Amsterdam Business School
University of Amsterdam
MSc Business Economics, Finance
Master Thesis
Does the Liquidity Coverage Ratio, implemented by
Basel III, influence U.S. Banks’ CDS Spreads?
Cédric Kishan Haselier
July 2016
Statement of Originality
This document is written by Cédric Kishan Haselier 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 text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
The Basel Committee on Banking Supervision announced a new regulatory framework named Basel III. Part of the new regulation is a minimum liquidity requirement known as the Liquidity Coverage Ratio (LCR). The LCR is subject to a stepwise implementation, requiring a minimum ratio of 60 percent in 2015, which increases in yearly 10 percent intervals until it is fully enforced, corresponding to an LCR of 100 percent by 2019. This thesis investigates the effect of the LCR announcement and first two implementation periods on U.S. bank CDS spreads. Spreads declined on the regulation’s announcement with no significant evidence for changes during the actual implementation phases. Banks with lower liquidity levels prior to the regulation’s announcement experience the biggest changes in spreads.
Table of Contents
1. Introduction ... 1
2. Literature Review ... 5
2.1 Basel Committee on Banking Supervision ... 6
2.2 Debate on bank regulation ... 7
2.3 Determinants of Credit Default Swap Spreads ... 12
2.4 Effect of Basel III on bank liquidity and asset prices ... 15
2.5 Relation to existing literature and hypotheses ... 17
3. Liquidity Coverage Ratio ... 19
4. Methodology ... 21
4.1 Control Variables and bank fixed effects ... 22
4.2 Separating treatment and control group ... 25
4.3 Difference-in-Difference Regression Model ... 26
5. Data and Descriptive Statistics ... 29
5.1 Summary Statistics ... 31
5.2 Correlation Between Variables ... 33
5.3 OLS Regression ... 34
6. Results ... 36
7. Robustness Checks ... 42
7.1 Macroeconomic events ... 42
7.2 Omitting variables ... 44
7.3 Interaction Term with Bank Size ... 46
8. Conclusion ... 49
References ... 53
1. Introduction
One of the many reasons that led to the financial crisis of 2007 – 2009 was the lack of financial regulation. The destruction of value and disruptions caused by deficiencies overseeing effectiveness called for a regulatory framework, such that banks and other financial institutions become more resilient in times of financial and economic distress. Das et al. (2007) speak of the clustering of default correlations among commercial banks, which is also referred to as credit contagion. The G20 and the Basel Committee on Banking Supervision therefore introduced the Third Basel Accord in December 2010. Basel III is a banking regulatory framework introduced in order to strengthen regulation, improve supervision of global bank practices and enhance the financial stability in the banking sector in general. As banks have become more interdependent and connected over the years, their resilience to economic or financial shocks have become of upmost importance. In fact, banks have become so large that a fall in liquidity could jeopardize entire financial systems and economies around the world. Subsequently, it endorses the reduction in moral hazard of these Systemically Important Financial Institutions, known as SIFIs. The Financial Stability Board (2010) defines SIFIs as financial institutions that could cause significant disruptions to the financial system and global economy as a whole due to their size, complexity and interconnectedness. These SIFIs should be regulated in order to mitigate or ideally even prevent further market wide adverse effects when one significantly important institution suffers from financial distress.
This thesis focuses on the liquidity aspect enforced by Basel III. The third Basel Accord introduces a minimum percentage that requires banks to hold a sufficient amount of highly liquid assets to survive a 30-day stress scenario (Bech and Keister, 2012). This minimum requirement is known as the Liquidity Coverage Ratio (LCR). One of the functions of the LCR, which has been firstly implemented in January 2015, is to assure that banks have sufficient
liquid assets at all times to meet short-term obligations and ride out short-term liquidity disruptions. Subsequently, the rationale behind the passage of the LCR is to ensure that banks, even in times of severe stress, can provide sufficient liquidity. To provide liquidity in the first place, banks need to be solvent and functioning, thus they can evidently not be bankrupt. The declaration of insolvency or bankruptcy of banks can be ascribed to several reasons, one of them the inability to meet their debt obligations. When the borrower is unable to repay the credit, loan, mortgage et cetera, the borrower is said to default.
Banks and investors can ensure themselves and the credit they lend to other banks through a credit derivative that serves as an insurance policy in case the reference entity, i.e. the borrower of the loan, defaults. This financial derivative is known as a Credit Default Swap (CDS), whereas the spread is the percentage that the CDS buyer pays to the CDS seller. The simplest form of a CDS would be a scenario where the protection buying (seeking) bank gives out a credit to a reference entity, like a company or another bank. In order to insure itself against the loss that the bank would incur in case the reference entity defaults on the loan, the protection seeking bank can engage in a contract with another institution to purchase a CDS from that institution, which is the protection seller or the so called protection provider. The protection provider would be obligated to pay the outstanding amount of the loan to the protection seeker in case the reference entity defaults on the loan. Therefore, the cost of purchasing the insurance policy is the so called CDS spread1, a fee measured in basis points that the protection buyer pays to the protection provider under negotiated terms. The buyer of the CDS protection does not need to own the underlying bond or credit. The transaction can be made as a speculative trade as it’s based on the expectation of the deterioration of credit quality (Schwager, 2012).
1 Throughout this thesis, spreads will refer to the spread paid on credit default swaps. CDS spreads and spreads will therefore be used interchangeably.
The relationship between the LCR and CDS spreads written on banks’ debt may not be evident at first. The LCR is a framework that ensures that banks are liquid enough to operate for 30 days under financially stressful scenarios, thus not be insolvent or bankrupt. Spreads paid for Credit Default Swaps can be seen as indications of how the market values the bank’s credit risk, being unable to repay its debt obligations and ultimately being insolvent. Consequently, this thesis investigates the relationship between a liquidity framework2 and a financial credit derivative3. The corresponding research question in this thesis is: “what is the
effect of the LCR announcement and implementation on U.S. banks’ CDS spreads?”.
As this specific relation has not been investigated by any prior research, this thesis should yield results that are of interest for academics, investors and regulators. The LCR is due to a stepwise implementation, meaning that an LCR of 60% was required by 2015, 70% by 2016, 80% by 2017, 90% by 208 and must be fully implemented, meaning an LCR of 100%, by January 2019. This thesis is based on historical events, as it covers the announcement of Basel III, which was on December 16th 2010 and the first two transition periods in 2015 and 2016. This makes my research very recent and could therefore lead to potential hypotheses for the later years of the LCR implementation process.
Figure 1. plots the CDS spreads for each bank from June 30th 2010 until March 30th 2016.
2 The liquidity framework mentioned here refers to the Liquidity Coverage ratio. 3 The credit derivative refers to credit default swaps.
Spreads between banks strongly differed depending on the time period taken into account. As an example, Morgan Stanley’s spreads were more volatile than i.e. Wells Fargo or PNC Financial spreads. The first red vertical line from left to right corresponds to the LCR announcement date on December 16th 2010. Almost all spreads declined after the announcement and yet changes do not seem to be substantial, considering the surges around seven months later. Furthermore, shortly after the 2015 and 2016 phase-in periods, which correspond to the second and third red vertical line, respectively, changes in CDS prices seem to be even smaller but slightly positive for both years. Figure 1. therefore enables to make the following predictions: Slightly positive changes in CDS spreads are predicted during the announcement period and even smaller, yet negative changes are expected for both transition phases4.
