementing the (Expanded) Asset Purchase Programme (EAPP).

The EAPP includes all purchase programmes of public and private sector securities to avoid risks of deflation. Under this programme, the ECB has been buying sovereign bonds from euro-area governments and securities from European institutions and national agencies (Claeys, Leandro & Mandra, 2015). The ECB has already purchased public and private assets worth more than two trillion euros. At the peak of its programme3_{monthly purchases were}

conducted on an average pace of 80 billion euros. Currently, the monthly average purchases amount to 30 billion euros. However, on 14 June 2018, the Governing Council stated that it

“anticipates that, after September 2018, subject to incoming data confirming the Governing Council’s medium-term inflation outlook, the monthly pace of the net asset purchases will be reduced to €15 billion until the end of December 2018 and that net purchases will then end”.

Taking into consideration that real GDP grew moderately, HICP inflation estimates increased to 1.9 percent, unemployment reduced, and the growth of loans is proceeding, the Governing Council is convinced that, even without its monthly net asset purchases, convergence of

2_{Under the fixed-rate full allotment policy, banks can get all the liquidity they demand from the ECB, on the} condition that they provide adequate collateral and are financially stable.

3_{€60 billion March 2015 – March 2016, €80 billion April 2016 – March 2017 and €60 billion April 2017 –} December 2017. 80 percent of the securities are purchased by NCBs (Claeys et al., 2015).

(10)

inflation to levels below, but close to, 2 percent in the medium term, will continue. To which extent monetary policy, particularly QE, contributed to these developments is empirically hard to disentangle but there are a variety of theoretical models that connect non-standard policy measures with economic growth and the level of inflation (Haldane et al., 2016). These models are built as structural frameworks presenting several channels through which QE can potentially have effect on financial markets and economic activity. It is essential to

understand these channels to evaluate whether a given policy was successful (Krishnamurthy & Vissing-Jorgensen, 2011). The following section discusses these channels.

2.3 Transmission channels

The first channel is the portfolio-rebalancing channel. The large-scale asset purchases of the ECB reduce the relative supply of the assets being purchased and at the same time raise the amount of base money. Consequently, prices of the purchased assets will increase, and the yields will go down. Since lower yields decrease the return on assets, investors are triggered to buy alternative assets in search of higher returns. This increases demand for a variety of assets. Thus, QE not only depresses the yields of the assets being purchased but also lowers interest rates and the associated term premia on other securities (Fiedler et al., 2017). Lower interest rates lead to lower borrowing costs, less saving and more investment.

The second channel in question is the signalling channel. Similar to conventional monetary policies, the communication of non-standard policies is an essential part of their transmission mechanism (forward guidance). Communications about future policies can only have a beneficial effect in lowering long-term yields if such a policy serves as a credible commitment to keep interest rates low. By announcing large-scale asset purchases the central bank can achieve such a credible commitment. Consequently, the central bank can affect public expectations about the likely path of future policies. Market participants may interpret the ECB’s willingness to purchase long-term assets as a signal that it will hold its policy rates low for an extended period. Moreover, the announcement that the central bank will engage in long-term asset purchase programmes, boosts investor confidence in those assets, thereby depressing liquidity premia. In conclusion, the signalling channel affects all bond market interest rates, with quantitative effects depending on bond maturity (Krishnamurthy & Vissing-Jorgensen, 2011).

(11)

Third, QE increases the liquidity in the hands of investors and enhances market functioning. Because the ECB acts as a reliable and substantial buyer of long-term assets, it triggers investors to take bigger positions in these securities, knowing that they will be able to sell them to the ECB when necessary. This results in improved trading opportunities, reduced liquidity risk premia and increases assets prices. Although QE has restored market liquidity through this liquidity premium channel, the effects may only last while the programme is still in progress (Joyce et al., 2011).

The fourth channel is the exchange rate channel. When QE lowers domestic4 interest rates relative to foreign interest rates, it induces a capital outflow and a depreciation of the domestic exchange rate. As a result, goods in the QE economy compared to foreign goods become cheaper and demand for domestic goods rises. Consequently, imports from trade partners, whose currency relative to the domestic currency has been appreciating, decrease and exports grow. Higher exports increase aggregate demand and stimulate economic growth.

The fifth channel is the credit channel. This channel can be broken down into two channels: the bank lending channel and the balance sheet channel. The bank lending channel works as follows: by buying long-term government bonds from banks, the ECB improves the liquidity of the banking sector and reduces term premia and, accordingly, it encourages banks to extend more credit at lower costs. If bank rates fall, bank-dependent borrowers will be able to acquire cheaper additional lending, which is likely to boost credit demand and,

consequently, economic activity. According to the credit channel, bank lending is essential for the transmission of monetary policy to the economy (Wang, 2016).

The balance sheet channel is based on the idea that the accessibility of external finance depends on the financial circumstances of borrowers (Bernanke & Gertler, 1995). By

increasing the prices of assets and reducing (long-term) yields, quantitative easing alleviates the borrowers’ burden of debt and improves the value of their collateral. This affects their investment and spending decisions. Overall, better finance conditions cause aggregate demand and economic growth to increase.

To summarize, through these five transmission channels, QE can boost consumer confidence, reduce market interest rates and push up assets prices. This increases the wealth of investors and reduces the cost of borrowing, thereby stimulating households and

corporations to increase borrowing and investments. Eventually, the ECB tries to boost economic growth and reduce unemployment to reach its primary objective: an inflation of

(12)

Central Bank Quantitative Easing

Inflation Target & Real Economic Growth

levels below, but close to, 2 percent in the medium term. Figure 1 presents a summary of the transmission channels. Nevertheless, the relative importance of these main channels is undetermined (Krishnamurthy, Nagel & Vissing-Jorgensen, 2017).

Figure 1: Transmission channels for QE

Confidence

Bank Lending & Balance Sheet Channel Intermediate targets Ultimate objectives

(13)

2.4 QE around the world

In this paragraph short overviews of quantitative easing measures implemented by the Bank of Japan, the U.S. Federal Reserve and the Bank of England are given. Figure 2 shows time-plots for total assets as a percentage of GDP for the ECB, BoJ, Fed and BoE.

