The bank-sovereign nexus and the impact of the zero risk weight-rule for EU sovereign debt

1

The bank-sovereign nexus and the impact of the zero risk

weight-rule for EU sovereign debt

Master’s thesis for the Msc. Finance & the Msc. International Economics and Business Faculty of Economics and Business, University of Groningen

12th June 2016

Mathijs Marten Lindemulder m.m.lindemulder@student.rug.nl

Student number: 2218380

Abstract

This thesis investigates the link between sovereign and bank credit default risk in Europe for the period 2010-2016. I identify the link between credit default risks of banks and sovereigns as the bank-sovereign nexus. The bank-sovereign nexus is measured with excess correlations. Excess correlation is defined as the quarterly correlation between residuals of daily changes of bank and sovereign CDS spreads which are not explained by a set of fundamental market factors (MSCI Europe, Vstoxx and Itraxx). I find significant evidence for an increase in the excess correlation if a bank increases its sovereign exposure to the sovereign counterpart. Furthermore, this thesis investigates the impact of the zero risk-weight rule for EU sovereign debt. This study finds that after the introduction of capital requirements for EU sovereign debt holdings of banks, the impact of sovereign exposures on excess correlations decreases, compared to the period in which risk-weights are absent.

2

1. Introduction

The global financial and consecutive Eurozone economic crises have shown that there is still a strong relationship between the default risks of banks and sovereigns (European Systemic Risk Board (ESRB), 2015). The ESRB (2015) claims that the large sovereign debt holdings of banks and the preferential treatment of EU sovereign exposures in capital requirements for banks drive the link between default risks of banks and sovereigns. This thesis attempts to identify the impact of sovereign exposure of banks on the link between bank and sovereign credit default risks. Furthermore, the key objective of this thesis is to identify what the impact of sovereign debt holdings of banks is on the link between bank and sovereign credit default risks in the period with zero risk-weights, compared to the period in which capital requirements for sovereign exposures are introduced.

In this thesis I find that the link between credit default risks of banks and sovereigns increases with the amount of sovereign debt holdings of banks. Furthermore, I find that sovereign exposure in the period where risk-weights to EU sovereign exposure are absent increases the link between bank and sovereign credit default risk, compared to the period where capital requirements for sovereign exposures are introduced. My findings have important implications for policy makers since the preferential treatment of risk-weights for EU sovereign debt has a considerable and significant increasing effect on risk spillovers between sovereign and bank credit default risks.

I find these results by using Credit Default Swaps (CDS) of banks and sovereigns as a measure of credit default risks. The link between credit default risks of banks and sovereigns is established by calculating the excess correlation of CDS spreads in line with De Bruyckere, Gerhardt, Schepens and Vander Vennet (2013). I define the excess correlation as: the quarterly correlation between residuals of daily changes in bank and sovereign CDS spreads which are not explained by a set of fundamental market factors (MSCI, Vstoxx, Itraxx).

3

In a first step, I analyze impact of the degree of sovereign bond holdings to total assets of banks on the excess correlations. My results show that banks that have a higher degree of exposure to a sovereign counterpart have a higher excess correlation between their default risk and that of the particular sovereign counterpart. I verify that this effect is not solely driven by countries in distress.

More specifically, to evaluate the impact of zero-risk weights to sovereign exposures I analyze the impact of an announcement of the European Banking Authority (EBA) in September 2011 on excess correlations. The EBA is a specialized institution that works alongside of the main EU institutions and is established by the European Council and EU parliament to maintain financial stability in the European Banking Sector (EBA, 2016). Prior to September 2011 banks did not have to hold a capital buffer against their EU sovereign exposure. The regulations on risk-weights for banks are stipulated in the Capital Requirements regulation (CRR) and Capital Requirements Directive (CRD) which follow from the guidelines from the Basel Committee. For a detailed summary of the most important institutional background with regard to capital requirements for sovereign exposures I refer the reader to Appendix A.

In September 2011 the EBA announces that banks have to strengthen their capital position by building up a temporary capital buffer against sovereign debt to reflect current market prices (EBA, 2011). As Korte and Steffen (2015) stress, the announcement of the EBA can be interpreted as a de facto implementation of assigning risk-weights to sovereign debt exposures of banks. Since the start of the European Union, this is the first time that EU institutions formally acknowledge that Eurozone sovereign debt is not risk-free and that banks have to hold capital against their sovereign debt exposure.

4

My thesis relates to research in the relatively new field of research on risk spillovers and the link between credit default risks of banks and sovereigns, which I define as the bank-sovereign nexus. Theory on risk spillovers between banks and sovereigns is relevant to theory in the field of International Economics and Business. Furthermore, my thesis is related to research in the field of regulatory capital arbitrage, which relates to theory in the field of Finance.

My research contributes to existing literature in these fields by explicitly analyzing the impact of the announcement of the EBA on the bank-sovereign nexus. Hence, my paper is the first paper that compares the impact of sovereign debt holdings of banks on excess correlations in the absence of regulatory risk-weights prior to September 2011 with the period after September 2011 in which capital buffers against sovereign debt are introduced.

Existing literature on the bank-sovereign nexus mainly focuses on bank-specific and country-specific drivers of the link between credit default risks of banks and sovereigns (Breckenfelder and Schwaab, 2015; De Bruyckere et al., 2013). Another stream of literature (Acharya, Drechsler and Schnable, 2012; 2014) focuses on a feedback loop between the deteriorated value of a bank and the sovereign counterpart its creditworthiness. The loop between the default risks of banks and sovereigns is best explained by a distressed financial sector that induces a bail-out from a government, which costs increase the sovereign credit risk. The increased sovereign credit risk deteriorates the creditworthiness of the financial sector in turn, due to a substantial share of sovereign bond holdings of banks.

5

banks prior to and after the introduction of capital requirements on the bank-sovereign nexus.

Therefore, my thesis contributes to research in the fields of regulatory capital arbitrage and research on the bank-sovereign nexus by finding that in the period where risk-weights are absent, the impact of sovereign exposure increases the bank-sovereign nexus compared to period in which banks have to hold capital buffers against their sovereign exposures. These findings imply that the link between sovereign and bank credit default risk is increased due to the absence of regulatory risk-weights for sovereign debt.

This thesis is structured as follows. In Section 2 I review the existing literature that is relevant to the bank-sovereign nexus and regulatory capital arbitrage. Furthermore, the hypotheses of this thesis are also presented in Section 2. I elaborate on the methodology and empirical strategy in Section 3. I present a description of my data collection in Section 4. In Section 5 I present and discuss my results from the regressions. Robustness checks of the results in Section 5 are discussed in Section 6. My conclusions on the findings of this thesis, the implications of these findings and subsequent policy recommendations are discussed in Section 7.

2. Literature review

In this section the most relevant existing literature on the bank-sovereign nexus and regulatory capital arbitrage are reviewed. In appendix D I include a table with a detailed

overview of the most relevant literature on the bank-sovereign nexus that I review below.

After a discussion of the most relevant literature, Section 2.3 presents the hypotheses of this thesis.

