BIS Working Papers
CoCo issuance and bank fragility
by Stefan Avdjiev, Bilyana Bogdanova, Patrick Bolton, Wei Jiang and Anastasia Kartasheva
Monetary and Economic Department
JEL classification: G01, G21, G28, G32
Keywords: CoCos, Contingent Convertible Capital, Bank Capital Regulation, Basel III
BIS Working Papers are written by members of the Monetary and Economic Department of the Bank for International Settlements, and from time to time by other economists, and are published by the Bank. The papers are on subjects of topical interest and are technical in character. The views expressed in them are those of their authors and not necessarily the views of the BIS.
This publication is available on the BIS website (www.bis.org).
© Bank for International Settlements 2017. All rights reserved. Brief excerpts may be reproduced or translated provided the source is stated.
ISSN 1020-0959 (print)
ISSN 1682-7678 (online)
CoCo Issuance and Bank Fragility
Stefan Avdjiev Bilyana Bogdanova Patrick Bolton Wei Jiang Anastasia Kartasheva
The promise of contingent convertible capital securities (CoCos) as a ëbail-inísolution has been the subject of considerable theoretical analysis and debate, but little is known about their e§ects in practice. In this paper, we undertake the Örst comprehensive empirical analysis of bank CoCo issues, a market segment that comprises over 730 instruments totaling $521 billion. Four main Öndings emerge: 1) The propensity to issue a CoCo is higher for larger and better-capitalized banks; 2) CoCo issues result in statistically signiffcant declines in issuersíCDS spreads, indicating that they generate risk-reduction beneÖts and lower costs of debt. This is especially true for CoCos that: i) convert into equity, ii) have mechanical triggers, iii) are classiffed as Additional Tier 1 instruments; 3) CoCos with only discretionary triggers do not have a signiffcant impact on CDS spreads; 4) CoCo issues have no statistically signiffcant impact on stock prices, except for principal write-down CoCos with a high trigger level, which have a positive effect.
Avdjiev, Kartasheva, and Bogdanova are a¢ liated with the Bank for International Settlements, Basel, Switzerland; Bolton and Jiang are a¢ liated with Columbia Business School, New York, U.S.A. The study has beneÖted from discussions by Felipe Cortes, Mark Flannery, Florian Heider, George Pennacchi, Klaus Schaeck, Theo Vermaelen, and comments from Ruth Doubleday, Louise Eggett, Ingo Fender, Ulf Lewrick, Noel Reynolds, and Kostas Tsatsaronis, as well as seminar participants at the American Economic Association, Bank for International Settlements, Deutsche Bundesbank, the Committee on the Global Financial System workshop on contingent capital instruments, Conference on Financial Regulation at the University of Chicago, De Nederlandsche Bank, European Finance Association, European Winter Finance Conference, Imperial College of London, NBER Risks of Financial Institutions, Research Task Force of the Basel Committee on Banking Supervision, Western Finance Association, University of Geneva, and University of St. Gallen. We thank Minchen Zheng (Columbia) for excellent research assistance. Avdjiev can be reached at Stefan.Avdjiev@bis.org; Bogdanova at Bilyana.Bogdanova@bis.org; Bolton at firstname.lastname@example.org; Jiang at email@example.com; and Kartasheva at Anastasia.Kartasheva@bis.org. The views expressed in the paper are those of the authors and do not necessarily represent the views of the Bank for International Settlements.
If there is one term that epitomizes the infamy of the Global Financial Crisis of 2007-09 it is “too big to fail”. Even when they sought to impose financial discipline on banks that had assumed excessive levels of leverage, financial regulators were more often than not forced to accept bailouts as the only possible stabilizing intervention because the alternative of putting the failed banks through some form of resolution and imposing losses on depositors and other bank creditors was seen to be too dangerous in the middle of the crisis. It is not surprising, therefore, that a major priority of regulators following the crisis, on top of significantly tightened bank capital and liquidity regulations, was to create new instruments to facilitate the write-down of the debt of distressed institutions. In short, the priority became to make sure that, in the event of another crisis, bail-ins would be a credible alternative to bailouts.
One way of bringing about a swift and seamless bail-in is through the conversion of contingent convertible capital securities (CoCos) previously issued by the bank. The automatic loss absorption by CoCos when a pre-specified trigger has been breached was the main appeal of these securities according to the early proponents, in particular, Flannery (2005; Raviv (2004); Duffle (2009); McDonald (2013); Coffee (2011); Pennacchi, Vermaelen and Wolff (2014); and the Squam Lake Working Group (2009). These advocates of CoCos disagreed on the design of CoCos, in particular on whether the trigger should be a stock price floor or a minimum equity-capital ratio, and their proposals were criticized by other commentators, most notably Admati, DeMarzo, Hellwig and Pfleiderer (2013) and Sundare- san and Wang (2015), who argued that CoCos were excessively complex and their conversion could have destabilizing effects.
In the immediate aftermath of the Global Financial Crisis, these debates among academics were mirrored by differences in opinions and preferences among national bank regulators. Nevertheless, the introduction of the Basel III framework, which allows banks to meet part of their regulatory capital requirements with CoCo instruments, created strong incentives for banks to seriously explore the possibilities for issuing CoCos. Furthermore, the inclusion of specific contingent capital elements in the Basel III framework brought some homogeneity in CoCo design across jurisdictions. As the number of jurisdictions implementing Basel III grew, banks responded by raising a substantial amount of capital in the form of CoCo issues. Between January 2009 and December 2015, banks around the world issued a total of $521 billion in CoCos through 731 different issues.
The market for bank CoCos is now large enough to undertake a systematic empirical analysis of bank CoCo issuance. We do this by assembling the first comprehensive data set of all CoCo issues by banks between 2009 and 2015. Beyond providing an overview of the main characteristics of CoCos and their issuers, our study seeks to address two fundamental questions: What drives bank CoCo issuance decisions? And how do CoCo issues affect issuers’ balance sheets?
To determine the factors driving CoCo issuance, we perform a duration analysis and estimate which bank characteristics are associated with a higher propensity to issue a CoCo. Our first main broad finding is that larger banks and banks with relatively strong balance sheets, were among the first wave of CoCo issuers. But banks with impaired balance sheets that were in greater need of recapitalization were less likely to issue CoCos. Even though this result is surprising at first sight, it is consistent with the predictions of our simple theoretical model (presented in Appendix B). Issuance decisions are made with shareholder interests in mind; and shareholders may pass on a CoCo issue if it does not enhance shareholder value. As our theoretical model predicts, a CoCo issue by a bank with an impaired balance sheet mostly benefits the bank’s senior unsecured debt holders (a version of the classic "debt overhang" problem). Viewed in this light, our result is less surprising. Consistent with this interpretation, we also find that statistical significance is greater for CoCos that specify a principal write-down when the trigger is breached rather than a conversion of debt into equity. Principal write-down CoCos are, in effect, junior to equity in distress states and, therefore, particularly attractive to shareholders.
We also conduct an event study to evaluate the impact of CoCo issuance on the issuer’s balance sheet. In particular, we estimate the announcement effect of a CoCo issue on the issuer’s CDS spreads and equity prices, breaking the effect down by key issuer characteristics and CoCo contract features. Several striking results emerge from this analysis. First, the overall impact of a CoCo issue on the issuer’s CDS spread is negative and strongly statistically significant, indicating that, according to market participants, a CoCo issue unambiguously strengthens the issuer’s balance sheet. When we break down the effect by CoCo loss absorption mechanism, we find that the impact on CDS spreads of CoCos that convert into equity is much stronger than the respective impact of principal write-down CoCos. This latter result is again consistent with the predictions of our simple model. As we show, a
CoCo that converts into equity disciplines risk-taking by shareholders because conversion may dilute existing equity holders’ claims. In contrast, risk-taking is rewarded at the conversion margin when CoCos absorb losses via a principal write-down. This explains why the net reduction in credit risk following the issue of a principal write-down CoCos is lower.