2.4.1 Japan

Japan’s first QE programme began in March 2001 and ended in 2006. As a result, the economy recovered but the programme failed to get rid of the persistent deflation risks. In response to the financial crisis of 2008, the BoJ announced the implementation of new, relatively small programmes followed by much larger easing programmes from 2013 onwards. These programs mostly consisted of monthly purchases of Japanese Government Bonds. Nowadays, the BoJ still applies QE, it has even purchased local stocks, because though the economy is expanding steadily, price developments remain weak. The governor of the Bank of Japan, Kuroda (2017), states that there is still a long way to go to achieve the price stability target of two percent.

2.4.2 The United States of America

In 2008 the Federal Reserve launched its first quantitative easing programme (QE1), with the hope to recover from the “Great Recession”. They announced the purchase of large quantities of agency debt and mortgage-backed securities to provide support to the mortgage and housing market. The Fed continued with a second programme (QE2) in 2011 to further spur economic activity and reach their inflation target of two percent. Under this programme, they purchased around $600 billion in long-maturity U.S. Treasuries. Finally, the Federal Open Market Committee announced the third round of QE in 2012 (QE3) which it terminated in October 2014 based on the following expectation: “with appropriate policy

accommodation, economic activity will expand at a moderate pace, with labour market indicators and inflation moving toward levels consistent with the Fed’s dual mandate”.

(14)

2.4.3 The United Kingdom

Beginning in 2009, the Bank of England’s Monetary Policy Committee proclaimed that it would begin a programme of large-scale purchases of public and private assets. It ended its first programme in 2010 but growing concerns about the macroeconomic

circumstances led the BoE to resume its sizeable asset purchases in October 2011 (Haldane et al., 2016). The current value of assets bought is around 435 billion pound and mostly contains of UK government bonds (gilts).

Figure 2: central banks’ total assets (stock) to GDP

Note: this figure presents time-plots for the ratio total assets (stock) to GDP for the Bank of Japan (BoJ), the European Central Bank (ECB), the Bank of England (BoE) and the US Federal Reserve (Fed), since the beginning of the “Great Recession”. The vertical line on the left depicts the moment the Fed launched its first QE programme, the first line right of this line points to the moment the BoE launched its first QE programme and the third line from the left depicts the moment the BoJ began its biggest QE programme since the onset of the crisis. Finally, the last (right) line depicts the moment the ECB launched its PSPP.

0 10 20 30 40 50 60 70 80 90 100 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 BoJ ECB BoE Fed

(15)

2.5 Summary of empirical evidence

In the empirical literature there are two main approaches for evaluating the effects of unconventional monetary policies: vector autoregressive regressions (VAR) and event studies. The VAR approach mostly focuses on the effectiveness of balance sheet measures on

economic activity. In contrast, event studies generally look at the effectiveness of policy announcements (forward guidance) and balance sheet measures on financial markets. A large part of earlier research investigated the effects of QE programmes on economic activity and financial markets in the United States. Williams (2014) presents an overview of fourteen empirical studies that did research into the effects of the Fed’s Large-Scale Assets Program (LSAP). From these studies two conclusions can be drawn. First, all analyses find that assets purchases have substantial effects on long-term yields. With estimations ranging around 90 to 150 basis points drop in ten-year Treasury notes, following the three rounds of QE. Second, there still is a large amount of uncertainty about the magnitude and overall effects on the general economy. For the euro area, there are several studies that qualitatively support the findings for United Kingdom, Japan and the United States. Because the euro area consists of nineteen different member states, several studies highlight the differences in impact on government bond yields across countries.

Fiedler et al. (2016) summarize findings of studies that investigated the effects of the ECB’s unconventional monetary policies on financial markets and on economic activity. They state that unconventional monetary policies have a considerable effect on a wide array of financial market variables and that the Euro system can improve economic growth by increasing its balance sheet or monetary base (Peersman, 2011). Most studies evaluate the effects of programme announcements on sovereign bond yields. However, Gibson et al. (2016) establish that the implementation of unconventional monetary policies measures also causes sovereign yields to fall. In general, policy impacts appear to be stronger in the GIIPS5 countries relative to the core countries; Austria, Finland, Germany and the Netherlands.

Additionally, the effects of UMP on economic variables have proven to be significant. To elaborate on this, various studies show that QE programmes positively affect inflation and output, thereby stabilizing financial markets and the economy as a whole. Though, the size of these effects differs per study. Furthermore, it is empirically hard to disentangle which of the ECB’s unconventional monetary policy programmes had the largest macroeconomic impact.

(16)

Nevertheless, it is commonly assumed that QE programmes are the most effective in enhancing general conditions of the financial market (Fiedler et al., 2016).

Recent researchers (Albertazzi et al., 2018; Burriel & Galesi, 2018; Duindam et al., 2018), are especially concerned with the cross-country differences in the effects of UMP. Results show that these heterogeneities are caused by differences in the health of the banking system, the degree of financial market stress, macroeconomic imbalances and risk premia. To be more specific, findings show that the economically healthier core countries benefitted the most in terms of output and inflation, whereas the more vulnerable southern countries mostly gained from reductions in government bond yields.

To conclude, regarding the effectiveness of balance sheet measures, there are three things researches have not agreed upon: the actual size of the effects, which programme has the largest macroeconomic impact and which countries benefit the most. Nevertheless, there is a worldwide consensus that expansive balance sheet measures cause long-term interest rates to fall and asset prices to rise. Moreover, research shows that the implemented programmes, particularly QE programmes, are effective in stimulating credit lending, output and inflation. Ultimately, enhancing price stability and financial stability.

Since improving financing conditions is one of the goals of quantitative easing, it is important to explain by what means financing can be conducted and what its role is in the transmission of monetary policy into the real economy. The next two sections clarify these matters.

(17)

3. FINANCIAL SYSTEM ARCHETYPES

Financial systems are crucial for the allocation of resources, for pricing and managing risks and for absorbing shocks. Most developed financial systems fulfil these functions. Nonetheless, when assessing the financial systems of various countries, there are large

differences in the structure of these systems (Allen & Gale, 2000). The financial structure of a country depends on a mixture of financial instruments, markets and intermediaries. Based on this mixture, a broad distinction can be made between two archetypes: bank-based or market-based financial systems.

