Quantitative easing : an empirical study on the rebalancing of eurozone bank assets

(1)

An Empirical Study on Eurozone Bank Assets.

Quantitative Easing: An Empirical Study on

the Rebalancing of Eurozone Bank Assets.

Tom van Gool

Thesis Submitted for BSc Economics & Business Ms. I. Sakalauskaite MSc

University of Amsterdam January 2018

(2)

An Empirical Study on Eurozone Bank Assets

2 Statement of Originality

This document is written by Student Tom van Gool who declares to take full responsibility for the contents of this document.

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

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

This empirical study researches the effects of Quantitative Easing (QE) on the structure of bank earning assets. The research is done through a panel regression containing 175 Eurozone banks from 2012 to 2016 and their loan-to-earning assets ratio. This paper finds a positive relationship between the size of QE and a rebalancing effect towards lending. Furthermore, this thesis presents findings providing evidence of a weaker rebalancing effect of QE for cooperative banks and banks in crisis nations. Also, this paper provides evidence of differences in rebalancing for larger and smaller banks. Besides the differences in ownership and location this thesis also presents evidence of a variation in operating mechanisms for asset purchase programs and the rebalancing effect for government owned banks. The results of this paper contribute to the debate about effective monetary policy and literature on QE and monetary policy transmission.

1. Introduction

This thesis is meant to provide an empirical study into the rebalancing effects of quantitative easing (QE) on the structure of earning assets for banks in the Eurozone, over the period 2012-2016. With interest rates at near zero banks have increasingly started looking at other options to attain higher earnings, potentially diverting the gains from QE to financial markets rather than the real economy. With the European Central Bank (ECB) having started its third round of QE during 2014 and most other central banks limiting their QE purchases during the selected years, this period provides an opportunity to look at the effects of QE with little foreign interference. Since the Eurozone is a relatively bank dependent region the effects of QE on the earning assets structure are important to investigate, as bank lending has a larger effect on the economy compared to other regions. Through researching the earning assets structure results can be found with regard to whether QE draws funds towards lending or away from lending. Also, with the region having a highly diverse banking sector it could be

(3)

3 the case that previous research from Japan, the USA, or the UK, might not be applicable given the European context.

Research about the effects of QE has become increasingly popular in recent years. Several papers have successfully shown a rebalancing effect in investment portfolios and increases in lending. Previous research shows a positive relation between QE, asset prices and financial markets performance. The link of QE to the lending itself has been discussed to a lesser extent. Recent papers from the ECB and others from the US show an increase in lending related to QE. However, little research has been done about the rebalancing of the broader asset portfolio to see if QE causes a shift towards increased lending in bank assets.

The main research question of this thesis is; ‘Does QE influence the structuring of earning assets for Eurozone banks and did these effects differ across the region? And if so, which asset purchase programs (APPs) were most effective in doing so?’ The answer to the question is found through multiple panel data regressions of the loans-to-earning assets ratio on QE for a total of 175 banking institutions located and operating in the region. The results will provide a better insight into the effects QE has on lending and would therefore allow authorities to increase the effectivity of their QE programs, should they enact new programs like QE. The results would enhance the current literature on QE by adding additional

knowledge to the effects it has in the Eurozone, with regard to the rate at which banks lend compared to other financial activities.

The thesis is structured in the following way. Section 2 gives an overview and discussion of the relevant literature about monetary policy transmission and QE. Section 3 provides the methodology, variables, and sample selection. Section 4 presents the results and tables of the empirical study. Section 5 discusses the implications of the research and the conclusions that can be made through it.

1. Literature Review

This thesis can be placed among literature on the transmission of monetary policy on through the bank lending channel and bank asset structure, with QE being the monetary policy tool which is being analyzed.

The rebalancing effect of bank assets is one of the most prominent features of the monetary transmission of QE, as it is linked to economic growth and inflation. This is because QE causes a shock in demand for the assets that are bought through the program and has significant effects on relative asset prices and spreads. Through the lower interest rates the

(4)

4 relative attractiveness of certain assets decreases causing banks to reallocate their money to other areas, such as mortgages. Through the rebalancing caused by QE there is increased economic and inflation stimulus, originating from a shift from other earning assets to lending, according to theoretical modelling by Jouvanceau (2016). The rebalancing effect is caused by the attractiveness and risk of certain assets over other assets and banks consequently

channeling funds from one to the other asset category (Tobin, 1953). Lower interest rates on assets tied to QE, such as MBS, ABS, and covered bonds, would channel more funds to those assets through a rebalancing effect of bank investment portfolios. This effect is confirmed by Joyce, Lasaosa, Stevens, and Tong (2010), in combination with various other trading classes such as currencies, ETFs and bond indices. However, the paper by Joyce, et al. (2010) only looks at a banks other earning assets, whereas this thesis investigates the broader composition of a bank’s asset portfolio. A paper by Butt, Domit, Kirkham, McLeay, and Thomas (2013) shows that when the rebalancing effect of QE is investigated through looking at broad money measures in the UK lending did not increase and was becoming less compared to total

financial activity. However, Butt, et al. (2013) looked at rebalancing through broad money, which would also include changes in household balance sheets, whereas this thesis tries to find evidence of rebalancing using mainly bank level data.

