Loan renegotiations and credit default swaps : an empirical analysis

(1)

MSc Business Economics: track Finance

Loan Renegotiations and Credit Default Swaps:

an empirical analysis

Master Thesis by Emiel van der Wekken (10871349) Supervised by Dr. R. Matta May 2016

(2)

Page 2

Statement of Originality

This document is written by the student Emiel van der Wekken who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are 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)

Page 3 Abstract:

Credit default swaps (CDSs) are widely discussed in the existing literature. Hu and Black (2008a) argued theoretically that CDSs negatively affect loan renegotiations, because a lender that is “insured” with a CDS has less incentive to renegotiate the loan. In my thesis I empirically investigated this and found statistical evidence for the described relation. However, I did not find convincing statistical evidence for a similar relation between CDSs and borrower friendly renegotiations. The relation between CDSs and loan renegotiations has a reverse causality and thus has an endogeneity issue, because it can be expected that if a borrower is in financial distress, the lender wants to “insure” the increased risk with a CDS. I coped with the endogeneity, using an instrumental variable and a two stage least squares model with the instrument being the seasonality of CDSs. Namely all CDSs mature on an international money market date.

(4)

Page 4

1. Introduction

Loan renegotiations, which if successfully renegotiated result in loan amendments, are a critical topic in corporate finance theories as was pointed out by Roberts and Sufi (2009). To start my thesis I describe the aim of it one sentence: empirically investigate the effect of the derivative credit default swaps (CDSs) on the renegotiations of loans.

Before I continue my thesis, I will introduce the derivative CDSs and its definition: consider three different parties: A, B and C. Party A issues bonds, which are bought by party B. This exposes party B to risk; if party A defaults, party B looses (part of) his money, depending on the terms of the bonds. In order to takeaway that risk party B can acquire a CDS from a third party, party C. The CDS then is an agreement between B and C and “insures” the risk of party B. In practice, when party A defaults, party C has to pay a certain notional amount to B, which is agreed in the terms of the CDS. Hence, the default risk moves from party B to party C. In exchange for this, party B has to pay a periodic coupon to party C, which is called the spread of a CDS. In short, a CDS is a derivative that moves the risk moves from the initial lender to a third party, in exchange for a periodic coupon (White, 2013). Notable is that it is not necessary for the lender to own the bond; in that case we talk about “naked” CDS. The position the lender takes is in that case purely speculative and not hedging the default risk.

CDSs were first designed and traded in 1994 and from the beginning of the 2000’s the CDS outstanding notional amounts increased substantially and peaked obviously in 2007, when the CDS gross notional amount outstanding was 62 trillion USD. When the recent financial crisis hit, the CDS market changed and the outstanding notional amount decreased tremendously (Colozza, 2013), mainly because in the aftermath of the crisis many derivatives, including CDSs, received a lot of criticism from both the media and researchers. Especially, naked CDSs, which take a purely speculative position, were highly questioned. Moreover, CDSs became a so-called “hot-topic” among researchers.

Before the crisis, the move of default risk to a third party was argued to reduce the cumulative default risk for banks. Hedge funds and similar companies now could easily share these risks with the banks. However, the cons of CDSs are widely questioned in the literature by several researchers after the crisis. One of the first criticasters were Hu and Black (2008a), who explained that the creditor is empty when it has a CDS, meaning that it has less incentive to keep a borrower alive and not bankrupt. Bolton and Oehmke (2011) further elaborate this so-called empty creditor problem. The empty creditor problem argues that a buyer of CDS has less

(6)

Page 6

incentive and sometimes does not even want to renegotiate the terms of a loan with a financial distressed firm, because it is “insured” with its CDS. When the financial distressed firm defaults, the CDS buyer receives the compensation that is arranged with the seller of the CDS. Also, the seller of the CDS has an incentive to renegotiate the loan between the buyer of the CDS and the borrower, because he has to pay a compensation if the borrower defaults, but the seller is in no position to do make these renegotiations. Bolton and Oehmke (2011) argue that the buyer of the CDS has two options in case the borrower is in financial distress:

1. It can renegotiate the loan with the borrower

2. It can receive the compensation from the seller of the CDS

Simply put; if the payoff of the second option is larger than the payoff of the first option, the buyer of a CDS will choose the compensation instead of renegotiation. Hence, one can expect that CDSs lead to less loan amendments.

Besides findings on CDSs, more general findings on loan renegotiation are closely related to my thesis. Roberts and Sufi (2009) found among others the relation between credit ratings, assets and other accounting data with loan renegotiations. Moreover, the study of Roberts and Sufi (2009) is closely related to my research and I use their investigation of the effects on loan renegotiation in my control variables. After the recent financial crisis, the focus moved to the effect of securitization in general on distressed loan renegotiations. E.g. Agarwal, Amromin, Ben-David, Chomsisengphet and Evanoff (2011) and Piskorski, Seru and Vig (2010).

A successful loan renegotiation and an amendment of a loan are in practice very similar, because a successful loan renegotiation directly leads to an amendment of a loan. Theoretically, I mostly use the term loan renegotiations, but in my empirical analysis I mostly use the term amendments in order to make the boundaries somewhat clearer.

For the examination of my baseline hypothesis, the negative effect of CDSs of loan amendment (formally formulized in Section 3). With my model I found empirical evidence that CDSs do negatively affect loan amendments.

I use a probit regression model with the dependent variable being an amendment dummy and the independent variable being CDS net notional amount. However, it is arguable that a model where the effect of CDSs on loan renegotiations is tested has an endogeneity problem in the form of reverse causality. This means that besides the effect of CDSs on loan renegotiations, loan renegotiations can have an effect on CDSs, because a lender can anticipate to the financial distress of the borrower and insures its risk with a CDS (Peristiani & Savino,

(7)

Page 7

2011). In order to tackle the endogeneity, I use an instrumental variable and hence a two stage least squares model. The CDS net notional amount has a seasonal pattern, because the maturity date of the CDSs is set on International Money Market (IMM) dates. Thus, a CDS contract of 5 years does not mature on the same date 5 years later, but on the first IMM date after 5 years. I prepared dummies of these IMM date, which I use as an instrument for the CDS net notional amount.

The following sections of my thesis are structured as follows: in the second section I will discuss the related literature, continuing to the third section where I describe the methodology and the collecting and structuring of my data. Then in the fourth section I present basic statistics. Further, the results are discussed in the fifth section, which is followed by a robustness check in the sixth section. In the last and seventh section I end my thesis with a conclusion.

2. Literature review

The literature review of my thesis consists mainly of recent literature, although somewhat older literature can be used to understand more basic principles and/or confirm later studies. I divided the literature review in several sub sections, in order to make it a clearer overview, which helps to understand the theories. I start the review wide with a sub section about financial contract renegotiations, then I narrow the review down and continue with a sub section about related literature on CDSs follows. In the following sub section I conclude with literature about the relation between loan renegotiation and CDSs.

2.1. Financial contract renegotiation

In practice, it is most of the time the borrower’s initiative to amend a loan / financial contract. Borrowers want “to go outside the deal” for several reasons. E.g. a borrower may want to merge or acquire, pay dividends, is in financial distress, etc.