In a bank-based financial system, financial intermediation is mainly conducted by monetary financial institutions. By channelling funds from savers to borrowers, banks turn liquid liabilities into illiquid long-term loans. Banks bear risks and therefore mostly perform intermediation through close relationships. To avert problems of information asymmetries and contract enforcements, banks use the knowledge they obtain from close relationships to

‘screen’ and ‘monitor’ (Pagano et al., 2014). In contrast, markets avoid close relationships

with investors and savers, by operating as a platform where debt and equity securities are priced and traded. Also, markets bear fewer risks as they primarily channel funds from

lenders to borrowers. To overcome problems caused by information asymmetries and contract enforcement, markets protect borrowers and lenders through legal covenants and the courts (Gambacorta et al., 2014).

The question whether one financial structure outperforms the other in terms of

economic growth is much debated. Because a direct measure of the services that markets and banks make available does not exist, empirical studies rely on indicators that represent the two financial structures (Bats & Houben, 2017; Gambacorta et al., 2014; Langfield & Pagano, 2016). Bats & Houben (2017) highlight the contrasting stance taken in the literature published before and after the great financial crisis of 2008. Before 2008 both financial structures were found to be of comparable importance for economic growth. By contrast, since the financial crisis of 2008, researchers have preferred market-based financial systems to bank-based financial systems. This is for two reasons. One, crises have been found to be economically more harmful in bank-based financial systems (Gambacorta et al., 2014; Langfield & Pagano, 2016), and two, bank-based financing generates systemic risk whereas market-based

financing reduces system risk (Bats & Houben, 2017; Langfield & Pagano, 2016). However, examining the effects of financial structure on the effectiveness of QE is an unexamined topic, therefore it is important to set out the views on the two opposing structures.

(18)

Proponents of the bank-based view argue that banks positively affect economic

growth for several reasons (Levine, 2002). First, by gathering data and processing information (screening) of managers and firms, banks improve capital allocation and control corporate managers. Second, by building long-term relationships banks can reduce transaction and informative expenditures, thereby allowing new companies to be established and existing ones to grow. Banks are particularly important for SMEs6 and capital-intensive enterprises, as they generally lack alternative financing sources and have less collateral (Klein, 2014). In addition, as relationship lenders, banks can alleviate adverse selection and credit rationing (Stiglitz & Weiss, 1981). Lastly, banks prevent misallocation of capital by providing an opportunity to deposit reserves. Consequently, they reduce the fraction of savings held in unproductive liquid assets. As a result, investment efficiency increases, and market liquidity is well managed (Bencivenga & Smith, 1991). Proponents of the bank-based view point to the importance of managing market liquidity. Because when markets become too liquid, investors will be able to sell their shares quickly and cheaply, which reduces the incentive to apply disciplined corporate governance (Levine, 2002).

In contrast, advocates of the market-based view emphasize the growth-enhancing role of well-functioning markets. They claim that markets are much more efficient when the problems and costs of acquiring information are less severe, which is likely the case in developed countries. Furthermore, the transmission of information through markets is more efficient and timely, and information is easily accessible, which augments firm financing and economic growth (Boot & Thakor, 1997). Also, because capital markets become more important the closer an economy is to its technological frontier, they are more advantageous for technologically innovative enterprises (Pagano et al., 2014). Additionally, banks are controlled by regulatory requirements, which restraints them. Finally, markets incite corporate control by smoothening takeovers and by facilitating companies to link managerial

reimbursements to firm performance (Jensen & Murphy, 1990).

Besides the bank and market-based view there is an additional view, namely, the financial services view, or as Bodie and Merton (1995) state it: the functional perspective of financial systems. Supporters of this view argue that it is not a matter of banks or markets, as long as there are markets and intermediaries that offer reliable financial services. Thus, the soundness of the financial system relies on how well its core functions are performed. Which are: i) facilitate trade, ii) transfer resources through time, across borders, and among

(19)

industries, iii) manage risk, iv) provide price information, v) reduce information asymmetries and vi) provide a mechanism for pooling resources and dividing shares (Bodie & Merton, 1995).

To summarize, there are many arguments favouring market-based or bank-based financing. Proponents of the bank-based view stress that banks improve corporate control, reduce adverse selection and prevent misallocation of resources. By contrast, the market-based view holds that markets increase efficiency by providing easily accessible information and liquidity. Moreover, markets are likely to be better risk-bearing financiers of innovation.

Having presented the theoretical framework of quantitative easing and financial structure, this thesis proceeds with explaining the link between the two concepts.

(20)

4. THE LINK BETWEEN QUANTITATIVE EASING AND FINANCIAL

SYSTEM ARCHETYPES

After the financial crisis, major central banks used unconventional monetary policy measures as new instruments to stabilize the financial system and promote price stability. How these measures were implemented and transmitted to the real economy largely depended on the financial structure. While all central banks have in common that they injected a large amount of liquidity in the financial system, the way they did this differed. In line with the market-based financial structure, the Fed’s operations consisted mostly of purchases and sales in the open financial market. These operations were designed to affect the prices and yields on bonds that were issued to finance lending to corporations and households (Joyce et al., 2011). By contrast, the BoJ supplied a large amount of funds directly to commercial banks, in line with the Japanese bank-based financial system. Banks were expected to extend new credit with the additional liquidity provided by the BoJ. However, much of the supplied funds remained on their balance sheet as reserves (Wang, 2016).

The ECB’s initial non-standard response to the crisis had a similar approach to the BoJ. As banks play a predominant role in the euro area, the ECB adopted several

non-standard measures directed at alleviating refinancing concerns of the euro-area banking sector and providing additional liquidity to banks. Namely, the fixed-rate full allotment policy and longer-term refinancing operations7 (Cour-Thimann & Winkler, 2012).

In a bank-based financial system, non-financial corporations, particularly small and medium – sized firms, find it difficult to substitute bank lending for market financing. Therefore, a reduction in bank credit in Europe, where employment and economic activity is mostly created by SMEs, could be increasingly harmful (Cour-Thimann & Winkler, 2012). However, the initial measures proved to be insufficient when the euro area again came under strain during the sovereign debt crisis. Consequently, additional purchase programmes were implemented. These programmes can affect economic activity and financial markets through several channels. Markets and banks act as intermediaries within these channels. In particular, markets value assets and risks and channel funds from lenders to borrowers, whereas, banks mostly turn liquid liabilities into (long-term) loans. If neither markets nor banks are

functioning well, the transmission of QE is hampered.