When we further separate the announcement effects by the type of trigger mechanism, we strikingly find that the presence of a mechanical trigger makes all the difference. More concretely, only CoCos that have a mechanical trigger (in addition to a discretionary trigger) have a significant negative impact on the issuer’s CDS spread. In other words, the issuance of a CoCo with only a discretionary trigger is seen to have no significant effect on the bank’s credit risk by the market. The most likely explanation for this market reaction is that these CoCos are pure "gone concern" instruments, with a lot of uncertainty surrounding the regulator’s conversion decision. CoCos that also have a mechanical trigger combine both “gone concern” and “going concern” features. The market reaction to the announcement of CoCo issues with a mechanical trigger suggests that the triggering event for these CoCos is more predictable, at least for the "going concern" part.
We also find that the level of the mechanical trigger matters. The impact of high-trigger CoCos on CDS spreads is stronger than that for low-trigger CoCos. These results shed light on an important controversy around banks’ true motivation for issuing CoCos. Ever since the first CoCo issue by Lloyd’s in November 2009, which specified a low trigger, the concern was raised that the contingent component of these instruments was purely theoretical and that regulatory intervention to save the issuer would take place long before the low trigger level would be breached (see Alessandri and Haldane, 2009 and Glasserman and Perotti. Our results only partially confirm these concerns and indicate that, at least for hightrigger CoCos, market expectations do factor in a significant probability of conversion. It is also worth noting that these results also partially dispel another myth about CoCos: that investors were only focused on the high yields and entirely ignored the risk of conversion.
All in all, we find that CoCo issuance has generally contributed to reducing bank fragility. But not all instruments are equal in this regard. The CoCos that have the strongest strengthening effect on bank balance sheets are the ones that absorb losses by converting to equity as well as those that have a mechanical trigger and an Additional Tier 1 (AT1) designation.1
We also look at the announcement effects on bank equity prices and find that for the full sample there is no significant impact of CoCo issuance on the issuer’s stock price. This is again consistent with the predictions of our theoretical model and reflects the fact that while a CoCo issue may strengthen a bank’s balance sheet it may do so at the behest of shareholders. Interestingly, however, the one type of CoCo whose main design features are most beneficial to shareholders - a principal write-down CoCo with a high trigger - does have a positive and statistically significant impact on the issuer’s equity price.
Our paper contributes to the wider literature of empirical studies of financial contracts (see Roberts and Sufi, 2009, for a survey). Because our financial contracts are publicly traded we are able to shed light on market reactions to particular designs, which is typically not possible for venture capital contracts and syndicated bank loans.
None of the CoCos in our sample have converted or been in danger of converting during our sample period. We therefore cannot investigate the effects of conversion.2 However, the study by Vallee (2016) of hybrid bonds issued by European banks between 1998 and 2012, which have similar features to CoCos structured as reverse convertible bonds, shows how conversion can provide a capital cushion to issuing banks in a crisis. Vallee (2016) studies the effects of the ex-post conversion of these hybrid bonds during the financial crisis of 2007-09 and finds that conversion had the intended effects predicted by CoCo theories.
The remainder of the paper is organized as follows. Section 2 provides the institutional background, describes the current state of CoCo issuance, and presents an overview of the sample. Section 3 describes the post-crisis policy debates around bank regulatory reform. It also summarizes the predictions derived from the theoretical framework developed in Appendix B. Section 4 presents the empirical analyses. Section 4.1 contains several duration analyses on the propensity of a bank to issue a CoCo. Section 4.2 reports the estimates of the impact of CoCo issuance on the CDS spread and the stock price of the issuing bank, as well as the differential effects of the main CoCo contract features. Section 5 discusses open
questions about the future design of CoCos and offers concluding comments.
2 Institutional Background and Sample Overview
2.1 The CoCo Market
CoCos have two defining features -a trigger modifying the debt repayment terms and a loss- absorption mechanism. We describe the different forms these features can take in Figure 1.
A CoCo can have one or more triggers. In the case of multiple triggers, the loss absorption mechanism is activated when any trigger is breached. Under Basel III, all CoCos are discretionary trigger. This trigger type allows regulators to activate the loss absorption mechanism if they decide that the issuer has reached the point of non-viability. In addition, some CoCos also have a mechanical trigger, which is defined relative to the capital of the CoCo-issuing bank. In theory, the capital measure could be based on book values or market values. In practice, however, the mechanical triggers of all the CoCos that have been issued so far are book value-based. In most cases, those are defined in terms of the ratio of the issuing bank’s Common Equity Tier 1 (CET1) capital to its risk-weighted assets (RWA).
[Insert Figure 1 here.]
The loss-absorbing mechanism is the second key characteristic of a CoCo. Recapitalization through conversion can occur in two ways. A mandatory conversion into equity (MC) CoCo increases CET1 capital by converting the CoCo debt into equity at a pre-defined conversion rate. The conversion rate can be based on the market price of the stock at the time the trigger is breached, a pre-specified price (e.g. the stock price at issuance), or a combination of the two prices. Principal write-down (PWD) CoCos repair the bank’s balance sheet by writing down the CoCo’s principal amount, either permanently or temporarily.3
The Basel III framework contains two key contingent capital elements (Figure 2).4 First, all (ATI and T2) CoCos must include a discretionary trigger, also known as a point of non-viability (PONV) trigger. Second, all ATI CoCos classified as liabilities must have a mechanical trigger, with a minimum trigger level of 5.125% (in terms of CET1/RWA).5 However, the Basel III framework does not specify a particular loss absorption mechanism, implicitly treating PWD CoCos and MC CoCos in a symmetric fashion.6
[Insert Figure 2 here.]
More recently, two other regulatory changes have further expanded the potential market footprint of bank CoCos. The first change is the upcoming total loss-absorbing capacity (TLAC) resolution requirements for global systemically important banks (G-SIBs). The new TLAC requirements can be (at least partially) satisfied through the issuance of CoCos as well as other (non-CoCo) "bail-in-able" debt.8 The second change was introduced by the European Bank Recovery and Resolution Directive (BRRD), which includes the Minimum Requirement for Own Funds and Eligible Liabilities (MREL) stipulating that a sufficient fraction of bank debt liabilities be bail-in-able.8
Nearly as important as bank regulations are institutional investors’ constraints and issuer incentives to create notable clientele effects on CoCo design. In particular, the persistently low interest rates in the post-financial crisis environment incentivized fixed income investors to invest in high-yield instruments. However, fixed income investors are often restricted by mandate from investing in instruments that have a non-negligible probability of converting to equity or face higher compliance costs for these instruments. In contrast, PWD instruments typically do not fall under such restrictions.
1 The implicit assumption behind the above conclusions (and throughout our analysis) is that CoCo issuance is not substituting for CET1 issuance, i.e. that CoCos are issued on top of (rather than instead of) CET1.
2 Outside our sample period, there has been one episode in which coupon suspension was feared for Deutsche Bank’s CoCos in February 2016, and another episode in which conversion of Banco Popular’s CoCos was triggered by the regulator in June 2017.
