• No results found

Evidence from Credit Default Swap demand : a study on the motivation behind CDS trading and the impact of credit rating changes on CDS demand

N/A
N/A
Protected

Academic year: 2021

Share "Evidence from Credit Default Swap demand : a study on the motivation behind CDS trading and the impact of credit rating changes on CDS demand"

Copied!
49
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

I

Evidence from Credit Default Swap demand: A study on the motivation behind CDS trading

and the impact of credit rating changes on CDS demand.

ABSTRACT

This thesis aims to examine the motivation behind CDS trading and the impact of credit rating changes on CDS demand. To do so I analyze a dataset of CDS demand for corporate single-name CDS ranging back to 2003 which is based on CDS position data obtained from the Depository Trust & Clearing Corporation (DTCC). I find some evidence that safe companies experience less CDS demand compared to companies in distress. This is indicative of hedging being one of the economic function of the CDS market. In addition to that I find heightened CDS trading volume around both rating downgrades and rating upgrades but few evidence of market risk transfer around rating changes. Furthermore, CDS demand tends to be higher for financial companies, companies with a higher proportion of unsecured debt and equity return volatility. However, CDS demand does not seem to change significantly around changes in the risk-weight categories for corporate debt based on the Basel capital requirement framework which would be indicative for banks engaging in regulatory arbitrage. This paper adds to a small set of literature examining the information content of CDS volume and a set of literature investigating the motivation of CDS trading and the economic function of the CDS market.

Universiteit van Amsterdam

MSc. Business Economics: Finance July 2016

Masterthesis by: Richard Mock Supervised by: Dr. Tomislav Ladika

(2)

II

Statement of originality

This document is written by Richard Mock who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

III

CONTENTS

1 INTRODUCTION... 1

2 BACKGROUND ON CDS ... 3

3 LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT ... 4

3.1 LITERATURE REVIEW ... 5

3.1.1 The role of financial corporation in CDS markets ... 5

3.1.2 Determinants of CDS volume ... 8

3.1.3 Evidence from initiation of CDS trading ... 9

3.1.4 Impact of credit rating changes on CDS spreads ... 9

3.2 HYPOTHESIS DEVELOPMENT ... 10

4 DATA AND METHODOLOGY ... 12

4.1 DATA AND VARIABLE CONSTRUCTION ... 12

4.1.1 CDS demand ... 12 4.1.2 Independent variables ... 15 4.1.3 Control variables ... 17 4.2 EMPIRICAL METHODOLOGY ... 18 4.3 DESCRIPTIVE STATISTICS ... 21 5 RESULTS ... 22 6 ROBUSTNESS CHECKS ... 26

7 CONCLUSION AND DISCUSSION ... 28

APPENDIX ... 32

(4)

1

1 INTRODUCTION

After being introduced in the early 1990s by JPMorgan1, usage of Credit Default Swaps (CDS) in the financial industry grew rapidly reaching its peak prior to the financial crisis in the mid-2000. According to data from the Bank of International Settlement (BIS), total gross notional outstanding CDS globally increased from $6 trillion in 2004 to $60 trillion in the second halve of 2007, just before the onset of the financial crisis2. In the following time CDS came under heightened scrutiny regarding their contribution to systemic risk, the overall stability of financial markets and how they have contributed to the financial crisis. Stulz (2010) provides a detailed overview of the role of CDS during the financial crisis discussing social costs and benefits of the CDS market.

Academic literature on CDS developed in a variety of different streams, focusing on valuing CDS contracts and the determinants of CDS spreads (Das (1995), Duffie (1999), Huang and Zhou (2008), Gamba and Saretto (2013), Gamba, Aranda, and Saretto (2015), Longstaff, Neis, and Mithal (2005)), on how the use of CDS can lead to a separation of the creditor’s control rights from his cash-flow rights (Hu and Black (2008a), Hu and Black (2008b), Bolton and Oehmke (2011)) and implications on corporate finance (Subrahmanyam, Tang, and Wang (2014), Subrahmanyam, Tang, and Wang (2014), Campello and Matta (2012), Saretto and Tookes (2013), Hirtle (2009)).

Despite a large body of research on the aforementioned topics, the precise economic function of the CDS market is still unknown. Theoretical and empirical literature predicts the use of CDS as a mean to hedge company risk, counterparty risk, perform regulatory arbitrage or capital structure arbitrage. Others stress the use of CDS as an outright tool for speculation and insider trading. Only all small set of this literature focusses on the intensity of CDS trading and CDS demand to draw inference on the motivation behind trading CDS and the determinants of CDS positions.

I apply a unique approach to estimate single-name CDS demand from 2003 to 2015 based on CDS position data from the Depository Trust & Clearing Corporation (DTCC) and its Trade Information Warehouse (TIW). The position data obtained from the DTCC database allows to back out the net and gross maturing CDS in weeks with an IMM date3

1 Please refer to: Philips (2008) and Lanchester (2009) in additional sources. 2 Please refer to Bank of International Settlement Statistics Explorer, retrieved from http://stats.bis.org/statx/srs/table/d10.1

3 IMM dates are quarterly dates which most CDS use as their maturity date. These dates are standardized to be on the 20th of March, 20th of June, 20th of September and 20th of December.

(5)

2

from 2008 onwards. In each IMM week, outstanding CDS drop in the amount of the maturing CDS. From this it is possible to estimate CDS demand prior to 2008. For instance, assuming that 100% of maturing CDS are 5-year CDS contracts4, $10m maturing CDS in the week of the 20th March 2008 would lead to $10m CDS demand in the 1st quarter of 2003. I follow this procedure for all IMM dates from 2008 to 2015 and compile a dataset for corporate single-name CDS demand from 2003 onwards. Based on the demand estimations I want to analyze CDS trading activity in order to draw conclusions on the use of CDS as a mean to hedge company risk, counterparty risk or perform regulatory arbitrage.

By doing that so, this study is able to examine CDS demand before 2008 and add to the literature that examines the use of CDS in the context of hedging or speculation and the determinants of CDS volume. I hypothesize that CDS demand should be higher for companies that are in distress and that demand increases around credit rating changes. Both would be indicative for hedging company risk. Furthermore, I look for evidence that suggests active counterparty risk management activities or regulatory arbitrage in CDS markets. These topics are important as they add to our understanding on CDS trading volume and whether the CDS market is primarily driven by hedging or speculation. This gives insights into the economic function that CDS markets perform in general.

I find some evidence that safe companies experience lower demand for CDS and that demand for distress companies tends to be higher. Both is indicative for hedging activities in the market. Furthermore, I find increased gross CDS demand in quarters where the long-term credit rating of the reference entity changes. However, this evidence on rating changes merely allows to draw inference whether there is significant market risk transfer activity around rating changes as gross CDS demand is better interpreted as a measure of trading intensity and CDS trading volume. Demand on a net and gross basis is significantly higher for financial companies, indicating that the CDS market is used as a tool for counterparty risk management. In contrast to that, I do not find strong evidence for regulatory capital arbitrage in our analysis. Furthermore, the evidence suggests that CDS demand scaled by total long-term debt is decreasing in leverage and firm size and increasing in equity return

volatility and the proportion of unsecured debt.

