The influence of Basel III on banks exposed to repos:
An investigation of the concave relationship between capital and risk
University of Amsterdam
MSc Business Economics, Finance track Master Thesis, July 2017
Name: Loes van der Jagt Student number: 10244611 Supervisor: R. Almeida da Matta
1
Statement of originality
This document is written by Loes van der Jagt 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.
2
Abstract
The recent global financial crisis exposed the weaknesses of banking regulation. The run on the repo market is seen by some researchers as one of the main causes of the crisis. To overcome the deficiencies in regulation, Basel III was developed. This study tries to investigate the relationship between the repo market and the capital requirements of Basel III. Building upon the paper of Matta and Perotti (2016), this thesis contributes to existing literature by conducting an empirical investigation of the concave relationship between risk and repos. Based on a panel analysis of 12 US banks, no evidence is found for this concave relationship. There is no evidence that the credit default swaps of banks more exposed to the repo market experience a larger increase than those of the banks less exposed to repos. Furthermore, only weak evidence is found for a decrease in the credit default swaps of banks due to the Basel III capital requirements. Due to the limitations of this research, the results are questionable.
3
Table of contents
1. Introduction ... 5
2. Literature review ... 6
2.1. Basel III ... 7
2.2. Repos and the repo markets ... 8
2.3. Repos during the financial crisis ... 9
2.4. The influence of repos on bank runs ... 10
2.5. Relation of literature to the hypothesis ... 11
3. Methodology ... 12
3.1. Determination of treatment and control group ... 13
3.2. Influence of Basel III ... 14
4. Data and descriptive statistics ... 14
4.1. CDS ... 15
4.2. Repo rate ... 15
4.3. Control variables – bank specific ... 15
4.4. Control variables – market wide factors ... 16
4.5. Descriptive statistics ... 17
5. Results ... 19
5.1. Classification of the treatment and control group – announcement date ... 19
5.2. Impact of the announcement of Basel III on banks exposed to repos ... 22
5.3. Classification of the treatment and control group – enactment date ... 23
5.4. Impact of the enactment of Basel III on banks exposed to repos ... 25
6. Robustness checks ... 27
6.1. Change of the treatment and control group ... 27
6.1.1. Excluding Charles Schwab and JP Morgan ... 27
6.1.2. Assigning all banks with insignificant or negative coefficients to the control group 28 6.1.3. Shifting the line between the control and treatment group one bank up or down ... 29
4
6.1.4. Banks with a medium sensitivity excluded from the sample ... 30
6.2. Other announcement and enactment dates ... 30
6.2.1. July 26, 2010 as the announcement date ... 31
6.2.2. September 12, 2010 as the announcement date ... 32
6.2.3. July 2, 2013 as the enactment date ... 32
6.3. Shifting the event date one day ... 33
7. Conclusion and discussion ... 33
References ... 36
5
1. Introduction
The financial crisis of 2007-2009 was the consequence of the rapid expansion and the subsequent collapse of the shadow banking system. Shadow banks are financial institutions that provide maturity, liquidity and credit transformation; they offer cheap funding by converting risky and opaque long-term assets into seemingly risk-free liabilities. One important difference between the shadow banking system and the traditional banking system is that the latter has no access to public sector guarantees nor to central bank liquidity (Pozsar et al., 2010). In the shadow banking system, a severe run on several forms of “safe” short-term debt occurred. There were runs on the market for repurchase agreements (repos), asset-backed commercial papers (ABCP) and money-market mutual funds (MMMFs). Previously, these types of short-term debt had been considered risk-free and “money-like” debt, but eventually, investors realized this was not the case. Indeed, people realized that the short-term debt was not perfectly collateralized and unsafe, which led to bank runs (Gorton and Metrick, 2010). Gorton and Metrick (2012) have even argued that the crisis was primarily caused by the run on repos. Others have suggested that the reduction in repos did not necessarily cause the crisis, although it is evident that these runs had a big and destroying impact on several dealer banks (Krishnamurthy et al., 2014).
Since repos are supposed to have significantly influenced the financial crisis, it is interesting to examine how bank performance is influenced by new regulations with the purpose of ensuring a more stable financial system. Therefore, this paper investigates the impact of Basel III capital requirements on the performance of 12 U.S. banks, while focusing on their exposure to the repo market. The theoretical paper of Matta and Perotti (2016) provides an important theory used in this study. Their paper is the first to suggest that there is a concave relation between the level of repos and the probability of a bank run by unsecured debtholders. Two important factors that influence the decision for the level of repos by a bank are: (i) the probability of default at a certain level of repos and (ii) the advantage of this cheap form of funding. A riskier bank will choose a lower level of repos because the bank has a higher probability of default. Therefore, the bank chooses a minimum level of repos. A safer bank will issue relatively more repos because the probability of going bankrupt is very small and repos are a cheap way of funding. Hence, the bank will undertake the maximum amount of repos possible. Therefore, the capital requirements of Basel III should have more effect on the level of repos for the relatively safe banks than for the riskier banks. This is because the riskier banks already choose a lower level of repos. Thus, the question to be answered is:
6
the riskiness of banks with relatively more exposure to the repo market? The hypothesis is
that safer banks react more strongly to the capital requirements of Basel III than do the riskier banks, since they are supposed to have a greater exposure to the repo market.
To answer the research question, a regression is conducted to distinguish the banks with a large exposure to the repo market (the treatment group) from those that are less exposed (the control group). After separating these two groups, a difference-in-difference regression is conducted to test the influence of Basel III on the performance of U.S. banks, based on their dependence on the repo market. After the basic regression, several regressions are conducted to check the robustness of the results. Changes are made to the construction of the treatment and control group, different dates are used for the announcement and enactment of Basel III and the event date is switched one day to check for noise around the announcement or enactment date and to see if this influences the results.
The results of this research are relevant for policy makers, central banks, commercial banks and investors. This paper contributes to existing literature since it is the first to search for evidence to support this concave relation. It is also the first to assess the impact of this relation on the capital requirements set by Basel III.
The remainder of this paper is organized as follows. The next chapter reviews the relevant literature to provide a theoretical basis and discuss outcomes of previous research. Moreover, the hypotheses are formulated in this chapter. In Chapter 3, the methodology is explained. The data and descriptive statistics are discussed in the fourth chapter. Chapter 5 provides the results and answers the research question. In the sixth chapter, several robustness checks are conducted. Finally, in Chapter 7, a conclusion is given and the limitations are discussed with suggestions for further research.
2. Literature review
This chapter describes the relevant literature useful for this study. First, a short review of relevant information for this paper about Basel III is provided. Second, literature about repurchase agreements will be discussed. This could be divided into three parts: (i) literature about repos and the functioning of the repo market, (ii) research about the role of repos during the financial crisis and (iii) a description of the relation of repos to the probability of a bank run by unsecured debtholders. Finally, the hypothesis will be formulated based on the literature provided.
