Contingent convertible issues in Europe : what is the impact on CDS spreads of European banks?

(1)

Supervisor: Mark Dijkstra 15/08/2016

Master Thesis

Contingent convertible issues in Europe. What is the

impact on CDS spreads of European banks?

Ewout Schikker

11096225

MSc Business Economics, Finance track

(2)

Statement of Originality

This document is written by Student Ewout Schikker 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.

(3)

Acknowledgements

I would like to thank my supervisor, Mark Dijkstra, who has guided me in the starting phase of my thesis and gave valuable reccomendations.

I also would to like to thank mister Ted Welten, who has provided me with guidance during the theoretical writing process this summer.

(4)

1

Abstract

This thesis analyzes the effects associated with introducing contingent convertible bonds (CoCos) in the capital structure on the default probability of European banks. For this purpose, I have conducted a panel data analysis to investigate the effect of CoCo issuance on the credit default swap spreads of European banks. Previous research has focused on the ex-ante and ex post announcement effects of contingent convertible issuance by carrying out an event study. This study mainly focus on the ex post effects of CoCo issuance. This study analyzes the CoCo market from the first issue in 2009 to July 2016. The findings indicate that the CDS spreads are significantly negatively related with CoCo issues suggesting that CoCo issuance will decrease the default probability of European banks.

(5)

2

1. Introduction

In this first chapter an introduction to the research topic is provided. First I will present a brief summary of the background of contingent convertible bonds. In this paragraph I will explain how and why contingent

convertibles are originated. In the following paragraph I will present the problem discussion and research question following from the background. The chapter ends with an outline of the thesis.

1.1 Background

The financial crisis of 2007-2009 has revealed the fragility of the world’s financial system. Many banks had to suffer losses that harmed their reserves as well as their equity bases. This led to under-capitalized financial institutions and even to bankruptcy of well-known banks like Lehmann Brothers. The crisis revealed that financial institutions had built up credit and liquidity risks from their investments (Calomiris and Herring, 2012). Most banks were unable to raise additional equity capital and to provide any loss absorption capacity to be able to survive even slight negative macro-economic shocks (Avdjiev et al., 2015). As a result, regulators and governments had to intervene and bail out multiple banks to decrease the risk of an even bigger financial crisis. The worldwide public rescues exposed the

importance of ‘global systematically important banks’. By nationalizing and using taxpayers’ money to re-capitalize banks, governments and regulators wanted to safe banks that are called ‘too-big-to-fail’. This led to a classical moral hazard problem where banks had no incentive to reduce their balance-sheet risk, since public funds are used as insurance (Calomiris and Herring, 2011). To address the moral hazard risk, and to address to problem of the too-big-to-fail-banks, financial regulation needed a major revision. The Basel

Committee on Banking Supervision had addressed these problems by issuing a set of proposals with the main purpose to enhance the quality, consistency and transparency of the regulatory capital base. This so called Basel 3 agreement has increased the level of regulation and supervision on risk management (Basel Committee on Banking Supervision, 2011). In July 2013, the European Commission announced to Capital Requirements Directive 4 package to increase the level of regulation. As a result, the Basel 3 agreement was

(7)

4

implemented into the EU law. Besides higher liquidity and capital requirements, the new EU law stated that all capital instruments need to absorb losses in distress situations to be recognized as regulatory capital consisting of retained earnings, common equity and contingent convertibles (Niedrig and Grundl, 2015). The loss absorption mechanism means that capital instruments should be fully or partially written down the principal of the instrument or be converted into common equity when the institution’s capital ratio falls below a specified level of insolvency. Before the financial crisis many large financial institutions have issued capital instruments with loss absorbing mechanism. They mainly have issued the ‘going-concern capital’ and the ‘gone-concern capital’. The going-concern capital is a form of hybrid capital, which involves equity and debt features and absorb losses by postpone coupon payments or extend the maturity of the capital instruments. Gone-concern capital is a form of subordinated debt, which absorbs losses in case of bankruptcy (Martino et al., 2010). The financial crisis and government interventions have proved that most of those instruments did not have the desired effect. Because most financial

institutions did not want to send negative signals to capital markets by postponing coupon payments, the loss absorbing mechanism of these instruments performed poorly

(Pazarbasioglu et al. 2011). As a result of these fundamental shortcomings, most former forms of hybrid capital would no longer qualify as Tier 1 or Tier 2 capital under the Basel 3 agreement. Because common equity qualified capital is relatively costly, a new form of hybrid capital has been introduced to meet the higher capital requirements (Avdjiev et al. 2015). Flannery (2002) introduced a new instrument called ‘’Reverse Convertible

Debentures” that was arguably the first contingent convertible proposition. This instrument would automatically convert from debt to equity when the issuer’s equity ratio falls too low. Flannery (2002) stated that this instrument facilitates a mechanism for preventing financial distress by reducing leverage in a bank without encouraging risk taking by shareholders. After the financial crisis contingent convertible bonds (CoCos) received attention from various economists, academics, regulators and banks as a potential instrument to ease the impact of the financial crisis and to reduce the need for public rescues. Hilsher and Raviv (2014) stated that the addition of contingent convertibles to banks’ regulatory capital was one of the best-suggested solutions for capital shortfall in times of distress. A contingent convertible bond (a CoCo bond) is a form of long-term debt with a fixed

(8)

5

coupon rate that absorbs losses on a going concern basis when the issuing firm or bank reaches a specified level of insolvency. Based on the contractual agreement, the loss absorption mechanism can take different forms. They could be either converted to equity, or the face value of the bond could be written down partially or completely (Schmidt and Azarmi, 2015). The trigger that activates the conversion could also take different forms. It could be either a predefined mechanical trigger, based on market or book values, or it could be a discretionary trigger, based on the regulators’ discretion about the level of insolvency of the issuer (Avdjiev et al. 2015). Unlike other hybrid capital instruments, the mechanism of swapping debt into equity or writing of the principal is considered potentially valuable since CoCo bonds are intended to automatically stabilize a bank’s balance sheet in times of financial distress and reduce the overall systemic risk (Hilsher and Raviv, 2014).

1.2 Problem discussion and research topic

The contingent convertible is a relatively new and complex capital instrument. There is still a lot to investigate in this area to get a better understanding of the structure and their related effects on the financial system. Since 2009 the number and size of CoCo issues have

increased exponentially. Until July 2016, 291 contingent convertibles have been issued with an issue amount of around 305 billion Euros. CoCos could also play an important role in stabilizing the financial system. Policymakers and regulators are interested in monitoring the risk of a financial institution due to the potentially harmful effect of default on the economy. Therefore it is interesting to analyze this new capital instrument.

Only since 2009, several studies have focused on the impact of CoCo issuance on banks’ default risk. This issue has been controversially discussed in prior literature. On the one hand, CoCos should decrease banks’ default probability due to their construction of automatically loss absorption (Avdjiev et al. 2015, Hilscher and Raviv, 2014). On the other hand, Cocos could create moral hazard problems and could be also unable to provide adequate loss absorbing capacity to banks (Admati et al. 2013, Chan and van Wijnbergen, 2015). Therefore, CoCos could also raise the default probability of banks. However, theoretical and empirical evidence on the impact of CoCo issuance on banks’ default probability remains scarce. Academics have not achieved broad and balanced consensus regarding the effect of CoCo issuance on banks default probability.

