The empty creditor problem and contract renegotiation

(1)

The Empty Creditor Problem and

Contract Renegotiation

Master Thesis

Aleksander Gwizdak 10824847 July 2015 MSc Business Economics Finance Track

(2)

2

Statement of Originality

This document is written by Aleksander Gwizdak 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)

3

Abstract

Credit Default Swap contracts have recently attracted a fierce debate in the academic world. Numerous researchers blame them for the exacerbation of the financial crisis. One interesting aspect of the CDS market is the Empty Creditor Problem, firstly referred to in the paper by Hu and Black (2008). It implies that creditors, who protect themselves against default of their counterparties, may refuse to restructure a loan in the case of financial difficulties of a borrower, in order to trigger the credit event and receive a CDS payment. This thesis investigates this issue empirically to complement the inconclusive findings from the currently available research. Using a sample of all major US companies in years from 2003 to 2013, I find that following the inception of CDS trading the probability of successful loan restructuring for financially distressed firms decreases by 8,61%. This result is statistically significant and robust to the removal of restructurings only marginally changing the loan terms. Furthermore, additional analysis implies that the Empty Creditor Problem affects mostly restructurings of a small and medium size and restructurings that alter the loan terms in favor of borrowers. Interestingly, the results also indicate that the intensity of the Empty Creditor Problem is the strongest during the prosperous years before the financial crisis.

(4)

4

1. Introduction

Credit Default Swap is a type of derivative instrument that facilitates the transfer of credit exposure between two entities – protection buyer and protection seller. In exchange for periodic payments, the provider of CDS is obliged to compensate the purchaser if the credit event occurs. The credit event is usually associated with the failure to repay debt obligation. CDS can be perceived as an insurance contract, it does not however require the protection buyer to be exposed to the counterparty risk. This differentiates it from the standard insurance and opens the possibility for speculation.

Although CDS contracts were designed in the mid-1990’, the expansion of the market for these derivatives did not start until 2003 (Mengle (2009)). Since then the size of CDS market has grown exponentially doubling in volume year by year and reaching the tremendous value of over 62 trillion US dollars of notional amounts outstanding at the end of 20071. Starting with the inception of the recent financial crisis the magnitude of CDS market began shrinking with the notional amounts of CDS falling to 21 trillion US dollars at the end of 20132.

The initial purpose for creation and usage of CDS contracts was to help creditors mitigate risks related to default of their counterparties. It appears however that these derivatives may have adverse effects on debtors. Recent theoretical and empirical research (Hu and Black (2008), Bolton and Oemhke (2011), Subrahmanyam, Tang and Wang (2014)) suggests that availability of CDS trading may itself influence the probability of debtor’s bankruptcy and credit rating downgrade. This phenomenon arises from the Empty Creditor Problem which implies that in some cases creditors will not be willing to renegotiate loan terms with financially distressed borrowers in order to receive compensation from the providers of CDS contracts. Bolton and Oehmke (2011) postulate that this scenario is likely to happen if borrower’s resources for restructuring are lower than a potential CDS payoff. In such a case it would be more profitable for the creditor to refuse to restructure and in turn push the borrower into further problems and possibly a bankruptcy. Moreover, Hu and Black (2008) emphasize that CDS protected lenders can have skewed incentives as they are not motivated to properly control and monitor debtors.

1_{ISDA Market Survey, Notional amounts outstanding at year-end, all surveyed contracts, 1987-present; See:}

http://www.isda.org/statistics/pdf/ISDA-Market-Survey-historical-data.pdf, Accessed: June 1, 2015

2

BIS (2014), Statistical release: OTC derivatives statistics at end-December 2013 ; See: http://www.bis.org/publ/otc_hy1405.pdf, Accessed: June 1, 2015

(6)

6

The issue of the Empty Creditor Problem has drawn the attention of media as well as practitioners. In 2009 the article in the Economist warned that financial innovation such as CDS contracts has undermined the assumption made in bankruptcy codes that “creditors always attempt to keep solvent firms out of bankruptcy”3. Furthermore, a famous investor George Soros associated the bankruptcies of General Motors and AbitibiBowater with CDS protected creditors. Additionally, he argued that purchasing insurance in the form of CDS is “like buying life insurance on someone else’s life and owning a license to kill him”4

. On the other hand International Swaps and Derivatives Association (ISDA) issued a research note in 2009 that aimed to rebut the allegations made in the theoretical literature and call for further empirical investigation of the issue.

Hitherto, only a few empirical studies analyzed the Empty Creditor Problem (Subrahmanyam et al. (2014), Peristiani and Savino (2011), Bedendo, Cathcart, and El-Jahel (2012)). In addition, the findings from these papers do not provide a clear evidence for the existence of “empty creditors”. The article that is perceived to be of the biggest contribution to the research on this topic is the one by Subrahmanyam et al. (2014). The conclusions from this paper suggest that the inception of CDS trading increases the likelihood of bankruptcy and credit rating downgrade. This result is in line with predictions from theory but does not answer a more fundamental question – whether CDS trading reduces the probability of successful loan renegotiation. This issue was also empirically investigated (Danis (2013), Bedendo et al. (2012)), yet the findings are inconclusive and indicate a need for further analysis.

It is important to analyze the effect of the Empty Creditor Problem on the probability of successful restructuring as renegotiation is usually the first serious attempt to escape financial troubles. If the renegotiation fails then it is likely that the company would go bankrupt. There are however also other reasons for a company to declare bankruptcy which do not necessarily relate to the Empty Creditor Problem. The relation of this phenomenon to loan renegotiation can be perceived as more primary than its relation to bankruptcy. It might be then more accurate to investigate the Empty Creditor Problem in terms of its influence on renegotiation outcome. Furthermore the analysis of this issue would complement the existing research on the effect of CDS on debt financing (Ashcraft and Santos (2009), Saretto and Tooks (2013)).

3_{The Economist, No Empty Threat, June 20, 2009. Accessed: June 2, 2015.; See:}

http://www.economist.com/node/13871164

4

Financial Times, The three steps to financial reform, June 16, 2009. Accessed: June 2, 2015.; See: http://www.ft.com/intl/cms/s/0/b62b1bd4-5aa3-11de-8c14-00144feabdc0.html#axzz3bpb4XGoB

(7)

7

Additionally a study of this particular aspect of the Empty Creditor Problem is worth researching as the findings could be useful for policy makers who can try to limit the negative effects of “empty creditors” and restore the assumption that lenders should attempt to keep solvent firms out of bankruptcy. The conclusions would also interest practitioners and could influence their financing choices.

