THE IMPACT OF UNCONVENTIONAL MONETARY POLICY ON SYSTEMIC RISK IN THE EUROZONE BANKING SECTOR

THE IMPACT OF UNCONVENTIONAL

MONETARY POLICY ON SYSTEMIC

RISK IN THE EUROZONE BANKING

SECTOR

Master Thesis Economics

Nienke Oudenampsen

n.oudenampsen@student.rug.nl | S2756390

1.

I

NTRODUCTION

The Great Recession showed the world that contagion of financial institutions during a crisis is severe. Especially the collapse of Lehman Brothers indicated the interconnectedness between banks and showed that a failure of one bank can have a severe impact on the whole financial system. Due to globalization of financial markets which already initiated in the 1980s, interconnectedness became considerably stronger. Although globalization led to efficiency gains, risk in the system amplified significantly (Engle et al., 2014). Currently, undercapitalized financial institutions can still impose major negative externalities on the real economy (Brownlees and Engle, 2016). Furthermore, in times of economic downturn, other financial institutions cannot absorb the bankruptcy of another firm. As a consequence, contagion and amplification effects occur, making capital shortfalls harmful for the whole economy. Increasing financial stability by reducing risks faced by financial institutions, called systemic risk, have become more and more important (Galati and Moessner, 2013).

The International Monetary Fund (2009) defines systemic risk as: “A risk of disruption to

financial services that is caused by an impairment of all or parts of the financial system and that has the potential to cause serious negative consequences for the real economy”. After the

systemic risk in the financial system at that period. Brownlees and Engle (2016) find that SRISK is a strong predictor of the capital injections done by the Fed to combat the crisis. Consequently,

SRISK is an adequate estimator for pre-crisis warning signalling.

The crisis of 2007-2009 and the years after, had an enormous impact on the economy and changed the environment for financial institutions deeply. The risk in the financial system increased significantly. Besides, the crisis indicated that the policies present at that time were not sufficient to help financial institutions and stimulate the economy. During normal (cyclical) crises, conventional monetary policy (CMP) was sufficient. However, with the interest rate at the zero lower bound and the economy still in crisis, central banks switched to forms of unconventional monetary policy (UMP). Central banks used different forms of UMP to combat the crisis. Numerous studies evaluate the effectiveness and impact of these policies on the macro economy, but concrete evidence about the bank level impact of UMP lacks (Avalos and Mamatzakis, 2018). Furthermore, existing studies have not yet determined a link between different forms of UMP and systemic risk in the Eurozone banking sector.

Although there is limited research on the effects of UMP on financial stability, some studies focus on the impact of UMP on individual banks. While UMP contributes to stabilizing the economy after the crisis, some forms of UMP, especially negative interest rate policy, lead to a higher risk-taking of banks (Chodorow-Reich, 2014; Heider et al., 2019; Jobst and Lin, 2016). Furthermore, although UMP improves the overall bank resilience, it also increases the fragility of some banks (Avalos and Mamatzakis, 2018). Consequently, higher risk-taking and an increase in the fragility of banks possibly increases systemic risk in the financial system.

three subsamples of banks and three different forms of UMP: longer term refinancing operations (LTROs), asset purchasing programs (APPs) and negative interest rate policy (NIRP). For the period this thesis analyses, 2006 till 2018, I do not find statistical evidence for the impact of different forms of UMP on systemic risk in all different subsamples.

This thesis contributes to the existing literature in linking the effects of UMP to financial stability and system wide risk effects. Furthermore, this thesis aims to investigate whether the impact varies across subsamples of banks, by including an examining on global systemic important banks and non-systemic important banks. Since current literature mainly focusses on investigating systemic risk in the United States, this thesis contributes to the existing knowledge by investigating systemic risk in Europe. Furthermore, previous research mainly uses systemic risk measures as pre-warning signals for a crisis, this thesis contributes by investigating policy effects on these systemic risk measures.

This thesis continues as follows; the next section provides an overview of the existing literature about the subject. Consequently, it introduces the hypotheses. Section 3 introduces different measures for systemic risk. Additionally, it introduces the econometric model and methodology this thesis uses. Section 4 presents the data and describes the summary statistics, I discuss data limitations and tests the models for heteroscedasticity and serial correlation. In the fifth section, I present the results of my analysis. Lastly, Section 6 discusses the main conclusions and possible limitations of this study.

2.

T

HEORETICAL FRAMEWORK

stability. Due to these asset bubbles, excessive credit creation developed, which ultimately resulted in the financial crisis (De Grauwe, 2018; Joyce et al., 2012). This financial crisis, the Great Recession, resulted in the need for new policies, referred to as ‘unconventional’. The purpose of CMP was to attain a low and stable rate of inflation, through altering the short-term nominal interest rate. However, when the effective lower bound on the short-term nominal interest rate was hit, CMP was no longer effective. Central banks switched to alternatives in an attempt to stimulate the economy, hence, in the years after the Great Recession central banks introduced different forms of UMP. Whereas CMP mainly used the interest rate channel, the new policies stimulated the economy through different channels, e.g. the portfolio rebalancing channel and the liquidity channel. Besides, the interdependencies between banks and the risk of a collapse of the entire financial system showed the importance of the term ‘systemic risk’. The severity of the crisis indicated the need of prudential rules. Regulation became tighter and both micro- and macroprudential rules sharpened (Galati and Moessner, 2013).

The current debate acknowledges the friction between price stability and financial stability. The Dutch Central Bank (2019) explains the conflict between macroprudential policy and UMP in their semi-annual financial stability report. According to this report, the prolonged period of UMP is effective in keeping inflation low, however, the risks for financial stability increases. Macro prudential policy is effective in stabilizing banks to a certain extent, however, policy makers should consider the possible side effects of this prolonged period of expansionary monetary policy, e.g. an increase in risk-taking of financial investors and the possibility of asset bubble creation.