When central banks engage in QE the transmission effects of the policy are mainly constricted to the type of asset being purchased. QE would increase the demand for the asset for purchase in the program, effectively lowering the yield of this asset. This would

consequently make production of the asset more attractive for financial institutions as it offers them a relatively easy way to increase funds. Lending tied to the financial product in question would then again go up, creating a monetary transmission effect limited to the asset. The effects of QE being limited to only certain asset types have been researched by

Krishnamurthy and Vissing-Jorgensen (2013) in relation to the Federal Reserve’s large scale purchases of MBS. The lending channel tied to MBS has then been further analyzed by Darmouni and Rodnyansky (2017), who prove that lending did increase for banks with higher MBS holdings. For this thesis it would implicate that the effects of programs within the ECB’s full QE are not homogenous in rebalancing asset structures.

With regard to lending itself, QE has a positive effect on funds directed at lending in general. This was found through the lower interest rates which make consumers and banks more willing to engage in debt, either for large purchases which could be done with low costs or to roll over previous debt with cheap loans. The effects of increased lending where

(5)

5 has been confirmed by Christensen and Rudebusch (2012) in the UK and USA, and by Ugai (2006) in Japan. With regard to this thesis it would implicate that for many banks in the Eurozone to loan-to-asset ratio might have gone up through the increased lending, given that effect on interest rates are consistent around the world.

The relationship between loans and assets, which is can be seen as a bank asset structure ratio, is influenced by changes in lending interest rates in a negative way. Higher lending rates would indicate a higher risk premium on loans, therefore lowering the

attractiveness of lending compared to other financial investments. The negative relationship between the loan-to-asset ratio and lending rate changes has been found by Müller,

Schwaninger, and Vithessonthi (2017). Taken together with evidence of QE causing a decline in interest rates tied to the asset for purchase it could be assumed that QE would increase the loan-to-asset ratio, given the negative relationship between rates and lending and the lowering of risk premiums on the purchased assets. Given the information of lower rates causing rebalancing towards certain lending and QE causing a rebalancing effect directed at the purchased assets through lower rates it could be that QE itself raises the loans-to-earning assets ratio through this channel.

Monetary policy transmission effects are not consistent through all types of banks. Previous research shows that smaller banks are more affected by monetary policy than larger banks. The reasoning behind this is that smaller banks are less liquid and are therefore less able to find money to generate new loans, during monetary tightening (Kashyap and Stein, 2000). Differences in lending caused by monetary shocks for small and large banks has also been further confirmed by Dewally and Shao (2014), in relation to the crisis and C&I lending. For this thesis it would mean that if there is monetary easing banks with lower

asset-to-securities ratios would have an increased effect of easing on lending, since less liquid banks would get more money available for lending which they previously did not have. Another point on which effectivity of monetary policy differs is ownership. Research shows that government owned banks and cooperative banks are less affected by monetary policy.

Government owned banks are less affected because of guidelines set by government agencies which own the banks (Andries and Billon, 2010). Whereas cooperative banks are less affected because they draw most of their funds for lending from within their cooperative, rather than the broad financial markets (Ehrmann and Worms, 2004). This would implicate that the effects of QE would be weaker for government owned and cooperative banks, compared to private banks. The effects of QE on lending were also weaker for banks in nations more severely hit by the European sovereign debt crisis. The reasoning behind this is that banks in

(6)

6 these nations were less focused on expanding their lending and more focused on using QE to gain reserves to meet reserve and solvency requirements. The influence of economic

conditions on monetary policy transmission were confirmed by Bernanke and Blinder (1992). The weaker effect tied to the European sovereign debt crisis has been confirmed and

explained by Reichlin (2014).

The hypothesis of this research, given the previous literature, is that QE increases the loan-to-assets ratio. This effect is caused by a rebalancing effect toward financial products purchased by central banks. The rebalancing operates through a narrow channel of

transmission linked to the specific type of assets purchased by the ECB. However, these effects differ across various types of banks yielding different results across banks in the Eurozone.

2. Empirical Analysis

2.1 Sample

The sample consists of the current active banks in the Eurozone. The companies have first been selected through the Orbis database based on the criteria that companies should be active since at least 2012. Banks currently in reorganization, default, or in a rescue plan are not included. The second criterium is that the selected companies are located in the Euro Area. The third criterium is that balance sheet data is available for at least the past five according to bank balance sheet formatting. The fourth criterium is that the companies are part of the following NACE Rev. 2 industry classifications a) 6419 – Other monetary intermediation, indicating institutions with deposits and lending possibilities which are not central banks, b) 6420 – Activities of holding companies, indicating the companies that are the owners of subsidiaries engaging in financial services, c) 6491 – Financial Leasing, indicating companies that engage in financial leasing activities, d) 6492 – Other credit granting and e) 6499 – Other financial service activities, except insurance and pension funds. The classifications mean that trusts, funds, and other similar financial entities are excluded, the same goes for central banks. The fifth and last criterium is that the bank is either the global ultimate owner (GUO) or that the GUO is a eurozone government institution or a family or private individual that resides in the eurozone. This leaves a sample of 175 banks, which consists are currently active in the Eurozone, with available balance sheet data, which are either independent or under

(7)

7

3.2 Methodology

To find an answer to the research question this thesis will look at how QE affects the loan-to-earning assets ratio using panel data from 175 banks during 2012-2017. The analysis will be done through the following model:

𝐿𝑜𝑎𝑛𝑠

𝐸𝑎𝑟𝑛𝑖𝑛𝑔 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡 = 𝛽0+ 𝛽1ln(𝑄𝐸𝑖,𝑡 + 1) + 𝜃1𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡 + 𝜀𝑖,𝑡 (1)