7
2.1. Basel III
The banking industry is for sure one of the most regulated industries in the world (Santos, 2001). The capital requirements are one of the main components of bank regulation. The strict regulation is justified by the central role banks take in financial intermediation and the importance of capital for the stability of banks. The amount of capital banks have to hold has an influence on the competiveness of banks. This is one of the reasons for a system of harmonized capital requirements (Santos, 2001).
The Basel Committee on Banking Supervision (BCBS) was founded at the end of 1974 by the central bank Governors of the Group of Ten Countries. They are an autonomous committee providing a forum for international cooperation about the regulation of banks (Goodhart, 2011). The Basel Accord developed by the BCBS in 1988 could be seen as the onset of the international regulation of capital standards (Santos, 2001). Even though the Accord only addressed credit risk, the statement that the level of capital should depend on the riskiness of an assets, is still endorsed by the policy makers of today (Hannoun, 2010). From 1996 not only credit risk but also market risk became subject to the capital requirements. Later, with the introduction of Basel II in 2004, operational risk became part of regulation. (Hannoun, 2010).
The financial crisis of 2007-2008 exposed the deficiencies of the regulation of banks. To overcome these weaknesses, Basel III was developed. The aim is to strengthen regulation by improving banks to deal with shocks caused by financial and economic stress and improving the supervision on banks. Also improvement of risk management, governance and transparency of banks belong to the goals of the new regulation (BIS, 2016).
The Group of Governors and Heads of Supervision from the Basel Committee on Banking and Supervision assembled on the 26th of July 2010, where they reached an agreement about the overall design of a reform package of existing regulation to increase the long-term stability of the banking sector. According to Nout Wellink, the head of the Dutch Central Bank at the time, the transparency of the design of the regulations reduced market uncertainty (BIS, 2010c). This meeting and the subsequent press release could be considered as the first announcement date of Basel III in this research.
At a subsequent meeting, they affirmed the agreements reached on the 26th of July 2010. They also announced a substantive reinforcement of the capital requirements already in place. The minimum common equity requirement will be increased from 2% to 4.5% and a conservation buffer of 2.5% will be required, bringing the total common equity requirement to 7%. Jean Claude Trichet has called it a “fundamental strengthening of capital standards”
8 (2010b). On the 16th of December 2010, the official Basel III rules were issued. They
comprised the details of the global standards for the regulation of bank capital and liquidity. The higher capital requirements and liquidity framework were proposed to decrease the probability of future financial stress (BIS 2010a). On July 2, 2013, the Federal Reserve Board approved the final rule for higher capital requirements for U.S. banks (Federal Reserve Board, 2013). This was followed by the release of the “New Capital Rule: Community Bank Guide” (Federal Reserve Board, 2013b).
2.2. Repos and the repo markets
This paper investigates the impact of the capital requirements of Basel III on bank performance, based on their exposure to the market of repurchase agreements. A repurchases agreement, hereafter referred to as repo, is “the simultaneous sale and forward agreement to repurchase the same, or a similar, security at some point in the future” (Krishnamurthy et al., 2014). Thus, it is a collateralized loan whereby, until the lender is repaid, the lender receives securities as a collateral. The borrower pays the lender interest, called the repo rate. For example, the investor buys an asset from a bank for $100, whereby the bank agrees to repurchase this asset at some point in the future for $105. In this example, the repo rate is 5%, since $105 minus $100, divided by $100 is 5%. It is similar to the interest rate on a bank deposit. In the majority of cases, the market value of the securities exceeds the value of the cash the investor lent, the difference is known as the haircut. Following the example above, whereby the bank sells an asset for $100, we assume the market value is $120. Therefore, the haircut is approximately 17%, because $120 minus $100, divided by $120 is 17% (Gorton and Metrick, 2012). The collateral provider repurchases the security for $100 plus the interest rate, known as the repo rate.
If the bank does not repurchase the loan, the investor has the right to keep the collateral. But there is a risk for the investor that if the bank defaults, the collateral is worth less than the money he lent to the bank. The investor can protect himself from this collateral risk by a higher haircut. (Krishnamurthy et al., 2014). Therefore both the haircut and the repo rate can be used as a measure for the cost of funding.
At least one of the counterparties in the repo market in the U.S. is a securities dealer. The repo market is, inter alia, used to finance their inventories of securities. Initially, all transactions in the repo market were bilateral (Copeland et al., 2012). Bilateral repos are between dealer banks or between dealer banks and hedge funds (Krishnamurthy et al., 2014). However, there are some complications in this market that demands operational expertise and
9 systems for particularly large investors who do a significant amount of repo lending. It requires them to (i) keep track of the collateral they received, (ii) check if the collateral is adequate and correctly valued and (iii) verify that the right margin was used (Copeland et al., 2012).
There is also a market where repo settlement is facilitated by a clearing bank. Such a market is called the tri-party repo market. In addition to the aforementioned settlement, these banks also offer collateral management services: they revaluate assets and remargin the collateral, but they also protect the cash lender for the default of the dealer. If the dealer defaults, the lender gets the collateral. The time to maturity for these repos is typically one day, but many of them are “rolled over” for a number of days (Copeland et al. 2012). Hence, the major difference between the tri-party and bilateral repo market is the safety of the collateral. With tri-party repos, a custodian bank safeguards the collateral on behalf of the cash lender, while bilateral repos are without such a custodian bank (Krishnamurthy et al., 2014). Before 2010, over half of the tri-party market consisted of repos between dealer banks and money market funds (MMFs) or securities lenders (SLs) (Copeland et al., 2010). The repos could serve as a secured alternative of bank deposits for investors. Since these investors care about the type of collateral, but not about the specific security, the repo market is a general collateral market. For dealers, repos are mostly used to lend a substantial amount of short-term financing for their securities inventories at a low cost. The tri-party repo market, where money is provided by MMFs or SLs, is the largest in secured funding for U.S. dealers. (Copeland et al. 2012).
2.3. Repos during the financial crisis
Earlier papers have suggested that the financial crisis was a system-wide bank run; a run driven by the withdrawal of repos. During the crisis, many doubts were raised about the collateral of repos and the riskiness of banks. At some point in 2007, this led to the first run on repos, whereby cash providers were not willing to provide short-term financing. Therefore, the crisis has been described by Gorton and Metrick (2012) as a run on the repo market.
There is much literature that confirms the theory of a sharp contraction of repos during the financial crisis and thus, repos can be considered a very unstable way of funding during an economic crisis (Martin et al., 2014). The sharp decrease in repos was unexpected since they are collateralized by securities (Bernanke, 2009). However, although there was a sharp decrease in the repo market, it was not as dramatic as Gorton and Metrick (2012) have suggested. While the contraction in the repo market was not that big, it had a significant effect
10 on a few important dealer banks (Krishnamurthy et al., 2014). The sharp decrease in funding from the repo market triggered the default of Bear Stearns and Lehman Brothers (Martin et al., 2014).
