The empty creditor problem : the impact of Credit Default Swaps on loan renegotiations

(1)

The Empty Creditor problem: the impact of Credit

Default Swaps on loan renegotiations

Elisa Carnevale

University of Amsterdam

Faculty of Economics and Business

Master Thesis

December, 2015

Abstract

Credit Default Swaps, introduced to transfer the financial risk between two counterparties, are often considered a double-edged sword for the economy. In particular, the theory of the Empty Creditor problem, by Hu and Black (2008) predicts that lenders with CDS insurance are less likely to help borrowers avoid bankruptcy. That is, creditors have a lower interest in monitoring the borrower, but are also less prone to sustain the debtor and sacrifice to avoid his default. This thesis empirically investigate this issue by looking at the effect of CDS on loan renegotiations for US companies with outstanding syndicated loans over the period 2004-2013. This work does not find evidence of lower likelihood of loan renegotiations for CDS firms, except for the year 2004. However, it documents more unfavorable loan renegotiation terms for CDS firms, especially for firms already in distress. This result suggests the actual existence of the Empty Creditor problem, as it implies that creditors prefer to impose conditions that can push debtors into bankruptcy.

MSc Business Economics, Finance track

Thesis supervisor: Tomislav Ladika

(2)

Statement of Originality

This document is written by Student Elisa Carnevale 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)

1 Introduction

In the early 1990s, forms of the derivatives known today as Credit Default Swaps (CDS) first ap-peared on the credit risk markets. The first modern CDS contract was negotiated in 1994. Since then, the market for CDS, alongside other credit risk instruments, has grown exponentially, starting from a total notional amount of $0.9 trillion in 2001 to a peak of $62 trillion in 2007. Following the recent financial crisis, their usage constantly decreased, but the outstanding contracts are still plentiful. Appendix A shows the size of credit risk transfer markets using various instruments from 2001 to 2014.

A CDS is an insurance-like contract that aims to transfer the financial risk between two counter-parties. The buyer of protection receives a compensation from the seller of the CDS in case of a negative credit event, generally defined as the filing for bankruptcy or the non-payment of debt by the reference entity. In return for the financial risk protection, the seller of CDS demands a premium payment over the life of the transaction, (Bolton and Oehmke, 2011). Nowadays, CDS are considered one of the most controversial financial innovations. Many analysts claim that these kind of derivatives were one of the major factors that exacerbated the recent financial crisis. For instance, it is known that CDS played a prominent role in the bankruptcy of Lehman Brothers, the collapse of AIG, and the sovereign debt crisis of Greece. These observers highlight the lack of sufficient regulation in the CDS market, mostly over-the-counter, and the potential speculation deriving from negative judgments about a company financial position. In particular, the Empty Creditor problem, first introduced by Hu and Black (2008), predicts that lenders with CDS insur-ance have little incentive to agree to a restructuring that would avoid bankruptcy. That is, creditors have a lower interest in monitoring the borrower, but are also less prone to sustain the debtor and sacrifice to avoid his default by rolling over debt, extending new financing, or proposing voluntary debt restructurings.

This thesis mainly tries to find evidence for the Empty Creditor problem advanced by Hu and Black (2008). In particular, it is focused on the effect of outstanding CDS contract for a particular reference entity on its loan renegotiations. In fact, empty creditors would prefer a firm to file for bankruptcy rather than to be involved in an out-of-court renegotiation. Hence, the main research

(5)

question: Is the likelihood of loan renegotiations decreasing for firms with CDS contracts referenc-ing their default? Furthermore, Bolton and Oehmke (2011) point to the fact that CDS-protected creditors are even less willing to restructure debt when the reference entity is already in distress. Therefore, the Empty Creditor problem is likely to be more prominent for troubled borrowers. This leads to the second research question of the paper: Is the probability of loan renegotiation falling for firms with CDS contracts that are already in financial distress? Last but not least, in case of loan renegotiations, another related question is: Can the presence of CDS negatively impact the loan renegotiation terms?