4 Positive CDS spread changes correspond to declining spreads and negative changes to increasing spreads.
0 100 200 300 400 500 600 700 6/30/10 12/31/10 6/30/11 12/31/11 6/30/12 12/31/12 6/30/13 12/31/13 6/30/14 12/31/14 6/30/15 12/31/15 CDS S pr ea d (b ps ) Date
Figure 1. U.S. Bank CDS spreads
American Express Bank of America Bank of New York Mellon Corp.
Capital One Financial Citigroup HSBC
Goldman Sachs JP Morgan Chase Morgan Stanley
The rest of this thesis is structured as follows. Section 2 relates this thesis to existing literature and results from other studies. This section gives an insight in the ongoing debate on the effectiveness on bank regulation, how CDS prices and spreads are determined, the effects of Basel III on asset prices and consequently states the hypotheses to be tested in this research. Section 3. describes the LCR and its components, while section 4. introduces the model and describes the variables. Descriptive and summary statistics and the OLS regression output are presented and discussed in section 5. This will be followed by section 6, which presents the main results and interprets the output by giving economic meaning. To test the viability of the model, robustness checks are conducted in section 7. Lastly, section 8. recapitulates this thesis with a conclusion.
2. Literature Review
Ever since the existence of banks, regulatory frameworks and legislations were enforced in order to assure the safety of the financial system and economies as a whole. Nevertheless, the effectiveness, adequacy and type of bank regulation is still subject to ongoing debates. Even the very first banking regulation, the National Banking Act of 1864 substantially improved the safety in the banking sector and yet resulted in the closure of more than 250 state banks (Jaremski, 2013). Over the years, regulatory frameworks have shifted towards more market transparency and restricting speculative activities. Examples are the 2002 Sarbanes Oxley Act that requires consistent disclosure of financial information and the Volcker Rule that forbids banks to engage in certain proprietary trading activities in securities, derivatives et cetera. Although bank runs and other liquidity threatening scenarios are everything but new, just with the Basel III accord and the LCR specifically, bank liquidity is addressed for the first time in bank regulation history. Subsequently, the issue of encountering liquidity issues is as old as
banks themselves and yet regulating their liquidity positions is very recent. As a result, existing literature regarding the effectiveness of liquidity is very limited. Especially prior literature regarding Basel III’s LCR is thus far almost non-existent, therefore this thesis is touching upon a new and very recent new field. Nevertheless, literature regarding capital and bank regulation as a whole is very widespread as its effectiveness and adequacy is extensively debated not only among academic but also regulators and many others.
2.1 Basel Committee on Banking Supervision
The Basel Committee on Banking Supervision (BCBS) is an independent committee of banking authorities that aims at improving the quality of worldwide banking supervision. It grew out of the globalization of financial markets and financial intermediation and was established by the G10 countries in 1974. The state of financial markets around the 1970s is described as a time that markets became more global, whereas regulation and monetary control was maintained at the national level, emphasizing the need for communication between national authorities (Goodhart, 2011). The BCBS’ governing body consists of central bankers and banking sector supervisors of the member countries. Its membership expanded to 27 member countries in total, which meet around four times a year. It sets standards for prudential bank regulation and is a provider of a forum for the bank supervisory cooperation.
Its first major regulatory framework was the Basel I Accord, implemented in 1988. Basel I enforced the “tier1” and “tier2” minimum capital requirement for banks due to concerns about the stability of the global financial system as bank capital levels were low (Tarullo, 2008). However, the emergence of securitization of mortgages and advances in risk management techniques called for a revision of the first framework, resulting in Basel II in 2004. Although
Basel II separated bank regulation into three pillars5, liquidity regulation was still neglected. Hence, few years after, the 2008 subprime mortgage crisis proved Basel II to be ineffective and led to a recession. Even the well positioned banks with adequate capital levels were heavily struck by the crisis. The reason was not insufficient capital but a lack of ready cash or liquidity (Levinson, 2010), setting an incentive for the BCBS to once again revise its prior frameworks. The largest change was the new emphasis on liquidity standards through the announcement and implementation plan of the LCR. Basel III, which was announced late December 2010, requires higher capital standards, more risk-weighted assets and additional capital buffers. Finally, just with the implementation of Basel III that was firstly implemented in 2015, bank liquidity has firstly been addressed through a regulatory framework. Some academics like Levinson (2010) argue that Basel I and II even destabilized the financial system as banks were incentivized to securitize instead of backing up these liabilities with sufficient capital. It is therefore evident that effective regulation is highly complicated and is therefore subject due to an ongoing debate.
2.2 Debate on bank regulation
This section summarizes the main arguments represented in academic literature. According to Peltzman (1970), bank capital has two main functions: Firstly, it serves as an input that enables the bank to provide services like liquidity services, accounting and information services et cetera. Secondly, it provides insurance for depositors that protects them against the risk of assets losing their value, implying that the higher the amount of capital the bank holds, the
5 The three Basel II pillars were a minimum capital requirement, regulatory supervision and market discipline. The minimum capital ratio required banks to hold a certain proportion of regulatory capital to risk weighted assets. Secondly, the supervisory review pillar provided a framework for how to deal with various risks and lastly, the market discipline pillar enforced a set of disclosure requirements
more depositors are protected against consequences from asset value declines. Therefore, from a regulator’s standpoint, one needs to weigh the effect of the riskiness of assets against the adequacy of bank capital, signifying somewhat of a substitution effect. Nevertheless, Peltzman (1970) argues that regulatory agencies put too much emphasis on the adequacy of capital, which they see as the single most important indicator. Instead, they would be better off reviewing the details of banks’ asset portfolios. Interestingly, the new Basel III regulation and the LCR in particular shift this emphasis away from the adequacy of capital versus assets to the evaluation of the riskiness of assets. The riskiness of what the authors call the asset portfolio shifts more to the foreground by ensuring that financial institutions do not only hold a sufficient amount of assets but high quality liquid assets specifically. Thus, in addition to regulating capital through capital ratio requirements like the Common Equity Tier 1 (CET1) ratio6, new regulations also take into account liquidity aspects through mitigating the riskiness of assets. Through this process, the substitution effect diminishes as the instruments through which depositors and the banking system can be protected are no longer viewed as mutually exclusive. Regulators no longer need to decide whether tackling the issue of the capital adequacy or mitigating asset riskiness is more important, they simply do both. Mingo (1975) connects to Peltzman’s (1970) line of reasoning in a way that he also defines capital as an input into the production process, they both share the same view about the function of capital for financial institutions and banks. Yet, opposed to Peltzman’s results, Mingo (1975) finds statistical significance for substantial regulatory effects on bank capital. The actual level of bank capital is higher under bank capital regulation than without. Mingo (1975) finds that the difference in findings stem from specification errors and the use of aggregate data in Peltzman’s (1970) research. Dietrich and James (1983) build on Mingo’s study but improve deficiencies by taking
6 The CET1 is a measurement of a bank’s financial strength, calculated as the ratio of the bank’s core equity capital over total risk-weighted assets.
into account the role of uninsured depositors and other bank regulations. Contrarily, they find no evidence that regulation is effective in determining changes in bank capital. Therefore, they draw the same conclusion as Peltzman (1970) and contradict the findings of Mingo (1975). The exact reasons for why these findings differ are beyond the scope of this thesis, what is important is to note that there is a clear debate regarding the effectiveness of regulation.