The model indicates that bank asset structure is dependent of the size of QE and various other control variables. According to a model specified by the ECB bank asset structure consists of three parts; loans, securities, and interbank loans (ECB, 2013). In this research the ECB model is used as a basis to for the dependent variable. Therefore, the main dependent variable is the loan-to-earning asset ratio. Through this variable the rebalancing effect on bank asset structure can studied. Changes in the value of the dependent variable would indicate a shift in a banks choice of lending compared to other investment activities. Loans in this thesis is specified as mortgages, C&I lending, and other consumer and retail loans. Earning assets indicate the total amount of assets generating a return, including loans, interbank loans, securities, derivatives, real estate investments, and insurance assets. The choice of earning assets instead of total assets has been made to keep in line with the ECB paper. Also, it results in the exclusion of cash and due from banks, fixed assets, foreclosed real estate, goodwill, tax assets, and discontinued operations from the ratio. In their working paper the ECB also exclude non-earning assets from bank asset structure. The main explanatory variable is the logarithm of the combined value of the yearly asset purchases done by the ECB under its QE programs during the selected time frame. In the extended regressions ownership is taken into account as a factor that influences the transmission effects of QE, resulting in interaction variables between QE and government ownership, QE and cooperative ownership. An interaction between QE and the banks having its main Eurozone operations based in Greece, Ireland, Italy, Portugal, or Spain, which are defined as the worst hit nations of the European sovereign debt crisis by Lane (2012), is included to investigate differences in transmission effects caused by economic environment.

The first main control variable is the logarithm of total assets, to control for bank size. Controlling for bank size is very common throughout literature on banking and monetary

(8)

8 policy, as a banks size significantly influences investment policy (Kashyap and Stein, 2000; Kishan and Opiela, 2000). The expected coefficient for bank size is negative, as larger banks have more access to other earning assets than loans compared to smaller banks. The second main control variable is the leverage ratio, which indicates the relative size of a banks equity compared to liabilities. This variable is included because it enables to control for differences in capital structures (Kishan and Opiela, 2000). A higher leverage might influence the loans-to-earning assets ratio in a negative way, as banks can engage in more risky investments. The third main control variable is the return-on-assets (ROA) as a measure to control for

differences in profitability (Darmouni and Rodnyansky, 2017). Differences in profitability might influence the loans-to-earning assets ratio as less profitable banks might engage short-term investments to boost profitability levels. Therefore, it is expected that the ROA and loans-to-earning assets ratio are positively related. The fourth main control variable is the logarithm of GDP per capita of the nation where the bank mainly operates in the Eurozone as a way to control for general economic conditions. When economic conditions are better banks are more inclined to lend to the public as the chances of defaulting on loans go down. The expected coefficient for GDP per capita is therefore positive (Kakes and Sturm, 2002). Finally, the lending interest rate is added as a control variable in the base regression as a measure to control for yields on lending. Based on previous literature the variable is expected to influence the rebalancing effect in a negative way (Müller, Schwaninger, and Vithessonthi, 2017).

One of the additional control variables is the reserves-to-assets ratio to control for differences in a bank’s financial buffer. A higher reserve ratio might indicate a lower loan-to-earning assets ratio as banks can take on more risk. The second additional control variable is the loan-loss-provision (LLP) ratio, as a measure to control for general loan performance. A higher LLP-ratio indicates loans that underperform. Having more underperforming loans might increase the need the engage in more additional lending, increasing the loan-to-earning assets ratio. Thirdly, there is the loan-to-deposits (LTD) ratio as an additional control variable to control for the general state of lending within a bank. A high LTD-ratio indicates that a bank has little deposits to back additional lending, while a low LTD-ratio indicates a bank can lend more. The expected coefficient of the LTD-ratio is therefore negative. Another additional control variable is the non-earning-assets ratio, to control for the relative size of a banks fixed assets, goodwill, foreclosed real estate, tax assets, bank infrastructure investments, and other assets that do not generate returns. A higher non-earning-assets ratio might indicate a

(9)

9 The effects of monetary policy are expected to be different for larger banks compared to small banks. Therefore additional regressions are added to look at the effects for banks below the 25th percentile (€1,473.2 million) in terms of size and banks above the 75th

percentile (€43,394.2 million) based on total assets. The definition of size on which the banks were placed in a group was made using the average size over the 5 years included in the sample.

Since the effects of different QE programs are expected to vary with each program, due to the narrow channel of transmission, final regressions will be done where QE is split into the various programs. This results in the variable ABSPP, which measures the total yearly ECB purchases of high grade ABS from banks. Another variable is CBPP, which measures the total yearly ECB purchases of covered bonds from banks under CBPP2 and CBPP3. The next variable is CSPP, which measure the total yearly ECB purchases of

corporate bonds from non-banking corporations in the Eurozone. Lastly, the variable PSPP is added, which measures the total yearly purchase of government agency bonds.

