Systemic risk contribution of Global Systemically Important Banks.
An assessment on the influence of the introduced additional loss
absorbency requirement
Amsterdam Business School
Name J.L.C. Alberts
Number 10190090
MSc in Business Economics Specialization Finance
Supervisor prof. dr. E.C. Perotti Completion 15-08-2016
Abstract
This paper assesses whether the additional loss absorbency requirement which only applies to G-SIBs has the desired effect of reducing the negative externalities and enhancing the financial system. By applying quantile regression methodology the CoVaR is constructed as a measure of systemic risk contribution. Using OLS regression methodology to a sample of the 70 largest banks globally it is assessed whether G-SIBs have reduced their systemic risk exposure more than other banks by increasing their capital. Overall, this paper determines that increasing capital does not affect the change in systemic risk contribution directly.
Statement of originality
This document is written by student Laura Alberts who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the texts and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1. Introduction ... 3
2. Literature review ... 5
2.1. Systemic Risk ... 6
2.2 Basel III ... 8
2.3 Conditional Value at Risk (CoVaR) ... 13
2.4 Hypothesis ... 15
3. Methodology and Data ... 16
3.1. Methodology ... 16
3.1.1 ΔCoVaR estimation ... 16
3.1.2 Models ... 18
3.2. Data and descriptive statistics ... 21
4. Empirical results ... 23
4.1.1. Results on assessing the VaR ... 24
4.1.2. Results on assessing the CoVaR ... 25
4.1.3. CoVaR versus VaR ... 26
4.2. Robustness ... 27
5. Conclusion and discussion ... 28
References ... 30
Appendices ... 32
Appendix A: List of G-SIBs ... 32
Appendix B: Figures ... 33
Appendix C: Empirical results ... 34
1. Introduction
As experienced in the latest global financial crisis from 2008, although financial institutions were failing, but being too big to fail or even too systemic to fail a
government bail-out was inevitable. Besides this, these banks also passed on shocks to other banks and the financial system which eventually led to a spillover to the real economy and causing the global recession. These institutions, now benefitting from this status, can get funding more easily and at lower costs. Banks involved, are favored by the uninsured creditors and other participants in the market, even though they have no claim in case of failure (Brewer and Jagtiani, 2007). Because of this favored position, these financial institutions are incentivized to grow even bigger and become more complex and interconnected. Although these banks benefit from their position, bailing out these banks in case of distress comes at high costs, eventually to be paid by society as a whole. Systemically important financial institutions are aware of their importance and incentivized to take excessive risks to expropriate their wealth, knowing they will be bailed out anyway (Stern and Feldman, 2004). To reduce this moral hazard problem - preventing banks from failing and letting the taxpayer absolve this - the Basel Committee on Banking Supervision (BCBS) and the Financial Stability Board (FSB) have introduced two special measures for these systemically important financial institutions in the Basel III agreements. The first measure involves tighter supervision and resolution regimes. The second measure is a capital adequacy surcharge, which only applies for the identified global systemically important banks (G-SIBs). The aim of this requirement is to guard the financial system against the risks of spillovers of G-SIBs and the negative externalities they have (BCSB, 2011b). This capital adequacy requirement is called the additional loss absorbency requirement. This loss absorbency surcharge consists of additional Common Equity Tier 1 (CET1) capital ranging from 1% to 3,5% as a percentage of risk-weighted assets banks have to hold additional to the other capital requirements. Every year the Financial Stability Board (FSB) publishes an updated list of the G-SIBs identified by the BCBS. On this list, the G-SIBs are placed into one of the five buckets, where each bucket has a different level of systemic risk and percentage of additional required capital.
This paper investigates whether this additional loss absorbency requirement not only serves its purpose of paying off more creditors in case of impairments or default, but also of reducing the probability of default or impairment and therefore enhancing the financial system reducing spillover risks. It is tested whether increasing capital reduces the systemic risk contribution of banks. Therefore, the research
question is as follows:
Has the systemic risk contribution of Global Systemically Important Banks reduced due to the implementation of the additional loss absorbency requirement in 2011?
This study uses the conditional value at risk (∆CoVaR), defined by Adrian and Brunnermeier (2011), as a measure of systemic risk contribution. The CoVaR estimates the risk of the financial system, conditional on banks in a stress situation. The systemic risk contribution of a bank (∆CoVaR) is defined as the difference between the CoVaR of the financial system conditional on a bank in distress and the CoVaR of the financial system conditional on a bank in the median situation. To estimate the ∆CoVaR, quantile regressions are performed on weekly data of the growth in market valued assets. An OLS regression is conducted to answer the research question using a panel data with a yearly frequency on 70 global banks. Figure 1 shows that the systemic risk contribution for all banks has reduced, since 2011 and therefore the financial system has become safer since 2011. This paper has analyzed whether this effect is greater for G-SIBs due to the introduction of the additional loss absorbency requirements.
Figure 1: This figure shows the 1%-CoVaR for the sample over time. This risk measure is in percent returns to market valued assets.
Results from this assessment reveals that G-SIBs have not reduced their systemic risk contribution by raising their capital as required by the additional loss absorbency requirement. G-SIBs in general have reduced their systemic risk contribution less than other banks. This means that the additional loss absorbency requirement itself does not serve its purpose of reducing the negative externalities that G-SIBs pose, when using the ∆CoVaR as measure for systemic risk. However, in case of impairment or default more creditor can be paid with the additional capital.
The analysis in this paper are relevant for multiple reasons. In the first global banking reforms, Basel I, capital adequacy requirements were introduced to reduce the risks that banks pose because their leverage ratio is very high. Blüm (1999) however, has found that these capital requirements may increase risk instead of reducing it. Therefore, it is important for the regulators to know whether this new this new requirement, which only applies to G-SIBs, reaches its desired effects of
reducing the negative externalities of the G-SIBs. Since this requirement is quite new, the relation of the effect of the additional loss absorbency requirement and systemic risk exposure has not been investigated yet. This is also a limitation to this analysis, since this requirement is not fully implemented yet.
To answer the research question this paper follows several steps. This thesis is structured as follows: the second section provides an overview of the literature
relevant to this research where after the hypotheses are defined. The third section provides the methodology and data used to assess the research question. Section four provides an analysis of the resulting regression estimates and the final section
provides a conclusion of this research.
2. Literature review
In this section the relevant literature regarding systemic risk, global systemically important banks and the additional loss absorbency requirement is discussed. First, the existing literature on systemic risk is discussed. Thereafter, the rules for banks, stated in Basel III are reviewed. Thirdly, the systemic risk variable conditional value at risk (CoVaR) is discussed. Finally, the results of the existing literature is converted into two hypotheses which are used to answer the research question of this paper.
2.1. Systemic Risk
There are many definitions of systemic risk. This paper uses the definition set by the International Monetary Fund, the Bank for International Settlements, and the
Financial Stability Board (2009). They have defined systemic risk as “a risk of
disruption to financial services that is (i) caused by an impairment of all or parts of the financial system and (ii) has the potential to have serious negative consequences for the real economy”.