Different studies define the concept and elements of systemic risk. According to Benoit et al. (2017), systemic risk is “The risk that many market participants are simultaneously affected by

severe losses, which then spread through the system”. In their paper, they distinguish different

sources of systemic risk, including systemic risk-taking, contagion mechanisms and amplification mechanism. Furthermore, the paper describes systemic taking as the risk-taking of financial institutions, who choose to be exposed to a higher risk, which may reinforce amplification mechanisms and additionally may lead to contagion of other institutions. According to them, contagion is the phenomenon when a loss of a financial institution spill over to other financial institutions. Amplification mechanisms describes the event when a considerable small shock have substantial large impact. UMP can affect systemic risk of banks through different channels, which can affect all different parts of systemic risk. The most important channel is systemic risk-taking channel. Correspondingly, Colletaz et al. (2018) investigates this systemic risk-taking channel. They examine the impact of monetary policy on long-run systemic risk-taking. Risk perception and tolerance of banks and financial institutions change if central banks keep the interest rate too low for a longer period. The build-up of risk is a long process. Some risk-taking channels have an incubation period, which makes long-term causality effects more important. They investigate the period 2000-2008, and find a causality between monetary policy and systemic risk in the long run. The period they analyse is mainly pre-crisis. In this period central banks primarily uses CMP by altering the short-term interest rate. Whether the negative interest rates even stronger impact systemic risk-taking is still ambiguous.

the ECB used several measures in an attempt to improve long-term liquidity through the liquidity channel, e.g. LTROs with fixed-rate full allotment (FRFA) tender procedure. These programs allowed for lending from the central bank with a longer maturity, for a fixed rate tender procedure and with full allotment. The bank executed different rounds of APPs, including all forms of quantitative easing (QE), e.g. asset backed securities purchasing program (ABSPP), but also credit easing programs (CE), e.g. covered bond purchasing program (CBPP1; CBPP2; CBPP3). Those policies worked through the bank lending and the portfolio rebalancing channel. Also, the ECB tried to stimulate the economy through the interest rate channel, with allowing for very low and even negative interest rates (NIRP). The paper concludes that UMP enhances the overall bank resilience in the euro-area. However, the effect varies across subsamples of banks. For countries highly affected by the financial crisis, the loss absorbing buffers are weaker, and the fragility of these institutions increases. Chodorow-Reich (2014) investigates the impact of UMP on financial institutions in the United States. The paper states that UMP affects the real economy not only directly, but also indirectly through financial institutions, which makes investigating the impact of UMP on financial institutions important. The paper draws three conclusions. First, UMP was specifically important for life insurers during the first period, the winter of 2008-2009, since it helped recovering their balance sheets. Second, for the money market funds and the pension funds, a low interest rate environment stimulated risk-taking through the ‘reach for yield’ opportunity, but only for the period 2009-2011. Third, both banks and insurers benefit from UMP in 2013, during that period there was no trade-off between UMP and financial stability. Overall, he concludes that UMP supports the stabilization of some sectors of the economy, e.g. life insurance companies, but triggers risk-taking behaviour in other sectors, e.g. pension funds.

funding costs and gives a boost to the assets prices. However, there is a substantial probability of a negative effect on bank’s profitability in the future. Since the expected profitability of banks influences the bank lending channel, it is likely that a further decline of the interest rate diminish returns for banks. The profitability decline can constrain credit expansion of banks. Heider et al. (2019) study the impact of NIRP on bank lending in the Eurozone. They highlight the importance of banks’ funding structure, and find that risk-taking is larger for weakly capitalized banks. High-deposit banks reduce lending, but lend to riskier borrowers. Low-deposit banks attract safer borrowers. This higher risk-taking behaviour can deteriorate financial stability. Furthermore, the paper suggests that it is interesting to take a closer look at heterogeneity of banks and how the impact of UMP changes for different structured banks. Besides lower profitability and higher risk-taking of banks, low interest rates environments encourage a larger endogenous risk build-up, and can result in a larger systemic risk (Brunnermeier and Sannikov, 2014). The economy will always face instability because leverage and risk-taking are endogenous in the financial system. There exists an amplification loop between aggregate risk and equilibrium leverage. Hence, when prudential tools want to manage risk, this can possibly increase systemic risk.

investigates heterogeneity of banks in responses to monetary policy announcements, and examines the main objectives of this heterogeneity. Ricci (2015) concludes that the impact of monetary policy shocks is stronger for banks with weaker balance sheets and banks with higher risk. Furthermore, he concludes that systemic risk in a banking system is an important factor that determines the heterogeneous response of banks to a monetary policy announcement.

H

Taking into consideration the theoretical framework, this section introduces the hypotheses. The main objective of this thesis is to identify whether UMP has an impact on systemic risk in the Eurozone banking sector. The main question of this research is: ‘What is the impact of

unconventional monetary policy on systemic risk in the Eurozone banking sector?’. The

hypotheses aim to answer the main question as accurate as possible. Because of the scope of this research, this thesis focusses only on three key forms of UMP. These three policies impact systemic risk through three different channels. The first channel, the liquidity channel, includes the LTROs with FRFA. The second channel, the portfolio rebalancing channel, includes the different forms of APPs. The third channel, the risk-taking channel, includes NIRP. Although I distinguish between different channels, this does not indicate that these policies only impact systemic risk through this specific channel. Policies might overlap in time and it is possible that some policies impact through other channels, or through different channels at the same time. Table A.1, A.2 and A.3 in Appendix A provides the key policies of the ECB between 2006-2018.

central bank against a fixed interest rate with unlimited provision, hence full allotment. Darracq-Paries and De Santis (2015) concludes that LTROs with FRFA ease credit friction for banks, and make lending easier. Easing lending could increase the riskiness of loans, and therefore raises credit risk (Ciccarelli et al., 2013). A higher credit risk can result in a higher systemic risk for banks. The first hypothesis is as follows:

Hypothesis 1: Liquidity provision operation programs i.e. LTROs combined with FRFA tender procedures increase systemic risk in the banking sector.

Note that refinancing operations are also a form of CMP, only with a shorter maturity. To examine the impact of the longer maturities, this thesis investigates LTROs with one-year maturity, three-year maturity and the two targeted longer term refinancing operation (TLTRO) programs with a maximum maturity of four years.

The second channel, the portfolio rebalancing channel, comprises multiple APPs, including CBPP, ABSPP, the public sector purchasing program (PSPP) and the corporate sector purchase programme (CSPP). According to Kandrac and Schlusche (2017), QE affects bank balance sheet through portfolio substitution channel, the total share of riskier loans increase due to QE. Both interest rates and risk spreads lower due to QE, therefore QE increases risk-taking (Van den End, 2016). Furthermore, QE increases the risk of an asset price bubble. Overall, this increase in risk-taking and the possibility of asset price bubbles due to QE might result in an increase of systemic risk in the banking sector. Therefore, the second hypothesis is as follows:

Hypothesis 2: Multiple asset purchasing programs increase systemic risk in the banking sector.

risk in the banking system (Brunnermeier and Sannikov, 2014). The third hypothesis is as follows:

Hypothesis 3: NIRP increases the risk-taking behaviour of banks and therefore increases systemic risk in the banking sector.