This thesis intends to complement the existing literature and contribute to the research on the Empty Creditor Problem by exploiting a unique collection of data on loan renegotiations. I use a dataset provided by my supervisor dr. Tomislav Ladika that enables me to recognize which firms out of all major US companies successfully restructured any loans in years from 1989 to 2013. Together with the information on the availability of CDS trading in years 2003 - 2013 obtained from Datastream database I create a dataset that allows me to perform econometrical analysis. I investigate the Empty Creditor Problem by verifying the conjecture that CDS referenced companies have a lower probability of successful loan restructuring than non-CDS companies (Hu and Black (2008), Bolton and Oemhke (2011)). Furthermore, I analyze whether the Empty Creditor Problem affects mostly restructurings that change the loan terms in favor of borrowers and whether it is mostly pronounced among small and medium sized restructurings. The dependent variable in this research is a dichotomous variable equal to 1 if a particular firm successfully renegotiated any loans in a given year and it is equal to 0 otherwise. The main explanatory variable is an indicator variable for CDS trading. The specification was complemented with a set of control variables obtained mainly from the Compustat database. The model that most accurately suits this type of analysis is the probit model as it allows assessing the differences in probability.

The main results of this thesis provide a useful insight into the Empty Creditor Problem. According to the estimation results companies referenced by CDS trading have roughly 2% lower probability of successful loan restructuring than non-CDS firms. This result is however exposed to the omitted variable bias stemming from the differences between these two types of firms. Furthermore, the result is statistically insignificant when the specification is supplemented by year fixed effects. The problem of omitted variable bias is limited in a test that focuses exclusively on companies that at some point in time were referenced by CDS. The model indicates that following the inception of CDS trading on companies’ default, the likelihood of successful restructuring for financially distressed firms decreases by 8,61%. This result is both statistically and economically significant. Moreover, the econometrical analysis

(8)

8 is robust to the removal of “trivial” restructurings5

. The robustness check provides also supplementary regressions proving that the Empty Creditor Problem concerns only the entities in financial troubles. Additionally, the investigation of the intensity of the Empty Creditor Problem reveals that it varies over time. Surprisingly it is the strongest during prosperous years before the financial crisis.

The further analysis of the Empty Creditor Problem provides more insights into the issue. Firstly, the results imply that the difference in probability of successful loan restructuring between CDS and non-CDS firms is lower for restructurings beneficial for borrowers than for restructurings beneficial for lenders. This conclusion suggests that the Empty Creditor Problem affects mostly firms in financial distress that need to obtain more attractive loan terms (Hu and Black (2008)). Next, the empirical investigation in this thesis also indicates that the Empty Creditor Problem influences mostly restructurings implementing small and medium changes. The difference in likelihood of successful loan restructuring between CDS and non-CDS firms is lower for restructurings implementing small and medium changes as compared to restructurings significantly altering the loan terms. Overall, the findings of this thesis contribute to the research on the Empty Creditor Problem and suggest a need for further analysis that would support the results with the use of more advanced empirical approach.

The remainder of this thesis is organized as follows. The next section provides a review of the literature relating to CDS market, the Empty Creditor Problem and loan renegotiations. Section 3 presents the hypotheses and their theoretical foundation and then discusses the empirical approach adopted. Section 4 describes the data and the variables used in the research. In the further three sections I discuss the findings of the econometrical analysis. Section 8 elaborates on the potential omitted variable bias and finally section 9 concludes.

5

Restructurings only marginally changing the loan terms. A more detailed explanation is provided in further parts of the thesis.

(9)

9

2. Literature Review

I present the most relevant literature to my research dividing it into four subsections. Firstly, I shortly discuss the important articles relating to the market and economics of Credit Default Swaps. This allows for a better understanding of the topic. In the second subsection I analyze the findings of the theoretical papers discussing the Empty Creditor Problem which in turn allows me to move to the next part – the summary of the empirical evidence. Lastly, I describe other relevant literature, which mainly relates to loan renegotiation determinants and relationship between creditors and debtors. In that subsection I also discuss the influence of CDS trading on the capital structure of reference companies.

2.1 Market and Economics of Credit Default Swaps

Academic literature on Credit Default Swaps grows very rapidly due to the massive size of the market and the relation of CDS contracts to the recent financial crisis. Apart from the articles discussing the Empty Creditor Problem, which I analyze in the next part, there is also a vast research on other topics connected to CDS. In order to fully understand all characteristics and rules of CDS market it is worth to acquaint more generally with research in this field.

Oehmke and Zawadowski (2014) try to establish the economic role of Credit Default Swap market by analyzing the determinants of the volume of CDS bought. Their findings suggest that this market serves as an alternative for speculation and hedging in the reference asset. Jarrow (2010) analyzes the economics of CDS using the literature on insurance as a base for the research. He claims that the existence of CDS market enhances welfare as it allows for a more economical allocation of risks. Contrarily Brown and Hao (2012) claim that the concept of CDS is unsound and that it may lead to widespread escalation of leverage in companies’ capital structure and in turn to financial fragility. Terzi and Ulucay (2011) analyze the influence of CDSs on the stability of financial markets and state that this influence depends on the capital liquidity requirements as well as on the market mechanisms. Stulz (2010) focuses on the connection of CDS and the recent credit crisis. He analyzes the mechanics of the market to determine its contribution to risk of the financial system and he rejects the accusations against credit derivatives.

Finally Bolton and Oemhke (2013) provide a thorough review of the most interesting recent research in the area of CDS. They discuss literature concerning many interesting topics,

(10)

10

among others Empty Creditor Problem, dealing with counterparty risk and strategic behavior in CDS auctions.