The present work gives a contribution to existing literature in the understanding of the role of CDS on the relationship between creditors and borrowers, following the steps of Subrahmanyam, et al. (2011), Danis (2012) and Peristiani and Savino (2011), among others. The divergence in results was a stimulus for this research, which could bring new insights into the mechanism of the Empty Creditor problem thanks to the wide dataset of US firms available and referring to more recent years. Nevertheless, the Empty Creditor problem is studied from an innovative point of view: while the aforementioned papers mainly analyze the effect of CDS on the probability of bankruptcy, this thesis examines two strictly related effect of CDS, that is, the probability of loan renegotiations and the impact on their terms. The advantage of considering loan renegotiations instead of bankruptcy rates is that they can be contracted multiple times and with different terms for a single company, providing a wider sample. Furthermore, loan renegotiations are a direct expression of the will of creditors and borrowers, whereas bankruptcy can depend on a variety of factors among which the willingness of creditors is important but less relevant. The confirmation of the stated hypotheses would demonstrate that CDS modify creditors attitude towards the borrowers, leading to an unfair penalization of firms with CDS trading on their default. This has important ethical implications, and signals the need for specific regulation.

To answer the research questions an unbalanced panel data set is used. The dataset contains ob-servations of 2140 US firms that received a syndicated loan in the period between 2004 and 2013. For each firm and year, it is determined whether the company renegotiated any loan and whether it had any outstanding CDS contract trading on its default. To answer the second question, the dataset was then split based on whether the firm was in distress (determined using the Altman

(6)

Z-score). Finally, for the third research question, an index that quantifies the favorability of the loan renegotiation terms is computed. The sources of these data are Thomson One, Datastream and Standards & Poor's Compustat. A logistic regression on the panel data tests if there exist a causal relationship between the presence of CDS contracts on a companys debt and the probability of loan renegotiations, whereas a linear regression establishes whether the presence of CDS nega-tively impacts the loan renegotiation terms.

The first part of this analysis aims at testing whether firms that may be affected by the presence of empty creditors show a lower probability of loan renegotiations with respect to other firms, once controls for different firms' characteristics are introduced. Contradicting the empty creditor theory, there is no evidence that companies whose bondholders might be insured via CDS are less likely to renegotiate their debt. This part also studies whether the presence of empty creditors may be more prominent for firms already in distress relatively to safe firms. Again, the results are not definitive, as, although they are coherent within the different regressions, they are statistically non-significant. Additionally, when analyzing the sample by single years, this study could not reach conclusive find-ings, except for the year 2004, when CDS firms show a 36% lower probability of undergoing a loan renegotiation compared to non-CDS firms.

This thesis found evidence of the Empty Creditor problem when analyzing the impact of CDS on loan renegotiation terms. For a firm with outstanding CDS, the mean amendment index, which ranges from -4 to 4, is lower than that of non-CDS firms by about 0.47. The negative coefficient supports the third research hypothesis by showing that terms of loan renegotiation are less favor-able for firms with outstanding CDS contracts, suggesting the presence of Empty Creditors. As expected, the effect is more prominent in case of firms already in distress, the coefficient falls to -0.61, and non-significant in case of non-distressed firms. These results are consistent also when a propensity score matching exercise is introduced and different definitions for the amendment index are used in the regressions.

This work contributes to the debate on the Empty Creditor problem and CDS role in general by providing further proof that CDS are a penalizing factor for companies. In fact, they are subjected to tougher loan renegotiation terms that can eventually push them into default. This effect is even

(7)

stronger for firms already in distress. Hence, the results are in line with the theory delineated by Hu and Black (2008), and confirms the theoretical and empirical works of Bolton and Oehmke (2011), Danis (2012) and Subrahmanyam et al. (2011).

The remainder of this paper is structured as follows. Section 2 overviews the existing theoretical and empirical papers referring to the ambiguous role of CDS in the economy and in particular to the Empty Creditor problem. Section 3 presents the two hypotheses that arise from the conflict of interpretations on the Empty Creditor problem observed in the literature review. Moreover, it describes the methodology implemented to test these hypotheses. Section 4 specifies the data collection procedure, and comments on the characteristics of the sample. Section 5 presents the em-pirical analysis and results. Section 6 contains the robustness checks performed on the regressions. Lastly, section 7 concludes and gives insights for further research.

2 Literature Review

This section discusses relevant literature related to the Credit Default Swaps market and the Empty Creditor problem. The literature review is divided into four main parts. The first paragraph provides an overview of the literature related to the market and the costs and benefits of Credit Default Swaps. The second paragraph focuses on the definition of the Empty Creditor problem, from its inception to its economic implications, and summarizes the papers that developed the subject using merely theoretical models. The third paragraph focuses on the empirical evidence of the Empty Creditor problem. The final paragraph discusses further empirical data relevant to this thesis.

2.1 The CDS market

Credit Default Swaps (CDS) are contracts that offer protection against the risk of a negative credit event by a particular debtor, in exchange for a premium. The premium is proportional to the credit event risk of the reference entity, the issuer of the CDS underlying security, and needs to be paid periodically. Typical credit events are bankruptcy, non-payment, and, in some cases, debt restructuring or a credit rating downgrade.