(9)

6

In this thesis I will analyze the effect of introducing contingent convertible bonds in the capital structure on the default probability of European banks. In order to analyze this issue I will answer the following research question:

Does CoCo bond issuance have an impact on the probability of default of European banks?

I expect to encounter a negative reaction from the credit default swap spread (CDS spread) to CoCo issues. In order to measure the effect, I will run a panel regression on a sample of 32 European banks (see table 8) that issued 72 CoCo bonds over the time period of January 2009 to July 2016.I will use the credit default swap spread as a proxy of banks’ default probability. Control variables are included in the regression models to control for fixed effects that could impact or bias the outcome variables. Based on prior literature I have included the control variables that correspond with the most explanatory power of CDS spreads.

Studies on the impact of CoCo issuance on banks’ default probability are scarce. My research will provide several contributions to the existing literature. First of all, I believe, this will be the first research that analyzes CoCo issues of European banks only. My research will cover 32 large and small banks headquartered in Europe. The reason for analyzing European banks is that these banks all have to meet the new Basel 3 requirements. In Europe the regulations are different from regulation in other parts of the world. Therefore it is interesting to investigate whether the intention of the Basel 3 framework to strengthen the quality of the capital base has been successful. Due to these similar regulations, European banks will have more similar characteristics, which will limit bias in the research. Secondly, previous studies have focused on the ex-ante and ex post announcements effects of CoCo issuance while I will mainly focus on the ex post effects. The few studies that have measured the issue effect on the CDS spread have focused on issue announcements. Therefore to my knowledge, this is the first study that measures the effect on the CDS spread based on a panel regression instead of carrying out an event study on the issue announcements.

(10)

7

1.3 Outline

The rest of the thesis is structured as follows. The next chapter will present a more accurate explanation of CoCos. Subsequently the theoretical framework and the related research will be provided to get a better understanding of the research problem. Based on the theoretical framework and related research, the hypothesis will be stated which will be the base of my empirical analyses. Chapter 3 will present the econometric model that is going to be tested. Chapter 4 will describe the data sources and present descriptive statistics of the variables used in the analysis. In the following chapter the results will be presented. Economic meaning of the results will be given. Finally I will discuss the limitations of the study. In chapter 6 I will present robustness tests in order to critically assess presented results. In the last chapter I will present the conclusion. In this chapter I will explain the economic

meanings of the results. Subsequently I will describe the limitations and provide some important aspects that should be investigated in the future.

(11)

8

2. Theoretical framework and literature review

This chapter presents a more accurate explanation of CoCos and its main structure. In the first paragraph I will discuss the regulatory framework of CoCos based on the Basel 3 framework. The second paragraph will provide a more developed explanation of CoCos and its features. Subsequently, the theoretical framework related to the research question will be introduced and discussed. The fourth paragraph will focus on the chosen measure of banks’ default probability. After that, a summary of related studies will be presented including their drawbacks. This will finally be used to set the hypothesis of the research

2.1 Basel 3 framework and CoCo regulation

As mentioned before the Basel 3 framework is a comprehensive set of measures proposed to increase the level of regulation and to strengthen the supervision on risk management in the financial industry. The banking sector’s ability to absorb losses and shocks arising from distress situations should be increased (Basel Committee on Banking Supervision, 2013c). The framework was a response to the poor performance of banks revealed by the financial crisis. It underlined the importance of a high quality capital base in order to decrease the risk. The most important reason that the financial crisis became so severe was the low quality capital base of banks in many countries (Basel Committee on Banking Supervision, 2011). The crisis has revealed that losses had to be covered out banks’ common equity base, especially out of retained earnings. Furthermore, public rescues were consisting of

injections in form of common equity or other forms of Tier 1 capital. This meant that Tier 2 capital, and in some cases Tier 1 capital, did not absorb losses incurred by the banks. As a consequence, the criteria for classification as Tier 1 or Tier 2 capital were tightened. Basel 3 required that the highest quality component of bank’s equity capital, the best-ranked form of Tier 1 capital, must be retained earnings and common shares (Basel Committee on Banking Supervision, 2011). The Basel 3 framework separates regulatory capital into Tier 1 capital (going concern capital) and Tier 2 capital (gone concern capital). Tier 1 capital is divided in Common Equity Tier 1 (CET1) and Additional Tier 1 (AT1). The objective of Tier 2 capital is to provide loss absorption on a gone-concern basis, whereas additional Tier 1 is

(12)

9

designed to recapitalize banks at an earlier stage (Flannery, 2009). For capital instruments to be concluded in these categories, they need to meet requirements made in the Basel 3 agreement. One of the criteria is that the instruments need to absorb losses when a predefined trigger point is breached. For additional Tier 1 capital, if the trigger level is breached, the amount that must be written down or converted should be at least the amount necessary to return the Common Equity Tier 1 capital ratio to the trigger level. (Basel Committee on Banking Supervision, 2011). Another core concept of the Basel

agreement is that of the Risk Weighted Assets (RWA). The RWA ratio includes all assets and weights with regard to their credit risk. This approach has been introduced to compare risks of banks and encourage banks to hold low risk assets (Basel Committee on Banking

Supervision, 2013c).

Avdjiev et al. (2015) stated that the main driver of issuance of CoCos has been the

regulatory treatment of the bonds, and to boost their capital ratio to meet the new capital requirements. Under Basel 3, CoCos can either be qualified as additional Tier 1 capital or Tier 2 capital, both ranked below the predominant form of equity (CET1). A set of minimum criteria has to be met for CoCos to be concluded as additional Tier 1. According to the Basel Committee on Banking Supervision (2011), ‘it should be perpetual and there should be no incentives to redeem. It may be callable, only after a minimum of five years. Instruments classified as liabilities must be able to absorb losses through either conversion to equity at a pre-specified trigger or a write down of the principal at a pre-specified point’. For CoCos to be recognized as Tier 2 capital, the instrument should not have to be perpetual but instead have a minimum maturity of at least 5 years. In 2011, the Basel committee on banking supervision issued additional requirements about loss absorption at the point of non-viability (PONV). This proposition states that the capital instruments must have a provision that requires the instrument to absorb losses at the option of the relevant authority, if a trigger event occurs. This trigger has been known as the PONV trigger (Avdjiev et al. 2015).

2.2 CoCo design

As mentioned above, contingent capital instruments are hybrid debt securities that are able to absorb losses (Avdjiev et al. 2015). More specifically, CoCos are debt instruments

(13)

10

minimum of five years that pay tax-deductible interest and are automatically converted into equity or written down at the conversion price when pre-specified level of insolvency has been reached (Flannery, 2009). This definition of CoCos illustrates that the conversion trigger and the loss absorption are key characteristics of the CoCo structure. These characteristics crucially determine the effectiveness of CoCos in stabilizing banks

(Pazarbasioglu et al. 2011).Hence, a good understanding of the CoCo design is critical for regulators in order for CoCos to fulfill their function to stabilize banks as well as to price the CoCos (Hilsher and Raviv, 2014).