Gavin and Hausmann (1996) explain that bank characteristics are making the financial sector more vulnerable to a crisis than other sectors. The most important characteristic that causes vulnerability to the financial system is that banks are
astoundingly financed by debt. Banks are levered for at least 90%, which implies that the bank’s equity is only 10%. This has two implications. First, because shareholders have limited liability, risk-shifting often occurs. This means that the bank’s
management is not acting in the best interest of the debt holders, by taking excessive risks because this maximizes the shareholders’ value. These shareholders do not bare the downside of risk. The reason for this is that in case of default, debt holders are paid first and the shareholders will end up with nothing, because not enough money is left. Therefore taking high risks benefits the shareholders more on average than taking safer investments. Second, a small shock can easily lead to an insolvent situation. Where other institutions can use their capital as a buffer, banks cannot bear this because of their limited capital. Another bank characteristic is illiquidity of a bank’s assets. Gavin and Hausmann (1996) argue that most banks’ business models are based on the transformation if short term maturities on the liability side of the balance to the long term maturities to the asset side of the balance. They take short-term deposits, which are highly liquid and sell long-term loans and mortgages. As long as the short-term depositors do not withdraw their money, the bank has no funding problem. However, if the depositors do not roll over their deposits, the bank has no money to finance the long-term investments. The risk associated with this is called liquidity risk.
The latest financial crisis started with a shock on the financial markets, causing large financial institutions to become insolvent. However, some of these institutions were too big to fail (TBTF), because the impact of this failure would have affected so many stakeholders that the government has acted as lender of last resort in order to prevent them from the negative effects. Other financial institutions where
even too systemic to fail (TSTF). This does not only means that the institution is large, but also that it has a central function in the market, has a complex structure, is interconnected with other institutions and has cross border activities (Bongini et al. 2015). Not bailing out and letting this bank go into bankruptcy would lead to contagion to other banks.
Nier et al. (2007) describe four contagion mechanisms in which failure or weakness of parts of the financial system arise. The first is weakness is correlated exposures to different sources of risk. Second, cross holdings and direct exposures among banks which can cause a domino-effect of banks failing because of their interconnectedness. Thirdly information contagion, which implies that when a bank is in default it will adversely affect other banks because one might think default or impairment will happen to other banks. Last, fire sales of assets, implying that one bank starts to sell its assets, the asset prices will depress causing margin calls and other banks needing to sell their assets as well. This can end up in a downwards fire sale spiral. This may not only affect other banks, but it may affect the whole
economy, as experienced in the latest crisis. Therefore, governments and regulators try to prevent this from happening, also because bailing out comes at a high cost payed by society.
Although there are many negative effects from this status of TBTF and TSTF, with default being the worst effect, these financial institutions also benefit from this status. The reason for this is that it is easier to get new funding possibilities and the costs of funding are lower than for banks without the TBTF/TSTF status. These banks are favored by all parties, but what is striking, is that they are also favored by the uninsured creditors and other participants in the market, even though they have no claim in case of failure (Brewer and Jagtiani, 2007). Since this failure will always tried to be circumvented, the chance of failure will be reduced. G-SIBs try to maintain the status of TBTF/TSTF and other banks try to become TBTF/TSTF to profit from the cheaper funding as well, by growing bigger, become more complex and
interconnected, because they will be bailed out anyway (Stern and Feldman, 2004). To gain and maintain the importance status, these banks take excessive and,
sometimes, irresponsible risks, causing the chance of failure to increase. Therefore regulators must try to regulate these banks in order to prevent defaults leading up to the consequences of a breakdown of the financial system and possibly the economy.
2.2 Basel III
To prevent a breakdown of the financial system and spillovers to the real economy, among countries itself, the Basel committee, is trying to regulate the global financial system. After the breakdown of the Bretton Woods system in 1974, a committee was set up by the G10 central bank governors as a response to the disruption in the international financial markets. The aim of this committee was strengthen the
financial system by ameliorating supervision and regulation. This led to the first Basel rules (Basel I), which was implemented in December 1992. The aim of this accord was setting minimal capital requirements for banks in order to absorb losses and guard against credit risk (BCBS, 2001). Basel I was superseded by Basel II in 2004 to extend the existing rules in order to guard banks against operational and financial risk as well. The revised the capital framework consisted of three pillars which are:
1. minimum capital requirements;
2. supervision which reviews the capital adequacy and internal assessment process of each financial institution and;
3. market discipline including disclosure requirements which are transparency and enhanced comparability among banks.
Although Basel II has been a sufficient improvement compared to Basel I, there are shortcomings. The most important shortcoming is the pro-cyclicality of the
requirements. This means that during a boom, the perceived risks are small leading to low capital requirements. During a bust, perceived risks are large leading to high capital requirements. The latter can provoke insolvency which could lead into a credit crisis (BCBS, 2001).
The latest financial crisis and the spillover to the real economy has led to the new reforms, Basel III, developed by the Basel Committee on Banking Supervision (BCBS). Basel III has been introduced in 2011 to promote a more resilient financial system supersede the Basel II. The new reforms should improve financial institutions’ ability to absorb losses, which could be arising from financial or economic shocks, and reduce the risk of a spillover to the financial system or the real economy. Such a resilient and stable financial system is the basis for economic growth, because banks and other financial institutions are the intermediary between investors and savers. Their role is essential in supplying credit to consumers and all sizes of corporations (BCBS, 2011a). The regulatory capital framework in Basel III builds on the framework with the three pillars from Basel II.
The first pillar, which are the capital requirements, now consists of three parts. The first part is capital, in which not only the quantity, but also the quality of the capital requirements is enlarged. A capital conservation buffer is introduced which contains a buffer for additional 2.5% common equity of the risk weighted assets. Also a countercyclical buffer has been brought in. When the authorities decide that the growth of credit has no proper build up then banks can be imposed with holding additional common equity up to 2.5% as a percentage of risk weighted assets. With this buffer the authorities address the pro-cyclicality of systemic risk (BCBS, 2011a). The second part is risk coverage, which has been introduced due to large losses for the trading book and the exposures of securitization by banks operating internationally. The third part is the leverage ratio which is not based on any risk weights.
The second pillar is fortified with additional requirements focusing on
governance throughout institutions and risk management. This includes a framework for dealing with different types of risk. The requirements of the third pillar of market discipline are revised and are more detailed. Besides the disclosure requirements, the focus lies now on the off-balance sheet items and on clarification of the calculations institutions use to compute the equity ratios (BCBS, 2011a).
In addition to these three pillars, liquidity standards are established. These are the net stable funding ratio(NSFR), the liquidity coverage ratio (LCR) and
supervisory monitoring. The purpose of the NSFR is to tackle liquidity mismatches and to encourage financial institutions to adopt stable funding. The NSFR requires financial institutions to have enough stable funding to overcome one year of stress (BCBS, 2014). The purpose of the (LCR) is to have enough high quality liquid assets to exceed the cash outflows the coming 30 days conditional on a stress scenario (BCBS, 2013). All these reforms are compulsory for all banks, however, due to the fact that large global financial institutions are TBTF and TSTF, they pose greater risk to the financial system, the BCBS has introduced additional measures for these banks. These are defined below.
2.2.1 Global Systemically Important Banks
The new reforms in general are neither sufficient to protect the system from the negative externalities these G-SIBs pose, nor to protect the system from spillover risks to other financial institutions or the real economy. Therefore, two special policy measures for global systemically important banks are adopted in the Basel III accords.
According to their business models, these banks are more involved in capital market activities, which impose more risks (BCBS, 2013) .
The first measure is the improvement of the global recovery and the resolution frameworks which has the aim to reduce the impact of a failure of SIFIs. These come in the form of tighter supervision and enhanced resolution regimes. The banks and other financial institutions which are subjected to the resolution regime should pose significant spillover risk to the financial system when this bank is deteriorating or failing (FSB, 2011).