Moreover, the amendment of financial contracts is not an uncommon event, and the amendments can lead to significant changes in the terms of the financial contract. Hart and Moore (1989) already argued that long term financial contracts are not renegotiation proof. Main ex ante problems when creating a financial contract are asymmetric information, and unable to foresee contingencies. New information can alter the view on the credit quality, investment opportunities of the borrower, which possibly lead to renegotiations and actual

(8)

Page 8

changes in the original contract. Roberts and Sufi (2009) find that renegotiations depend on the economic climate. Renegotiations are pro cyclical and depending on the state of the financial institution its market. Furthermore, Roberts and Sufi (2009) show that ex ante contingencies are used for the intent of amendment, and not for reducing the probability of a renegotiation, which could be too costly.

Moreover, the borrower may be able to renegotiate more favorable terms, if the credit quality of the financial contract is increased during the term of the contract (Garleanu & Zwiebel, 2009). The main problem for the borrower here is that it is too hard (i.e. too costly) to make its increase in credit quality credible to the lender. Having outside options and alternative sources of financing could be possible ways to show the credibility. Hence, competition between lenders improves the bargaining power of the borrower.

Deterioration of the credit quality of the borrower can also lead to renegotiation of a financial contract. A lender cannot always foresee the deterioration, due to asymmetrical information and sudden market changes. Ideally poor performance leads to a higher fee for the lender, because of the extra risk. Hence, in that case the borrower has to pay some form of penalty. This increases the likelihood of bankruptcy for the borrower that is already facing financial difficulties and bankruptcy of the borrower is an even worse scenario for the lender. Thus, the threat of future bankruptcy of the borrower is no longer important, because it is ex post Pareto-inefficient. In turn, the lender wants to renegotiate the initial terms, because it is a good option to prevent the borrower’s bankruptcy (Roberts & Sufi, 2009). In line with this conclusion, Jostarndt and Sautner (2010) empirically investigate successful financial contract renegotiations for German firms in financial distress. They find that companies with more debt, thus having higher leverage, have a higher probability to successfully renegotiate the financial contract with the lender, because these companies owe the lender more debt, which in turn makes the lender more concerning about these companies.

The original terms of the contract are important for later renegotiations. Making the party with the most bargaining power not exogenously, but on forehand arranged in the contract. That is, each possible contingency must be contractually specified. Aghion, Dewatripont and Rey (1994) show the “renegotiation game”, as this phenomenon is called, leads to the right marginal incentives for both the lender and the borrower.

Covenants can lead to amendments in the financial contract as well, in order to anticipate on a covenant breach or a direct covenant breach. Nowadays, most of the financial contracts have some sort of relation to changes in market conditions (such as a floating rate, or hedging

(9)

Page 9

strategies by the borrower) (Roberts and Sufi, 2009). Moreover, Roberts (2015) find that approximately a quarter of renegotiations are due to a (anticipated) covenant breach. Also, covenant breaches are likely to successfully lead to the renegotiations. Namely, three quarter of all covenant breaches lead to renegotiations.

After the financial crisis, the relation between securitization in general and renegotiation decisions by lenders became a topic in the loan renegotiation literature. Piskorski et al. (2010) empirically proof the existence of such a relation. Agarwal et al. (2011) find similar proof that successful renegotiation of loans is harder with securitization. Different from Piskorski et al. (2010). Agarwal et al. (2011) solely focus on mortgage renegotiations. This emphasizes that CDSs, which is a form of securitization, is likely to have an effect on loan renegotiations. Although both articles use a fairly different methodology than I present in my research.

2.2. Related literature on CDSs

As previously mentioned, CDSs became a hot-topic in the media and with researchers after the recent financial crisis. Also, the market of CDS has a substantial notional amount trading, making it an important topic for traders. This resulted in publications of many media and academic articles about CDSs. Although these publications are interesting literature, they do not fall directly within the scope of my thesis and have limited added value. Hence, I do not discuss such articles in this literature review. Instead I focus solely on the directly related CDS literature.

Despite their law point of view, Hu and Black (2008a) discuss in their article relevant CDS theories for my thesis. Hu and Black (2008a) is an extension on their previous articles (Hu and Black, 2006, 2007, 2008b). Hu and Black (2008a) argue that there is a simple tradeoff for lenders: if a company that is in financial distress is worth more when it is not bankrupt, compared to when it is bankrupt, the lender is willing to reorganize the financial contract. Vice versa, when the company in financial distress is worth more bankrupt, compared to when it is

not bankrupt, the lender is not willing to reorganize. When there is no CDS in the tradeoff,

financial contracts, with companies that are actually worth more when they are not bankrupt, will be renegotiated. A CDS changes this scenario, then it becomes possible for the lender to gain from a bankruptcy even if the company is under normal circumstances worth more when it is not bankrupt. This results in that the financial contract is less likely to be renegotiated, if the lender has a CDS. This phenomenon is also called the empty creditor problem. Moreover, Hu and Black (2008a) argue that the market of debt securitization is “fairly new” and “imperfectly known”, which confirms the importance of my topic.

(10)

Page 10

Potential incidences of the empty creditor problem include large companies such as Six Flags, GM and Chrysler. All three companies failed to renegotiate a deal with its lenders, despite that the renegotiation deal was at first sight the better payoff. It is likely that the lenders had CDSs, resulting that defaulting of the companies was the better payoff, instead of the renegotiation with the lenders (Bolton & Oehmke, 2012). Besides Six Flags, GM and Chrysler, there are many similar examples, which emphasize the importance of the problem even more. Campello and Matta (2012) are more critical to the empty creditor problem and discuss that the relation between CDS and the economy is needed in order to empirically find proof for the phenomenon. Furthermore, they discussed that buyers of CDS are likely to exceed the optimal amount of CDS insurance during economic prosperity, and vice versa, meaning that the empty creditor problem and CDSs are pro cyclical.

On the contrary to the legal article by Hu and Black (2008a), which examines the ex post inefficiencies of CDSs. Bolton and Oehmke (2011) discuss ex-ante results relating CDSs. They find that CDSs result in that existing can be more efficient and they increase investments. On the contrary, they also find that lenders tend to over-insure with CDSs, even when the CDS market is able to perfectly anticipate on the negative ex post inefficiencies as described by Hu and Black (2008a), again resulting in too little renegotiations and too much bankruptcies.

Peristiani and Savino (2011) also empirically investigate the problem of Hu and Black (2008a), implying that companies who insure their debt with CDSs are more likely to go bankrupt than companies without CDSs. Their time frame, 2001 – 2008 is slightly longer than mine, but almost entirely exclude the years after the recent financial crisis. Also, they did not find a relation between corporate defaults and CDSs in this time frame. However, they did find the relation in a shorter time frame, 2004 – 2008.

Similar to Peristiani and Savino (2011), Subrahmanyam, Tang and Wang (2012) empirically investigate CDSs and its relation bankruptcies. In complement to Peristiani and Savino (2011) they also investigate the relation between CDSs and credit risk. Unlike Peristiani and Savino (2011) they find strong evidence that both the probability of a down grade in credit risk and the probability of a bankruptcy increases when CDSs trade begins. Moreover, after the inception of CDS trading the probability of bankruptcy doubles. Their sample consists of North American companies in the time frame June 1997 – April 2009

In line with this, but rather using a different methodology, Danis (2013) finds evidence consistent with the empty creditor hypothesis, meaning that the incentives of bondholders

(11)

Page 11

change, if they are insured by CDSs. In that case, it is more difficult for financially distressed borrower to reduce the debt.