2.2.1 The trigger

The design of the conversion has a direct effect on the probability of conversion or write-down of the CoCo bond. Therefore it is one of the key determinants of the risk of the CoCo and a key element for pricing the instrument (Flannery, 2009). The trigger defines an event, which leads to conversion or write-down of the CoCo bond. The trigger determines multiple events that force the CoCo bond to automatically write-down or convert its nominal value into common equity (Skinner and Ioannides, 2011). The conversion trigger can take 2 different forms. It can either be a mechanical or a discretionary trigger. The discretionary trigger (PONV) is activated based on the critical decision of a regulator about the issuers’ solvency level. The mechanical trigger is characterized by a trigger variable and a trigger level often is defined in terms of a specific capital ratio. The most used ratio is the book value of Common Equity Tier 1 as a percentage of the risk weighted assets (Avdjiev et al.

2015). The literature has distinguished a few different triggers. The regulatory trigger, or PONV trigger, belongs to the category of discretionary triggers,

since this trigger leaves the decision to convert to the judgment of the regulator (Maes and Schoutens, 2012). The inclusion of this trigger has increased the past years due to the current Basel 3 framework, which requires that a CoCo contains this element to be included in AT1 and Tier 2. This feature decreases the risk of unreliable bookkeeping, but at the same time increases uncertainty about the timing of the loss absorption mechanism (Avdjiev et al.

2015).

The systemic trigger is linked to a system-wide condition and is almost independent of an individual bank’s financial condition. This trigger would be based on the condition of the

(14)

11

whole financial system. Variables that measure liquidity conditions, volatility or declarations of regulators should be included in the trigger (Pazarbasioglu et al. 2011). The effectiveness of the systematic trigger has been analyzed and discussed in the literature. Flannery (2009) states that the conversion should depend on bank specific variables because it is not possible to measure all banks by the same standard. Pazarbasioglu et al. (2011) state that the systematic trigger might be more efficient since capital increases simultaneously across

banks.

The accounting value trigger is a mechanical trigger using an accounting ratio as conversion trigger, typically the Core Equity Tier 1 capital to risk weighted assets. The effectiveness of the book value triggers has been controversially discussed in several papers. The

effectiveness of accounting value triggers depends on the accuracy and consistency of banks’ risk models and the frequency at which capital ratios are calculated. Vallée (2015) and Calomiris and Herring (2012) stated that accounting triggers are sensitive to

manipulation by bank managers. Pazarbasioglu et al. (2011) pointed out that capital ratios tend to be lagging indicators of a bank’s financial condition.

The market based trigger could address the shortcomings of accounting triggers since it represents the opinion of a large number of market participants and it is less sensitive to unreliable bookkeeping. Academics have discussed several market-based measures for potential use as triggers. Flannery (2009) proposed to use the market value of equity as measure, whereas Calomiris and Herring (2012) suggested using the credit default swap spread or share price movement as market based measures. However, there are also concerns about market-based triggers since market arbitrageurs could manipulate the market price, the so-called death spiral risk (Maes and Schouten, 2012).

2.2.2 Loss absorption

The conversion mechanism can take 2 forms. They could be either converted to equity, or the face value of the bond could be written down partially or completely (Avdjiev et al. 2015). Spiegeleer and Schoutens (2014) stated that there are three different forms of write-down mechanism.

First, the principal of the CoCo could be fully write-down. This means that the face value of the CoCo bond would be completely written off when the pre-specified level of insolvency

(15)

12

has been reached (Flannery, 2009).

Second, the CoCo could have a partial write-down. This means that a haircut is applied to the face value of the CoCo, and the difference will be repaid to the CoCo investor (De

Spiegeleer and Schoutens, 2014).

The third option would be a staggered write-down. This is a flexible write-down mechanism were the CoCo investor would be imposed to losses up to the point were capital

requirements are satisfied. Principal write-down CoCos is getting more common. The

reason for this could be the concern about the threat of dilution of current shareholders and the concern that the CoCo investor would own a controlling stake after the trigger is

breached. (De Spiegeleer and Schoutens, 2014). Equity conversion is the other possible mechanism. The conversion rate could be either linked to a predefined price or to the market price at conversion or to a combination of both (Avdjiev et al. 2015).

The conversion mechanism is an essential part of the design of the CoCo bond. When the trigger is breached, dilution of the current shareholder could become reality. The current shareholders thus prefer a higher conversion price, so that conversion is less likely. On the other hand, CoCo investors will prefer a low conversion price. This can result in adverse incentives among shareholders and CoCo investors (De Spiegeleer and Schoutens, 2014).

2.3 Credit Default Swap Spread

There are different possibilities to measure the default probabilities of banks. Common measurements used, are the value at risk or published corporate ratings produced by credit rating agencies e.g. Standard and Poor’s or. When it comes to new securities, a great

challenge for credit rating agencies is the assessment of the creditworthiness of the bank, and the risk that investors’ claims cannot be entirely paid back. Especially in the first years (2009-2012) CoCo bonds have not been rated by credit rating agencies. The marketability of CoCos in Europe is nowadays questionable due to the fact that the major rating agencies currently abstain from rating all CoCos (Schmidt and Azarmi, 2015). Therefore, credit rating is not a good proxy for default probability in relation with CoCo issuance. In this research, I will use another possible measure of institutions’ credit risk: the credit default swap spreads (CDS spreads). During the financial crisis, CDS spreads have become increasingly popular as

(16)

13

a measure of a banks’ default risk (Avdjiev et al. 2015). CDS spread consists of an agreement between two parties, the protection buyer and the protection seller. In exchange for a premium paid by the buyer, the seller agrees that, if a debt issuer defaults or another credit event occurs, he will pay the protection buyer the premium as well as well as all interest payments (Galil et al. 2014). CDS spreads are thus prices for an insurance against insolvency of a firm and therefore a good proxy of a firm’s default risk. Another advantage of using the CDS spread as default risk measure, is that CDS spreads reflect the market sentiment on a more frequently basis than credit ratings (Schmidt and Azarmi, 2015).

2.4 Literature review of CoCos and risk

Over the past years, academics have controversially discussed the impact of issuing CoCos on the banks’ risk. There are several articles that have stated that issuing CoCos will decrease the risk profile of banks. The financial crisis has proved that investors are not willing to provide capital in times of financial distress (Spiegeleer and Schoutens, 2014). CoCo bonds have been widely considered as a possible solution for this problem since it includes a mechanism for recapitalizing over levered financial institutions (Flannery, 2009). CoCo bonds behave like a straight bond in times of economic wellbeing but once the issuing bank’s capital ratio falls below a specified level of insolvency, the CoCo bond is converted into equity or is written down. Due to this conversion, the bank’s debt to equity ratio will decline and should lead to a reduction of the bank’s default probability since it provides higher capital cushions when it is most needed (Hilscher and Raviv, 2014). Due to the

conversion risk, CoCos offer a risk premium to investors paid by the issuing bank. The higher yields attract potential investors and the growing CoCo market offers an attractive

investment opportunity, which will decrease banks’ risk (Ammann et al. 2015). CoCos also have a fiscal advantage because they are treated as straight bonds before the conversion so that they are tax deductible, which makes it a much cheaper option compared to equity for holding an additional regulatory capital base. In accordance with the conversion, bankruptcy costs are obviated because the banks recapitalize automatically (Hilsher and Raviv, 2014). Avdjiev et al. (2015) stated that due to the lower costs of debt rollovers, the lower risk of default not only accrue to bond holders, but also to equity holders. Therefore equity holders

(17)

14

can have a positive incentive to issue CoCos, which will reduce the debt overhang problem. Debt overhang is a debt burden that is so large that a bank cannot issue additional debt to finance new investments, even though they are profitable (Albdul, 2012).