Further, Koehn and Santomero (1980) also seek to answer the question regarding the viability of regulation in terms of capital requirements. Regulatory constraints are designed in order to not only protect those who deposit their money at the bank but also the banking system in general. They predict that the effect of capital regulation on bank safety is unclear and depends on factors like whether or not the capital ratio requirement is binding. Thus, according to the authors, higher capital requirements do not necessarily lead to lower bank failures but they, in fact, can. When the regulation requirement is not binding, the consequences are opposed to their initial intent, which means higher capital ratios lead to higher bankruptcy risk. They find that capital ratio requirements are not an adequate tool to protect the financial system as it does not sufficiently mitigate the risk of bankruptcy. It does not control for the riskiness of banks and the probability of failure as it assumes that the effect would be too similar on all banks, meaning that regulations are too general to be effective instead of tailoring requirements to banks individually. Consequently, they suggest that other instruments should control for the probability of failure like asset restrictions. Nevertheless, the authors identify the bindingness of the regulation as the differentiating characteristic for the effectiveness on banks. This thesis, as will be explained in the methodology, differentiates between banks that are more and those that are less affected by the implementation of the liquidity requirements. For some banks these liquidity requirements serve as constraints as for others they do not, which, in this thesis, is the differentiating characteristic between treatment and control group. Thus, I adapted part of the author’s methodology by examining how binding the requirement should be prior to
implementation in order to separate those who should be more and those who should be less affected. Koehn and Santomero (1980) therefore look at the amount of regulatory capital, the risk of the bank portfolio and chance of bankruptcy in order to evaluate the effect of capital regulation. Their approach differs from this research in the sense that in this thesis, the CDS spread is used as a proxy for what the authors call “the chance of bankruptcy”. They identify a variable using the Chebyshev Inequality model7 to find that the probability of bank failure can decrease, increase or stay the same. Subsequently, they conclude that the effectiveness of capital regulation on the reduction of bankruptcy risk depends on whether or not the ratio requirement is binding or not.
Compared to the previously mentioned research, a more recent study by Allen et al. (2012) investigates the economic impact of Basel III on economic activity. As banks are required to maintain more liquid assets on their balance sheet and thus cannot lend it out or gain any return, credit supply would decline. Basel III could therefore jeopardize credit availability and lead to the reduction of economic activity. They furthermore emphasize the importance of coordination between banks and investors such that regulatory frameworks can be efficiently implemented instead of having adverse effects. If this is not the case, then what they call “the cure”8, will turn out to be worse than “the disease” by impairing the current situation without any regulatory capital and liquidity requirements. Therefore, the authors also do not take any position regarding the viability of capital and liquidity regulation. They suggest that the effect depends on how well banks and regulators manage to implement these requirements in the transition phase, requiring close coordination of monetary policy in combination with liquidity provisions from central banks and the government. A connection can be drawn between their
7 Chebyshev’s Inequality is a measure of the distance from the mean, expressed as a probability. At least 1 - 1
k2 of
the distribution’s values are within k standard deviations of the mean. Koehn and Santomero applied this measure to identify the probability of bank failure.
research and this thesis, which is that both put weight on what Allen et. al (2012) call the “transition phase”. This transition phase corresponds to the two-year phase-in periods of the LCR. Based on this, one can hence say that the line of reasoning found in Allen et. al (2012) is continued in this thesis. The output on banks’ CDS spreads can serve as an answer of how well they were able to overcome the difficulties of the transition phase up to March 2016.
In a study by Choi and Choi (2016), they find that banks tend to rely less on safe funding sources like retail deposit funding when central banks engage in monetary tightening. Initially, this would adversely influence the banking sector’s aggregate lending. To mitigate this effect, banks must rely more on wholesale funding, meaning that a change in the central bank’s monetary policy affects the funding composition of banks. Choi and Choi (2016) find a substitution effect between deposit funding and wholesale funding caused by the monetary requirements. This substitution effect is more predominant for larger banks because they already have access to wholesale funding markets and thus can increase their funding by more than smaller ones. As a consequence of the substitution of funding sources, funding liquidity risk increases. This is especially critical for large banks or SIFIs that are already heavily exposed to liquidity risk, amplifying the adverse effect of monetary tightening. Thus, according to the authors, liquidity requirements stimulate a substitution effect more for larger banks, which in turn increases liquidity risk and could consequently threaten the economy. They find an inverse relationship between bank size and the level of their liquidity ratio. The ratio the authors used is one by Choi and Zhang (2014) that measures the liquidity mismatch on bank’s balance sheets. The Liquidity Stress Ratio’s (LSR)9 mismatch is represented by the proportion of how long-term assets are funded with short-term liabilities, exposing banks to liquidity risk.
9 The Liquidity Stress Ratio calculates the potential liquidity shortfall of an individual bank holding company, or aggregated bank holding company, during a liquidity stress scenario, using publicly available data sources (Federal Reserve Reporting Forms FR Y-9C). The data and information regarding the LSR was provided by the author Dong Beom Choi and Lily Zhou.
The LSR consists of the sum of liquidity-adjusted liabilities and off-balance-sheet exposures in the numerator and liquidity-adjusted assets in the denominator. A bank is therefore more likely to encounter liquidity difficulties when its LCR is higher. Choi and Choi (2016) find that because of the enforcement of liquidity regulation, large bank’s LSRs increase significantly.
Contrarily, Dubois and Lambertini (2016) find that liquidity regulation does not reduce bank lending but instead increases it. Due to higher amounts of safe and liquid assets, the bank’s asset portfolio becomes safer. According to the authors, this leads to a decrease in moral hazard, meaning these banks are able to leverage more. As they can lend more while keeping the same probability of default, the authors find that enforcing liquidity requirements does not crowd out credit to firms but enforces it instead. Their findings contradict with those by Choi and Choi (2016) in a way that Dubois and Lambertini (2016) find that liquidity requirements do not contract bank lending. Due to regulation, banks borrow more by increasing both deposits and wholesale funding, which in turn reduces their LCR. Despite being more leveraged, they are subject to a smaller liquidity mismatch on their balance sheet due to the higher amount of required liquid assets.
2.3 Determinants of Credit Default Swap Spreads
In the last years, credit derivatives like CDSs have emerged to become essential instruments to hedge against or to take on credit risk. As the CDS market has expanded immensely in the last two decades, so has the amount of academic literature covering the determinants of CDS prices and spreads. CDS prices peaked during the 2008 financial crisis due to the market perception of very high credit risk, which were valued for an outstanding gross notional value of around $60 billion in 2007. Peaking spreads caused by the crisis were mainly brought upon by banks that were engaging in speculative actions. Thus, historic events have proven that crises play a significant role in the pricing of CDS. Nevertheless, there are many factors that
impact credit derivative spreads, whereas this thesis investigates whether liquidity regulation is one of the determinants.