The second measure is the additional loss absorbency requirement which has the aim of encouraging these GSIBs to lower their systemic risk exposure over time (BCBS, 2013). This capital surcharge seeks to reduce the probability and impact of default of these G-SIBs. The loss absorbency requirement requires G-SIBs to increase their ability to absorb losses and therefore need to hold more capital additional to the other capital adequacy requirements Basel III oppose. This loss absorbency surcharge consists of additional Common Equity Tier 1 (CET1) capital ranging from 1% to 3,5% as a percentage of risk-weighted assets banks have to hold additional to the other capital requirements.
The G-SIB banks are identified by the FSB based on the indicator-based measurement approach, see table 1. This approach consists of five elements of global systemic importance (BCBS, 2013),
1. Size: It is evident that the larger the bank the more likely it is damaging the entire financial system. Therefore, size and systemic importance are positively related.
2. Complexity: When bank is structurally and operationally complex, it will cost more time and money to resolve the bank. This is a positive relation between complexity and systemic importance.
3. Interconnectedness: The greater the interconnectedness of a defaulting bank, the more likely other banks which are connected to this bank through contracts will be in distress as well. Higher interconnectedness means higher.
4. Substitutability: With this element the FSB looks at whether a bank’s activities can easily be replaced by other banks’ activities in case of defaulting. The higher the substitutability, the lower the systemic importance rating.
5. Cross-border activities: The higher the degree of global activities, the more arduous it is to resolve this bank and therefore the degree of cross-border activities is positively related to systemic importance.
Table 1: Indicator-based measurement approach
Category (and weighting) Individual indicator Indicator weighting Cross-jurisdictional activity
(20%)
Cross-jurisdictional claims 10%
Cross-jurisdictional liabilities 10% Size(20%) Total exposures as defined for
use in the Basel III leverage ratio
20%
Interconnectedness (20%) Intra-financial system assets 6.67% Intra-financial system
liabilities 6.67%
Wholesale funding ratio 6.67%
Substitutability/financial institution infrastructure (20%)
Assets under custody 6.67%
Payments cleared and settled
through payment systems 6.67%
Values of underwritten
transactions in debt and equity markets
6.67%
Complexity (20%)
OTC derivatives notional value 6.67%
Level 3 assets 6.67%
Held for trading and available
sale value 6.67%
Based on the scores of systemic importance, the G-SIBs are allocated into five buckets where each bucket has a different level of additional loss absorbency
requirement. Every year the Financial Stability Board (FSB) publishes an updated list of the identified G-SIBs. The first list was published in 2011 (see Appendix A) and the banks were not placed into buckets yet. There are five buckets where one is the lowest bucket and five is the highest, see table 2.
This surcharge is implemented through an extension of the capital
conservation buffer. Meaning that every year, starting in 2015, the G-SIBs have to suffice a percentage of the surcharge to be fully implemented in 2019.
Table 2: Global systemically important banks bucketing approach Bucket number Additional surcharge as a % of CET1
5 3.5%
4 2.5%
3 2.0%
2 1.5%
1 1.0%
Note: no G-SIBs have been allocated in the fifth bucket
The existing literature is inconclusive about whether a capital adequacy requirement provides the right effects and incentives of reducing risks. Blüm (1999) has assessed this issue and found that capital adequacy requirements, in general, may increase risks which is the opposite of the desired effects. The reason for this is that increasing capital reduces a bank’s profits, since most banks’ business models are based on the maturity transformation. Therefore for every dollar of equity, means a dollar less to transform maturities with. To gain as much profits as before the capital requirements, one must invest in riskier assets.
Abreu and Gulamhussen (2013) have tested whether the additional capital surcharge for G-SIBs provides the right incentives to decrease their systemic risk contribution over time. Using an event study, they have assessed the market reaction to the labeling of banks as G-SIBs. Their expectation was a negative reaction from the market on the announcement of the FSB, because of the addition capital these banks need to hold because of the additional loss absorbency requirement and the extra supervision and resolution schemes. However, with their sample of 119 banks worldwide, they have found no evidence of abnormal returns on the stock markets. Therefore the market did not believe that these G-SIBs did not lower their moral hazard issue.
Bongini et al. (2015) also used an event study to test whether the stock prices of banks classified as SIFIs reacted in another way than other large banks which were not classified as being systemically important. They hypothesized that the SIFI banks are burdened with additional and tighter regulation and therefore, their stock prices will act negatively on the announcements. This is also called the ‘regulatory burden hypothesis’. They have reviewed two announcement dates: July 2011, which was the first announcement on the new regulation for SIFIs and the second announcement was on November 4th in 2011 in which the FSB published the list of G-SIBs. With a
sample of the 70 largest banks in the world, they have found that the stocks do not react unambiguously to the first announcement. Nevertheless, they have found that the reaction to the second announcement that systemically important banks with lower capital adequacy than what is required within one year experience negative abnormal returns, where systemically important banks with high capital adequacy benefits with positive abnormal returns. This supports their the regulatory burden hypothesis. These two papers do not give a univocal conclusion on how the market has reacted to the announcements of banks being labeled as systemically important.
2.3 Conditional Value at Risk (CoVaR)
Before the financial crisis in 2008, Value-at-Risk was a methodology commonly used by authorities to statistically measure and quantify the exposure of an individual (financial) institution to market risk for regulatory purposes (Linsmeier & Pearson, 2000). The q%-Value-at-Risk is the maximum loss with a confidence level of q% (Jorion, 2006). For example, the q%-VaR for a variable X can be estimated by,
𝑝𝑟 (𝑋𝑖 ≤ 𝑉𝑎𝑅
𝑞) = 𝑞,
where Xi is the variable of individual bank I for which the VaRq is calculated. This measure has been under severe criticism since it only gauge the risk of a single financial institution in an isolated situation. However, it does not measure the risk of the financial system conditional on the institution’s risk (Girardi & Ergün, 2013). Therefore, many different studies have tried to develop an alternative measure for risk, which does include the systemic nature of risk.
Adrian and Brunnermeier (2011) have developed such an alternative systemic risk measure which they have called the Conditional Value at Risk (CoVaR). This paper follows Adrian and Brunnermeier’s (2011) paper in defining the CoVaR and in this section a short review on this measure is provided. The construction and
estimation of the CoVaR is provided in section 3.1.1. They have described the CoVaRj|i as the VaR of institution j, conditional on the VaR of institution i. In this case institution j is the financial system thus CoVaRsystem|i shows what happens with the VaR of the financial system when an individual institution reaches its VaR, which means that this institution is under stress. The other way around, CoVaRi|system
describes the institution’s VaR when the financial system reaches its VaR, thus when a financial crisis happens. With this measure, one can see which financial institutions should fear most for default in case of a financial crisis.
With the CoVaRsystem|i, the contribution to systemic risk of each financial institution to the financial system can be estimated. The purpose of measuring systemic risk contribution measuring the spillover effects of a financial institution to the financial system. This could serve as a fundament for regulatory actions against systemically important financial institutions, such as an additional capital surcharge or tighter supervision (Adrian & Brunnermeier, 2011). This systemic risk contribution is defined with the following definition,
∆𝐶𝑜𝑉𝑎𝑅𝑠𝑦𝑠𝑡𝑒𝑚|𝑖 = 𝐶𝑜𝑉𝑎𝑅𝑠𝑦𝑠𝑡𝑒𝑚|𝑉𝑎𝑅𝑎𝑡 𝑟𝑖𝑠𝑘𝑖 − 𝐶𝑜𝑉𝑎𝑅𝑠𝑦𝑠𝑡𝑒𝑚|𝑉𝑎𝑅𝑠𝑡𝑒𝑎𝑑𝑦 𝑠𝑡𝑎𝑡𝑒𝑖 With this definition, Adrian & Brunnermeier (2011) have set up three different models for estimating a unconditional and time consistent CoVaRsystem|i, a conditional time-varying CoVaRsystem|i and a forward- CoVaRsystem|i. The latter is used to estimate future systemic risk contributions.