The remainder of this thesis is structured as follows. Section 2 is a brief introduction into CDS and the functioning of the CDS market. Section 3 summarizes existing literature on

4 This assumption is for illustrational purposes only. Please refer to the Methodology section for the tenor distribution as applied in the CDS demand estimation process.

(6)

3

CDS and results in the formulation of four hypotheses. Section 4 describes the data sources, the CDS demand estimation process and the empirical methodology to analyze our hypotheses. In section 5 all empirical results are presented, followed by robustness checks in section 6. I conclude in section 7 and discuss the validity of my empirical findings.

2 BACKGROUND ON CDS

Based on Longstaff, Neis, and Mithal's (2005) introduction into Credit Default Swaps, this section aims to provide the necessary background on the nature of CDS and the characteristics of the CDS market.

Single-name Credit Default Swaps (CDS) are credit derivatives that allow the buyer to protect himself from the default of a bond. Typically, the insured bond is called reference obligation and the issuer of the reference obligation is called reference entity. CDS are bilateral contracts between a buyer of the derivative and a seller. The buyer of the CDS pays an insurance premium to the seller of the CDS until a predefined credit event5 occurs or the CDS matures. In return the CDS seller agrees to buy back the reference obligation from the protection buyer at face value in case a credit event triggers the settlement of the derivative contract. The CDS premium paid by the buyer of the CDS on a regular basis is meant to compensate the seller of the CDS for the credit risk he assumes. The premium is typically quoted in basis point per $100 insured notional of the reference obligation. Essentially, the protection buyer has a credit risk exposure similar to being short a corporate bond and the protection seller has a credit risk exposure similar to being long in a corporate bond.

CDS can be settled physically or by exchanging cash. In case of a physical settlement the protection buyer delivers the defaulted reference obligation with the face value equal to the notional amount of the CDS contract to the seller. The protection seller in return delivers the notional amount of the CDS contract to the protection buyer in cash. In case of cash settlement, the contract can be settled on the basis of current market prices of the reference obligation. The protection buyer would then receive the difference between the market value of the reference obligation and the notional value of the CDS contract.

CDS and regular insurance contracts differ in two important aspects. First, in contrast to a car insurance for instance, it is possible to buy insurance on a corporate bond without

5 A credit event can be defined as bankruptcy, failure of payment, restructuring, repudiation/moratorium or obligation acceleration.

(7)

4

owning the bond itself or to buy more insurance on your bond than your outstanding credit exposure would require to be fully hedged. It is therefore possible to use CDS not only for hedging purposes but also for speculation (Lee (2012)). Furthermore it is notable that over insurance with CDS ultimately can alter the relationship between the creditor and the borrower by separating creditor control rights from cash flow rights (Bolton and Oehmke (2011)). Secondly, CDS are actively traded. In contrast to an insurance contract most CDS are much more standardized, which makes them easy to trade on the market.

Credit Default Swap contracts are entirely traded on the over-the-counter market (OTC). In OTC markets, market participants directly trade with each other via telephone, eMail or electronic trading systems without a central exchange that matches demand and supply. OTC markets are generally considered to be less transparent with less regulation and reporting. To reduce counterparty risk, dealers can trade through a central counterparty (CCP) (Cecchetti, Gyntelberg, and Hollanders (2009)). CCPs are legal entities that act as middleman between buyer and seller of OTC securities. In case a trade is done via a CCP the buyer and seller of the security are no longer counterparties to each other. The CCP steps in as central counterparty for both the seller and the buyer, the initial contract between the buyer and seller is therefore replaced by two contracts between the CCP and each of the original traders. This practice allows for better management of counterparty risk and increases market transparency (Cecchetti, Gyntelberg, and Hollanders (2009)).

3 LITERATURE REVIEW AND HYPOTHESIS DEVELOPMENT

This paper is related to a small set of literature that attempts to examine the information content of CDS volume and trading intensity. This literature stream aims to provide insights on the motivation of CDS trading, determinants of CDS positions and the broader market structure that determines the CDS market. In order to derive the hypotheses, I use insights from mostly theoretical literature on the use of CDS by financial corporations such as banks, corporate bond funds and hedge funds, and empirical findings from papers examining the impact of initiation of CDS trading on reference entities and from papers that analyze CDS volume. This thesis adds to the literature by introducing a novel approach to estimate CDS demand before 2008 and examine determinants of CDS demand in order to draw inference on the use of CDS as a tool for hedging company risk, counterparty risk or perform regulatory arbitrage.

(8)

5

3.1 LITERATURE REVIEW

3.1.1 The role of financial corporation in CDS markets

Unlike for foreign exchange, interest rate and commodity linked derivatives, credit derivatives are mostly used by financial corporations (Smithson, Rutter Associates, and Mengle (2006)). Consistent with the findings of Smithson, Rutter Associates, and Mengle (2006) data from the Bank of International Settlement (BIS) suggests that on average less than 2.5% of total gross notional outstanding is traded by non-financial corporations. Furthermore, the data provides insights into the net positions of certain financial institutions. Hedge funds for instance consistently have been net sellers of protection whereas insurance companies mostly act as net buyers of CDS over the period of 2004 to 2015. Similarly, Siriwardane (2015) find evidence that in the time between 2010 and 2014 a large part of the net selling activity in the CDS market has moved from large dealers to hedge funds and asset managers.

A survey by Fitch Ratings (2010) shows that 43% of the banks name trading as a dominant objective for using CDS. 38% name hedging/credit risk management, 32% market making, 17% the use of an alternative asset class and 10% regulatory capital as being a dominant motivation for their use of CDS. This indicates that CDS are not only used for managing credit risk, but also to take views on credit market movements and benefit from regulatory arbitrage.

Since the majority of CDS are traded by financial corporations it is worthwhile to examine theoretical literature on the potential use of CDS through banks and other financial intermediaries. Duffee and Zhou (2001) consider the implications of CDS on bank’s risk sharing behavior. They introduce a model were a loan giving bank is exposed to a risk of bankruptcy in case the loan defaults. The bank is able to sell the loan on the loan sell market, but faces an asymmetric information problem when doing so. They show that in case of little asymmetric information CDS contracts can decrease the banks risk of financial distress and that in case of severe asymmetric information banks will use credit derivatives to hedge credit risk rather than selling the loan on the loan sell market. Zawadowski (2013) introduces a model to analyze asset and counterparty risk in banks. His model suggests that banks treat asset and counterparty risk differently. While banks trade OTC derivatives with each other in order to reduce asset risk they are not willing to enter into derivatives contracts to reduce counterparty risk, even though this would be social beneficial. According to the model OTC derivatives thereby create systematic risk by transforming

(9)

6

asset risk into counterparty risk, as counterparty risk is too costly compared to its benefits for the bank in case of counterparty default.