2.1 Bank-sovereign nexus

My thesis is related to research in the field of the link between the credit default risks of banks and sovereigns. I define the link between credit default risk of banks and sovereigns as the bank-sovereign nexus, which is measured in line with the methodology of De Bruyckere et al (2013) as elaborated on in Section 3. The bank sovereign nexus is also identified as the

deadly embrace, vicious circle or doom loop (Acharya, Drechsler and Schnabl, 2012; 2014;

6

Acharya, Drechsler and Schnabl (2012) find a connection, which they define as a nexus, between the credit risks of banks and sovereigns measured by their CDS spreads for the period November 2008 – September 2010. Their study focuses on banks with total assets larger than USD 50 billion, that have an investment grade rating and trade in Credit Default Swaps. Acharya, Drechsler and Schnabl (2012) establish the nexus by regressing the log of bank CDS spreads on the change in the log of a sovereign CDS spread, after controlling for the credit rating of the bank. Acharya, Drechsler and Schnabl (2012) find that after controlling for the credit rating of banks, an increase in sovereign CDS spreads significantly increases bank CDS spreads. They continue with stating that this effect is stronger for banks with a lower credit rating.

Acharya, Drechsler and Schnabl (2012) interpret these results as two adverse feedback effects of credit risks between banks and sovereigns that lead to a nexus between the CDS spreads of banks and sovereigns. Acharya, Drechsler and Schnabl (2012) point out that the nexus is best explained as a two-tale debt overhang problem. Acharya, Drechsler and Schnabl (2012) argue that when a financial institution faces undercapitalization, sovereigns fear a contraction of bank-lending that possibly results in a liquidity crunch, along with severe consequences for economic growth. Sovereigns are then forced to bail-out financial institutions that are undercapitalized to prevent a credit crunch. The sovereign has to issue additional sovereign debt to bail-out financial institutions that are in distress, thereby raising their own debt overhang problem. In turn, the credit risk of the sovereign increases through the liability side of the balance-sheet of the sovereign.

Additionally, Acharya, Drechsler and Schnable (2012) identify increased credit risk through the asset-side of the balance sheet via lower investments of households and firms due to higher anticipated future taxes by the sovereign as a consequence of the issuance of new debt and declined economic prosperity. Increased credit risk of sovereigns is transmitted via two channels that lead to the nexus according to Acharya, Drechsler and Schnabl (2012).

7

Consequently, they stress that the deteriorated value of credit guarantees of the sovereign adversely influences the credit quality of the financial sector. Acharya, Drechsler and Schnable (2012) claim that the direct holdings of government bonds is a channel that intensifies the nexus. However, they do not regress on the actual sovereign bond holdings of banks, which my thesis does explicitly.

Acharya, Drechsler and Schnabl (2012) interpret the presence of a direct holdings channel from deteriorated credit ratings of banks, whereas my research explicitly uses the sovereign debt holdings of banks to make claims on the impact of the holdings of banks on the bank-sovereign nexus. Furthermore, their measure of the nexus differs from my preferred measure of the nexus since I use the excess correlation between bank and sovereign CDS spreads as dependent variable. My measure is based on the methodology of De Bruyckere et al. (2013) that is highlighted Section 3. De Bruyckere et al. (2013) claim that a correlation measure is preferred in identifying a link between credit default risks rather than regressing the CDS spread of the sovereigns on the bank’ CDS spreads.

In their follow-up research Acharya, Drechsler and Schnabl (2014) find a feedback loop between sovereign and bank credit risk for the period 2007 – 2011. Their method and banks sample are similar to the Acharya, Drechsler and Schnabl (2012) paper. However, they add bank bail-outs of sovereigns in their analysis. Their main contribution compared to Acharya, Drechsler and Schnabl (2012) is that Acharya, Drechsler and Schnabl (2014) find that bank bail-outs trigger sovereign credit risk. In turn the change in credit risks leads to a deterioration of bank credit risk, similar to the analysis in their 2012 paper. They describe the feedback loop as a consequence of bank outs as follows. A distressed financial sector induces bank bail-outs by national governments. When governments issue new debt to bail-out banks, sovereign credit default risk increases as a consequence.

As a result of increased sovereign credit default risk, bank credit default risk increases since banks have a substantial exposure to the deteriorated sovereign debt. Therefore, Acharya, Drechsler and Schnabl (2014) claim that the banks’ balance sheet value erodes as a consequence of the deteriorated sovereign debt holdings of banks.

8

Their aim is to find the effect of bank credit risk on sovereign risk both within and across countries. A detailed summary of the Breckenfelder and Schwaab (2015) study is presented in Appendix D.

The study of Breckenfelder and Schwaab (2015) is focused around the comprehensive assessment of the EBA in October 2014. The comprehensive assessment of the ECB is a health check of the banks the ECB supervises (ECB, 2016). The data sample of the Breckenfelder and Schwaab (2015) study covers data from September 2014 - November 2014. Their sample of banks covers 130 banks and is restricted to banks that participate in the comprehensive assessment of the ECB and that trade in CDS spreads. Breckenfelder and Schwaab (2015) find that sovereign risk increases in non-stressed countries, when the risk in the financial sector in stressed countries erupted. Hence, they argue that contagion risks are spread across borders and are not related to a particular sovereign counterpart. Their results are in contrast with Acharya, Drechsler and Schnable (2012, 2014), which find a feedback loop between the banking sector and the sovereign counterpart.

Farhi and Tirole (2016) develop a theoretical model for the deadly embrace between sovereigns and banks. Their model has similarities to the interpretation of the loop defined by Acharya, Drechsel and Schnabl (2014), since it focuses on an adverse shock that deteriorates the value of sovereign debt. The model of Farhi and Tirole (2016) is best summarized as follows: due to an adverse economic shock, the value of sovereign debt holdings of banks deteriorates. Hence, bail-outs of the financial sector by governments are triggered by deteriorated sovereign debt holdings of banks. Farhi and Tirole (2016) conclude with stating that bail-outs are costly and will therefore result in a further deterioration of sovereign debt value.

9

De Bruyckere et al. (2013) conclude that there are risk spillovers between banks and sovereigns. Particularly, they examine which bank- or country-specific factors drive these risk spillovers. Furthermore, they find a strong home bias of banks’ sovereign debt holdings, which they conclude from higher excess correlations between banks and the sovereign in which the bank is headquartered. De Bruyckere et al. (2013) argue that there are two driving forces for this home bias. First, the substantial share of domestic sovereign debt holdings in a banks’ portfolio. Secondly, they find that the total assets of banks increase excess correlations and they conclude from this findings that large banks trigger bailouts by the governments.

Furthermore, De Bruyckere et al. (2013) analyze the impact of the sovereign exposure to total sovereign exposure of a bank on the bank-sovereign nexus and find that a higher degree of sovereign exposure results in an increase in the excess correlation of the bank and its sovereign counterpart. My method is similar to the approach of De Bruyckere et al. (2013). However, I use the sovereign exposure to total assets of a bank instead of the sovereign exposure to total sovereign exposure as an explanatory variable since this better represents the degree of exposure a bank has relative to its size.

The research of De Bruyckere et al. (2013) is mainly focused on identifying bank-specific and country-bank-specific drivers of excess correlations between bank and sovereign CDS spreads. For example, De Bruyckere et al. (2013) find that a higher tier 1 capital of banks decreases excess correlations between banks and their sovereign counterpart. Furthermore, the results of the De Bruyckere et al. (2013) show that a lower reliance on short-term funding decreases the excess correlation between bank and sovereign CDS spreads. Additionally, they find that a higher debt-to-GDP ratio of the sovereign increases excess correlations.

10 2.2 Regulatory capital arbitrage

Next to research in the field on the bank-sovereign nexus as elaborated on in Section 2.1, this thesis also relates to academic research in the field of regulatory capital arbitrage. Steffen and Acharya (2015) define regulatory capital arbitrage as an incentive for banks to invest in high-yielding assets with low risk weights to increase the short-turn return on equity without the need to issue economic capital.