2.4. The influence of repos on bank runs
Repos are a safe and cheap source of funding for banks. However, repos have a detrimental effect for unsecured debt holders, which makes their position riskier because part of the collateral is pledged by these repos. Consequently, there remains less collateral for the unsecured debtholders. This increases the riskiness for the unsecured debt holders and therefore, traditional theory has stated that the higher the level of repos, the higher the probability that unsecured lenders will run.
Matta and Perotti (2016), however, have suggested that this positive relation might not hold for all levels of repos. Instead, they have suggested a concave relation. The assumption is that a mandatory stay is triggered for the remaining lenders as soon as the bank has no more liquid assets to meet the withdrawals. Therefore, the remaining illiquid assets are shared equally with the lenders who did not run. The reasoning is that if a substantial amount of the collateral is pledged to repos, there is less capital for unsecured debtholders. This leads to the theory that if all unsecured debtholders run, the chance the debtholders will be paid will be minimum. By rolling over instead of running, they improve their chances of receiving the fixed value if the bank does not default. Therefore, at some point, unsecured debtholders might choose to roll over. This leads to the theory of a concave relation between the probability of a bank run and the level of repos, presented in Figure 1
The vertical axis in Figure 1 represents the probability of a bank run by unsecured debt holders, called p(R). The amount of collateral pledged by repos, called s, is on the horizontal axis. The diagonal line represents the traditional theory, where the higher the amount of s (collateral pledged by repos), the higher the chance of a bank run. The theory of Matta and Perotti (2016) is represented by the concave line, since they have suggested that at some point, the chance of a bank run starts declining. The point max(s) is the maximum amount of capital available for repos and is determined by the capital requirement.
11 Figure 1: In this figure the relationship between the probability of a bank run by unsecured debtholders and the amount of
collateral pledged by repos is shown. The diagonal line represents the traditional theory, the curved line represents the concave relation suggested by Matta and Perotti (2016).
2.5. Relation of literature to the hypothesis
The topic of this research is to investigate whether an increase in capital requirements has a different impact on banks with a lot of exposure to the repo market (the safer banks) compared to banks that are less exposed to this market (the riskier banks). The paper of Matta and Perotti (2016) assumes banks to be either in point one or point two of Figure 2. Banks in point one are supposed to be riskier while the banks in point two are assumed to be safer based on their decision about the exposure to the repo market.
Since Basel III requires banks to withhold more capital, the amount of collateral pledged by repos must decrease to meet the capital requirements. Hence, the banks in point two will shift along the concave line to point three. Figure 2 shows that the probability of a bank run for these banks will increase. This means that the credit default swap (CDS) of the safer banks will increase more than the CDS of the riskier banks after the introduction of the capital requirements of Basel III. Therefore the hypothesis is:
H0: The credit default swaps of banks more exposed to the repo funding increases more after the introduction of Basel III than the CDS for banks with less exposure to the repo market.
12 Figure 2: In this figure the effect of a decrease in the amount of collateral pledged by repos is shown. Before an increase in
the capital requirements, risky banks are considered to be in point 1 and safe banks in point 2. After an increase in the capital requirements, banks in point 2 should hold more capital will move to point 3.
3. Methodology
The purpose of this study is to test the influence of the Basel III capital requirements on the performance of banks in the US. This study focuses solely on U.S. banks, because banks in other countries may be subject to more or different regulations than U.S. banks. Moreover, continent- or country-specific effects may play a role. For example, the Euro crisis impacted the performance of banks in Europe differently than banks in the rest of the world. Because of the availability of data, the sample consists of 12 listed U.S. banks.
The influence of the capital requirements is examined by constructing a difference-in-difference regression. To test the influence of the capital requirements, the banks were separated into a control group and a treatment group. The treatment group is the group of banks that is more exposed to repos, while the control group is the group of banks with less exposure. As previously explained, banks are deciding on the level of repos by balancing the chance of default and the advantage of low funding costs. Therefore, a riskier bank should choose point one in Figure 2, because otherwise the probability of default will become too high. Conversely, a safer bank (a bank with a low probability of default) is more likely to choose more repos and could end up in point 2 in Figure 2. Therefore, additional capital requirements should affect the number of repos of the safe banks – with more exposure to the repo market – more than for the riskier banks. Based on this assumption a natural separation between the treatment and control group is generated.
13
3.1. Determination of treatment and control group
To divide the banks into a treatment group and a control group, a regression was conducted to test the banks’ exposure to the cost of repo funding. The assumption is that a bank that relies more on repo funding should be more affected by an increase in the repo rate than a bank that relies less on repo funding. To test to what extent a bank is exposed to the repo market, and to create a treatment and control group, the following regression was conducted:
𝐶𝐷𝑆𝑖𝑡 = 𝛼 + 𝛽 𝑟𝑒𝑝𝑜 𝑟𝑎𝑡𝑒𝑡+ 𝑋𝑖𝑡 ,
𝛾 + 𝛿𝑖+ 𝛿𝑡+ 𝜀𝑖,𝑡 (1)
where the dependent variable 𝐶𝐷𝑆𝑖𝑡 is the credit default swap of bank i at time t. The repo rate is the overnight rate at time t. 𝑋𝑖𝑡 is a vector of control variables that control for both bank-specific as well as market-wide factors that, according to Galil et al. (2014), have a significant impact on credit default swaps. The constant is expressed by 𝛼, 𝛿𝑖 denotes bank fixed effects and 𝛿𝑡 denotes time fixed effects, while 𝜀𝑖,𝑡 is the disturbance term.
If the repo rate increases, it will be more costly for banks to finance their activities through repo. This should have a greater effect on the probability of default for banks that are more exposed to repos, than for the ones that are less. Stated differently, banks with a high level of repos should be more sensitive to repo rates and therefore, have a higher coefficient for β. Thus, banks with a higher β belong to the treatment group, while firms with a lower β belong to the control group. The two groups are separated by taking the median of all coefficients of the repo rate. Banks with a higher coefficient than the median are assigned to the treatment group. Banks with a lower coefficient than the median are assigned to the control group.
For each of the announcement or enactment dates, observations one year before the specific date are taken into account. This means that for every announcement and enactment date, a new determination of the treatment and control group is investigated. This is done to ensure accuracy, since the banks that are exposed more to the repo market in 2010 do not have to be the same banks with a high exposure in 2013. It is also incorrect to use the whole sample size from 2009 to 2013 to determine to which group the banks belong, since data after the announcement is in that case included in the analysis for the purpose of determining the groups. This will obfuscate the results, because the groups should be defined based on data before the “event” – in this case the announcement or enactment of Basel III.