Acharya and Johnson (2007) examine insider trading in CDS markets. They find that the CDS market potentially reveals non-public information via informed trades of financial intermediaries that use private information from their lender relationship to buy protection on their credit portfolio. These findings are consistent with the prediction of Duffee and Zhou (2001) and suggest that banks indeed use the CDS market to hedge credit risk and that there is an information flow from the CDS market to the equity markets. However, the evidence suggests that the information flows only when the gains from hedging for the bank are large, which the authors interpret as evidence for insider trading. Minton, Stulz, and Williamson (2009) in contrast provide evidence that only a few banks use credit derivatives. Those banks who use credit derivative contracts seem to take these position rather for dealer services than for hedging. Furthermore, they find evidence that the use of CDS for hedging is more pronounced for investment grade corporations and that the market for CDS on non-investment grade corporations is less liquid. Private information and asymmetric information should be higher for non-investment grade corporations, therefore hedging these corporations will be costlier compared to investment grade companies. Parlour and Winton (2013) model the use of CDS versus loan sales of banks and find that loan sales dominate CDS for riskier credit but not for safer credit. When they include repeated borrowing and reputation concerns in their model their work suggests that CDS begin to dominate loan sales for safer credit. Parlour and Winton (2013) finding is consistent with research by Campello and Matta (2012). Campello and Matta (2012) develop a theoretical model of CDS demand implicating that CDS contracts are more beneficial for safer firms and firms with higher continuation values.

Yorulmazer (2013) examines the use of CDS for regulatory capital relief and its consequences for systemic risk. According to Basel capital regulation CDS can be used to mitigate credit risk, meaning that banks have to hold less capital against a risky investment when they buy protection via CDS. In times of rising cost of capital and increasing capital constraints banks may be forced to forgo positive NPV projects because of capital requirements. However, in these cases CDS allow banks to lower capital requirements and continue to invest into positive NPV projects. On the other hand, banks may exaggerate the use of CDS for regulatory capital and overinvest into high risk projects. Furthermore, in times of financial distress when default probabilities of banks and its derivative

(10)

7

counterparties are likely to be correlated CDS will fail to be a proper insurance against the default of the investment. In these times, banks will be worse off because they have high risk investments on their balance sheet, uncertain insurance and low regulatory capital requirements at the same time. Shan, Tang, and Yan (2016) find that total bank assets increased after they started using CDS while simultaneously risk-weighted assets decreased. They argue that banks use CDS to shift loans from high risk-weight categories to lower risk-weight categories in order to be able to hold less regulatory capital and thereby that banks merely use CDS for risk management purposes but rather for managing regulatory capital requirements.

These previously described paper essentially show four different uses of CDS contracts. First, to meet credit risk hedging demand by banks and investors (e.g. Duffee and Zhou (2001)), secondly to hedge counterparty risk (Zawadowski (2013)), thirdly for trading inside information (Acharya and Johnson (2007)) and fourth for regulatory arbitrage (Yorulmazer (2013)).

In contrast to the studies above Adam and Guettler (2015) examine the use of CDS in corporate bond funds rather than in banks. They find evidence that the use of CDS in corporate bond funds has increased significantly alongside the development of the general CDS market. 50% of the largest corporate bond funds use CDS and most of them hold both long and short positions. However, the aggregate corporate bonds funds were always net short of CDS. Siriwardane (2015) examines DTCC data on single-name and multi-name CDS and finds that hedge funds and asset managers are among the largest sellers in the CDS market, taking an increasing share from dealers. This implies that on average corporate bond funds and hedge funds in their sample used CDS positions to speculate rather than to hedge. Hedging would require a long position in a CDS contract. Yu (2004) and Duarte, Longstaff, and Yu (2007) examine the use of CDS by hedge funds and provide evidence that hedge funds might use CDS to engage in capital structure arbitrage6 and earn abnormal risk-adjusted returns by doing that. Furthermore, anecdotal evidence by Lewis (2010) suggests that hedge funds used CDS to short certain sub-price mortgage-backed securities during crisis in order to speculate on the housing market. These papers suggest that non-bank financial corporations, such as corporate bond funds or hedge funds, use the

6 Capital structure arbitrage refers to a fixed-income arbitrage strategy that exploits mispricing between a companies’ debt and its equity (Duarte, Longstaff, and Yu (2007)).

(11)

8

CDS market primarily to engage in fixed-income arbitrage and speculation rather than hedging.

3.1.2 Determinants of CDS volume

Oehmke and Zawadowski (2016) use position data on single-name CDS contracts from the DTCC to examine the economic role of the CDS market. Analyzing net notional CDS positions they find that the CDS market is larger for firms with more debt outstanding, which is indicative that at least some market participants use CDS to hedge credit risk. They estimate the impact on net notional positions outstanding to be 7.77 cents per dollar insurable interest. Bank debt however, does not seem to be a significant determinant of net notional CDS outstanding. Furthermore, Oehmke and Zawadowski (2016) examine the impact of speculative trading motives on net notional CDS outstanding. They find that earnings disagreement on the future earnings of the reference entity between analysts is an economically and statistically significant determinant of net notional CDS outstanding. Interestingly, they find that the impact of speculative motives, such as earnings disagreement, occurs almost exclusively in the CDS market and not in the underlying bond market. This shows that short-term speculators value the liquidity advantage of the CDS markets and rather trade there than in the underlying bond market. The authors thereby conclude, that CDS markets act as an “alternative trading venue” with both hedging and speculation and that the CDS market primarily has a standardization and liquidity function. Chen et al. (2011) examine trading frequency in CDS markets over a three months’ period in order to derive implications for regulation and greater market transparency. They find evidence that dealers typically hold on to risk taken by customer trades for a few days before engaging in hedging activity. Furthermore, their data suggests that trading activity in single-name CDS is often driven by specific credit events or economic events that alter the credit risk of the reference entity.

Peltonen, Scheicher, and Vuillemey (2014) examine the network structure and its determinants of the sovereign CDS market. They use DTCC trading data for their analysis and find evidence for both speculation and hedging in the CDS market although it is not possible to disentangle both. Their evidence suggests that a larger volume of underlying bonds is associated with an increasing network size and higher network activity which could be evidence for hedging. Furthermore, they find that network size and activity decreases with increasing debt maturity which they interpret as signs for banks rolling-over

(12)

9

their hedges. Evidence which can be viewed as indicators for speculation include that CDS volatility and beta are better determinants of the CDS market than CDS spreads as a measure for absolute risk. Similar to the findings of Peltonen, Scheicher, and Vuillemey (2014), Berg and Streitz (2015) find that CDS markets tend to be larger for countries that are smaller, riskier and that have weaker creditor rights. According to their evidence, common law countries, where hedging motives are potentially low, have significantly smaller CDS markets.

3.1.3 Evidence from initiation of CDS trading

Ashcraft and Santos (2009) find that after inception of CDS trading, cost of debt increased for riskier firms and firms with are more informationally opaque. For firms that are more transparent and safer however, spreads tend to decrease after initiation of CDS trading. Saretto and Tookes (2013) examine whether the ability to hedge credit risk by banks has an impact on a firm’s capital structure. They find statistically and economically significant evidence that the onset of CDS trading increases firm leverage and debt maturity. These impacts are most prevalent in periods with high credit supply constraints. This suggests that the ability of credit suppliers to hedge their exposure leads to an increase in credit supply in terms of both, quantity (leverage) and debt maturity. Subrahmanyam, Tang, and Wang (2014) examine the impact of CDS trading on corporate cash holdings and find that due to increased bargaining power of insured creditors, companies become more cautious and hold more cash. This is more pronounced for firms with higher refinancing risk and firms that have higher credit constraints.