2.2 Theoretical Research on the Empty Creditor Problem

In this part I present the theoretical contribution to the Empty Creditor Problem as it is a fundament for the empirical research. Bolton and Oemhke (2011) provide a formal analysis of this issue building a three time period limited commitment model to thoroughly examine it. They show that the presence of CDS protection can influence the result of debt renegotiation. The outcome of renegotiation might be affected because a creditor that purchased insurance in the form of CDS is hedged against the default of the counterparty. This creditor receives a payoff from the CDS if the credit event occurs. If the debtor’s resources for restructuring are lower than potential payoff from CDS contract, then it is preferable for the creditor to force the firm into bankruptcy and receive CDS payment (Bolton and Oehmke (2013)). These arguments are in line with findings of Hu and Black (2008) who also indicate that lenders protected by CDS have skewed incentives as they have lower eagerness to monitor debtors and may have no interest in agreeing to voluntarily restructure debt which in turn may cause the bankruptcy of a borrower. Hu and Black call such lenders “empty creditors”.

In addition to the general predictions about the Empty Creditor Problem, Bolton and Oehmke (2011) also point to the fact that classifying (voluntary) restructuring as a credit event6 in the CDS contract may significantly impair (or even completely eliminate) this phenomenon. If that clause is included in the swap, then creditors should have no incentives to impede the renegotiations as they would receive the default payment regardless of the final outcome. This finding is essential as it implies that the intensity of the Empty Creditor Problem may differ depending on the terms of a contract. It is then clear that various types of CDS contracts should be priced differently and this expectation was proven by Packer and Zhu (2005).

Campello and Matta (2013) provide another interesting theoretical contribution to the study of the Empty Creditor Problem. The authors analyze the issue focusing on the possible procyclicality. They claim that lenders tend to overinsure (using CDS) during economic upswings and protect themselves less during times of recession. This in turn would cause the Empty Creditor Problem to be more pronounced during times of high overinsurance.

6_{There are four main types of CDS contracts: XR (no restructuring) – a contract with such a clause does not treat}

restructuring as a credit event; CR (complete restructuring), MR (modified restructuring), MMR (modified-modified restructuring) – these three types of CDS contract treat restructuring as a credit event however with certain restrictions in MR and MMR types.

(11)

11

Conclusions from these theoretical papers clearly suggest a need for an empirical research which would enable a better and deeper understanding of the topic and would allow creditors and debtors to understand their situation and incentives of their counterparties.

2.3 Empirical Evidence on the Empty Creditor Problem

Predictions from Bolton and Oemhke (2011) analysis seem to be partially confirmed in the empirical study by Subrahmanyam et al. (2014). These authors investigate the effect of CDS trading on probability of credit rating downgrade and probability of bankruptcy. They find a strong relationship between the inception of CDS trading and the risk of bankruptcy (the risk of bankruptcy is more than doubled after the inception of CDS trading). The authors also support the implications made by Bolton and Oehmke (2011) regarding the intensity of the Empty Creditor Problem being dependent on the restructuring clause contained in the contract. They show that reference companies with a larger number of “no restructuring” clauses in CDS tend to be more adversely affected by the Empty Creditor Problem. The possible flaw of the study by Subrahmanyam et al. (2014) is however the use of Instrumental Variables regression model. The lack of reliable data on CDS trading and the endogeneity problem forced the authors to use instruments that might seem not very convincing (they use Foreign exchange hedging positions of lenders and their Tier One capital ratio as instruments for CDS trading). Although the Sargan (1958) overidentification test did not reject the null hypothesis that both instruments are exogenous, the p-value was low and very close to 10% significance level (p-value was 0,119). Despite the shortcomings this article is considered to be the biggest empirical contribution to the Empty Creditor Problem and it constitutes a base for the analysis in this thesis.

In a similar manner, Peristiani and Savino (2011) explore the effects of CDS trading on corporate distress risk. The authors determine the default likelihood for nonfinancial U.S companies by adapting a proportional hazard model of bankruptcy together with Merton’s contingent claims approach. They conclude that expected default frequency is higher for companies with available CDS contracts (they however find this relationship only for a part of the time period they analyze after decomposing the effect by year). Insights from these two papers were of a great help in writing this thesis as they research a similar issue. The effect of the Empty Creditor Problem on the probability of bankruptcy is closely related to its effect on the probability of successful loan renegotiation. The effect on loan renegotiations can be

(12)

12

considered to be primary as if restructuring fails companies are likely to declare bankruptcy (it is however not the sole determinant of it).

There are also two other interesting papers on the effect of CDS trading on the outcome of debt restructuring. One paper was written by Danis (2013) and the other by Bedendo et al. (2012). The general concern of these articles may seem to be similar to the topic of my thesis, however the data and methodology used is very different. Danis (2013) focuses mainly on the economic costs created by CDS and takes into account only public debt. He finds that companies with CDS traded on their default have lower participation rate to distressed exchanges (supporting the Empty Creditor Problem Hypothesis). Bedendo et al. (2012) concentrate on the role CDS trading plays in the choice between in- and out-of-court debt restructuring. Their results are opposing the Empty Creditor Problem as they demonstrate that CDS reference companies are not more likely to restructure debt in court (through Chapter 11 or Chapter 7 bankruptcy) as compared to companies without CDS trading on their default. What encourages researching the topic in more detail is the fact that conclusions from paper by Danis (2013) and paper by Bedendo et al. (2012) are conflicting. Both articles contribute to the literature on the Empty Creditor Problem, none of them however precisely addresses the research question of my thesis.

2.4 Other Relevant Literature

The Effect of CDS on Capital Structure

Another interesting aspect of CDS trading is related to the changes it might cause in debtors’ capital structure (and in turn in their risk of default). Danis and Gamba (2014) build a theoretical model to assess whether the introduction of CDS on companies’ default alters their financing choices and access to capital. In their model the authors implement the predictions of the Empty Creditor Problem (the increased probability of bankruptcy). Their results are consistent with previous research on this topic by Saretto and Tooks (2013) who find that companies are able to preserve a higher ratio of leverage if there are CDS contracts traded on their default (Saretto and Tooks (2013) also indicate that these companies have longer debt maturities of debt). Additionally Danis and Gamba (2014) imply that the inception of CDS trading enables companies to reduce the cost of debt which is however in opposition to findings of Ashcraft and Santos (2009) who reject this hypothesis in their paper on that topic.