Acharya, Engle and Pierret (2014) stress that the regulatory risk weights that are employed by the EBA have no link with the realized risks that banks encounter in times of a crisis. Hence, Acharya, Engle and Pierret (2014) argue in accordance with Arnold, Borio, Ellis and Moshirian (2012) that banks are incentivized to invest a substantial part of their portfolio in a single asset category or increase their exposure to an asset class that has an inaccurate risk measure. In this section existing literature on regulatory capital arbitrage will be discussed.

Korte and Steffen (2015) argue that the zero-risk weight to sovereign debt as stipulated in the CRR gives a perverse incentive for the financial sector to invest in sovereign debt. They argue that banks are incentivized by the zero risk-weight rule for EU sovereign debt to increase their exposure to EU sovereign debt, without the need to increase their capital reserves. Korte and Steffen (2015) build a data set from EBA bank portfolio data and consolidated banking statistics from the BIS for the period March 2010 – June 2012. Therefore, their sample of banks is similar to my sample of banks. However, their sample of banks is not restricted to banks for which CDS spreads are available. The variable of interest of Korte and Steffen (2015) is the sovereign subsidy. They stress that European banks can take advantage of zero risk-weights for EU sovereign debt by accumulating holdings of EU sovereign debt. Accordingly, Korte and Steffen (2015) claim that the zero risk-weight to EU sovereign debt results in a subsidy to EU sovereign debt.

11

the national sovereign CDS spread and that this relationship is amplified by a larger value for their measure of the sovereign subsidy of the national financial sector. Accordingly, Korte and Steffen (2015) conclude that when a financial sector has a higher sovereign subsidy, risk spillovers between the domestic sovereign and other sovereigns increase.

Korte and Steffen (2015) conclude with stating that the co-movement of CDS spreads of sovereigns is reduced if banks employ a higher risk-weight on their sovereign exposure to non-EU sovereigns that is not subject to zero risk-weighting. Their results suggest that the presence of regulatory capital arbitrage leads to an increase risk spillovers in the Eurozone. Korte and Steffen (2015) analyze risk spillovers between sovereigns, whereas my thesis analyzes risk spillovers between the banking sector and sovereigns. My thesis and the study of Korte and Steffen (2015) are similar in the sense that they both aim to analyze the impact of the zero-risk weights on risk spillovers in the Eurozone.

In particular, Korte and Steffen (2015) find that the coefficient of the sovereign subsidy prior to the EBA announcement is positive and significant and loses its significance after the EBA announcement in September 2011. Therefore, they come to the conclusion that risk spillovers can be mitigated if regulators introduce proper risk weights to sovereign exposures. Korte and Steffen (2015) stress that risk-weights have to reflect underlying risk of the sovereign exposure. My research has similar findings to the research of Korte and Steffen (2015). However, my thesis compares the impact of the actual sovereign bond holdings of banks on risk spillovers prior to and after the announcement of the EBA instead of correcting for implied risk-weights suggested by the EBA. For emphasis, my study researches the impact of zero risk-weights on risks spillovers between banks and sovereigns, whereas Korte and Steffen (2015) solely analyze risk spillovers between sovereigns.

12

Acharya and Steffen (2015) analyze the relation between stock returns of 35 European banks and sovereign CDS spreads of Euro-countries over the period 2007-2013. The sample of banks in the study of Acharya and Steffen (2015) includes banks that report to the EBA and that are listed. Therefore, the sample of banks differs from my sample since they are not restricted by the availability of CDS spreads. The research of Acharya and Steffen (2015) differs from the methodology of my research and the research of Korte and Steffen (2015) since they focus on stock returns of banks rather than CDS spreads. In Appendix D, further details of the studies are presented.

Acharya and Steffen (2015) state that banks finance high-yielding bond positions with short-term wholesale deposit funding. Particularly, Acharya and Steffen (2015) find that European banks invest positively in long-term GIIPS-countries bonds and negatively in short-term German sovereign bonds.

Additionally, Acharya and Steffen (2015) analyze three channels through which carry trade behavior by banks can be explained. The regulatory capital arbitrage and risk shifting motive is according to Acharya and Steffen the most important transmission channel. Acharya and Steffen (2015) explain that the regulatory capital arbitrage motive originates from the zero risk-weight under Basel II guidelines. This is not completely adequate, since the Basel accords are not translated directly into national and international regulations as explained in Appendix A. The Basel accords give direction to the directives and regulations issued by the European Commission (CRR / CRD IV), however they are still subject to national and international democratic forces.

Acharya and Steffen (2015) find evidence for the regulatory capital arbitrage motive by concluding that banks with a lower tier 1 ratio have a higher share of their portfolio holdings invested in high-yielding GIIPS sovereign debt. Moreover, Acharya and Steffen (2015) find that undercapitalized, large banks with lower short-term debt with a higher ratio of RWA’s are more likely to engage in carry-trading They conclude that this is further evidence for regulatory capital arbitrage within European banks. Acharya and Steffen (2015) argue that increased risk-taking in sovereign bond yields generally leads to higher expected returns, which is in line with the findings of Battistini, Pagano and Salmonelli (2014)

carry-13

trade findings of Acharya and Steffen (2015) as it also focuses on regulatory capital arbitrage and incentives for banks to increase their risky sovereign exposures. However, Acharya and Steffen (2015) focus on stock returns of banks, whereas my thesis focuses on excess correlations between bank and sovereign credit default risks.

To summarize, as far as the author is aware, none of the literature empirically researches what influence the absence of regulatory risk-weights to sovereign exposure has on the bank-sovereign nexus. Existing literature analyzes this by developing a theoretical model (Farhi and Tirole, 2014) or by examining only the co-movement of CDS spreads of sovereigns (Korte & Steffen, 2015) or banks and sovereigns (Breckenfelder and Schwaab, 2015) over a certain timeframe. De Bruyckere et al. (2013) identify bank-specific and country-specific drivers of the bank-sovereign nexus by examining excess correlations between bank and sovereign CDS spreads.

Another stream of literature claims that the zero-risk weight to EU sovereign debt gives an incentive for banks to invest in high-yielding sovereign debt without the need to hold capital against that sovereign exposure (Acharya, Engle and Pierret, 2014; Acharya and Steffen, 2015). However, this stream of literature does not identify what impact the zero risk-weight has on risk spillovers between CDS spreads of banks and sovereigns. The findings of Korte and Steffen (2015) have the greatest similarity to my research since they find that risk spillovers between sovereign CDS spreads can be mitigated if regulators introduce risk-weights to sovereign exposures.

14 2.3 Hypotheses

In this section I formulate my hypotheses based on my key objective and the literature discussed in Section 2.1 and Section 2.2.

I hypothesize that if banks increase their degree of sovereign exposure to total assets to a particular sovereign counterpart, the excess correlation between the bank’s default risk and that of the sovereign counterpart increases (hypothesis 1). De Bruyckere et al. (2013) find that if banks increase the sovereign exposure to their total sovereign exposure to a particular counterpart the excess correlation between the bank’s default risk and that of the sovereign counterpart increases. Furthermore, Korte and Steffen (2015) find evidence for a stronger sovereign subsidy of banks if they have a higher degree of sovereign exposure to a particular counterpart. Hence, since I expect a positive coefficient for the sovereign exposure of banks based on existing literature I perform a one-sided test.