3.1.4 Impact of credit rating changes on CDS spreads

Corporate rating changes are considered to have a significant impact on credit and equity markets, however the evidence of previous research seems to be mixed. While Katz (1974) finds that investors do not anticipate rating changes and that prices of corporate bonds react to rating change only with a small delay, Steiner and Heinke (2001) find evidence for significant price movements up to 100 trading days prior to the rating change. However, this seems only to be true for rating downgrades and not for rating upgrades. The results seem to be similar in the equity markets. Pinches and Singleton (1978) find abnormal equity returns both for up- and downgrades. Similar to the findings of Steiner and Heinke (2001), Griffin and Sanvicente (1982) find significant evidence for price changes only for rating downgrades and not for rating upgrades.

(13)

10

Hull, Predescu, and White (2004), Finnerty, Miller, and Chen (2013) and Micu, Remolona, and Wooldridge (2004) examine the impact of credit rating changes on CDS spreads. Hull, Predescu, and White (2004) base their analysis on credit rating announcements by Moody’s and find evidence that reviews for downgrades convey significant information, whereas both negative outlooks and rating downgrades do not contain any significant information. However, all three different types of rating announcements are anticipated by the CDS market. Similar to the findings in equity and bond markets, the results for positive rating announcements where much less significant. Finnerty, Miller, and Chen (2013) obtain similar results in a more extensive dataset for rating downgrades, negative outlooks and reviews for downgrades but note that their results for positive rating announcements differs from the results of previous studies. They find that rating upgrade announcements consistently had a highly significant impact on CDS spreads since 2003 although the impact of negative announcements is still much greater.

We will add to the latter literature by measuring the impact of rating changes on CDS demand rather than CDS spread. For instance, an increase in net demand for CDS around a negative rating change could suggest that banks and investors use the CDS market to hedge credit or counterparty risk.

3.2 HYPOTHESIS DEVELOPMENT

This paper aims to examine the determinants of CDS demand in order to draw conclusions on what type of companies are more often insured by CDS and whether CDS demand spikes around corporate credit rating changes. Therefore, this thesis adds to the literature that sheds lights on the motivation behind CDS trading and that examines the determinants of CDS demand. Previous literature on the activities of financial corporations in the CDS market predicts that CDS are used to hedge against credit or counterparty risk, perform regulatory arbitrage, capital structure arbitrage or for outright speculation by taking a view on the credit risk prospects of a company. Taking the outcome of the literature review about the potential use of CDS, empirical evidence on the determinants of CDS volume and the effect of CDS trading initiation on companies, I establish the following hypotheses.

Hypothesis 1: High credit risk companies (indicated by a lower Altman’s Z-Score),

(14)

11

As show in the reviewed literature several theoretical paper (e.g. Duffee and Zhou (2001)) predict that banks or investors might use CDS in order to hedge against credit risk of firms. Oehmke and Zawadowski (2016) find evidence for CDS as hedging tool by examining CDS position data. In the sovereign CDS market Berg and Streitz (2015) find evidence that CDS markets of riskier countries tend to be larger compared to the CDS market of safer countries. In contrast to that, Campello and Matta (2013) find a negative correlation between CDS protection and firm risk measured by Altman’s Z-Score which indicates that lower risk companies tend to have more of their debt insured via CDS than higher risk companies. However, following Duffee and Zhou (2001) and Oehmke and Zawadowski (2016) I would expect that ceteris paribus a higher company risk of the reference entity, measured by the companies Z-Score, should imply higher CDS demand reflecting a higher need to hedge. A failure to reject Hypothesis 1 would be suggestive of hedging as one motivation of CDS trading.

Hypothesis 2: CDS demand spikes around a rating changes.

Previous literature by Hull, Predescu, and White (2004), Finnerty, Miller, and Chen (2013) and Micu, Remolona, and Wooldridge (2004) shows that CDS spreads increase in anticipation of a negative rating announcement and decrease in anticipation of positive announcements. Again ceteris paribus I would expect that net CDS demand increases around a rating downgrade in case investors indeed use CDS as a tool to hedge increased credit risk. Furthermore, I would expect to see heightened gross CDS demand both for rating upgrades and rating downgrades as gross CDS demand is better interpreted as a measure of CDS volume.

Hypothesis 3: CDS demand tends to be higher for financial corporations such as banks,

dealers and monoline insurers in case of a rating change because high interconnectedness of the CDS markets demands increased counterparty risk management.

Getmansky, Girardi, and Lewis (2014) examine the single-name CDS market and find that the top 10 sellers of protection account for 73% of all contracts traded in 2012. They stress that a failure of one of these market participants may impose significant contagion effects and systemic risk. Furthermore, Siriwardane (2015) documents an even higher concentration of the CDS market after the financial crisis with only a handful of sellers and buyers dominating the market. According to a report of the European Central Bank (2009) the CDS market has experienced an increasing demand for CDS written on financial

(15)

12

companies with six dealers being among the most insured non-sovereign reference entities on the market. These papers suggest that when the underlying reference entity of a CDS is a financial corporation the effects on counterparty risk can be substantial, because it is most likely that the intermediary in the CDS market is a financial corporation as well. The need to hedge this counterparty risk could increase if the rating of a financial corporation changes. Therefore, a failure to reject Hypothesis 3 would be indicative for counterparty risk hedging activities in the CDS market.

Hypothesis 4: The spike in CDS demand is higher when the rating change alters the

company’s risk category status in the loan book of banks.

Based on Yorulmazer (2013) and Shan, Tang, and Yan (2016) I argue that banks might use CDS to perform regulatory arbitrage. If regulation requires to hold less regulatory capital for CDS-protected assets I would expect banks to actively engage in CDS trading to lower risk weighted assets both in the trading and in the loan book. Therefore, in case a reference entity is downgraded into a higher risk weight category I would expect that demand increases more compared to a downgrade that does not change the risk weight category of the company. In case we find a positive significant effect this would be indicative for banks using CDS as a mean to engage in regulatory arbitrage.

4 DATA AND METHODOLOGY

In order to test all hypotheses, a dataset on CDS demand is needed. As CDS demand is not readily available from a database we come up with a methodology to estimate demand for CDS ranging back to the end of 2003. The following section describes the construction of the main CDS demand variables on which I will base my empirical analysis as well as analysis methodology.

4.1 DATA AND VARIABLE CONSTRUCTION 4.1.1 CDS demand

The data for the dependent variables comes Depository Trust & Clearing Corporation (DTCC) and its Trade Information Warehouse (TIW). According to DTCC their proprietary TIW processes 98% of the global credit derivatives transactions and includes all major

(16)

13

global derivatives dealers and over 2500 buy-side firms in over 70 countries7. From this database all raw data for compiling our two main dependent CDS demand variables is extracted.

Data on net notional outstanding and gross notional outstanding are obtained from Section I (Table 6) which provides data on all live positions in the warehouse. New trades, full terminations and partial terminations are taken from Section III (Table 22 and Table 23) and change in gross notional outstanding by year of termination date can be obtained from Section II (Table 16).