Girardi and Ergün (2013) have used univariate and bivariate GARCH models to estimate the ∆CoVaR to test which industry (depository institutions, broker dealers, insurance companies and other financial institutions) contributes most to systemic risk during the sample period June 2000 to February 2008. Using 74 financial institutions in the United States, they have found that depository institutions contribute most to systemic risk, then the broker-dealers, insurance companies and last the other financial institutions.
Lopez-Espinosa et al. (2012) have used the same ∆CoVaR to identify the main determinants of systemic risk using a sample of 54 large and complex banks from 18 different countries from 2001Q3 till 2009Q4. They have found that wholesale funding is positively and significantly related to the ∆CoVaR and thus systemic risk, whereas other firm characteristics they have tested, including size and leverage, are not as strongly related as wholesale funding. This variable could be closely linked to interconnectedness and exposure to liquidity risk. Therefore, this paper supports the way the Basel committee identifies the G-SIBs and penalizing these with requiring additional capital.
Castro and Ferrari (2014) have used the ∆CoVaR from Adrian and
Brunnermeier (2011) to set up a model with a time variant (daily) ∆CoVaR to test if a financial institution should be classified as being a systemically important financial institution based on this institution’s contribution to systemic risk. They have also tested whether banks in higher buckets contribute more to systemic risk than banks in lower buckets. Their sample consists of 26 large European banks over a period from the year 1999 to 2012. They have found that, that only some banks in their sample, can be classified as systemically important based on their contribution to systemic risk. They do find that bank properties, in this case size, support the statistical
significance of the bank rankings. Their recommendation to policy makers is that they should use statistical tests in assigning banks into one of the five buckets.
2.4 Hypothesis
Both studies on the market reaction on the SIFI regulation announcements as well as studies on the systemic risk contribution of SIFIs to the financial system have not been univocal in their results.
Two hypothesis are set up to test the risk exposure of banks. The first
hypothesis is used to test whether G-SIBs have reduced their individual risk exposure and therefore reduced their own chance of default, more than banks which are not subjected to the loss absorbency requirement. Therefore, the hypothesis is,
H0: Global systemically important banks have not experienced a reduced individual
risk exposure by having to increase their capital ratio more than non-GSIB banks since 2011.
If the regression results are consistent with this hypothesis then increasing capital does not affect the individual risk exposure and therefore does not make the G-SIBs any safer. However, with more capital more creditors can be paid off in case of impairment or default.
For the additional capital requirement to work not only as a loss absorbency buffer, but also to reduce the negative externalities and spill risks, G-SIBs should have reduced their systemic risk exposure more than other banks. Therefore, the
second hypothesis tests whether G-SIBs have reduced their risk exposure more than other banks.
H0: Global systemically important banks have not experienced a reduced their
systemic risk contribution by having to increase their capital ratio more than non-GSIB banks since 2011.
When the regression results are consistent with the hypothesis, this means that the G-SIBs have not reduced they systemic risk contribution. However, this is not in line with the intentions of additional loss absorbency requirement for the G-SIBs, to address the negative externalities these G-SIBs pose and to protect the financial system from spillover risks.
3. Methodology and Data
This section describes the methodology and data used to test the relation between change in equity and systemic risk contribution. First, the construction of the ΔCoVaR is provided. Afterwards, the models for testing the hypothesis are defined and the last part consists of the descriptive statistics.
3.1. Methodology
3.1.1 ΔCoVaR estimation
There are several ways to calculate the ΔCoVaR. To capture all forms of risk, in this paper, the same measure as Adrian and Brunnermeier is used, which is the quantile regression model. In this way funding liquidity risk and adverse assets price
movements are captured by basing the CoVaR on changes in total asset values of publicly traded financial institutions. In this analysis, only publicly available data is used. The banking supervisors are able to calculate the VaR and the ΔCoVaR from a wider and a more detailed definition of total assets containing items which do not appear on the balance sheet such as derivatives exposures. Since this thesis has tried to capture the year-to-year differences in this ∆CoVaRsystem|i, the unconditional and time consistent CoVaRsystem|i is calculated for each year in the sample period. For estimating the value at risk of the system conditional on the value at risk of institution i, a quantile regression model is adopted. The ‘standard’ regression
method Ordinary Least Squares regression views the average relation between the dependent (outcome) variable and one or more independent variables. This depends on the conditional mean function E(y|x). Since the CoVaR captures the risk of the financial system conditional on an institution being at risk (at its VaR level), the independent variable should be conditional on x being in a certain quantile instead of being at its mean which is the steady state. The quantile used where institution is at its VaR. The ΔCoVaR is estimated in three steps. First, the change in market-valued assets is calculated, then the VaR is estimated based on a historical simulation and last, the ΔCoVaR is calculated. This ΔCoVaR is at its 1% level which captures the 1% losses.
In this paper, X, and therefore the VaR, is based on the growth rate of market-valued total assets. To calculate the change in the market value of assets, the
following formula is used:
(1) 𝑋𝑡𝑖 =𝑀𝐸𝑡
𝑖∗𝐿𝐸𝑉
𝑡𝑖−𝑀𝐸𝑡−1𝑖 ∗𝐿𝐸𝑉𝑡−1𝑖
𝑀𝐸𝑡−1𝑖 ∗𝐿𝐸𝑉
𝑡−1𝑖
where 𝑋𝑡𝑖 is the change in the market value of assets, 𝑀𝐸
𝑡 𝑖 is the market value of equity and 𝐿𝐸𝑉𝑡𝑖 is the market-to-book ratio.
After this the change in the market value of assets of the system (𝑋𝑡𝑠𝑦𝑠𝑡𝑒𝑚 ) is calculated by using the sum of all market-valued assets of each bank.
With these two variables, for a particular quantile, the predicted value of 𝑋𝑡𝑠𝑦𝑠𝑡𝑒𝑚 conditional on institution i is estimated by,
(2) 𝑋̂𝑞𝑠𝑦𝑠𝑡𝑒𝑚,𝑖 = 𝛼̂𝑞𝑖 + 𝛽̂𝑞𝑖𝑋𝑖+ ɛ𝑞
For this estimation, quantile regression is used to estimate the change in the market value of assets conditional of Xi at its median (q = 0.50) and at the 1% quantile where q = 0.01. Thus two variables are estimated, which are 𝑋̂0.50𝑠𝑦𝑠𝑡𝑒𝑚,𝑖 and 𝑋̂0.01𝑠𝑦𝑠𝑡𝑒𝑚,𝑖.