As far as I am aware, the impact of regulatory risk-weights on the bank-sovereign nexus has not been researched before. My research contributes to existing literature in the field of regulatory capital arbitrage and the bank-sovereign nexus by empirically comparing the impact of sovereign exposures on the excess correlation between bank and sovereign CDS spreads prior to and after the EBA announcement. Korte and Steffen (2015) argue that risk spillovers between sovereign CDS spreads can be mitigated if regulators introduce risk-weights to sovereign exposures. Korte and Steffen (2015) find a decrease in the co-movement of sovereign CDS spreads after September 2011 compared to the period prior to the EBA announcement.

Hence, I hypothesize that after the announcement of the EBA In September 2011 in which capital requirements for EU sovereign exposure are introduced, the excess correlation between bank and sovereign CDS spreads decreases compared to the period prior to September 2011 (hypothesis 2). Thus, I perform a one-sided test to evaluate hypothesis 2.

Furthermore, I distinguish between exposures of banks to the group of countries that is identified in existing literature (Algieri, 2013; De Grauwe, 2012; Mayer, Möbert and Weistroffer, 2012; Shambaugh, 2012) as the GIIPS (Greece, Ireland, Italy, Portugal and Spain) countries and exposures to non-GIIPS countries. The same literature argues that the GIIPS countries are countries that were in distress during the sovereign debt crisis.

I expect that exposures of banks to the countries that are in distress have a higher

15

al. (2013) find that excess correlations between banks and sovereigns are higher in GIIPS countries compared to non-GIIPS countries. However, they do not test whether this difference is significantly different from zero. My thesis performs a one-sided test to assess whether exposures to GIIPS-countries significantly increase excess correlations compared to non-GIIPS exposures. Hence, I hypothesize that if banks increase their GIIPS sovereign exposure to total assets find an increase in their excess correlation between their credit default risk and that of their sovereign counterpart compared to non-GIIPS sovereign exposure (hypothesis 3).

Furthermore, I expect that after the announcement of the EBA in September 2011, the impact of exposures to GIIPS-countries decreases more than exposures to non-GIIPS countries compared to the period prior to September 2011 due to the introduction of additional capital requirements against sovereign exposures. As stressed, there is currently no literature that is related to the impact of introducing capital requirements against sovereign exposure on risk spillovers. Therefore, I perform a one-sided test for the following hypothesis (4): after the announcement of the EBA in September 2011, the coefficient of exposure to GIIPS countries decreases more than exposure to non-GIIPS countries compared to the period prior to September 2011.

3. Methodology

In order to study the key objective and sub questions mentioned in the last paragraph of previous Section (2.3), I analyze the impact of sovereign exposures of banks on the excess correlations between bank and sovereign CDS spreads in a first step. In a next step, I include a dummy that takes a value 1 for the period after September 2011. Hence, I compare the impact of sovereign exposures prior to and after September 2011 on the excess correlation between the credit default risks of banks and sovereigns. This model is presented in equation 3 and is explained in more detail in section 3.2. Furthermore, I incorporate a dummy for exposure to GIIPS-countries.

16

increase in the bank-sovereign nexus (De Bruyckere et al., 2013). I attempt to measure whether the degree of sovereign exposure banks hold increases the bank-sovereign nexus.

In Section 3.1 below, I first elaborate on the methodology of De Bruyckere et al. (2013) on calculating excess correlations between bank and sovereign credit default risks. Subsequently, I analyze the impact of the degree of sovereign exposure of banks on the bank-sovereign nexus in Section 3.2. In Section 3.3 I compare the impact of bank-sovereign exposure of banks in the period with zero-risk weights on EU sovereign exposures prior to September 2011 with the period after September 2011 in which capital buffers against sovereign exposures are introduced.

3.1 Determining excess correlations between bank and sovereign credit default risk

To start with, in line with the Bruyckere et al. (2013) I attempt to investigate the link between bank and sovereign credit default risk. I measure the default risks of banks and sovereigns by their CDS spread which I retrieve from Datastream, as elaborated on in Section 4.1. As De Bruyckere et al. (2013) state, a natural starting point to investigate the link between risk spillovers of bank and sovereign CDS spreads would be to measure the correlation between the two default risk indicators. However, in times of distress simple correlations can be misleading since higher volatility in markets lead to higher CDS spreads (Boyer, Gibson, and Loretan (1997); Forbes and Rigobon (2002) and Loretan and English (2000)). Therefore, excess correlations are my preferred measure of the bank-sovereign nexus. Excess correlations are determined by the factors adopted in the factor model, which is explained below. This a potential drawback of the method.

17

𝐶𝐷𝑆_𝑖𝑡 = 𝐶 + 𝛽₁ 𝑀𝑆𝐶𝐼_𝑡+ 𝛽₂ 𝑉𝑆𝑡𝑜𝑥𝑥_𝑡+ 𝛽₃ 𝐼𝑡𝑟𝑎𝑥𝑥_𝑡+ ε_𝑖𝑡

𝐸𝑞𝑢𝑎𝑡𝑖𝑜𝑛 (1)

where 𝐶𝐷𝑆𝑖𝑡 is the daily change in the CDS spread of a bank or country 𝑖 retrieved

from Datastream, 𝑀𝑆𝐶𝐼𝑡 is the daily change in the MSCI-Europe index, 𝑉𝑆𝑡𝑜𝑥𝑥𝑡 is the daily

change in the volatility index and 𝐼𝑡𝑟𝑎𝑥𝑥𝑡 is the daily change of the economy-wide CDS index.

ε_𝑖𝑡 is the residual term of daily changes in sovereign and bank CDS spreads for bank or

sovereign 𝑖.

To control for the business climate in the European market I include the daily changes in the MSCI-Europe index in the factor model. The MSCI-Europe index has 446 constituents and covers around 85% of the free float market capitalization across European equities markets (MSCI, 2016). I expect that a better overall business climate reduces default probabilities and therefore CDS spreads. Hence, I expect the sign of the MSCI-Europe to be negative. I use the MSCI-Europe index rather than the total stock market index from Datastream which De Bruyckere et al. (2013) use since the MSCI-Europe’s constituents are large and mid-caps. The Total stock market index has substantial share of small-caps constituents. As my sample of banks mainly consists of large listed banks, the MSCI-Europe is more in line with my sample.

I include the daily changes in the Vstoxx Volatility indicator to control for volatility expectations in the Eurozone (Berndt, Douglas, Duffiee, Ferguson and Schranz (2005); De Bruyckere et al. (2013); Tang and Yan (2010)). The Vstoxx Volatility indicator is based on option prices of the EURO STOXX 50. The Vstoxx is often regarded as an indicator for market sentiment and fear in the Eurozone. As the fear and thereby economic uncertainty increases, uncertainty about credit default probabilities will increase as well. Hence, I expect a positive association between changes in the Vstoxx and changes in CDS spreads.

18

the expected signs of the changes in the fundamental variables of my factor model on the changes in CDS spreads of banks and sovereigns.

Figure 1. Expected signs of fundamental market variables to explain the daily changes of CDS spreads of banks and sovereigns.