Gross notional values are the sum of CDS contracts bought (or sold) for a single reference entity within one week. For instance, a CDS contract between buyer A and seller B of $10m notional will amount to a gross notional amount of $10m and not $20m. Net notional is defined as the sum of protection bought by net buyers and equivalently the sum of protection sold by net sellers in the market. For any given agent in the market the net buys and sells can be added up to obtain the net position of the agent. The sum of the positive net positions of buyers and the negative net position of sellers will add up to zero. Net notional outstanding is therefore the amount that will be collectively transferred between all buyers and sellers in case of a credit event. It is therefore considered to be the most precise measure of credit risk transferred on the CDS market.

New trades represent the gross notional amount of newly traded contracts within one week. Full terminations represent the gross notional amount of fully unwinded8 contracts (100% of the notional is terminated) within one week. Partial terminations are equivalent to full terminations with the difference that only a portion of the gross notional is terminated.

Change in gross notional outstanding by year of termination indicates the change in gross notional for different tenors. For instance, in 2015 when change in gross notional for 2020 as year of termination is -$10m this indicates that $10m of 5-year tenor CDS have been terminated in this week. The weekly data is available from December 05, 2008 in the case of gross notional amounts outstanding, net notional amounts outstanding and change in gross notional by year of termination. The data on new trades and full and partial terminations dates back to April 3, 2009.

The variables described above form the basis of our CDS demand variables. First, we compile a database with net and gross notional amounts outstanding per week and company

7 See: http://dtcc.com/derivatives-services/trade-information-warehouse

8 To unwind a contract means to close out the position by entering in an offsetting contract to your previous position.

(17)

14

and add new trades, full terminations and partial terminations. In the next step I identify each week which contains an IMM date (International Monetary Market). IMM dates are quarterly dates which most CDS use as their maturity date. These dates are standardized to be on the 20th of March, 20th of June, 20th of September and 20th of December. Standardized CDS contracts will jointly mature on one of these dates depending on when both parties entered in the contract. For instance, if buyer A and seller B agree to enter into a 5-year CDS contract somewhere in the period between January 1st 2005 and 19th March 2005 the contract will mature on March 20th 2010. If they enter a 3-year contract in April 1st 2005 the contract matures on June 20th 2008. This is an important feature for our demand estimation, because on each of these dates the amount of net notional and gross notional outstanding drops due to the maturing contracts. From this drop in CDS outstanding I estimate the demand for CDS in the past with the following methodology.

For net CDS demand, I first compute the change in net notional outstanding as this is my estimate of the maturing CDS from the week prior to an IMM date to the IMM with an IMM date. Here I have to make a first assumption, which is, that new trades of CDS that would alter the net notional amount outstanding will be in the week prior or after the IMM week. This would mean that the drop of net notional outstanding is solely due to maturing CDS in the IMM week. In the next step we have to estimate the CDS tenor distribution in order to come up with an estimate on how much of the maturing CDS is due to the different contracts. For instance, if we would assume that 100% of CDS contracts outstanding are 5-year contracts we could conclude, that $5m maturing CDS on 20th June of 2010 was due to an overall net demand for CDS in Q1 2005 of $5m. Chen et al. (2011) find that 47% of all single-name CDS transactions were traded in the 5-year tenor and that approximately 10% of CDS contracts were traded in 1-4 years tenor. Therefore, I assume that 40% of the maturing CDS comes from 1-4-year tenors (equally distributed) and that 50% of the maturing CDS comes from 5-year tenors. The remaining 10% comes from tenors >5 years which will be neglected in our CDS demand computation. With the tenor distribution and the estimate of maturing CDS it is now possible to compute CDS demand prior to 2008. The methodology is as follows: With $10m maturing CDS in September 20th 2010 we would add $5m CDS demand to Q3 2005 (50% 5-year tenor), $1m to Q3 2006 (10% 4-year tenor), $1m to Q3 2007 (10% 3-year tenor), $1m to Q3 2008 (10% 2-year tenor) and $1m to Q3 2009 (10% 1-year tenor). This methodology is applied to each IMM week from 2008 to 2015 to obtain a CDS demand estimate from 2003 onwards.

(18)

15

For gross CDS demand, the change in gross notional outstanding for each of the companies is computed. In contrast to the methodology with net CDS demand I subtract new trades and add full and partial terminations to account for newly traded contracts in each week. By doing so I account for newly traded contracts and terminations in the IMM week and are able to compute gross maturing CDS outstanding consistently. I will assume new trades and full and partial terminations to be zero for those companies that appear in Table 6 but not in Table 22 and 23. Volumes for remaining single-names, as reported on the bottom of Table 22 and 23 usually account for less than 1% of the entire reported CDS volume and is therefore negligible. After that I follow the methodology as done for the estimation for net CDS demand.

Gross and net CDS demand differ in what we can learn from them in our analysis.

Demand based on net notional represents the amount that will be transferred between all buyers and sellers in case of a credit event. A change in net demand will therefore alter the risk perception of the CDS market on a particular reference entity. In contrast to that, gross

CDS demand is better interpreted as a measure of trading volume as it does not allow to

draw inference on how the market perceives the risk of a certain reference entity (Lee (2012)).

4.1.2 Independent variables

All of the independent variables used to test the four hypotheses are based on two variables. First, Altman’s Z-Score as a measure for company bankruptcy risk and second a company credit rating variable that will indicate whether the company experienced a rating change in the particular quarter.

Following Altman (2013) we compute company Z-Scores with data obtained from CapitalIQ. To account for differences in industries we use Altman’s Z-Score for manufacturing companies, the Z”-Score for non-manufacturing companies and excluded financial companies from all analyses that include independent variables on the basis of company Z-Scores.

𝑍 − 𝑆𝑐𝑜𝑟𝑒 = 1.2𝑋1+ 1.4𝑋2+ 3.3𝑋3+ 0.6𝑋4+ 1.0𝑋5 𝑍" − 𝑆𝑐𝑜𝑟𝑒 = 6.56𝑋1+ 3.26𝑋2+ 6.72𝑋3+ 1.05𝑋4

(19)

16 X2 = Retained Earnings / Total Assets

X3 = EBIT / Total Assets

X4 = Market Value of Equity / Total Book Value of Debt X5 = Sales / Total Assets

For further analysis of CDS demand I will use indicator variables which specify whether the company is considered to be safe or in distress based on computed Z-Scores and relevant cutoffs based on Altman (2013). For manufacturing companies z-Score thresholds are: Distress: Z<1.81, Gray 1.82<Z<2.99, Safe: Z>2.99. For non-manufacturing companies z-Score thresholds are: Distress: Z<1.1, Gray 1.1<Z<2.6, Safe Z>2.6.