(21)

To conclude, the design and transmission of QE depend on the structure of the financial system. Markets and banks act as intermediaries in the transmission of quantitative easing to the real economy. Thus, their functioning determines the effectiveness of

(22)

5. METHODOLOGY

This section starts by discussing the model specifications and methodology. The final paragraph presents two alternative models as robustness checks. Appendix D presents a third alternative specification.

5.1 Model specifications

To empirically investigate how a financial structure influences the transmission of quantitative easing, this thesis distinguishes between two linear fixed effects regression models. The first model uses an economic indicator as dependent variable; the second model uses a monetary indicator. Depending on the indicator variable, additional regressors are included. The baseline model is structured as follows:

𝐼𝑁𝐷𝐼𝐶𝐴𝑇𝑂𝑅𝑖,𝑡= 𝛽0+ 𝛽1𝑄𝐸(𝑖),𝑡+ 𝛽2𝐵𝐴𝑁𝐾𝑖,𝑡+ 𝛽3(𝑄𝐸(𝑖),𝑡∗ 𝐵𝐴𝑁𝐾𝑖,𝑡) + 𝛽4𝑋𝑖,𝑡+ 𝛼𝑖 + 𝑢𝑖,𝑡 (1)

5.1.1 The economic indicator

The economic indicator chosen as the dependent variable, 𝐼𝑁𝑇𝐸𝑅𝐸𝑆𝑇𝑖,𝑡, is the natural

logarithm of the 10-year (nominal) interest rates on government bonds, because reducing long-term interest rates is one of the intermediate targets of QE. The variable 𝑄𝐸(𝑖),𝑡 is

separated into three stock8 indicators, one flow9 indicator and a dummy variable. Each QE indicator is used as independent variable in separate regressions10.

Similar11 to the existing literature (e.g. Boeckx, Dossche & Peersman, 2017; Burriel & Galesi, 2018), to account for the effect of 𝑄𝐸(𝑖),𝑡 on long-term interest rates the natural

logarithm of ECB’s stock of assets in millions of euro is used. Additionally, to look at

country-specific quantitative easing measures, the model includes the natural logarithm of the National Central Banks’ stock of assets in millions of euro. The third QE stock indicator is defined as the stock of purchases of the Public Sector Purchase Programme in billions of euro.

8_{Outstanding amount (holdings) at the end of each period.} 9_{Transactions/quarterly conducted purchases.}

10_{Additional QE indicators were used. Namely, country-specific stock of PSPP purchases, country-specific flow} of PSPP purchases, stock of purchases of the EAPP and flow of purchases of the EAPP. In the interest of brevity and because all indicators yielded qualitatively the same results, estimations are not reported.

(23)

In addition, the model uses one flow indicator, defined as the flow of purchases conducted under the PSPPin billions of euro. Lastly, this thesis analyses QE by using a dummy during the period QE is applied (from Q1-2015 onwards). Since QE programmes are expected to affect interest rates directly (Hausken & Ncube, 2013), this model does not include a lagged 𝑄𝐸(𝑖),𝑡.

The independent variable that represents the financial structure is 𝐵𝐴𝑁𝐾𝑖,𝑡, and is

defined as the ratio of bank credit to total credit to the private non-financial sector. The “private non-financial sector” includes non-financial corporations (both private-owned and public-owned), households and non-profit institutions serving households as defined in the System of National Accounts 2008. The higher 𝐵𝐴𝑁𝐾𝑖,𝑡, the more bank-based a country’s

financial system is assumed to be.

Furthermore, an interaction term between 𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑄𝐸(𝑖),𝑡 is included. The sign and

significance of this term determines the outcome of this paper. A negative (positive) 𝛽₃ gives indication of an improved (worsened) transmission of QE in bank-based financial systems. To show that multicollinearity does not pose a threat the mean of 𝑄𝐸(𝑖),𝑡∗ 𝐵𝐴𝑁𝐾𝑖,𝑡is subtracted

from its original values. This transformation does not alter the interaction term’s coefficient or standard error but merely reduces its correlation with the main effects 𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑄𝐸(𝑖),𝑡.12

Moreover, a set of additional determinants of sovereign bond yields, 𝑋_𝑖,𝑡, is added to the model (Ciocyte et al., 2016; Dröes, van Lamoen & Mattheussens, 2017; Poghosyan, 2014).

The first determinant, 𝐸𝑂𝑁𝐼𝐴𝑡, is defined as the level of the effective overnight interest

rate for European interbank lending. 𝐸𝑂𝑁𝐼𝐴𝑡 measures the risk-free interest rate and is

included following Dröes et al., 2017, who assume that government bond yields consist of three elements; a risk-free element, a risk premium and a residual. A higher 𝐸𝑂𝑁𝐼𝐴𝑡 is

expected to push up government bond yields.

The second determinant, 𝜋_𝑖,𝑡, is defined as the 12-month average rate of change of inflation measured by the Harmonised Index of Consumer Prices (HICP). According to the Fisher equation, that estimates the relationship between nominal and real interest rates under inflation, nominal government bond yields and inflation are positively related.

(24)

The third determinant, 𝐺𝑂𝑉𝐷𝐸𝐵𝑇𝑖,𝑡, is defined as the natural logarithm of government

debt to GDP. This determinant is included because government debt is a measure of default risk. In turn, a higher risk of default is typically associated with higher interest rates.

The fourth determinant, Δ𝐺𝐷𝑃𝑖,𝑡, indicates the 15-month13 moving average growth rate

of gross domestic product. This term measures the influence of developments in output growth on long-term bond rates. The sign of Δ𝐺𝐷𝑃𝑖,𝑡’s coefficient can be either positive or

negative. A rise in GDP growth may increase tax revenues and reduce sovereign risk. As a result, GDP growth is expected to have a negative effect on sovereign bond yields. However, an increase in GDP growth can reduce the demand for bonds and consequently lower the prices of these bonds. Thus, GDP growth may positively affect sovereign bond yields.