The last hypothesis investigates heterogeneity between banks and to what extent this heterogeneity results in different impact of UMP on systemic risk. Only large banks contribute substantially to systemic risk. To investigate impact on different banks, I distinguish between a sample of global systemic important banks (GSIBs) and compare this to a sample of non-systemic important banks (non-GSIBs). According to Brissimis and Delis (2010), banks with more market power, higher liquidity and more capital are less affected by monetary policy. Additionally, Ricci (2015) finds that the impact of monetary policy is higher for weaker banks. Hence, I expect a stronger impact of UMP on non-GSIBs. The fourth hypothesis is as follows:

Hypothesis 4: The impact of UMP on systemic risk of non-GSIBs is larger than the impact of UMP on systemic risk of GSIBs.

To test the fourth hypothesis, I divide the sample of banks into subsamples, the total sample, GSIBs, and non-GSIBs. Next, I compare the different subsamples and test for significant differences in the models.

3.

M

ETHODOLOGY

Numerous researches introduce measures for systemic risk. Girardi and Ergün (2013) present Value-at-Risk (VaR), which is a measure financial institutions use to quantify the risk of an institution. One major drawback of this risk measure is that it only considers the risk of an institution in isolation. Besides presenting VaR, Girardi and Ergün (2013) elaborate on a measure by Adrian and Brunnermeier (2011), the conditional Value-at-Risk (CoVaR). CoVar investigates the returns of the financial system by comparing periods of distress with periods in median state. Additionally, Girardi and Ergün (2013) discuss the Marginal Expected Shortfall (MES). In contrast to the CoVaR, MES investigates the returns of an institution when the financial system as a whole is in a period of distress. Acharya et al. (2012) and Brownlees and Engle (2016) extend on the MES model by adding the size and the liabilities of the firm. They introduce SRISK, which measures the expected capital shortfall of a financial institution during a period of severe market decline. It includes leverage, size, and the Long Run Marginal Expected Shortfall (LRMES).

The SRISK measure looks as follows:

(1) 𝑆𝑅𝐼𝑆𝐾_%& = 𝑘𝐷_%&− (1 − 𝑘)𝑊_%&(1 − 𝐿𝑅𝑀𝐸𝑆_%&)

Where D denotes the book value of debt, W the market value of equity, and LRMES the long run marginal expected shortfall, all for bank i and time t. Furthermore, k denotes the prudential capital fraction, according to Engle et al. (2014), this prudential capital fraction in Europe must be set at 5.5%, in line with the Basel III Accords.

For this measure, first the LRMES is calculated. A market index is set up, RM, to represent the whole market, besides an individual bank is represented by a return index, Ri. When the market index falls by 40% semi-annual, it is considered as a crisis. The MES is calculated as the expected return of an individual bank when the overall market return decreases with 2%. This is calculated as follows:

(2) 𝑀𝐸𝑆_%& = −𝐸_&2𝑅_%,&456𝑅_7,&45≤ −2%;

In this formula, 𝑅_%,&45is the market return of bank i, and 𝑅_7,&45 the market return index for the total market. The average of the firm’s loss of equity value during this crisis period is called the

LRMES. The LRMES is calculated as follows:1

(3) 𝐿𝑅𝑀𝐸𝑆_%& = 1 − 𝑒𝑥𝑝(−18 ∗ 𝑀𝐸𝑆_%&)

This SRISK measure can be used for a ranking of systemically risky institutions. The highest

SRISK ratio is the bank which is the largest contributor, in terms of undercapitalization, of the

financial system, during a period of financial distress. The total systemic risk in the system is

1 _{Note that the parameter in this formula, 18, is estimated with the extreme value theorem.}

the sum of all the individual firm’s SRISKs. Additionally, the sum is the total amount of capital that the government need for a bailout of the whole financial system during a crisis.

To investigate the impact of UMP on systemic risk, this paper uses the sum of all the individual banks’ SRISK.

(4) 𝑆𝑅𝐼𝑆𝐾& = ∑B%C5𝑆𝑅𝐼𝑆𝐾%,&

Brownlees and Engle (2016) state that negative values of capital shortfalls (capital surpluses) can be ignored, because during a crisis it is highly unlikely that financial institutions utilizes capital surplus for mergers or loans. However, Laeven et al. (2016), state that it does make sense to consider negative values of SRISK. These negative values, capital surpluses, can absorb systemic shocks. Highly capitalized banks with a negative SRISK, can help the undercapitalized banks in the system. Though, during the crisis interbank lending decreases because of uncertainty issues, and central banks replaced the interbank market (Giannone et al., 2012). Taking into consideration the sample period in this thesis, which is for a considerable time a crisis period, I agree with the statement by Brownlees and Engle (2016) and ignore negative values of SRISK.

This thesis uses an approach by Ehrmann et al. (2019) to measure a macroeconomic shock. Ehrmann et al. (2019) calculate the measure by the difference between the actual value and the expected value of a macroeconomic indicator. This thesis uses inflation (𝜋_&F_{) and expected}

inflation (𝜋_&G_{), to capture a macroeconomic shock, in this case inflation uncertainty. The}

standard deviation of the actual inflation is added to stabilize the measure. The model by Ehrmann et al. (2019) is as follows:

(5) 𝑥_&= HIJKHIL

I use this macroeconomic shock indicator to analyse interactive effects with an introduction of a form of UMP. When the difference between the two values is different from zero, there is a macroeconomic shock. This macroeconomic shock indicator can test whether the introduction of forms of UMP, i.e. a macroeconomic shock, impacts systemic risk.

3.1

R

To investigate the impact of UMP on SRISK, the following model is used:

(6) 𝑆𝑅𝐼𝑆𝐾& = 𝛽5+ 𝛽Q∗ 𝑥&+ 𝛽R∗ 𝐷&+ 𝛽S∗ 𝑥&∗ 𝐷&+ 𝛽T∗ 𝑉𝑆𝑇&+ 𝜀&

In this model, 𝛽₅ is a constant, 𝑥_& is the macroeconomic shock and 𝐷_& is a dummy for the active policy. A control variable is added to control for the overall volatility in the market, VSTOXX (VSTt), which is the volatility index of the EuroStoxx. Furthermore, an error term 𝜀& is added.