On the contrary, Bedendo, Cathcart and El-Jahel (2014) do not find evidence for the same empty creditor hypothesis. They empirically investigated this using a sample of US companies in the time frame 2007 – 2011. Their main hypothesis is that in order to avoid bankruptcy, borrowers offer better renegotiation deals with lenders that are insured with CDSs. Hence, the outcome of the renegotiation is less borrower friendly. Although, as mentioned, they do not find evidence that confirm this.

Besides the main argument discussed so far that CDSs can lead to less successful renegotiations, one can also argue that firms in financial distress lead to more CDSs, because if a lender observes that the borrower is going to or is suffering financially, the lender has an incentive to buy insurance with a CDS, anticipating on the situation. This results in a reverse causality issue, which is an endogeneity problem in my model (Peristiani & Savino, 2011).

O’Kane (2011), White (2014) and Colozza (2013) explain that CDSs have a standardized maturity date. The maturity date matches with the International Money Market (IMM) date and is on the 20th_{of March, June, September and December, sometimes it also}

referred to as the third week of those months. This results in a seasonal pattern in the net notional amount, which I use as an instrumental variable to solve the endogeneity problem. Moreover this issue is discussed and explained in section 4 about methodology and data.

2.3. Summary and conclusion

The theory and findings on renegotiations of financial contracts are of importance to both lenders and borrowers. Initially, articles such as Roberts and Sufi (2009) and Jostarndt and Sautner (2009) discussed this and found that asymmetrical information and leverage are drivers of renegotiations. Especially, the model of Roberts and Sufi (2009) is closely related to my thesis and I use their methodology as a guideline for my empirical methods.

With the emerging credit securitization, including CDSs, starting begin 2000, there is a third party, the seller of the CDS, who need to understand the theory and findings on renegotiations of financials contracts with CDS insurance. This previous related literature on CDSs is mainly focused on the relation between default and CDSs. Instead I focus on the relation between loan renegotiations and CDSs, which is slightly different. Also, the empirical investigated literature not all lead to the same results and are contradictory, resulting in that the topic is not yet properly investigated and making it an interesting topic for my thesis.

(12)

Page 12

An important conclusion of the literature review is that one can expect that CDSs have a negative impact on loan renegotiations, because a lender who is insured with a CDS has less incentive to renegotiate the financial contract, than a lender who is not insured with a CDS. I.e. when a borrower defaults, the payoff of the lender who is insured with a CDS is higher compared to the payoff of the lender who is not insured.

Noteworthy is that the literature review frequently consists of working papers besides references to A-journal literature. This is especially the case in the Section 2.2 that focuses on the literature of CDSs. Although, the authors are mainly researchers at well-respected universities. Also, this emphasize the fact that the territory is relatively undiscovered, and hence I have an original topic for my thesis.

3. Methodology and data

In this section, I will first formulize my two hypotheses with a brief background based on the literature of Section 2. In the second sub section, I discuss the econometric methods that I use to test these hypotheses. This also includes discussing the endogeneity issue of CDSs. In the last and third sub section I introduce my data and how it was collected and structured.

3.1. Hypotheses

As previously elaborated, the focus of my thesis is on the relation between CDSs and loan renegotiation. More specific, the effect of CDSs on loan renegotiations. Theoretical literature implies that the effect of CDSs is negative, because the CDS insures the lender from the borrower defaulting, resulting in a relatively higher payoff when the borrower defaults, compared to a lender without a CDS. Hence, the odds for successful financial contract renegotiations and thus loan amendments are lower when the lender has a CDS, leading to my baseline and first hypothesis:

Hypothesis 1 Companies whose debt is insured with CDSs have a lower probability to amend a financial contract, than companies whose debt is not insured.

From a different angle, one can make a similar argument regarding a borrower friendly outcome of the renegotiation. In other words a lender with protection from a CDS is less willing to renegotiate the loan. In exchange, the borrower offers more to compensate the higher payoff. Hence, the bargaining power of the lender with protection from a CDS is relatively high, compared to the lender without a CDS. This all leads to the conclusion that CDSs have a

(13)

Page 13

negative impact on borrower friendly renegotiations, which is formulized in my next hypothesis:

Hypothesis 2 CDSs have a negative impact on a borrower friendly outcome of the financial contract amendment.

3.2. Econometrical methods and endogeneity

As discussed in the literature review, CDS is likely suffering from reverse causality and thus is there is an endogeneity problem. To cope with the problem, I use the seasonality of CDSs as an instrumental variable to predict the CDS net notional amount. The seasonality refers to the fact that CDSs mature on one of in total four International Money Market (IMM) dates. The IMM dates are in the third week of March, June, September and December. E.g. a 5 year CDS agreement written on February 1, 2016, matures in the third week of 2021. Hence, on those dates both the net and gross notional amounts outstanding of CDSs decrease substantially, resulting in a seasonal pattern of the notional amounts of CDSs. Hence, I can use the IMM dates as an instrumental variable to predict the CDS net notional amounts and then I use a regression of the predicted notional amounts on an amendment dummy. This method is called a two stage least squares model. The first stage is defined as in formula 1, below.

𝑛𝑒𝑡_𝑛𝑜𝑡𝑖𝑜𝑛𝑎𝑙_𝑡,𝐶 = 𝜋₀ + 𝜋₁∗ 𝐼𝑀𝑀_𝑑𝑎𝑡𝑒1_𝑡,𝐶 + +𝜋₂∗ 𝐼𝑀𝑀_𝑑𝑎𝑡𝑒2_𝑡,𝐶 + 𝜋₃∗

𝐼𝑀𝑀_𝑑𝑎𝑡𝑒3_𝑡,𝐶+ 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝑣_𝑡,𝐶 (1) Where 0 is the intercept and v is the error term. This first stage of the two stage least squares

predicts the net notional CDS amount outstanding, net_notional. The IMM date variables are dummy variables that equal 1 on an IMM date and 0 otherwise. Hence I created four IMM date variables, each equal 1 on a different IMM date.

Now that the first stage formula is set out, I can focus on the two hypotheses and in order to successfully test the first baseline hypothesis discussed in Section 3.1., I use a probit regression model, because the dependent variable, an amendment dummy, is binary. More specific, the outcome of the dummy can be either that a loan is amended or is not amended, respectively 1 or 0. With a binary dependent variable, the probit model is a good fit, because it makes more sense to implement a nonlinear regression model than a linear model (Stock & Watson, 2012). Then in formula 2,  is the cumulative standard normal distribution function.