Chiaramonte and Casu (2012) investigate the determinants of CDS spreads before, during and after the financial crisis. They find that bank CDS spreads are a good proxy for bank risk as they can be seen as a bet on an institution’s strength (Hart and Zingales, 2010). Their study relates to this thesis as they regress CDS spreads against not only variables like capital requirements and balance sheet ratios but also against liquidity measures. Their results show that prior to the crisis, a period they define until 30 June 2007, liquidity indicators were insignificant. Yet, they became significant during and after the crisis. Their findings can serve as a first impression on how spreads react as they examine how changes in liquidity determine changes in CDS spreads.
Galil et al (2012) investigate the determinants of the change in credit spreads by presenting four different models. They find that stock returns, the change in stock return volatility and the median spread change of all firms within the same rating group, carry the most explanatory power regarding CDS spreads. Other variables with some explanatory power are the change in spot rates, the change in term slope, the change in VIX, the growth rate in industrial production and the change in the term structure between the 20-year U.S. yield rate and the 1-year U.S. yield rate. Most regression results in their research turn out to be significant and therefore, what they find to be direct determinants, are used as control variables in this thesis. Specifically, the change in stock returns, the term slope, and the VIX volatility measure are implemented as control variables in order to control for omitted factors and avoid endogeneity.
Greatrex (2009) compares CDS adjustments upon earnings announcements relative to stock market reactions by taking credit ratings into account. According to her findings, not the reference entity’s credit quality or industry effects are the most significant factors driving CDS price changes, but the magnitude of the earnings announcement surprise. Furthermore, CDS
markets do not react efficiently to good information in quarterly earnings announcements. Instead, markets seem to overreact to negative earnings news. Subsequently, markets overreact to bad news and do not react to good news when setting CDS prices. On the contrary, Zhang and Zhang (2011) find support for the efficiency of incorporating information from earnings announcements in CDS markets. Furthermore, they find that CDS spread changes prior to earnings announcements serve as good predictors of the direction of earnings surprises. These two studies by Greatrex (2009) and Zhu and Zhu (2011) can be related to my research in a way that they both relate to changes in CDS spreads due to the announcement of earnings announcements or, in this case, Basel III and the LCR. Although there are no direct parallels between a quarterly earnings announcements and a onetime regulatory framework announcement and implementation, comparing the results of the magnitude of CDS spread changes to these different events should yield interesting results.
Norden and Weber (2009) examine the CDS and stock market response to credit rating announcements and find that both markets anticipate rating downgrades ranging up to 90 days prior to the announcement date. In case of a downgrade review by rating agencies, the CDS market reacts earlier than the stock market. Blanco et al. (2003) find similar results such that the CDS market leads the bond market as most of the price discovery takes place in the CDS market. Hull et al (2004) find that CDS spread changes have predictive power for downgrades by rating agencies. Downgrade reviews comprise significant information whereas actual downgrades and negative outlooks do not. The aspect of anticipation can be related to the preparation of banks to the implementation of the LCR. Just as CDS prices adjust prior to anticipated changes in credit ratings, CDS prices adjust to enforcement dates of Basel III regulations.
2.4 Effect of Basel III on bank liquidity and asset prices
Hong et. al (2007) find that liquidity risk has an impact on banks through systematic and idiosyncratic channels, which can have different effects on bankruptcy risk. They develop a model that relates liquidity risk to the probability of bank defaults, which yields that both the LSR and Net Stable Funding Ratio10 (NSFR) have limited effects on bank failures. An effective liquidity regulation therefore needs to target liquidity risk management not only at the individual bank level but at the system level in general. The authors confirm that banks increase their LCR when they anticipate financial distress, they are subject to high insolvency risk and when anticipating downturns in economic conditions. When banks increase their liquidity standards as a consequence of anticipating worsening economic conditions, it is referred to as liquidity hoarding (Hong et. al, 2007). Their research is similar to this thesis as they investigate the impact of the LCR on a measure of bank failures. Nevertheless, they only take a time period from 2001 until 2011 into account, which hardly captures the effect of the Basel III announcement. Therefore, this research can be seen as somewhat of an extension of Hong et. al (2007), using different models and capturing the actual effect of the regulation. Gârleanu and Pederson (2007) find that liquidity hoarding can lead to negative externalities like illiquidity at the market aggregate level. Market level illiquidity corresponds to drying up of financial markets, which can ultimately cause bank failures (Hong et. al, 2007). De Haan and van den End (2013) investigated the reaction of 62 Dutch banks upon the Liquidity Balance (LB) rule11, which is similar to LCR. They find that most banks Dutch tend to hold more liquid
10 The NSFR is a standard developed to promote the median and long-term funding stability for banks. It is calculated as the ratio of available stable funding and required stable funding, which depend on the definitions of each of these two terms, such as differences in classifications of stable assets and liabilities. This NSFR is required to be above 100%.
11 This Dutch liquidity supervisory framework, which was enforced in 2003, is the difference of available liquidity and required liquidity over required liquidity. This ratio needs to be equal or higher than zero at all times.
assets against liquid liabilities, than is required under the LB rule. This is especially the case for smaller Dutch banks and foreign subsidiaries. Banks that are more solvent on the other hand, hold less liquid assets against liquid liabilities, indicating a relationship between capital and liquidity buffers. Depending on the expectation of the bank’s one-month future cash outflows, they tend to hold the respective amount of liquid assets on their balance sheet. Subsequently, they find that banks with higher capital positions tend to hold less liquid assets in proportion to liquid liabilities.
Furthermore, a study by Slovik and Cornède (2011) showed that Basel III has an impact on lending rates of banks. They estimate the impact of Basel III on economic output, bank capital levels and lending spreads. In order to be able to meet Basel III requirements, banks would increase their lending spreads by approximately 50 basis points. They find that a one percentage point increase in the ratio of capital to risk weighted assets pushes up bank lending spreads by 20.5 basis points. Consequently, the increase in bank lending spreads influences the interbank borrowing behavior. This ultimately impacts the implied credit risk and could potentially cause changes in CDS spreads. Slovik and Cornède’s (2011) research can be related to a study by Bech and Keister (2012) who find that the impact of the LCR depends on how close banks are to the LCR requirement once it is implemented. The closer the bank’s asset margin is to the LCR requirement, the smaller the aggregate effect. Furthermore, loans with a maturity longer than 30 days improves the LCR while overnight interbank borrowing does not12. Therefore, longer term maturities are more valuable than loans with a maturity of 30 days or less, which increases the term premium in the very short end of the yield curve (Bech and Keister, 2012). Subsequently, banks engage more in interbank borrowing in the long-term and therefore increase their leverage exposures in order to meet LCR requirements. As a consequence of
12 Bech and Keister (2012) find that interbank loans with a maturity of longer than 30 days increases the numerator while not falling under the 30-day total net cash outflows, thereby not decreasing the LCR.
more and riskier long term borrowing, lenders endure more credit risk, which they might want to ensure by purchasing insurance policies like credit default swaps.