14
3.2. Influence of Basel III
After the separation of the banks between a treatment and control group, a second regression is conducted to investigate the impact of the Basel III regulation on the performance of banks. To test if the credit default swaps of banks more exposed to the repo market (the safer banks), increases by more than the credit default swaps of the banks less exposed (the riskier banks), the following regression was conducted:
𝐶𝐷𝑆𝑖,𝑡 = 𝛼 + 𝛽 𝐵𝑎𝑠𝑒𝑙 𝐼𝐼𝐼 + 𝜇 𝐵𝑎𝑠𝑒𝑙 𝐼𝐼𝐼 × 𝑇𝑟𝑒𝑎𝑡𝑒𝑑 + +𝑋𝑖𝑡 ,
𝛾 + 𝛿𝑖 + 𝛿𝑡+ 𝜀𝑖,𝑡 (2)
The dependent variable 𝐶𝐷𝑆𝑖,𝑡 is used as proxy for the riskiness of bank i at time t. The dummy variable Basel III is a dummy variable, where observations before the introduction of Basel III are zero, and variables after the introduction of Basel III are equal to one. The dummy variable treated is one if the bank belongs to the treatment group and zero if the bank belongs to the control group. The main variable of interest is the interaction term1, Basel III ×
Treated. Because requiring more capital should only affect banks that significantly rely on
repo and not those who do not, a positive coefficient for 𝜇 is expected. Conversely, if there is no evidence that these banks react differently, 𝜇 should be zero. Moreover, if banks in the treatment group react less to the capital requirements of Basel III, 𝜇 is negative.
For this analysis, several dates for the introduction of Basel III were used. The basic analysis uses the announcement date of December 16, 2010 and July 9, 2013. Robustness checks are also conducted for the other announcement and enactment dates, as mentioned in section 2.1. As previously explained, a new analysis will be conducted for all these different dates to determine which banks belong to the treatment group and which banks belong to the control group. After the control and treatment group were defined, the main regression, i.e. formula (2), is conducted for these different dates.
4. Data and descriptive statistics
Panel data was used to answer the research question. Data was downloaded for the period between August 20, 2009 and January 1, 2017.2 This is the longest time horizon possible given the availability of the data at the moment this research was conducted. The sample
1 “Interaction term” and “Treated Banks × Basel III” are used interchangeably throughout this text.
2 CDS data for all 12 banks is only available from August 20, 2009 onwards. Therefore, this date was chosen as the starting
15 consisted of 12 U.S. banks.3 The banks involved in this research were selected based on the availability of CDS data in the Thomson Reuters CDS database.
The variables and the data sources are described below. For the independent variable as well as the control variables, the expected signs of the coefficients are explained. A brief overview of the variables and the direction of their effect on credit default swaps are summarized in Table 6 in the Appendix.
4.1. CDS
The CDS data is used to measure the banks’ probability of default and is the dependent variable in this study. The data was retrieved from the Thomson Reuters CDS database, which was accessed by Datastream. The market selected is the United States and the currency is USD. Only 5-year Senior XR CDS rates were used in this research. These 5-year contracts are most often used in research about credit default swaps and are considered the most liquid on the market (Meng and Ap Gwilym, 2008).
Since the XR CDS rates for Barclays and U.S. Bancorp were not available from this
source, these banks were not considered within this research. For the majority of the banks, data was available from September 1, 2009 to January 1, 2017. Data for BB&T, is only available from mid-2011 and therefore this bank is not included in this paper.
4.2. Repo rate
The overnight repo rates are retrieved from Thomson Reuters via Datastream. This rate is used to determine which banks are more and which banks are less exposed to the repo market. A positive correlation is expected between CDS and the repo rate, because both are likely to increase when there is more uncertainty in the market.
4.3. Control variables – bank specific
Based on the Merton model – named after economist Robert Merton – a bank’s leverage was introduced as a control variable in the analyses. The Merton model suggests a negative relation between stock return and the probability of default (Merton, 1974). In accordance with Christie (1982), Collin-Dufresne et al. (2001) and Alexander and Kaeck (2008), bank stock returns are used as a proxy for financial leverage. If the stock returns are positive, the leverage measured in market value decreases, which would lead to lower credit default swaps.
3 Banks included in the sample are: American Express, Bank of New York Mellon, Bank of America, Capital One, Charles
16 Therefore, a negative correlation is expected between stock returns and CDS spreads (Anneart et al., 2013).
However, it is important to note that this variable could also capture factors other than financial leverage. For instance, the stock returns might incorporate future expectations of a bank’s profits and positive returns could be a sign of lower default risk. However, for this analysis, the driving force behind the proxy is not of interest. This variable was only included in the sample to control for bank-specific effects that influence CDS and, if not included, could cause an omitted-variable bias.
Daily stock returns were calculated based on the banks’ stock price, and adjusted for stock splits.4 These prices are retrieved from The Center for Research in Security Prices (CRSP) and were used to manually calculate the daily stock returns. As previously stated, the relation between CDS spreads and stock returns is expected to be negative.
4.4. Control variables – market wide factors
After controlling for firm-specific variables, the majority of studies have found that the residuals in their models have, to some extent, a common variation, suggesting that the models could be improved by controlling for common factors (Collin-Dufresne et al., 2001). Therefore, some market-wide factors are included in the models of this paper as control variables. Galil et al. (2014) have determined that these market-wide factors as the change in the spot rate, the change in the slope of the term structure and the change in the VIX, can be used to improve the fit of a model about credit default swaps.
Spot rate
The spot rate is used as a reinvestment rate (Galil et al., 2014). It is determined as the 5-year Treasury Constant Maturity Rate and was retrieved from the St. Louis Federal Reserve with a daily frequency. The 5-year rate is consistent with the 5-year CDS spread.5 A negative relationship between the spot rate and credit default swaps is expected, since a higher reinvestment rate increases future value (Longstaff and Schwartz, 1995) and reduces the probability of default (Collin-Dufresne et al., 2001). This negative relationship has been confirmed by the research of Longstaff and Schwartz (1995) and that of Galil et al. (2014).
4 For example, a stock split on May 9, 2011 for Citigroup.
17
The slope of the term structure
The slope of the term structure is included as a market-wide factor because it is a sound proxy of the business cycle (Estrella and Mishkin, 1997) and influences the number of projects with a positive net present value (Galil et al., 2014). The term structure slope was calculated as the difference between the 10-year Treasury Constant Maturity Rate and the 2-year Treasury Constant Maturity Rate, which were both derived from the St. Louis Federal Reserve website.6 On the one hand, Fama and Fench (1989) have stated that an increasing yield-curve slope indicates more economic growth and therefore, a negative relationship should be expected between the CDS spread and the slope of the term structure. On the other hand, Galil et al. (2014) are aware of an opposite effect: a steeper slope reduces the projects with a positive net present value and therefore, leads to an increase in the CDS spread. According to their theory, there is a positive relationship between the CDS spread and the slope of the term structure. In their empirical research, however, they found a negative relationship, which is more in accordance with the research of Fama and Fench (1989). Therefore, a negative sign is expected for this research as well.