3 Avdjiev et al. (2013) provide further details on CoCo contract features and bank capital regulations related to CoCos.
4 As illustrated in Figure 2, the regulatory minimum capital requirements under the Basel III framework are (RWA-based) ratios of 4.5% for CET1, 6% for Tier 1 capital (CET1 + ATI), and 8% for total capital (CET1 + AT1 + T2). All instruments other than those in CET1 are optional -i.e. banks have the option to meet all their capital requirements using only CET1.
5 The latter requirement does not apply to AT1 CoCos classified as equity (i.e. preferred shares).
6 In some jurisdictions (e.g. Switzerland), national supervisors supplement the Basel III minimum capital levels by requiring G-SIB banks to have additional capital layers, filled with CoCos. See FINMA (2015).
7 The TLAC standard defines a minimum requirement for the instruments and liabilities that should be "bail-in-able" (i.e. readily available for bail-in within resolution) at G-SIBs. See FSB (2015).
8 See EBA (2016).
2.2 Sample Overview
Our data set is compiled using information from Bloomberg and Dealogic. We have manually augmented the data set with additional information from supplementary sources. The sample that we use in our benchmark analysis consists exclusively of CoCos issued by banks and excludes those issued by insurance companies and other non-bank financial institutions. Furthermore, we focus on CoCos that have at least one (mechanical or discretionary) contractual trigger.9
The data comprises the main elements of CoCo term sheets, their regulatory classification, and key issuer characteristics. We collect information on the most important CoCo contractual terms - the loss absorption mechanism (principal write-down or mandatory conversion to equity), the trigger type (mechanical or discretionary), and the trigger level (for
CoCos that have a mechanical trigger). In addition, we collect the following variables: maturity; issued amount; currency of denomination; regulatory capital classification (AT1 or T2); issue date; price and yield at issuance; coupon rate; and payment rank classification (subordinated debt, preferred equity, etc.).
As for issuer characteristics, we collect daily data on issuer equity prices and CDS spreads (for senior unsecured debt). We construct benchmark indices in which banks are grouped based on their country/region of operations - “Australia and New Zealand,” “Canada”, “Europe,” and “Japan.” For equity prices, we use market capitalization-weighted Datastream bank country/regional equity indices. For CDS spreads, we calculate equally-weighted averages of the CDS spreads of banks operating in the respective county/regions. In order to obtain a more representative set of banks included in each country/regional CDS index, we augment the constituents of the Markit iTraxx and CDX indices with additional banks, which (i) operate in the respective CoCo-issuing jurisdictions and (ii) have liquid CDS for senior unsecured debt.10
For the empirical exercise on the determinants of CoCo issuance, we take the universe of potential CoCo issuers to be the top 500 banks (by total assets in 2009) in CoCo-issuing advanced economies. This sample encompasses both those banks which eventually do issue one or more CoCo securities and those banks for which CoCo is a potential option, but choose not to issue. We obtain annual data of the major balance-sheet variables for the actual and potential CoCo issuers from Bankscope.
An overview of the CoCo market through the end of 2015 is presented in Table 1. Between 2009 and 2015, there were 731 CoCo issuances for a combined issuance volume of $521 billion. The size of the average CoCo issue is $713 million. Nevertheless, that figure conceals a wide variation in individual CoCo issue sizes, which range from $2 million to $7 billion.
[Insert Table 1 here.]
The last column of Table 1 provides a break-down of CoCo issuance by region. European issuers account for 39% of the CoCo market. Another 14% of CoCo issuance has come from banks from non-European advanced economies. Following a period of rapid growth in CoCo issuance in the last years, emerging market economies (EMEs) account for over 46% of the CoCo market by the end of 2015.
The left-hand panel of Figure 3 displays the geographical distribution of CoCo issuance by individual countries. In the early years of our sample, CoCo issuance was heavily concentrated in advanced European countries; issuers headquartered in the UK and Switzerland were particularly active. In the later years of our sample, CoCo issuance in non-European advanced economies (Australia, Japan, and Canada) gradually picked up in tandem with the implementation of Basel III. Issuance by banks in EMEs remained relatively low until the last couple of years in our sample, when Chinese banks became particularly active issuers of CoCos. One notable absence from the list of CoCo-issuing jurisdictions is the United States, where banks have issued AT1 and T2 instruments that have only statutory (as opposed to contractual) triggers.11
[Insert Figure 3 here.]
Banks tend to issue CoCos either in their respective home currency or in US dollars (Figure 3, right-hand panel). Virtually all CoCos denominated in British pounds, Australian dollars, Japanese yen, Canadian dollars, and Chinese yuan were issued by banks from the respective currency-issuing jurisdictions. Also, roughly two thirds of the $84 billion in euro- denominated CoCos were issued by banks from the euro area. The US dollar-denominated issuance totaled $148 billion and accounted for 28% of global issuance.
The majority of the CoCos in our sample have a mechanical trigger in addition to the discretionary trigger (present in all CoCos in our sample). The total volume of CoCos with a mechanical trigger is $292 billion (or 56% of the global aggregate), and the amount of CoCos with only a discretionary trigger is $229 billion (or 44% of the global aggregate). The majority of CoCos issued by banks in advanced economies tend to have mechanical triggers.
The majority ($205 billion) of the CoCos with a mechanical trigger have trigger levels that do not exceed 5.125%, which is the minimum trigger level (in terms of CET1/RWA) required for a CoCo classified as a liability to qualify as AT1 capital under Basel III. About $87 billion worth of CoCos have a mechanical trigger level that is higher than 5.125%.12
Slightly more than half (55%) of all CoCos are classified as AT1 capital. The rest are classified as T2 capital. AT1 CoCo issuance has dominated in most advanced economies, where nearly three quarters of the CoCos are classified as AT1 capital. The only exceptions are Switzerland and Canada (where issuance has been roughly evenly split between AT1 and T2 CoCos) and Japan (where T2 CoCos have accounted for two thirds of all issuance). In contrast, EME banks have issued mostly T2 CoCos, which represent 63% of their total CoCo issuance so far.
Figure 4 tracks the evolution of the CoCo market between 2009 and 2015. It shows that the volume of CoCo issuance has grown rapidly since 2009. The left-hand panel reveals that, in the early years, issuance of AT1 instruments was slightly more prevalent than that of T2 instruments. As issuance by EME banks picked up in 2014 and 2015, so too did the share of T2 CoCos.
[Insert Figure 4 here.]
The right-hand panel of Figure 4 breaks down the evolution of CoCo issuance by the loss absorption mechanism. Even though MC CoCos dominated in the early years, the issuance of PWD instruments picked up over time. Growing demand by fixed income investors for CoCos and shareholder incentives to issue PWD CoCos could be important factors that contributed to this trend.
9 In this paper, we focus on loss absorption mechanism triggers and we do not analyze coupon cancellation triggers. We also do not include CoCos that have only a statutory trigger.
10 All the additional banks that we include in our augmented country/regional CDS indices have (senior unsecured debt) CDS spreads reported by Markit on at least 75% of the business days in our sample. In our benchmark empirical exercises, we use the values of each country/regional CDS index on a given date only if a CDS spread is reported for at least 75% of the banks included in the respective country/region on that date.
11 We have not included the AT1 preferred shares and the T2 subordinated debt instruments issued by US banks since all of them have only statutory triggers. As discussed above, we focus on contingent convertible capital instruments with at least one contractual trigger.