(14)

Page 14

The independent variable comes out of the first stage and is the predicted CDS net notional amount. I also include control variables to prevent an omitted variable bias. These are based on the model of Roberts and Sufi (2009). The observations are on a week company level, which is discussed further in 3.3. Hence, the t represents the week and C the company. Moreover, a specific description of the control variables in given in the next sub section. The model is defined as follows:

Pr(𝑙𝑜𝑎𝑛_𝑎𝑚𝑒𝑛𝑑𝑒𝑑_𝑦𝑒𝑎𝑟𝑡,𝐶 = 1| 𝑛𝑒𝑡_𝑛𝑜𝑡𝑖𝑜𝑛𝑎𝑙̂ 𝑡,𝐶, 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠) = Φ(𝛽0+ 1∗

𝑛𝑒𝑡_𝑛𝑜𝑡𝑖𝑜𝑛𝑎𝑙̂ _𝑡,𝐶+ 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠) (2)

In order to test the second hypothesis, I use the same first stage of the two stage least squares model, which is the predicted CDS net notional amount, as defined in formula 1. The second stage in this model is an index that represents the level of borrower friendliness of the amendment. The index has a range between -3 and 3. An amendment is classified as borrower friendly, if the index is positive, and vice versa an amendment is classified as non-borrower friendly, if the index is negative. Also, a larger positive index number indicates borrower friendlier amendment and a larger negative index number indicates a less borrower friendly amendment. Hence, -3 and 3 are respectively the least borrower friendly number and the borrower friendliest number. In this regression the OLS applies and similar to formula 2, I also included control variables. Moreover,  is the constant and  is the error term. The model is defined as follows:

𝑎𝑚𝑒𝑛𝑑_𝑖𝑛𝑑𝑒𝑥_𝑡,𝐶 = 𝛼 + 𝛽₁∗ 𝑛𝑒𝑡_𝑛𝑜𝑡𝑖𝑜𝑛𝑎𝑙̂ _𝑡,𝐶+ 𝑐𝑜𝑛𝑡𝑟𝑜𝑙 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝜀_𝑡,𝐶 (3)

3.3. Data

The dataset that I used contains 300,499 observations. Each observation is a company in a week of the year. The datasets consists of 1,797 US companies in the time frame from week 44, 2008, until week 12, 2013. I used US companies, because the data is only available for US companies. Although, CDSs are for a large part traded in the US, making it a good sample.

3.3.1. Collecting and structuring the data

For the large part, the data was collected by my thesis supervisor, Dr. R. Matta, and his fellow researchers from several different databases. First, the data on loans and the amendments was collected from Thomson One. Second, the CDS data is collected from DTCC, and thirdly accounting data for the control variables is collected from Standard and Poor’s Compustat, which was collected by myself.

(15)

Page 15

All three the datasets contain the variable Global Company Key Identifier or gvkey, which is a unique number that is assigned to each company. Similar to for example a TICKER symbol. First, the loans outstanding and accounting data were merged on a year-company basis, meaning that each observation represents a company in a given year. Afterwards I expanded this dataset in a way that observations became on a week-company basis. The expansion means that I assumed that the values of a variable in a specific year are the same for all weeks. It was necessary, because in that way I could merge this dataset including loans outstanding data and accounting data with the CDS data, which was already structured on a week-company basis. In the final merged dataset each observation represents a company in a certain week in the time frame week 44, 2008 until week 12, 2013, resulting in that a company can have multiple observations in multiple weeks throughout the years.

The amendment dummy is prepared based on the amendment date. If the amendment date was in that specific week the amendment dummy equals 1 and in all other scenarios the dummy equals 0.

As discussed in Section 2. Literature review, Roberts and Sufi (2009) investigated several effects on loan renegotiation. In order to cope with an omitted variable bias, I use their statistically and economically significant variables as control variables. These include market capitalization, the logarithm of the asset, the book leverage and the EBITDA/assets ratio. Like Roberts and Sufi (2009), I made a quarterly lagged variable for these variables, in order to capture the effect. I also use their formulas to calculate these control variables with the accounting data, presented in Appendix I. Moreover, the control variables are checked on outliers and are Winsorized, if necessary.

Moreover, following Roberts and Sufi (2009), I also included fixed effects. These include the industry, credit rating and year fixed effects. The year fixed effect is self-explanatory, the industry fixed effect is based on the Fama French 12 industry standard and the credit rating is the S&P long term credit ratings. The credit rating is alternated in an index as follows between 0 and 4. A higher index indicates a better credit rating.

4. Basic statistics

The summary statistics of the dependent and independent variables are presented in Table 1. As discussed in the previous section, the observations of the dataset are on a week-company level and the dataset runs from week 44 in 2008 until week 12 in 2013. The dependent

(16)

Page 16

variable is a dummy variable that equals 1, if the company has amended a loan in that fiscal year.

Panel A of Table 1 shows that of all observations 82% does not have a CDS trading and 18% have a CDS trading. Moreover, at first sight there are rather a few amendments in the data set. However, this can be clarified, because observation is on week-company level and the amendment dummy only equals 1, if the amendment is made for a company in the specific week. This also leads to a low ratio amendments divided by observations in the last column. Relatively more amendments are made for observations with CDS than without CDS. The fractions are respectively 0.0011 and 0.0009.

The difference between CDS gross notional amount and CDS net notional amount is that the net notional amount is the net position of the company’s notional amount. For example a company buys a CDS for 10 USD, but the company also sells a CDS for 5 USD. The net position is then 10 USD minus 5 USD equals 5 USD. In practice, we always see a much larger CDS gross notional amount, as is the case in my data set, which can be seen from Panel B of Table 1. Logically for companies without CDSs trading, the notional amounts are 0.

As discussed in Section 3. Methodology and data the amendment index is a borrower friendly index. An amendment is classified as borrower friendly, if the index is positive, and vice versa an amendment is classified as not borrower friendly, if the index is negative. Panel B of Table 1 shows that the amendment index mean of observations with CDS is lower than the mean of observations without CDS. This is not yet a result, but it follows my second hypothesis that CDSs have a negative impact on borrower friendly amendments.

(17)

Page 17

Table 1: Loan renegotiations, CDS and borrower friendly renegotiations summary statistics

The observations are on a week-company level, meaning that each observation represents a company in a given week. Panel A presents the observations and how many of them are amended divided into observations with CDS, without CDS and total. The observations with CDS account for approximately 18% of the total dataset and of all the amendments 21% include a CDS agreement. Moreover, 0.11% of the observations with CDS is amended. This seems rather low, but keep in mind that the observations are on a week company level, resulting in that there is only an amendment, if a company amended a loan in that specific week. The 0.11% is slightly more compared to the observations without CDS and the total, both 0.09%. Panel B presents the mean, standard deviation and median for dependent and explanatory variables for observations with CDS, without CDS and all the observations. The amendments is similar to what I discussed in Panel A. Logically the CDS notional amounts are all 0 for observations without CDS. The amendment index is a borrower friendly indices that is positive if the amendment is borrower friendly and is negative if it is not borrower friendly. Also, the more positive the index, the friendlier to the borrower and vice versa the more negative the index, the less borrower friendly.

Panel A: CDSs and loan renegotiation

Observations with CDS Observations without CDS Total

Panel B: Amendment and CDS characteristics

Mean Median Mean Median Mean SD Median

Amendments 0.0011 0.000 0.0009 0.000 0.0009 0.030 0.000

CDS gross notional amount ($ mln.) 11,391 9,162 0.0000 0.000 2,024 5,513 0.000

CDS net notional amount ($ mln.) 894 753 0.0000 0.000 159 418 0.000

Amendment index 0.722 1.000 1.050 1.000 0.983 1.517 1.000 0.000 0.000 8,021 573 1.542 1.509 SD SD 0.033 0.030 301,309 1.000 275 1.000 0.0009

Observations with CDS Observations without CDS Total observations

53,401 0.177 57 0.207 0.0011

247,908 0.823 218 0.793 0.0009

All observations Amendments Amendments/

observations

(18)

Page 18

Table 2 shows characteristics of the observations of my dataset. In Panel A of Table 2 the mean, standard deviation and median for the credit rating and the control variables are presented Unlike as in the regressions in the next section, the credit rating here is not an index with 5 baskets. It is the same index and equals 20 with an AAA rating and 0 with unrated. Every credit rating value between these values is set pro rata. In Panel B of Table 2 the industry distribution of the companies according to the 12 industries of Fama French is shown. In total all observations include 1,797 companies.

