1
Thesis MSc Finance
Faculty of Economics and Business, University of Groningen January, 2018
The CoCo Effect in a Nutshell: A Study about the Influence of Hybrid Capital
Instruments on the Stability of the Issuing Bank1
Abstract:
Contingent Convertible Bonds (CoCos) are hybrid securities that absorb losses when the capital of the issuing bank reaches a threshold level. As one of the first empirical studies, this paper investigates the effect of CoCo issuances on the stability of the bank using two market-based risk indicators. The main findings show that the issuance of CoCos leads to a statistically significant decrease in the CDS spread, indicating that CoCos provide risk reducing advantages. Furthermore, banks that have issued CoCos are related with a lower exposure towards systemic risk, as proxied by the Marginal Expected Shortfall.
JEL classification: E58, G14, G21, G28
Keywords: Contingent Convertible bonds, Capital requirements, Financial stability, Systemic
risk, Risk-shifting incentives
Author: Gibran Borst
Student number: S2407388
Mail: g.l.borst@student.rug.nl
Phone: +316 38245121
Supervisors: Dr. M.A. Lamers University of Groningen Drs. C.J.C. Vos MA De Nederlandsche Bank
2 Table of Contents
Abbreviations 3
1. Introduction 4
2. CoCos and design structures 6
2.1 Trigger event 6
2.2 Loss absorption mechanism 7
2.3 Regulatory classification 8
3. Related literature and hypotheses 10
4. Methodology 12
4.1 The influence of CoCo issuances on the default probability 12 4.2 The influence of CoCo issuances on systemic risk exposure 16
5. Data 17
5.1 The CoCo dataset 17
5.2 Data description for the event study 18
5.3 Data description for the panel study 19
5.4 Control variables for the panel study 19
5.5 Descriptive statistics for the panel study 20
6. Results 23
6.1 The influence of CoCo issuances on the CDS spread 23 6.2 The influence of CoCo issuances on systemic risk exposure 25
6.3 Robustness checks 28
7. Conclusion 28
7.1 Main conclusions 28
7.2 Policy implications 29
7.3 Limitations and future research 29
Bibliography 30
3
Abbreviations
AT1 Additional Tier 1
BCBS Basel Committee on Banking Supervision CAAR Cumulative Average Abnormal Return
CAR Cumulative Abnormal Return
CDS Credit Default Swap CET1 Common Equity Tier 1 CoCo Contingent Convertible Bond
EC Equity Conversion
GDP Gross Domestic Product
G-SIB Global Systemically Important Bank MES Marginal Expected Shortfall
OLS Ordinary Least Squares
PONV Point Of Non-Viability PWD Principal Write-Down RWA Risk Weighted Assets
T2 Tier 2
US United States
4
1. Introduction
During the global financial crisis of 2008 the common approach to rescue global systemically important banks (G-SIBs) was through bailouts in which governments provided capital injections to prevent the bank from liquidation. After the crisis discussion arose about the utilization of tax payers’ money to save financial institutions. Subsequently, in 2010 the Basel Committee on Banking Supervision (BCBS) agreed upon an international framework of post-crisis policy reforms known as Basel III2. This framework addresses stricter minimum capital requirements aiming to increase the quantity and quality in the level of loss absorbing capacity in the banking sector (Basel Committee on Banking Supervision, 2011). Basel III states that
financial institutions can partially satisfy their regulatory capital requirements by issuing contingent convertible bonds (CoCo) which results in an expanding security segment (Avdjiev et al., 2015). Namely, this instrument can qualify as additional tier one (AT1) going-concern capital, or as tier 2 (T2) gone-concern capital (Basel Committee on Banking Supervision, 2011).
AT1 instruments specifically, must provide loss absorbency when the bank still operates as a viable, “going” institution.
A CoCo is a hybrid security which functions as a bond requiring coupon payments under stable circumstances, until a trigger event is breached (European Parliament, 2016). This trigger can be predefined based on a capital threshold or either at the regulators’ discretion when the point of non-viability (PONV) is reached. Subsequently, the CoCo can absorb losses via equity conversion (EC), or a principle write-down (PWD). Therefore, the bank can recapitalize through the CoCo in economic downturn when raising new capital is often difficult for financial institutions (Flannery, 2005). Furthermore, the instrument provides the option to increase the capital buffer at lower funding cost than equity (Boermans and Van Wijnbergen, 2017).
Between 2009 and 2015 the CoCo market expanded from nonexistent to 450 billion dollar of total issuances around the world (Avdjiev et al., 2015). Despite this level, the market is still limited relative to equity and subordinated debt issuances. Therefore it is relevant to assess whether CoCos can contribute to bolster the financial sector. A burgeoning theoretical literature examines the impact of different CoCo characteristics on bank stability which leads to mixed findings. Apart from providing additional loss absorption capacity, these instruments are associated with risk shifting incentives and wealth transfers (Hilscher and Raviv, 2014; Chan and Van Wijnbergen, 2017). However, empirical work is rather scarce. Consequently, the aim of this study is to partially fill the gap by answering the empirical research question:
How does the issuance of different CoCo instruments influence the stability of banks?
In this study, the stability of the bank is measured by two market-based risk indicators. First, event study methodology is applied to measure the impact of CoCo issuances on the credit default swap (CDS) spread, which functions as a proxy for the default probability of the issuing bank following Avdjiev et al. (2015). Second, using panel data, this study assesses if the issuance of CoCos has an influence on the Marginal Expected Shortfall (MES), functioning as
5
a proxy for the exposure towards systemic tail-risk as proposed by Acharya et al. (2012). After analyzing the overall effect of CoCo issuances on these two risk measures, subsamples are made to distinct on the main design elements including the regulatory classification (AT1 or T2 capital) and the loss absorption mechanism (PWD or EC).
This study focuses on CoCo issuances undertaken by banks from advanced economies with active CoCo markets, resulting in a comprehensive dataset of CoCo issues between 2009 to late 2017. As mentioned before, this paper contributes to the existing literature by providing one of the first empirical studies about the effect of CoCo issuances on bank stability3.
The results indicate that the overall effect of CoCo issuances can increase the stability of the issuing bank based on the two market-based risk measures. The effect on the default probability and the exposure towards systemic risk is negative and statistically significant at the 5%, and 10% respectively. More specifically, the issuance of CoCos is related to a reduction in the CDS spread of the bank, implying that the market regards the CoCo as an instrument to shield the more senior debtholders of the bank. However, when one distinguishes on the design elements of CoCos, this result remains statistically significant for PWD CoCos whereas no statistical evidence is provided for EC CoCos. Furthermore, the reduction in the CDS spread is statistically significant for CoCos qualifying as AT1 instruments, but not for T2 issuances. The results of the panel analysis indicate that some statistical evidence is found for a linkage between a CoCo issue and a reduction in the MES. However, when one discriminates on the design elements, the results only remain statistically significant for PWD CoCo issuances and instruments that qualify as T2 capital. Furthermore, the size of the issuance does not have an impact on the MES.
From a regulatory perspective, this empirical study is relevant to assess whether different CoCo structures lead to the desirable effects where these are initially intended for. This study captures how financial markets react towards CoCo issuing banks depending on the structure of the instrument which is difficult to predict ex ante. Debate exists in the theoretical literature about the possible negative effects of PWD CoCos, however the findings in this empirical study show that these CoCos can increase the stability of the bank.