12 Another potentially relevant critical capitalization level is the Capital Conservation Buffer Requirement (CBR). When an issuer’s capital buffer falls below the CBR, it is constrained in its ability to continue its existing dividend and coupon payouts.
3 Policy Debates and Hypotheses
For most advocates, the main case for CoCos is the quick and simple bail-in mechanism they provide. Early critics of CoCos have argued, however, that an even better and simpler bail-inable instrument is equity (Admati et al., 2013). Another early concern was that the design of some types of CoCos could result in destabilizing bail-ins (Sundaresan and Wang. These observations, and others we discuss below, have been at the heart of policy debates around CoCos. Our analysis focuses on how the design of CoCos matters for the overall costs and benefits from issuing such instruments.
3.1 Market-based or Accounting-based Triggers?
While CoCo issuance reduces the issuing bank’s probability of failure, it also changes the distribution of payoffs over different claims in states of distress. As a result, the CoCo trigger may be subject to gaming by bank managers, regulators, and investors. Different designs give rise to strategic gaming problems. One prominent issue that has been widely discussed in the literature is the pros and cons of book-value versus market-value triggers in terms of the promptness and accuracy of conversion.
The responsiveness of book-value triggers depends on the frequency with which book values are reassessed and disclosed. It also depends on the accuracy and robustness of the internal risk models that are used to measure the risk exposure of bank balance sheets. These factors reduce the ability for CoCos with accounting-based triggers to respond in a timely way to the onset of a sudden crisis. In contrast, market-value triggers, as Flannery (2005, and Calomiris and Herring (2013) have argued, are that they immediately respond to news about realized or prospective bank losses. In addition, they are somewhat more immune to accounting obfuscation.13
However, a potential shortcoming of market-based triggers is that they are exposed to price volatility generated by noise trading. Moreover, they may invite stock price manipulation or speculative attacks. For example, market-based triggers may be vulnerable to strategic short-selling for the purposes of forcing the conversion of a CoCo and thereby transferring value from equity holders to CoCo investors, as Hillion and Vermaelen (2004) have argued. A broader concern with using market prices for regulatory purposes is that this can distort the information content of prices (see Faure-Grimaud, 2002, and Bond et al., 2010). For CoCos specifically, when a market price triggers conversion, the share price reflects both the value of the issuer’s total assets and the redistribution of value among initial shareholders and CoCo investors, as discussed in Glassermann and Nouri (2016) and Berg and Kaserer (2015).
There are several proposals that address these latter concerns, including: (i) inducing gradual conversion into shares (Bulow and Klemperer, 2014); (ii) fixing the number of new shares at conversion (Flannery, 2016); (iii) letting the trigger be based on a historical moving average of stock prices (Duffle, 2009, and Calomiris and Herring, 2013); and (iv) fixing the conversion price below the share value at conversion, with an option for initial shareholders to purchase the new shares at the lower price (Pennacchi et al., 2013).
Another important feature of CoCos is their maturity. As Pennacchi and Tchistyi (2016a) have argued, CoCos with infinite maturity are much more robust to price manipulation. They show that when CoCos are perpetuities, the multiple equilibrium problem identified by Sundaresan and Wang (2015) disappears and equilibrium is unique regardless of other CoCo features.
3.2 Loss Absorption and Risk-taking Incentives
The effect of CoCo issuance on risk-taking incentives of the issuing bank is an important aspect of CoCo design. In theory, PWD CoCos could encourage excess risk-taking around the conversion trigger, whereas MC CoCos that dilute existing shareholders upon conversion could encourage accounting manipulation to delay conversion. Moreover, CoCos with conversion terms that transfer value to equity holders upon conversion are likely to encourage additional risk-taking. Indeed, Berg and Kaserer (2015) analyze the conversion price of CoCos issued by Lloyds in an option-pricing context and show that those bonds create perverse risk incentives for banks’ equity holders.
Excess risk-taking incentives associated with PWD CoCos can be mitigated by exposing bank executive compensation to CoCo price risk (Bolton et al., 2015 and Hilscher and Raviv, 2014). Approaches to address the concerns regarding MC CoCos have been more diverse. Flannery (2005) and Calomiris and Herring (2013) urge that the design of CoCos should take the form of MC CoCos that are dilutive to preexisting equity holders and that convert well before the bank faces insolvency. Similarly, Martynova and Perotti (2015) show that CoCos lower bank risk-taking in equilibrium only when they take the form of high-trigger MC CoCos. But the problem with these CoCo designs is that bank equity holders have little incentive to issue such CoCos, because doing so mostly benefits outstanding unsecured creditors with higher seniority in the event of a bankruptcy state. In other words, bank equity holders may face a “debt overhang” problem with respect to high-trigger, dilutive, MC CoCos. However, Chen et al. (2017) and Albul et al. (2013) have shown that equity holders can have a positive incentive to issue CoCos to the extent that they also benefit when the issuer has a lower default risk (for example, via lower costs of debt rollovers).
In summary, the literature highlights that CoCo contract features affect bank risk-taking incentives. Though CoCo contract design could be influenced by regulators, the reality is that issuing banks have substantial leeway in selecting the contract features they want. Hence it is important to understand the effects of CoCo designs on bank incentives in order to induce CoCo contracts that are desirable from the perspective of financial stability.
3.3 CoCos as a Macroprudential Tool
Several CoCo design proposals have focused on the question of how CoCos should be modified to deal with a systemic risk event. The essence of these proposals is that the CoCo trigger should be contingent on some measure of systemic risk. However, this contract design could paradoxically lead to higher systemic risk because it would increase moral hazard frictions. That said, CoCos whose conversion is triggered by a systemic event do offer the desirable property that they deliver a bail-in only when it is needed, in a crisis when restructuring a bank’s liabilities and raising new capital is difficult.
One early such proposal came from Kashyap et al. (2008), who suggest that banks purchase capital insurance, with a trigger that is based on a systemic event. Similarly, the Squam Lake Working Group (2009) proposes that CoCos convert to equity under two conditions: (i) a breach of a trigger based on an accounting capital ratio; and (ii) a declaration by regulators that there is a systemic crisis. Caballero and Kurlat (2008) have also proposed that the central bank could issue tradable insurance credits that entitle the holder to attach a central bank guarantee to assets on its balance sheet during a systemic crisis. Their proposal eliminates the need to make the trigger contingent on a specific event, which they argue is a better way of dealing with a surprise shock.
CoCos can also be a macroprudential tool if they implement counter-cyclical equity buffers. One way of achieving this, suggested by Bolton and Samama (2012), is to do away with automatic triggers altogether and design the CoCos as reverse convertible bonds granting the issuer the option to convert the bond into equity. Issuers would only convert when the put option embedded in the CoCo is in the money, which is more likely to be the case in recessions or in a financial crisis. When the issuer exercises the option, the bank is effectively recapitalized at more favorable terms than those available in the market. A CoCo structured as a reverse convertible bond would be equivalent to giving the issuer a commitment to augment its equity capital at will, and at favorable terms, in recessions, thus implementing a form of countercyclical equity buffer. Vallee (2016) shows that European banks that had issued hybrid debt before the crisis took advantage of the convertibility option to convert the debt issues in the middle of the crisis of 2007-09, which enabled them to partially recapitalize their stressed balance sheets. Finally, Zeng (2013) also derives optimal countercyclical bank capital requirements, and shows that they can be implemented using CoCo instruments.