ΔCDS spreads

ΔMSCI Europe ΔVstoxx ΔItraxx

+

Accordingly, in line with De Bruyckere et al. (2013) I calculate the quarterly correlation between residuals of daily CDS spreads changes for every bank-country pair which are not explained by the factor model in Equation 1. My dataset contains 32 banks and 24 sovereigns for 2010-2016 as described in Section 4. Therefore, I obtain 32 matrices with excess correlations for every bank in my sample. The matrices have 24 rows and 24 columns. The columns are identified by the excess correlation of the bank to a particular sovereign counterpart, whereas the rows are identified by the 24 quarters from 2010-2016. In Figure 2 below I aim to present how I calculate the excess correlations from raw daily CDS spreads.

Figure 2. Overview of steps to calculate the excess correlations of bank and sovereign CDS spreads.

19

Table 1. Descriptive statistics of daily changes in bank and sovereign CDS spreads, daily changes in fundamental variables of the factor model and the quarterly excess and correlation of bank and sovereign CDS spreads for 2010-2016.

Bank and sovereign CDS spreads, Vstoxx, Itraxx and MSCI Europe quotes are retrieved from Datastream.

VARIABLES Δ Sovereign

CDS spreads

Δ Bank CDS spreads

Δ Vstoxx Δ Itraxx Δ MSCI

Europe Excess correlation Correlation Mean .0005 .0005 .0020 .0005 .0001 0.1602 0.2752 Standard deviation .0207 .0246 .0647 .0310 .0129 0.1948 0.2457 Maximum .1315 .1215 .3505 .1827 .0866 0.9982 0.9991 Minimum -.1986 -.1745 -.2206 -.2781 -.0622 -0.6572 -.7691 Median -.0003 0 -.0021 0 0 0.1511 0.2505 Observations 1565 1565 1565 1565 1565 18292 18292

Frequency Daily Daily Daily Daily Daily Quarterly Quarterly

3.2 Analyzing the impact of the degree of sovereign exposure of banks on the bank-sovereign nexus

In order to test for the impact of the degree of sovereign exposure of banks, I include the sovereign exposure to total assets of every bank to a particular sovereign counterparty. The actual sovereign holdings of banks are available from March 2010 onwards in different stress-tests, transparency exercises and capital exercises in the EBA databases. The impact of sovereign exposures of banks on the bank-sovereign nexus is analyzed with the following regression equation:

𝜌_𝑏𝑠𝑡 = 𝐶 + 𝛽₁ 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒_𝑏𝑠𝑡+ ε_𝑏𝑠𝑡

𝐸𝑞𝑢𝑎𝑡𝑖𝑜𝑛 (2)

where 𝜌𝑏𝑠𝑡 is the excess correlation between bank and sovereign CDS spreads.

Explicitly, 𝜌𝑏𝑠𝑡 is the quarterly correlation of the residuals of daily changes in CDS spread of

bank 𝑏 and residuals of the daily changes in CDS spreads of sovereign 𝑠 which are not explained by fundamental market factors at time 𝑡, as described in section 3.1. This variable is defined as a measure of the bank-sovereign nexus. Therefore, I interpret a higher value for the excess correlation of CDS spreads of banks and sovereigns as a measure for the presence

20

bank 𝑏 to sovereign counterpart 𝑠 at time 𝑡 divided by the total assets of that particular bank

𝑏 at time 𝑡. ε𝑏𝑠𝑡 is the residual term. I take a step away from the method the Bruyckere et al.

(2013) by correcting the reported sovereign exposures for the total assets of a bank. In this way I circumvent possible scale effects. I perform the test also without correcting for total assets as a robustness check in Section 6.

3.3 Identifying the impact of the Zero-risk weight rule

To assess the impact of the zero-risk weight rule to sovereign exposure to EU member states I compare the impact of sovereign exposure of banks in the period with zero risk-weights for EU sovereign exposures prior to September 2011, with the period after September 2011 in which capital buffers against sovereign exposures are introduced. Particularly, I include a dummy in the model that takes a value 1 for the period after September 2011.

As stressed in Section 4.2, the EBA publishes detailed sovereign debt holdings of banks in Europe from March 2010 onwards. Hence, the dummy takes a value 0 for observations from March 2010 – September 2011. The reporting dates of the EBA where the dummy takes a value 0 are the following: March 2010, December 2010 and September 2011. The observations where the dummy takes a value 1 for the time period September 2011 – June 2015 contains the following EBA reporting dates: December 2012, June 2013, December 2013, December 2014 and June 2015. Comparing the time periods provides important insights in the regulatory treatment of capital against sovereign exposure on the link between credit default risks of banks and sovereigns. Accordingly, the following equation is used to estimate whether the announcement from the EBA in September 2011 significantly reduces the excess correlation between bank and sovereign CDS spreads:

𝜌𝑏𝑠𝑡 = 𝐶 + 𝛽1 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑏𝑠𝑡+ 𝛽2 𝑍𝑒𝑟𝑜𝑊𝑒𝑖𝑔ℎ𝑡 + 𝛽3 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑏𝑠𝑡

∗ 𝑍𝑒𝑟𝑜𝑊𝑒𝑖𝑔ℎ𝑡

𝐸𝑞𝑢𝑎𝑡𝑖𝑜𝑛 (3)

where 𝜌𝑏𝑠𝑡 is the excess correlation between the default risks of banks and sovereigns.

The excess correlation interpretation in equation 3 is similar to the interpretation in equation

2. 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑏𝑠𝑡 is the sovereign exposure of bank 𝑏 to sovereign counterpart 𝑠 at

21

the EBA in the September 2011. 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑏𝑠𝑡 ∗ 𝑍𝑒𝑟𝑜𝑊𝑒𝑖𝑔ℎ𝑡 is an interaction

variable.

The interpretation of the coefficient of an interaction variable specifies how the effect on the coefficient of one explanatory variable depends on the level of one or more other explanatory variables (Hill, Griffiths and Lim, 2012). The interaction variable between the dummy variable that takes a value 1 for the period after the EBA announcement and the banks’ sovereign debt exposure to total assets measures the impact of this announcement along with the degree of sovereign exposure of banks on the excess correlation between bank and sovereign CDS spreads. The economic interpretation of this coefficient is as follows: the extent to which the nexus between the default risks of banks and sovereigns would decrease or increase additionally with an increase in the degree of sovereign exposure to total assets after the announcement of the EBA in September 2011, compared to the period before the announcement.

A negative coefficient for the interaction between 𝑆𝑜𝑣𝑒𝑟𝑒𝑖𝑔𝑛𝑒𝑥𝑝𝑜𝑠𝑢𝑟𝑒𝑏𝑠𝑡∗

𝑍𝑒𝑟𝑜𝑊𝑒𝑖𝑔ℎ𝑡 implies that introducing additional capital buffers against sovereign debt exposure of banks lowers the excess correlation between bank and sovereign credit default risk when banks increase their sovereign exposure, compared to the period prior to the EBA announcement in 2011 in which risk-weights are absent.

4. Data

In this section I will briefly discuss which sources are used to collect the data. Furthermore, I will elaborate on specifications and restrictions of the different data used in this thesis.

4.1 Credit Default Swap spreads

22

Aizenman, Hutchison, and Jinjarak (2013) argue that CDS spreads have three main advantages compared to sovereign bond spreads. They claim that CDS spreads provide a better and timelier market based price. Secondly, they state that CDS spreads avoid the problem of dealing with interest rate spreads, since in that case zero-coupon bonds are preferred. Thirdly, they contend that bond spreads include market expectations for demand and supply for lending conditions next to default risk. Since I am mostly interested in the default risk of banks and sovereigns I prefer CDS spreads.