The remainder of this study is organized as follows. Section 2 describes the various design structures of CoCos and its current role in the regulatory Basel III framework. Section 3 provides an overview of the relevant literature about the impact of CoCos on bank stability. In addition, the hypotheses are defined. The methodology is discussed in section 4, followed up with a description of the data in section 5. Section 6 discusses the empirical results of this paper. Section 7 concludes.
6
2. CoCos and design structures
This section starts by introducing the concept of CoCos, in which its main design features are explained, containing the trigger event and the loss absorption mechanism. Additionally, the debates under scholars about the CoCo design are provided. Subsequently, the function of CoCos in the current regulatory framework Basel III is discussed.
2.1 Trigger event
The CoCo market established in 2009 by the first CoCo issuance of Lloyds Bank even before Basel III was implemented (EU Parliament, 2016). The current low yield environment make CoCos an attractive instrument to yield seeking investors. CoCos namely, provide higher coupon payments to investors than other types of debt (Jaworski et al. 2016). However, "There
ain't no such thing as a free lunch" as Milton Friedman would say. CoCos are the most risky
bonds that a bank can issue.
CoCos are hybrid debt securities providing loss absorbency when a certain trigger event is breached. At this point, the balance sheet of the financial institution will automatically deleverage and the capital level will rise through a debt-to-equity swap of the CoCo (Hilscher and Raviv, 2014). The trigger event thus refers to the point at which the loss absorption mechanism will be activated. This can either occur mechanically when a predetermined minimum capital ratio is reached and on the discretionary judgement of the supervisory authority. The first trigger of two AT1 CoCos with a combined face value of 1.25 billion euros occurred in June 2017 when Banco Popular entered into resolution4.
The mechanical trigger is based on a fixed percentage of the book value of Common Equity Tier 1 (CET1) divided by the risk weighted assets (RWA) of the bank. In contrast to this accounting based trigger, some of the literature has stressed the need for market-based triggers (Flannery, 2005, 2016; Calomiris and Herring, 2013). They argue that market prices of equity are more accurate in reflecting the contemporaneous status of the underlying institution, which enables a continuous monitoring process. The book value of equity can imperfectly represent information due to subjective interpretation of accounting values. In addition, the inconsistency of the internal models results to variability in the calculation of the RWA between banks. This provides banks with the opportunity to modify the trigger moment. Furthermore, the dependency on discrete reporting intervals can result in a considerable time lag. Accurately monitoring the bank is especially relevant when the institution encounters financial difficulties. Therefore, CoCos based on book-value triggers are probably converting too late, when the bank already faces funding challenges (Flannery, 2014). However, there are certain drawbacks associated with market-based triggers. Hillion and Vermaelen (2004) have shown that incentives can arise for short selling the stock of the bank in distress, resulting in a downward pressure on stock price. This so called death-spiral increases the probability of a trigger event and raises funding difficulties for the institution (Flannery, 2016). Reaching the trigger event
7
can be beneficial for investors when a wealth transfer between CoCo holders and shareholders will be formed (Sundaresan and Wang, 2015). Furthermore, a market based trigger can be imperfect because a trigger event can occur when additional capital is not necessarily required.
Apart from a mechanical trigger, the trigger can also hit on a discretionary base when the supervising authority judges that the point of non-viability (PONV) of the financial institution is reached (Avdjiev et al., 2015). The PONV, which is not specifically defined under the Basel III standards, refers to the moment at which the bank is no longer considered to continue as an autonomous viable entity.
2.2 Loss absorption mechanism
The Basel III framework states that the loss absorption mechanism of CoCos can be distinguished into two types namely, EC or via a PWD (Basel Committee on Banking Supervision, 2011). This mechanism specifies how losses are absorbed when a trigger event has been breached.
In case a EC CoCo converts into equity, the amount of common shares will increase by the amount of the converted CoCo instrument. Then, the conversion rate determines to how many common shares the bond is converted at this trigger event. Note that the conversion rate depends on the contractual conversion share price. Although the conversion rate can be based on a predefined share price, this rate is not always known ex ante the trigger is breached. The reason is that the conversion can also be based on the market share price at the day of conversion. However, conversion based on the market price contemporaneous to the trigger event can result in considerable ownership dilution of the existing shareholders. This is caused by market share prices which are often particularly low in distressed periods (Flannery, 2016). For this reason, the conversion rate can also be based on the market price including a minimum floor to prevent excessive dilution.
A complex element of EC CoCos is the existence of potential wealth transfers between CoCo holders and initial shareholders when a trigger event occurs. Namely, if the conversion price is below the contemporaneous market share price then a wealth transfer arises from the initial shareholders to the CoCo holders. Conversely, a conversion price which is higher than the market share price transfers wealth towards the initial shareholders (Flannery, 2014).
8
partial principal write-down as loss absorption mechanism. However, attention arose about the contractual possibility of a cash payout to the CoCo holder of the remaining value after a trigger event. The concern is that the bank might end up in liquidity shortages when a cash payout should occur in a period of distress (Avdjiev et al., 2013).
Figure 1. An overview of the CoCo design
This figure presents the main design elements of CoCos
Source: Avdjiev et al. (2013) 2.3 Regulatory classification
Under the Basel III framework, CoCos can either contribute to fulfill the AT1 or T2 regulatory capital requirements of financial institutions (Basel Committee on Banking Supervision, 2011). These layers of minimum capital requirements complement the core capital of the bank which is defined as Common Equity Tier 1 (CET1). The purpose is that instruments qualifying as CET1 capital, provide the highest level of loss absorbing capacity. The capital requirements are expressed in terms of the percentage of the RWA held by the financial institution. This risk weight is based on the riskiness of the underlying asset with the rationale that higher capital cushions are held for riskier activities.
According Basel III, CoCos can, on top of 4.5% CET1 qualify up to 1.5% of RWA to account as AT1 capital. In addition, CoCos can suffice the minimum T2 capital requirement of 2% of RWA. An illustration of the stacking order of these capital layers is provided in Figure 2. Since AT1 capital must provide a higher ability to absorb losses than T2 capital, the requirements for a CoCo to qualify as AT1 instrument are stricter. The mechanical trigger level of AT1 CoCos is generally higher and has a required floor set at a CET1 to RWA ratio of 5.125% (Basel Committee on Banking Supervision, 2017b). The higher trigger level is related to higher related funding costs since the probability of conversion increases. The rationale for this minimum level is that the AT1 CoCo can function as a loss absorbing instrument while the financial institution still operates as a going concern. In contrast, T2 CoCos can either be triggered in
going concern or gone concern dependent on the trigger level. Going concern capital refers to
9
capital is needed for a situation in which the institution reaches, or is close to default (Jaworski et al., 2017). Therefore, there are two different authorities in place to monitor the institution. The supervisory authority has the objective preserve the bank as a going concern and the resolution authority is responsible for managing the process of the bank which is in default or “gone”.
Figure 2. CoCos in the regulatory framework under Basel III
This figure provides a simplified illustration of the role of CoCos in the capital requirements framework under the standards of Basel III (Basel Committee on Banking Supervision, 2011). Note that both AT1 and T2 capital requirements may also be fulfilled with CET1 capital. Furthermore, the capital add-ons including the capital conservation buffer of 2.5%, the G-SIB surcharge of 2.5% and countercyclical buffer between 0-2.5%, that must be met with CET1 capital are not included in this example.
Source: Avdjiev et al. (2013)
Another distinctive aspect is that AT1 CoCos are perpetual and callable after five years, where T2 instruments are required to have a fixed maturity of at least five years. Furthermore, the coupons on AT1 CoCos may voluntarily be canceled upon the decision of the supervising authority or the financial institution without consequences (Basel Committee on Banking Supervision, 2011). This makes AT1 CoCos more similar to common equity.