With these two, the Value at Risk (VaR) for each institution is,
For each institution, for every year, a historical simulation is set up and only the data with a 𝑉𝑎𝑅0.50 𝑖 and 𝑉𝑎𝑅
0.01 𝑖
are kept in the sample. The VaR of the financial system conditional on i is simply,
(4) 𝑉𝑎𝑅𝑞𝑠𝑦𝑠𝑡𝑒𝑚,𝑖 = 𝑋̂𝑞𝑠𝑦𝑠𝑡𝑒𝑚,𝑖 The last step is to estimate the ΔCoVaR where
(5) 𝐶𝑜𝑉𝑎𝑅𝑞𝑠𝑦𝑠𝑡𝑒𝑚|𝑉𝑎𝑅q
𝑖
= 𝑉𝑎𝑅𝑞𝑠𝑦𝑠𝑡𝑒𝑚|𝑉𝑎𝑅𝑞 𝑖
And thus,
(6) ∆CoVaR qsystem | i = CoVaR q
system | VaRqi
− CoVaR 0.50system | VaR0.50i
With this, the ∆CoVaR 0.01system | i is calculated and defined as the systemic risk contribution variable. Since this thesis has used the change in ∆CoVaR qsystem | i, the dependent variable would be stated as ∆ΔCoVaR qsystem | i. For simplicity, the ΔCoVaR qsystem | i is now CoVaRq and ∆ΔCoVaR qsystem | i is now ΔCoVaRq.
3.1.2 Models
To test the first hypothesis, assuming that G-SIBs did not experience a reduced individual risk exposure more than non G-SIBs by raising capital, the following model is used,
(7) ∆𝑉𝑎𝑅𝑞𝑖𝑡= β0+ β1ΔCapital𝑖𝑡+ β2D1,𝑖𝑡+ β3ΔCapital𝑖𝑡* Dit+ [𝛽𝛽4
7] 𝐶𝑖𝑡+ εit The second hypothesis, assuming that G-SIBs did not experience a reduced individual risk exposure more than non G-SIBs by raising capital, is tested with the following model,
(1) ∆𝐶𝑜𝑉𝑎𝑅𝑞𝑖𝑡= β0+ β1ΔCapital𝑖𝑡+ β2GSIB𝑖𝑡+ β3ΔCapital𝑖𝑡* GSIBit+ [𝛽𝛽4
8] 𝐶𝑖𝑡+ εit
The models are tested with an OLS regression on the panel data. In the first model, the year to year change in the Value-at-Risk for each individual bank (∆VaRqit) is calculated in equation (3). This is the ∆1%-VaR (∆VaR0.01it), capturing the losses at a 1% level. Since this thesis has tried to capture the year-to-year differences in this (∆CoVaRqit), the unconditional and time consistent (∆CoVaRqit) is calculated for each year in the sample period. The year to year change in the systemic risk conditional to each individual bank (∆CoVaRqit) is calculated in equation 6. This dependent variable ∆CoVaRqit will also be tested on a 1% level. Since the dependent variables in both models are based on the change in market-valued assets, they are given as ratios.
ΔCapitalit, is the year to year difference in common equity relative to assets. The reason for this is that the additional loss absorbency surcharge is given as common equity (CET1) as a percentage of risk weighted assets. This variable could be improved by applying the risk weights of the assets, but this information is not publicly available. Therefore, the ΔCapital is set equal to Δ Common equityAssets . If the regression estimator of this variable is positive, then capital has a positive effect on the dependent variable risk, which means that this risk is lowered. The expectation is that an increase in equity leads to a reduction in the systemic risk contribution. The dummy variable D1,it is a dummy variable, where this variable is 1 if a bank is a G-SIB and a 0 if the bank is not. This dummy variable is added to capture the effect whether G-SIB’s have a different effect on the dependent variable. Bongini et al. (2015) have used this dummy variable as well to assess the market reaction the publication of the list of G-SIB’s. If the regression estimator is significantly different this means that the systemic risk contribution changes more (or less) than non-G-SIB banks. According to the purpose of the additional loss absorbency requirement and tighter supervision, it is expected that systemic risk contribution positively changed their systemic risk exposure.
The interaction variable ΔCapitalit*D1,it is added to the model to test whether G-SIBs have experienced a different change in systemic risk exposure due to an increase in
equity. The expectation is that G-SIBs in general have reduced their systemic risk exposure more than other banks due to the additional loss absorbency requirement. With the regression estimator the hypotheses are either rejected or not rejected. As with the G-SIB dummy a positive change is expected.
An additional regression is done to test whether G-SIBs in different buckets have different effects on the individual risk for the first hypothesis and for systemic risk contribution in the second hypothesis. There are two additional variables. The first variable is a set of dummy variables [𝛾𝛾1
5] [ 𝐷2,𝑖𝑡
𝐷6,𝑖𝑡], where each dummy variable contains one bucket, including bucket 0 containing the non-G-SIBs. Bucket 5 is left out of the regression, because this bucket is empty. To prevent multicollinearity, the dummy containing bucket 1 is also left out of the regressions. With these dummies, the effect of SIBs in each can be estimated separately. The expectation is that G-SIBs in higher buckets have changed their risk exposure more than non G-G-SIBs. There is also an interaction term added to capture the effect of the additional loss absorbency surcharge per bucket which is ΔCapital * [𝛾𝛾1
5] [ 𝐷2,𝑖𝑡
𝐷6,𝑖𝑡]. The same buckets, 1 and 5, are left out because of multicollinearity and no banks in the last buckets. This interaction term is added to test whether G-SIBs in different buckets have experienced different change in the systemic risk contribution. Because the higher the bucket allocation, the more additional capital is required. The expectation is that the higher the bucket, the more these G-SIBs have reduced their systemic risk contribution.
Cit contains four control variables in equation (7) and five control variables in equation (8). The first control variable is the logarithm of Sizeit, which is consistent
with Adrian & Brunnermeier (2011), Lopez-Espinosa (2012) and Castro & Ferrari (2014). Wholesale funding is the second control variable, since this variable captures the interconnectedness and liquidity risk of a bank and it is a key determinant to systemic risk according to Lopez-Espinosa et al. (2012). Maturity mismatchit captures the mismatch on a bank’s balance sheet where the short term liabilities is greater than the short-term assets. Maturity mismatch is defined as 𝑆ℎ𝑜𝑟𝑡 𝑡𝑒𝑟𝑚 𝑑𝑒𝑏𝑡−𝑐𝑎𝑠ℎ𝑡𝑜𝑡𝑎𝑙 𝑙𝑖𝑎𝑏𝑖𝑙𝑖𝑡𝑖𝑒𝑠 .The fourth control variable is Market-to-bookit ratio which is a proxy for the growth rate of a bank and possibly mispricing (Lopez-Espinosa et al., 2012). In equation (8), the Cit contains an additional control variable which is Value at Risk (VaRit). The reason for this is that the ∆CoVaR is defined by using the VaR. They are only loosely linked in
the cross section but they have a strong relationship over time (Adrian &
Brunnermeier, 2011). To prevent serial correlation in the error term, the errors are clustered per bank. To check the robustness of the results, the dependent variable ∆1%-VaR is replaced by ∆5%-VaR and the dependent variable ∆1%-CoVaR is replaced by ∆5%-CoVaR. The 5% dependent variables contain extreme losses, but not as rare as the 1% dependent variables.
3.2. Data and descriptive statistics
This paper assesses the effect of the increasing capital reduces the systemic risk contribution of banks, where, due to the additional loss absorbency requirement, it is expected that G-SIBs have reduced their exposure more than other SIFIs. The Basel Committee on Banking Supervision has applied their methodology of identifying G-SIBs on a sample of 73 banks which are systemically important. However, they did not enclose the banks that they have used in their sample. Therefore, this study uses the sample of banks as Bongini et al. (2015). Their paper constructs the sample based on three conditions:
1. the headquarters of these banks are in the following countries or regions: North- South- and Central America; Western Europe; Scandinavia and Far East- and Central Asia.