However,other academics stated that CoCos might become a source of risk. While the conversion of CoCos may bring the bank back into compliance with capital requirements, it could nevertheless raise the probability of the bank being run. The conversion of the CoCo is a negative signal to depositors that asset quality has deteriorated. In principal write down CoCos, the conversion provides no comfort to deposit holders since upon conversion wealth transfers go in the wrong direction from junior debtors to equity holders (Chan and van Wijnbergen, 2015). Because existing equity holders will not be diluted by a conversion, this will lead to low incentives to supply capital in times of financial distress (Flannery, 2014). Moral hazard may also arise when CoCos converting into equity dilute current shareholders insufficiently. If shareholders would not be diluted by the conversion, they might choose a risky investment ex ante, which increase systemic risk even before conversion has taken place. This will lead to higher probabilities of bank runs. (Chan and van Wijnbergen, 2015). Conversion of the CoCos may also effect the complication of contagion, when assets are correlated across banks. A conversion of one bank may lead to increased probabilities of runs in other banks. This is a neglected channel of risk, especially when banks hold each other’s CoCos (Chan and van Wijnbergen, 2015). Other skeptics of CoCos have argued that especially principal write-down CoCos is complex and unlikely to provide adequate loss absorbing capacity to banks. The size of losses that could be absorbed will be too small. Therefore, CoCo investors will suffer losses they are less equipped to manage than equity holders. This will increase the risk of default (Admati et al. 2013). According to Calomiris and Herring (2012), the decrease of banks’ default probability promotes inefficiency as failing banks and incompetent managers have not been replaced. Finally, according to Maes and Schoutens (2012), CoCo issuance might increase the systemic risk as smaller banks and private investors, which are more sensitive for slight shocks are the main issuers of this instrument. All in all, the design of the contingent convertibles are determined by their main role in decreasing the probability of a run of the issuing bank and thus their role in

decreasing systematic risk. However, it is still unclear if the contingent convertibles will fulfill their expectations in times of distress since no CoCo has been tested in financial crises.

(18)

15

(Schmidt and Azarmi, 2015). Therefore, Pazarbasioglu et al. (2011) stated that CoCos, even though they meet criteria for regulatory capital and might have the potential to play an important role in preventing coming crises, should be considered only as a part of a framework that would make the financial system stronger.

2.5 Related research

It is only since 2009, after the financial crisis, that academics have focused on contingent convertibles and their effects on systematic and banks’ default risk. After the regulatory changes of the Basel 3 framework, the literature concerning CoCos focused mainly on three directions of research. Firstly, there exists quite some literature with a focus on the

structures and design of the CoCos. This literature has given a better understanding of this new financial instrument and has analyzed the instrument’s role as financial insurance for banks (Flannery 2009, Pazarbasioglu et al. 2011, Maes and Schoutens, 2012). Secondly, several academics have focused on drawbacks of CoCos from distorted risk-taking

incentives, which could lead to a potential increase of the banks’ default probability (Hilsher and Raviv, 2014). Thirdly, academics have introduced different pricing models of contingent convertibles. These papers give a better understanding of the CoCos’ risk indicators (Albul et al. 2010, Spiegeleer and Schoutens, 2014). However, theoretical and empirical evidence on the effect of issuing CoCos on banks’

default probability measured by the credit default swap spread, remains scarce. A few articles made the attempt to measure the effect of CoCo issuance on banks’ default probability with different approaches. Several studies have estimated the announcement effect of CoCos on the CDS spread. In 2014, the first published article covering this subject was from Schmidt and Azarmi (2015). They analyzed the effects of the use of CoCos in Europe by Lloyds Banking Group in 2009. Before they empirically analyzed the

announcement effect, they stated that the direction of the CDS spread after the

announcement could go both ways. On the one hand, CoCos make banks more resistant to economic shocks, and therefore decrease the probability of default. On the other hand, after conversion the shareholders will be diluted, which will decrease the firm value and increase the default probability. In order to analyze the event they carried out an event

(19)

16

study. They used a method proposed by MacKinlay (1997). This method sets up a market model in order to define a reference point for normal CDS spread, presuming the issuance of CoCo had never happened and use this as benchmark. The event period covers a period of 20 days before the announcement to 20 days after the announcement. They have documented an increase in the CDS spreads following the announcement to issue CoCos. Schmidt and Azarmi (2015) suggested that CoCos could have a negative effect on a bank’s creditworthiness and thus increase the probability of default. The main drawback of this research is that it only has a small scope, since it only measured the effect of Lloyds Banking Group. Avdjiev et al. (2015) have presented another study on the effect of CoCo issuance. In their overview they highlighted that there has been a lively debate in academia and among policymaker on the stabilizing effect of CoCos on the banking system. Nevertheless, they developed their hypothesis based on the theory that CoCos should lower the probability of default. In order to test the effect, Avdjiev et al. (2015) used a sample of CoCos issued by banks from all advanced economies with the exception of the euro area periphery (Greece, Ireland, Italy, Portugal and Spain). In the empirical research, they tested to what extent CoCo issuance changes banks default risk and they analyzed how the effect depends on main contract features as the trigger level and the loss absorption. They also carried out an event study with an event period of 21 days, covering the pre-issuance 15 days and the post issuance 6 days. In order to measure the announcement effect they calculated the average prediction error by comparing the actual CDS spread with the weighted average of the ITraxx Senior Financials CDS spread. The main finding that emerged from this study is that the impact of CoCo issuance on CDS spreads is significantly negative, indicating that CoCo issuance decreases banks’ default probability. This study also contains a few drawbacks. First of all, they measure the effect using an event study. But unlike typical event studies, the announcement of the CoCo issuance is not at a clearly defined point in time. An

upcoming CoCo issue is not publicly announced at a single point in time. Therefore an event study might not be the best way to measure the effect of CoCo issuance on default risk. Also when calculating average standardized prediction errors, Avdjiev et al. (2015) assume that a bank is acting independently, while in reality banks’ decisions are highly correlated.

Ammann et al. (2015) have also investigated the announcement effect of CoCos on the CDS spread. They used a sample of 34 financial institutions. They examined abnormal CDS spread

(20)

17

changers before and after the announcement dates. The event period covers a period from ten days before the announcement date to 20 days after the announcement date. Their hypotheses and findings were almost the same as from Avdjiev et al. (2015). Hilscher and Raviv (2014) have investigated the effects of financial institutions issuing contingent capital and have quantified the reduction in banks’ default probability associated with issuing CoCos without measuring the CDS spread. They investigated the stabilizing abilities of CoCos by comparing it to two alternatives: capital structures that include subordinated debt or include additional equity. They constructed a model where in each case the probability of default is expressed as a function of claim parameters, market parameters and regulatory parameters. The most important parameters included are the banks’ leverage ratio, asset value, maturity, issue amount and conversion ratio. This model proves that the inclusion of contingent capital can have a decreasing effect on the default probability of banks. They have shown that a bank with subordinated debt has a higher default probability because the CoCo has a stabilizing effect on the capital structure. In case of deterioration in asset value, the automatic provision of already available capital will decrease the default probability. The main drawback of this study is the use of a subjective self-constructed model. This model is constructed in a way that CoCo inclusion or additional equity always has a decreasing effect on the default probability. Chan and van Wijnbergen (2015) have investigated the effect of CoCos on systemic risk. They especially focused on the effect of conversion on the risk. Therefore, they have added Coco issuers to an existing agent model. This model of Goldstein and Pauzner (2005) relies on the fact that depositors fund the majority of banks assets. Therefore the behavior of depositors plays an important role in the banks default

probability. By using this model they were able to show the impact of conversion on ex ante incentives of depositors, regulators and existing equity holders but also the default

probability of banks. Chan and van Wijnbergen (2015) proved that an unexpected decline in asset returns leads to conversion, which lead to an increase of the default probability. They have also shown that assets are correlated across banks. Therefore, the negative signal of conversion increases contagion across banks and thus increases the systematic risk. The main drawback lies in the fact that the model assumes that all contingent convertibles are discretionary triggered, which is not a realistic assumption.