10
3. Related literature and hypotheses
This section provides a discussion on the most related studies how CoCo issues affect the stability of the bank. By doing so, the theoretical literature is discussed first, related to the rationale to issue CoCos and the potential risk shifting incentives that can arise. In addition, the hypotheses are defined in combination with a reflection on the most related empirical study.
With the reverse convertible debenture, Flannery (2005) has introduced the predecessor of the CoCo as a tool to bolster the viability of financial institutions. The main function of a CoCo is similar, namely to provide more loss absorbing capacity and potential capital cushions in circumstances of distress when banks normally faces difficulties with equity issuances (Jaworski et al., 2017). Therefore, CoCos are able to function as a tool to automatically deleverage the balance sheet when the bank still operates as a going concern. Nevertheless, CoCos can also contribute to an orderly resolution process for a bank as a gone concern (Chen et al., 2017).
Compared to issuing new equity, CoCos do not dilute the shareholders’ ownership of the institution ex ante a trigger event. Furthermore, since CoCos are regarded as bonds in most jurisdictions, banks can deduct the coupon payments from pre-tax earnings which increases the tax shield. As a result, CoCos are attractive instruments to comply with regulatory capital requirements (Avdjiev et al., 2013). Flannery (2016) argues that compared to equity financing, the instrument will reduce incentives to disinvest in safe activities with lower returns since the funding costs are lower. CoCos therefore combine the cost efficient advantage of debt financing in financial good states and the potential loss bearing capacity in periods of financial distress.
11
However, the theoretical paper of Hilscher and Raviv (2014) is based on the assumption that shareholders have a direct influence on the risk level of the bank. Furthermore, an increase in the riskiness of the assets will generally require higher regulatory capital requirements which will dampen the return on equity. This will reduce the marginal benefits of higher risk levels. In addition, if shareholders are assumed to have a direct influence on the activities of the bank, it is unclear why they would permit CoCo issuances that have unfavorable effects upon a trigger event. In contrast, practice shows that investors still want to invest in CoCos that provide benefits for shareholders as discussed in the literature.
The issuance of CoCos increases the loss bearing capacity which is expecting to reduce the probability of default of the bank. Nevertheless, according the theoretical literature the design of the CoCo seems to be relevant due to potential risk shifting incentives that can arise (Calomiris and Herring, 2013). An increase in risk level due to the issuance of CoCos has a deteriorating effect on bank stability. These ambiguous effects therefore provide relevance for an empirical analysis. As a result the first hypothesis is:
Hypothesis 1: The issuance of CoCos changes the default probability of the issuing bank.
The most related study stems from 2015, in which Avdjiev et al. found a statistically significant negative relationship between CoCo issuances and the CDS spread. Their dataset contains a sample of 72 CoCos issuances in the first years after the CoCo market established in 2009. The event study is based on CoCos issued by banks in advanced economies with the exception of the euro area periphery5. The intuition is that the increase in loss absorbing capacity provided by CoCos, increases the market value of higher ranked debt instruments due to a decrease in the probability of default. A stronger effect is found for EC CoCos than for PWD CoCos. Furthermore, the negative effect on the CDS spread became weaker when the deal value of the CoCo increased, suggesting that CoCos with a smaller size are more effective. In addition, they found no effects of CoCo issuances on the stock price of the issuing bank.
The study of Avdjiev et al. (2015) provides room for improvement. In their study, no distinction is made between the regulatory tier of the CoCo. As discussed in section 2.3, substantial differences between AT1 and T2 CoCos are present. AT1 instruments are intended to provide loss absorbency when the bank still functions as a going concern. Furthermore, their sample of 72 CoCos is limited due to the emergence of a new asset class during their sample period. Therefore it is too early to generalize on their findings. The market has developed in the last years, parallel with the regulation that stimulates the issuance of CoCos. As a result, a more representative analysis can be made about the issuance effects on the default probability.
Although event study methodology is effective to analyze the direct effects of CoCo issuances, it is at least as important to analyze the longer term effects on bank stability. In particular, it is meaningful to know if financial institutions that issued CoCos are indeed more resilient towards shocks due to increased loss absorbing capacity. According to Vallée (2016), banks in Europe that have issued convertible bonds benefited from the conversion during the global financial
12
crisis of 2008. These banks had the possibility to recapitalize in a period in which raising new equity was difficult.
Chen et al. (2017) found that CoCos can reduce the debt overhang problem, which refers to the unwillingness of shareholders to issue new equity in critical circumstances since most of the benefits will be reaped by the debt holders. If CoCos are designed as such that shareholder will refrain the occurrence of the trigger event, then the shareholders will maintain the capital to an adequate level. Avoiding a trigger event seems rational since a trigger may send a bad signal to the market about the quality of the balance sheet. As shown by Chan and Van Wijnbergen (2014) a well-capitalized bank can still reach insolvency after a trigger event due to liquidity shortages caused by a bank run.
Overall, it is still unclear how banks perform once CoCos have been issued. Risk-shifting incentives can increase the vulnerability of the institution, whereas CoCos are able to provide stability in financial distressed periods when equity refinancing is difficult. Therefore, second hypothesis is:
Hypothesis 2: The issuance of CoCos changes the exposure towards systemic risk.
As discussed above, most of the existing literature focuses on the theoretical aspects of CoCos while limited empirical research has been disclosed. This paper contributes by providing a better understanding if and in which manner CoCo issuances can enhance the stability of the bank.
4. Methodology
This section describes the methodology of the empirical analysis in two parts. First, the methodology of the event study is described in which the effect of a CoCo issuance on the CDS spread is measured. Second, the panel regression model is provided to measure if CoCo issuances affects the systemic risk exposure of the issuing bank.
4.1 The influence of CoCo issuances on the default probability
The aim of the first hypothesis is to test whether the issuance of CoCos can change the default probability of the issuing bank. The CDS spread is selected an appropriate market based proxy for the default probability since it reflects the insurance premium in basis points for the exposure towards fixed income securities issued by the institution (Vallée, 2016; Avdjiev et al., 2015). Additionally, the CDS spread is a market-based risk measure which has the advantage of a forward looking approach as compared to a static accounting-based risk measure (Flannery, 2016). Consequently, if a CoCo issuance changes the default probability of the issuing bank, it is expected to be reflected in the CDS spread.
13
to examine the influence of new economic information on the value of a security (MacKinlay, 1997). New relevant economic information of an institution will be reflected in security prices at the short notice. The event study conducted in this research is based on the methodology of James (1987) and MacKinlay (1997). Furthermore, Avdjiev et al. (2015) later adopted this methodology to measure the impact of CoCo issuances on the abnormal returns of the CDS spread.
The timetable in figure 3 contains an estimation window and an event window. During the estimation window the normal return can be observed. This window should contain a substantial number of days to measure the daily changes in CDS spreads under normal circumstances. Following MacKinlay (1997) an estimation window W1 from T0 to T1 containing 120 business
days is appropriate. The aim is to test whether the returns in the event window are abnormal with respect to the normal return.
Figure 3. The timeline of the event window
This figure depicts the timeline in the event study including the estimation window W1 containing 120 trading
days and the event window W2 consists of 15 trading days. The event at T3 represents the issuance day of the
CoCo.