To test the impact of a rating change on the demand of CDS rating data on the Domestic

Long-Term Issuer-Credit Rating from Compustat Capital IQ Monthly Rating Updates for

all companies in our sample from 2003 to 2015 is obtained. In order to be able to test the impact of a rating change on quarterly CDS demand the rating variable is computed as follows. I assign numeric values to the corporate credit rating (e.g. 1 to “AAA”, 2 to “AA+”, 3 to “AA” etc. until 21 to “D”). The rating of the third month in each quarter will be equal to the quarterly rating. By doing so I lose all rating changes that are offset in one quarter. For instance, when company A is downgraded from BB to BB- in the first month of a quarter and then upgraded again to BB from BB-. However, I assume that this rarely happens.

On the basis of these quarterly credit ratings I compute several independent variables that are used in the analysis. Downgrade 1 (1 Step) is a dummy variable indicating whether the company experienced a downgrade over 1 step (e.g. from BB to BB-). Downgrade 2

(2+ Steps) is defined as a dummy variable indicating whether the company experienced a

rating downgrade over two or more rating levels (e.g. from BB to CCC). Upgrade 1 (1

Step) is defined as a dummy variable indicating whether the rating level increased by one

level or not. Upgrade 2 (2+ Steps) indicates whether the rating increased by two or more levels.

In order to test the implications of being a financial corporation on CDS demand I construct a Fin dummy variable on the basis of company SIC codes. The Fin indicator will be equal to one if the SIC codes is between 6000 and 6799 and zero otherwise.

In order to compute our risk-weight indicators we refer to the Bank for International Settlements (2004) risk-weight framework. According to that, a 20% risk-weight applies

(20)

17

for AAA to AA- rated corporate bonds, a 50% risk-weight for A+ to A- rated corporate bonds, a 100% risk-weight for bonds with a rating class between BBB+ and BB- and 150% risk weight for lower rate corporate bonds. We construct a dummy variable RWd being equal to one if the risk-weight category decreases (the company experiences a rating upgrade and the bank has to hold less regulatory capital) and zero otherwise and a second dummy variable RWi being equal to one if the risk-weight category increases (the company experiences a rating downgrade and the bank has to hold more regulatory capital).

4.1.3 Control variables

In order to test the relative relationship between the independent and dependent variables properly I introduce control variables mostly taken from literature on the determinants of CDS spreads as determinants of CDS spreads and CDS demand should be closely related. Das (1995) and Duffie (1999) follow a basic arbitrage pricing approach which claims that a CDS contract can be replicated with a long position in a par default free floating-rate note and a short position in a par floating rate note in the underlying reference entity. Huang and Zhou (2008) use structural models on the basis of the work of Black and Scholes (1973) and Merton (1974) to examine the determinants of the CDS premium. They use equity volatility, risk-free interest rate, leverage ratio and asset payout ratio as key variables in evaluating different structural models of credit risk. Controlling for leverage seems to be particularly important as Subrahmanyam, Tang, and Wang (2014) and Saretto and Tookes (2013) find that leverage tends to increase after initiation of CDS trading. Besides control variables from the CDS spread literature we follow Subrahmanyam, Tang, and Wang (2014) and include ROA as a measure for company profitability, ln(Assets) to control for firm-size, WC to control for working capital of the company and Ret as quarterly equity return. Furthermore, I add the fraction of unsecured debt to total debt as control variable as the need for CDS as a hedging tool should we smaller when the creditor is already protected by collateral (Peltonen, Scheicher, and Vuillemey (2014)). To control for factors that change over time but apply for all companies equally, e.g. risk-free interest rate, we control for time-fixed effects in our models. For some models I include company fixed effects to control for time-invariant variables that have not been measured but affect our dependent variable CDS demand (e.g. industry). For the control variables we would expect that an increase in ln(assets), ROA, Ret and unsecured debt decreases CDS demand. We would expect leverage and equity return volatility to have a positive influence on CDS

(21)

18

demand, as an increase in leverage or volatility might indicate higher risk companies. All rations are winsorized at the 2.5% level to account for potential outliers in the dataset. Please refer to Table I for a detailed description of the main control variables used in our regression analysis.

4.2 EMPIRICAL METHODOLOGY

The analysis of the data in order to test all three hypotheses is done via a set of simple ordinary least square regressions on a panel data set. Based on aforementioned hypotheses I want to test the impact of company risk variables on CDS demand, in order to derive conclusions on the role of the CDS market and how market participants use CDS. The dependent variables will reflect the demand for CDS per quarter on a single reference entity and the main independent variables will be Z-Score, a distress indicator variable, and a set of indicator variables that will distinguish whether the rating change was an up- or a downgrade and whether the rating changes by one, or two or more levels. Furthermore, I will assess whether the impact of the explanatory variables is significantly different when the reference entity is a financial corporation as I would expect if market participants use the CDS market to hedge against counterparty risk. CDS demand in the equation specifications stands for both net and gross CDS demand.

Hypothesis 1: High credit risk companies (indicated by a higher Altman’s Z-Score),

experience a higher demand for the CDS written on their debt.

To test the first hypothesis, I perform ordinary least square regression (OLS) analysis with CDS demand as dependent variable and indicator variables that indicate whether the company is in financial distress or not on the basis of Altman’s Z-Score cutoffs. I test both for net and gross CDS demand. This specification excludes financial companies from the analysis as Z-Score is only computed for non-financial companies. In Equation 1.1 I control both for time and entity fixed effects, return on assets (ROA), leverage (lev), logarithm of assets (ln assets), equity return (ret), working capital (WC) and equity return volatility

(vola).

(22)

19

To overcome potential issues that are due to differences in the amount of outstanding CDS on distress companies or safe companies, changes-on-changes regressions without entity fixed effects are performed. Fixed effects regression compare above average means of the dependent variable to above average means of the independent variables, which could lead to biased results in case outstanding CDS on e.g. distressed companies are already high and changes in demand might therefore be smaller than average. In Equation 1.2 we control only for time fixed effects, return on assets (ROA), change in leverage (lev), change in logarithm of assets (ln assets), equity return (ret), working capital (WC) and equity return volatility (vola).

1.2 ∆ CDS demandi,j = β0+ β1distressi,j+ β2safei,j+ γ ∆controlsi,j+ εi,j

We would reject Hypothesis 1 in case we do not find a significantly positive coefficient for out distress indicator variables. A failure to reject Hypothesis 1 would be indicative for CDS being used to hedge company risk. See Table III for results.

Hypothesis 2: CDS demand spikes around a rating changes.

To test the second hypothesis, OLS regression analysis with CDS demand as dependent variable and rating change indicators variables that indicate whether the company experienced a rating up- or downgrade and whether the rating change was over one or more rating levels is performed. Again I perform the analysis on both net and gross CDS demand. In equation 1.3 and 1.4 it is controlled for entity- and time-fixed effects, return on assets (ROA), change in leverage (lev), change in logarithm of assets (ln assets), working capital (WC), equity return (ret), equity return volatility (vola) and fraction of unsecured debt (unsecured debt). Similar to the methodology used for Hypothesis 1 I perform changes-on-changes regression on both variables.