In this research three forms of UMP are tested, hence, different sets of dummies are used. Furthermore, this research tests for differences between GSIBs and non-GSIBs. Therefore, each model is tested for three different subsamples. For all three forms of UMP two situations are distinguished:

(7) [𝑆𝑅𝐼𝑆𝐾_&|𝐷& = 0] = 𝛽5+ 𝛽Q∗ 𝑥&+ 𝛽T ∗ 𝑉𝑆𝑇&+ 𝜀&

(8) [𝑆𝑅𝐼𝑆𝐾_&|𝐷& = 1] = (𝛽5 + 𝛽R) + 𝑥&∗ (𝛽Q+ 𝛽S) + 𝛽T∗ 𝑉𝑆𝑇&+ 𝜀&

In this model, the overall impact of a macroeconomic shock on systemic risk is presented by 𝛽_Q. For periods when a new form of UMP is active, the value of the dummy is one. The value of coefficient 𝛽_S represents the impact of the macroeconomic shock, during the period when the new form of UMP is active, on systemic risk. Following the hypotheses, I expect that 𝛽S is

larger than zero for all policies. Furthermore, I expect that 𝛽S is larger for non-GSIBs than for

3.2

E

:

Different policies are executed at different times in the period analysed, hence some policies might overlap in time. In 2014, LTROs and APPs are combined with NIRP, therefore it can be interesting to investigate the interaction of two policies. The following interaction models are used:

(9) 𝑆𝑅𝐼𝑆𝐾 _& = 𝛽_\+ 𝛽_] ∗ 𝑥_&+ 𝛽_^∗ 𝐿𝑇𝑅𝑂_&+ 𝛽_`∗ 𝑥_&∗ 𝐿𝑇𝑅𝑂 + 𝛽_5a ∗ 𝑁𝐼𝑅𝑃_&+ 𝛽₅₅∗ 𝑥_&∗ 𝑁𝐼𝑅𝑃_&+ 𝛽_5Q ∗ 𝐿𝑇𝑅𝑂_&∗ 𝑁𝐼𝑅𝑃_&+ 𝛽_5R ∗ 𝑥_&∗ 𝐿𝑇𝑅𝑂_&∗ 𝑁𝐼𝑅𝑃_&+ 𝛽_5S∗ 𝑉𝑆𝑇_&+ 𝜀_&

(10) 𝑆𝑅𝐼𝑆𝐾 & = 𝛽5T+ 𝛽5\ ∗ 𝑥&+ 𝛽5] ∗ 𝐴𝑃𝑃&+ 𝛽5^∗ 𝑥&∗ 𝐴𝑃𝑃&+ 𝛽5`∗ 𝑁𝐼𝑅𝑃&+ 𝛽Qa ∗

𝑥&∗ 𝑁𝐼𝑅𝑃&+ 𝛽Q5 ∗ 𝐴𝑃𝑃&∗ 𝑁𝐼𝑅𝑃&+ 𝛽QQ ∗ 𝑥&∗ 𝐴𝑃𝑃&∗ 𝑁𝐼𝑅𝑃&+ 𝛽QR ∗ 𝑉𝑆𝑇&+ 𝜀&

These models are simplified as follows:

(11) [𝑆𝑅𝐼𝑆𝐾&|𝑁𝐼𝑅𝑃& = 1; 𝐿𝑇𝑅𝑂 = 1] = (𝛽\+ 𝛽^+ 𝛽5a+ 𝛽5Q) + 𝑥&∗ (𝛽]+ 𝛽`+ 𝛽55+

𝛽5R) + 𝛽5S ∗ 𝑉𝑆𝑇&+ 𝜀&

(12) [𝑆𝑅𝐼𝑆𝐾_&|𝑁𝐼𝑅𝑃_& = 1; 𝐴𝑃𝑃 = 1] = (𝛽_5T + 𝛽_5] + 𝛽_5`+ 𝛽_Q5) + 𝑥_&∗ (𝛽_5\+ 𝛽_5^ + 𝛽_Qa+ 𝛽_QQ) + 𝛽_QR∗ 𝑉𝑆𝑇_&+ 𝜀_&

4.

D

ATA

The impact of different forms of UMP on systemic risk in the Eurozone banking sector is investigated in this thesis. The period analysed is 2006 till 2018, to include pre-crisis, crisis and post crisis periods. Data about systemic risk of banks is retrieved from the Stern-NYU’s V-Lab, which calculates the daily SRISK for all significant important institutions in USD. In 2006, the following countries were member of the Eurozone: Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, and Spain. A set banks with its headquarter in these countries are selected, the full list of 24 banks included in this analysis is shown in Appendix B. Also, in Table B.1 the GSIBs are indicated. Note that for Luxembourg there are no banks in the list, since there are no active significant important banks in this country to analyse. Furthermore, smaller banks and banks which are (partly) government owned are excluded from the list. To balance the data set, only banks for which data is available for the whole period are analysed. The sum of the daily positive SRISK of the selected banks is calculated to represent the total daily systemic risk of this sample of banks.

As a control variable, the VSTOXX index (𝑉𝑆𝑇_&) is used. This is a daily volatility index based on the Eurostoxx 50. In this thesis, it is used to capture the overall volatility in the European market.

4.1

S

Table 1 presents the descriptive statistics of the variables in this analysis. I find that SRISKGSIB on average accounts for 72% of the total amount of SRISKt in the sample. Hence, these six GSIBs account for approximately 72% of all systemic risk in this sample. Graph 1 shows this major share of GSIBs as well. The macroeconomic shock indicator (𝑥_&) uses the standard deviation of actual inflation to stabilize the indicator, this standard deviation is slightly larger than one. Furthermore, the macroeconomic shock indicator is on average somewhat above zero, implying that the actual inflation is on average slightly higher than the expected inflation during the period I analyse. Additionally, Graph 2 represents the macroeconomic shock indicator over the whole period. The macroeconomic shock indicator highly fluctuates over time. The 𝑉𝑆𝑇&

variable is on average around 23 index points. The maximum of this 𝑉𝑆𝑇& variable was during

the crisis; at that day the value was over 87 index points.

Table 1: Descriptive statistics of the variables in the Eurozone in the period 2006-2018

N Mean Med. Max. Min. St. Dev.

Graph 1:Total SRISKt for the different samples of banks during period 2006-2018, in million USD

The tables in Appendix C show the correlation matrices for the different samples of banks. These tables are useful to check whether the correlations between variables are as expected. As expected, there is a relatively high and positive correlation between the control variable 𝑉𝑆𝑇_& and SRISKt, implying that if the volatility in the market is high, systemic risk is also high. Furthermore, for the macroeconomic shock indicator, the relation is negative, relatively low, and not significant. When comparing the different tables, I conclude that the correlations are approximately similar for the different samples.