(21)

18

2.6 Hypothesis

In order to test the effect of the CoCo issues on the CDS spreads I first have to state the hypothesis. This hypothesis is based on the background, the presented theoretical background and related research. The proposed hypothesis is:

H0: CoCo issuance has a significantly decreasing effect on the credit default swap spread

H1: CoCo issuance does not has not a significantly decreasing effect on the CDS spread

That is, I expect that CoCo issuance will decrease the default probabilities of European banks. By automatically providing higher capital cushions in times of distress the banks debt to equity ratio will decline, which will decrease the CDS spread. Therefore issuing contingent convertibles will lead to a reduction of the bank’s default probability. I expect that this effect will outweigh the negative effects of issuing CoCos.

(22)

19

3. Methodology

In this chapter I will describe the econometric model. First a brief description of the collected data is provided. Subsequently I will explain the relationship and the econometric model that will be tested. Finally, the most important control variables and the expected relations will be discussed.

The market of CoCo bonds is relatively new, since the first CoCo was only issued in 2009. Therefore the market is still small but it is rapidly increasing in size with around €305 billion issued by 154 issuers between 1 January 2009 and 30 June 2016.Until July 2016, 291

contingent convertibles have been issued. In order to measure the effect of CoCo issuance, I have collected data from all contingent convertibles that have ever been issued around the world in euros including their main features. My dataset covers the period from January 2009 to July 2016 and all amounts are measured in euros. The most important features of the CoCo issues considering this study are the issue date and the issue amount. I have also collected several other CoCo features as the trigger type, the trigger level, the loss

absorption mechanism and the Basel 3 designation. Since this study measures the effect of CoCo issuance of European banks, a selection of specific issues has been made. Therefore 133 issues of insurance companies or banks outside Europe have been sorted out. For the remaining European banks I have collected the monthly CDS spreads. Finally I have collected bank specific data that could be used as control variables. First I have collected several accounting specific data including the Tier 1 and Tier 2 capital, the risk weighted assets, the total assets, the total debt and the total equity of the corresponding banks on a monthly basis. I have also collected other bank specific information as the Standard & Poor’s credit rating, the banks’ stock price and the equity volatility of each bank. These data are also collected on a monthly basis from January 2009 to July 2016. Summarized, this research is most interested in the effect of CoCo issuance on the credit default swap spreads of European banks. In order to test the ex post effect of CoCo issuance on the CDS spread I have constructed four panel regression models with the banks’ CDS spread as the

dependent variable. Panel regressions have the advantage to capture both time and cross sectional information. Therefore, panel regressions are more informative than other

(23)

20

regressions that only capture the variation across firms or time. Panel regressions allow for an increased number of measurement points with more variation and it also mitigates the collinearity problems between firms or time. This is the reason why panel regressions are the best to run for testing a causal relationship, rather than just a correlation.

In each regression the CoCo issue amount of the corresponding bank has been used as the only explanatory variable. Finally, control variables are included in the regression models to control for fixed effects that could impact or bias the outcome variables. I have performed a Hausman test. The Hausman test for fixed effects versus random effects shows significant results (Prob>chi2 = 0.0000). Therefore the equations are estimated in fixed effects. All equations are estimated with clustered standard errors as well as to correct for both heteroskedasticity and autocorrelation.

(1) CDSit = α1 + 𝛿1D1 CoCo issue amounti,t-1 + β1 Volatilityi,t + β2 Stock priceit + β3 Debt to equity ratioi,t-1 +β4 Tier 1 capital ratioi,t-1+ β5 Credit rating classi,t-1 + Time fixed effects +µit

𝑖 = 1, . . . N ; t = 1, . . . ,90

(2) CDSit = α1 + 𝛿1D1 CoCo issue amounti,t-1 + β1 Volatilityi,t + β2 Volatilityi,t-1 + β3 Stock priceit + + β4 Stock pricei,t-1 + β5 Debt to equity ratioi,t-1 +β6 Tier 1 capital ratioi,t-1+ β7 Credit rating classi,t-1 + Time fixed effects +µit

𝑖 = 1, . . . N ; t = 1, . . . ,90

(3)CDSit = α1 + 𝛿1Issue amount ratio to assetsi,t-1 + β1 Volatilityi,t + β2 Stock priceit + β3 Debt to equity ratioi,t-1 +β6 Tier 1 capital ratioi,t-1+ β7 Credit rating classi,t-1 + Time fixed effects +µit

𝑖 = 1, . . . N ; t = 1, . . . ,90

(4)CDSit = α1 + 𝛿1Issue amount ratio to assetsi,t-1 + β1 Volatilityi,t + β2 Volatilityi,t-1 + β3 Stock priceit + + β4 Stock pricei,t-1 + β5 Debt to equity ratioi,t-1 +β6 Tier 1 capital ratioi,t-1+ β7 Credit rating classi,t-1 + Time fixed effects +µit

𝑖 = 1, . . . N ; t = 1, . . . ,90

Where i denotes the individual bank, t denotes the time in months, CDSit is the banks’ CDS spread,CoCo issue amountit represents a dummy variable of the CoCo issue amount of the

(24)

21

bank which takes the value 0 when no CoCo has been issued and the value 1 when a CoCo has been issued. In the third regression the Issue amount ratio to assetsi,t-1 has been used as the explanatory variable. This ratio represents the banks’ CoCo issue amount to its total assets. This ratio is used to correct for the total assets of the bank, which could affect the banks’ CDS spread. Therefore, including a CoCo issue ratio as percentage of the total assets would eliminate bias. Volatilityit and Volatilityi,t-1 are the banks ‘equity volatility and the one month lagged equity volatility. TheStock priceit and the Stock pricei,t-1 are the banks stock price and the one-month lagged stock price. TheDebt to equity ratioi,t-1 and theTier 1 capital ratioi,t-1 are the banks’ one month lagged accounting control variables. The Credit rating

classit denotes the bank’s Standard and Poor’s credit rating from AA to B concerning the used banks. Note that this is not the bond specific credit rating. In order to include those ratings in the regression model I made the following categories to measure the creditworthy of different banks where AA denotes the most creditworthy category, and B the less

creditworthy category.