The fifth determinant is the Composite Indicator of Sovereign Stress, 𝑆𝑜𝑣𝐶𝐼𝑆𝑆𝑡. This is

a composite indicator that measures the level of sovereign bond market stress in the euro area. The higher the stress in sovereign bond markets, the higher the interest rates on those

sovereign bonds.

The final determinant the model incorporates is a dummy, 𝐷𝑈𝑀𝑀𝑌𝑡, that captures the

effect of the sovereign debt crisis and equals one from Q1-2010 onwards. 𝐷𝑈𝑀𝑀𝑌𝑡allows for

a shift in the level of sovereign bond interest rates.

To eliminate omitted variables bias arising from unobserved variables that vary across entities but are constant over time, country fixed effects 𝛼𝑖 are added. The model excludes

time fixed effects because including them absorbs much of the variation in 𝐼𝑁𝑇𝐸𝑅𝐸𝑆𝑇𝑖,𝑡

caused by movements in 𝐸𝑂𝑁𝐼𝐴𝑡, 𝑆𝑜𝑣𝐶𝐼𝑆𝑆𝑡, 𝐷𝑈𝑀𝑀𝑌𝑡 and four QE indicators14, as these are

constant across countries but vary over time. Lastly, 𝑢_𝑖,𝑡 represents the error term. The previous explanation results in the following regression:

𝐼𝑁𝑇𝐸𝑅𝐸𝑆𝑇𝑖,𝑡 = 𝛽0+ 𝛽1𝑄𝐸(𝑖),𝑡+ 𝛽2𝐵𝐴𝑁𝐾𝑖,𝑡+ 𝛽3(𝑄𝐸(𝑖),𝑡∗ 𝐵𝐴𝑁𝐾𝑖,𝑡) + 𝛽4𝐸𝑂𝑁𝐼𝐴𝑡+ 𝛽5𝜋𝑖,𝑡+

𝛽6𝐺𝑂𝑉𝐷𝐸𝐵𝑇𝑖,𝑡+ 𝛽7Δ𝐺𝐷𝑃𝑖,𝑡+ 𝛽8𝑆𝑜𝑣𝐶𝐼𝑆𝑆𝑡+ 𝛽9𝐷𝑈𝑀𝑀𝑌𝑡+ 𝛼𝑖 + 𝑢𝑖,𝑡 (2)

Where 𝑄𝐸(𝑖),𝑡depicts the five quantitative easing indicators.

13_{This is done to smooth GDP fluctuations (just as HICP). The 15-month moving average is calculated as} follows: the quarter of interest, two quarters before and after the quarter of interest, are summed and divided by five for every point in time.

(25)

5.1.2 The monetary indicator

The monetary indicator, 𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡, is defined as the natural logarithm of total

credit to the non-financial sector15. Market-based and bank-based financial systems differ in the way financial intermediation is conducted. By implementing QE, the ECB increases market liquidity and extracts duration, which reduces the cost of borrowing and enhances credit growth. Accordingly, QE aims to spur economic activity. Therefore, examining QE, in interaction with the financial structure indicator, on total credit, provides valuable insights into the research of this paper. This leads to the following model:

𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡 = 𝛽0+ 𝛽1𝑄𝐸(𝑖),𝑡−4+ 𝛽2𝐵𝐴𝑁𝐾𝑖,𝑡+ 𝛽3(𝑄𝐸(𝑖),𝑡−4∗ 𝐵𝐴𝑁𝐾𝑖,𝑡) + 𝛽4𝜋𝑖,𝑡+

𝛽5Δ𝐺𝐷𝑃𝑖,𝑡+ 𝛽6𝐶𝐼𝑆𝑆𝑡+ 𝛽7𝐷𝑈𝑀𝑀𝑌𝑡+ 𝛼𝑖 + 𝑢𝑖,𝑡 (3)

Where 𝑄𝐸(𝑖),𝑡−4depicts the five quantitative easing indicators.

In contrast to interest rates, lending shows a delayed and varied response to monetary policy. However, because there is no comparable research that analyses the effects of

monetary policy on total credit lending, the exact lag of response is not determined. Although, the effect of monetary policy on bank lending, and the corresponding lag response of bank lending, have been thoroughly examined. Therefore, this thesis assumes the (lagged) response of total lending to monetary policy to be like bank lending. In empirical research, policy lags vary from one quarter (Morais, Peydro & Ruiz, 2015) to eight quarters (Cetorelli & Goldberg, 2012). In line with Gräb and Żochowski (2017), this model uses a four-quarter policy lag16, 𝑄𝐸(𝑖),𝑡−4, to capture a relatively slow adjustment of lending aggregates to changes in liquidity

conditions. 𝑄𝐸_{(𝑖),𝑡−4} reflects the same five quantitative easing indicators as in regression (2), though the indicators in this model are lagged four quarters.

Moreover, 𝐵𝐴𝑁𝐾𝑖,𝑡 is equivalent to 𝐵𝐴𝑁𝐾𝑖,𝑡 in the first model. Likewise, a (demeaned)

interaction term between 𝑄𝐸(𝑖),𝑡−4 and 𝐵𝐴𝑁𝐾𝑖,𝑡 is included. If this term turns out to be

significantly negative, this research shows that bank-based financial systems are less effective in the transmission of QE to the real economy.

15_{𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇}

𝑖,𝑡 is only weakly correlated with 𝐼𝑁𝑇𝐸𝑅𝐸𝑆𝑇𝑖,𝑡 : - 4.65%.

16_{As robustness checks, instead of using QE with a four-quarter lag, regressions were ran with no lags,} one-quarter, two-quarter and three-quarter QE lag. All tests yielded qualitatively the same results.

(26)

Lastly, the model includes four variables indicating macroeconomic/financial conditions and affecting total credit (Antoshin et al., 2017; Ivanović, 2016; Jiménez et al., 2012).

The first variable, 𝜋𝑖,𝑡, is defined as the 12-month average rate of change of inflation

measured by the Harmonised Index of Consumer Prices (HICP). Since higher inflation is expected to raise nominal interest rates and higher interest rates typically reduce lending, higher 𝜋𝑖,𝑡 is anticipated to decrease total credit lending.