Some rebalancing effects caused by QE and the other control variables might not be visible immediately or during the year that a variable is recorded. Therefore an additional regression is done with lagged effects:

𝐿𝑜𝑎𝑛𝑠

𝐸𝑎𝑟𝑛𝑖𝑛𝑔 𝐴𝑠𝑠𝑒𝑡𝑠𝑖,𝑡+1 = 𝛽0+ 𝛽1ln(𝑄𝐸𝑖,𝑡+ 1) + 𝜃1𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡+ 𝜀𝑖,𝑡 (2)

3.3 Descriptive Statistics

Table 1 provides descriptive statistics for the loans-to-earning assets ratio as well as the control variables which are ratios. The table also presents descriptive statistics for the logarithmic variables. The statistics reported are the mean of each variable, the standard deviations, and the value of the 25th percentile, the 50th percentile, and 75th percentile. Also the amount of observations is shown. Data on bank specific variables has been provided using the Orbis database, whereas data on QE comes from the ECB. Data on GDP has been

retrieved from the Eurostat database. With regard to the descriptive statistics for the separate APPs, it should be noted that no QE program was active during the entire time frame. Table 2 provides an overview of percentages of banks that are private, government owned or

cooperative owned in the sample. The percentage of for banks located in crisis regions in the sample is also presented. The tables also show that the panel unbalanced, for three banks on

(10)

10 year of information was lacking from the dataset. Figure 1 provides an oversight of the active QE programs during 2012-2016, using quarterly data.

Table 1: Summary statistics

Variables Mean Std.Dev. P25 P50 P75 Obs.

Loans-to-EA Leverage ROA Reserve/Assets LLP-ratio NEA/Assets Interest-rate Ln(QE) Ln (ABSPP) Ln (CBPP) Ln (CSPP) Ln (PSPP) Ln(TA) Ln(LTD) Ln(GDP/Cap) 0.6148 0.1305 0.0034 0.0063 0.3489 0.0692 0.0153 10.4305 6.3284 10.2490 2.4552 5.3537 15.8453 1.0244 10.3543 0.2116 0.3990 0.0153 0.0064 0.7629 0.0692 0.0141 5.4507 5.4265 5.3153 4.9027 6.5434 2.5425 1.1342 0.2599 0.5288 0.0630 0.0012 0.0026 0.0288 0.0302 0.0108 10.9025 0.0000 10.8629 0.0000 0.0000 14.2029 0.5965 10.1924 0.6657 0.0893 0.0045 0.0048 0.1329 0.0438 0.0153 11.8332 7.6559 11.8332 0.0000 0.0000 15.9099 0.7543 10.4043 0.7676 0.1284 0.0085 0.0085 0.4356 0.0868 0.0187 14.3839 11.6183 13.9634 0.0000 13.1122 17.5858 0.9803 10.5401 872 872 872 872 872 872 872 872 872 872 872 872 872 872 872 Summary statistics are based on the consolidated bank data provided by Orbis. Data on QE is provided by the ECB. Data on GDP per capita is provided by Eurostat. The data is on a yearly basis, for the years 2012 to 2016. A more specific view of the separate APPs is provided in figure 1. Full descriptions of the variables and the correlations table are provided in the appendix.

Table 2: Descriptive statistics ownership and location

Total TA < P25 TA > P75 Private (%) Government (%) Cooperative (%) Crisis (%) 44.15% 20.07% 35.78% 41.86% 34.57% 2.30% 63.13% 39.17% 62.79% 32.56% 4.65% 48.84% Observations Groups 872 175 217 44 215 43 Summary statistics are based on information on the GUO and the mean of the binary variables. Private was obtained through subtracting 1 – mean govt. – mean coop. The difference in observations for each group is caused by an imbalance in the panel.

(11)

11 Figure 1: Quarterly QE purchases, for separate APPs. Using data retrieved from the ECB.

The data provided in Figure 1 show that QE was not active during all years of the sample. CBPP2 ended during late 2012 and CBPP3 started during early 2014. ABSPP started around a fairly similar time as CBPP3. PSPP became an active part of QE during 2015, while the CSPP is the most recent program, which has only been active since 2016. During 2013 there were no observations for QE as there were no active programs during this year. The figure also shows that the largest QE program is the PSPP, followed by CBPP2 and CBPP3. ABSPP is the smallest program in terms of yearly purchases.

4. Results

The results of the panel data regression of QE on the loan-to-earning assets ratio are presented in Table 3. The table shows that QE is positively related to the loan-to-earning assets ratio. This effect is consistent throughout the table, however the significance becomes less when taking year fixed effects into account. The most significant result is in the extended regression without year fixed effects where a 1% increase in QE increases the loan-to-earning assets ratio by 0.0011. These results implicate that QE rebalances bank earning asset structures towards loans, rather than other financial activities. The effect is consistent with the research hypothesis, being QE raising the loan-to-earning asset ratio. Furthermore, table 3 also

-10000 0 10000 20000 30000 40000 50000 60000 70000 80000 90000 14-9-2011 1-4-2012 18-10-2012 6-5-2013 22-11-201310-6-201427-12-201415-7-2015 31-1-2016 18-8-2016 6-3-2017

Asset Purchase Programs (x1.000.000€)

Asset-backed securities purchase programme Covered bond purchase programme 3 Corporate Sector purchase programme Public sector purchase programme Covered bond purchase program 2

(12)