Table 2: Characteristics of the observations and company distribution across industries

The table shows summary statistics of the observations and companies. Panel A shows the mean, standard deviation and median for the credit rating and the control variables. Panel B shows the distribution of all 1,797 companies included in the dataset according to 12 industries as classified by Fama and French.

The seasonal pattern of CDS can be seen from the data as well. My data runs from 2008 until 2013, but for illustration purposes I take 2011 as an example. Figure 1 shows that before the IMM week (i.e. week 12, 25, 38 and 51) the average CDS net notional amount of all companies declines.

Panel A: Observation characteristics

Mean SD Median

S&P Credit rating 10.83 3.34 11.00

Market capitalization 6099 18922 1274

Log assets 7.549 1.606 7.470

Book leverage 0.303 0.256 0.266

EBITDA/assets 0.081 0.091 0.072

Panel B: Company distribution according to Fama and French

Business equipment

Chemicals and allied products Consumer durables

Consumer nondurables Finance

Healthcare, medical equip. and drugs Manufacturing

Oil, gas, and coal

Telephone and television trans. Utilities

Wholesale, retail, and some srvs. Other Total companies 0.214 1,797 0.152 0.043 0.053 0.135 0.070 0.020 Companies 0.098 0.033 0.059 0.087 0.035

(19)

Page 19

Figure 1: Average CDS net notional amount of all companies in 2011

After the IMM weeks (week 12, 25, 38 and 51) the average CDS net notional amount drops, because the maturity dates of CDSs is set on IMM dates. As an example I used 2011, but the maturity date of CDSs are standardized on the IMM dates.

5. Results

I have used an instrumental variable to cope with the endogeneity issue of CDS, which is in-depth described in Section 2 and Section 3. The results of the two stage least squares model is discussed in this section and the section is organized such that 5.1. presents the results of the first stage of the model and 5.2. follows with the second stage of the model. In order to test both hypotheses, I prepared two second stage regressions, respectively discussed in 5.2.1. and 5.2.2.

5.1. First stage regression

The first stage regression of the two stage lease squares model has as independent variable dummies equaling 1 on the IMM dates regressed on the dependent variable, CDS net notional amount in millions of USDs. The control variables and fixed effects are also included,

(20)

Page 20

as is discussed in Section 3. Methodology and data. The results of the first stage presented in the regression table below:

Table 3: First stage regression: effect on the CDS net notional amount

The table shows the first stage of the two stage least squares model. The dependent variable is the CDS net notional amount in millions of USD and is predicted with this model. Then it can be used in Table 4, 5 and 6. The four International Money Market (IMM) dates are the independent variables and the remaining variables are control variables. The IMM dates are dummy variables that equal one on each a different IMM date. CDS contracts mature on an IMM date, which results in the negative effect of IMM dates on CDS net notional amounts. The other variables are control variables. Off all four regression all variables are statistically signification with one exception the IMM date (3) in regression (4). In order to predict the CDS net notional amount, I used regression (1), because it has the highest R2_{and the IMM dates are both statistically and economically} significant. P-values are in parentheses (* significant at 10%, ** significant at 5%, and *** significant at 1%). Heteroskedatisticity is taken into account, which resulted in the use of robust standard errors.

The effect of the IMM dates is, as expected, negative on the CDS net notional amount. More specific in regression (1), which includes all three fixed effects. Following regression (1), the CDS net notional amount is 17.67 million USD lower, if IMM date (1) equals one and the other variables are constant. The same principle applies to the other IMM dates. Hence, if IMM date (2), (3) or (4) equals one, the CDS net notional amount is respectively 15.92, 10.85 or 9.75

(1) (2) (3) (4) IMM date (1) -17.67*** -17.21*** -14.81*** -8.044* (0.000) (0.000) (0.001) (0.067) IMM date (2) -15.92*** -14.38*** -12.51** -17.01*** (0.001) (0.004) (0.011) (0.000) IMM date (3) -10.85** -10.72** -9.654* -10.90** (0.025) (0.029) (0.050) (0.025) IMM date (4) -9.751** -10.37** -10.54** 4.635 (0.028) (0.022) (0.020) (0.296) Control variables Market capitalization 0.00348*** 0.00342*** 0.00388*** 0.00343*** (0.00) (0.00) (0.00) (0.00) Log assets 105.5*** 124.8*** 98.99*** 102.4*** (0.00) (0.00) (0.00) (0.00) Book Leverage 73.77*** 42.98*** 67.17*** 78.24*** (0.00) (0.00) (0.00) (0.00) EBITDA/Assets -311.2*** -291.2*** -238.5*** -343.5*** (0.00) (0.00) (0.00) (0.00) Fixed effects

Credit rating Yes No Yes Yes

Industry Yes Yes No Yes

Year Yes Yes Yes No

Observations 280,370 280,370 280,370 280,370

R2 0.355 0.333 0.329 0.347

(21)

Page 21

million USD lower. This is in line with the expectations and the results are both statistically and economically significant.

Moreover, in the other three regressions the IMM dates are also statistically significant, except for regression (4). In that regression I did not control for the year fixed effect, resulting in that IMM date (3), is not statistically significant. More specific, IMM date (1) is statistically significant at the 1% level in regression (1), (2) and (4). In all other cases the IMM date variables are statistically significant at the 5% level. The control variables are all statistically significant at the 1% level. Hence, the regressions (2) and (3) can be interpreted the same as regression (1) from Table 3.

In order to predict the instrumental variable, CDS net notional amount, which I require for the second stage of the model, I used regression (1), because it includes all the fixed effects, making it theoretically best option. Also, statistically regression (1) is a viable option.

5.2. Second stage regressions

With the regression made in 5.1. First stage regression, I predicted the CDS net notional amount that I use in the second stage of my two stage least square model. I use the instrumental variable for testing both hypothesis.

5.2.1. Results of hypothesis 1

The results of the second stage probit regression are presented in Table 4. Moreover, it shows that the CDS net notional amount has a negative impact on the amendment of a loan. This can be concluded from regression (1) and (4), because these regressions are statistically significant at the 1% level, and not from regression (2) and (3), because these regressions are statistically insignificant.

The results of regressions (1) and (4) are in line with my expectation and hypothesis. The pseudo-R2 is fairly low in all 4 regressions, although I included a set of control variables, following Roberts and Sufi (2009). This means that in regression (1), (2), (3) and (4) the independent variables, including control variables and fixed effects, predicts the dependent variable for respectively 2.7%, 2.6%, 2.3% and 2.0%.

I cannot interpret these results yet, because the results are of a probit regression, which is a nonlinear model. In order to interpret the results, I derived the marginal effects, which are presented later on in Table 5. In Table 4’s regression (1) which includes all three fixed effects, the CDS net notional amount is statistically significant at the 1%. This is also the case for regression (4) that excludes the year fixed effects. However, in regressions (2) and (3),

(22)

Page 22

respectively without credit rating and industry fixed effects, the CDS net notional amount is not statistically significant, so are most of the control variables, and thus these results cannot be used to draw any conclusions.