Our empirical analysis focuses on two key questions. What bank characteristics determine whether a bank issues a CoCo of a given design? And what is the impact of CoCo issuance on the market price of other claims on the issuing bank? To guide our empirical analysis, we develop a simple theoretical model (presented in Appendix B) that generates several predictions on these two questions. The model considers a bank issuer with assets and liabilities in place that currently satisfy regulatory capital requirements, but risks violating these requirements in a future crisis state in which it incurs a loss. The decision the issuer faces is whether to issue a CoCo now, so as to buttress its balance sheet sufficiently to be able to withstand a loss without violating its capital requirements, or risk having to go through a costly recapitalization in the crisis state. As a trade-off to fire-sale recapitalization costs in a crisis, the issuer faces CoCo issuance costs. These issuance costs include underwriting fees and dilution costs due to asymmetric information.
The model assumes that the decision whether to issue a CoCo or not rests with the shareholders of the issuing bank, who evaluate the relative costs of CoCo issuance today versus contingent recapitalization tomorrow, with the goal of shareholder value maximization. The analysis implies that issuing a CoCo is more attractive to shareholders the less the CoCo issue is underpriced by the market, the lower the fraction of equity promised to CoCo investors upon conversion, and the higher the asset sales discount in a crisis.
Accordingly, there are two important differences between an MC CoCo and a PWD CoCo for a bank issuer. First, the conversion of the PWD CoCo always increases the value of equity, while the conversion of an MC CoCo could result in dilution of equity holders. Second, PWD CoCos are less likely to be underpriced by the market, because the loss upon conversion is contractually specified and is therefore less likely to be overestimated by investors. Moreover, PWD CoCos are 100% fixed income securities, which are more straightforward to value and which have greater appeal to institutional investors specializing in fixed income securities.
The key predictions of the model about banks’ propensities to issue CoCos are as follows. There is an inverse U-shaped relation between the issuing bank’s incentives to issue a CoCo and the bank’s equity capitalization. When the bank’s capitalization is very low, its shareholders have little incentive to issue CoCos, especially MC CoCos, since most of the benefits of the CoCo issue go to senior unsecured debt holders and not equity holders. When the bank’s capitalization is very high, it has no need to raise costly equity, whether in the form of CoCos or common stock. It is mostly for intermediate equity capitalization levels that equity holders have an incentive to issue CoCos, for then equity holders, along with senior unsecured debt holders, benefit from the CoCo issue by avoiding a future costly recapitalization in a crisis.
The model analysis also delivers the following predictions about CoCo-issuing banks’ incentives to engage in excess risk-taking.14 A sufficiently dilutive MC CoCo contract mitigates bank shareholders’ incentives to engage in excess risk-taking, as they would bear a cost as a result of CoCo conversion following the realization of a loss. This effect is stronger for CoCos with higher conversion triggers. In contrast, PWD CoCos encourage bank excess risk-taking by decreasing the size of the losses for bank shareholders in the crisis state.
Finally, the model delivers key predictions about the effect of CoCo issuance on the value of the bank’s other claims. First, CoCo issuance, whatever its design, should increase the recovery value of senior unsecured debt, and therefore lower the bank’s CDS spreads. However, there is no such clear prediction for the issuing bank’s share price. The reason is that the decision to issue a CoCo conveys a "mixed" signal about the health of the issuer’s balance sheet. If the market prior is that the issuer has a very strong balance sheet, then the issuance of a CoCo is bad news, but if the market prior is that the bank is undercapitalized, then the issuance of a CoCo is good news. The formal analysis considers a situation where the issuer may be of three different types: (i) poorly capitalized banks that become insolvent upon the realization of a loss; (ii) adequately capitalized banks that require recapitalization upon the realization of a loss; and (iii) well capitalized banks that do not require a recapitalization upon the realization of a loss. As we pointed out above, our model predicts that poorly and well capitalized banks should have no incentives to issue a CoCo. Thus, an announcement of a CoCo issue can trigger a positive or a negative stock price reaction, depending on the prior beliefs about the distribution of bank types in the economy.
13 As we have already noted, all AT1 and T2 CoCos must include a discretionary trigger. Such triggers potentially have regulatory forbearance problems, and generate excessive sensitivity of the market to regulatory announcements.
14 Our analysis on risk-taking incentives aligns with the literature reviewed in Section 3.2.
4 Empirical Analyses
We conduct two main sets of empirical exercises. First, we perform duration analyses in order to investigate the main determinants of the propensity of a bank to issue a CoCo. Second, we estimate of the impact of CoCo issuance on the CDS spread and the stock price of the issuing bank, as well as the differential effects of the main CoCo contract features.
In both of our benchmark empirical analyses, we focus on CoCos classified as liabilities and issued by banks from advanced economies. We do not include CoCos issued by banks from EMEs since the considerable degree of heterogeneity in the timing of the Basel III implementation among EME jurisdictions could introduce a significant amount of noise in the empirical analyses. We do not include CoCos classified as preferred shares in our benchmark empirical analyses for a couple of reasons. First, as discussed above, CoCos that are classified as preferred shares are exempt from the requirement to have a mechanical trigger with a minimum level of 5.125% in order to be eligible to qualify as AT1 capital. This makes such CoCos considerably different from AT1 CoCos classified as liabilities. Second, CoCos classified as preferred shares are concentrated in a very small number of jurisdictions and represent a tiny fraction of our overall sample.
4.1 Determinants of CoCo Issuance
The first question we analyze concerns banks’ propensity to issue CoCos. We focus on an “unconditional” sample of banks that could be potential CoCo issuers. We restrict the sample to the top 500 banks (ranked by total assets in 2009) in the set of CoCo-issuing advanced economies, plus all the CoCo issuers from advanced economies that are not among the top 500. The resulting sample includes a total of 523 banks.
We take January 2009 as the beginning of our sample period and track CoCo issuance of all the banks in our sample until December 2015.15 For our first question, we estimate the expected duration from January 2009 until the time when a bank with given characteristics issues a CoCo for the first time. We perform several duration analyses that are reported in Table 2. The first set of results, reported in Panel A, is from a two-sided Tobit regression of time to issue in the cross-section, the number of months from January 2009 to the time of the first CoCo issue, as a function of key bank balance sheet characteristics. The duration, by construction, is bounded from below at zero. For the non-issuers, the time of the first CoCo issue is coded as being censored at the end of the sample period.
[Insert Table 2 here.]
The independent variables we focus on are: (i) total assets (in logs); (ii) Tier 1 (Tier-1 capital); (iii) G-SIB (a dummy variable for G-SIB status); (iv) gross loans; (v) deposits; (vi) trading securities; and (vii) long-term funding. All variables, except total assets, are scaled by total assets and expressed as percentage points. All bank characteristics are measured at the beginning of the sample period, in January 2009. We break down deposits into customer deposits and bank deposits on the grounds that the risk profile of banks with mostly customer deposits is likely to be significantly different from that of banks with mostly bank deposits. We also track interbank lending on both the liability and asset sides through the variables interbank borrowing and interbank assets.