2.5 Relation to existing literature and hypotheses
Prior literature has covered the general effects of regulatory requirements on different asset classes, their prices and spreads. The views on adequacy and their findings strongly differ and are sometimes contradicting. Most research has been conducted on minimum capital requirements whereas liquidity requirements have mostly been neglected. With the implementation of the LCR, the issue about the competence of liquidity regulation has firstly been tackled. Although there are some predictions about the effect of the LCR on asset prices, empirical research is limited. This leaves an unfilled gap that this thesis is going to fill by providing empirical evidence on the direct consequence of the LCR announcement and implementation on a credit derivative. Although Hong et. al (2014) discuss the implications of the LCR on U.S. bank bankruptcy risk, the time period is limited only until 2011. This thesis therefore extends their research, yet by applying different methodological models. Not only does this research explore a new field of the effect of a new regulation by looking at CDS spreads, it also combines the difference in magnitude of reactions between announcement and actual implementation. Therefore, this study connects to and extends research by Allen et. al (2012) in a way that it seeks to empirically answer the question whether banks and regulators managed to successfully overcome difficulties that they encounter during transition phases. This question was left open and unanswered by the authors as the implementation phase has not yet taken place by the time their research was published.
H1: The announcement of the LCR leads to a decline in U.S. bank CDS spreads.
H2: The implementation of the LCR leads to a decline in U.S. bank CDS spreads.
H3: Banks with lower liquidity levels experience a stronger change in CDS spreads than
banks with higher liquidity levels in the announcement period.
H4: Banks with lower liquidity levels experience a stronger change in CDS spreads than
banks with higher liquidity levels in the implementation period.
Whilst CDS spreads are expected to adjust upon the announcement of Basel III and thereby the announcement of the LCR, the relationship of the actual implementation and CDS spread adjustments are less clear. On the one hand, when Basel III was announced, investors had the chance to react quickly, right when the news were published. Therefore, an inverse relationship between the announcement of the LCR and CDS spreads is expected. An inverse relationship corresponds to declining CDS spreads upon the Basel III announcement, which is in line with H1. On the other hand, banks prepare themselves for prior to actual implementation dates like the LCR by i.e. increasing the value of their stock of high quality liquid assets. Subsequently, CDS spreads are not expected to change by as much as when Basel III was initially announced. Thus, the effect on CDS spreads is expected to be smaller the closer the bank’s pre-announcement and pre-implementation margins are to the required LCR. Lastly, the liquidity levels referred to in H3 corresponds to the level of each bank’s Liquidity Stress Ratio as explained in section 3.
3. Liquidity Coverage Ratio
An asset’s liquidity is measured by the ease at which the asset is convertible into cash. Banks and financial institutions have many different types of assets on their balance sheets that have different liquidities. Cash, for example, is the most liquid asset, whereas other very liquid assets are e.g. Treasury Bills because they can be easily traded at their market price. Treasury Bills are very liquid because the buyer of the asset is able to resell the same asset again without a loss, thus the exchange of a Treasury Bill into cash is easy. The more of a loss the seller would incur in the sale of the asset, the less liquid or the more illiquid an asset becomes. An example for an illiquid asset would be real estate. When a company sells its real estate, it typically incurs a loss because real estate cannot be sold quickly in the market. This is because the value of the real estate may be based on subjective valuation and complexity, which potentially causes unwillingness to purchase the real estate. Subsequently, liquid assets can easily be traded because they can be sold at their fair value whereas illiquid assets are harder to sell due to their difficulty to convert into cash.
The amount of liquid assets a bank or financial institution holds on its balance sheet are of upmost importance because they can be easily and quickly exchanged for cash if needed, which determines its liquidity position. On the one hand, banks do not have strong incentives to hold on to cash and other liquid assets because they can be lent out as a credit, gaining interest or could be invested differently for a return on that investment. Thus, when liquid assets are simply held on the balance sheet without any further use, they do not gain any return. On the other hand, banks do have an incentive to hold on to these assets. When, for example, short-term or long-short-term debt matures and needs repaid, they can serve as a buffer. More importantly, in times of financial distress, the availability of easily convertible assets can make the difference between being solvent or bankrupt. Consequently, banks hold on to liquid assets for liquidity reasons.
This is where the rationale of the LCR comes into play. The Liquidity Coverage Ratio not only requires the institution to hold any liquid assets but High Quality Liquid Assets (HQLA). The LCR standard aims at improving banks’ standalone liquidity positions such that central banks do not become lenders of first resort. In the past and specifically prior to the financial crisis, banks were allowed to include many different securities as liquid assets. In times without any financial distress, instruments like mortgage-backed securities were regarded as liquid. Yet, in times of financial distress, these markets tend to dry up, making them illiquid. For this reason, the Basel Committee on Banking Supervision and regulators redefined the characteristics that make them fall under high quality liquid assets. In essence, these liquid assets should be and remain liquid, independent of the state of the market and economy and be qualified for operations by the central bank. Examples of HQLA are cash, central bank reserves and certain marketable securities backed by sovereigns and central banks. These assets are typically of the highest quality and most liquid. The Liquidity Coverage Ratio is defined as:
High Quality Liquid Assets (HQLA)
Total Net Cash Outflows ≥ 100%
This ratio corresponds to a required LCR of 100 per cent, which conforms to the full implementation by January 2019. Next to identifying the exact assets that fall under the HQLA characteristics, measuring the amount of total net cash outflows, is an even more complicated accounting process. The amount of total net cash outflows is defined as the difference between total expected net cash outflow and expected net cash inflows in a specified stress scenario for the subsequent 30 calendar days. Examples are cash outflows needed to comply with derivative, investment, debt, and other contractual obligations that are not included elsewhere. The definition of HQLA is continuously subject to new reforms and rules, i.e. as of July 1st 2016, investment grade U.S. general investment state and municipal securities may be included
in the calculation of HQLA. Calculating the LCR is a complex accounting process as identifying each asset that falls under HQLA characteristics is subject to regulatory liquidity metrics. Although measuring the exact amount of HQLA is a very complex process, identifying SIFIs that are concerned is more simple.
4. Methodology
In order to test the stated hypotheses, certain panel datasets are required. The first repeated cross-sectional dataset consists of the daily CDS spreads of each bank. In addition, bank fixed effects, variables or characteristics that differ across banks but are time invariant, are included as well. By adding control variables to the model, the regression is less likely to suffer from omitted variable bias. Not controlling for omitted variable bias would distort the regression model as determinants of CDS spreads that are also correlated with the independent variable of interest, are not included in the model. When including controls, the the coefficient of the regressor becomes a more precise estimator of the dependent variable. More specifically, some variables determine CDS spreads and at the same time are related to the Basel III announcement. When e.g. markets anticipated the Basel Committee on Banking Supervision to publish with a new regulatory framework, markets might react by becoming more volatile prior to the announcement. This in return impacts the spreads on CDS, which makes it necessary to include a variable that controls for this volatility. As a consequence, this ensures that the mean independence assumption holds, such that the regression model is valid and unbiased.