1.3 CDS demandi,j = β0+ β1UG1i,j+ β2UG2i,j+ β3DG1i,j+ β4DG2i,j+ γ controlsi,j+ εi,j

1.4 𝛥 CDS demandi,j = β0+ β1UG1i,j+ β2UG2i,j+ β3DG1i,j+ β4DG2i,j+ γ Δ controlsi,j+ εi,j

Downgrade dummy 1 (DG1) is equal to 1 if the company experienced a rating downgrade over one rating level (e.g. from BB+ to BB) and 0 otherwise. Similarly, downgrade dummy 2 (DG2) will be equal to 1 if the company experienced a rating

(23)

20

downgrades over two or more levels (e.g. from BB+ to BB-) and 0 otherwise. Upgrade dummy 1 (UG1) and (UG2) are defined equally for rating upgrades. To accept Hypothesis 2 coefficients on the upgrade indicators should be negative and significant, whereas the coefficients for our downgrade indicator variables should turn out to be significantly positive when testing the impact on net CDS demand. Furthermore, the impact of a downgrade by two or more levels should have a greater impact on demand compared to a downgrade by one level. Testing the impact on gross CDS demand could lead to different results. As gross CDS demand can be considered to be a better measure of CDS trading volume than market risk transfer one could expect to find a positive coefficient both for upgrades and downgrades. I run all regressions for non-financial companies and financial companies separately to grasp potential different effects for financial companies. A failure to reject Hypothesis 2 would suggest that market participants react to rating downgrades by buying more CDS which would be indicative for a use for hedging. See table IV for results.

Hypothesis 3: CDS demand tends to be higher for financial corporations such as banks,

dealers and monoline insurers in case of a rating change because high interconnectedness of the CDS markets demands increased counterparty risk management.

The methodology for the third hypothesis is based on the methodology in Equation 1.4 to test Hypothesis 2. I interact our rating change dummies (UG, DG) with a dummy variable indicating whether the company is a financial corporation or not. UG and DG are defined as a dummy variables indicating whether the company experienced an upgrade or downgrade or not. I therefore try to assess whether financial corporations experience a higher CDS demand spike than regular companies, which would be indicative of the CDS market of a tool to hedge counterparty risk. I control for the same factors as in Hypothesis 2 and Hypothesis 1.

1.5 CDS demandi,j= β0+ β1UGi,j+ β2𝑈𝐺i,j+ β3UGi,j∗ FINi+ β4DGi,j∗ FINi+ β5FINi+ γ controlsi,j+ εi,j

1.6 ∆ CDS demandi,j = β0+ β1UGi,j+ β2𝑈𝐺i,j+ β3UGi,j∗ FINi+ β4DGi,j∗ FINi+ β5FINi+ γ ∆controlsi,j+ εi,j

I would accept our Hypothesis 3 in case I find positive and significant coefficients for the interaction terms of our financial corporation dummy variable and the downgrade dummies. Furthermore, I would expect to find a significantly positive coefficient for our

(24)

21

financial corporation dummy. Failure to reject Hypothesis 3 would be indicative counterparty risk management activities on the CDS market. See table VI for results.

Hypothesis 4: The spike in CDS demand is higher when the rating change alters the

company’s risk category status in the loan book of banks.

Similar OLS regression analysis as in Equation 1.5 is used to test Hypothesis 4. In Equation 1.6 I use RWi and RWd as dummy variables indicating whether a rating change lead to an increase in the risk-weight category (RWi) or a decrease in the risk-weight category (RWd).

1.7 CDS demandi,j= β0+ β1UGI,j+ β2DGI,j+ β3RWii+ β5RWdi+ γ controlsi,j+ εi,j

1.8 ∆ CDS demandi,j = β0+ β1UGI,j+ β2DGI,j+ β3RWii+ β5RWdi+ γ∆controlsi,j+ εi,j

Hypothesis 4 would be rejected in case the data does not show significant coefficients

for the risk-weight dummy variables. A failure to reject Hypothesis 4 would be indicative for CDS being used to perform regulatory arbitrage. See Table VIIfor results.

4.3 DESCRIPTIVE STATISTICS

The dataset includes 927 companies for which I am able to compute quarterly CDS demand from 2003q4-2015q4. The dataset is an unbalanced panel with gaps and includes 41.826 firm quarter observations. However, the dataset is complete for 68%, meaning we have data from 2003q4 to 2015q4 without any gaps. For 515 of these companies I am able to compute Altman’s Z-Score on a quarterly basis. And for 648 companies I am able to obtain quarterly rating data from S&P. Summary statistics for the major variables used in the regressions are presented in Table II.

[INSERT TABLE II HERE]

Panel A shows the mean for the industry classification dummies and the risk category dummies. 13% of all companies in our dataset are financial companies, 25% are

manufacturing and the remaining companies are classified to be non-manufacturing. With

49% almost halve of all sample companies fall into the distress category. 20% fall into the safe category and the remaining 31% in the gray area in between.

Panel B shows descriptive statistics for main independent variables. The mean Z-Score is at 1.682 with a median at 1.530. The average quarterly equity returns (Ret) for all our sample companies is at 2.2% and the average return on assets (ROA) is 2.1%. The mean

(25)

22

leverage of all companies is at 28% with leverage being defined as total long-term debt divided by total assets. Approximately 84% of the sample companies’ debt is unsecured and approx. 20% of total debt is bank debt. The sample companies have a mean revenue of $6.6bn and a mean EBIT of $980mn, the average market capitalization is at $28.6bn.

Descriptive statistics for the dependent variables are presented in Panel C. Average net demand for CDS ranges from $8.98mn for safe companies to $11.39mn for companies is the distress category. The average gross demand for safe companies is $160.27mn, for gray companies $199.26 and for companies in distress $212.83m. Scaled by total long-term debt as an approximation for insurable interest, net CDS demand and gross CDS demand decreases with the risk categories. Mean gross CDS demand scaled by total long-term debt for companies in the safe risk category is 0.0987, for companies in the gray risk category 0.0725 and for companies in distress 0.0637. For net CDS demand scaled by total long-term debt the mean is 0.0055 for safe, 0.004 for gray and 0.0032 for distress companies.

In Panel D I conduct one-way analysis of variance to compare the means of the CDS demand variables grouped by the company risk categories and the industry categories. Throughout all classifications the results indicate that there are significant differences between at least two of the three means. However, this test does not indicate between which two means.

5 RESULTS

In this section, the regression results as proposed in the methodology section are presented and discussed, separately for each of the hypotheses.

[INSERT TABLE III HERE]

I test the first hypothesis by regressing both net and gross CDS demand on dummy variables indicating whether the company is considered to be safe, in a gray area or in distress. Gray is the omitted variable in this specification. As described in Table I, the company indicator variables are based on the companies’ Z-Score. This sample excludes financial companies as the Z-Score models are not fit to measure company risk for financial companies.

Column (1) and (2) show fixed effects regressions with time and firm fixed effects. I find small evidence that net CDS demand, normalized by the lagged value of total long-term debt, is smaller for safe companies. This is indicative for some hedging in the CDS market as firms that are considered to be safe should experience less demand from hedgers.