The second variable, Δ𝐺𝐷𝑃𝑖,𝑡, is measured by the 15-month17 moving average growth

rate of gross domestic product. The GDP growth rate reflects the state of the economy. When GDP growth rises the demand and supply of credit are expected to rise also. Thus, GDP growth should be positively related to 𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡. However, higher credit may lead to

higher GDP growth, resulting in reversed causality. By using the 15-month moving average of GDP growth this thesis aims to avoid reversed causality.

The third variable, 𝐶𝐼𝑆𝑆𝑡, is defined as the Composite Indicator of Systemic Stress,

which is an indicator of instability or stress in the financial system. 𝐶𝐼𝑆𝑆𝑡increases in periods

of financial stress with depressed economic activity. Therefore, 𝐶𝐼𝑆𝑆𝑡is expected to be

negatively related to total credit.

Finally, the model incorporates a dummy variable, 𝐷𝑈𝑀𝑀𝑌𝑡, that captures the effect of

the sovereign debt crisis and equals one from Q1-2010 onwards. 𝐷𝑈𝑀𝑀𝑌𝑡allows for a shift in

the level of total credit.

To eliminate omitted variables bias arising from unobserved variables that vary across entities but are constant over time, country fixed effects 𝛼𝑖 are added. The model excludes

time fixed effects because including them absorbs much of the variation in 𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡

caused by movements in 𝐶𝐼𝑆𝑆𝑡, 𝐷𝑈𝑀𝑀𝑌𝑡, and four QE indicators18, as these are constant

across countries but vary over time. Lastly, 𝑢𝑖,𝑡 represents the error term.

17_{See footnote 8}

(27)

5.2 Robustness

As robustness checks, three alternative specifications of baseline model (1) are calculated. This section discusses two alternative specifications19. Appendix D presents the third alternative specification.

5.2.1 Market-based financial structure indicator

First, the model adds an indicator that represents the degree of market-based stock financing to the baseline regression, 𝑆𝑇𝑂𝐶𝐾𝑖,𝑡. This indicator is defined as the stock market

capitalization to GDP. The higher 𝑆𝑇𝑂𝐶𝐾𝑖,𝑡, the more market-based a country’s financial

structure is assumed to be. Besides adding a new variable to the model, the indicator 𝐵𝐴𝑁𝐾𝑖,𝑡

is defined differently. 𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡 still reflects the degree of bank-based financing but is now

measured as private bank credit to GDP. Both indicators follow from earlier studies (Bats & Houben, 2017; Gambacorta et al., 2014).

Second, the model uses another variable that indicates the degree of markets in a financial system, 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅𝑖,𝑡. 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅𝑖,𝑡is defined as the ratio total value of shares

traded to average real market capitalization (Beck & Levine, 2004; Gambacorta et al., 2014). Instead of using the stock of purchases of the PSPP and the flow of purchases of the PSPP, these models use the stock of purchases of all asset purchase programmes and the flow of purchases of all asset purchase programmes as two QE indicators, because data on the latter variables cover a longer period. The other three 𝑄𝐸(𝑖),𝑡indicators, ECB’s stock of assets,

NCBs’ stock of assets and QE DUMMY are equivalent to the 𝑄𝐸_(𝑖),𝑡indicators used in equations (2) and (3).

In line with regression equations (2) and (3), the models include the interaction terms 𝑄𝐸(𝑖),𝑡∗ 𝑆𝑇𝑂𝐶𝐾𝑖,𝑡, 𝑄𝐸(𝑖),𝑡∗ 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅𝑖,𝑡and 𝑄𝐸(𝑖),𝑡∗ 𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡. The signs and significance

levels of these terms determine the outcome of the robustness check. This gives the following linear fixed effects regressions:

𝐼𝑁𝐷𝐼𝐶𝐴𝑇𝑂𝑅𝑖,𝑡 = 𝛽0+ 𝛽1𝑄𝐸(𝑖),𝑡+ 𝛽2𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡+ 𝛽3(𝑄𝐸(𝑖),𝑡∗ 𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡) + 𝛽4𝑆𝑇𝑂𝐶𝐾𝑖,𝑡+

𝛽5(𝑄𝐸(𝑖),𝑡∗ 𝑆𝑇𝑂𝐶𝐾𝑖,𝑡) + 𝛽6𝑋𝑖,𝑡+ 𝛼𝑖 + 𝑢𝑖,𝑡 (4)

(28)

𝐼𝑁𝐷𝐼𝐶𝐴𝑇𝑂𝑅𝑖,𝑡= 𝛽0+ 𝛽1𝑄𝐸(𝑖),𝑡+ 𝛽2𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡+ 𝛽3(𝑄𝐸(𝑖),𝑡∗ 𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡) +

𝛽4𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅𝑖,𝑡+ 𝛽5(𝑄𝐸(𝑖),𝑡∗ 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅𝑖,𝑡) + 𝛽6𝑋𝑖,𝑡+ 𝛼𝑖 + 𝑢𝑖,𝑡 (5)

Where 𝐼𝑁𝐷𝐼𝐶𝐴𝑇𝑂𝑅𝑖,𝑡 represents 𝐼𝑁𝑇𝐸𝑅𝐸𝑆𝑇𝑖,𝑡 and 𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡. 𝑄𝐸(𝑖),𝑡depicts the five

quantitative easing indicators. In case 𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡 is used as dependent variable, 𝑄𝐸(𝑖),𝑡 is

lagged one period: 𝑄𝐸_{(𝑖),𝑡−1}. 𝑋𝑖,𝑡represents the set of additional determinants that is added to

(29)

6. DATA

The models are fixed effects panel regressions with quarterly time data “t” on country “i”. The panel consists of nineteen eurozone countries namely, Austria, Belgium, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Portugal, Slovakia, Slovenia and Spain.

Since not all countries simultaneously became member of the euro area, the timespan covered differs across countries. For the eleven original member states time series begin in 1999, followed by Greece in 2001, Slovenia in 2007, Cyprus and Malta in Q1-2008, Slovakia in Q1-2009, Estonia in Q1-2011, Latvia in Q1-2014 and finally, Lithuania in Q1-2015. All series end in Q3-2017.