Table 1: Standard and Poor’s credit rating and corresponding category class

To measure the ex post impact of CoCo issuance on the CDS spread I match the CoCo issue dates of the bank to the bank’s CDS spread. I have collected the monthly CDS spread data. I use the CDS spreads the month after the CoCo issue as dependent variable to measure the impact of the issue. In this study, the maximum of days between the issue date and the measurement of CDS spreads is 21 days. For this reason I use lags in the regression to link

Credit rating Category class

AA 1 AA- 1 A+ 2 A 2 A- 2 BBB+ 3 BBB 3 BBB- 3 BB+ 4 BB 4 BB- 4 B+ B 5 5

(25)

22

the control variables to the CoCo issue date. So the control variables represent banks specific information at the moment the CoCo has been issued. Only for the stock price and the equity volatility I use both the normal and the lagged variables because these variables can change each day. Therefore those variables could have both an impact at the measure date of the CDS as the month before the measure date.

Prior research has argued for endogeneity problems in financial research. With the inclusion of fixed effects I have mitigated the problem of bias as much as possible. However, the most reasonable bias to address in my empirical research is the omitted variable bias. There is a probability to leave out important explanatory or control variables especially in the case of measuring the CDS spread. An infinite number of factors could influence banks’ CDS spreads. Therefore results could be able to only partly reflect the extent of CoCo issues affecting the CDS spreads of European banks. After reviewing prior literature I have included the variables that corresponds with the most explanatory power when it comes to

measuring the effect of CDS spreads. These control variables should eliminate bias as much as possible to study the effect of CoCo issuance on CDS spreads. Prior literature has

controversially discussed the most important determinants of CDS spreads. Abid and Naifar (2006) have distinguished five different variables influencing CDS spreads including credit ratings, time to maturity of the CDS contract, slope of the yield curve, equity volatilities and risk free interest rates. They have stated that these 5 variables explain more than 60 per cent of the total level of credit default swap spreads. Aunon-Nerin et al. (2002) have used the institutions’ credit rating, asset volatility, stock prices, leverage and market

capitalization in combination with market information to explain the CDS spreads. They have found that these variables drive up more than 80 per cent of the variation in CDS spreads. In order to determine the most important variables I have selected the most significant variables of different studies. These variables could be separated in accounting variables and other bank specific information. Prior literature has proven that the two most important accounting variables are the Tier 1 capital ratio and the Debt to Equity ratio. The Tier 1 ratio is the ratio of banks’ core equity capital (Tier 1 capital) to its risk weighted assets. Since CoCos are regulated in the Basel 3 framework this ratio should be included in the model. The debt to equity ratio has been calculated by dividing banks’ total leverage by

(26)

23

the shareholders equity capital. The other bank specific control variables included in this study are the credit rating, the stock price and the equity volatility.

I expect that the all included variables will have an effect on the CDS spreads of European banks. First, as mentioned the previous stated hypothesis I expect that CoCo issues will have a negative and significant impact on the CDS spreads. Therefore I expect a negative

coefficient, which suggest that when CoCo issue amount increases, corresponding CDS spread will significantly decrease. I expect that the Tier 1 capital ratio and the CDS spread be negatively related with each other. The higher Tier 1 capital ratio, the higher core equity capital to its risky assets. This will decrease the probability of a bank run, and so will lower the CDS spread (Avdjiev et al. 2015). Contrary to the Tier 1 capital ratio, I expect that the debt to equity ratio will be positively related to the CDS spreads. A higher level of leverage will increase the default probability of a bank (Chan and van Wijnbergen, 2015). Same logic applies for the stock prices of banks. Higher stock prices signal a healthier state of the bank, which lowers the credit risk. Higher equity volatility reflects higher equity uncertainty, which accompanies higher risks, and thus will be most certain positively related to the CDS spread (Abid and Naifar, 2006). Finally I expect a positive relation between the Standard and Poor’s credit rating and the CDS spread. As showed in table 1 the least credit worth banks have the highest rank (category class 5). Therefore it is plausible to expect a positive relation between those variables (Aunon-Nerin et al. 2002).

Table 2: Bank CDS spread descriptives and the expected coefficient sign

Variable Expected Coefficient sign

CoCo issue amount -

Volatility +

Stock price -

Debt to equity ratio +

Tier 1 capital ratio -

(27)

24

4. Data and descriptive statistics

In this chapter I will describe the data sources I have used in this study. It will present a more accurate explanation of which data are used. Subsequently, I will present descriptive statistics of the variables. Finally I will test the correlation between variables and present those results in a correlation matrix.

4.1 Data collection

The bond data used in this study consists of CoCo bonds issued by different European banks during the time period of January 2009 to July 2016. The sample contains 291 issues, which is the whole sample before sorting out non-banks and limitations for missing values. Since the first CoCo has been issued in 2009, around €305 billion worth of CoCos have been issued by 154 different issuers. For measuring the effect of CoCo issuance on the CDS spread of European banks a selection has been made. 154 issues of non-banks or banks outside Europe (including Russia) have been sorted out. Eliminating small subsidiaries from the dataset has made further alternations. For example, a lot of subsidiaries from SpareBank (Norway) have issued contingent convertibles that were not representative to include in the model. After this selection, banks specific data has been collected for the remaining

European banks. The collected panel data is unbalanced due to several gaps in both bank specific data and the CDS spreads. The average amount of CDS data available throughout the whole data is 94.8%. Almost 60% of the data is totally complete for all necessary variables included in the model. There are multiple reasons why several banks had to be eliminated. For example, Rabobank and Coventry Building society are not publicly listed and therefore do not provide stock prices. All required CoCo bond related panel data is

obtained from Bloomberg Professional. Several previous studies like De Spiegeleer and Schoutens (2014), Avdjiev (2015) and Schmidt and Azarmi (2015) have used the same data source to get both quantitative and qualitative data on contingent convertibles. In addition to Bloomberg, I have utilized Thomson Reuters DataStream to collect the bank specific information except the credit ratings, which were only available in the Bloomberg

(28)

25

DataStream (e.g. the year 2016) I have used Bankscope to obtain additional values. Due to the combined use of those three databases I have collected a unique and up-to date dataset that is required to present a representative study on the CoCo issuance effects on banks’ CDS spread.

4.2 Descriptive statistics

Table 3: CDS spread descriptives of the measured European banks

Variable Mean Std. Dev. Min Max Observations CDSspread overall 216,4323 321,5325 19,865 6031,82 N = 2731 between 275,1373 58,00071 1499,114 n = 32 within 219,8262 -923,2015 4749,138 T-bar = 85.3438 issued_dummy overall 0,025 0,1561521 0 1 N = 2880 between 0,0187205 0 0,0777778 n = 32 within 0,1550608 -0,0527778 1,013889 T-bar = 90 issued_ratio_to_assets overall 0,0033938 0,0061661 0,0003762 0,0400129 N = 72 between 0,0090728 0,0009146 0,0400129 n = 30 within 0,0006628 0,0013846 0,0051202 T-bar = 2.4 Volatility overall 0,5814267 0,7743514 0,08 7,108951 N = 2617 between 0,6546273 0,2451111 3,02754 n = 30 within 0,417827 -1,439602 4,662837 T-bar = 87.2333 Stockprice overall 29,9456 80,08234 0,0182 779,999 N = 2616 between 72,63904 0,1252695 400,2387 n = 30 within 33,9234 -135,1507 710,8042 T-bar = 87.2 Debttoequityratio overall 6,913753 5,230022 0,632886 30,82514 N = 2712 between 4,761391 0,8276222 28,85845 n = 32 within 2,106282 2,336229 20,68876 T-bar = 84.75 Tier1capitalratio overall 0,1376803 0,051497 0,0428783 0,4245999 N = 2556 between 0,0425987 0,0983123 0,331491 n = 32 within 0,0276328 0,0492061 0,2769243 T-bar = 79.875 Creditratingclass overall 2,1875 0,9141637 1 5 N = 1968 between 0,6933252 1 4,066667 n = 23 within 0,6067045 0,1208333 3,654167 T-bar = 85.5652