VIX
To control for market volatility, the VIX index from the Chicago Board Options Exchange (CBOE) is included in the regressions. This is in accordance with the research conducted by Collin-Dufresne et al. (2001) and Galil et al. (2014). The VIX contains the option-implied volatilities of a range of S&P 500 index options. It is commonly used to measure market risk. More volatility means more risk for investors, which increases the CDS spread. Therefore, a positive relationship between market volatility and the CDS is expected (Galil et al., 2014).
4.5. Descriptive statistics
Table 1 presents the summary statistics for both the dependent and the independent variables. The first rows presents the number of observations, mean, median, standard deviation and the minimum and maximum observation for the credit default swaps, which is the dependent variable in this research. The second part of the table displays the same characteristics for the used dependent variables.
18 The CDS (Total) represents the observations of all 12 banks included in the sample. In the second row of the table, the credit default swaps are spitted by bank. During the sample period (August 20, 2009 to December 31, 2016) the mean credit default swap was 99.36 basis points with a minimum of 25.85 and a maximum of 609.81 basis points. The highest credit default swap was for Morgan Stanley on October 3, 2011. Morgan Stanley’s exposure to Europe’s sovereign debt worried investors, which lead to a sharp decrease in the company’s share price and a peak in the credit default swap. The credit default swaps for the whole sample period are presented in Figure 3 in the Appendix. This figure also displays Morgan Stanley’s peak on October 3, 2011. The period of time between mid-2011 and mid-2012 experienced the largest credit default swaps of the entire sample period.
The repo rate was, on average, 0.19 with a standard deviation of 0.16. A negative repo rate was even observed. The change in the spot rate, slope of the term structure and the market
Table 1 – Summary Statistics
Observations Mean Median Std. Dev. Min Max
Independent Variables CDS (Total) 22,260 99.36 86.26 61.00 25.85 609.81 CDS American Express 1,855 68.58 60.51 28.87 34.49 200.40 CDS Bank of NY Mellon 1,855 89.96 103.18 22.77 35.72 104.01 CDS Bank of America 1,855 135.86 110.93 81.37 58.25 487.54 CDS Capital One 1,855 92.07 93.11 31.42 43.38 184.21 CDS Charles Schwab 1,855 49.97 40.30 11.31 32.57 70.17 CDS Citigroup 1,855 132.27 109.47 60.96 58.54 358.30 CDS Goldman Sachs 1,855 137.14 113.44 66.66 63.81 423.52 CDS HSBC 1,855 86.25 54.01 64.29 25.85 298.27 CDS JPMorgan 1,855 83.04 76.57 24.95 44.17 187.13 CDS Morgan Stanley 1,855 159.32 134.30 96.73 59.30 609.81 CDS PNC Financial 1,855 82.06 73.65 16.26 59.52 136.52 CDS Wells Fargo 1,855 75.75 69.99 28.07 35.39 181.85 Dependent variables Repo Rate 1,855 0.19 0.16 0.14 -0.01 1.34 Δ Spot Rate 1,855 0.00 0.00 0.05 -0.20 0.21 Δ Slope 1,855 0.00 0.00 0.04 -0.17 0.17 Δ VIX 1,855 -0.01 -0.10 1.65 -12.94 16.00 Stock return 22,260 0.05 0.04 1.85 -20.32 16.74
Notes: This table provides the summary statics for the data used in this research. The observations range from 20/08/2009 to
31/12/2016. The sample consists of 12 banks from the United States. All credit default swaps are presented in basis points. The first row, CDS (total) consists of all observations for the 12 banks, below the observations are divided to the different banks. The
repo rate is the overnight repo rate retrieved from Thomson Reuters. Δ Spot Rate is the change in the 5-year Treasury Constant
Maturity Rate derived from the FRED. Δ Slope is the change in the difference between the 10- and 2-year Treasury Constant maturity rate, derived from the FRED. Δ VIX is the change in the CBOE Volatility Index, derived from The Chicago Board Options Exchange (CBOE). The stock return is the daily return for each bank stock, derived from CRSP.
19 volatility index was, on average, zero. Furthermore, the VIX experienced a much higher standard deviation than the other two control variables, indicating a higher variance in the observations for the VIX index. The dependent variables discussed above are market-wide, while the stock return is firm-specific. The stock return measures the change in the stock price of each bank. Therefore, the observations are 12 times higher than for the other dependent variables. The stock return was 0.05% during this period, while the median was 0.04. Moreover, the standard deviation was high, meaning that there is a significant amount of variation in this variable. The largest decrease in the daily stock return was more than 20%, while the largest increase in the daily return was almost 17%.
5. Results
This chapter describes the results. First, the treatment and control group are defined according to their sensitivity to the cost of repo funding. Thereafter, the results regarding the impact of Basel III on bank risk are discussed. For this analysis, December 16, 2010 was used as the announcement date and July 9, 2013 was used as the enactment date.7
5.1. Classification of the treatment and control group – announcement date
The treatment and control group were classified according to formula (1). The sensitivity of a bank’s CDS to the repo rate was determined using this regression. Banks with a high sensitivity belong to the treatment group, whereas banks with a low sensitivity belong to the control group.
The results are presented in Table 2, where the banks are ordered according to their sensitivity. The ranking of the banks’ coefficients is the same for all three regression specifications. Column 1 displays the results for the regression with bank and time fixed effects but without control variables. Column 2 displays the results for the regression with control variables and bank fixed effects but without time fixed effects. Column 3 displays the results for the regression in which control variables, time and bank fixed effects are all included.