I use CDS spreads of banks and sovereigns on senior debt contract with 5 year maturities since these are the most traded and liquid CDS securities compared to other maturities (Aizenman, Hutchison, and Jinjarak (2013); Alter and Schuler (2012); Anderson (2011) and Barrios, Iversen, Lewandowska, and Setzer (2009)). I obtain daily CDS spreads for 32 banks and 24 sovereigns from Datastream. My sample of banks and countries is therefore restricted to the availability of CDS spreads in Datastream. In Appendix B and C I listed for which banks and sovereigns in the EBA sample CDS spreads on senior debt contracts with 5 year maturities are available.

4.2 Sovereign exposure

I extract the data on sovereign exposures from the EBA databases of stress-tests, capital exercises and transparency exercises. The data from the EBA is publicly available. The EBA data contain bank-by-bank data with exposures to all EU sovereign counterparties. The EBA published detailed bank-by-bank portfolio holdings on the following dates: March 2010, December 2010, September 2011, December 2012, June 2013, December 2013, December 2014, June 2015.

Unfortunately, the EBA uses different metadata and formats for identifying the banks and sovereign counterparts in their stress tests, capital exercises and transparency exercises. Hence, I match and merge all the bank identifiers from the different reporting dates in accordance with the EBA codes of the list presented in Appendix B. Furthermore, I transform all the bank-by-bank sovereign exposures to values in millions of Euro in order to be consistent. Next to that, I match all the sovereign identifiers from the different stress tests, capital exercises and transparency exercises in line with the codes in Appendix C.

23

and ISIN-numbers in order to match the EBA data on sovereign exposures with data from Datastream and Bankscope.

The data of the EBA contain information of the 91 largest banks in the Eurozone. However, in the 2010 and 2011 stress-test and transparency exercise only 60 banks were obliged to report to the EBA. After matching the 60 banks that report their portfolio holdings to the EBA with the availability of CDS spreads in Datastream, the sample reduces to 32 banks. In Appendix B I highlight for which banks I obtain EBA exposures for the transparency exercise and stress-test in 2010 and 2011 and for which banks CDS spreads are available in Datastream.

The EBA data contains bank-by-bank data of exposures to all EU member states. However, similar to my bank sample, the countries in my sample are also restricted to the availability of CDS spreads in Datastream. Therefore, my sample of sovereign counterparts is reduced to 24 countries which I present with the EBA codes in Appendix C.

I retrieve the total assets of the bank sample from Bankscope. I have quarterly data from 2010-2016 of the total assets of my bank sample in millions of Euros. I match this with the EBA exposures to calculate the explanatory variable of the degree of sovereign exposures to total assets that I use in equation 2 and 3. Again, I match the identifiers that are compatible with Bankscope with the EBA bank identifiers. In Table 2 descriptive statistics of the degree of sovereign exposure to total assets to a particular sovereign counterpart per EBA reporting date are presented.

Table 2. Descriptive statistics of the average exposures to total assets of a bank to a sovereign counterpart per EBA reporting date.

The exposures of a bank to a particular counterpart are retrieved from the EBA databases. These exposures are divided by the total assets of that bank, which are quarterly available in Bankscope.

24

5. Results

In Figure 3 the average sovereign exposures on the EBA reporting dates and the average of excess correlations for all banks in my sample in every quarter are presented. The announcement of the EBA in the third quarter of 2011 is highlighted. Over time, I see a steady rise in the average sovereign bond holdings of banks. The average excess correlation of all banks in the sample fluctuates substantially over time. There seems to be a downward trend in the excess correlations over time, whereas average sovereign exposures seem to have an increase over time. Especially, the downward spikes of the excess correlations from 2010 to September 2011 appear to have higher values compared to the downward spikes period from September 2011 to 2016.

Interestingly, there seem to be lower average excess correlations in the period after September 2011 compared to the period prior to September 2011, although average sovereign exposures have increased. This suggests that introducing capital requirements for sovereign exposures decrease excess correlations and the bank-sovereign nexus. In table 4, I find significant evidence for a decreasing effect of the announcement of the EBA on the excess correlation between bank and sovereign credit default risk. This is the main result of my thesis. In this section, I will discuss this main result and other results of my study on the bank-sovereign nexus.

Figure 3. Averages of quarterly excess correlations and sovereign exposures in millions for the EBA reporting dates for all banks in the sample.

.0 5 .1 .1 5 .2 .2 5 Ave ra g e exce ss co rre la tio n 5 00 1 00 0 1 50 0 2 00 0 2 50 0 Ave ra g e so v. e xp o su re (mi lli o ns) 2010Q1 2011Q1 EB 2012Q1 2013Q1 2014Q1 2016Q1 A 2015Q1 Quarter

25

Since my data contain excess correlations for every bank-sovereign pair for every quarter between 2010 and 2016 as explained in section 3.1, I can evaluate the impact of bank characteristics such as sovereign exposures on the bank-sovereign nexus. Furthermore, for every bank I have exposures to all sovereign counterparts separately for every EBA reporting date. I start with analyzing the impact of the degree of the sovereign exposure of a bank on excess correlations. In table 3 I regress the sovereign exposure to total assets of banks on the excess correlation between bank and sovereign CDS spreads.

Table 3. The impact of the degree of sovereign exposure to total assets of banks on the excess correlation between bank and sovereign CDS spreads for the period 2010-2016.

This table shows the impact of the degree of sovereign exposure to total assets of banks on the excess correlation of bank and sovereign CDS spreads. Excess correlation is calculated as the quarterly correlation in residuals of daily changes in bank and sovereign CDS spreads which are not explained by a set of fundamental market factors (MSCI, Vstoxx, Itraxx). In all of the regressions I control for bank-time fixed effects. The first column identifies the effect for my full sample of banks and sovereign counterparts. The second column only takes into account the exposures of banks to non-GIIPS sovereign counterparts. The third column shows only the results for the exposures of banks to GIIPS sovereign counterparts. By separating these two models, I want to distinguish between the impact of distressed countries on the excess correlations of banks and sovereigns. VARIABLES (1) Full sample Excess correlation (2) Non-GIIPS Excess correlation (3) GIIPS Excess correlation Sovereign exposure 0.802*** 0.702*** 1.770*** (0.261) (0.177) (0.558) Constant 0.158*** 0.149*** 0.230*** (0.015) (0.013) (0.030) R2 0.14 0.12 0.39 N Bank-time FE Cluster 4088 YES Bank 3236 YES Bank 852 YES Bank

26

Before running the regression I perform a Hausman test. I find significant evidence for using a fixed effects model instead of a random effects model. The p-value of the Hausman test is 0.0006. Therefore, the null hypothesis of the coefficients estimated by the efficient random effects estimator being equal to the ones estimated by the fixed effects estimator is rejected. As a consequence, a fixed effect model is preferred over random effects model (Hill, Griffiths and Lim, 2012). The Hausman-test is included in Appendix F. Bank-time fixed effects are included in all regression specifications to control for differences in the intercepts of banks and the different quarters. Using a bank-time fixed effects model is plausible since I am interested in analyzing the effect of sovereign exposures of banks that changes over time. Hence, my results are not biased by differences in constants of banks or specific quarters.

Additionally, a test for all dummy variables of the bank and time fixed effects being jointly equal to zero is done. The results for these tests are reported in appendix G. Since the p-values are significant at a 1% significance level, I reject the null hypotheses of all dummies for bank fixed effects and time effects being jointly equal to zero. Hence, bank and time fixed effects are needed in the fixed effects regression (Hill, Griffiths and Lim, 2012).