Descriptive statistics for all variables in the regression can be found in table 3. The statistics give a good picture over the number of observations of the used variables. The big N represents the total amount of observations. These are the monthly banks observations between January 2009 and July 2016. So the numbers of observations are approximately 2500 for each variable except for the credit rating class. Due to the unbalanced panel data it differs per variable. The small n represents the amount of measured banks, which also differs per variable. For each variable the mean, standard deviation, the minima and

(29)

26

maxima are given for three categories. First the overall category is given. In this category all possible observations are measured as can be seen in the observations column. Second the category between is given. This category measures the variables per bank and subsequently compares the different banks. Therefore the maximum of observations is 32 as shown in the last column, which means that 32 different banks are measured in this study. Finally the within category is given, which present descriptive statistics within the same month. The T-bar represents the average available months per variable. As can be seen in the last row, the credit rating class is divided in five classes as mentioned before. The mean of 2.1875 means that on average the banks have a credit rating between A/A- and BBB+/BBB/BBB-. Finally the equity volatility is measured as the percentage change of the equity price index over 1 month. The range between maxima and minima are

suspicious at first sight. The difference of the maximum and minimum measured CDS spread is more than 6000 while the average CDS spread equals 216. CDS spreads are quoted in basis points of the contract’s notational amount.Since the sample contain both small banks with problems during the covered period and so called banks that are too-big-to-fail it is reasonable that these amounts differ a lot. Same logic applies for all other variables especially the stock price of banks. The mean stock price is €29.95. However there is quite some variation in the stock prices, as can be seen from the standard deviation, which is around €80. Due to the differences between banks, some standard errors are even higher than the mean of several variables. Therefore it is very important to eliminate outliers in our regression model because those could bias the effects.

4.3 Correlation

Table 4: Correlation matrix of measured variables

Variable CDSspread issued_dummy Issued_to_assets Volatility Stockprice L1.Debt_to_equity L1.Tier1ratio Creditrating

CDSspread 1 issued_dummy -0.0849 1 issued_to_assets -0.0718 0.8159 1 Volatility 0.3222 -0.0537 -0.046 1 Stockprice -0.2761 0.0474 0.0202 -0.1689 1 L1.Debt_to_equity -0.0297 -0.055 -0.0585 -0.0481 0.0081 1 L1.Tier1ratio -0.3443 0.0868 0.1029 -0.1994 0.1015 0.1329 1 Creditrating 0.3937 0.0306 0.0399 0.2663 -0.2274 -0.2246 0.0031 1

(30)

27

Table 4 provides a correlation matrix of the used variables. The table shows the level and the sign of correlation between the measured variables. The second column is the most interesting since it measures the correlation of all variables with the CDS spread. As can be seen in the table, the volatility, stock price, Tier 1 capital ratio and credit rating are most linked with the CDS spread. These variables each explain about 30 percent of the variation in the CDS spread. Also the signs of the correlation almost meet all the expectations presented in table 2. Only the relation of debt to equity ratio to CDS spreads gives odd results. The correlation of the two variables is low, but also not representative. It is not reasonable that a relative higher level of leverage will decrease the CDS spread and thus the credit risk. Furthermore, this matrix could indicate a possible multicollinearity problem. This is the case when the correlation among the used variable is too high. In this matrix the issued dummy and the issued ratio to assets are logically highly correlated but will never be used in the same regression. Apart from this, the highest correlation as that between credit rating and equity volatility. This correlation is only 26.63% and therefore this study does not contain multicollinearity problems.

(31)

28

5. Results

In this chapter I will present my main results. First I will present and give meaning to the empirical results of the used regressions. After that, I will interpret the results in the context of previous mentioned research. Finally I will discuss several implications of the study.

5.1 Regression output

Table 5: CDS spread determinants for European banks

This table looks at the determinants of CDS spreads for CoCos issued by European banks. In columns 1-2 the issue dummy is used as explanatory variable. This dummy variable takes the value 1 if a European bank has issued a contingent convertible bond. It takes the value 0 if no issue has occurred. In columns 3-4 issued ratio to assets are used as explanatory variable. The regressions use monthly data from January 2009 to July 2016. Robust t-statistics are reported in parentheses. Clustered standard errors are used. The variable _cons represents the constant or the intercept. Additionally, the adjusted R-squared (adj. R-sq) and the number of observations (N) are given. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Dependent variable: CDS Spread

(1) (2) (3) (4) issued_dummy -19.31** -19.62** (-2.13) (-2.15) issued_ratio_to_assets2 -9883.4* -9894.2* (-1.81) (-1.84) Volatility 61.22*** 43.43*** 59.49*** 41.40*** (4.03) (4.14) (3.79) (3.58) L.Volatility 31.12*** 28.68*** (3.81) (4.25) Stockprice -5.512*** -3.043*** -4.828*** -2.296*** (-5.53) (-3) (-4.82) (-2.60) L.Stockprice -2.547** -1.798* (-2.35) (-1.81) L.Debttoequityratio 9.590* 9.509* 7.488* 7.438* (1.95) (1.93) (1.86) (1.83) L.Tier1capitalratio 169.7 171.0 -25.39 -25.89 (0.69) (0.70) (-0.12) (-0.14) Creditratingclass 27.53 25.15 31.23* 30.47* (1.33) (1.22) (1.89) (1.82) time -0.301 -0.216 -0.341 -0.309 (-0.86) (-0.62) (-1.21) (-1.12) _cons 76.14 73.06 84.87** 82.26** (1.64) (1.57) (2.08) (2.03) Firm Fixed Effects Yes Yes Yes Yes

N 1717 1717 1717 1717

(32)

29

The descriptive statistics table has shown that in order to run representative regressions to measure the impact of CoCo issuance on CDS spreads, some outliers has to be removed. By removing the outliers, the bias of the regression would be eliminated as much as possible which will increase the R squared. Therefore, Hellenic Bank has been sorted. Also the 10 highest observed CDS spreads and the 10 lowest observed CDS spreads have been removed. Eliminating outliers before the robustness test was necessary to run representative

regressions. Due to both the elimination of the outliers and the unbalanced panel data the total observations used in the regression has decreased to 1717 as can be seen in table 5. Table 5 illustrates the determinants of CDS spreads, respective regression coefficients and related t-statistics.The first column of the table represents regression results of the first equation where the one-month lagged values of volatility and stock prices of the banks are excluded. Moreover, the regression includes the dummy variable of the CoCo issue as the main explanatory variable to test the hypothesis. The second column represents the second equation where both the lagged values of volatility and stock prices are included. The third and fourth column has the same structure as the first two columns. In these regressions the issue ratio to assets is used as the explanatory variable. In all four regressions time effects are included to eliminate general time trend that could bias the results. As can be seen in table 5, the time effects are not significantly related to the CDS spread. However, they show a negative relation with the CDS spread. This means that the observed CDS spreads have had a tendency to decrease over time. In this case this seems reasonable because it the study covers the period of 2009-2016. Due to the financial crisis, the world’s financial system was more fragile in 2009. Since then, the worldwide financial system has stabilized, which declares the lower CDS spreads. The last row of table 5 illustrates the R squared for the regressions. As can be seen, the R squared of the four regressions all approximately equal 0.2. This means that the models are predicting more than 20% of the change in CDS spread. When measuring the effect on CDS spreads, it is hard to contain a high R squared. First of all, in this case the CDS spread and banks’ default risk do not completely correlated. That is, CDS spreads do not fully measure just default probabilities of banks. Second, as mentioned earlier, an infinite number of factors have an influence on CDS spreads, which cannot be fully measured in a regression model. Therefore, the R squared in this models give a representative picture of the effects CoCo issues on CDS spreads.