To assign the banks to two different groups, the sensitivity of the median bank to the repo rate was used. The median is between Charles Schwab and JP Morgan. According to his method, Citigroup, Bank of New York Mellon, PNC, Capital One, American Express,
20
Table 2 – Classification of the treatment and control group (16/12/2010)
(1) (2) (3)
Bank × repo rate
Citigroup -2.494*** -0.890*** -2.393*** 0.538 0.003 0.491 Bank of NY Mellon -2.240*** -0.637*** -2.138*** 0.538 0.003 0.491 PNC -1.613** -0.014** -1.501** 0.538 0.006 0.490 Capital One -1.298** 0.305*** -1.194** 0.538 0.004 0.491 American Express -1.091* 0.516*** -0.997* 0.538 0.003 0.492 Wells Fargo -1.012* 0.589*** -0.904* 0.538 0.005 0.490 Charles Schwab -0.353 1.257*** -0.266 0.538 0.004 0.493 JP Morgan 0.486 2.087*** 0.593 0.538 0.004 0.491 Goldman Sachs 1.183** 2.788*** 1.280** 0.538 0.003 0.492 Bank of America 1.484** 3.077*** 1.604*** 0.538 0.009 0.489 HSBC 1.690*** 3.299*** 1.781*** 0.538 0.004 0.493 Morgan Stanley 2.490*** 4.093*** 2.592*** 0.538 0.003 0.491 Stock return -0.003 0.006* 0.002 0.003 Δ Spot Rate -0.065 0.282** 0.076 0.122 Δ Slope -0160* -0.407** 0.089 0.183 Δ VIX -0.002 0.002 0.002 0.006
Time fixed effects Yes No Yes Bank fixed effects Yes Yes Yes
Observations 3036 3036 3036
R2 0.553 0.228 0.554
Adjusted R2 0.511 0.224 0.511
Notes: This table is used to assign the banks to either the treatment or the control group. The banks are ordered
based on their coefficients (from smallest to largest). Column 1 includes all time and bank fixed effects but no control variables. Column 2 includes all control variables and bank fixed effects but no time fixed effects. Column 3 shows the results of the most accurate model that includes control variables, time and bank fixed effects. The 5-year credit default swap is the dependent variable. The repo rate is the overnight repo rate. The stock return is the daily return for each bank stock. Δ Spot Rate is the change in the 5-year Treasury Constant Maturity Rate. Δ Slope is the change in the difference between the 10- and 2-year Treasury Constant maturity rate. Δ VIX is the change in the CBOE Volatility Index. Robust standard errors are reported in italics. Constants were included in the regressions but are not reported. *, **, and *** indicate significance at 10%, 5%, and 1% respectively.
21 Wells Fargo and Charles Schwab belong to the control group, since their coefficients are below the median. The banks belonging to the treatment group are JP Morgan, Goldman Sachs, Bank of America, HSBC and Morgan Stanley.
Four banks exhibit a significant, positive effect of the repo rate on credit default swaps. For example, for Morgan Stanley, an increase in the overnight repo rate by 1 basis point leads to an increase in the CDS rate by 2.5 basis points. Except the change in the market volatility index (VIX), all other variables suggested by previous literature have a significant effect on credit default swaps. Moreover, 55% of the variation in the credit default swaps could be explained by this model, regardless of whether the control variables are included. If the model is not controlled for time-fixed effects, the explanatory power decreases to 23%. It should be noted that the results do not entirely meet the expectations. According to the theory of Matta and Perotti (2016) banks are either in point one of Figure 2 or in point two. Banks in point two experience a negative impact for an increase in the repo rate, and therefore the credit default swap (as a measure of performance and probability of default) increases. Conversely, banks that are in point one of Figure 2 are not, or otherwise marginally, exposed to the repo market. Therefore, these banks in point one are not significantly impacted by the repo rate. Hence, the coefficients were expected to be either positive or insignificant. The banks with insignificant or very low coefficients belonged to the control group and the banks with significantly, positive, coefficients belonged to the treatment group.
Therefore, the results should be interpreted with caution and this method for distinguishing the two groups should be considered critically. Although it is not perfect, the results can be interpreted under the assumption that banks can be exposed to two types of markets: the repo market and another market, called market X, and a negative correlation is assumed between those two markets. This means that when the repo rate increases, the credit default swap for banks that are exposed to the repo market increases, while the credit default swap for the banks exposed to market X decreases. Under this assumption it were possible to determine the groups in the treatment and control group as described above.
As a robustness check the banks with insignificant coefficients (Charles Schwab and JP Morgan) were excluded from the sample. A regression in which these two banks are included in the control group is also conducted as a robustness check. Because if the two assumptions are combined (banks in the control group do not significantly react on changes in the repo rate or are negatively correlated to the ones with a large exposure) it is justified to include Charles Schwab and JP Morgan in the control group. Both strategies are applied in the robustness section, paragraph 1 of chapter 6.
22
5.2. Impact of the announcement of Basel III on banks exposed to repos
Based on the determination of the treatment and control group by the previous subsection, the main regression was conducted. The results for this regression are presented in Table 3. The variable of interest is Treated Banks × Basel III, which is a dummy variable which is one for the banks that are part of the treatment group, times a dummy variable which is one if the observation is after the announcement of Basel III. Although a positive relationship was expected, this could not be concluded from the evidence presented in Table 3. The interaction term is insignificant for all different time windows, meaning that, after the announcement date of Basel III, there is no significant difference between the impact on the CDS of banks that are sensitive to the repo rate compared to banks that are less sensitive. The hypothesis that there is a larger increase in banks’ CDS, because of the capital requirements of Basel III, for those that are more exposed to repo funding than those that are less exposed is rejected.
Table 3 – Impact of Basel III on the riskiness of banks (16/12/2010)
+/- 13 weeks +/- 8 weeks +/- 4 weeks +/- 2 weeks +/- 1 week Basel III -8.688 -7.845* 11.451*** 0.909 1.205 8.877 4.040 2.711 1.237 0.943 Treated Banks × Basel III -8.545 -3.561 -2.574 -0.596 0.007 6.228 3.304 2.774 1.973 1.516 Stock return 0.001 0.002 0.001 0.000 0.000 0.003 0.002 0.002 0.001 0.002 Δ Spot Rate -0.130*** -1.036** 0.730** 0.226 -0.060 0.040 0.412 0.323 0.229 0.110 Δ Slope -1.585*** 2.758** 0.392 -0.413 0.099 0.369 1.026 0.344 0.412 0.232 Δ VIX -0.013*** -0.079* 0.117*** -0.039 0.000 0.004 0.040 0.026 0.040 0.001
Time fixed effects Yes Yes Yes Yes Yes Bank fixed effects Yes Yes Yes Yes Yes Observations 1524 936 468 240 132 R2 0.452 0.636 0.650 0.553 0.171
Adjusted R2 0.401 0.603 0.618 0.510 0.088
Notes: This table looks at the impact of Basel III on the credit default swaps of banks, dependent on their exposure to the repo
market. The variable of interest is the interaction term: Treated Banks × Basel III. The variable Treated Banks is one if the bank belongs to the treatment group (if the bank is exposed more to the repo market than the median bank) and zero if the bank belongs to the control group. The variable Basel III is one if the observation is after the announcement date, and zero otherwise. The different columns provide the results for different event windows. Column 1 shows the results for a +/- 13 weeks, meaning that observations of three months before and 3 months after the announcement of Basel III are included. The 5-year credit default swap is the dependent variable. The stock return is the daily return for each bank stock. Δ Spot Rate is the change in the 5-year Treasury Constant Maturity Rate. Δ Slope is the change in the difference between the 10- and 2-year Treasury Constant maturity rate. Δ VIX is the change in the CBOE Volatility Index. Robust standard errors are reported in italics. Constants were included in the regressions but are not reported. *, **, and *** indicate significance at 10%, 5%, and 1% respectively.