4.1 Control Variables and bank fixed effects
Control variables that are included in the model are the VIX, Term Slope, Fama-French (F&F) factors, Stock Price Returns and the 1-month Treasury Yield. Since these are also recorded on a daily basis, they serve as an adequate match for the regression.
The VIX control variable is a measurement of near term implied volatility of index options in the S&P 500 from the Chicago Board Options Exchange (CBOE) (Collin-Durfense et al., 2001). It is perceived as an indicator for investor sentiment and market volatility, which makes is a convenient measure of controlling for volatility in markets. According to Figuerola-Ferretti and Paraskevopoulos (2010), the CDS market is correlated to the VIX in a way that there is a lead price discovery process in the VIX market over the CDS market. This proven relationship requires inclusion of the VIX as control variable to overcome omitted variable bias. As higher volatility is correlated with higher risk, the expected coefficient of the VIX is expected to be positive. This is because higher volatility could relate to more uncertainty and ultimately to higher credit risk.
The term slope control variable is the difference between the yield on a 10-year U.S. Treasury bond and a 2-year U.S. Treasury note (Campello et. Al, 2016). Fama and French (1989) found that a steepening in the yield curve13 slope and therefore a higher term slope, anticipates stronger economic growth. Moreover, an inverse relationship between the term slope variable and CDS spreads is expected. This is analogous to a negative regression coefficient for the term slope.
The three F&F factors used in this regression are High-Minus-Low (HML), Small-Minus-Big (SMB) and the market excess return (MKT). The HML factor is the difference in the return on the high-book-to-market-ratio stocks portfolio and the low-book-to-market-ratio stocks
13 The yield curve is a line that plots interest rates on the Y-axis and maturity dates on the X-axis of bonds with the same credit quality at a certain point in time.
portfolio. The SMB factor is the difference in the small market capitalization portfolio and the big market capitalization portfolio. Market capitalization is calculated by multiplying the company’s share price with the amount of outstanding shares. Lastly, the MKT factor is the difference between the market return, which is the value weighted return of all NASDAQ, AMEX and NYSE stocks, and the risk free rate, which is the one-month Treasury bill rate. Thus, the MKT factor accounts for the excess return on the market. By including the F&F factors, the model accounts for company size, company price-to-book ratio, market risk and market efficiency. Higher factor values correspond to better economic conditions due to higher asset values. These higher asset values make banks safer, which should reduce spreads. Therefore, the coefficients of the three F&F factors are expected to be negative.
Each bank’s daily stock prices were retrieved to yield daily stock returns. When a bank’s stock prices increase, so does its market capitalization. Subsequently, the bank’s market value increases as a consequence of higher stock prices. The Merton model finds that increasing firm values should be negatively related to the probability of default. Thus, spreads should be inversely related with the market value of equity, which means that a negative correlation coefficient is expected.
Furthermore, the TED spread is the difference between the month LIBOR rate and the 3-month Treasury Bill rate. It is the difference between the rate that banks pay on interbank loans and the yield on short term U.S. government debt. The TED spread has a direct effect on CDS spreads as it can be seen as an indicator for credit risk. As Treasury Bills are considered risk free and the LIBOR represents an interbank rate, increasing TED spreads indicate more credit
risk14. Consequently, an increase in the TED spread should lead to increasing CDS spreads and therefore a positive coefficient is expected.
Bank size, profitability and tangibility are bank fixed effects, signifying time invariant bank characteristics. The larger the amount of assets on the bank’s balance sheet, the larger the bank’s size. With more assets and therefore a better protection against default, the bank should be seen as safer than a bank with less assets, everything else equal. Consequently, as assets can serve as an indication of security and protection against default, credit risk should be lower. The lower credit risk should decrease CDS spreads, which indicates a negative relationship between bank size and spreads. The return on assets is used as a proxy to measure the bank’s profitability. As more profitable banks are less likely to default on their debt, holding all else equal, an inverse relationship is expected between CDS spreads and profitability. Lastly, the correlation tangibility coefficient could be negative or positive. Bank fixed effects impact the level of CDS spreads while not changing within the treated time period. These bank fixed effects therefore do not produce any change in spreads over the time period but are important determinants of CDS spreads, making them essential variables to account for time invariant bank characteristics.
Appendix A. summarizes each variable included in the regression model with its respective data source and predicted coefficient.
14 The TED spread is measured in basis points and can be seen as an indicator of credit risk as it is driven by the difference between interbank rates and the risk free rate. Whether the 3-month Treasury Bill rate is actually risk free is debatable but it is assumed here for simplification.
4.2 Separating treatment and control group
In order to conduct a difference-in-difference (DiD) regression, the treatment- and control group need to be specified. The differentiating factor for this research is the bank’s liquidity as the LCR is a framework that aims at improving liquidity standards. In this research, the Liquidity Stress Ratio (LSR) is used as the representative measurement of bank liquidity prior to the Basel III announcement in December 2010.
The LSR is a measure of a bank’s balance sheet liquidity mismatch. It assesses the liquidity shortfall in case of a liquidity stress test scenario (Choi and Zhou, 2016). The numerator represents adjusted liabilities and off balance sheet exposures whereas liquidity-adjusted assets are in the denominator. Thus, banks with a lower LSR are more liquid and should experience less liquidity issues during liquidity stress test scenarios. Contrarily, banks with a higher LSR should experience more severe liquidity problems due to their high amount of liabilities, lower levels of liquid assets, or both. The Liquidity Stress Ratio is defined as follows:
LSR = Liquidity-Adjusted Liabilities and Off-Balance Sheet Exposures
Liquidity Adjusted Assets
Using the LSR as a differentiating measure of more and less liquid banks prior to the announcement yields the following regression model:
(1) CDSit = α + βLSRit + εit
This model regresses each bank’s LSR against its respective CDS spread, with β being the coefficient of interest. It measures the effect of the treatment indicator, the LSR, on the outcome variable, the CDS spread. Banks that have higher ratios, are less liquid than those with lower
ratios and intuitively, the implementation of a minimum liquidity requirement should affect those more that have lower liquidity standards, meaning higher LSR. Hence, in the future they must stack up their balance sheet with more liquid assets in order to meet requirements. Consequently, banks with below median LSR coefficients in regression (1) belong to the treatment group and those with above median LSR coefficients belong to the control group. The allocation between treatment and control group is the same for all regression models.
4.3 Difference-in-Difference Regression Model
This model estimates the change in pre-announcement (or pre-implementation) and post-announcement spreads (or post-implementation) between treatment and control group banks. This model thereby captures the effect of the event and simultaneously adjusts for differences in pre-treatment spreads between groups. The resulting estimator is a difference-in-difference (DiD) estimator since the estimator is the difference across groups over time. The corresponding variables for establishing the regression estimator are Y treatment, before , which is the average CDS spread of all banks that are in the treatment group before the event ,
Y treatment, after, the sample average spread of the treatment group after the event, Y control, before, the sample average of all banks in the control group before the event and lastly Y control, after, the average spread of the control group banks after the event (Stock and Watson, 2015). Therefore, the DiD estimator, β3DiD is defined as:
(2) Β3 DiD
= Y treatment, after - Y treatment, before – Y control, after - Y treatment, before
Subsequently, the DiD estimator is the difference of the average change in the treatment groups’ CDS spread change and the control groups’ CDS spread change.