(13)

13

Debtors and Creditors Relationship and Loan Renegotiation

In this part I summarize the important articles relating strictly to loan renegotiation. It can be noticed that the number of papers researching this issue is very small, which could be caused by the lack of easily accessible and reliable data on loan renegotiation. A further investigation is needed to better understand this interesting area of finance. All this encourages me even more to study the effect of the Empty Creditor Problem on the restructuring likelihood as such an analysis would be of a significant contribution to the currently available research.

In order to gain a deeper understanding of the Empty Creditor Problem it is important to understand the main drivers of banks’ willingness to restructure debt – costs related to it. Campello, Ladika and Matta (2014) provide a model that analyzes how changes in costs of debt restructuring influence bankruptcy risks and financing costs. To do that, they exploit a modification in U.S. tax regulation which reduced taxes paid by creditors while restructuring syndicated loans. The authors find that a decrease in costs of out-of-court debt restructuring increases the probability of successful loan renegotiation.

Roberts and Sufi (2009) analyze the incentive conflicts between creditors and debtors and their influence on the corporate financing policy. They find that following a covenant violation the debt issuing activity declines significantly and terms of contracts deteriorate. The managers of companies (debtors) are forced to modify their financial policies, which would not happen if not for the covenant violation. Despite that borrowers rarely switch to other lenders as they cannot obtain contracts at more favorable terms. In a more recent paper Roberts (2014) analyzes the role of loan renegotiation claiming that it plays a crucial role in bank lending. More specifically, the ability of borrowers to renegotiate loans in the future encourages them to accept restrictive contracts at their origination. This paper is of a significant contribution to the research on loan renegotiation as it relates to the renegotiation of debt outside of financial distress.

In relation to the effect of leverage on loan renegotiation, Jostarndt and Sautner (2009) find that companies with more debt and higher leverage are more likely to conduct a successful restructuring. Although Jostarndt and Sautner (2009) do not differentiate between companies with and without traded CDS contracts, their evidence suggests a useful control variable for the regression model in my thesis (leverage ratio).

(14)

14

2.5 Conclusion

All above-mentioned papers contribute significantly to the literature on CDS and its connection to loan restructuring. None of them however clearly answers the research question of my thesis. The existing literature, both empirical and theoretical, encourages me to form hypotheses that I present in the following section and then investigate empirically.

3. Hypotheses and Methodology

3.1 Hypotheses

The literature review presented in the preceding section demonstrates that the up to date available evidence on the existence of the Empty Creditor Problem is still insufficient and does not provide a thorough examination of the issue. The conclusions from various papers are conflicting and dependent on the methodology and time period used. In this thesis I intend to further analyze this interesting phenomenon and supplement the existing literature by concentrating on the differences in probability of successful loan renegotiation between CDS referenced companies and non-CDS firms. The answer to the main hypothesis (which is the Empty Creditor Hypothesis) is provided by researching various aspects of the issue and it is gradually formed in the subsequent sections in which I interpret the results.

The first hypothesis that I test derives from the two main theoretical papers on the subject of the Empty Creditor Problem (Bolton and Oehmke (2011), Hu and Black (2008)). The authors of these articles postulate that due to the skewed incentives of CDS protected creditors, the reference entities should be less likely to restructure their debt out of court. These indications, together with conflicting results of the currently available research (Subrahmanyam et al. (2014), Bedendo et al. (2012)), encourage me to set the following hypothesis, which I investigate empirically in the further sections:

H1: The probability of successful loan renegotiation is lower for companies with active CDS

trading as compared to non-CDS companies.

The results from an econometrical analysis of this hypothesis may however be biased due to the possible differences between companies referenced by CDS and companies without CDS

(15)

15

trading. It would also be insufficient to merely compare the general probability of successful restructuring for these two types of firms, and even the statistically significant results would have to be interpreted with caution. In order to directly observe the decline in the probability of successful loan renegotiation for CDS referenced companies, it is interesting to revise the hypothesis and also ask whether following an introduction of trading in CDS on companies’ default their likelihood of restructuring is lowered. This issue is also analyzed in this thesis and it provides further evidence to the investigation of the main hypothesis. It is useful to provide such a test as it relates to the comparison of the same companies before and after the start of CDS trading. If there are no other significant changes that influence the outcome of renegotiations and the probability of success is lowered then the introduction of CDS trading would likely be the only reason for this decline. This could be an evidence for the emergence of “empty creditors”, who would prefer to impede the renegotiations in order to receive the CDS payment.

In order to analyze the Empty Creditor Hypothesis from a different angle and to relate it to a type of renegotiation and the financial condition of borrowers I set the second hypothesis:

H2: The difference in probability of successful loan renegotiation between CDS and non-CDS

firms is higher for restructurings beneficial for borrowers as compared to restructurings beneficial for lenders.

The general idea about “empty creditors” originally defined by Hu and Black (2008) implies that CDS protected lenders would not be willing to restructure debt if it meant receiving a lower final payoff than a possible payment from the CDS contract. Similarly, Bolton and Oehmke (2011) demonstrate that CDS contracts allow lenders not to renegotiate unless the renegotiation terms are beneficial for them. It follows then, that companies which would be mostly affected by the Empty Creditor Problem should be the firms in financial distress and wanting to restructure debt on more attractive terms. Such restructurings would be bad for lenders, and would create a threat of them becoming “empty creditors”. It is interesting to test it then, whether actually the Empty Creditor Problem is driven by restructurings disadvantageous for creditors.

Another interesting aspect of the Empty Creditor Problem is its relation to the size of renegotiation. Hu and Black (2008) claim that unhedged creditors set the loan covenants in such a way that their violation by the borrower provides an early warning of financial trouble. Following such a warning it is likely that a restructuring of a small size would suffice to avoid

(16)

16

exacerbation of the problems. In the situation however when lender is protected by CDS contract, he would not care that much about monitoring the borrower and borrower’s condition. In such a case it is more likely that the restructuring would be done at the later stage of financial troubles and would require larger change in loan terms. CDS referenced companies could be more likely to restructure in the further stages of distress as creditors also take into consideration their reputation (Bedendo et al. (2012)), which could be worsened if they continuously refused to renegotiate out-of-court. It might be then possible that the Empty Creditor Problem is manifested mostly through the lowered probability of successful renegotiation of a small size.