Data for 𝐼𝑁𝑇𝐸𝑅𝐸𝑆𝑇𝑖,𝑡, 𝐺𝑂𝑉𝐷𝐸𝐵𝑇𝑖,𝑡, 𝑄𝐸(𝑖),𝑡, 𝐸𝑂𝑁𝐼𝐴𝑡, 𝑆𝑜𝑣𝐶𝐼𝑆𝑆𝑡 and 𝐶𝐼𝑆𝑆𝑡are taken from

the European Central Bank (ECB) Statistical Warehouse. Data for 𝜋_𝑖,𝑡 and 𝐺𝐷𝑃𝑖,𝑡 are obtained

from DataStream.

𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡 data for Austria, Belgium, Finland, France, Germany,

Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal and Spain are collected from the Bank for International Settlements (BIS). 𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑇𝑂𝑇𝐴𝐿𝐶𝑅𝐸𝐷𝐼𝑇𝑖,𝑡 data for Cyprus,

Estonia, Latvia, Lithuania, Malta, Slovakia and Slovenia are the author’s own calculations and are constructed following the paper of Dembiermont, Drehmann and Muksakunratana (2013). The underlying data for these calculations are taken from the European Central Bank (ECB) Statistical Warehouse. Appendix A presents the computation method.

Data for the alternative indicators 𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡, 𝑆𝑇𝑂𝐶𝐾𝑖,𝑡 and 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅𝑖,𝑡 are

obtained from the World Bank’s Global Financial Development and Structure Dataset. Data are on yearly basis and cover the timespan from 1999 to 2015.

Table 1 gives a summary of the statistics. Total credit, four QE indicators, EONIA, inflation (HICP) and CISS have no missing values. The variable interest shows missing values because there is no suitable proxy indicator for government bonds yield in Estonia. Furthermore, there are no available bank credit data for the period 1999-2002 in Luxembourg. Finally, data on the variable sovereign CISS are only available from Q3-2002 onwards.

(30)

Table 1: Descriptive statistics

Variables Unit of measurement Obs. Mean Std. Dev. Min Max

Dependent variables Interest Total credit % Millions of euro 1,074 1,101 3.85 1,394,426 2.33 1,679,087 -0.12 14,164.38 25.4 6,870,506 Independent variables

ECB’s stock of assets NCBs’ stock of assets QE stock of purchases QE flow of purchases BANK EONIA Inflation (HICP) Government debt GDP growth CISS Sovereign CISS Millions of euro Millions of euro Billions of euro Billions of euro % of total (PNFS) credit % % % of GDP % Pure number Pure number 1,101 1,101 1,101 1,101 1,085 1,101 1,101 1,073 1,091 1,101 1,035 1,905,538 155,182.8 170.40 30.79 63.45 1.51 1.83 70.78 0.47 0.20 0.21 980,348 221,437.6 431.30 66.67 17.98 1.63 1.36 35.36 0.90 0.17 0.16 695,644 2,140 0 0 15.81 -0.36 -2.50 6.07 -2.46 0.04 0.03 4,318,624 1,675,408 1,784.14 230.24 98.30 4.84 5.97 180.93 6.08 0.76 0.62 Note: The dependent variable total credit is total credit to the non-financial sector whereas the unit of

measurement of BANK is in % of total credit to the private non-financial sector. The four QE indicators: ECB’s total assets is the ECB’s stock of assets, NCBs’ total assets are the NCBs’ stocks of assets, QE stock of

purchases reflects the stock of asset purchases made under the PSPP and QE flow of purchases is the flow of purchases conducted under the PSPP. The variable BANK depicts the financial structure indicator and is measured as the ratio bank credit to total credit to the private non-financial sector.

Appendix B presents five correlation matrices for all (transformed) independent variables. A risk of multicollinearity appears between the variable 𝐸𝑂𝑁𝐼𝐴𝑡 and the

quantitative easing indicator ECB’s total assets. The correlation equals -0.8178. However, including 𝐸𝑂𝑁𝐼𝐴𝑡 in the model slightly changes the coefficients and standard errors of the

variables of interest. Thus, multicollinearity issues seem to be minor. Furthermore, the QE indicators, QE stock of purchases, QE flow of purchases and QE DUMMY, are highly correlated with the demeaned interaction term, which is caused by the fact that most of the observations are zero. Adding the interaction term to the regression does not change the significance or sign of the main effect QE.

Compared to the United States, Europe is relatively bank-based. Nevertheless, within Europe there are large differences in financial structure. Figure 3 illustrates these differences in time-plots for the financial structure indicators of the six euro-area countries with the highest GDP. The reason these countries are selected is to show that there are considerable differences in financial structure while other country characteristics are similar. The financial systems of Germany, Italy and Spain are the most bank-based whereas in the Netherlands,

(31)

Belgium and France they rely less heavily on bank credit. The figure also shows the minimum and maximum observations of the entire panel.

Figure 3: financial structure indicator

Note: the figure presents time-plots for the financial structure indicator, 𝐵𝐴𝑁𝐾𝑖,𝑡, for the six euro-area countries

with the highest GDP. 𝐵𝐴𝑁𝐾𝑖,𝑡 is defined as the ratio of bank credit to total credit to the private non-financial

sector. The thick top line represents the maximum observations for the financial structure indicator 𝐵𝐴𝑁𝐾𝑖,𝑡 and

the thick bottom line represents the minimum observations.

Figure 4 displays the time variation of the dependent variable 10-year government bond yields. Interest rates had been moving closely together until the sovereign debt crisis hit in 2010, thereafter yields diverged enormously. Interest rates on Italian and Spanish

government bonds reached values of above six percent while those of the Netherlands, France and Germany only increased a little. After the implementation of multiple rescue programmes by the ECB, interest rates converged again and moved to below one percent except for those of Italy and Spain.

0 20 40 60 80 100 120 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 2 201 3 201 4 201 5 201 6 201 7 Belgium France Germany Italy the Netherlands Spain

(32)

Figure 4: 10-year government bond interest rates

Note: the figure shows time-plots for the depended variable ten-year government bond interest rates for the six euro-area countries with the highest GDP. The dotted line on the left represents the quarter the ECB announced and implemented the CBPP3 and ABSPP. The solid line on the right represents the quarter the ECB announced and implemented the PSPP.