12

Table 3: Panel regression results

Dependent variable: Loan-to-Earning Assets ratio. For bank i at time t

Variable (1) (2) (3) (4) Ln(QE) Govt (QE) Coop (QE) Crisis (QE) Ln(TA) Leverage ROA Ln(GDP/Cap) Interest Rate Reserves/Assets LLP-ratio Ln(LTD-ratio) NEA/Assets Constant 0.0006* (0.0003) -.0666** (0.0272) -.0261*** (0.0039) 0.3150 (0.3444) 0.0129 (0.0612) 3.0722*** (1.4030) 1.4855* (0.7976) 0.0004 (0.0004) -.0713*** (0.0268) -.0271*** (0.0039) 0.3017 (0.3537) -.0630 (0.0818) 3.2580** (1.6127) 2.3447** (0.9773) 0.0011** (0.0005) -.0011 (0.0007) -.0001 (0.0006) -.0008 (0.0006) -.0674** (0.0273) -.0259*** (0.0040) 0.5395 (0.4710) 0.0489 (0.0632) 3.4574** (1.5624) 0.0434 (0.8232) 0.0107** (0.0053) 0.0003 (0.1040) 0.1829* (0.1040) 1.1025 (0.8192) 0.0010* (0.0006) -.0010 (0.0007) -.0001 (0.0006) -.0007 (0.0006) -.0707*** (0.0265) -.0269*** (0.0038) 0.5163 (0.4765) -.0132 (0.0840) 3.5248** (1.7178) 0.0099 (0.8034) 0.0103* (0.0053) 0.0016 (0.0042) 0.1358 (0.0998) 1.8012* (0.9806) Fixed Effects (Regression) Fixed Effects (Year) Yes No Yes Yes Yes No Yes Yes Observations Groups R-Squared 872 175 0.1219 872 175 0.1511 872 175 0.1529 872 175 0.1744 This table presents the coefficient estimates of the full regression for the period 2012-2016. The value below the coefficients is the corresponding standard error. In columns 1-4 a fixed effects regression is run. Year fixed effects is assumed in column 2 and 4. ‘*’ indicates significance at a 10% level, ‘**’ indicates

(13)

13 significance at a 5% level, ‘***’ indicates significance at a 1% level. Full descriptions of the variables can be found in the appendix.

indicates a weaker rebalancing effect caused by QE for government owned banks, cooperative banks, and banks in nations hit by the European sovereign debt crisis. According to the

extended regression, the marginal effect of QE on the loan-to-earning assets ratio for government owned banks is 0.0000, for cooperative banks 0.0010 (0.0009 using year fixed effects), and 0.0003 for banks in crisis regions, compared to 0.0011 (0.0010 using year fixed effects) for private banks in non-crisis regions. However, these coefficients are not

significantly different from 0. The weaker effect of monetary policy transmission is also consistent with previous literature.

Table 4 provides the coefficients for the regression taking into account the assumption that rebalancing takes time and does not happen immediately. The results presented in table 4 show that QE has a larger and more significant rebalancing effect on the composition of earning assets over time. The coefficient is positive for all regressions, including those taking year fixed effects into account. The marginal effect of QE is most significant in the base regression without year fixed effects (0.0013), this effect is also significantly different from the value it had in table 3. The highest value of the coefficient is 0.0015 in the extended regression without year fixed effects. This implicates that the rebalancing effect is not an immediate effect, but rather a process that occurs over time. The extended regression also provides additional insight into ownership effects on QE. For government owned banks QE increases to loan-to-earning assets ratio more than for private banks over time (0.0018 and 0.0015 assuming year fixed effects). An explanation for this might be that government owned banks react more slowly to QE than private banks because of differences in the speed of decision making. The inflexibility of government banks would therefore explain that in table 3 the interaction for government ownership and QE is negative, compared to the positive coefficient in table 4. Furthermore, table 4 shows that the rebalancing effect for cooperative banks is weaker. The marginal rebalancing effect caused by QE is 0.0005 for cooperative banks (0.0001 when assuming year fixed effects). This might be explained through the effect previously mentioned in the literature review, where cooperative banks are less effected by monetary policy transmission because of a more secluded funding network. Additionally, the extended regression indicates that over time the rebalancing effect of QE for banks in crisis hit nations is 0.0002 (0.0003 when using year fixed effects) higher than for banks in other nations of the Eurozone.

(14)

14

Table 4: Lagged panel regression results

Dependent variable: Loan-to-Earning Assets ratio. For bank i at time t+1

Variable (1) (2) (3) (4) Ln(QE) Govt (QE) Coop (QE) Crisis (QE) Constant 0.0013*** (0.0003) 0.6808 (0.6073) 0.0010*** (0.0003) 1.2702 (0.8536) 0.0015*** (0.0005) 0.0003 (0.0006) -.0010* (0.0005) 0.0002 (0.0005) 1.0086 (0.6290) 0.0011** (0.0005) 0.0004 (0.0006) -.0010* (0.0005) 0.0003 (0.0005) 0.0263 (0.1044) Fixed Effects (Regression) Fixed Effects (Year) Control Variables (Base) Control Variables (Extended) Yes No Yes No Yes Yes Yes No Yes No Yes Yes Yes Yes Yes Yes Observations Groups R-Squared 697 175 0.0993 697 175 0.1049 697 175 0.1188 679 175 0.1265 This table presents the coefficient estimates of QE and the interactions of QE on the loan-to-earning assets ratio using data from 2015 to 2012. The loan-to-asset ratio take the values of 2016-2013. The value below the coefficients is the corresponding standard error. In column 1-4 a fixed effects regression is run. Year fixed effects are assumed in column 2 and 4. Column 1 and 2 present the base regression, whereas 3 and 4 present the extended regression. ‘*’ indicates significance at a 10% level, ‘**’ indicates significance at a 5% level, ‘***’ indicates significance at a 1% level. The total regression and a full description of the variables can be found in the appendix.