2. Total assets of these banks exceeded $200 million 3. These banks are listed during the entire testing period.
This has led to a panel of 70 banks, including all the G-SIBs on the lists published and potential SIFIs. The list of G-SIBs is provided in Table 1 in Appendix A and contains the bucket allocation for each bank ever identified as a G-SIB. Table 3 represents the distribution of these banks by their classification as G-SIB or SIFI.
Table 3 Sample distribution Group 2011 2012 2013 2014 2015 G-SIBs 29 28 28 30 30 Peer group 41 42 42 40 40 All SIFIs 70 70 70 70 70
This table represents the distribution of banks by classification
In this paper’s sample the time period starts in 2011Q1 and ends in 2015Q4. This period is used because the Basel III accords have been introduced in 2011.
Weekly data on stock prices is retrieved from Datastream. The weekly data is needed to give a solid estimate of the CoVaR. Balance sheet data is retrieved from
Datastream as well, but this is on a quarterly frequency. Missing data of balance sheet information is found on Bankscope and in the actual quarterly reports on the banks’ websites. To generate weekly data of balance sheet items, the quarterly data is linear interpolated. The interpolated values are computed only if less than two consecutive quarters are missing. This results in 311 weeks of data per bank. The weekly data is used to estimate an unconditional yearly CoVaR. These VaR and CoVaR values are not winsorized since this paper estimates the 1% and 5% quantiles, which contain rare losses. When data of these are winsorized, it would state that the losses in the far left tail are assumed as outliers, which in this case cannot be reduced to a more normal level since the distribution is not expected to be normal. Because this paper tests the effect of a change in equity on the systemic contribution from year-to-year, hereafter, data in the sample has a yearly frequency.
Figure 1, in Appendix B, shows that the 1%-CoVaR and the 1%-VaR as average of all banks from the sample. On the vertical axis, the values of both
measures of risk are negative. The 1%- CoVaR and the 1%-VaR tend to move in the same direction over time, and since 2011 both measures are increased. This means that both the systemic risk contribution and the individual risk exposure are reduced, since 2011. This relationship is also found by Adrian & Brunnermeier (2011). Figure 2 provides two separate evolutions of the systemic risk contribution of G-SIBs and non G-SIBs. One can see that the G-SIBs have a lower 1%-CoVaR associated with a higher systemic risk contribution than the non G-SIBs have. Since 2011, both G-SIBs and non G-SIBs have reduced their systemic risk contribution and therefore the financial system has become more resilient.
Adrian and Brunnermeier (2011) have also found that the relationship between the 1%-CoVaR and the 1%-VaR is not strong in the cross-section. Figure 2 confirms that in this sample the cross-sectional relationship between the CoVaR and the 1%-VaR is only loosely linked. Therefore, the individual risk exposure of a particular bank is not equal to the same banks systemic risk contribution.
Figure 4 shows the evolution of the capital ratio defined as Common equityAssets . On average the capital ratios of G-SIBs are lower than the capital ratios of other SIFIs in this sample. This means that G-SIBs have a lower absorbency capacity in case of
losses. This also means that risk-shifting is more likely to occur because of the limited liability (Gavin and Hausmann, 1996). This validates the reason of the Basel
Committee to charge the G-SIBs with an additional capital requirement such as the loss absorbency surcharge assessed in this paper. This figure also shows that the capital ratio, consisting of common equity relative to total assets, has increased since 2011.
Table 4 shows the means of the bank characteristics used as control variables in the models. The first column mean is the average of all banks in the sample. The second and column mean represent the G-SIBs and the peer group with SIFIs respectively. The penultimate column displays the difference between the G-SIB means and the peer group means. A T-test is conducted to assess whether the bank characteristics of G-SIBs differ significantly from the bank characteristics of the peer group. The individual risk measures 1%-VaR and 5%-VaR do not differ significantly from being a G-SIB or not. Concluding from this, on average, both G-SIBs and SIFI experience the same individual risk. For all other bank characteristics, G-SIBs have significantly larger values than the peer group.
Table 4
Bank characteristics
Variable All SIFIs G-SIBs Peer group G-SIB vs Peer
Mean Mean Mean Difference T-test
1%-VaR -0.125 -0.113 -0.120 0.008 0.474
5%-VaR -0.062 -0.063 -0.062 -0.001 -0.242
Wholesale Funding 0.158 0.235 0.117 0.118 2.911***
Size 20.348 20.954 20.019 0.936 13.291***
Market to book ratio 7.478 12.334 4.843 7.491 2.259** Maturity mismatch 0.148 0.308 0.062 0.246 4.476***
Note: Total number of observations is 418 where 147 observations classified as G-SIB and 271 are classified observations as peer group. *, ** and *** are the significance at the levels 10%, 5% and 1% respectively
4. Empirical results
In this section, the relation between both individual and systemic risk contribution and change in capital is assessed with several regressions. First, the regression estimates on relation between the individual risk exposure, VaR, and change in capital are discussed. Secondly, the regression estimates of the relation between the systemic risk
contribution, CoVaR, and the change in capital are discussed. Afterwards, the results between the individual and systemic risk exposure are compared. Finally, the
robustness of the results are checked.
4.1.1. Results on assessing the VaR
The main regression results based on equation (7) are stated in Appendix C, Table 5 in columns 1 to 3. Column 1 provides the OLS regression estimates of equation (7) without any control variables. In column 2 the control variables are added and column 3 provides the additional regression including the separate buckets the G-SIBs are allocated in to test whether different buckets have different effects on the individual risk exposure.
The results show that positive changes in common equity relative to assets has a significantly positive effect on the individual risk exposure, 1%-VaR, across all banks since 2011. Adding control variables in column 2 and buckets in column 3 does neither change the impact, nor the significance of the change in estimator of the differenced capital. For all banks hold, that increasing equity reduced the individual risk exposure in an isolated situation since 2011. The results also show that G-SIBs, in general, have positively changed their individual risk exposure more than the peer group, which consists of banks which are not identified as G-SIBs. However, the significance in the second column is at a 5% level whereas the significance in the first column is at a 1% level. In column 3, the level of significance is at a 10% level. It can be concluded that the G-SIBs have reduced their individual risk exposure more than other banks by raising capital, which is in line with the intentions of the Basel Committee on Banking supervision, to reduce the risk exposures of these TBTF and TSTF banks (BCBS, 2013). However, it should be kept in mind that the VaR
measures the individual risk exposure in an isolated situation. It does not measure the effect on the financial system. With interaction variable consisting of the G-SIB dummy and the differenced capital, the first hypothesis is tested. The regression estimator of this interaction term in columns 1 and 2 does not have any significant value. This is consistent with the hypothesis stating that G-SIBs have not experienced a more reduced individual risk exposure by having to increase their capital ratio more than non-GSIB banks since 2011. Thus, G-SIBs have experienced a reduced
individual risk exposure by increasing capital, but not more than other banks. Column 3 provides the additional regression test whether G-SIBs in different
buckets have different effects on the dependent variable. Comparing the coefficients on the bucket dummies, none of these significantly differ from the coefficient of the G-SIBs in the first bucket. The effect of a GSIB being in bucket one is captured in the coefficient on the GSIB-dummy as all GSIBs are at least placed in the first bucket. It can be concluded, that G-SIBs, in general, have significantly changed their individual risk exposure more than the peer group banks, but there are no reciprocal differences among the buckets. For the coefficient of the G-SIBs in the second bucket holds that, comparing to the coefficient of the G-SIBs in the first bucket, an increase their equity has a negative effect on the changes in individual risk exposure. The same holds for the coefficient of the G-SIBs in the fourth bucket, but the coefficient of the G-SIBs in bucket 3 is insignificant. Therefore, the null hypothesis, stating that G-SIBs have not experienced a reduced individual risk exposure due to capital increase since 2011, is rejected for G-SIBs in buckets 2 and 4. However, one must be careful to interpret these results, since the results for the G-SIBs in the third bucket are insignificant, while this bucket is between the second and fourth bucket. There is neither an explanation from the existing literature, nor intuitively. The regression estimates are not in the right direction as intended by the additional loss absorbency requirement. Equity increase causes an increase in the individual risk exposure. A reason for this could be that capital undermines the business model of a bank and reduces its profits (Blüm, 1999).