(3) β3 DiD = ΔY treatment - ΔY control
Equations (4), (5) and (6) below are the DiD regression models that test the hypotheses. The dependent variable of interest in these models is the CDS spread of bank i at time t. All three regressions contain the same set of fixed effects, dependent, independent and control variables. The only difference lies with time dummy as the DiD regressions are conducted in different time periods. The three difference-in-difference regression models are defined as:
(4) CDSit= β0+ β1DAnnouncement+ β2DTreatment+ β3DiDDAnnouncement* DTreatment +
β4TermSlopet + β5VIXt + β6HMLt+ β7SMBt + β8MKTt +
β9Stockreturnit + β10TED t + β111monthYieldt + β12BankFixedEffectst + εit
(5) CDSit= β0+ β1DImplementation2015 + β2DTreatment + β3DiDDImplementation2015* DTreatment+
β4TermSlopet + β5VIXt + β6HMLt + β7SMBt+ β8MKTt +
β9Stockreturnit + β10TED t + β111monthYield t+ β12BankFixedEffectst + εit
(6) CDSit= β0+ β1DImplementation2016 + β2DTreatment + β3DiDDImplementation2016* DTreatment+
β4TermSlopet + β5VIXt + β6HMLt + β7SMBt+ β8MKTt +
β9Stockreturnit + β10TED t + β111monthYield t+ β12BankFixedEffectst + εit
Regression (4) estimates the outcome variable during the announcement time period, which lies between December 16th 2010 and January 15th 2011. By implementing a 30-day time window, the outcome should capture the immediate short run reactions as well as the
adjustments within the next month. A 30-day time window is used to match the 30-day total net cash outflow time period as indicated in the LCR. Model (4) thereby regresses the CDS spreads against the time dummy, treatment dummy, the interaction term and control variables. Regression model (5) resembles regression (4) except for the fact that the time period lies between 01/02/2015 (the first working day in 2015) and 02/02/2015. Lastly, regression (6) is similar to (5) and (4) except for the time period, which ranges from 01/04/2016 (first working day in 2016) until 02/01/2016.
The first dummy variable, D Announcement , is a time dummy that corresponds to whether the regression takes place during the announcement period, thus D Announcement = 1 if it lies in
the time period between December 16th 2010 and January 15th 2011 and 0 otherwise. The same applies to the regression (5) and (6) time dummy variables DImplementation2015 and
DImplementation2016. The 2015 dummy equals 1 if regression (5) is between January 1st 2015 and February 2nd 2015 and 0 otherwise. The 2016 dummy equals 1 if regression (6) is between January 1st 2016 and February 1st 2016 and 0 otherwise. By implementing a time dummy variable, the model ensures that only the correct time periods affect the outcome variable.
The second dummy variable is the treatment dummy denoted as DTreatment. DTreatment equals 1 if bank i belongs to the treatment group and DTreatment= 0 if it belongs to the control
group with β2 as the corresponding coefficient.
The third coefficient of interest, β3 DiD, is the coefficient of the interaction term D Basel * D Treatment. The interaction term equals one for banks in the treatment group and within the respective time window. Therefore, β3 DiD is the difference of the change in the treatment and the change in the control group in the respective points in time.
Following this regression model, potential endogeneity issues need to be addressed. It is unlikely to run into simultaneity problems as the LCR impacts CDS spreads15 but vice versa, CDS prices do not drive the passage of this legislation. During the development phase of Basel III, regulators implausibly based the design of the liquidity requirement on bank CDS spread levels. Instead, other factors were taken into account that served as important determinants for the LCR design. Therefore, CDS spreads improbably have a causal effect on the LCR but the LCR has a causal effect on CDS spreads, meaning reversed causality issues are unlikely. Besides, by adding bank fixed effects to the regression model, heterogeneity in bank characteristics are accounted for. In order to control for further endogeneity problems, all control variables are lagged by one day16.
5. Data and Descriptive Statistics
Data for the mean 5-year CDS spreads were accessed via DataStream under the Thomson Reuters Composite Database. This 5-year CDS spreads panel dataset ranges from March 2010 until March 2016 for each U.S. bank. All spreads are denoted in basis points, were recorded on a daily basis for every working day and are written on senior debt only. Furthermore, all CDS comply with the ISDA “No Restructuring” clause standards, meaning out-of-court restructurings are not counted as credit events, therefore only CDSs that did not undergo any restructurings were used in this dataset. The daily yields on 10-year and 2-year Treasury Bond yields were retrieved from the U.S. Treasury database. The difference between the 10-year and 2-year U.S. Treasury Bond rate yields the daily Term Slope variable. The VIX volatility data
15 The effect of the LCR on CDS spreads is in fact scope of this thesis. For simplicity and clarity reasons, it is assumed that there is a causal relationship.
16 All control variables except for the TED spread are manually lagged as the TED dataset is already lagged due to the lagged recording of the LIBOR series (Federal Reserve Bank of St. Louis, 2016)
was accessed from the Chicago Board Options Exchange and is also recorded daily. Data for the F&F three-factor model was retrieved from the Kenneth-French database and stock prices for each bank were taken from the CRSP database and were used to manually calculate daily returns. In order to control for heterogeneity in bank characteristics like bank size, profitability and tangibility, the required data was retrieved from the Compustat database. Most balance sheet items that are needed to calculate bank fixed effects are published annually or quarterly and since this research is based on short time windows, quarterly data is used to control for characteristics that differ from one bank to another but are time invariant. Bank size is calculated as the natural logarithm of one plus total assets (Compustat item ATQ). Profitability is calculated as the ratio of net income before interest expenses (Compustat item NIQ minus XINTQ) to total assets. Lastly, Tangibility is the ratio of plant, property and equipment (Compustat item PPENT) to total. The LSR panel dataset was provided by Dong Beom Choi from the Federal Reserve Bank of New York.
Although the retrieved data is for the six-year time span, the three most important periods for the regressions are the following: Firstly, the announcement date of Basel III and the LCR, that took place on 12/16/2010. Second, the first phase in date, which corresponds to the 30-day time period starting on 01/02/2015. Thirdly, the second transition phase sets in starting 01/04/2016. Dates regarding announcement and implementation dates of the regulatory Basel III framework and the LCR were collected manually from various sources like the Bank for International Settlements, Bloomberg and the Federal Reserve website.
5.1 Summary Statistics
Table 1. presents summary descriptive statistics for each bank and independent variable in the regression model. Panel A. summarizes each bank’s CDS spreads showing the number of observations, the mean, median, standard deviation and minimum and maximum CDS spread between March 2010 and March 2016. Panel B. examines summary statistics for each fixed effect and control variable.