The regressions presented in Table 5 reveal that a 1% increase in QE has a stronger rebalancing effect for smaller banks (0.0021) compared to larger banks (0.0015). An explanation for this might be that smaller banks increasingly start producing financial

products related to QE, something they might have done in a lesser extent before. Through the purchases of securities and bonds with loans as collateral the ECB increased the money supplied to smaller banks to a relatively larger extend than to larger banks whom might have already been more active in the process of covering bonds and securitizing their loans. The rebalancing effect caused by QE is -0.0006 for large government owned banks.

(15)

15

Table 5: Panel regression results separated

Dependent variable: Loan-to-Earning Assets ratio. For bank i at time t

Total Assets < P25 Total Assets > P75

Variable (1) (2) (3) (4) Ln(QE) Govt (QE) Coop (QE) Crisis (QE) Constant 0.0021*** (0.0008) 0.0033** (0.0013) -.0005 (0.0009) -.0022** (0.0010) 2.8975 (3.1202) 0.0021** (0.0009) 0.0033** (0.0013) -.0006 (0.0008) -.0023** (0.0010) 4.0475 (5.7485) 0.0016** (0.0006) -.0022** (0.0010) 0.0016 (0.0018) -.0021*** (0.0007) 1.3531 (0.8904) 0.0011 (0.0007) -.0020** (0.0009) 0.0016 (0.0017) -.0019** (0.0007) 1.6115* (0.8942) Fixed Effects (Regression) Fixed Effects (Year) Control Variables (Base) Control Variables (Extended) Yes No Yes Yes Yes Yes Yes Yes Yes No Yes Yes Yes Yes Yes Yes Observations Groups R-Squared 217 44 0.3009 217 44 0.3089 215 43 0.3020 215 43 0.3602 This table presents the coefficient estimates for QE and the QE interactions on the loan-to-earning assets ratio for the years 2012-2016, for banks which have their size below the 25th_{percentile and banks which have their}

size above the 75th_{percentile. In columns 1-4 a fixed effects regression is run. Year fixed effects are assumed}

in columns 2 and 4. Columns 1-4 present the extended regression.‘*’ indicates significance at a 10% level, ‘**’ indicates significance at a 5% level, ‘***’ indicates significance at a 1% level. The value below the coefficients is the corresponding standard error. The total regression and a full description of the variables can be found in the appendix.

The lower rebalancing effect might be caused by less effective decision making for and by the fact that a few large government owned banks are owned by the state as a bailout measure. Therefore they might be more focused on recapitalizing rather than increasing lending. The high coefficient for small government banks might be caused by the fact that there is only 1 government bank in the small sample. The difference in coefficients for cooperative

ownership for small and large banks might be explained through the fact that only two cooperative banks are included in the larger banks group (Rabobank and Raiffeisen Zentral Bank). The two large cooperative banks also have more access to the broad financial market

(16)

16 compared to their smaller counterparts. Furthermore, table 5 shows how having main banking operations settled in a crisis hit nation lowers the rebalancing effect for both groups. The effect of QE on the loan-to-earning asset ratio is 0.0022 less for small banks in these regions, and 0.0021 less for large banks in these regions. This effect might be caused by increased exposure to the sovereign debt crisis and a more distressed local financial system. This reduction in the rebalancing effect might therefore be explained through the need to recapitalize rather than increase lending. Also, when taking a demand point of view, these banks might have had less opportunities to increase lending due to a lower demand for capital caused by recessionary forces.

Table 6 provides an overview of the separate APPs and the rebalancing effect each causes individually. Taking year fixed effects into account was impossible due to collinearity issues, therefore the choice has been made to table present immediate and lagged effects instead. Because CSPP has only been active in 2016 the program was omitted from the lagged regression. ABSPP causes an immediate rebalancing effect towards other earning assets, rather than lending through a decrease in the loan-to-earning assets ratio of 0.0013 per 1% increase of ABSPP. This effect might be caused by the condition the ABSPP only allows for ABS to be bought, making banks increase their ABS holdings, to later sell them to the ECB. The table implicates that when considering a lagged effect, ABSPP only minimally affects the composition of earning assets. The regression shows that for CBPP the immediate effects are relatively minor on the loans-to-earning assets ratio. However, when considering lagged effects CBPP2 and CBPP3 appear to increase the division of earning by the highest margin and the highest significance. A 1% increase in the CBPP programs results in a 0.0010 increase in the loan-to-earning asset ratio when considering the lag. A reason behind the lagged effects for ABSPP and CBPP2-3 might be that it takes time to produce these financial products, thereby explaining the lag. CSPP and PSPP are both influential and significant on the short term, increasing the loan-to-earning asset ratio by 0.0012 and 0.0013 respectively for every 1% increase in these programs. In the extended regression these numbers are 0.0011 for CSPP and 0.0012 for PSPP. This could be caused by the fact that banks hold bonds for companies and government agencies that are being purchased by the ECB. This causes an increase in currency available to banks without any further conditions to the money they receive. This extra cash would then be repurposed into lending rather than new bonds for which the ECB effectively lowered yields. The lagged effect for PSPP also shows that this program has a

(17)

17

Table 6: Panel regression results (Separate APPs)

Dependent variable: Loan-to-Earning Assets ratio.