Our main findings are as follows. First, the coefficient on total assets is negative and significant for all specifications. Note that a positive coefficient indicates a longer delay to issuance, or a low propensity to issue, so that our first result implies that larger banks are quicker to issue CoCos over our sample period. This confirms the anecdotal evidence that smaller banks take longer to test the market for new financing vehicles.16
Second, the coefficient on Tier 1 is negative and statistically significant in most specifications. Recall that our theoretical prediction is that inadequately capitalized banks are less likely to issue CoCos. Based on the coefficient in column (1), a one-percentage-point increase in Tier 1 capital (including CET1 and AT1, with an average ratio of 7.0% in 2009) is associated with a 3.8-month shortening in the time to issuance, out of a sample average of 55.7 months (from January 2009) for all issuers. To take into account the potential nonmonotonicity in the relation, we add the square of the variable Tier 1 (or Tier 1 " 2), in specification (3). Although the coefficient on the squared term is negative, it is insignificant. This is likely due to the fact that the increasing regulatory benchmark during our benchmark window implied issuance needs for most banks. As a consequence, hardly any banks in our sample were in an “extremely well funded” state, which prevented the non-monotonicity in the theoretical model from taking effect.
Third, on the asset side of bank balance sheets, the coefficients on both gross loans and trading securities are negative and significant. A higher value of the trading securities variable is typically interpreted as an indication of greater risk-taking by a bank (Roengpitya et al, 2014). Every percentage point increase in trading securities (with an average of 17.8%) predicts a 0.9 months shortening of the expected duration to first issuance.
Fourth, on the liability side, deposits are commonly considered to be a reliable and sticky source of bank financing, as opposed to wholesale funding which is more sensitive to changes in interest rates and more prone to “runs” in response to negative information about bank
profitability. Indeed, Huang and Ratnovski (2011) suggest that wholesale funding was one of the major sources of bank vulnerability during the financial crisis. Consistent with their analysis, we find that banks that are predominantly financed by deposits are significantly (at the 5% level) less likely to issue CoCos. The coefficient of 0.810 for deposits suggests that for every 10-percentage point increase in deposits relative to total assets, a bank waits on average 8.1 more months before issuing its first CoCo. Further breaking down deposits into customer deposits and bank deposits, we observe that both elements are highly significant.
Finally, column (5) in Panel A shows that an increase in interbank assets has no significant effect, but that banks more reliant on interbank borrowing are significantly less likely to issue CoCos. In the interbank market, a few banks tend to play the role of market makers by channeling the excess deposits of, usually smaller, banks that do not have access to lending opportunities to medium-sized and larger banks that have ample lending opportunities (Fecht et al., 2011).
Panel B of Table 2 reports the estimates of a Cox (1972) proportional hazards analysis of banks’ propensity to issue CoCos in each month during our sample period of January 2009 to December 2015. Here a higher coefficient means a higher probability of issuing a first CoCo in a given month, conditional on no-issuance until the previous month. Each coefficient, once exponentiated, could be interpreted as an “odds ratio” of issuance by a bank in a given month.
The results of the Cox hazards analysis mostly reinforce those reported in Panel A: larger banks, as well as banks with more loans and marketable securities are more likely to issue a CoCo. The opposite is true for banks with more deposits and more interbank borrowing. Note, however, that Tier 1 capital adequacy no longer matters at the monthly frequency, except when both Tier 1 and (Tier 1) "2, are included as regressors, suggesting that the negative relation between core capital and propensity of CoCo issuance is most prominent among banks with low levels of Tier 1 capital.
Panels C and D of Table 2 break down CoCo issuance into PWD CoCo and MC CoCo issues at the bank-month level using the Fine and Gray (1999) competing risk model. Under this model, the first issue of a CoCo with a particular loss-absorbing mechanism (PWD or MC) is considered the realization of one type of risk, with the other possible form of loss- absorbing mechanism interpreted as a “competing risk.” Accordingly, this model estimates the propensity to issue a particular type of CoCo instrument, taking into account that another type of CoCo is also available in the issuer’s choice set.
A finding that emerges from this analysis is that total assets and Tier 1 tend to be more consistently significant for PWD CoCos. This finding is consistent with our theoretical prediction that CoCos are issued when they primarily benefit shareholders, but not when they primarily benefit senior unsecured debt investors. The larger banks, and the better capitalized banks, are likely to be more resilient in a crisis. For these banks there is therefore a low credit risk associated with their senior unsecured debt, and hence a low debt overhang problem. In other words, most of the potential benefits of a CoCo issue go to incumbent shareholders, who tend to benefit more from a PWD CoCo issue than from an MC CoCo issue, since the latter has the potential to dilute the value of their shares upon conversion.
4.2 Impact of CoCo Issuance on Senior Debt and Equity
4.2.1 Empirical set-up
The issuance of a CoCo can affect the CDS spread of the issuing bank through two main channels. First, it reduces the probability of default by providing an additional layer of loss- absorbing capacity.17 Second, issuing a CoCo affects the risk-taking incentives of the bank’s management and equity holders, which can alter the probability of default and thereby have an impact on the riskiness of senior debt. While the second effect could go in either direction depending on the main CoCo design features, the first effect should be weakly beneficial for banks’ senior debt holders. Overall, CoCo issuance is expected to lower CDS spreads, as our theoretical analysis suggests.
As we explain in Section 3.4, the effect of issuing a CoCo on the issuer’s stock price is more difficult to determine, a priori. However, the relative effect of different CoCo designs on the issuer’s stock price is easier to pin down. Given that PWD CoCos do not have any negative equity dilution effects, we expect to see a more positive stock price response to the issuance announcement of a PWD CoCo than to an MC CoCo.
We assess the announcement effect of issuing CoCos on CDS spreads and equity prices using two different empirical approaches. First, we follow the estimation methodology used in James (1987) to determine the impact of CoCo issuance. In addition to assessing the overall effect, we also examine how it depends on the main CoCo contract features (loss
absorption mechanism, trigger type, trigger level, etc.) and on issuer characteristics (size, capital cushion over trigger level, G-SIB status, geographical location, etc.).18 Second, we estimate the impact of CoCo issuance on CDS spreads and equity prices in a standard cross-sectional regression framework.
The event date for CoCo issuance is not always clearly defined. Unlike in other event studies, where all relevant information is simultaneously announced to all market participants at a clearly defined point in time, for CoCo issues there is no single point in time when an upcoming issue is publicly announced. Instead, information about an upcoming issue spreads among market participants in a diffusion-like process. According to market participants, the information about the intention of a bank to issue a CoCo is revealed to a small group of potential buyers over the two weeks prior to the date of the issuance. As the book is built, the information is likely to diffuse to a wider set of investors and to be incorporated into equity prices and CDS spreads prior to the actual date of the issue. Still, the issue date often reveals additional value-relevant information, such as the over-subscription status of the issue.
Due to the above reasons, we adopt a conservative approach in selecting the event date and the window size for our benchmark exercise. More concretely, when measuring the impact of CoCo issuance on CDS spreads, we select an 11-day window (in units of trading days), which starts ten business days before the issuance date (t-10) and ends at the issuance date (t): [t-10, t]. When measuring the impact on equity prices, we select a shorter, six-day window, which starts five business days before the issuance date (t-5) and ends at the issuance date (t): [t-5, t]. We select a shorter event window for equity prices than for CDS spreads since the former tend to be more informationally sensitive than the latter. As a result, equity prices are more likely than CDS spreads to have their reaction to CoCo issuance be “contaminated” by other information that is revealed during the event window.19
15 CoCos were only beginning to be considered as a financing option by banks in 2009.
16 See, e.g., "Coco bonds: Mass conversion," Economist, September 13th, 2014.
17 As discussed in the Introduction, our analysis is based on the implicit assumption that CoCos are issued on top of (rather than instead of) CET1.
18 Appendix A contains a detailed description of the estimation methodology.
19 We also conduct robustness checks, in which we explore a number of alternative window sizes and event dates. While we do not report those results in the paper due to space constraints, all of them are available upon request.