The event window contains an anticipation period, from T2 to T3, the event date on which the
CoCo is issued at T3, and a post issuance market reaction between T3 to T4. The anticipation
period starts when potential investors are informed about the pricing of the CoCo and the trading book will be written. Accordingly, we expect that information about the CoCo issuance of bank is partly reflected in the CDS spread before the CoCo is issued at T3. The time interval
of the anticipation period has to be set in such that the time period between the pricing date and the issuance date is captured. Based on the sample of CoCo issuances, the average number of days between the pricing and the issuance is 9 trading days. In addition, Avdjiev et al. (2015) argue that it takes some time before the market has access to information about the issuance. Therefore the market reaction is continuing to be reflected in the CDS spread a few days after the issuance. At T4 we expect that all the information about the issuance is reflected in the CDS
spread. Therefore, the event window is set at 15 business days starting at T3-9 and ending at T3
+5.
To detect a potential abnormal return on the CDS spread during the event window, the estimated normal return derived from the estimation window is subtracted from the actual return during the event window. The abnormal return on the CDS spread for bank i on day t can be calculated as:
14
Where 𝐴𝑅𝑖𝑡, 𝑅𝑖𝑡 and 𝐸(𝑅𝑖𝑡|𝑋𝑡) are the abnormal, the actual and the estimated normal returns on the CDS spread respectively. 𝑋𝑡 contains the conditioning information for the normal return model.
There are several models to estimate the normal returns. In this study, the Market Model is applied to estimate the normal returns since it is widely used in event studies (James, 1987; MacKinlay, 1997). MacKinlay (1997) argues that more advanced models are generally not effective to decrease the variance of the abnormal returns. In addition, Avdjiev et al. (2015) has shown that the Market Model is appropriate to apply on CDS spreads. From equation 2, the normal return can be derived by applying the Market Model which is given by the following equation:
𝑅𝑖𝑡 = 𝛼𝑖 + 𝛽𝑖𝑅𝑚𝑡+ 𝜀𝑖𝑡 (2)
In which the normal return 𝑅𝑖𝑡 for each CDS spread is estimated via an Ordinary Least Squares (OLS) regression over the estimation window. The normal return is based on a constant factor 𝛼𝑖 and the beta 𝛽𝑖 multiplied by the return on the corresponding market index 𝑅𝑚𝑡 on the respective day t and the addition of the zero mean error term 𝜀𝑖. Here 𝛽𝑖 measures the systematic sensitivity of the CDS spread of bank i towards the market index. The market index is based on a regional CDS index, described in the following data section 5.2. Furthermore, the daily changes in the CDS spreads are calculated using logarithmic returns.
Accordingly, the abnormal return in the event window can be calculated following equation 3.
𝐴𝑅𝑖𝑡 = 𝑅𝑖𝑡− (𝛼̂ + 𝛽̂𝑖 𝑖𝑅𝑚𝑡) (3)
Here, the abnormal return 𝐴𝑅𝑖𝑡 represents the difference between the actual return in the event window 𝑅𝑖𝑡 for bank i on trading day t and the estimated normal return (𝛼̂ + 𝛽̂𝑖 𝑖𝑅𝑚𝑡).
To analyze the average effect of CoCo issuances on the CDS spread during the event window, the average abnormal return can be calculated. The equation of the average abnormal return on day t during the event window is obtained by the average of all abnormal returns related to 𝑁 CoCo issuances in the sample which is given as:
𝐴𝐴𝑅𝑡 = 1
𝑁∑ 𝐴𝑅𝑖𝑡
𝑁
𝑖=1
(4)
Accordingly, the cumulative average abnormal return can be generated to detect the overall effect during the event window by taking the sum of the average abnormal returns.
𝐶𝐴𝐴𝑅(𝑇2,𝑇4) = ∑ 𝐴𝐴𝑅𝑡
𝑇4
𝑡=𝑇2
15
In order to test for the statistical significance of the abnormal returns, the returns have to be standardized. The standardized cumulative abnormal return of bank i with an event window starting at T2 and ending at T4 is given as:
𝑆𝐶𝐴𝑅𝑖 = ∑ 𝐴𝑅𝑖𝑡 𝑆𝑖𝑡 𝑇4 𝑡=𝑇2 (6)
In which the standard error 𝑆𝑖𝑡 is corrected for the number of days in the event and estimation window respectively and corrected for the variance in the market return during the event window (James, 1987). The equation is given as:
𝑆𝑖𝑡 = [𝑇𝑉𝑖2[1 + 1 𝑀+ (𝑅𝑚𝑡 − 𝑅̅𝑚)2 ∑𝑀 (𝑅𝑚𝑗 − 𝑅̅𝑚)2 𝑗=1 ]] 1/2 (7)
In which 𝑆𝑖𝑡 is the standard error for bank i on day t, 𝑉𝑖2 is the residual variance in the estimation window, 𝑇 stands for the number of days in the event window and 𝑀 for the number of days in the estimation window. 𝑅𝑚𝑡 is the market return during day t in the event window and 𝑅̅𝑚 represents the mean market return over the estimation window. 𝑅𝑚𝑗 is market return of day j in the estimation period.
Now the standardized cumulative average abnormal return during the total event window of all issuances in the sample can be derived where 𝑁 represents the number of CoCo issuances in the sample. 𝑆𝐶𝐴𝐴𝑅 = 1 𝑁∑ 𝑆𝐶𝐴𝑅𝑖 𝑁 𝑖=1 (8)
Assuming the individual abnormal returns are independently distributed, the Z-statistic with a normal distribution N(0,1) can be calculated as:
𝑍 = √𝑁(𝑆𝐶𝐴𝐴𝑅) (9)
16
4.2 The influence of CoCo issuances on systemic risk exposure
The aim of the second hypothesis is to test whether the issuance of a CoCo can change the systemic risk exposure of the bank in the year after the CoCo has been issued. Furthermore, the analysis is extended by estimating if the relative issuance amount is of influence.
In order to conduct the analysis, an OLS regression will be performed using a panel dataset including banks that issue CoCos and banks that do not. Based on the study of Acharya et al. (2012) the Marginal Expected Shortfall (MES) is the dependent variable of the model as a proxy for systemic risk exposure. The tail risk measure is computable with public market data and is widely used in the literature see e.g. De Jonghe et al., (2015) and Idier et al., (2014). In equation 10, the MES is presented for bank i at time t given the quantile Q = 5%.
𝑀𝐸𝑆𝑖𝑡(𝑄) = −𝐸[𝑅𝑖𝑡|𝑅𝑚𝑡< 𝑉𝑎𝑅𝑚𝑡 𝑄
] (10)
In the model 𝑅𝑖𝑡 represents the daily stock (log) return of bank i at time t. 𝑅𝑚𝑡 represents the
(log) return on a regional banking sector index at time t conditional on the threshold level Q bound to the Value at Risk (VaR). In words, the MES can be explained as the average return of bank i at time t conditional on the five percent lowest returns of the market index. The MES is calculated on a yearly basis. Additionally, the risk measure is multiplied by -1 such that higher values of the variable indicate a higher risk level.
The general empirical model to be estimated is represented by equation 11. All independent variables have a one year lag to prevent reverse causality.
𝑀𝐸𝑆𝑖𝑡 = 𝛽0+ 𝛽1𝐶𝑜𝐶𝑜𝑖𝑡−1+ 𝛽2𝑋𝑖𝑡−1+ 𝛼𝑖 + 𝛾𝑡+ 𝜀𝑖𝑡 (11)
Where 𝐶𝑜𝐶𝑜𝑖𝑡−1 is a dummy variable which equals to one if at least one CoCo has been issued by bank i in year t-1 and equals to zero for the years of no issuances. 𝛼𝑖 is the time-invariant fixed or random effect, 𝛾𝑡 is the year fixed effect and 𝜀𝑖𝑡 is the error term. 𝑋𝑖𝑡−1 represents a vector for the country and bank specific control variables which are described in the data section 5.4.