(26)

23

As expected, coefficients for distress companies are positive in column (1) and (2) but lack any significance. Furthermore, the evidence suggests that CDS demand, both for net and gross demand, is decreasing in leverage, which is significant at the 1% level. This seems counterintuitive first, but seems reasonable as CDS demand is scaled by the lagged value of total long-term debt. Columns (1) – (4) show that equity return volatility is positively related to gross CDS demand which is consistent with what literature on CDS spread predicts. Collin-Dufresne, Goldstein, and Martin (2001) that the debt claim through owning a CDS is similar to holding a put option, whose prices is increasing in asset return volatility. In addition to that, the logarithm of total assets is negatively related to CDS both for net and gross CDS demand. This is significant at the 1% level and intuitive as we would expect that bigger companies are perceived to be safer.

In column (3) and (4) I remove firm fixed effects and perform changes on changes regressions. Similar, to what I find in column (1) and (2) I find that the change in net CDS demand seems to be lower for safe companies, although this lacks high significance. The control variables show a more mixed picture in column (3) and (4). Here I find a significantly positive coefficient for leverage, indicating a larger CDS demand in quarters where leverage increases. Furthermore, in column (3) the coefficient for equity return is negative with a significance at the 10% level. This suggests that the change in net CDS demand tends to be lower if the company experiences high equity returns. Throughout all specifications, coefficients for ROA are statistically insignificant and show mixed signs.

Overall, Table II suggest that there is some evidence that the CDS market is used as a tool to hedge company risk. However, the evidence is not very strong and we have to reject Hypothesis 1 at the 1% significance level.

In the following table I examine the impact of rating changes on CDS demand, again to draw inference on whether market participants use the CDS market to hedge company risk and how trading volume behaves around rating changes. According to Hypothesis 2 I would expect to find significantly higher CDS demand around rating changes.

[INSERT TABLE IV HERE]

In Table IV evidence on gross CDS demand seems to be stronger than for net CDS demand in Table III. In column (2) and (4) I find that downgrades are positively related to gross CDS demand at the 5% and 10% significance level. This effect seems to be stronger for a downgrade over two or more rating levels compared to a downgrade over one level in column (2). Interestingly, the analysis suggests that an upgrade by one level is also

(27)

24

indicative for higher gross CDS demand. This is not necessarily contradictive to the hypothesis as gross CDS demand can be interpreted as a better proxy for CDS volume than for credit risk transfer on the market. I find that a rating upgrade over two or more steps in column (3) is associated with a higher change in net CDS demand. The evidence on the control variables are similar to what I find in Table III. CDS demand decreases in firm size and leverage and increases in equity return volatility, unsecured debt (albeit not significantly) in column (1) and (2). Again this seems to be inconsistent with the changes-on-changes regression in (4) here we find small evidence that a change in logarithm of assets and a change in leverage is positively related to gross CDS demand. Furthermore, I find a significantly positive relationship between return on assets (ROA) and CDS, which is counterintuitive as I would expect that more profitable reference entities should attract less demand from hedgers.

Overall, Table III suggests that gross CDS demand is higher around both rating upgrades and rating downgrades. This is indicative of a higher CDS volume around rating changes as gross CDS demand is a better proxy for CDS volume than for credit risk transfer on the market. I confirm Hypothesis 2, that CDS demand increases around rating changes at the 1% level. However, I find no significance that net CDS demand, as a better proxy for risk transfer, changes around a change in company rating, which might indicate that the market is not used to transfer significant amounts of credit risk around a rating change. Please refer to the discussion section for a detailed discussion on what drives the validity of the empirical results.

[INSERT TABLE V HERE]

I run the same specifications again on financial corporations. Interestingly, all coefficients for the rating change variables loose significance and/or change the direction. Column (1) and (3) suggest that net CDS demand decreases for financial companies in case they experience a rating downgrade over two or more rating levels. In column (2) the coefficient for ROA matches my expectation in this specification as it indicates that gross CDS demand decreases with a higher ROA. An at the 10% significance level negative coefficient for unsecured debt is puzzling in column (2) as I would expect that CDS demand increases in the fraction of unsecured debt.

For Hypothesis 3 I argue that CDS demand should be higher for financial corporations around a rating change, as market participants might want to hedge counterparty risk. The results on the analysis of Hypothesis 3 are presented in Table VI.

(28)

25

[INSERT TABLE VI HERE]

Consistent with the results from the analysis on upgrades and downgrades for Hypothesis 2, the results in Table IV show significantly positive coefficients for upgrades and downgrades in column (2) and (4). This again suggests that CDS volume increases significantly around rating changes. The results suggest that there is no effect on net CDS demand in column (1) and (2). The coefficient for the financial corporation dummy (fin) shows consistently a positive relationship between being a financial corporation and CDS demand in column (1) and (2). These results are statistically significant at the 1% level for net CDS demand and gross CDS demand and at the 10% level for the change in gross CDS demand. This suggests that financial corporations experience higher CDS demand on their debt than non-financial corporations, possibly due to counterparty risk hedging activities. However, results from the coefficients for the interaction terms between rating changes and the financial corporation dummy variable in column (2) and column (4) indicate that the impact on rating changes is less pronounced for financial corporations compared to non-financials. This effect is stronger and more significant for downgrades compared to upgrades. This can possibly be explained with the interconnectedness of the CDS market and how this affects the demand for CDS on financial corporations. In case of a rating downgrade of a bank A, market participants may be reluctant to buy CDS on that bank because the rating downgrade of A most probably also affects bank B, that acts as counterparty for the CDS, through the interconnectedness of the CDS market. However, this would be true for all reference entities in case of a rating downgrade of a bank. Another explanation could be that the information revealed in a downgrade or an upgrade of a bank is less strong than for a non-financial corporation. In the spirit of Acharya and Johnson (2007) it might be possible to argue that dealers have particularly strong information on the other dealers in the market and thereby are not surprised by a rating change. Furthermore, Zawadowski (2013) predicts that banks are willing to trade CDS to reduce asset risk, but they might not be willing to reduce counterparty risk as counterparty risk hedging is to costly compared to its benefits. Overall, the evidence on financial corporations is mixed. Results from Table VI suggest that financial corporations experience a higher demand for CDS written on their debt, possibly due to a counterparty risk hedging, but in case of a rating change the impact on CDS demand is less strong when being a financial corporation. This is similar to what the results from Table V suggest.

Referenties

GERELATEERDE DOCUMENTEN

Understanding the signaling mechanisms that underlie the metabolic programming of the chondrogenic cell fate might al- low for the efficient production of either permanent

In short, birth cohorts can differ in their job satisfaction level because older cohorts are replaced by younger cohorts who have higher and more idealistic expectations, different

accessibility evaluation guidelines suggests to test more than 50 pages of any given site, we use the total number of unique problems found when testing 53 pages (13 + 4 times 10)

To make a single specification of AtomicInteger that can capture all exclu- sive access patterns, the specification is parametrised by (1) a set of roles, which basically is

This paper discusses the design and implementation of Sylvan, especially an improvement to the lock-free unique table that uses bit arrays, the concurrent operation cache and

By temporality is implied the dominant temporal orientations of situated actors towards either the past, present, or future that shape the way practical judgment unfolds

It can be concluded that a bond issue during a low business cycle is a valuable addition to the model explaining the credit default swap spread since the coefficient is significant

Credit rating announcements published before the financial crisis show large negative cumulative average abnormal returns for downgrades in all event windows except for