Furthermore, standard errors of the coefficients are clustered per bank to control for heteroscedasticity and autocorrelation (De Bruyckere et al., 2013; Hill, Griffiths and Lim, 2012). In the second and third column of table 3 I run the regression separately for exposures of banks to non-GIIPS and GIIPS countries in order to disentangle the effect of exposures to distressed GIIPS-countries on the excess correlations.

In table 3 I find that an increase in sovereign exposure to total assets of banks with 1% to a sovereign counterpart, increases excess correlation to the sovereign counterpart with 0.802%. This result for the complete sample of banks and sovereign counterparts is significant at 1% level. The economic interpretation of this coefficient is as follows: if a bank increases its degree of exposure to total assets to a sovereign counterpart with 1%, the excess correlation of the banks’ credit default risk and the credit default risk of the sovereign counterpart increases with 0.802%. The economic interpretation of the coefficients when regressing on exposures to GIIPS countries and non-GIIPS countries is similar to the interpretation above.

27

by exposures to distressed GIIPS-countries. Column 3 of table 3 indicates that the impact of the degree of sovereign exposure to total assets on the excess correlation is higher for the exposure of banks to GIIPS-countries compared to non-GIIPS countries. Particularly, when banks increase their GIIPS-sovereign exposure to total assets with 1%, the excess correlation increases with 1.77%. The coefficient is also significant at a 1% significance level again when running the regression only for exposures of banks to GIIPS-countries. In table 4, I test whether exposure to GIIPS-countries significantly increases excess correlations compared to non-GIIPS sovereign exposure.

From the results above, I reject the null hypothesis of hypothesis 1 of the coefficient for sovereign exposure being equal to zero with a one-sided test at a 1% significance level. Therefore, I conclude that banks that have a higher degree of sovereign exposure to total assets to a particular sovereign counterpart have a higher excess correlation between their default risk and that of their sovereign counterpart. Additionally, I find that this effect is not solely driven by exposures of banks to GIIPS-countries but it does also hold for exposures to non-GIIPS countries. These results imply that the bank-sovereign nexus as measured by excess correlations of bank and sovereign CDS spreads increases with the degree of sovereign exposure to total assets of banks.

De Bruyckere et al. (2013) find similar results when regressing the excess correlations of bank and sovereign CDS spreads with bank-time fixed effects on EBA exposures. I define the sovereign exposure to a particular sovereign counterpart as the gross sovereign exposure to total assets, whereas De Bruyckere et al. (2013) define this variable as the sovereign exposure to total sovereign exposure. Therefore, the magnitude of the coefficient in my study and the coefficient De Bruyckere et al. (2013) find is somewhat different. De Bruyckere et al. (2013) do not distinguish between exposures to GIIPS and non-GIIPS sovereign exposures, which I do explicitly.

28

To analyze the impact of the zero risk-weight rule on the bank sovereign nexus, a dummy that takes a value 1 for the period after the announcement of the EBA in September 2011 and the interaction term between the dummy and the sovereign exposure variable are incorporated in the model of equation 3 compared to equation 2. The results of this regression are presented in the first column of table 4.

In the second column of table 4 I add a dummy for exposures to GIIPS countries and the interactions all between the variables. The fourth regression assesses the impact of sovereign exposures to GIIPS-countries of banks without differentiating for the period prior to and after the EBA announcement. To check for possible problems with multicollinearity I calculate and present the correlation matrix of all the explanatory variables used in the second column of table 4 in Appendix H.

Multicollinearity can have consequences for the standard errors and coefficients of the model being inflated. Therefore, the sign and significance of the coefficients can be misleading (Hill, Griffiths and Lim, 2012). Especially, the interaction variables have a high pairwise correlation as Appendix H shows. Further analysis of the Variance Inflation Factor (VIF) that I present in Appendix I indicates that multicollinearity of the variables can be present for the regression in column 2. When the VIF value takes a value higher than 10 or the 1/VIF takes a value lower than 0.1, then there is an indication for the presence of multicollinearity (Craney and Surles, 2002; Hill, Griffiths and Lim, 2012). Therefore I omit two variables in the third column of table 4 to avoid problems with multicollinearity. The first, third and fourth specification do not seem to have issues with multicollinearity as none of the VIF values as presented in appendix I exceed the critical value.

29 Table 4. The impact of the degree of sovereign exposure to total assets, the zero risk-weight rule and GIIPS-sovereign debt on excess correlations between bank and sovereign CDS spreads for 2010-2016. This table shows the impact of a dummy that takes a value 1 after the announcement of the EBA in September 2011 on excess correlations. Excess correlation is calculated as the quarterly correlation between residuals of daily changes in bank and sovereign CDS spreads which are not explained by a set of fundamental market factors (MSCI, Vstoxx, Itraxx). In all regression specifications I control for bank-time fixed effects. Furthermore, the standard errors in all regressions are clustered by bank before and after the announcement of the EBA in September 2011. Except for column 4 where standard errors are clustered by bank since it covers the complete time period. The first column shows the impact of the Zero risk-Weight dummy and the interaction of Sovereign exposure with the Zero risk-weight dummy. The second column incorporates exposures of banks to GIIPS countries and the interactions between the variables. The third column omits two variables in order to avoid problems with multicollinearity. The fourth column identifies the impact of GIIPS sovereign bond holdings.

VARIABLES (1) Excess Correlation (2) Excess Correlation (3) Excess Correlation (4) Excess Correlation Sovereign exposure 1.276*** 0.882*** 0.942*** 0.353* (0.058) (0.080) (0.066) (0.206) ZeroWeight -0.115*** -0.112*** -0.115*** (0.013) (0.014) (0.013)

Sovereign exposure * ZeroWeight -0.669*** -0.703*** -0.747***

(0.168) (0.200) (0.185)

GIIPS 0.120*** 0.111*** 0.117***

(0.001) (0.010) (0.015)

GIIPS * Sovereign exposure 0.045 0.273*

(0.095) (0.144)

ZeroWeight * GIIPS -0.015

(0.014)

ZeroWeight * GIIPS * Sovereign exposure 0.291

(0.172) 0.302* (0.161) Constant 0.157*** 0.135*** 0.136*** 0.136*** (0.022) (0.022) (0.023) (0.014) R2 0.14 0.20 0.20 0.21 N 4088 4088 4088 4088

Bank-time FE YES YES YES YES

Cluster Bank * ZeroWeight Bank * ZeroWeight Bank * ZeroWeight Bank

30

In the first column of table 4 I find that the main coefficient of interest, the interaction of the degree of sovereign exposure to total assets with the dummy variable has a negative sign and is significant at a 1% significance level. The coefficient can be economically interpreted as follows. After the announcement of the EBA in September 2011, a 1% point increase in the sovereign exposure to total assets of banks is, ceteris paribus, related to a decrease in the excess correlation of bank and sovereign default risk of 0.669% compared to the period prior to the announcement of the EBA in September 2011.

Next to that, the dummy variable that takes a value 1 for the period after September 2011 has a negative sign and is significant at a 1% significance level as well. The negative value for the dummy coefficient implies a decrease in the intercept of 0.115% on the excess correlations after the announcement of the EBA compared to the period prior to September 2011. The economic interpretation of the coefficient of the dummy is as follows: the average excess correlation between bank and sovereign credit default risk decreases with 0.115% after the EBA announcement in September 2011 compared to the period prior to the announcement.