17
The advantage of a panel study is that is enables to observe variability in the MES across sections and time. Furthermore, including banks in the dataset that do not issue CoCos at all, enables reliable comparisons. To correct for possible unobserved heterogeneity, a fixed effects or random effects model will be used. The selection for the specific model is based on a Hausman test. The null hypothesis of the Hausman test states that the error term is uncorrelated with the explanatory variables which is assumed under a random effects model. Conversely, the alternative hypothesis states that the error term is correlated with the explanatory variables which opts for a fixed effects model. In addition, to correct for potential heteroskedasticity, cross-section white standard errors are used in the regression estimation. Furthermore, to mitigate potential effects of outliers, all independent variables apart from the dummy variables, are winsorized at the 1st and 99th percentile.
5. Data
This section briefly describes the data used. First, information is provided about the dataset containing the CoCo issuances that are used throughout this study. Subsequently, the data requirements are described for the event study and the panel study respectively. In addition, the control variables and the summary statistics for the panel study are discussed.
5.1 The CoCo dataset
Data on CoCo issuances is gathered from Dealogic and contains the main contractual elements namely: the pricing date, the issuance date, the issuance amount, the issuance currency, the regulatory classification as AT1 or T2 and the loss absorbency either via a PWD or EC. Table_1 provides the total sample of CoCos issued by banks including the distribution per country and region. The dataset contains 382 CoCos with an aggregated deal value of more than 216 billion euros issued in the period between December 2009 and September 2017. The interest of financial institutions to issue CoCos arose significantly from 2013. An explanation for this rise is that banks have to comply with the implementation of the Basel III guidelines in their particular jurisdictions (Basel Committee on Banking Supervision, 2011). The majority of the CoCos are issued by listed banks, although some CoCos issued by unlisted banks are incorporated in the event study. Most CoCos in the sample are issued in Europe. Furthermore, the distribution of CoCo issuances in terms of the regulatory tier and the loss absorption mechanism is quite balanced. However, as can be seen in the table, for two issuances the regulatory tier is not provided, and for 27 CoCos the loss absorbency is not defined.
18
Table 1. CoCo issuances by banks in the total region of interest between the period 2009-2017.
Regulatory Tier Loss Absorption
Country Issuances AT1 T2 PWD EC Amount (billion €)
Australia 60 23 37 23 40 33.72 Austria 3 3 - 1 - 0.52 Belgium 2 1 1 1 - 2.15 Canada 42 26 14 2 38 21.58 Denmark 5 4 1 4 - 3.00 France 17 15 2 14 - 16.77 Germany 6 6 - 4 - 12.05 Hong Kong 6 3 3 6 - 2.39 Ireland 5 3 2 2 3 2.38 Italy 6 6 - 4 - 5.55 Netherlands 10 9 1 5 3 13.89 New Zealand 3 1 2 1 2 0.94 Norway 4 4 - 3 1 1.78 South Korea 34 8 26 32 - 6.21 Spain 11 10 1 - 9 10.73 Sweden 13 13 - 11 1 8.67 Switzerland 22 14 8 15 4 27.03 Taiwan 53 15 38 50 - 2.58 United Kingdom 80 29 51 4 72 44.33 Total 382 193 187 182 173 216.27 Region Americas 42 26 14 2 38 21.58 Asia Pacific 156 50 106 112 42 45.84 Europe 184 117 67 68 93 148.85 Total 382 193 187 182 173 216.27
5.2 Data description for the event study
19
CoCos issued by Canadian banks are not included in the event study, since liquid data on CDS spreads is unavailable. In addition, the Markit iTraxx CDS market index for Europe is only available from 2012 onwards. Nevertheless, this causes only a limited reduction in the sample. One limitation is the availability of liquid CDS data for the banks. Since the event study is based on daily changes, the effect of CoCos issued by banks with illiquid CDS spreads cannot be observed. This introduces a bias since larger financial institutions often have more liquid CDS spreads. Since no CoCos are issued in the Asian Pacific region before 2012, the total sample consists out of 174 CoCo issuances between 2012 and 2017.
5.3 Data description for the panel study
The panel study requires bank-specific balance sheet data and daily share prices. Annual consolidated balance sheet information and regional stock market indices are retrieved from SNL Financial. The regional stock market indices to derive the MES, the dependent variable, include the Euro Stoxx Banking Index for Europe, SNL TSX bank index for Canadian banks, and the SNL Asia Pacific ex Japan bank index for banks located in Asia Pacific. Furthermore, the dataset includes both banks that issue CoCos, and banks that do not. Banks are included with a minimum threshold value of 1.5 billion Euros in total assets for the year 2015 according the SNL database. This results in an unbalanced panel dataset from the period 2009-2016 including 131 publicly listed banks. In addition, the independent variables in the regression model have a one year lag with respect to the dependent variable resulting in the observation of 253 CoCo issuances between 2009 and 20156. Finally, because this study has an international scope, macro-economic data at the country level is obtained from the IMF World Economic Outlook database. The macro-economic data is assigned to the bank based on the provided home country according the SNL Financial database.
5.4 Control variables for the panel study
In order to be more certain that the variability in the MES is accurately explained, several control variables are introduced. The selection of the control variables is based on the commonly applied bank performance indicators used in the academic literature see e.g. De Jonghe et al., 2015; Laeven and Levine 2009; Demirgüç-Kunt and Huizinga, 2013. The first control variable is bank size, as measured by the natural logarithm of total assets to adjust for normality. Mainly the largest institutions, which are associated as too-big-to-fail, are associated with a higher systemic risk exposure (De Jonghe et al., 2015). As a second control variable, the capital ratio is added, which is measured by the total capital divided by the RWA. Well capitalized banks are generally more resistant towards systemic shocks. The profitability of the bank is measured by the return on average equity. The cost-to-income ratio controls for the operational efficiency of the bank. Liquidity is measured by the liquid assets to total assets. The quality of the loan portfolio and credit risk is proxied by the ratio of non-performing loans to total loans. Furthermore, the deposits to total assets ratio indicates to what extend the bank depends on deposit funding. The total loans to total assets ratio is added as an indicator of the
20
retail activities. In addition, macro-economic control variables at the country level are incorporated since the economic condition of a country can influence the stability in the financial sector. The size of the economy is measured by the natural log of the gross domestic product (GDP). Contagion risk may be greater for a small countries, since the loss of confidence in one institution can lead to distrust of the system more easily. Inflation, which can deteriorate the value of an asset portfolio, is measured by the yearly consumer price index inflation rate. The GDP growth is the inflation-adjusted growth in an economy.
5.5 Descriptive statistics for the panel study
21
Table 2. Descriptive statistics for the panel study
This table depicts the descriptive statistics for all variables used in the model. The full sample contains yearly data over the period 2009-2016 for 131 banks. All independent variables have a lag of one year with respect to the dependent variable. The MES, the dependent variable in the model, is a market based measure for systemic risk exposure. A higher percentage of the MES indicates a higher risk exposure. The CoCo variables contain dummy variables which are equal to 1 if the specific type of CoCo is issued by the bank in a particular year. Furthermore, the issuance amount divided by total equity enables to measure the impact of the CoCo size. All independent variables are winsorized at the 1st and 99th percentile.