Figure 5 and 6 show time-plots for bank credit to the private non-financial sector and total credit to the non-financial sector, respectively. Before the crisis, credit showed an upward trend in almost all countries, which can be attributed to the excessive credit provided by banks. However, when the financial bubble burst banks were unwilling and unable to lend and with few sources of alternative financing, credit growth stagnated. This was especially the case in bank-based financial systems like Italy and Spain. Conversely, Germany’s total credit continued to grow. Nonetheless, after 2010, total credit in France bypassed that of Germany, which gives an indication of a more resilient financial system in France. To conclude, total credit and bank credit remained roughly constant in the Netherlands and Belgium.

-1 0 1 2 3 4 5 6 7 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 2 201 3 201 4 201 5 201 6 201 7 Belgium France Germany Italy the Netherlands Spain

(33)

Figure 5: bank credit to the private non-financial sector

Note: the figure shows time-plots for bank credit to the private non-financial sector in millions of euro for the six euro-area countries with the highest GDP. The vertical line points to the beginning of the “Great Recession”.

Figure 6: total credit to the non-financial sector

Note: the figure shows time-plots for the depended variable total credit to the non-financial sector in billions of euro for the six euro-area countries with the highest GDP.

0 500000 1000000 1500000 2000000 2500000 3000000 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 2 201 3 201 4 201 5 201 6 201 7 Belgium France Germany Italy the Netherlands Spain 0 1000 2000 3000 4000 5000 6000 7000 8000 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 2 201 3 201 4 201 5 201 6 201 7 Belgium France Germany Italy the Netherlands Spain

(34)

Figure 7 shows the total assets of the European Central Bank. Total assets increased when the ECB adopted the fixed-rate full allotment procedure in 200820. Thereafter, it implemented several purchase programmes. Starting with the covered bond purchase programme (CBPP) in 2009, Securities Markets Programme (SMP) in 2010 and CBPP2 in 2011. These three programmes ended in June 2010, September 2012 and October 2012, respectively. Then, in October 2014, it launched a fourth programme, CBPP3. From October 2014 onwards, assets have risen enormously, which can be attributed to the three latest programmes; asset-backed securities purchase programme (ABSPP), started in November 2014, public sector purchase programme (PSPP), began in March 2015, and corporate sector purchase programme (CSPP), launched in June 2016. All four programmes together are called the expanded asset purchase programme (EAPP). Lastly, figure 8 shows the flow of purchases conducted under the ABSPP, CBPP3, CSPP and PSPP programme (QE).

Figure 7: total assets (stock) of the European Central Bank

Note: this figure presents the time-plot for ECB’s stock of assets in millions of euro. The vertical lines point to the quarter of implementation of (from left to right): FRFA, CBPP, SMP, CBPP2, CBPP3/ABSPP, PSPP and CSPP.

20_{Under the fixed-rate full allotment policy, banks can get all the liquidity they demand from the ECB, on the} condition that they provide adequate collateral and are financially stable.

0 500000 1000000 1500000 2000000 2500000 3000000 3500000 4000000 4500000 5000000 199 9 200 0 200 1 200 2 200 3 200 4 200 5 200 6 200 7 200 8 200 9 201 0 201 1 201 2 201 3 201 4 201 5 201 6 201 7

(35)

Figure 8: flow of purchases of ongoing asset purchase programmes

Note: the figure shows the flow of purchases in millions of euro conducted under the ABSPP, the CBBP3, the CSPP and the PSPP. -10000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 201 4 201 5 201 6 201 7 201 8 Asset-backed securities purchase programme Covered bond purchase programme 3

Corporate Sector purchase programme

Public sector purchase programme

(36)

7. RESULTS

This section analyses the empirical findings obtained from the two linear fixed effects regression models, (2) and (3). Both models include robust standard errors and country fixed effects.

7.1 The economic indicator model

Table 2 presents the results for fixed effects regression (2). Column I of table 2 reports the results for the panel regression of the logarithm of the 10-year government bond yields on the logarithm of ECB’s stock of assets and the financial structure indicator. Column II adds the interaction term of the two variables. These results prove the importance of financial structure in the effect of quantitative easing measures on interest rates. Column II shows that at the 1% significance level, an increase in the ECB’s stock of assets, negatively affects the 10-year government yield21. However, the estimations suggest that, when QE is implemented in bank-based economies, it is less effective. Based on the positive sign and 10% significance level of the interaction term, it can be stated that the more bank-based a financial structure is, the less effective quantitative easing measures are. Including the set of additional

determinants in the model hardly changes the significance of the interaction term, which proves that the negative influence of a bank-based financial structure on QE is robust to the set of controls. In turn, this strengthens the evidence for a reduced effectiveness of QE measures in bank-based financial systems.

Furthermore, columns IV-VII present the regression results of model (2) with the four other independent variables that indicate QE. From these estimations a similar conclusion can be drawn. Particularly, no matter which QE indicator is used, the outcomes indicate that quantitative easing is less able to reduce long-term interest rates in countries with a higher degree of bank-based financing22_.

As a robustness check, based on earlier research (Bats & Houben, 2017; Gambacorta et al., 2014), 𝐼𝑁𝑇𝐸𝑅𝐸𝑆𝑇𝑖,𝑡is regressed on a new indicator, 𝑆𝑇𝑂𝐶𝐾𝑖,𝑡, reflecting the degree of

market-based stock financing, and an indicator representing the degree of bank-based

21_{The same logic applies to all QE indicators.}

22_{Coefficients and standard errors in column V, VI and VII hardly change after including the set of additional} determinants. Coefficients and standard errors in column IV change considerably after including the set of additional determinants, which could be a matter of collinearity.

(37)

financing, 𝐵𝐴𝑁𝐾𝑅𝑂𝐵𝑖,𝑡. In addition to table 2, these results show that a market-based financial

system has a significant positive impact on the effects of quantitative easing23. By contrast, the estimations of regression (5), with 𝑇𝑈𝑅𝑁𝑂𝑉𝐸𝑅𝑖,𝑡 as indicator

reflecting the degree of markets in a financial structure, give mixed results. Particularly, the results show that financial market development has an insignificant positive or negative influence on the effect of QE24.

23_{See appendix C, table 1} 24_{See appendix C, table 3}

The influence of financial structure on the effectiveness of quantitative easing