To test for stationarity, I perform the Augmented Dickey-Fuller (ADF) test, which tests the null hypothesis of the existence of a unit root against the alternative hypothesis of no unit root. Appendix D (Table D.1) provides the test results of this test. SRISKt, SRISKGSIB,t and

SRISKnonGSIB, t are non-stationary and integrated of order I(1). The other two variables,𝑥_& and 𝑉𝑆𝑇&, are stationary. To match with the other variables, I transform SRISKt by taking the natural logarithm. Furthermore, I take the first difference of this value. Now, SRISKt is as follows:

(13) ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_& = (𝐿𝑜𝑔(𝑆𝑅𝐼𝑆𝐾_&) − 𝐿𝑜𝑔(𝑆𝑅𝐼𝑆𝐾_&(−1))) ∗ 100

Consequently, I transform Model (6) as follows:

(14) ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_& = 𝛽₅ + 𝛽_Q∗ 𝑥_&+ 𝛽_R∗ 𝐷_&+ 𝛽_S ∗ 𝑥_&∗ 𝐷_&+ 𝛽_T ∗ 𝑉𝑆𝑇_&+ 𝜀_&

All ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_& series are stationary. To deal with potential outliers in these new series, I remove the top and bottom 0.5% from those series by winsorizing. Appendix E shows graphs for all ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾& samples (Graph E.1; E.2; E.3). Furthermore, Table E.1 provides descriptive

statistics of these series. Table E.2, E.3 and E.4 provides correlation matrices with ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾&.

dependent variable decreases. Furthermore, the correlation between 𝑥_& and the dependent variable switch sign, however these values are insignificant.

Table F.1 in Appendix F describes the most important announcement shocks of the ECB in the period this thesis analyses. Graph F.1 shows the values of ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_& on the day of an ECB announcement, to investigate the impact of an announcement shock. The most notable shock in the graph is the introduction of PSPP, especially the impact on non-GSIBs systemic risk is large. Although it is interesting to see how an announcement shock impacts systemic risk, this only demonstrates a temporary effect. Graph 1 displays more structural breaks in the trends, which might imply permanent effects on systemic risks. Therefore, I analyse the whole period for which the policy is active.

E

This subsection discusses two econometric problems, serial correlation and heteroscedasticity. First, I introduce the concepts of these problems. Afterwards, I discuss the results of the tests.

With heteroscedasticity, the variance of the error term is not constant, but it depends on the observation. To test for this problem, I perform the White test. Additionally, I test for serial correlation in the models, using the Breasch-Godfrey Serial Correlation LM Test. Serial correlation occurs if there is a correlation between the observations ordered in time.

5.

R

ESULTS

This section describes and discusses the main results of the models in this analysis. This study investigates the different policies individually, in which the coefficient for the interaction between the macroeconomic shock indicator and the dummy variable (𝑥_&∗ 𝐷_& ) is most interesting. This variable describes the marginal effect of a policy related macroeconomic shock on systemic risk. Furthermore, this study compares the three subsets to test for heterogeneity between banks (Hypothesis 4). Table 2, 3 and 4 provide the empirical analysis results.

Table 2 provides the empirical analysis results for the sample of all banks. I do not find significant results for all models in this sample, which implies that for this sample of banks, a macroeconomic shock and a macroeconomic shock of the introduction of a new form of UMP not significantly affects systemic risk.

Table 2: Empirical analysis results for all banks in the period 2006-2018

DEPENDENT VARIABLE: ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲𝒕

DUMMY (𝑫𝒕)

Explanatory variable

LTRO APP NIRP

Table 3: Empirical analysis results for all GSIBs in the period 2006-2018

DEPENDENT VARIABLE: ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩,𝒕} DUMMY (𝑫_𝒕)

Explanatory variable

LTRO APP NIRP

𝑪𝒐𝒏𝒔𝒕𝒂𝒏𝒕 -0.113 (0.098) -0.117 (0.072) -0.126* (0.073) 𝒙𝒕 0.101 (0.069) 0.114** (0.057) 0.079 (0.049) 𝑫_𝒕 -0.016 (0.054) -0.007 (0.046) 0.007 (0.054) 𝒙_𝒕∗ 𝑫_𝒕 -0.092 (0.100) -0.148 (0.101) -0.092 (0.134) 𝑽𝑺𝑻𝒕 0.005* (0.003) 0.005* (0.003) 0.005* (0.003) METHOD LS LS LS R-SQUARED 0.002 0.002 0.001 F-STATISTIC 1.127 1.334 1.059 OBSERVATIONS 2984 2984 2984 Note: standard deviation is showed in the parentheses. Significant at 10%(*), 5%(**), or 1% (***). HAC Standard Errors are applied to all models

Table 3 provides the results of the empirical analysis for the sample of GSIBs. The coefficient for variable 𝑥_&∗ 𝐷_& is not significant in all different models. This implies that a macroeconomic shock of a form of UMP does not significantly impact systemic risk of GSIBs. In the APP-model, the coefficient for variable 𝑥_& is significant at a 5% significance level. During a macroeconomic shock, systemic risk in this model is 0.114 percentage points higher.

To investigate differences between the different subsamples (Hypothesis 4), I compare Table 2, 3, and 4. In all models, the coefficient for variable 𝑥_&∗ 𝐷_& is not significant. Which implies that in this study, a macroeconomic shock of a policy change does not significantly affect systemic risk for all different samples of banks. The coefficient for variable 𝑥& for non-GSIBs

in the APP-model is higher than the same coefficient in the GSIBs APP-model, implying that non-GSIBs react stronger to a macroeconomic shock in this model than GSIBs. The control variable (𝑉𝑆𝑇_&) correlates significantly with systemic risk for all GSIBs models, whereas for the non-GSIBs models, this control variable does not significantly correlate with systemic risk. In the subsample containing all banks, only the coefficient for variable 𝑉𝑆𝑇_& of the APP-model and NIRP-model correlates significantly with the dependent variable. However, for all different samples, the value of the coefficient for variable 𝑉𝑆𝑇_& is relatively small, which implies that it has a small impact on systemic risk.

Table 4: Empirical analysis results for all non-GSIBs in the period 2006-2018

DEPENDENT VARIABLE: ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝑵𝒐𝒏𝑮𝑺𝑰𝑩,𝒕} DUMMY (𝑫_𝒕)

Explanatory variable

LTRO APP NIRP

E

Table 5 provides the results of the empirical analysis for the extended models. For all models, a macroeconomic shock of a new form of UMP does not significantly impact systemic risk. The coefficient for variable 𝑥& for the sample of GSIBs and the sample of non-GSIBs are both

significant at a 10% significance level. When comparing those two coefficients, the value of the non-GSIBs is higher, which implies a higher impact of a macroeconomic shock on systemic risk of non-GSIBs.