Variable N Mean Std. Dev. Min Max
Dependent variable MES (in %) 1183 2.287 2.260 -4.521 16.423 CoCo variables CoCo dummy 1376 0.084 0.277 0 1 AT1 dummy 1376 0.062 0.241 0 1 T2 dummy 1376 0.037 0.189 0 1 PWD dummy 1376 0.044 0.204 0 1 EC dummy 1376 0.041 0.199 0 1
CoCo to Equity (in %) 1376 0.428 1.841 0 20.166
AT1 to Equity (in %) 1376 0.289 1.514 0 20.166
T2 to Equity (in %) 1376 0.134 0.926 0 14.842
PWD to Equity (in %) 1376 0.208 1.383 0 20.166
EC to Equity (in %) 1376 0.176 1.110 0 19.334
Bank variables
Ln(Total Assets) (in thousands €) 1327 17.539 1.880 13.707 21.414
Capital ratio (in %) 1269 14.836 3.995 9.215 32.193
Return on Equity (in %) 1306 5.717 13.874 -83.834 31.114
Cost-to-Income (in %) 1319 61.593 15.064 28.062 116.235
Liquid Assets to Assets (in %) 1142 29.965 18.023 5.526 92.998
NPL ratio (in %) 1179 5.043 7.433 0.064 42.126
Deposits to Assets (in %) 1326 59.519 19.182 10.323 90.244
Loans to Assets (in %) 1318 60.042 19.005 4.302 90.760
Country variables
Inflation (in %) 1183 1.311 1.357 -1.140 4.463
GDP Growth (in %) 1183 1.260 2.884 -5.568 10.631
22
Table 3. Correlation statistics of the control variables and the MES
This table indicates the correlation statistics for all control variables used in the model including the dependent variable. The MES as the dependent variable in the model, is a market proxy for systemic risk exposure. A higher value of the MES indicates a higher sensitivity of the bank towards systemic risk. LN(TA) is the natural logarithm of total assets (in thousands €) to control for size. CAP is the total capital divided by the risk weighted assets as a proxy for the capital ratio. ROE is the return on average equity controlling for profitability. CTI is the cost to operating income ratio, whereas LIQ denotes the liquid assets to total assets to control for liquidity. NPL is the ratio of non-performing loans to the total loans to control for the quality of the loan portfolio. DEP controls for the deposit to total asset ratio, LOAN controls for the loans to total asset ratio. The macroeconomic variables INF, GDPG and LN(GDP) stand for inflation, growth in GDP and the size of the economy as the natural logarithm of GDP (in billion €). All independent variables have a one year lag and are winsorized at the 1st and 99th percentile. The significance levels at 1%, 5% and 10% are indicated by ***, ** and * respectively
Variable MES LN(TA) CAP ROE CTI LIQUID NPL DEP LOAN INFL GDPG LN(GDP)
23
6. Results
This section presents the empirical results in this study. Section 6.1 starts with the results of the impact of CoCo issuances on the CDS spread, applying an event study methodology. Thereafter, section 6.2 continues with the results for the panel regressions measuring the impact of CoCo issuances on the systemic risk exposure. Section 6.3 provides additional robustness checks
6.1 The influence of CoCo issuances on the CDS spread
Table 4 reports the results of the CoCo issuances on the CDS spread derived by the Market Model7. Starting with a total sample of 174 CoCos issued between 2012 and late 2017, the cumulative average abnormal return (CAAR) for the total event window of 15 trading days is -1.185% and statistically significant at the 5% significance level. Figure 4 plots the CAAR including the 95% confidence interval the during the complete event window. The graph shows that the CDS spread is decreasing in the specified period towards the issuance date. In addition, it can be observed that the CDS spread of the issuing bank decreases after the issuance has taken place. Hence, the CoCo issue is negatively related to the default probability of an institution. Therefore one could argue that the increase in loss absorbing capacity dominates the potential rise in risk-taking incentives.
In addition to the general effect of CoCo issuances on the CDS spread, one can discriminate among the design elements. As discussed in section 3, the design of the loss absorption mechanism, especially PWD CoCos, can induce risk-shifting incentives of the shareholders. In addition, AT1 CoCos have a higher capital ranking providing banks more flexibility than T2 instruments. Therefore, table 4 also presents the results for subsamples based on the contractual loss absorption mechanism and the regulatory tier of the CoCo. Although the CAAR on the CDS spread is negative for all CoCos in the subsamples, the results are not all statistically significant. Neither the reduction in the CDS spread for EC CoCos, nor for T2 classified instruments is statistically significant. In contrast, CoCo issuances with a contractual PWD and instruments which are classified as AT1 capital, show evidence for a reduction in the CDS spread, significant at the 10% level. The loss in significance when discriminating among the conversion mechanism and regulatory capital classification may be attributable to the decreased sample size.
To some extend these findings are not in line with the outcomes of Avdjiev et al. (2015). Although the overall findings are similar, they find that EC CoCo issuances do result in a statistically significant decrease in the issuers’ CDS spread, whereas PWD CoCos do not. However, there are several potential explanations to address these controversies. The sample from Avdjiev et al. (2015) only consists of 72 CoCos, issued between September 2009 and March 2015. Furthermore, their geographic scope deviates from our study since the authors included countries from Latin America and excluded south European countries.
24
Table 4 The influence of CoCo issuances on the CDS spread
This table reports the cumulative average abnormal returns (CAAR) in percentage derived from the Market Model for various samples. All 174 CoCos in the sample are issued in the period between 2012-2017. The event window of 15 days starts at t-9 and ends at t+5, where t indicates the day of issuance. The event window starts 9 days before the issuance date to capture the market reaction between the pricing date and the issuance date. The null-hypothesis states that the CAAR is equal to zero. The significance of the CAAR is measured following the methodology of James (1987). The significance levels at 1%, 5% and 10% are indicated by ***, ** and * respectively based on a two-sided test.
The 15 Day Event Window (-9, +5)
CAAR Z-statistic P-value Sample Size
All CoCos -1.185 -2.161** 0.031 174
Loss Absorption EC -0.688 -0.698 0.485 67
PWD -1.358 -1.670* 0.095 80
Regulatory Tier AT1 -0.848 -1.667* 0.096 128
T2 -2.028 -1.425 0.154 46
Figure 4. The cumulative average abnormal return of the CDS spreads for the total sample
This figure plots the cumulative average abnormal return (CAAR) derived from the Market Model in percentage on the 5-year senior unsecured CDS spread of the issuing bank for the full sample of 174 CoCos issued between 2012 and 2017. The 95 percent confidence interval is indicated by the dotted lines. The event window contains 15 trading days starting at t-9 and ending at t+5, where t indicates the respective day of the CoCo issuance.
-2,50 -2,00 -1,50 -1,00 -0,50 0,00 0,50 1,00 t-9 t-8 t-7 t-6 t-5 t-4 t-3 t-2 t-1 t t+1 t+2 t+3 t+4 t+5 C A A R on the C D S spr ea ds (i n % )
25
6.2 The influence of CoCo issuances on systemic risk exposure
The outcomes of the panel regression analyses are presented in table 5 where the independent variable of interest is a dummy variable which equals one if at least one CoCo is issued by the bank in a given year. The dependent variable is the MES, as a proxy for systemic risk exposure. In all regressions (1-6), bank fixed effects and year dummies are included. The preference for a fixed effects model is based on the Hausman test. This test with the null hypothesis of no significant difference between the fixed effects and random effects model is rejected at a 1% significance level. Furthermore, the standard errors are corrected for heteroskedasticity and all independent variables have a lag of one year. In addition, a constant is added in each regression but not shown in the table. The regressions 1-3 capture the total sample of banks whereas the regressions 4-6 are duplicates for European banks only. From a regulatory perspective, the inclusion of the subsample of European banks has relevance to observe if the outcomes deviate from the total sample. In regressions 1 and 4, no distinction is made on design elements of the CoCo meaning that the aggregate effect of a CoCo issuance on the MES is measured. The regressions 2 and 4 discriminate on the regulatory tier of the CoCo, where regressions 3 and 6 distinct on the loss absorption mechanism.