4.2.2 Impact on bank CDS spreads
We estimate the change in issuers’ CDS spreads (on senior unsecured debt) around CoCo issuance dates. Table 3 reports the results of this estimation.
[Insert Table 3 here.]
Our main finding is that the overall impact of CoCo issuance on the CDS spread of the issuing bank is negative and strongly statistically significant. The z-value (—2.70) indicates that the cumulative change vis-à-vis the benchmark during the 11-day event window is negative and statistically significant at the 1% level.
The economic significance is also meaningful. The average prediction error (APE) for the full sample is approximately 3 bps. Once prediction errors are weighted by the (US dollar) volume of each issue, the estimated impact increases to 5 bps. This implies that issuing a CoCo reduces the annual interest costs associated with each $10 billion of (non-CoCo) bank debt by $5 million. Furthermore, the estimated economic impact varies considerably across CoCo sub-groups, ranging from 2 to 22 bps.
We next break down the full sample into subsamples sorted by the most important CoCo contract terms and issuer characteristics. Our first sorting variable is the loss absorption mechanism (mandatory conversion or principal write-down). In line with the predictions of Section 3.4, the impact of mandatory conversion (MC) CoCos on CDS spreads tends to be considerably stronger than the impact of principal write-down (PWD) CoCos. The estimated impact of MC CoCo issuance is negative and statistically significant at the 1% level (with a z-value of —3.07). Roughly two-thirds (63%) of the prediction errors in the MC CoCo subsample have a negative sign and the Wilcoxon signed rank statistic is statistically significant at the 1% level. The coefficient for the impact of PWD CoCo issuance has the right (negative) sign, but it is not statistically significant (according to both, the z-statistic and the Wilcoxon signed rank statistic).
In economic terms, the size of the estimated impact of MC CoCo issuance on CDS spreads is also considerably larger than the impact of PWD CoCo issuance. The simple average prediction error for MC CoCos (—5.0 bps) is roughly three and a half times higher than that for PWD CoCos (—1.4 bps). When we weight prediction errors by the size of each respective CoCo issue, the estimated impact of issuing MC CoCos is —8.4 bps, whereas the impact of issuing PWD CoCos is —2.4 bps.
The significant differences between the estimated impacts of issuing MC CoCos and PWD CoCos on the CDS spread of the issuing bank points in particular to the potential importance of the loss absorption mechanism for banks’ risk-taking incentives. The MC feature increases the cost of risk-taking for current shareholders and management due to the threat of equity dilution. But, no such costs are imposed by PWD CoCos. On the contrary, PWD CoCos actually increase the risk-taking incentives of shareholders, as PWD CoCo holders could absorb losses ahead of shareholders. Thus, in the case of PWD CoCos, the risk-taking incentive effect works in the opposite direction of the loss absorption effect (which has a negative impact on CDS spreads for both loss absorption mechanisms). In contrast, the two effects reinforce each other in the case of MC CoCos.
A second major finding is that the presence of a mechanical trigger in CoCo term sheets plays a very important role. The estimated impact of issuing a CoCo with a mechanical trigger is negative and statistically significant at the 5% level (with a z-value of —2.22). That subsample of CoCos has an average prediction error of —3.3 bps and weighted average prediction error of —4.8 bps. In contrast, the estimated impact of issuing a CoCo that only has a discretionary trigger, while also negative, is not statistically significant.
There are a number of explanations for this key finding. CoCos with only a discretionary trigger are more akin to pure "gone concern" instruments and inherit all the uncertainties associated with regulatory intervention in a bank resolution. At what point will the CoCo be triggered? What happens in resolution? Will the CoCo be totally or partially wiped out? Will it be the only debt instrument to be bailed in? All of these uncertainties compound to make the valuation of the CoCo and assessing the impact of the CoCo issue on the value of senior unsecured debt very difficult. In contrast, CoCos that also have a mechanical trigger combine features of both “gone concern” and “going concern” instruments. As a consequence, such CoCos are likely to absorb losses earlier, which is desirable from the perspective of senior unsecured bondholders.
The level of the mechanical trigger is also important for the CDS market reaction. The estimated impact of issuing CoCos with a high trigger (above the minimum trigger for ATI classification of 5.125%) on CDS spreads is negative and statistically significant at the 10% level. The estimated impact of CoCos with a low trigger (less than or equal to 5.125%) is also negative, but is not statistically significant. CoCos with a high trigger are closer to "going concern" CoCos, as they are more likely to convert before the PONV than low-trigger CoCos. Thus, they provide higher-quality protection to unsecured bondholders of the CoCo-issuing bank.
A third major result is that the impact of CoCo issuance on CDS spreads is affected by the interaction of the main CoCo design features. Namely, MC CoCos with a mechanical trigger have a negative and statistically significant impact, whose magnitude (-6.2 bps) is greater than those of the respective one-dimensional subcategories (MC CoCos and CoCos with a mechanical trigger). This implies that the above two characteristics interact with each other to boost the impact of CoCos on CDS spreads. Intuitively, the combination of reduced risk- taking incentives (associated with the MC feature) and automatic loss absorption (associated with the mechanical trigger feature) provide the greatest degree of protection for senior unsecured debt holders. Meanwhile, the impact of MC CoCos with a discretionary trigger is also negative, but it is only statistically significant at the 10% level and its magnitude (—1.5 bps) is considerably smaller.
A fourth important finding is that the size of the CoCo issue (as a share of RWA) also matters. Large CoCo issues have a negative and statistically significant (at the 5% level) impact on CDS spreads, but small issues have no statistically significant effect. The weighted average impact on the issuing bank’s CDS spread is considerably higher for relatively larger issues (—7.2 bps) than for relatively smaller issues (—1.7 bps). This is not surprising. The larger the issue, the thicker the layer of protection that CoCos provide to the senior unsecured bondholders. Moreover, the fact that the bank was able to place a relatively large issue in itself reveals that investors have confidence in the overall healthiness of the bank.
The regulatory classification of CoCo instruments also plays an important role. ATI CoCos have a negative and highly significant (at the 1% level) impact on CDS spreads. In contrast, the impact of T2 CoCos is insignificant. The weighted average prediction error for ATI CoCos is —6.7 bps, whereas that for T2 CoCos is only —0.4 bps. Intuitively, the enhanced design features associated with the higher quality AT1 capital instruments provide better protection for the senior creditors of the issuing bank.
Our results also suggest that CoCo issuance tends to have a bigger effect on the CDS spreads of relatively smaller issuers. The impact of CoCos issued by banks with total assets of less than $1 trillion is negative and statistically significant at the 1% level. In contrast, the impact of CoCo issuance on CDS spreads tends to be insignificant for banks whose total assets exceed $1 trillion. Splitting banks according to their G-SIB classification generates similar results. That is, the impact of CoCos issued by banks that are not classified as G-SIBs is highly significant while the impact of CoCos issued by G-SIBs is not statistically significant. This latter finding could be interpreted as indicating that market participants believe that only G-SIBs benefit from an implicit government guarantee.
We turn next to the results of our cross-sectional regression analysis of the impact of CoCo issuance on CDS spreads. These results, reported in Table 4, are in line with the results obtained using the James (1987) methodology.
[Insert Table 4 here.]