Table 5: Empirical analysis results for the extended models in the period 2006-2018

Literature states that UMP helps stabilizing the economy in the years after the crisis, however, some forms lead to a higher risk-taking of banks and increase fragility in the banking sector (Avalos and Mamatzakis, 2018; Chodorow-Reich, 2014; Heider et al., 2019; Jobst and Lin, 2016). LTROs with FRFA make lending easier, however it increases the riskiness of loans and therefore raises credit risk (Ciccarelli et al, 2013; Darracq-Paries and De Santies, 2015). Furthermore, APPs increase risk-taking of banks and increase the risk of asset price bubbles (Van den End, 2016; Kandrac and Schlusche, 2017). Besides, the NIRP further stimulates risk-taking behaviour and searching for yield (Brunnermeier and Sannikov, 2014). Although literature suggests that it is highly likely that systemic risk of banks increases due to the new forms of UMP, this study does not find significant impact. This insignificant relation could be explained by different approaches. Acharya et al. (2012) and Brownlees and Engle (2016) use

SRISK to investigate the contribution of an individual bank to the total systemic risk in the

that I do not find significant results, because the model I use investigate the impact on daily change in systemic risk over a longer period. Graph E.1, E.2 and E.3 in Appendix E show the high volatile daily difference of the SRISK series this thesis uses. This high daily volatility in

SRISK might explain insignificant results. Besides, the model to estimate a macroeconomic

shock contains monthly and quarterly data. Inflation expectations change on quarterly base, while a shock might occur on daily base. This gap between the shock and the change in inflation expectation is a plausible reason for the insignificant results.

Regarding the fourth hypothesis, I find a slight increase in systemic risk for the APP-models of the GSIBs and non-GSIBs during a macroeconomic shock. The increase for non-GSIBs is higher than for GSIBs, however in both cases the impact is considerably small. Furthermore, the impact of all policy related macroeconomic shocks do not significantly impact all different subsamples. Literature suggests that banks with more market power and higher liquidity and capital are less affected by monetary policy (Brissimis and Delis, 2010). Ricci (2015) finds that impact of monetary policy is higher for weaker banks. However, large banks, with high market power, can still be weakly structured. I distinguish between GSIBs and non-GSIBs, it is plausible that using these subsamples do not result in significant heterogeneity effects.

R

This sub-section test for robustness of the model by executing a robustness check. I investigate whether results are robust to a distinction between positive and negative inflation shocks.

I distinguish two different macroeconomic shocks, periods for which actual inflation is higher than expected inflation, hence a positive inflation shock (𝑥_&,•), and periods for which expected inflation is higher than actual inflation, in this case the inflation shock is negative (𝑥&,€). When

policy is possibly less effective than expected. Both situations result in a macroeconomic shock, the difference in impact of those different situations on systemic risk is interesting to investigate. I extent Model (14) with distinction between negative and positive shocks, to investigate whether distinction between negative and positive inflation shocks alter results considerably. The model is as follows:

(15) ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_& = 𝛽_\+ 𝛽_]∗ 𝑥_&,•+ 𝛽_^∗ 𝑥_&,€+ 𝛽_`∗ 𝐷_&+ 𝛽_5a∗ 𝑥_&,•∗ 𝐷_&+ 𝛽₅₅∗ 𝑥_&,€ ∗ 𝐷&+ 𝛽5Q ∗ 𝑉𝑆𝑇&+ 𝜀&

This model can be simplified; I distinguish four situations:

(16) [∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_&|𝐷_& = 0; 𝑥 > 0] = 𝛽_\+ 𝛽_] ∗ 𝑥_&,•+ 𝛽_5Q ∗ 𝑉𝑆𝑇_&+ 𝜀_& (17) [∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_&|𝐷& = 0; 𝑥 < 0] = 𝛽\+ 𝛽^ ∗ 𝑥&,€+ 𝛽5Q ∗ 𝑉𝑆𝑇&+ 𝜀&

(18) [∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_&|𝐷& = 1; 𝑥 > 0] = (𝛽\+ 𝛽`) + 𝑥&,• ∗ (𝛽]+ 𝛽5a) + 𝛽5Q ∗ 𝑉𝑆𝑇&+ 𝜀&

(19) [∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾&|𝐷& = 1; 𝑥 < 0] = (𝛽\+ 𝛽`) + 𝑥&,€ ∗ (𝛽^+ 𝛽55) + 𝛽5Q ∗ 𝑉𝑆𝑇&+ 𝜀&

Appendix H provides test results. Table H.2 and Table H.3 show that for GSIBs and non-GSIBs, the coefficients for variable 𝑥& and the coefficients for the interaction variables 𝑥&∗ 𝐷&, the sign

remains the same for a positive and a negative shock. For the sample of all banks (Table H.1), in the LTRO-model, the sign of the coefficients changes when distinguishing for a positive and a negative shock, for both the coefficient for variable 𝑥& and for the coefficient for variables

𝑥&∗ 𝐷&. In the NIRP-model, the coefficient sign only changes for variable 𝑥&. However, all

6.

C

ONCLUSION AND DISCUSSION

The aim of this research is to investigate the interaction between systemic risk and different forms of UMP between 2006-2018. For this purpose, this thesis examines three different forms of UMP; LTROs, APPs and NIRP. To test for the impact of these policies, I use a measure for systemic risk: SRISK. Furthermore, I use a macroeconomic shock indicator, which measures the level of inflation uncertainty as a proxy for a macroeconomic shock. The models in this thesis test for the impact of a macroeconomic shock of a new form of UMP on the level of systemic risk in three different samples. Although literature suggests a relation between forms of UMP and financial stability, I do not find a significant relation between systemic risk and forms of UMP. The results in my analysis are mainly insignificant. However, the absence of a relationship seems unlikely considering the existing literature. Hence, I explain these insignificant results with other plausible reasons; The SRISK measure is not suitable to capture UMP shocks; There exist a long run causality between UMP and systemic risk, which is not captured in my model; The gap in data between the macroeconomic shock indicator and systemic risk might explain why systemic risk in this model is not directly impacted by a macroeconomic shock; The distinction between GSIBs and non-GSIBs does not result in significant heterogeneous effects between banks, rather samples of different structured banks can possibly explicate heterogeneous impacts of UMP on systemic risk. Further research can elaborate on these above-mentioned reasons. However, to the extent of this research, I conclude that UMP does not significantly influence systemic risk in the Eurozone banking sector.

R

EFERENCES

Acharya, V., Engle, R., & Richardson, M. (2012). Capital shortfall: A new approach to ranking and regulating systemic risks. American Economic Review, 102(3), 59-64.

Adrian, T., & Brunnermeier, M. K. (2011). CoVaR (No. w17454). National Bureau of

Economic Research.

Avalos, F. H., & Mamatzakis, E. C. (2018). Euro area unconventional monetary policy and bank resilience. Bank for International Settlement, Working Paper, 754.

Benoit, S., Colliard, J. E., Hurlin, C., & Pérignon, C. (2017). Where the risks lie: A survey on systemic risk. Review of Finance, 21(1), 109-152.

Brissimis, S. N., & Delis, M. D. (2010). Bank heterogeneity and monetary policy transmission.