Starting with the results in regression 1, the relationship between a CoCo issuance and the MES is negative and statistically significant at a 10% significance level. Since a higher value of the MES indicates a higher systemic risk exposure of the given bank, the result provides evidence that the issuance of a CoCo is associated with lower tail risk exposure in the subsequent year. To be more precise, the average decrease in the MES is 0.33 percentage point for banks that have issued CoCos. Most coefficients on the control variables depict intuitive signs. The coefficient of the bank size is statistically significant and positively related with the MES. In addition the coefficient of non-performing loans is also significant and positive as expected. These results are also in line with other studies (De Jonghe et al., 2015; Idier et al., 2014). The current model is able to explain 36.4% of the variance in the MES as indicated by the adjusted R-squared. In regression 2 which distinguishes on the regulatory tier of the CoCo, the coefficients for AT1 and T2 dummy variables are both negative. Nevertheless, the result is only statistically significant at the 1% significance level for CoCos that qualify as Tier 2 instruments. The issuance of these instruments leads to an average reduction in the MES of 0.56 percentage point in the year after the CoCo is issued. Note however that the coefficients of AT1 and T2 issuances are not statistically different from each other as indicated by the F-statistic of 1.60.
26
Table 5. The effect of the CoCo issuance on the MES as systemic risk measure
This table reports the results of the OLS regression analysis measuring the effect of the CoCo issuance, using dummy variables, on the dependent variable MES, a measure for systemic risk exposure. The sample contains data over the period 2009-2016 for 131 banks. All regressions include a constant, bank fixed effects as well as year dummies. Control variables are described in Table 5.4. All independent variables have a one year lag and are winsorized at the 1st and 99th percentile. The significance levels at 1%, 5% and 10% are indicated by ***, ** and * respectively. The standard errors, corrected for heteroskedasticity, are reported in the parentheses. The regressions 1,2,3 capture the total sample, whereas the regressions 4,5,6 include European banks only.
Variable (1) (2) (3) (4) (5) (6)
Dummy CoCo Issuance (-1) -0.330* -0.437
(0.192) (0.278)
Dummy AT1 Issuance (-1) -0.122 -0.303
(0.220) (0.297) Dummy T2 Issuance (-1) -0.561*** -0.801 (0.206) (0.493) Dummy PWD Issuance (-1) -0.549** -0.625** (0.217) (0.314) Dummy EC Issuance (-1) -0.030 -0.216 (0.266) (0.421) Ln(Total Assets) (-1) 1.391*** 1.474*** 1.402*** 1.615** 1.648** 1.636** (0.463) (0.480) (0.460) (0.639) (0.645) (0.634) Capital Ratio (-1) 0.010 0.007 0.009 0.036 0.034 0.034 (0.038) (0.039) (0.039) (0.044) (0.043) (0.044) Return on Equity (-1) -0.005 -0.005 -0.005 -0.010 -0.010 -0.010 (0.018) (0.018) (0.018) (0.019) (0.019) (0.019) Cost-to-Income (-1) 0.005 0.006 0.005 0.003 0.004 0.003 (0.018) (0.018) (0.018) (0.023) (0.023) (0.023)
Liquid Asset Ratio (-1) 0.008 0.007 0.008 -0.001 -0.001 -0.001
(0.014) (0.014) (0.014) (0.019) (0.019) (0.019) Non-Performing Loans (-1) 0.071** 0.074** 0.071** 0.067** 0.068** 0.066** (0.029) (0.029) (0.029) (0.033) (0.032) (0.033) Deposit Ratio (-1) -0.001 -0.001 -0.002 0.003 0.003 0.003 (0.016) (0.016) (0.017) (0.024) (0.024) (0.024) Loans to Assets (-1) 0.020 0.019 0.020 0.013 0.013 0.013 (0.016) (0.015) (0.016) (0.022) (0.022) (0.022) Inflation (-1) 0.012 0.019 0.013 0.007 0.002 0.014 (0.096) (0.095) (0.094) (0.147) (0.147) (0.146) GDP growth (-1) 0.131*** 0.125*** 0.127*** 0.179** 0.179** 0.181** (0.046) (0.046) (0.043) (0.073) (0.074) (0.071) Ln(GDP) (-1) -2.466** -2.416** -2.439** -3.511* -3.568* -3.587* (1.107) (1.123) (1.105) (1.987) (2.008) (1.947)
Fixed Effects YES YES YES YES YES YES
Year Dummies YES YES YES YES YES YES
Observations 742 742 742 469 469 469
Banks 131 131 131 86 86 86
F-statistic AT1 = T2 1.586 0.616
F-statistic PWD = EC 2.542 0.701
27
Table 6 reports the results of the regressions which measure the influence of the relative issuance size on the MES. Fixed effects and year dummies, and all control variables, although not shown, are included in the regressions.
In the regressions 1-3 which includes the full sample of banks, the signs on the coefficients are all negative, indicating that a higher issuance size relative to the existing equity value, decreases the MES. However, these results are not statistically significant. The results for the European subsample depict similar signs on the coefficients. The results show only statistical evidence for T2 CoCo issuances in regression 5. The coefficient is negative and statistically significant at the 5% level. It indicates that a one percent increase in the issuance amount of the CoCo relative to the equity level of the bank reduces the MES with 0.14 percentage point.
Table 6. The effect of the CoCo issuance amount on the MES as systemic risk measure
This table reports the results of the OLS regression analysis measuring the effect of the CoCo issuance amount relative to the equity value, on the dependent variable MES, a market based measure for systemic risk exposure. Bank fixed effects, year dummies, control variables as well as an intercept are included in all columns. All independent variables have a one year lag and are winsorized at the 1st and 99th percentile. The significance levels at 1%, 5% and 10% are indicated by ***, ** and * respectively. The standard errors, corrected for heteroskedasticity, are reported in the parentheses. The regressions 1,2,3 capture the total sample, whereas regressions 4,5,6 include European banks only.
Variable (1) (2) (3) (4) (5) (6)
CoCo to Equity ratio (-1) -0.025 -0.068
(0.034) (0.046)
AT1 to Equity ratio (-1) -0.029 -0.063
(0.046) (0.047) T2 to Equity ratio (-1) -0.023 -0.140** (0.035) (0.057) PWD to Equity ratio (-1) -0.009 -0.043 (0.033) (0.039) EC to Equity ratio (-1) -0.028 -0.098 (0.078) (0.099)
Fixed effects YES YES YES YES YES YES
Year dummies YES YES YES YES YES YES
Control variables YES YES YES YES YES YES
Observations 742 742 742 469 469 469 Banks 131 131 131 86 86 86 F-statistic AT1 = T2 0.011 1.618 F-statistic PWD = EC 0.053 0.287 F-statistic 13.349*** 12.960*** 12.737*** 9.495*** 9.970*** 8.831*** R-squared 0.377 0.377 0.376 0.400 0.400 0.399 Adjusted R-squared 0.361 0.361 0.360 0.376 0.375 0.374
28
often difficult (Flannery, 2014). Therefore, from a regulatory perspective CoCos are instruments that can mitigate the countercyclical effects. Although this empirical study is conducted over a time period with relatively stable economic conditions, an early indication is provided that the benefits of CoCos are visible. Additionally, it is worthwhile to point the negative relationship between the issuance of PWD CoCos and the systemic risk exposure of the bank which conflicts the arguments of Hilscher and Raviv (2014) and Chan and Van Wijnbergen (2017).