4.1.2. Results on assessing the CoVaR
The main regression results based on equation (8) are provided in Table 5 in columns 4 to 6. Column 4 provides the OLS regression estimates of equation (8) without any control variables. In column 5 the control variables are added and column 6 provides the additional regression on the separate buckets the G-SIBs are placed in to assess whether different buckets have different effects on the systemic risk contribution. The results in column 4 show that, on average, raising the capital, as common equity relative to risk weighted assets, does not significantly change the systemic risk contribution of banks in this sample. This means that capital adequacy requirements for all banks do not act as an instrument to enhance the financial system. Adding the variables for control in column 5 and using buckets in column 6 does not change this insignificance. G-SIBs in all three columns have a negative effect on changing their systemic risk exposure, with a significance level of 1%. From this, it can be
concluded that SIB have changed their systemic risk contribution less than non G-SIBs. The coefficient of the interaction variable, which captures the effect of the change in equity for G-SIBs compared to non-G-SIBs, is insignificant in the three regression results. This is consistent with the hypothesis assuming that G-SIBs did not experience a reduced systemic risk exposure by having to increase their capital since 2011. However, this does also mean that the additional loss absorbency requirement, which only applies to G-SIBs, does not lower the systemic risk contribution of banks. This is in contrast with the intentions of the additional loss absorbency requirement to address the negative externalities and enhancing of the financial system (BCBS, 2011b).
Additionally, a separate regression is done to test the possible different effects of G-SIBs in different buckets as well. The result of this regression is provided in column 6 of Table 5. The coefficients of the G-SIBs in buckets 2 to 4 have a positive effect on the change in the systemic risk exposure compared to the bucket 1
coefficient. Since these banks are captured in the G-SIB dummy as well, e.g. G-SIB dummy equals the value 1 for all G-SIBs, for G-SIBs in buckets 1 and 3, it results that these G-SIBs have changed their systemic risk exposure less than non G-SIBs, while the coefficient of the G-SIBs in buckets 2 and 4 have a significantly positive effect on the change in systemic risk exposure compared to non G-SIBs. One must be careful to interpret this, since these results are not backed by the existing literature. The only significant interaction term is the one between bucket 4 and the differenced capital, but this regression estimate is significantly negative. It can be concluded that, on average, G-SIBs in the fourth bucket have positively changed their systemic risk exposure compared to other banks, but raising capital has a negative effect on this change. The added control variable 1%-VaR has as significant positive effect on the change systemic risk contribution. But since the 1%-VaR only contains negative values, the individual risk exposure has a negative effect on the systemic risk
contribution. This is in line with the expectation, since on average, banks with higher individual risks also have higher systemic risk contribution.
4.1.3. CoVaR versus VaR
Whereas the 1%-VaR measures the individual risk exposure of a bank in an isolated situation, the 1%-CoVaR measures the risk of the financial system conditional on each bank under distress and therefore the systemic risk contribution. Reducing the
individual risk exposure means that the chance of default or impairment is lowered. Reducing the systemic risk contribution means that the negative externalities G-SIBs pose to the financial system are reduced and therefore the financial system becomes safer.
Resulting from the analysis done by this paper, for the individual risk exposure holds that capital increase does reduce the individual risk exposure for all banks. G-SIBs have reduced their individual risk exposure more than non G-SIBs, but not by increasing capital. For the systemic risk contribution holds that capital increase does not affect the change in systemic risk contribution. G-SIBs, in general, even have changed their systemic risk exposure less than non G-SIBs. Raising capital and
therefore capital adequacy requirements do reduce the individual risk exposure in an isolated situation but do not enhance the financial system.
4.2. Robustness
To check for robustness in equation (7), the dependent variable ∆1%-VaR is replaced by the ∆5%-VaR. The results are expected to have the same sign but smaller in de amplitude, because the 5%-VaR in itself has higher values since it captures the losses at a 5% level whereas the 1%-VaR contains more rare and thus larger losses. The regression results are provided in Appendix D, Table 6 in columns 1 to 3. Results on the change in capital in columns 1 to 3 are not very different from the results provided in Table 5. However the regression estimators are of a lower amplitude when
regressing on the ∆5%-VaR, as expected. A different outcome does exists in the coefficient of the dummy variable for G-SIBs. Where the regression estimators of the 1%-VaR of the G-SIB coefficient are significantly positive, these are insignificant with the 5%-VaR.
The coefficient of the interaction variable between differenced capital and G-SIB dummy has a positively significant effect on the change in the 1%-VaR. These results were not found using the 1%-VaR. From this, it can be concluded that the null
hypothesis can be rejected when using the 5%-VaR but not for the 1%-VaR. In column 3 of Table 6, the same results hold for the interaction terms between the differenced capital and the buckets, compared to the regression results in column 3 Table 5. These also result in a negative effect of the coefficient of the G-SIBs allocated in buckets 2 and 4. As expected, the estimates are lower than the estimates in Table 5.
To check for robustness in equation (8), the dependent variable ∆1%-CoVaR is replaced by the ∆5%-CoVaR. The results are expected to have the same sign but smaller in de amplitude, for the same reason as with the robustness checks on the ∆5%-VaR. The regression results are provided in the columns 4 to 6 of Table 6. The results are not much different from the results in Table 5. The differenced capital has no significant effect on the changed systemic risk contribution, which is the same result as in Table 5. G-SIBs have significantly changed their systemic risk
contribution more negatively than non G-SIBs banks. Although these results are of a lower amplitude than in the regressions on the ∆1%-CoVaR, these are still consistent with the regression results in Table 5. The same results hold for the coefficients of the G-SIBs in the buckets 2 and 4 and for the interaction term consisting of the change in capital and the G-SIBs in bucket 4. Concluding from this, only G-SIBs allocated in the fourth bucket have a lower change in their systemic risk exposure because of increasing capital.
5. Conclusion and discussion
In this paper, it is investigated whether G-SIBs have reduced their systemic risk exposure more than other banks due to the influence of the additional loss absorbency requirement since 2011. Using a set of the 70 largest banks, over a time span of 2011 to 2015, for each bank the conditional value at risk (CoVaR) is estimated with the quantile regression methodology. Figure 2 shows that since 2011, the systemic risk has reduced over time for both G-SIBs as for non G-SIBs. Therefore, the financial system has become safer over time. That the systemic risk contribution of G-SIBs is higher than for non G-SIBs will always be the case, since that is the reason why they have the TBTF and the TSTF status. The OLS regression methodology is used to assess whether the additional loss absorbency requirement has caused this reduction. The results reveal that increasing capital does not reduce the systemic risk
contribution for G-SIBs or other banks.