Furthermore the conclusions from Bolton and Oehmke (2011) also imply that CDS companies should mostly be associated with restructurings that significantly alter the loan terms. These authors find that CDS trading prevents strategic default as the CDS protected lenders have bigger bargaining power (they are harder in debt renegotiations) and do not allow debtors to default only to obtain better loan terms or to divert cash flows (Bolton and Oehmke (2011)). Borrowers referenced by CDS need to offer conditions that make it advantageous enough for creditors to agree to restructure (large changes to loan terms). It means then that only non-CDS firms would be strategically defaulting on their commitments and the restructurings that would follow are likely to be smaller as compared to restructurings resulting from real default. Small renegotiations can more often be done for strategic reasons than bigger ones that usually result from financial problems. CDS companies in turn should have a lower probability of implementing a restructuring that does not change the loan terms considerably (as they cannot strategically default). The last hypothesis that I test follows from the two arguments presented above and is shown below:

H3: The difference in probability of successful loan renegotiation between CDS and non-CDS

firms is higher for restructurings implementing small changes as compared to restructurings implementing large changes.

All hypotheses that are investigated in this thesis relate generally to all companies in the data sample as well as separately to companies in financial troubles. It is worth to test these hypotheses for distressed firms exclusively as they should be more likely to be endangered by the Empty Creditor Problem (Bedendo et al. (2012)). The Empty Creditor Hypothesis would relate mostly to these companies and not to firms that restructure solely because they originally agreed to restrictive terms in the expectation of future renegotiations (Roberts

(17)

17

(2014)). In the following section I explain the empirical approach adopted in this thesis and elaborate on the methodology used to identify distressed firms.

3.2 Empirical Approach

In this thesis I assess whether there exists an economically and statistically significant difference in the probability of successful loan renegotiation between CDS referenced companies and firms without CDS trading. I also determine whether the inception of CDS trading indeed reduces the chances of loan restructuring as can be concluded from theory (Bolton and Oemhke (2011)) and research (Danis (2013)). I am able to analyze these issues using the dataset that contains information on loan renegotiations of all major US companies in years from 1989 to 2013 (I provide a more detailed description of the dataset in the next section of the thesis). The empirical approach that most accurately, in my view, suits this analysis is the probit model since it allows for the relationship to be nonlinear and a dichotomous outcome (failure of renegotiation vs. success of renegotiation) modeled as a linear relationship could produce inaccurate results. The time sample for testing the initial hypothesis is from 2003 to 2013 which practically is the whole period when CDS contracts were traded (some CDSs started to trade earlier however the liquidity might have not been sufficient). The baseline regression model that I estimate to verify the first hypothesis is as follows:

where F denotes the cumulative standard normal distribution function. The dependent variable amendment_dummy is an indicator that is equal to 1 if a particular firm successfully restructured any loans in a given year and 0 otherwise. The main independent variable CDS_dummy is also binary and it is equal to 1 if a particular firm was referenced by CDS in a given year and to 0 if it was not. The coefficient of interest in this research is the coefficient on CDS_dummy variable. Its sign informs about the direction of causal relationship. If the coefficient is negative and statistically significant then the results would be in line with the Empty Creditor Hypothesis. The negative coefficient would indicate that the probability of successful loan renegotiation is lower for companies referenced by CDS contracts. Although it is possible to draw this general conclusion only from the sign of the coefficient, one needs to calculate the marginal effect in order to assess the magnitude of the difference in probability (Stock and Watson (2012)). The result obtained from calculating the marginal effect of a change in a CDS_dummy variable (change from 0 to 1 relates to a change from

(18)

18

non-CDS firm to CDS firm) would illustrate how much smaller or bigger is the probability of out-of-court restructuring for CDS referenced firms.

In order to minimize the omitted variable bias I add to the regression model a set of useful control variables. I control for the riskiness of companies by including a leverage ratio variable. CDS companies are likely to be more levered (Saretto and Tookes (2011)) and this might affect the outcome of renegotiation. I also calculate Altman Z-scores for all companies in the sample using the equation provided in the paper written by Altman (2000). The equation is as follows:

where:

= Working Capital/Total Assets = Retained Earnings/Total Assets

= Earnings Before Interest & Tax/Total Assets = Market Value of Equity/Total Liabilities = Sales/Total Assets

This measure is very advantageous as it can be used to identify the companies that are in financial distress. Such firms should be more likely to be referenced by CDS and as suggested by theory could be prone to the Empty Creditor Problem (Bolton and Oehmke (2011), Subrahmanyam et al. (2014)). Additionally as the last measure of riskiness I use the S&P Domestic Long-Term Issuer Credit Rating. I use this rating as a separate variable together with indicator variables for upgrades and downgrades to capture the changes in credit worthiness of borrowers.

The other factor that might influence both the renegotiation outcome and the CDS trading is the size of a company. I follow Subrahmanyam et al. (2014) and include the logarithm of market value in the regression model in order to minimize the omitted variable bias. Furthermore, I use a variable controlling for the past performance of companies (I use the stock return over a previous year) as it influences the successful renegotiation likelihood and can stimulate the start of CDS trading. Moreover including lagged explanatory variables can help to limit the endogeneity problem of CDS trading (Peristiani and Savino (2011)).

(19)

19

Another important control variable that I add to the regressions is a variable reflecting the change in EBITDA/Total Assets ratio which is a measure of accounting performance (similarly as in Peristiani and Savino (2011)). Furthermore, to account for the possibility that CDS companies renegotiate less simply because they have a smaller amount of loans outstanding, I include a ratio of loans outstanding to total assets. In the next section of the thesis I provide a more thorough definition of each variable together with the sources used to calculate those variables.

The probit model that is used in this thesis is complemented in most of the specifications with both industry fixed effects and with year fixed effects. Industry fixed effects are added in order to account for unobserved heterogeneity (Gormley and Matsa (2014)) among industries that could bias the results. The companies in the sample are classified using the 2-digit Standard Industrial Classification codes (SIC) downloaded from the Compustat database. Year fixed effects are also very important as they allow to increase the accuracy of results by separating “the effect” of each year in the sample on the restructuring probability (Stock and Watson (2012)).