Table 4: Second stage probit regression on the amendment dummy

Table 4 presents the second stage of the two stage least squares model. The model is a probit regression with as dependent variable an amendment dummy that equals one, if an amendment was made in a specific observation. The independent variable is the estimated CDS net notional in millions of USD, which is predicted using the first stage presented in Table 3. The statistically significant regressions, (1) and (4), show that CDS net notional amount has a negative impact on the amendments of loans. However, the numbers cannot be interpreted, because the probit regression is nonlinear. The marginal effects are presented in Table 5. The table shows the P-values are in parentheses (* significant at 10%, ** significant at 5%, and *** significant at 1%). Heteroskedatisticity is taken into account, which resulted in the use of robust standard errors.

The marginal effects of the probit regression presented in Table 4 are presented in Table 5 below. These results can be interpreted, and as mentioned the dependent variable is an amendment dummy that equals 1 if a loan is amended in that particular week and 0 otherwise. In turn, the independent variable is the CDS net notional amount in USD millions. It logically follows that for companies without CDSs, this amount equals 0.

(1) (2) (3) (4)

Predicted CDS net notional amount -0.00418*** -0.000153 -0.000058 -0.00307***

(0.000) (0.583) (0.822) (0.000) Control variables Market capitalization -0.000017*** -0.0000020 -0.0000019 0.0000085*** (0.00) (0.246) (0.303) (0.00) Log assets -0.394** 0.0803** 0.0569* 0.371*** (0.033) (0.037) (0.069) (0.00) Book Leverage -0.297* 0.050 0.010 0.251** (0.061) (0.549) (0.915) (0.011) EBITDA/Assets 1.015** -0.309 -0.347* -1.238*** (0.022) (0.144) (0.098) (0.00) Fixed effects

Year Yes Yes Yes No

Observations 280,370 280,370 280,370 280,370

Pseudo-R2 0.027 0.026 0.023 0.020

(23)

Page 23

Table 5: Marginal effects of the probit regression on the amendment dummy

The table shows the marginal effects of the probit regression presented in Table 4. A 1 million USD increase in the CDS net notional amount leads to a decrease of the probability of the amendment dummy equaling 1 of 0.4%. This is statistically significant in regression (1). The observations are on a week company level, resulting in very few amendments. That is why despite the low percentage, the CDS net notional amount is also economically significant. Regressions (2) and (3) are not statistically significant, but regression (4) again is statiscally significant. P-values are in parentheses (* significant at 10%, ** significant at 5%, and *** significant at 1%). Heteroskedatisticity is taken into account, which resulted in the use of robust standard errors.

In regression (1) that is statistically significant at the 1% level, a 1 million USD increase of the predicted CDS net notional amount results in a decrease of the probability of a loan amendment equaling 1 of 0.00418. At first sight this number is low and one can question the economic significance of it, but one should keep the following point in mind: the observations are on a week company level, which results that for only very few observations the amendment dummy equals 1. Hence, I have empirically tested my first hypothesis and the results show that CDSs lead to less amendments of loans.

Regarding the other regressions; regression (2) and (3) cannot be interpreted as the predicted CDS net notional amount is not statistically significant. On the contrary, regression (4) is statistically significant and can be interpreted: an increase of 1 million USDs in the CDS

(1) (2) (3) (4)

Predicted CDS net notional amount -0.00418*** -0.000154 -0.000058 -0.00306***

(0.000) (0.583) (0.822) (0.000) Control variables Market capitalization -0.0000167*** -0.0000020 -0.0000019 0.0000085*** (0.00) (0.247) (0.304) (0.00) Log assets -0.394** 0.0803** 0.0569* 0.371*** (0.031) (0.037) (0.069) (0.00) Book Leverage -0.297* 0.050 0.010 0.251** (0.061) (0.549) (0.915) (0.01) EBITDA/Assets 1.015** -0.31 -0.347* -1.238*** (0.023) (0.144) (0.098) (0.00) Fixed effects

Year Yes Yes Yes No

Observations 280,370 280,370 280,370 280,370

Pseudo-R2 0.027 0.026 0.023 0.020

(24)

Page 24

net notional amount results in a decreases the probability of the amendment dummy equaling 1 by 0.00306. Regarding the economic significance, the same argument applies that is mentioned in the previous paragraph.

5.2.2. Results of hypothesis 2

The regression to test the second hypothesis is similar as the regression in 5.2.1. Both regressions are the second stage of the two stage least squares model. The difference is that the index is not a dummy variable and thus I use a normal OLS regression, instead of a probit regression.

Table 6: Second stage regression on the amendment index

Table 6 shows the second stage regression of the CDS net notional amount on the amendment index. The index is positive, if the amendment is borrower friendly and vice versa. The results of the regressions (1), (2) and (3) are not statistically significant. Probably, because my dataset is on a week company basis. Hence, very little observations are amendment and the index is not available for every amendment. Regression (4) is statistically significant and shows the negative effect of CDS net notional amount on the amendment index. P-values are in parentheses (* significant at 10%, ** significant at 5%, and *** significant at 1%). Heteroskedatisticity is taken into account, which resulted in the use of robust standard errors.

The dependent variable is an amendment index with a range between -3 and 3 as is described in Section 3. A positive index is a borrower friendly amendment and a negative index is not a borrower friendly index. The regression results are presented in Table 6. However, the results in regressions (1), (2) and (3) are not statistically significant, because the observations are on a week company level as is described in Section 3. Methodology and data. This leads to

(1) (2) (3) (4)

Predicted CDS net notional amount -0.00205 -0.0000836 0.000918 -0.0170***

(0.949) (0.962) (0.561) (0.000) Control variables Market capitalization 0.0000131 0.0000141 0.00000714 0.0000634** (0.908) (0.549) (0.795) (0.029) Log assets 0.0346 -0.179 -0.313 1.617*** (0.992) (0.464) (0.139) (0.00) Book Leverage -0.613 -0.754 -0.232 0.309 (0.803) (0.264) (0.745) (0.705) EBITDA/Assets 0.187 0.821 0.96 -4.715** (0.985) (0.635) (0.577) (0.03) Fixed effects

Year Yes Yes Yes No

Observations 168 168 168 168

R2 0.299 0.298 0.224 0.261

(25)

Page 25

few observations that are amendments. Moreover, the amendment index is not available for all observations that are amendments. These two reasons result in the statistical insignificance and thus the results cannot be interpreted.

The independent variable, CDS net notional amount, of regression (4) is statistically significant at the 1% level and the control variables are also statistical significant. An increase of 1 million USD of CDS net notional amount leads to a decrease in the amendment index of 0.017.

6. Robustness

In this section, I discuss the robustness of my results of Section 5. Moreover, I discuss the shortcomings of my dataset and discuss further research regarding my topic.

In order to cope with an omitted variable bias, I followed Roberts and Sufi (2009) and used their findings as control variables. I also checked whether adding other variables would increase the R2 and affect the significance levels of the other variables. Hence, I took as many control variables into account as seemed suitable. However, it remains possible that there are other factors that could significantly affect loan renegotiations, which results in an omitted variable bias. Hence, potentially my regressions and results suffer from an omitted variable bias.

Like Roberts and Sufi (2009), I used fixed effects in the regressions, but I also always made regressions that excluded one fixed effects, in order to see if this would change the results. Sometimes it indeed did changed some of the results, for example without the year fixed effect the results were always statistically significant in Tables 3, 4, 5 and 6. On the contrary, excluding the credit rating or industry led to statistically insignificant results in Tables 4, 5, and 6. Moreover, in Table 6, regressions (1), (2) and (3) are not statistical significant, but regression (4) is. Also, as discussed in Section 3, I Winsorized variables, if necessary, in order to prevent large and illogical outliers.