Before the relation between systemic risk exposure and capital changes is tested, the relation between capital changes the individual risk exposure, captured by the 1%-VaR, is assessed. Resulting from this, is that G-SIBs have reduced their individual risk exposure more than non G-SIBs, but not by increasing capital. In fact, increase in capital by G-SIBs in the buckets 2 and 4 have negative effects on the
change in the individual risk exposure. When checking for robustness with the 5%-VaR, G-SIBs have positively changed their individual risk exposure by raising capital. Therefore, the results of the regression estimates on the individual risk exposure are inconsistent. Raising capital, in general, positively affects the change in de individual risk exposure and therefore reduces the chance of failure or impairment of a bank in an isolated situation.
To answer the research question, the relation between the systemic risk
exposure and the capital increase of G-SIBs is assessed. The results from this analysis show that G-SIBs did not change their systemic risk exposure significantly by raising capital. G-SIBs, in general, have changed their systemic risk exposure less than non G-SIBs. G-SIBs allocated in the buckets 2 and 4 have changed their systemic risk exposure positively compared to G-SIBs in the first bucket. This is also the case in the robustness checks. Interpreting the latter result is more difficult, since this effect does not hold for G-SIBs in bucket 3. A possible reason for this is that G-SIBs already have met the capital adequacy requirements and banks raising capital above the capital adequacy requirements, are undermining their business model and possibly goes hand in hand with taking more risks to receive the same benefits (Blüm, 1999). One could also argue that the level of systemic risk contribution cannot be changed much by increasing capital, since this also depends on the banks’ characteristics which are used in the indicator based approach to identify these G-SIBs. However, this does not support an additional capital adequacy requirement specially for G-SIBs. This paper suggests to regulators to focus more on addressing the factors which distinguishes G-SIBs from non G-SIBs and to mitigate the risks and other negative externalities these G-SIBs pose to the financial system.
A limitation of this research is that the additional loss absorbency requirement is not fully implemented yet. After full implementation, one should test this
requirement on systemic risk again, to assess whether this requirement has served its aim. Another limitation is that in this paper, systemic risk is only measured by the ∆CoVaR, whereas there are other measures of systemic risk which could assess the relation between raising capital and systemic risk. Therefore, future research should use and compare different systemic risk measures to provide a conclusive result on the effect of the additional loss absorbency requirement.
References
Adrian, T. & Brunnermeier, M. (2011) CoVaR, Federal Reserve Bank of New York
Staff Reports (348), 1-45
Basel Committee on Banking Supervision (2001) Overview of the new Basel capital accord. Bank for International Settlements
Basel Committee on Banking Supervision (2011a) Basel III: A global regulatory framework for more resilient banks and banking systems. Bank for International
Settlements
Basel Committee on Banking Supervision (2011b) Global systemically important banks: updated assessment methodology and the higher loss absorbency requirement.
Bank for International Settlements
Basel Committee on Banking Supervision (2013) Basel III: The liquidity coverage ratio and liquidity risk monitoring tools. Bank for International Settlements
Basel Committee on Banking Supervision (2014) Basel III: The net stable funding ratio. Bank for International Settlements
Bongini, P., Nieri, L. & Pelagatti, M. (2015) The importance of being systemically important financial institutions. Journal of Banking & Finance 50, 562-574
Blüm J. (1999) Do capital adequacy requirements reduce risks in banking, Journal of
Banking and Finance 23, 755-771
Castro, C. & Ferrari, S. (2014) Measuring and testing for the systemically important financial institutions. Journal of Empirical Finance 25, 1-14
Financial Stability Board (2011) Key Attributes of Effective Resolution Regimes for Financial Institutions. Retrieved from:
http://www.financialstabilityboard.org/2014/10/r_141015/
Gavin, M. & Hausmann, R. (1998) The roots of banking crises: The macroeconomic context. Retrieved from:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1815948
Girardi, G. & Ergün, A. T. (2013) Systemic risk measurement: Multivariate GARCH estimation of CoVaR. Journal of Banking & Finance 37, 3169-3180
International Monetary Fund, Bank for International Settlements & Financial Stability Board (2009) Guidance to Assess the Systemic Importance of Financial Institutions, Markets and Instruments: Initial Considerations. Report to G20 Finance Ministers
and Governors
Linsmeier, T.J. & Pearson, N.D. (2000) Value at Risk. Financial Analysts Journal 56(2), 47-67
López-Espinosa, G., Moreno, A., Rubia, A. & Valderrama, L., (2012) Short-term wholsale funding and systemic risk: A global CoVaR approach, Journal of Banking &
Finance 36, 3150-3162
Nier, E., Yang, J., Yorulmazer, T. & Alentorn, A. (2007) Network models and financial stability. Journal of Economic Dynamics & Control 31. 2033-2060
Stern, G.H., Feldman, R.J.(2004) Too Big to Fail: The Hazards of Bank Bailouts. Washington D.C. The Brookings Institution.
Appendices
Appendix A: List of G-SIBs
Bank Bucket
2011 2012 2013 2014 2015
Agricultural Bank of China no 0 0 1 1
Bank of America yes 2 2 2 2
Bank of China yes 1 1 1 1
Bank of New York Mellon yes 2 1 1 1
Banque Populaire CdE yes 1 1 1 1
Barclays yes 3 3 3 3 BBVA no 1 1 1 0 BNP Paribas yes 3 3 3 3 China Constructionbank no 0 0 0 1 Citigroup yes 4 3 3 3 Commerzbank yes 0 0 0 0
Credit Suisse yes 2 2 2 2
Deutsche Bank yes 4 3 3 3
Dexia yes 0 0 0 0
Goldman Sachs yes 2 2 2 2
Group Crédit Agricole yes 1 2 1 1
HSBC yes 4 4 4 4
Industrial and Commercial Bank of China no 0 0 1 1
ING Bank yes 1 1 1 1
JP Morgan Chase yes 4 4 4 4
Lloyds Banking Group yes 0 0 0 0
Mitsubishi UFJ FG yes 2 2 2 2
Mizuho FG yes 1 1 1 1
Morgan Stanley yes 2 2 2 2
Nordea yes 1 1 1 1
Royal Bank of Scotland yes 2 2 2 1
Santander yes 1 1 1 1
Société Générale yes 1 1 1 1
Standard Chartered no 1 1 1 1
State Street yes 1 1 1 1
Sumitomo Mitsui FG yes 1 1 1 1
UBS yes 2 2 1 1
Unicredit Group yes 1 1 1 1
Wells Fargo yes 1 1 1 1
Note: This list contains all banks which ever have classified as global systemically important bank. In 2011 the list did not contain the buckets as in the years after. If a bucket states zero, than this bank is not
Figure 2: This figure shows the cross-sectional relation between the 1%-CoVaR and the 1%-VaR of all observations
Figure 4: This figure shows the evolution of the average capital ratio per year where, the black line is the average capital ratio of all banks in the sample, the red line shows average the capital ratio of banks classified as G-SIBs and the green line shows the capital ratio of banks identified as SIFI but not as GSIB.
Figure 3: This figure shows the relation between the 1%-CoVaR of G-SIBs and the 1%-CoVaR of the peer group over time
Appendix B: Figures
Figure 1: This figure shows the 1%-CoVaR (black line) and the 1%-VaR (red line) for the sample. Both risk measures are in percent returns to market valued assets