23 Moreover, for all the different event windows – except for 4 weeks – the time variable is insignificant. This suggests that the announcement of Basel III did not impact the banks’ credit default swaps, except for during the 4-week window, when the credit default swap increased by 11.5 basis points. While the variable that controls for the daily stock return does not appear to have a significant impact on the CDS, the change in the spot rate, the change in the slope of the term structure and the change in the market volatility index all appear to have an effect on the banks’ credit default swaps if certain event windows are taken into account. Surprisingly, the coefficients switch signs when different event windows are considered. Finally, the variance explained by the model ranges from between 17% to 65%, whereby the lowest explanatory power is for the shortest even windows; this could be due to the limited observations.
5.3. Classification of the treatment and control group – enactment date
To ensure accuracy, a new regression was conducted to determine the treatment and control group for the enactment of Basel III. To define the two groups, formula (1) was again used. This regression examines the sensitivity of a banks’ CDS to the repo rate. As previously stated, the banks with a high coefficient belong to the treatment group, while banks with a low coefficient belong to the control group.
The results for this regression are presented in Table 4. The banks are ordered based on their coefficients, from smallest to largest. Banks with a coefficient for the sensitivity to the repo rate below the median include Charles Schwab, The Bank of New York Mellon, Capital One, Wells Fargo, America Express and JP Morgan. Hence, these banks belong to the control group. The banks with a coefficient above the median are PNC Financial, The Bank of America, Goldman Sachs, Citigroup, HSBC and Morgan Stanley, and therefore belong to the treatment group.
Charles Schwab and JP Morgan, which belonged in the analysis for the announcement date to the control group, now belong to the treatment group. Conversely, PNC and Citigroup now belong to the treatment group instead of the control group. Moreover, the ranking in the groups differs. Therefore, it can be concluded that the banks in the treatment group are not the same for the analysis of the enactment date as for the announcement date (described in section 5.1) Hence, it is good to not just take the separation, determined in section 5.1 for the announcement date, as basis for the analysis of the effect of the enactment date.
24
Table 4 – Classification of the treatment and control group (09/07/2013)
(1) (2) (3)
Bank × repo rate
Charles Schwab -0.487 -0.237*** -0.790** 0.413 0.005 0.359 Bank of NY Mellon -0.278 -0.022*** -0.585 0.413 0.003 0.358 Capital One 0.164 0.417*** -0.141 0.413 0.003 0.359 Wells Fargo 0.340 0.591*** 0.036 0.413 0.004 0.359 American Express 0.870* 1.122*** 0.566 0.413 0.004 0.359 JP Morgan 0.878* 1.131*** 0.572 0.413 0.003 0.359 PNC 0.888* 1.138*** 0.586 0.413 0.005 0.359 Bank of America 2.182*** 2.443*** 1.870*** 0.413 0.005 0.357 Goldman Sachs 2.226*** 2.482*** 1.919*** 0.413 0.003 0.358 Citigroup 2.401*** 2.661*** 2.090*** 0.413 0.005 0.357 HSBC 3.226*** 3.484*** 2.916*** 0.413 0.004 0.357 Morgan Stanley 3.946*** 4.202*** 3.639*** 0.413 0.003 0.358 Stock return -0.003 0.002 0.002 0.002 Δ Spot Rate -0.305*** 0.130 0.096 0.160 Δ Slope -0.087* -0.531 0.046 0.351 Δ VIX -0.003 -0.032 0.002 0.022
Time fixed effects Yes No Yes
Bank fixed effects Yes Yes Yes
Observations 3012 3012 3012
R2 0.589 0.205 0.589
Adjusted R2 0.550 0.201 0.550
Notes: This table is used to assign the banks to either the treatment or the control group. The banks are ordered based on their
coefficients (from smallest to largest). Column 1 includes all time and bank fixed effects but no control variables. Column 2 includes all control variables and bank fixed effects but no time fixed effects. Column 3 shows the results of the most accurate model that includes control variables, time and bank fixed effects. The 5-year credit default swap is the dependent variable. The
repo rate is the overnight repo rate. The stock return is the daily return for each bank stock. Δ Spot Rate is the change in the 5-year
Treasury Constant Maturity Rate. Δ Slope is the change in the difference between the 10- and 2-year Treasury Constant maturity rate. Δ VIX is the change in the CBOE Volatility Index. Robust standard errors are reported in italics. Constants were included in the regressions but are not reported. *, **, and *** indicate significance at 10%, 5%, and 1% respectively.
25 It should again be noted that not all coefficients are significant if the model is controlled for time-fixed effects – as in column one and three. Only Charles Schwab’s CDS rate decreases if the repo rate increases, while the CDS increases in the same situation for Bank of America, Goldman Sachs, Citigroup, HSBC and Morgan Stanley. Other banks’ credit default swaps are not significantly impacted. This makes it necessary to conduct a robustness check, in which only the banks with a significant positive relationship between the repo rate and credit default swaps are in the treatment group, while the other banks belong to the control group.8
The results presented in Table 4 suggest that all control variables included in the model do not have a significant impact on credit default swaps if time-fixed effects are included in the model. A reason for this could be that the influence of these variables is captured by the time-fixed effects. However, another explanation could simply be that, for these banks and given this time interval, no significant effect exists. The explanatory power of this model is similar to the regression for the time period before the announcement date. In Table 4, the R2 is 59% while it was 55% for the other time period (Table 2). Without time-fixed effects in the model, the variance in the credit default swaps of the banks falls to only 21%, which was 23% for the analysis of the announcement date.
5.4. Impact of the enactment of Basel III on banks exposed to repos
Based on the classification of the treatment and control group of section 5.3, the main regression was conducted. The results are presented in Table 5. The variable of interest is
Treated Banks × Basel III. This term consists of a dummy variable that defines whether the
bank belongs to the treatment or the control group, times a dummy variable that explains whether the observation came after the enactment of Basel III in the United States.
As explained, a positive relationship is expected between Treated Banks × Basel III and the banks’ credit default swaps. However, the coefficient for all event windows is insignificant, suggesting that there is no difference between the effect of Basel III on banks that are sensitive to the repo market compared to those that are less exposed to the repo market. Based on these results, the hypothesis that banks that are more exposed to repo funding will experience a larger increase in their credit default swaps because of the capital requirements introduced by Basel III, is rejected.