Moreover, I performed a robustness check that excludes the observations in which the CDS net notional amount is 0, meaning the observations without a CDS trading were deleted. The econometrical methodology is the same as before. First, I estimated the CDS net notional amount, using the IMM dates, which is the first stage of the two stage least squares model. The results of this regression are presented in Appendix II. Similar to the methodology in Section 5, I used the results of regression (1) from Table A-1 in Appendix II to predict the CDS net

(26)

Page 26

notional amount. Then, I made a probit regression and included the predicted CDS net notional amount from the first stage in the regression as the independent variable, resulting in the table in Appendix III. Using the probit regression, I prepared the marginal effects, because a probit regression is a non-linear model and thus one cannot directly interpret the results shown in Appendix III, but one should rather calculate the marginal effects of the probit regression, as is presented in Table 6.

Table 6: Marginal effects of the probit regression on the amendment dummy using only observations with CDS trading

Table 6 is a robustness check and all observations have a CDS trading and thus a CDS net notional amount outstanding. The predicted CDS net notional amount is predicted using Table A-1 from Appendix II. Moreover, the predicted CDS net notional amount is statistically significant in regression (1) and (4). P-values are in parentheses (* significant at 10%, ** significant at 5%, and *** significant at 1%). Heteroskedatisticity is taken into account, which resulted in the use of robust standard errors.

In Table 6, the CDS net notional amount for regression (1) and (4) are statistically significant and are statistically insignificant for regression (2) and (3). The effects in regression (1) and (2) remain negative. Only regression (1) is substantially smaller; an increase of 1 million USD in CDS net notional amount, decreases the probability of an amendment by 0.000102 or as a percentage 0.0102%. However, in this case 57 observations were amendments, resulting in the low figure. The interpretation of regression (4) is that an increase of 1 million USD in CDS net notional amount, decreases the probability of an amendment by 0.0038 or as a percentage 0.38%. Notable is that in all four regressions of Table 6, most of the control variables are statistically insignificant.

(1) (2) (3) (4)

Predicted CDS net notional amount -0.000102*** 0.0000562 0.000428 -0.00380***

(0.003) (0.918) (0.383) (0.001) Control variables Market capitalization -0.00000175 -0.00000137 -0.00000144 -0.00000239 (0.330) (0.466) (0.465) (0.270) Log assets 0.0159** 0.0266 -0.0154 0.517*** (0.028) (0.797) (0.830) (0.001) Book Leverage 0.112 0.185 0.236 0.389 (0.389) (0.457) (0.359) (0.179) EBITDA/Assets 0.0146 0.00262 0.03 -1.119 (0.940) (0.997) (0.964) (0.148) Fixed effects

Year Yes Yes Yes No

Observations 47,540 47,814 48,034 49,256

Pseudo-R2 0.028 0.027 0.018 0.027

(27)

Page 27

As shown in Section 5, the results of my second hypothesis were in mostly not statistically significant, as discussed in Section 5. My dataset probably contains too few observations that are amended. Future research could use a larger time frame, leading to larger dataset and more amendments. In my opinion it is highly likely that this will result in statistically significant evidence that CDSs negatively affect borrower friendly amendments. Moreover, this contains important information, which can be used widely in practice.

Another interesting point for future research could be a behavioral point of view. Behavioral and experimental economics are growing fields and this might lead to interesting results CDSs. With the use of interviews, one can understand the motives of financial directors. Although the theory of CDSs and the empty creditor problem is already discussed in the current literature (E.g. Hu & Black 2008a; and Bolton & Oehmke 2011), it is interesting to use a practical point of view, which can lead to new insights.

7. Conclusion

My thesis is about the effect of CDSs on loan renegotiations. Theoretically, Hu and Black (2008a) argued that due to CDSs the, lender is insured against the borrowers’ default, resulting in that the lender has less incentive to renegotiate a loan. As a result, there are fewer amendments, if the lender has a CDS.

In order to empirically test this, I used a probit regression model. The dependent variable of my econometrical model is a dummy variable that equals 1, if a loan is amended and 0 otherwise, and the independent variable is the CDS net notional amount. However, because the relation between CDS and loan renegotiations suffer from endogeneity in the form of reverse causality, I had to use an instrumental variable. I used a two stage least squares model to include the instrumental variable, and the instrument I used was the seasonal pattern of CDSs net notional amount. The seasonal pattern of CDSs arises, because a CDS contract mature on one of the four IMM dates. Hence, I made dummy variables for each IMM date, which predicted the CDS net notional amount.

Roberts and Sufi (2009) discussed the importance of evidence on loan renegotiations. Also, they found several connections between loan renegotiations and other factors, which in turn I used as control variables, in order to prevent an omitted variable bias.

My results present empirical evidence in support of the theory set out by Hu and Black (2008a) and many others; CDSs negatively affect loan amendments and thus loan

(28)

Page 28

renegotiations. More specific, an increase of the CDS net notional amount of 1 USD million leads to a decrease in the likelihood of a loan to be amended of 0.4%. Although this percentage seems quite low, one should keep in mind that my observations are on a weekly company level, resulting in that very few observations are amendments.

Besides my baseline hypothesis, concluded above, I also tested a second hypothesis. The second hypothesis states that a CDS lead to borrower un-friendlier loan renegotiations. Moreover, this hypothesis is also constructed with a two stage least squares model and the first stage is the same as for my baseline hypothesis. Thus, the same predicted CDS net notional amounts are used. The second stage regresses the predicted CDS net notional amount on an amendment index. The amendment index measures the level of borrower friendliness of a loan amendment on a scale between -3 and 3. Where -3 is the borrower un-friendliest and 3 is the borrower friendliest. For this hypothesis, I found limited statistical evidence. Only one out of in total four regressions were statistical significant. Although this regression supported the hypothesis and showed the negative effect of CDS on borrower friendly loan amendments and thus loan renegotiations. The statistical insignificance probably is explained by the fact that only a very limited number of observations are amendments, due to the weekly company observations explained previously in the conclusion. Only for those observations an amendment index is applicable. Moreover, not for every amendment the amendment index is available. A possible solution for future research is to prolong the time frame. I used a time frame from week 44 in 2008 until week 12 in 2013. Hence, adding more years leads to more amendment and more observations, which can result in more convincing evidence for this hypothesis.

Another interesting point of view for future research can be derived from a behavioral point of view. Interviews with financial directors can perhaps result in new interesting and practical insights. This might especially be interesting with respect to the motives of the financial directors.

(29)

Page 29

References

Agarwal, S., Amromin, G., Ben-David, I., Chomsisengphet, S., & Evanoff, D. D. (2011). The role of securitization in mortgage renegotiation. Journal of Financial Economics, 102(3), 559-578.

Aghion, P., Dewatripont, M., & Rey, P. (1994). Renegotiation design with unverifiable information. Econometrica, 62(2), 257-282.

Altman, E. I. (2000). Predicting financial distress of companies: revisiting the Z-score and ZETA models. (Stern School of Business, New York University Working Paper).

Bedendo, M., Cathcart, L., & El-Jahel, L. (2014). Distressed debt restructuring in the presence

of credit default swaps. (SSRN Working Paper No. 1666101).