The main hypothesis is also tested by investigating whether following an inception of CDS trading on companies’ default the likelihood of successful restructuring is lowered. This aspect is analyzed using the same regression model however the sample is restricted to include only firms that at some point between 2003 and 2013 were referenced by CDS contracts. Using the start dates of CDS trading, which I obtained from Datastream database, I recognize the start years of CDS trading7 and only keep the observations of up to 2 years before and 2 years after the beginning of trading. This way I create a 5-year window for each company with the inception of CDS trading in the middle of a window. Such a dataset enables me to assess the difference in probability of successful loan renegotiation for the same companies before and after the start of CDS trading. In the case of this type of regressions, the coefficient on the main independent variable – CDS dummy – is expected to be negative. Negative coefficient would imply that following the inception of CDS trading reference firms are less likely to restructure as compared to the time before the start of trading. Such values of the coefficient would be in support of the main hypothesis and in turn in support of the Empty Creditor Hypothesis. The time period for these regressions is extended to 2001-2013 in order

7

I assume that a particular company had CDS trading in a given year if trading started at least 360 days before the end of a company’s fiscal year.

(20)

20

to be able to assess the renegotiation likelihood before the start of CDS trading for companies which begun to be referenced by CDS in 2003 and 2004.

In order to complement the findings from the testing of the first hypothesis I also investigate whether the Empty Creditor Problem is driven by restructurings disadvantageous for creditors. To do so I create two separate samples. In the first dataset I drop all observations with restructurings that changed loan terms in favor of creditors (in result the sample includes only restructurings beneficial for the borrowers). In the second dataset I do the opposite and only leave renegotiations that were bad for borrowers (good for creditors). I am able to do it using the variable called amend_index2 which takes positive values if a particular loan amendment was beneficial for the borrower and it is negative otherwise8. The second hypothesis is tested using the same regression model that is adopted in the analysis of the first hypothesis. I expect the coefficient on CDS_dummy to be negative and significant in the regressions run in the dataset with exclusively good-for-the-borrower restructurings. Contrarily I expect the analogous coefficient to be positive or insignificant in the results from regressions on the second dataset. Such results would imply that the Empty Creditor Problem is driven by restructurings that are bad for creditors (good for borrowers). This would be also in line with the general predictions from theoretical papers. If the results would in contrast be driven by restructurings that are beneficial for creditors then the base assumption of the Empty Creditor Hypothesis (that “empty creditors” would not willingly restructure debt to receive higher payoff from CDS contract) would be contrasted.

The methodology used for the empirical investigation of the last hypothesis is somewhat similar to the methodology used for the second one. Analogously I produce two separate datasets – one dataset that only includes restructurings that are large and the other that consists exclusively of renegotiations implementing smaller changes. In order to identify the large and small restructurings I use the information on markups in original loan terms and markups that were used in the amended loans9. I classify restructurings as being large if the markup changed by at least 25%. The remaining restructurings were classified as small (the loans with missing information on a change in markup were dropped from both datasets as they could bias the results). To test the third hypothesis I again use the same regression model and run it in both datasets. If the results of regressions in the dataset with only small restructurings would imply that CDS referenced firms are less likely to successfully

8

A more thorough definition of amend_index2 can be found in the next section.

(21)

21

renegotiate and if the results of the regressions in the second dataset would be weaker (or not statistically significant), then the hypothesis would be supported. Such results would indicate that CDS protected lenders are hard in debt renegotiations and lower the likelihood of restructuring for small renegotiations, in effect pushing the CDS referenced companies into further problems. This would be related to the limited interest in the condition of borrowers by CDS protected lenders (at the start of financial troubles of borrowers) and the increased bargaining power of creditors (limited possibility to strategically default for debtors). All this would indicate that companies referenced by CDS are associated with more sizeable restructurings.

Besides running the regressions using the data on all companies I also test all the hypotheses on the restricted sample. In those samples I include only firms that are in financial distress. I use the previously calculated Altman Z-score to recognize such companies. In the dataset used for testing the first hypothesis I simply exclude the gvkey-fyear observations for which Altman Z-score is bigger than 2,5. In the regressions with the 5-year window I exclude all the observations for a given company if in the first year of the 5-year window the particular firm had a Z-score higher than 2,5. The cutoff that I chose is not standard however it lies between the two standard values of 1,8 and 3 that are used in most of papers. The cutoff equal to 1,8 in my view is too low as restricting the dataset using such a low value of Z-score would significantly decrease the number of observations. In result, the dataset would not include enough observations to provide valuable insight into the problem (especially the datasets used for testing the second and third hypothesis). The Altman Z-score of 3 could be better, however companies with values of this measure below 3 are often recognized as being in a “grey” zone and this does not necessarily mean a financial distress. In the main dataset for testing the first hypothesis they constitute on average as much as 54% of firms in each year, which I think is too much. Due to these reasons I choose the cutoff equal to 2,5 which lies somewhat in the middle of the two standard values and allows me to create a sample of distressed companies that is not too small to perform a valid econometrical analysis.

Additionally, as a robustness check, I also run the regressions using the dataset that includes exclusively the companies with Altman Z-score higher than 2,5 and separately using the dataset with firms with Z-score higher than 3. This way I provide supplementary regressions that test if the Empty Creditor Problem is related only to distressed companies. In the light of the Empty Creditor Hypothesis I expect the coefficient on the CDS_dummy variable to be statistically insignificant.

(22)

22

4. Data

The sample used in this thesis consists of US companies and the time period chosen for the regressions testing the first hypothesis is from 2003 to 2013. For the regressions investigating the change in probability of successful loan renegotiation (regressions with a 5-year window) the sample is extended by two years and it is from 2001 to 2013. The duration of the sample used in this thesis is fairly long as compared to other research (Peristiani and Savino (2011), Bedendo et al. (2012)) and this is beneficial for the quality of the study. Including both years before and after the crisis can enhance the reliability of the findings as the recession that took place between 2008 and 2012 could distort the results if it constituted a large part of the time sample used.

The empirical investigation of the Empty Creditor Hypothesis required the necessary data to be obtained from various databases. The first part of this section provides information on the process of data retrieval and preparation. The next subsection presents the most important variables to the research together with their definitions and sources. The final part includes summary statistics that allow for the assessment of the differences between companies referenced by CDS and companies without CDS trading.