The variable Basel III, which is one if the observations belong to the period after the enactment of Basel III and zero before this date, has a significant, negative effect on three of
26 the five event windows. Stated differently, the introduction of Basel IIII has a negative effect on the banks’ credit default swaps: after the enactment of Basel III, the banks’ CDS spreads fell by approximately 3 to 11 basis points, depending on the chosen event window. For all the different windows, the stock returns are not significant at a minimum of a 5% significance level. The change in the spot rate and the slope of the term structure are significant for three of the five different windows, while the coefficients do switch signs. The effect of the change in market volatility is – when significant – negative, in contrast to the previous literature. This means that, when the change in market volatility is positive, more uncertainty, there is a decrease in the credit default swaps. Since credit default swaps can be seen as an insurance to default, this is the opposite of what is expected. The model presented explains roughly 35-60% of the variance in the credit default swaps of the 12 examined banks.
Table 5 – Impact of Basel III on the riskiness of banks (09/07/2013)
+/- 13 weeks +/- 8 weeks +/- 4weeks +/- 2 weeks +/- 1 week Basel III -10.091** -5.891 -2.970 -10.654** -2.995** 4.321 4.578 2.156 4.313 1.324 Treated Banks × Basel III -3.006 0.687 -5.849 -8.243 -4.198 1.997 1.039 3.782 4.796 3.008 Stock return -0.005* -0.008* -0.005 -0.003 -0.006 0.003 0.004 0.004 0.004 0.005 Δ Spot Rate 0.867** -0.275 -1.759 -0.546** 0.228*** 0.339 0.562 1.289 0.216 0.062 Δ Slope -1.878** 0.028 2.339 1.239** -0.800** 0.677 0.401 1.761 0.457 0.264 Δ VIX -0.036** -0.034 0.108 -0.058* -0.034** 0.012 0.040 0.079 0.028 0.012
Time fixed effects Yes Yes Yes Yes Yes Bank fixed effects Yes Yes Yes Yes Yes Observations 1536 936 468 228 120 R2 0.404 0.478 0.574 0.621 0.606
Adjusted R2 0.350 0.430 0.534 0.585 0.566
Notes: This table looks at the impact of Basel III on the credit default swaps of banks, dependent on their exposure to the repo
market. The variable of interest is the interaction term: Treated Banks × Basel III. The variable Treated Banks is one if the bank belongs to the treatment group (if the bank is exposed more to the repo market than the median bank) and zero if the bank belongs to the control group. The variable Basel III is one if the observation is after the enactment date, and zero otherwise. The different columns provide the results for different event windows. Column 1 shows the results for a +/- 13 weeks, meaning that observations of three months before and 3 months after the enactment of Basel III are included. The 5-year credit default swap is the dependent variable. The stock return is the daily return for each bank stock. Δ Spot Rate is the change in the 5-year Treasury Constant Maturity Rate. Δ Slope is the change in the difference between the 10- and 2-year Treasury Constant maturity rate. Δ
VIX is the change in the CBOE Volatility Index. Robust standard errors are reported in italics. Constant were included in the
27
6. Robustness checks
In this chapter, several robustness checks are conducted and all results are presented in tables included in the appendix. In section 6.1, the treatment and control groups are constructed differently to determine whether this impacts the results. In section 6.2, other announcement and enactment dates are used to examine whether the results change if other, justifiable dates are used as the event dates. Finally, in section 6.3, the results are presented for if the announcement and enactment dates are considered to be one day earlier or later than the actual event date.
6.1. Change of the treatment and control group
This section investigates whether the results change significantly if the treatment and control group are slightly altered. First, for the announcement date, this study investigates what happens to the results if two banks are excluded from the analysis, since their sensitivity to the repo rate was not significant. Second, a robustness checks is conducted by which the control group consists of all banks with either an insignificant or a negative relation to the repo rate. Third, for both the announcement and the enactment date, one bank is added (subtracted) from the control group and subtracted (added) to the treatment group. So that the control group consists of 7 (5) banks and the treatment group of 5 (7) banks. Finally, a robustness check is conducted in which the banks with a medium sensitivity to the repo rate are excluded from the sample.
6.1.1. Excluding Charles Schwab and JP Morgan
In section 5.1, the treatment and control group were determined for the announcement date of December 16, 2010. As previously mentioned, the coefficients for Charles Schwab and JP Morgan were insignificant, whereas the coefficients for all other banks were significant, either positively or negatively. Therefore, including these two banks in the analysis might affect the results. Therefore, a similar regression to the one in section 5.2 was conducted, excluding Charles Schwab and JP Morgan from the sample.
The results for this regression are presented in Table 7. It can be concluded that the results are not significantly affected by this change. The variable of interest, the interaction term between the treatment group and the announcement date of Basel III, is still insignificant for all time windows, except when the data is taken into account from 3 months before the announcement date to 3 months after the announcement date. However, since the results are still insignificant for the other columns, it still cannot be concluded that the announcement of
28 the capital requirements of Basel III affects banks in the treatment group differently than banks in the control group. Moreover, all other variables that were significant in the analysis conducted in section 5.1 are still significant if Charles Schwab and JP Morgan are excluded. The signs of the coefficients also remain the same. Furthermore, the variables with no significant effect remain significant if these two banks are excluded. The explanatory power changes marginally: an increase of only three to six percentage points. Thus, it could not be concluded that excluding these two banks affects the results significantly.
6.1.2. Assigning all banks with insignificant or negative coefficients to the control group
Another option to handle the insignificance of the coefficients for Charles Schwab and JP Morgan (for the announcement date) is to assign them to the control group, since they are not significantly impacted by a change in the repo rate.
The results for this regression are presented in Table 8. These results suggest that there exists a significant, negative relationship between Treated Banks × Basel III and banks’ credit default swaps. According to the outcomes for the 13- and 8-week window, banks that belong to the treatment group are subject to a negative shock in their credit default swaps of the announcement of Basel III compared to banks belonging to the control group. This result is the opposite of what was expected, and disappears if a smaller event window is taken into account. Therefore, the hypothesis that the treatment group experiences a larger increase in their credit default swaps than the control group, is rejected. On the other hand, there is not enough evidence for whether the relation can be considered negative or nonexistent. The other variables in this regression do not change compared to Table 3.
Results for the enactment date of July 9, 2013 are presented in Table 9; these results differ on important points from the main results presented in Table 5. If the control group consists of all banks that have a negative or insignificant coefficient in Table 4, the results for the interaction term for the 4-, 2- and 1-week window become significantly negative. This means that the CDS of banks belonging to the treatment group decreases by more after the enactment of Basel III than the CDS of banks that are less exposed to the repo marker. This is in contrast to the expected positive relationship. Hence, the hypothesis should be rejected, there is no evidence for a concave relationship as suggested by Matta and Perotti (2016). Furthermore, the evidence for a negative relation is weak since it only exist for the 4-, 2, and 1-week window.