Bolton, P., & Oehmke, M. (2011). Credit default swaps and the empty creditor problem. Review

of Financial Studies, 24(8), 2617-2655.

Campello, M., & Matta, R. (2013). Credit default swaps, firm financing and the

economy. (SSRN Working Paper No. 1770066).

Colozza, T. (2013). Standardization of Credit Default Swaps Market. (SSRN Working Paper No. 2366447).

Danis, A. (2013). Do empty creditors matter? Evidence from distressed exchange

offers. (SSRN Working Paper No. 2001467).

Garleanu, N., & Zwiebel, J. (2009). Design and renegotiation of debt covenants. Review of

Financial Studies, 22(2), 749-781.

Hart, O., & Moore, J. (1997). Default and renegotiation: A dynamic model of debt. (NBER Working Paper No. 5907).

Hu, H. T. C., & Black, B. S. (2006). The new vote buying: Empty voting and hidden (morphable) ownership. Southern California Law Review, 79, 811-908.

(30)

Page 30

Hu, H. T. C., & Black, B. S. (2007). Hedge funds, insiders, and the decoupling of economic and voting ownership: Empty voting and hidden (morphable) ownership. Journal of

Corporate Finance, 13(2), 343-367.

Hu, H. T. C., & Black, B. S. (2008a). Debt, equity and hybrid decoupling: Governance and systemic risk implications. European Financial Management, 14(4), 663-709.

Hu, H. T. C., & Black, B. S. (2008b). Equity and debt decoupling and empty voting II: importance and extensions. University of Pennsylvania Law Review, 156, 625-739.

Jostarndt, P., & Sautner, Z. (2009). Out-of-Court Restructuring versus Formal Bankruptcy in a Non-Interventionist Bankruptcy Setting. Review of Finance, 14(4), 623-668.

O'Kane, D. (2011). Force-fitting CDS spreads to CDS index swaps. Journal of

Derivatives 18(3), 61-74.

Peristiani, S., & Savino, V. (2011). Are credit default swaps associated with higher corporate

defaults? (ECONSTOR Working Paper No. 494).

Piskorski, T., Seru, A., & Vig, V. (2010). Securitization and distressed loan renegotiation: Evidence from the subprime mortgage crisis. Journal of Financial Economics, 97(3), 369-397.

Roberts, M. R. (2015). The role of dynamic renegotiation and asymmetric information in financial contracting. Journal of Financial Economics, 116(1), 61-81.

Roberts, M. R., & Sufi, A. (2009). Renegotiation of financial contracts: Evidence from private credit agreements. Journal of Financial Economics, 93(2), 159-184.

Stock, J. H. and Watson M. M. (2012). Introduction to Econometrics. Harlow, England: Pearson Education Limited.

Subrahmanyam, M. G., Tang, D. Y., & Wang, S. Q. (2012). Does the tail wag the dog? The

(31)

Page 31

White, R. (2014). The Pricing and Risk Management of Credit Default Swaps, with a Focus on the ISDA Model. Open Gamma Quantative Research, 16, 1-43.

(32)

Page 32

Appendix I

The data regarding the control variables is downloaded from Compustat on a quarterly basis. Also, like Roberts and Sufi (2009) the variables are lagged one quarter relative to the dependent variable in order to avoid mechanical associations. Also, if necessary the variables are Winsorized on the 1% level. The control variables are calculated as follows:

𝑀𝑎𝑟𝑘𝑒𝑡 𝑐𝑎𝑝𝑖𝑡𝑎𝑙𝑖𝑧𝑎𝑡𝑖𝑜𝑛 = 𝑐𝑠ℎ𝑜𝑞 ∗ 𝑝𝑟𝑐𝑐𝑞 (4)

𝐿𝑜𝑔 𝑎𝑠𝑠𝑒𝑡𝑠 = ln (𝑎𝑡𝑞) (5)

𝐵𝑜𝑜𝑘 𝑙𝑒𝑣𝑒𝑟𝑎𝑔𝑒 = (𝑑𝑙𝑐𝑞 + 𝑑𝑡𝑡𝑞)/𝑎𝑡𝑞 (6)

(33)

Page 33

Appendix II

Table A-1: First stage regression that only includes observations with CDS trading

Table A-1 is the first stage regression and predicts the dependent variable of the mode: CDS net notional amount. Moreover, the regression only includes observations with a CDS trading. The independent variables IMM date (1) until (4) are dummy variables that equal 1 on each another IMM date, and 0 otherwise. Regression (1), (2) and (3) are all statistically significant for all four IMM dates, but regression (4) is not. P-values are in parentheses (* significant at 10%, ** significant at 5%, and *** significant at 1%). Heteroskedatisticity is taken into account, which resulted in the use of robust standard errors.

(1) (2) (3) (4) IMM date (1) -37.97** -39.33** -32.29** -8.201 (0.012) (0.011) (0.049) (0.600) IMM date (2) -36.93** -38.40** -29.62 -48.72*** (0.028) (0.024) (0.102) (0.005) IMM date (3) -37.01** -38.79** -35.29** -42.99** (0.026) (0.022) (0.050) (0.012) IMM date (4) -54.64*** -53.36*** -56.72*** 6.898 (0.000) (0.001) (0.001) (0.660) Control variables Market capitalization 0.000676*** -0.000704*** 0.00129*** 0.000401*** (0.00) (0.00) (0.00) (0.00) Log assets 173.7*** 194.5*** 136.5*** 166.2*** (0.00) (0.00) (0.00) (0.00) Book Leverage 245.7*** 117.1*** 129.4*** 238.4*** (0.00) (0.00) (0.00) (0.00) EBITDA/Assets -544.2*** -597.3*** -378.9*** -722.4*** (0.00) (0.00) (0.00) (0.00) Fixed effects

Year Yes Yes Yes No

Observations 50,063 50,063 50,063 50,063

R2 0.264 0.236 0.141 0.212

(34)

Page 34

Appendix III

Table A-2: Second stage probit regression on the amendment dummy that only includes observations with CDS trading

All observations in the regression results in Table A-2 have a CDS trading. Using Table A-1, I predicted the CDS net notional amount, which I then regressed on the amendment dummy. The variable CDS net notional amount in regression (1) and (4) are statistical significant and are insignificant in regression (2) and (3), which is similar to Table 4. P-values are in parentheses (* significant at 10%, ** significant at 5%, and *** significant at 1%). Heteroskedatisticity is taken into account, which resulted in the use of robust standard errors.

(1) (2) (3) (4)

Predicted CDS net notional amount -0.000102*** 0.0000562 0.000428 -0.00380***

(0.003) (0.918) (0.383) (0.001) Control variables Market capitalization -0.00000175 -0.00000137 -0.00000144 -0.00000239 (0.330) (0.466) (0.465) (0.270) Log assets 0.0159** 0.0266 -0.0154 0.517*** (0.028) (0.797) (0.830) (0.001) Book Leverage 0.112 0.185 0.236 0.389 (0.389) (0.457) (0.359) (0.179) EBITDA/Assets 0.0146 0.00262 0.03 -1.119 (0.940) (0.997) (0.964) (0.148) Fixed effects

Year Yes Yes Yes No

Observations 47,540 47,814 48,034 49,256

Pseudo-R2 0.028 0.027 0.018 0.027

Loan renegotiations and credit default swaps : an empirical analysis