(33)

30

When testing the earlier stated hypothesis, the most interesting finding from the first column is significant value of the issued dummy. The issue dummy is significantly negative related to the CDS spreads on a 5% significance level. The first column presents a negative coefficient of 19.31, which means that the CDS spread will decrease with 19.31 in the first month after the CoCo has been issued. The first column also shows that the control

variables stock price and equity volatility are highly significant related to the CDS spread at a 1% significance level. The stock price has a negative coefficient of 5.512 which means that the CDS spread will decrease with 5.512 when the stock price of a European bank goes up with €1,-. The equity volatility has a positive coefficient of 61.22 which means that a 1% increase in the volatility will increase the CDS spread with 61.22. The first column also shows that, apart from the debt to equity ratio, other control variables are not statistically

significant. However, the coefficient signs correspond with the expected sign presented in table 2. The debt to equity ratio is statistically significant at a 10% level. However the measured relation between the debt to equity ratio and the CDS spread seems

unreasonable. That is, a relatively higher level of leverage should not lower the risk of a bank. However, debt could have different advantages (e.g. a fiscal advantage), which will decrease the risk profile of the bank (Hilscher and Raviv, 2014). Although this could be a plausible explanation, this result is controversial. In the second column, both one-month lagged values of the equity volatility and the stock price are included. Table 5 shows that the inclusion of those variables does not significantly change the results of the first regression. All other values remain the same. However, the second column shows that both the lagged equity volatility and stock price are statistically significant at a 1% and 5% significance level. This means that stock price level and the level of equity volatility corresponding with the CoCo issue date, both have a significant impact on the CDS spread. Due to the fact that volatility and stock prices could change every day both unlagged and lagged variables have a significant impact on the CDS spread. I have also tested the second lag of both variables, which were not statistically significant. This proves the causal impact of both variables on CDS spreads. The third and fourth regression differs from the first two regressions. That is, those regressions measure the impact of the CoCo issue amount rather than the occurring of a CoCo issue. The reason for using this ratio, rather than the pure amounts, is that the ratio ensures that the issues are relatively better comparable between banks. Therefore the

(34)

31

ratio is a more representative variable for measuring the effect on CDS spreads. The impact on CDS spreads will be way different when relatively small banks with a small asset base will issue a large amount of contingent convertibles then for a large bank with a large asset base. Due to use of the ratio as explanatory variable, columns 3 and 4 of table 5 presents different results. It shows that the issued ratio to assets is statistically significant at a 10% significance level. In first sight, the negative coefficients of 9894.2 and 9883.4 seem odd. Table 3 shows a mean issue amount to assets of 0.003394. Therefore the coefficients could be interpreted as -33.5809 (-9894.2*0.003394) and 33.5442596 (-9883.4 *0.003394). Since the ratios are statistically significant we could state that an increase in the issue amount will decrease the CDS spread. Moreover, regression 3 and 4 presents slightly different results than the regressions 1 and 2. Table 3 shows that the credit rating is significant at a 10% level. That means that if a bank is in a higher category class, the CDS spread will increase. This seems plausible, because higher category classes are related with the less creditworthy banks.

5.2 Interpretation of results

Summarized, the results for the separate regression models support the stated hypothesis of a negative relationship between CoCo issues and the CDS spread. Since both the issued dummy as the issued ratio to assets are statistically significant the conclusion could be drawn that issuing CoCos enhance the stability of issuing banks. This conclusion corresponds with the conclusion of papers of Avdjiev et al. (2015), Ammann et al. (2015) and Hilscher and Raviv (2014). Therefore, the additional layer of protection that contingent convertible bonds offer to creditors could explain the negative CDS spread reactions. This will reduce the probability of government bailouts. Since CoCos reduce the probability of default, the risk drops and therefore the CDS spreads narrow (Ammann et al. 2015, Avdjiev et al. 2015). This additional layer of protection most likely outweighs the additional moral hazard

problem, which arise when CoCo has been issued. (Chan and van Wijnbergen, 2015). Moreover a conclusion could be drawn that the Basel 3 regulations has successfully been implemented in the EU law since their main purpose was to enhance quality of the capital base and decrease the risk of banks.

(35)

32

Furthermore, the results have shown that both the control variables stock price and volatility have a high significant relationship with the CDS spread. This corresponds with prior literature where Chan and van Wijnbergen (2015) have stated that higher stock prices signal a healthier state of the bank, which lowers the credit risk. Abid and Naifar (2006) have illustrated that higher equity volatility reflects higher equity uncertainty, which accompanies higher risks, and thus will be positively related to the CDS spread.

5.3 Limitations

The proposed method and data could have some limitations. Analyzing new capital

instruments as contingent convertibles have the disadvantage of using a limited set of data. Although the dataset contains the entire available issued Cocos up to July 2016, the sample is still small. Therefore I acknowledge that using a larger sample would give more

representative results. Additionally, several measurement points in other used variables were also missing. This leads to unbalanced panel data. Moreover, the study probably omits notable CDS spread factors. By including the most significant control variables presented by prior literature, I have tried to limit the omitted variable bias as much as possible. Finally, the proposed measurement of the effects on the CDS spread is controversial. To my knowledge, this type of study has never been performed in prior academia. This study focus explicitly on the ex post effects on the CDS spreads of European banks. This study measures the CDS spread the month after the CoCo issue. However, CoCo bonds have been issued on different dates. Therefore the used observations contain difference in days between the CoCo issue and the measurement of the CDS spreads between banks. One bank measures the CDS impact 21 days after the issue while the other only measures 8 days after the issue. Due to this variation, this study could contain bias. I expect that the effect on CDS spreads is stronger when the difference in days is smaller. Prior studies have both estimated the ex ante and ex poste announcement effects of CoCos on the CDS spread by carrying out an event study (Avdjiev et al. 2015, Ammann et al. 2015, Schmidt and Azarmi, 2015). Carrying out an event study could eliminate this bias in my study. However, the announcement of the CoCo issuance is not at a clearly defined point in time. An upcoming CoCo issue is not

(36)

33

publically announced at a single point in time. Therefore an event study might not be the best way to measure the effect of CoCo issuance on CDS spreads.

Contingent convertible issues in Europe : what is the impact on CDS spreads of European banks?

Master Thesis