Table 1. Summary Statistics
Panel A. Summary Statistics on Bank CDS Spreads
Obs Mean Median Std. Dev. Min Max
US Bank
American Express 1,747 67.27 63.88 25.83 34.49 161.56 Bank of America 1,747 143.90 116.91 87.78 58.25 487.54 Bank of New York Mellon 1,747 94.61 103.18 19.29 48.81 104.01 Capital One 1,747 89.89 94.09 32.74 43.38 184.21 Citigroup 1,747 132.09 111.79 61.26 58.54 358.30 HSBC 1,747 88.25 63.06 65.85 25.85 298.27 Goldman Sachs 1,747 145.01 122.72 71.54 63.81 423.52 JP Morgan Chase 1,747 87.30 81.52 25.87 48.34 187.13 Morgan Stanley 1,747 169.75 139.31 104 59.30 609.81 PNC Financial 1,747 82.23 73.68 15.16 59.52 136.52 UBS AG 1,747 97.83 88.97 48.03 34.44 246.24 Wells Fargo 1,747 76.45 71.38 29.68 35.39 181.85 Mean 1,747 106.22 94.21 48.92 47.51 281.58 Median 1,747 92.25 91.53 40.39 48.58 216.69
Panel B. Summary Statistics of dependent variables
Obs Mean Median Std. Dev. Min Max
Variable VIX 17,688 17.67 16.52 4.35 12.73 26.25 Term Slope 17,688 0.02 0.02 0.00 0.01 0.03 Stock Return 17,688 0.00 0.00 0.02 -0.05 0.05 TED 17,688 0.26 0.23 0.00 0.13 0.56 MKTRF 17,688 0.01 0.00 0.01 -0.02 0.01 SMB 17,688 0.00 0.00 0.00 -0.01 0.01 HML 17,688 0.00 0.00 0.00 -0.01 0.01
One-month Treasury yield 17,688 0.06 0.04 0.05 0.00 0.32
Bank Size 216 13.53 13.53 0.97 11.83 14.83
Profitability 216 0.00 0.39 0.00 -0.01 0.03
Tangibility 216 0.01 0.01 0.01 0.00 0.04
This table represents summary statistics for the sample used in this research. The sample contains 12 different U.S. Banks between March 2010 and March 2016. Panel A. summarizes each bank's CDS spread, denoted in basis points (bps). Panel B. shows summary statistics for the dependent variables.
VIX is a near term implied volatility measure from the Chicago Board Options Exchange. The Term Slope is the difference between the ten-year and
two-year U.S. Treasury Bond yield. MKTRF, SMB and HML account for size, value and market risk and are indicators in the Fama-French three factor model. Bank Size is one plus the natural logarithm of the bank's total assets, Profitability is the return on Assets, which is net income plus interest expense, divided by total assets. Tangibility is the ratio of property, plant, and equipment to total assets. All variables are winsorized at the 1-99% level. *, **, and *** represent statistical significance at the 10%, 5%, and 1% levels.
Panel A. in Table 1. above shows that average CDS spreads tend to differ between banks. When purchasing a credit default swap on Morgan Stanley’s debt, the protection buyer had to pay a spread of 169.75 basis points on average between March 2010 and March 2016. Spreads even went as high as 609.81 basis points and as low as 59.30 and there is quite some variation with a standard deviation of 104 basis points. Goldman Sachs’ minimum spread of 63.81 basis points is almost as high as American Express’ median spread of 63.88, signaling that the market perceives Goldman Sachs to be riskier than American Express. Furthermore, volatilities also differ across banks. This could give a first impression of different reactions to events during this time period. Larger standard deviations could mean stronger or weaker investor’s beliefs regarding the riskiness of the bank caused by the event at hand. When, for example, investors believe that Bank of America is more affected by the announcement of the LCR than PNC Financial, then Bank of America spreads should increase or decrease with stronger magnitude than spreads of PNC Financial. The difference in reactions ultimately increases overall volatility. Therefore, higher volatilities can serve as a first impression for which banks are more affected by events like the announcement and implementation of LCR. Panel B. presents summary statistics of the independent variables, control variables and bank fixed effects. The number of observations for the daily recorded variables are equal because observations are omitted if values are missing in the panel data. As balance sheet data is retrieved quarterly, the number of observations for bank size, profitability and tangibility are much lower than the other daily recorded data. Stock returns are on average below 1% between March 2010 and 2016. This could be due to uncertainty in markets, representing the 2% volatility and returns ranging from 5% to -5%. The difference between 10-year U.S. Treasury Bond yields and 2-year U.S. Treasury Bond yields is around 2% on average, signaling a positive risk premium in the yield curve. To minimize the effect of outliers, all control variables are winsorized at the 1%-99% level.
5.2 Correlation Between Variables
Table 2. Correlation Matrix
Variable 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 1. CDS Spread 1.00 2. 10-year Yield -0.17 1.00 3. 2-year Yield -0.32 0.52 1.00 4. 1-month Yield 0.06 0.20 0.31 1.00 5. Term Slope -0.06 0.93 0.16 0.10 1.00 6. MKTRF 0.00 0.01 -0.02 -0.01 0.02 1.00 7. SMB 0.00 0.03 0.01 -0.01 0.02 0.41 1.00 8. HML 0.00 -0.02 -0.01 0.03 -0.02 0.14 -0.09 1.00 9. VIX 0.45 0.02 -0.01 0.15 0.02 -0.17 -0.06 -0.05 1.00 10. Stock Return 0.00 0.01 0.00 -0.01 0.01 0.20 0.08 0.08 -0.04 1.00 11. Bank Size 0.13 -0.07 -0.03 -0.06 -0.06 0.01 0.00 0.02 -0.03 -0.06 1.00 12. Profitability -0.08 0.21 0.09 0.12 0.18 0.00 0.00 -0.05 0.05 -0.03 -0.46 1.00 13. Tangibility -0.17 0.09 0.06 0.06 0.06 0.00 0.00 -0.03 0.01 -0.04 -0.64 0.48 1.00 The table below is a crosstabulation of correlation coefficients of each variable included in the DiD regression. The number of observations for per variable is equal for each bank. Each bank has 1470 observations per variable, thus n = 17641 observations in total. Variable 1. is retrieved from te Thomson Reuters Composite Database via DataStream.Variables 2 to 5 are taken from the U.S. Treausury database. 6, 7 and 8 are the Fama & French three factor models taken from the Kenneth French Data Library. Items 9 and 10 are taken from the Chicago Board Options Exchange and the CRSP database, respectively. Items 11, 12 and 13 are retrieved from Compustat.
Table 2. presents a correlation matrix showing a cross tabulation of the relationship among variables included in the DiD regression. Column 1. is of most interest as it shows the correlation coefficients of each regression item on the CDS spread. The U.S. Treasury 10-year and 2-year yields are negatively correlated with CDS spreads, whereas the more short-term yields are positively correlated. The three F&F factors in rows 6, 7 and 8 all have positive correlation coefficients, as expected. Higher factor values indicate better economic conditions due to higher asset values, meaning spreads should decline. Nevertheless, their correlation coefficients are rather small, thereby not being completely confirmative. Item 9, the VIX volatility measure is relatively strongly correlated with spreads. As expected, volatility is a strong determinant of spreads as volatility is a quantification of risk and CDS spreads are a measurement of credit risk, hence the comparatively high correlation coefficient. The positive correlation coefficient of bank size on spreads indicates that the larger the bank, the higher CDS spreads.