ECB Working Paper, 1233.

Brownlees, C., & Engle, R. F. (2016). SRISK: A conditional capital shortfall measure of systemic risk. The Review of Financial Studies, 30(1), 48-79.

Brunnermeier, M. K., & Sannikov, Y. (2014). A macroeconomic model with a financial sector. American Economic Review, 104(2), 379-421.

Chodorow-Reich, G. (2014). Effects of unconventional monetary policy on financial institutions. National Bureau of Economic Research, w20230.

Ciccarelli, M., Maddaloni, A., & Peydró, J. L. (2013). Heterogeneous transmission mechanism: monetary policy and financial fragility in the eurozone. Economic Policy, 28(75), 459-512. Colletaz, G., Levieuge, G., & Popescu, A. (2018). Monetary policy and long-run systemic risk-taking. Journal of Economic Dynamics and Control, 86, 165-184.

Darracq-Paries, M., & De Santis, R. A. (2015). A non-standard monetary policy shock: The ECB's 3-year LTROs and the shift in credit supply. Journal of International Money and

Finance, 54, 1-34.

Ehrmann, M., Gaballo, G., Hoffmann, P., & Strasser, G. (2019). Can more public information raise uncertainty? The international evidence on forward guidance. Journal of Monetary

Economics, 108, 93-112.

Engle, R., Jondeau, E., & Rockinger, M. (2014). Systemic risk in Europe. Review of

Finance, 19(1), 145-190.

European Banking Authority (2019). Global Systemically Important Institutions (G-SIIs).

Retrieved from: www.eba.europa.eu

European Central Bank (2019), Inflation Rate (HICP) for the Euro Area [Data Set]. Retrieved

from: www.sdw.ecb.europa.eu

European Central Bank (2019). HICP Inflation Forecasts [Data Set]. Retrieved from:

www.ecb.europa.eu/stats

Galati, G., & Moessner, R. (2013). Macroprudential policy–a literature review. Journal of

Economic Surveys, 27(5), 846-878.

Giannone, D., Lenza, M., Pill, H., & Reichlin, L. (2012). The ECB and the interbank market. The Economic Journal, 122(564), F467-F486.

Girardi, G., & Ergün, A. T. (2013). Systemic risk measurement: Multivariate GARCH estimation of CoVaR. Journal of Banking & Finance, 37(8), 3169-3180.

Heider, F., Saidi, F., & Schepens, G. (2019). Life below zero: Bank lending under negative policy rates. The Review of Financial Studies, 32(10), 3728-3761.

International Monetary Fund (2009). Global financial Stability report, responding to the financial crisis and measuring systemic risks. Retrieved from: www.imf.org

Jobst, A., & Lin, H. (2016). Negative interest rate policy (NIRP): implications for monetary transmission and bank profitability in the euro area. IMF Working Paper, 16/172.

Joyce, M., Miles, D., Scott, A., & Vayanos, D. (2012). Quantitative easing and unconventional monetary policy–an introduction. The Economic Journal, 122(564), F271-F288.

Laeven, L., Ratnovski, L., & Tong, H. (2016). Bank size, capital, and systemic risk: Some international evidence. Journal of Banking & Finance, 69, S25-S34.

NYU Stern School of Business VLAB (2019). SRISK [Data Set]. Retrieved from:

www.vlab.stern.nyu.edu

Ricci, O. (2015). The impact of monetary policy announcements on the stock price of large European banks during the financial crisis. Journal of Banking & Finance, 52, 245-255. Stoxx Digital (2019). EURO STOXX 50 Volatility (VSTOXX) [Data Set]. Retrieved from

www.stoxx.com

The Dutch Central Bank (2019). Financial stability report autumn 2019. Retrieved from:

www.dnb.nl

A

PPENDIX

A:

P

OLICY ANNOUNCEMENT LIST OF

ECB

Table A.1 LTROs and TLTROs with FRFA policy announcements of ECB between 2006-2018, retrieved from www.ecb.europa.eu

Date Announcement

25/06/2009 – 01/07/2010 01/10/2009 – 30/09/2010 17/12/2009 – 23/12/2010

LTRO with increased maturity of 12 months, FRFA

22/12/2011 – 29/01/2015 01/03/2012 – 26/02/2015

LTRO with maturity of 36 months, FRFA

01/09/2014 – 30/09/2018 01/12/2014 – 30/09/2018 01/03/2015 – 30/09/2018 01/06/2015 – 30/09/2018

Targeted Longer-term refinancing operations (TLTROs).

01/06/2016 – 31/12/2018 01/03/2017 – 31/12/2018

Targeted Longer-term refinancing operations (TLTROII)

Table A.2. Asset purchasing programs policy announcements of ECB between 2006-2018, retrieved from www.ecb.europa.eu

Date Announcement

01/07/2009 - 30/06/2010 01/11/2011 – 31/10/2012 15/10/2014 – 19/12/2018

Covered bond Purchasing Program (CBPP) Covered Bond Purchasing Programme (CBPP2) Covered Bond Purchasing Programme (CBPP3)

01/11/2014 – 19/12/2018 09/03/2015 – 19/12/2018 08/06/2016 – 19/12/2018

Asset Backed Securities Purchasing Programme (ABSPP) Public Sector Purchase Program (PSPP)

Corporate Sector Purchase Programme (CSPP)

Table A.3 NIRP policy announcements of ECB between 2006-2018, retrieved from www.ecb.europa.eu

Date Announcement

A

PPENDIX

B:

L

IST OF ALL BANKS

Table B.1: Banks included in the analysis.

Country Bank

Austria Erste Group

Raiffeisen Bank International Belgium Dexia SA

KBC Group Finland Nordea Bank Abp

France Credit Agricole BNP Paribas BPCE/Nataxis Societe Generale Germany Commerzbank

Deutsche Bank Greece Alpha Bank S.A.

Eurobank Ergasias Bank S.A. National Bank of Greece S.A. Piraeus Bank S.A.