4.1 Sources of Data and Preparation

The first dataset which is crucial for this thesis and constitutes a base for the empirical analysis was prepared by dr. Tomislav Ladika using the Thomson One database. It contains information on syndicated loans outstanding of all major US companies for years 1989 -2013. Observations in this dataset are uniquely identified by: gvkey (Global Company Key Identifier), fyear (Fiscal Year), facility_id, and tranche_id (unique identifiers of loans and tranches of these loans). There is also an indicator variable in this dataset that informs whether a particular loan was amended in a given year or not. I used this dataset to recognize firms which successfully renegotiated any loans in the time period from 2001 to 2013. I created a variable called amendment_dummy that is equal to 1 if a particular firm successfully renegotiated any loans in a given year (otherwise it is equal to 0). Additionally this dataset includes a number of variables which reflect key characteristics (markup, deal amount, maturity, issue date etc.) of all these loans as well as the information about the loan terms following a restructuring. It also contains variables that inform about the size of the changes that a particular restructuring implements. Furthermore it is possible to differentiate between

(23)

23

restructurings that were beneficial for the borrower and restructurings that were beneficial for the creditor. This can be done using amend_index2 variable. This variable takes big positive values if loan terms change significantly in favor of the borrower (e.g. markup decreases a lot, principal or maturity increases a lot) and small positive values if loan terms are positive but smaller. If the restructuring implements changes that are good for the creditor, then amend_index2 takes negative values.

In order to distinguish the companies which are referenced by CDS contracts I used the Datastream database. In this research I only use the US dollar denominated senior unsecured Credit Default Swap contracts with a 5 year maturity. The reason for concentrating exclusively on these contracts is that CDSs of this type are the most widely used ones and are the most liquid. I downloaded the names of all companies which had CDS trading of this type at some point between 2003 and 2013. Additionally I downloaded the start dates of CDS trading for each company together with a panel data including CDS spreads and restructuring types. Unfortunately the Datastream database does not provide the gvkey firm identifiers that distinguish the companies in the dataset with loan restructurings. In order to be able to merge the two datasets I manually matched gvkey identifiers to CDS dataset by comparing names of CDS contracts and names of companies (using the Compustat database10). The dataset created by merging the dataset on loan renegotiations and CDS dataset from Datastream included the information on restructurings together with information on the availability of CDS trading in each firm-year pair. Using this dataset I created a binary dependent variable that differentiated CDS firms and non-CDS firms. This variable is called CDS_dummy and can be associated with analogous variables in the paper by Subrahmanyam et al. (2014) – CDS Active or in the paper by Peristiani and Savino (2011) – CDS dummy. I assumed that a particular company had CDS trading in a given year if trading started at least 360 days before the end of the company’s fiscal year. This was done to avoid situations in which a company would be assigned as being referenced by CDS trading if for instance CDS trading lasted for only one or two months in a given year11.

The next step in the data preparation process consisted of downloading other useful variables from the Compustat database. I used this database to obtain the items necessary for the calculation of the leverage ratio, Altman Z-score, market capitalization and performance

10_{I used Compustat database to download information on names and gvkeys of all US companies.}

11_{This could happen for instance if a company ends its fiscal year in March and CDS trading begins in February}

of that year. Classification of such a firm in that particular year as a CDS firm could bias the results as the trading only happened for a very short part of a year.

(24)

24

indicators (stock returns). Additionally, I also downloaded information on the credit rating of companies in my dataset. Compustat database was also helpful in the calculation of the change in EBITDA/Total Assets variable which was used to control for accounting performance. In order to limit the bias that can be introduced by large outliers I winsorized all variables on the 5-95 level12.

4.2 Variables Description

The table presented underneath summarizes the most important variables that were used throughout the research for this thesis. The first column contains the names of all variables, the second column provides explanations of how these variables were calculated and finally the last column indicates the source of data.

INSERT TABLE 1 HERE

4.3 Summary Statistics

In this subsection I present the summary statistics that enable to grasp the differences between companies referenced by CDS contracts and non-CDS companies. It is important to understand these differences before using more advanced econometrical methods in order to be able to minimize the omitted variable bias in the regressions.

Table 2 presents the basic statistics relating to financial condition and performance of CDS companies and non-CDS companies. The bottom of the table indicates that the main dataset used for the baseline regressions includes 2309 firms of which 431 are referenced by CDS and the remaining 1878 are not. The statistics included in the first column with results imply that companies referenced by CDS tend to be more levered. This result is in line with findings of Saretto and Tooks (2013) who present similar conclusions in their paper. Furthermore, statistics in the second column of the table show that CDS firms are significantly bigger than companies without CDS trading on their default. This should not be surprising as large companies could simply be associated with more interest from business partners wanting to protect against default (as they could be related to a bigger number of partners), which in turn could stimulate CDS trading. If CDS contracts were also traded on smaller firms, the liquidity would likely be much lower. The conclusions from the first two columns of table 2 imply that the regression model should be complemented with a variable representing the leverage ratio

12

I did not winsorize the proxy for the size of a company as I calculated a logarithm of it which already limits the problem with outliers.

(25)

25

as well as the size of a company. These variables should be included as both of them can have an influence on the success of renegotiation.

INSERT TABLE 2 HERE

The next three columns in table 2 inform that CDS companies can be associated with higher average credit rating, lower value of Altman Z-score and slightly lower average returns. Lower value of Altman Z-score may indicate that companies referenced by CDS tend to be in a worse financial condition. This is an important issue as it is related to the possible endogeneity problem of CDS trading. It might be the case that firms start to be referenced by CDS contracts as they become distressed. This could negatively influence the likelihood of successful renegotiation and in turn bias the results. In order to minimize the risk of omitted variable bias I include in the regression model a variable for Altman Z-score as well as a variable reflecting the stock return over a previous year (similarly as in Peristiani and Savino (2011)). In contrast to the conclusions from the comparison of Altman Z-scores, it can be noticed that CDS referenced firms tend to have a higher credit rating than non-CDS firms. This can be related to the fact that CDS firms are larger in size and bigger companies could be perceived as safer by the rating agencies.

The last column of table 2 presents the statistics that inform about the relative amount of loans that CDS and non-CDS firms take. It can be seen that firms referenced by CDS contracts have relatively less loans outstanding. This is an important finding as it suggests that regression model should be complemented by a variable controlling for this difference to account for the possibility that CDS companies renegotiate less simply because they have a smaller amount of loans outstanding.