6.3 Robustness checks
Additional analyses are performed to check the robustness of the results. For the event study, the impact of the CoCo issuance on the CDS spread has been measured by two models. Namely, a comparison is made between the outcomes derived by the Market Model and the Constant Mean Return Model leading qualitatively to similar results. In addition, the event window is changed to verify the consistency of the coefficients. The results remain statistically significant although an event window of only two trading days, capturing the issuance date and the day after the issuance does not result in a statistical significant impact on the CDS spread. This suggest that most of the impact is priced in the anticipation period and post issuance period which is in line with Avdjiev et al. (2015) who apply an event window of 21 days. An extension of the event window can be found in the appendix B, which portraits comparable outcomes but an even higher significance level. Additionally, a multivariate regression is performed to gather a better understanding if bank and CoCo characteristics have an impact on the magnitude of the cumulative abnormal return during the event window. The outcomes are also reported in the appendix B. Based on these outcomes, the impact on the CDS spread diminishes when the issuance amount relative to the risk weighted assets increases, which aligns the findings of Avdjiev et al. (2015). However, no statistical evidence is found for differences in impact between EC and PWD, or AT1 and T2 CoCo issuances on the cumulative abnormal return. Interestingly, the CDS spread seems to decrease more if smaller banks issue CoCos.
7. Conclusion
7.1 Main conclusions29
7.2 Policy implications
In the recent years CoCos have gained popularity as a funding instrument among financial institutions. The demand side for this risky assets class is abundant caused by yield-seeking investors in the current low interest rate environment. Nevertheless, the intension of policy makers is not only to enhance stability of induvial institutions, but also the financial sector as a whole. CoCos are still complex instruments as a consequence of considerable variation in designs and a lack of transparency. Progress can be made by standardization of the instrument through limiting the loss absorption mechanism to a PWD. Compared to EC CoCos, the expected loss for PWD CoCos is less uncertain for CoCo investors upon a trigger event which simplifies the risk management. The complex element associated with EC CoCos is the conversion rate. As a consequence, EC CoCos create tension between shareholders and CoCo investors since a wealth transfer in any direction tends to be inevitable at conversion8. Transparency can be improved upon providing more disclosure of the CET1 ratios of banks. This can improve the monitoring process for CoCo investors and therefore enhance the market discipline. On the other hand, better insights about the background and purposes of CoCo investors is desirable for regulating authorities to measure the potential impact of a trigger event on the real economy. In addition, the minimum mechanical trigger level for AT1 instruments could be reconsidered. The current level of 5.125% CET1 is argued to come too late to rescue an institution as a going-concern.
7.3 Limitations and future research
This study contains some limitations. The sample size of the CoCos in the event study reduced substantially due to illiquid or non-available CDS spreads. The loss in significance when discriminating among the conversion mechanism and regulatory capital classification may be attributable to the decreased sample size. Additionally, some banks that have issued multiple CoCos during the period of interest, can have an over representing impact on the findings. Furthermore, no data was available about the trigger levels of CoCos. The height of the trigger may have an impact on the results. The dependent variable used in the panel study is the MES as an indicator for systemic risk exposure. However, note that the analysis does not give insights into the impact CoCos can have in true systemic crisis events, which can be seen as the left-tail of the left-tail. Therefore, the economic inference of the results should be taken with caution. The study focused on the relatively short term effects of CoCos on bank stability. One reason is the restriction to issuance data of CoCos only. By using sophisticated balance sheet data on CoCos one can give attention to the longer term effects of these instruments on bank stability.
Future research should focus on investigating the grounds why banks do issue certain types of CoCos. Is it simply based on market demand or do shareholders have strategic preferences for specific design elements? Furthermore, an important open question is the effect on the stability of the bank when CoCos are triggered. In addition, contrasting to the findings of Chan and Van Wijnbergen (2017), the results in this study suggest that PWD CoCos can enhance the stability of the bank based on the two risk measures used. Further research should give attention to the potential benefits of this relatively simplistic CoCo design.
30
Bibliography
Acharya, V, Pedersen, L., Philippon, T., Richardson, M., 2012. Measuring systemic risk. Centre for Economic Policy Research Discussion paper No. 8824, London.
Avdjiev, S., Kartasheva, A., Bogdanova, B., 2013. CoCos: a primer. BIS Quarterly Review. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2326334.
Avdjiev, S., Bolton, P., Jiang, W., Kartasheva A., Bogdanova, B., 2015. CoCo Bond Issuance and Bank Funding Costs. BIS and Columbia University working paper.
Basel Committee on Banking Supervision, 2017a. Basel III: Finalising post-crisis reforms. Available at: https://www.bis.org/bcbs/publ/d424.htm
Basel Committee on Banking Supervision, 2017b. Basel III definition of capital – Frequently asked questions. Available at: https://www.bis.org/bcbs/publ/d417.pdf
Basel Committee on Banking Supervision, 2011. Basel III: A global regulatory framework for more resilient banks and banking systems. Available at: https://www.bis.org/publ/bcbs189.htm
Barth, A., Schnabel, I., 2013. Why banks are not too big to fail–evidence from the CDS market. Economic Policy 28, 335-369.
Boermans, M., Van Wijnbergen, S., 2017. Contingent convertible bonds: Who invests in European CoCos? Applied Economics Letters, 1-5.
Breitenfellner, B., Wagner, N., 2012. Explaining aggregate credit default swap spreads, International Review of Financial Analysis 22, 18-29.
Calomiris, C., Herring, R., 2013. How to Design a Contingent Convertible Debt Requirement That Helps Solve Our Too‐ Big‐ to‐ Fail Problem. Journal of Applied Corporate Finance 25, 39-62.
Chan, S., Van Wijnbergen, S., 2014. Cocos, contagion and systemic risk. Tinbergen Institute Discussion Paper, Amsterdam.
Chan, S., Van Wijnbergen, S., 2017. CoCo Design, Risk Shifting Incentives and Financial Fragility. European Capital Markets Institute 2, 1-50.
Chen, N., Glasserman, P., Nouri, B., & Pelger, M., 2017. Contingent capital, tail risk, and debt- induced collapse. The Review of Financial Studies 30, 3921-3969.
De Jonghe, O., Diepstraten, M., & Schepens, G., 2015. Banks’ size, scope and systemic risk: What role for conflicts of interest? Journal of Banking & Finance 61, 3-13.
Demirgüç-Kunt, A., Huizinga, H., 2013. Are banks too big to fail or too big to save? International evidence from equity prices and CDS spreads. Journal of Banking & Finance 37, 875-894.
EU Parliament, 2016. Contingent convertible securities Is a storm brewing? Available at: europarl.europa.eu/RegData/etudes/BRIE/2016/582011/EPRS_BRI(2016)582011_EN.p df
Flannery, M., 2005. No pain, no gain: Effecting market discipline via "reverse convertible debentures". Capital Adequacy beyond Basel: Banking, Securities, and Insurance Oxford University Press, 1-24.
Flannery M., 2014. Contingent Capital Instruments for Large Financial Institutions: A Review of the Literature. Annual Review of Financial Economics 6, 225-240.
Flannery, M., 2016. Stabilizing large financial institutions with contingent capital certificates. Quarterly Journal of Finance 6, 1-26.