The first set of cross-sectional regressions examines the effect of the loss absorption mechanism (Table 4, Panel A). In line with the results obtained using the James (1987) methodology, the main finding is that the impact of MC CoCo on CDS spreads is much more negative and significant than that of PWD CoCos (specification (1)). In the baseline specification, the impact of MC CoCos is negative and statistically significant at the 5% level. Once additional dummy controls are included, the impact of MC CoCos is still negative and statistically significant at the 5% level for AT1 instruments (specification (2)) and for repeat issuers (specification (4)). In contrast, the estimated impact of PWD CoCos is not statistically significant in any of the examined specifications.
The second set of results provides insights into the impact of the trigger type (Table 4, Panel B). In line with the results obtained using the James (1987) methodology, the impact of CoCos with mechanical triggers is negative and statistically significant at the 5% level (Column 1). In contrast, the impact of CoCos that only have a discretionary trigger is not statistically significant. The negative impact of mechanical-trigger CoCos appears to be stronger for non-G-SIBs (Column 2) and for repeat issuers (Column 3). In sum, CoCos are most likely to be beneficial to senior unsecured bondholders if their automatic loss- absorption feature is combined with a thicker layer of protection (as in the case of repeat issuers) or with an issuer that is not systemically important (as in the case of non-G-SIBs). The last regression in Panel B examines two-dimensional splits of the main CoCo design features (Column 4). They indicate that the impact of MC CoCos with a mechanical trigger (MC/MT CoCos) is negative and statistically significant at the 5% level. The impacts of the other three CoCo types (MC/DT, PWD/MT, PWD/MT) are also negative, but not statistically significant.
The third set of regressions investigates the effect of changing the level of the mechanical trigger (Table 4, Panel C). The estimates from the baseline specification indicate that both high-trigger and low-trigger CoCos have a negative impact, which is statistically significant at the 10% level (Column 1). The specifications with additional dummy controls reveal that the high-trigger CoCos of repeat issuers have a negative and strongly statistically significant (at the 1% level) impact (Column 3). In contrast, the impact of repeat issuers’ low-trigger CoCos is statistically significant only at the 10% level. Interestingly, the impact of first-time issuers’ CoCos is not statistically significant, regardless of the trigger level. One possible explanation for these results is that the early loss absorption feature of high-trigger CoCos has a larger impact for repeat issuers, since they increase the size of the already existing layer of protection that CoCos provide for senior unsecured debt, making CoCos a more potent loss absorption tool. This hypothesis is further supported by the fact that the coefficient on the issued CoCo amount (as a share of RWA) in the same regression specification is negative and statistically significant.
4.2.3 Impact on bank equity prices
Our theoretical analysis does not yield clear-cut predictions on the effects of CoCo issuance on equity prices. That said, we should expect PWD CoCos to have a more positive effect on stock price than MC Cocos, all else equal, because they are more protective of shareholders.
Table 5 reports the results using the same methodologies as in Table 3, but replacing CDS spreads with stock prices as an independent variable. For the overall sample, the impact of CoCo issuance is not statistically significant. This most likely reflects the same ambiguity that we discuss in our theoretical predictions. The results for the one-dimensional subsample splits along the main CoCo design features (loss absorption mechanism, trigger type, trigger level) are also insignificant.
[Insert Table 5 here.]
The two-dimensional subsample splits generate the most interesting results for the impact of CoCo issuance on equity prices. The most important finding is that issuing a PWD CoCo with a high trigger has a positive and statistically significant (at the 5% level) impact on the equity price of the issuing bank. The combination of (i) the early loss absorption feature associated with high-trigger CoCos and (ii) the fact that PWD CoCo holders absorb losses ahead of equity holders appears to explain this result. Conversely, the impact of hightrigger MC CoCos on equity prices is negative, albeit not statistically significant. A likely explanation for this result is that the effect of possible dilution of equity holders may be (partially) offset by the effect of owning shares in a bank that is better capitalized and more stable as a result of the CoCo issuance. The fact that the estimated impact of high-trigger MC CoCos goes in the opposite direction to that of high-trigger PWD CoCos also explains why the overall impact of high-trigger CoCos (from the one-dimensional sample splits) is insignificant.
Finally, we examine the impact of CoCo issuance on equity prices in cross-sectional regressions. The main results, which are presented in Table 6, are in line with the ones obtained using the James (1987) methodology. When examined on a stand-alone basis, the main CoCo design features (loss absorption mechanism, trigger type, trigger level) do not have a statistically significant impact on equity prices.
[Insert Table 6 here.]
The regression specifications that include two-dimensional (CoCo design feature) dummies generate the most interesting and statistically significant results (Columns 1-3). Just as for the estimates generated using the James (1987) methodology, the cross-sectional regression coefficients imply that the impact of PWD CoCos with a high trigger on equity prices is positive and statistically significant (at the 5% level). Furthermore, the impact of PWD CoCos with a mechanical trigger is also positive and statistically significant (Columns 4-5). However, its estimated magnitude is somewhat smaller than the one for PWD CoCos with a high trigger. Thus, the positive impact of PWD CoCos on stock prices appears to be present for all PWD CoCos with a mechanical trigger. Nevertheless, it is stronger for high-trigger PWD CoCos (which absorb losses relatively early) than for low-trigger PWD CoCo (which absorb losses relatively late).
Overall, our analysis indicates that CoCos can contribute to reducing bank fragility. The CoCo market is no longer a small niche market. Although CoCos are deemed to be complex by many commentators, possibly too complex for retail investors, there appears to be a sufficiently large institutional investor clientele that stands ready to hold them. One reason why CoCos are perceived to be so complex is that there is a great variety of CoCo designs, as we have shown. What is more, the designs that have been chosen by issuers are quite different from those recommended by the large theoretical literature on CoCos (e.g., Albul et al (2013), Chen et al (2017), Glasserman and Nouri (2016), Pennacchi and Tchistyi (2016a, 2016b), Chan and van Wijnbergen (2016)).
The change in the mix of CoCo designs is primarily driven by a combination of experimentation, issuer incentives and investor demand. Now that the CoCo market is reaching maturity, it is important to find out which designs are desirable and which ones are not from a financial stability point of view, as well as where CoCo design can possibly be simplified with a view to standardizing this market.
Our study provides a first set of answers to these questions. One of our important findings is the shift over time from MC to PWD CoCos. Our other results suggest that this shift is driven by shareholder preferences, as the announcement effect of PWD CoCos on stock price is positive. This shift is also driven by fixed income investor clienteles, who prefer to hold pure fixed income products rather than hybrids, other things equal.
But is this a desirable development from a financial stability perspective? We have shown that the issuance of MC CoCos has a stronger impact on CDS spreads, which may suggest that MC CoCos have a superior design from the point of view of reducing bank fragility. However, we have also shown that shareholder risk-taking incentives tend to be stronger for instruments with PWD features. This points to a trade-off in terms of the combined effects of contractual features and overall issuance volumes for financial stability that any efforts to standardize CoCo instruments would have to take into account. Other potential avenues for standardization include: (i) reconsidering the benefits of CoCos with only discretionary triggers; (ii) requiring higher triggers, so that CoCos are more like "going concern" than "gone concern" instruments; (iii) revisiting the merits of T2 CoCos; (iv) considering whether to increase CoCo requirements with the goal of increasing their overall loss-absorbing capacity.
We have explored only a subset of questions on the effects of CoCos on issuing banks’ balance sheets. An important open question for the immediate future concerns the investor clientele of CoCos. Unfortunately, the lack of information on the buy side makes it difficult to know the distribution of CoCo holdings and assess whether CoCos reduce rather than redistribute risks in the banking system. CoCos would enhance the stability of issuing banks and the banking system at large only if the holders of CoCos were long-term investors unconnected to the banking system.