The results from the first column in table 4 are in line with hypothesis 2. I hypothesize that after the announcement of the EBA in September 2011 in which additional capital requirements for sovereign exposures are introduced, the excess correlation between the default risks of banks and sovereigns decreases. The null hypothesis of no impact of the announcement of the EBA on the bank-sovereign nexus of hypothesis 2 is rejected with a one-sided test at a 1% significance level. Hence, I conclude that the bank-sovereign nexus decreases as a result of the introduction of capital requirements for EU sovereign exposure of banks. This is the main result of my study. My findings imply that the bank-sovereign nexus is increased when risk-weights to sovereign exposures are absent compared to the period where capital requirements against EU sovereign exposures are introduced.

31

requirements for EU sovereign exposure decreases risk spillovers of bank and sovereign credit default risks.

To conclude, my study is the only study that empirically analyzes the impact of zero-risk weights to EU sovereign debt on the bank-sovereign nexus. Particularly, I find that the impact of sovereign exposures of banks in the period where risk-weights to sovereign exposures are absent prior to the announcement of the EBA in September 2011 increase the bank-sovereign nexus compared with the period after the announcement in which capital requirements for sovereign exposures are introduced. These are the main findings of my thesis. In a following step the impact of sovereign exposures to distressed GIIPS-countries is analyzed.

In the second column I include a dummy that takes a value 1 for exposure to GIIPS-countries. In the second column I find similar results for the coefficients that I use in column (1). However, due to multicollinearity problems as elaborated earlier in this section and indicated in Appendix H and I, the coefficients and significance of these coefficients can be inaccurate. Besides the dummy for exposure to GIIPS-countries, none of the added variables is significant at any reasonable significance level. By omitting two variables in the third column, the interaction between variables that represents the exposure of banks to GIIPS-countries after the announcement of the EBA in September 2011 becomes significant at a 90% confidence level. The coefficient has the similar sign and magnitude compared to column 2 of table 4.

The economic interpretation of this coefficient is as follows: after the announcement of the EBA in September 2011, a 1% increase in the exposure of banks to GIIPS-countries increases the excess correlation of bank and sovereign credit default risk by 0.302% more than non-GIIPS sovereign exposure compared to prior September 2011. I reject the null hypothesis of hypothesis 4 of no change in the coefficient for GIIPS-sovereign exposure after the announcement at a 10% significance level. However, the sign is the opposite of what I expect. Therefore, I am not able to confirm hypothesis 4 where I expect that the coefficient as GIIPS sovereign exposure as after September 2011 decreases more than non-GIIPS sovereign debt compared to the period prior to 2011.

32

sovereign counterpart is still increased if banks increase their exposure to the riskier GIIPS sovereign debt compared to non-GIIPS sovereign debt.

There is no existing literature that is related to this finding. Hence, I conclude that the riskier sovereign debt holdings of banks, even after the introduction of capital requirements, have an increasing effect on excess correlations compared to non-GIIPS sovereign exposure. In a following step I will empirically evaluate whether exposure to GIIPS-countries significantly increases excess correlations compared to non-GIIPS sovereign exposure over the complete time period.

In the fourth column of table 4 I find that sovereign debt holdings of GIIPS sovereign debt have a positive and significant effect on the excess correlations between the bank and the sovereign counterpart. The sovereign exposure coefficient to non-GIIPS countries remains significant, although at a 10% significance level. The economic interpretation is similar to the interpretation in table 3 and can be interpreted as follows: when a bank increases its sovereign exposure to total assets to a non-GIIPS country with 1%, the excess correlation to the sovereign counterpart increases with 0.353%. The dummy variable that takes a value 1 if the sovereign counterpart is a GIIPS-country is significant at a 1% significance level. The dummy can be interpreted as follows: exposure to GIIPS-countries increases the intercept of excess correlation between bank and sovereign credit default risk on average with 0.117% compared to the intercept of non-GIIPS sovereign exposure.

The coefficient for the interaction between the GIIPS-dummy and sovereign exposures is significant at a 10% significance level. This coefficient can be interpreted as follows: an increase of 1% of a banks’ GIIPS-sovereign exposure to total assets leads to an increase in the excess correlation of 0.273% compared to non-GIIPS sovereign exposure. Therefore, I reject the null hypothesis of hypothesis 3 of no change in GIIPS-sovereign exposures on excess correlations, compared to non-GIIPS sovereign exposure. Herewith, I conclude that exposure to GIIPS countries increases the bank-sovereign nexus compared to exposure to non-GIIPS countries. This finding has similarities to existing literature on investments in GIIPS-sovereign debt.

33

sovereign CDS spreads. Secondly, Acharya, Engle and Pierret (2014) find that banks are incentivized to invest in higher-yielding assets that have inaccurate risk-weights. Finally, Korte and Steffen (2015) find that risk spillovers between sovereigns are larger when the credit rating of the sovereign counterpart deteriorates. In my research significant evidence is presented for the actual sovereign bond exposures of banks to GIIPS-countries leading to an increase in risk spillovers of bank and sovereign credit default risks compared to non-GIIPS sovereign exposures.

To conclude, in this thesis I find significant evidence for an increase in excess correlations between bank and sovereign credit default risks due to sovereign debt holdings of banks. Furthermore, significant evidence is presented for an increase in excess correlations as a result of GIIPS-sovereign exposure compared to non-GIIPS sovereign exposure. My main finding is that after September 2011, where capital requirements for sovereign exposures are introduced, the impact of sovereign exposure on excess correlations decreases compared to the period in which risk-weights to sovereign debt are absent. Implications of these findings and policy recommendations will be discussed in Section 7.

6. Robustness

In this section I will briefly discuss robustness checks of my results in Section 5 by using other variables than in the regressions in Section 5. Instead of regressing the models in table 3 on the degree of sovereign exposures to total assets, I regress the excess correlations of bank and sovereign CDS spreads on the actual exposures as reported in the EBA stress-tests, capital- and transparency exercises. Furthermore, I include bank-time fixed effects and cluster the standard errors by bank. The results are presented in Appendix J. The signs and significance of the coefficients without controlling for the degree of the sovereign exposure to total assets are similar to my findings in table 3.

34

The same robustness checks are performed for my findings in table 4. In Appendix K the results by making use of gross exposures instead of the degree of sovereign exposures to total assets are presented. The interpretation and magnitude of the coefficients is the same as described above. Furthermore, the signs and significance for the models with and without correcting for the total assets of the bank are similar. The coefficient for the variable of the interaction between the zero-weight dummy and the gross sovereign exposures of banks is significant at a 1% significance level and can be interpreted as follows. After the announcement of the EBA in September 2011, a 1 million increase in the sovereign exposure of banks to a sovereign counterpart is, ceteris paribus, related to a decrease in the excess correlation of a bank and its sovereign counterpart default risk of 0.003256% compared to the period prior to the announcement of the EBA in September 2011. Furthermore, the results I find in column (4) of Appendix K have the same implications as my results in column (4) of table 4.

Additionally, the results in column (2) and (3) of Appendix K are similar to my findings in table 4. Interestingly, the coefficient for the interaction between the variables that represent the exposure of banks to GIIPS-countries after the announcement of the EBA in September 2011 is significant in both specifications at a 99% confidence level. The interpretation of this coefficient is as follows. After the announcement of the EBA in September 2011, a 1 million increase in the exposure of a bank GIIPS-countries increases the excess correlation of that bank and its sovereign credit default risk on average by 0.002387% compared to exposure to non-GIIPS countries.