Ireland Allied Irish Bank Bank of Ireland Italy Itesa Sanpaolo

Unicredit Netherlands ING

Portugal Banco Comercial Português Spain BBVA

Sabadell Santander

A

PPENDIX

C:

C

ORRELATION MATRICES

Table C.1: Correlation matrix for all banks

𝑺𝑹𝑰𝑺𝑲_𝒕 𝑽𝑺𝑻_𝒕 𝒙_𝒕

𝑺𝑹𝑰𝑺𝑲_𝒕 1

𝑽𝑺𝑻_𝒕 0.505*** 1

𝒙_𝒕 -0.013 -0.034* 1

Significant at 10%(*), 5%(**), or 1% (***)

Table C.2: Correlation matrix for all GSIBs

𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩} 𝑽𝑺𝑻_𝒕 𝒙_𝒕

𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩,𝒕} 1

𝑽𝑺𝑻_𝒕 0.508*** 1

𝒙𝒕 -0.016 -0.032* 1

Significant at 10%(*), 5%(**), or 1% (***)

Table C.3: Correlation matrix for all non-GSIBs

𝑺𝑹𝑰𝑺𝑲𝑵𝒐𝒏𝑮𝑺𝑰𝑩 𝑽𝑺𝑻𝒕 𝒙𝒕

𝑺𝑹𝑰𝑺𝑲𝑵𝒐𝒏𝑮𝑺𝑰𝑩,𝒕 1

𝑽𝑺𝑻𝒕 0.478*** 1

𝒙_𝒕 -0.011 -0,034* 1

A

PPENDIX

D:

S

TATIONARITY TESTS

Table D.1: Summary results of Augmented Dickey-Fuller test

INTERCEPT Level First difference

ADF T-STATISTICS PROB. ADF T-STATISTIC PROB.

𝑺𝑹𝑰𝑺𝑲_𝒕 -1.588 0.489 -55.720 0.000 𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩} -1.670 0.447 -59.709 0.000 𝑺𝑹𝑰𝑺𝑲𝑵𝒐𝒏𝑮𝑺𝑰𝑩 -1.427 0.570 -55.411 0.000 𝒙_𝒕 -4.749 0.000 -58.198 0.000 𝑽𝑺𝑻_𝒕 -4.558 0.000 -30.197 0.000 TREND & INTERCEPT

Level First Difference

ADF T-STATISTIC PROB. ADF T-STATISTIC PROB.

𝑺𝑹𝑰𝑺𝑲_𝒕 -1.474 0.839 -55.761 0.000

𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩} -1.574 0.803 -59.725 0.000 𝑺𝑹𝑰𝑺𝑲𝑵𝒐𝒏𝑮𝑺𝑰𝑩 -1.394 0.863 -55.450 0.000

𝒙_𝒕 -4.763 0.001 -58.190 0.000

A

PPENDIX

E:

G

RAPHS AND DESCRIPTIVE STATISTICS

USING

∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲

Graph E.1: ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾 for all banks in the sample over the period 2006-2018 ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾&

Graph E.3: ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾 for all non-GSIBs over the period 2006-2018

Table E.2: Correlation matrix for all banks

∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲𝒕 𝑽𝑺𝑻𝒕 𝒙𝒕

∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_𝒕 1

𝑽𝑺𝑻_𝒕 0.039** 1

𝒙_𝒕 0.0162 -0.037* 1

Significant at 10%(*), 5%(**), or 1% (***)

Table E.1: Descriptive statistics of systemic risk variables in the period 2006-2018

N Mean Med. Max. Min. St. Dev.

∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_𝒕 2701 -0,023 0,032 5.482 -8,062 1.581 ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩,𝒕} 3057 -0.005 0.040 5.110 -6.634 1.396 ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲𝒏𝒐𝒏𝑮𝑺𝑰𝑩,𝒕 2701 -0.016 0.041 9.058 -12.451 2.391

Table E.2: Correlation matrix for all GSIBs ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩,𝒕} 𝑽𝑺𝑻_𝒕 𝒙_𝒕 ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩,𝒕} 1 𝑽𝑺𝑻_𝒕 0.031* 1 𝒙_𝒕 0.017 -0.032* 1 Significant at 10%(*), 5%(**), or 1% (***)

Table E3: Correlation matrix for all non-GSIBs

∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝒏𝒐𝒏𝑮𝑺𝑰𝑩,𝒕} 𝑽𝑺𝑻𝒕 𝒙𝒕

∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝒏𝒐𝒏𝑮𝑺𝑰𝑩,𝒕} 1

𝑽𝑺𝑻_𝒕 0.032 1

𝒙_𝒕 0.021 -0.036* 1

A

PPENDIX

F:

A

NNOUNCEMENT SHOCKS

Graph F.1: ∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾_& during the most important announcement shocks of ECB

Table F.1: Most important policy announcements of ECB between 2006-2018, retrieved from www.ecb.europa.eu

08/10/2008 Initiation of fixed rate tender with full allotment (FRFA).

07/05/2009 LTRO increased maturity to one year, FRFA procedure

04/06/2009 Initiating the covered bond purchasing program (CBPP)

06/10/2011 Initiating two longer-term refinancing operations (LTROs), with maturity of 12 months and 12 months. Launch of Covered Bond Purchase Programme (CBPP2).

08/12/2011 Introduction of LTRO with maturity of 36 months.

26/07/2012 Draghi’s Whatever it takes speech

05/06/2014 Introduction of targeted longer-term refinancing operations (TLTROs). Extend the outright purchase of asset-backed securities (ABS). Introduction of negative deposit facility interest rate (NIRP), deposit facility interest rate decrease to

-0.10%

09/03/2015 Initiation of Public Sector Purchase Program (PSPP)

10/03/2016 Monthly purchase of APP expanded. New series of TLTROII. Introduction of Corporate Sector Purchase Programme (CSPP). Deposit facility interest rate decrease to at -0.40%.

14/06/2018 APP continue until end of September, after that net purchases will be reduced until the end of December, and then will end.

-8 -6 -4 -2 0 2 4 6 8-10-08 7-05-09 4-06-09 6-10-11 8-12-11 26-07-12 5-06-14 9-03-15 10-03-16 14-06-18

∆𝐿𝑁𝑆𝑅𝐼𝑆𝐾𝑡

A

PPENDIX

G:

E

MPIRICAL TEST RESULTS

Table G.1: Test results heteroscedasticity and serial correlation tests for all models

White test Breasch-Godfrey test

DEP. VARIABLE Dummy: F-Stat. Prob. F-Stat. Prob.

A

PPENDIX

H:

P

OSITIVE AND NEGATIVE SHOCKS

Table H.1: Empirical analysis results with distinction for positive and negative shocks

DEPENDENT VARIABLE: ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲𝒕

DUMMY (𝑫𝒕)

Explanatory variable

LTRO APP NIRP

Table H.2: Empirical analysis results with distinction for positive and negative shocks

DEPENDENT VARIABLE: ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝑮𝑺𝑰𝑩,𝒕} DUMMY (𝑫_𝒕)

Explanatory variable

LTRO APP NIRP

Table H.3: Empirical analysis results with distinction for positive and negative shocks

DEPENDENT VARIABLE: ∆𝑳𝑵𝑺𝑹𝑰𝑺𝑲_{𝑵𝒐𝒏𝑮𝑺𝑰𝑩,𝒕} DUMMY (𝑫_𝒕)

Explanatory variable

LTRO APP NIRP