Aside from presenting summary statistics relating to the financial condition, I also present a chart that illustrates the fraction of successful loan renegotiations for firms referenced by CDS and firms without CDS trading.

INSERT FIGURE 1 HERE

The figure presented above includes two comparisons. The first two posts relate to the statistics calculated for the whole sample. The next two posts relate exclusively to firms that at some point in time had CDS trading. It can be noticed that firms referenced by CDS have in general a higher fraction of successful loan restructurings (8,77% compared to 7,80% for non-CDS firms). The same is true for the analogous comparison of fractions in most of the years

(26)

26

in the sample (this is not presented in figure 1). Such statistics interpreted independently would stand in opposition to the Empty Creditor Hypothesis. The conclusions from table 2 however imply that CDS firms differ from non-CDS companies and a sole comparison of successful restructurings fraction cannot be considered meaningful. If CDS referenced firms differ from non-CDS firms and these differences affect the probability of successful restructuring then a comparison of fractions would be insufficient. A slightly more valid comparison is provided by the next two posts of figure 1. These posts compare the percentage of successful renegotiation for the same firms before and after the inception of CDS trading. The chart implies that following an introduction of CDS trading, on average the success rate decreases (this in turn would be in line with the Empty Creditor Hypothesis). Although the second comparison may seem to be more meaningful, it should also be interpreted with caution. The simple comparison of successful restructuring fractions does not take into account the differences between firms and changes happening in time. The further investigation of the issue that would control for these differences is needed. The results of a more advanced empirical analysis are presented in the next part of the thesis.

5. Baseline Results

This part of the thesis presents the results of the empirical analysis of the Empty Creditor Hypothesis. It is divided into three sections, each relating to one of the three hypotheses described previously. As mentioned in the methodology part it is worth to investigate the issue of the Empty Creditor Problem not only generally for all types of companies, but also exclusively for firms that are in financial distress. Such firms are the ones most likely to be facing the problem of “empty creditors” (Bedendo et al. (2012)). All the results presented in this section as well as in the robustness analysis are supposed to gradually provide an answer to the main concern of this thesis – the existence of the Empty Creditor Problem. In order to enable a quick assessment of the magnitude of results all the tables presented include marginal effects calculated after the estimation of probit regressions.

(27)

27

5.1 Empirical investigation of the first hypothesis

Table 3 presented below includes the results of probit regressions run on the main dataset prepared for the empirical research. The sample for the regressions contains information on loan renegotiations as well as the availability of CDS trading and was restricted to include observations from 2003 to 2013. The main dependent variable was not modified in any way and is equal to 1 if a particular company successfully restructured any loans (of any size and significance to creditors and borrowers) in a given fiscal year. The first two specifications relate to all companies in the sample and specifications 3 and 4 were obtained by regressing analogous model in the dataset that includes only observations of financially distressed firms (Altman Z-score<2.5). The bottom of the table informs that there were 12 744 gvkey-fyear pairs in the whole dataset of which 5 317 were relating to financially troubled companies.

INSERT TABLE 3 HERE

The first two specifications of table 3 differ only in the use of year fixed effects. Both specifications include industry fixed effects in order to control for the specifics of each industry. The results from running the first regression model, which does not include year fixed effects, imply a statistically significant relation between the availability of CDS trading and the probability of successful restructuring. The marginal effect of the change in CDS_dummy variable (a change from 0 to 1 reflects a change from a non-CDS firm to a firm referenced by CDS contracts) is equal to - 0,0191. Such a value indicates that according to the regression model firms referenced by CDS trading should have roughly 2% lower probability of successful loan restructuring as compared to non-CDS firms. This result is statistically significant at 1% level but initially may seem to be insignificant economically. One needs however, to realize that only around 8% of all loans (in my dataset) are renegotiated each year. If a company referenced by CDS trading is in general 2% less likely to successfully restructure a loan but only 8% of loans are renegotiated then the result appears to be significant economically. Unfortunately, it also needs to be noticed that following the inclusion of year fixed effects in the model, the coefficient on CDS_dummy variable becomes statistically insignificant. This can be caused by the endogeneity problem of CDS trading which would imply that there is an omitted variable bias in the regression model. I will examine this issue in the supplementary analysis at the end of the part with results. I will also elaborate on the problem of omitted variable bias in a small discussion that follows the results section of the thesis.

(28)

28

The next two specifications presented in table 3 help to verify the existence of the Empty Creditor Problem by focusing exclusively on distressed companies. The results from these regressions are very similar to the results of the first two specifications. The marginal effect of a change in CDS_dummy variable is slightly more negative (- 0,0236). It can be noticed, however that the coefficient on CDS_dummy is still statistically insignificant after the inclusion of year fixed effects in the model.

The problem of endogeneity of CDS trading that is reflected by the insignificant results of regressions including year fixed effects can be reduced by running a modified version of the regressions presented in table 3. It should be more accurate to compare the probability of successful loan renegotiation before and after the inception of CDS trading only for the firms that at some point were referenced by CDS. This way firms that never had CDS trading are removed from the sample. It can be advantageous as it is possible that such companies differ significantly from CDS companies and it would be difficult to control for all these differences. Instead, comparing the same companies reduces the risk of the omitted variable bias. If, for instance, the likelihood of successful loan renegotiation drops following the inception of CDS trading and there are no other significant events or changes during the time sample, the decrease in probability of restructuring could be attributed to the emergence of “empty creditors”.

Table 4 includes the results of the implementation of the same regression model (that was used for creation of table 3) on the sample that includes 5-year windows for the companies that at some point were referenced by CDS. Due to the fact that a large fraction of CDS contracts in the dataset started to trade in 2003, the time period was extended to additionally include information for years 2001 and 2002. This enabled me to create complete 5-year windows for companies that started to be referenced as early as 2003 and 2004. Similarly as in table 3 the first two specifications of table 4 relate to all companies in the sample and the next two were run on the dataset that only includes 5-year windows for companies that were financially distressed in the first year of the window (Altman Z-score<2,5). The condition for the restricted sample is crucial as it allows for the comparison of the probability of successful loan renegotiation for companies that were in financial distress already before introduction of CDS trading.

The empty creditor problem and contract renegotiation