Variable (1) (2) (3) (4) Ln(ABSPP) Ln(CBPP) Ln(CSPP) Ln(PSPP) Govt (QE) Coop (QE) Crisis (QE) Constant -.0013* (0.0007) 0.0004 (0.0004) 0.0012*** (0.0004) 0.0013*** (0.0004) 2.3447** (0.9773) 0.0002 (0.0007) 0.0010*** (0.0003) 0.0006* (0.0003) 1.2702 (0.8536) -.0013* (0.0010) 0.0010* (0.0006) 0.0011*** (0.0003) 0.0012*** (0.0003) -.0010 (0.0007) -.0001 (0.0006) -.0007 (0.0006) 1.8012* (0.9806) 0.0004 (0.0011) 0.0011** (0.0005) 0.0006* (0.0003) 0.0004 (0.0006) -.0010* (0.0005) 0.0003 (0.0005) 1.7970** (0.8857) Fixed Effects (Regression) Lagged Effects (Regression) Control Variables (Base) Control Variables (Extended) Yes No Yes No Yes Yes Yes No Yes No Yes Yes Yes Yes Yes Yes Observations Groups R-Squared 872 175 0.1511 697 175 0.1049 872 175 0.1744 697 175 0.1265 This regression presents the coefficient estimates for the separate APPs and the interactions with QE on the loan-to-earning asset ratio using data from 2012-2016. The table also shows the lagged effects of the APPs and the interactions with QE on the loan-to-earning asset ratio using data from 2012-2015 for the independent variables and data from 2013-2016 for the dependent variable. In column 1-4 a fixed effects regression is run. Column 2 and 4 show lagged effects. Column 1 and 2 present the base regression and 3 and 4 present the extended regression. ‘*’ indicates significance at a 10% level, ‘**’ indicates significance at a 5% level, ‘***’ indicates significance at a 1% level. The value below the coefficients is the corresponding standard error. The total regression and a full description of the variables can be found in the appendix.

weaker delayed effect. This could be caused by the fact that bonds from banks themselves are not eligible for the program and that through the narrow transmission channel banks are less affected.

(18)

18 Throughout the majority of the regressions it appears that the most determining factors for the composition of earning assets, besides QE, are a bank’s size, the leverage-ratio, the average interest they receive on their loans, and, to a lesser extent, the quality of the loans.

5. Conclusion

Through the empirical study the following conclusions can be made about the effect of QE on the composition of earning assets. First, QE causes a rebalancing effect towards loans from other earning assets. Secondly, QE operates through a mechanism causing the effects to be delayed. Thirdly, the effects of QE are different for private, government owned, and cooperative banks and also change with size of banks. Fourth, PSPP and CSPP cause an immediate rebalancing towards lending, whereas ABSPP causes and immediate rebalancing towards other financial activities. Through mechanisms CBPP2 and CBPP3 generate a large rebalancing towards lending over time. In line with the hypotheses made in the literature review this research provides further evidence that QE increased the loan-to-earning assets ratio for banks in the Eurozone. Furthermore, the hypothesis that the effects of QE differ across various types of banks in different Eurozone countries also proves to be true. Also, this research provides evidence of the separate APPs influencing the composition of earning assets through separate mechanisms.

The research enhances the existing literature by providing an extended view into the rebalancing effects of QE for various types of banks, operating in different economic conditions. The research also provides evidence of varying operating mechanisms given the asset for purchase under QE. Furthermore, through the empirical analysis this thesis might contribute to the debate of effective monetary policy and on how to optimize other large scale asset purchase programs, by targeting specific types of banks or assets where the effects are of increased value. Most important of this research is that through these new insights the

discussion about the unconventional tool called QE has become easier to understand, by providing both advantages and disadvantages of banks, bank ownership, and bank holdings, to the efficiency of QE. Further research might also look into the effects of QE on different types of lending, where this research takes lending as a whole other papers might use C&I lending, mortgages, and other consumer and retail loans as separate dependent variables, as the effects of QE might differ across specific types of loans. Extended research on QE in crisis hit nations might also be of interest to study effective monetary policy in case a similar situation occurs again.

(19)

19 A limit of this research might be the focus on the lending side of earning assets rather than the total, as there are more loan related variables controlled for compared to variables related to other earning assets. Also, this analysis suggests that QE is available to all banks in the sample and that every bank is exposed to the program in a similar manner, extended research could be done by looking at the effects of QE for banks with higher or lower

exposure levels to QE. Another limit of this paper is in analyzing the rebalancing effect of QE for small government owned banks and large cooperatives, since their results are limited by a low number of observations.

(20)

20

Appendix

The following table provides an overview of the individual correlations of the bank characteristics used as control variables. The table shows that the highest collinearity is between leverage and interest rate. However, the collinearity is reasonable for both variables to be included.

Correlations: Bank Characteristics

LtEA lTA Lev. ROA IR R/A LLP lLTD NEA/A

LtEA lTA Lev. ROA IR R/A LLP lLTD NEA/A 1.0000 -.0377 -.1732 -.0575 -.1511 -.1089 0.0686 0.1401 -.1851 1.0000 -.2449 -.0271 -.1767 -.0955 0.1055 0.1709 0.0208 1.0000 -.0968 0.6685 0.2571 -.0155 0.0102 0.1529 1.0000 0.0183 -.0837 -.5409 -.0144 -.0086 1.0000 0.0915 -.0457 -.1230 0.1302 1.0000 -.0612 0.0430 0.1102 1.0000 0.0589 0.0369 1.0000 -.1218 1.0000

The table below shows the variables used in the regressions and how they were calculated. The ROA and LLP-ratio were directly provided by the Orbis database. The other variables were all transformed to their functional form and ratios through Stata. The dummies were generated by analyzing the GUOs of banks and the country they are settled.

(21)

21

Variable descriptions

Loan-to-Earning Assets ratio Ln(QE) Government Ownership Cooperative Ownership Crisis Nation Ln(TA) Leverage-ratio ROA Ln(GPD/Capita) Interest rate Reserve-ratio LLP-ratio Ln(LTD-ratio) NEA-ratio

Loans / Total Earning Assets

Log of total yearly purchases under ABSPP, CBPP2, CBPP3, CSPP, and PSPP

1 = GUO being a government agency 1 = GUO being a cooperative

1 = GUO located in Greece, Ireland, Italy, Portugal, or Spain.

Log of total assets Equity / Liabilities

Earnings Before Interest and Taxes / Total Assets Log of GDP per capita

Net Interest Income / Loans Reserves / Total Assets

Loss Loan Provision / Net Interest Income Log of loans / deposits.

Non-Earning Assets / Total Assets

All bank specific data are obtained through Orbis and formatted through global reporting templates by Fitch.

(22)

22

References

Andries, N., & Billon, S. (2010). The Effect of Bank Ownership and Deposit Insurance on Monetary Policy Transmission. The Journal of Banking and Finance 34(12), pp. 3050-3054.

Bernanke, B.S., & Blinder, A.S. (1992). The Federal Funds Rate and Channels of Monetary Transmission. American Economic Review 82, pp. 901-921.

Butt, N., Domit, S., Kirkham, L., McLeay, M., & Thomas, R. (2013). What can Money Data Tell us about the Impact of QE? Bank of England Quarterly Bulletin 2012-Q4.

Christensen, J.H.E., & Rudebusch, G.D. (2012). The Response of Interest Rates to US and UK Quantitative Easing. The Economic Journal 122(564), pp. F385-F414.

Darmouni, O.M., & Rodnyansky, A. (2017). The Effects of Quantitative Easing on Bank Lending Behavior. The Review of Financial Studies 30(11), pp. 3858-3887. Dewally, M., & Shao, Y. (2014). Liquidity Crisis, Relationship Lending, and Corporate

Finance. Journal of Banking and Finance 39, pp. 223-239.

Ehrmann, M., & Worms, A. (2004). Bank Networks and Monetary Policy Transmission.

Journal of the European Economic Association 2(6), pp. 1148-1171.

European Central Bank (2013). Optimal Asset Structure of a Bank: Bank Reactions to

Stressful Market Conditions (Working Papers Series No. 1533). Retrieved from the

ECB website: http://www.ecb.europa.eu/pub/pdf/scpwps/ecbwp1533.pdf European Central Bank (2015). The Transmission of the ECB’s Recent Non-Standard

Monetary Policy Measures. ECB Economic Bulletin 7-2015, pp. 1-20.

Jouvanceau, V. (2016). The Portfolio Rebalancing Channel of Quantitative Easing. (GATE Working Papers Series No. 1625). Retrieved from EconPapers website:

ftp://ftp.gate.cnrs.fr/RePEc/2016/1625.pdf

Joyce, M., Lasaosa, A., Stevens, I., & Tong, M. (2010). The Financial Market Impact of Quantitative Easing in the United Kingdom. International Journal of Central Banking

7(3), pp. 113-161.

Kakes, J., & Sturm, J.E. (2002). Monetary Policy and Bank Lending: Evidence from German Banking Groups. Journal of Banking and Finance 26(11), pp. 2077-2092.

Kashyap, A.K, & Stein, J.C. (2000). What Do a Million Observations on Banks Say About the Transmission of Monetary Policy? American Economic Review 90(3), pp. 407- 428.

(23)

23 Channel. Journal of Money, Credit, and Banking 32(1), pp. 121-141.

Krishnamurthy, A., & Vissing-Jorgensen, A. (2011). The Effects of Quantitative Easing on

Interest Rates: Channels and Implications for Policy (NBER Working Paper No.

17555). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w17555.

Krishnamurthy, A., & Vissing-Jorgensen, A. (2013). The Ins and Outs of LSAPs.

Proceedings – Economic Policy Symposium – Jackson Hole, Federal Reserve Bank of

Kansas City.

Lane, P.R. (2012). The European Sovereign Debt Crisis. Journal of Economic Perspectives

26(3), pp. 49-68.

Müller, M.O., Schwaninger, M., & Vithessonthy, C. (2017). Monetary Policy, Bank Lending and Corporate Investment. International Review of Financial Analysis 50, pp. 129-142.

Reichlin, L. (2014). Monetary Policy and Banks in the Euro Area: The Tale of Two Crises.

Journal of Macroeconomics 39, pp. 387-400.

Tobin, J. (1958). Liquidity Preference as Behavior Towards Risk. The Review of Economic

Studies 25(2), pp. 65-86.

Ugai, H. (2006). Effects of the Quantitative Easing Policy: A Survey of Empirical Analyses (Bank of Japan Working Paper Series No. 06-E-10). Retrieved from EconPapers website: https://econpapers.repec.org/paper/bojbojwps/06-e-10.html