(8)

CDS have existed since the early 1990s to reduce banks exposure to corporate credit risk. Their usage strikingly increased thereafter, especially between institutional investors and hedge funds, from an outstanding amount of $1 trillion in the early 2000s to an all-time record of $62.2 trillion by the end of 2007 (ISDA Market Survey). One of the reasons for this remarkable growth is the fact that CDS can be traded independently from their underlying securities, allowing the CDS market to be much greater than the underlying debt market. Nevertheless, following the financial crisis, the outstanding amount of CDS fell to $26 trillion by the end of 2010 (ISDA market survey, 2010). Academic literature on Credit Default Swaps is extensive, and it evolved enormously after 2008, when the financial press openly accused them to be one of the reasons, if not the main reason, of the worldwide economic crisis. Just few years before, the economist Alan Greenspan publicly stated that credit derivatives and other complex financial instruments had contributed “to the development of a far more flexible, efficient, and hence resilient financial system than existed just a quarter-century ago.” (Greenspan 2004). Nowadays, the role of CDS is still puzzling and requires new insights, as researchers who have inspected the different aspects of this credit derivative came to divergent conclusions. However, the common sentiment towards CDS is that they are a double-edged sword for the economy.

Hull (2008) and Hu and Black (2008) noted that the introduction of CDS contracts transformed the characteristics of the traditional debtor-creditor relationship. Credit default swap contracts enable the decomposition and transfer of the underlying reference entity’s credit risk, therefore allowing the creditors control and cash flow rights to be partially or fully separated. A possible consequence of this separation is the emergence of holders of debt and CDS (empty creditors) that are no longer concerned with the debtors distress resolution, and may even push them into bankruptcy. The Empty Creditor problem just described is further developed in Paragraph 2.2, along with its implications, as this thesis contributes to the knowledge of this phenomenon.

Stulz (2010) and Jarrow (2010) are among the scholars who manifested mainly positive opinions about CDS. The former advocates that the recent financial crisis cannot be ascribable to an excessive use or exploitation of CDS. In fact, losses referring to these instruments occurred because of defaults on subprime mortgages and the waning of liquidity on such securizations. The latter, starting from

(9)

the similarity between CDS and insurance contracts, argues that CDS ultimately boost aggregate investment and economic growth. Another positive effect is that banks using CDS are able to provide lower corporate loan spreads (Norden, Silvia-Buston and Wagner (2012), Hirtle (2009)). On the other hand, Allen and Carletti (2006) illustrate that credit risk transfer can be unfavorable to welfare, leading to a possible inter-sector contagion and a more likely outset of a financial crisis. Finally, several studies discuss in depth the effect of asymmetric information and insider trading problems, specifically in the CDS market. According to Ashcraft and Santos (2009), the weakened monitoring incentives for banks leads to higher borrowing costs for firms with higher default risk. Acharya and Johnson (2007) find that banks could use credit risk transfer to abuse credit protection sellers, due to their information advantage on the borrowers ability to pay. Morrison (2005) argues that, as the monitoring activity by banks is adversely affected, borrowers may decide to issue low quality bonds instead, and thus reduce welfare. Minton et al. (2009) conclude that using CDS for hedging is limited owing to the same asymmetric information problems and the inability of banks to use hedge accounting when hedging with credit derivatives. Furthermore, Arping (2004) argues that CDS can help overcome an incentive problem between banks and borrowers, given that CDS contracts expire before maturity.

Summarizing, CDS markets have become the subject of various policy disputes, including their role in the recent financial crisis; their impact on the debtor-creditor relationship as well as firms costs of capital, financing choices, and credit risk. The result of CDS trading is uncertain: on one side, CDS can enhance risk sharing and therefore result in an advanced risk management instrument and an expansion of credit supply, with positive effects on welfare; on the other side, it could tempt bankers into a less scrupulous monitoring activity.

2.2 The theory of the Empty Creditor problem

The following paragraph will present the notion of Empty Creditor, first advanced by Hu and Black (2008), and the related empirical evidence, which is fundamental for this thesis. Hu and Black (2008) concept is purely theoretical, and is not backed by any empirical evidence. Nonetheless, it captured the interest of the entire financial community, and bankruptcies, like those of the Canadian paper manufacturer AbitibiBowater and General Motors, were attributed to the fact that some

(10)

bondholders owned CDS and stood to gain more by bankruptcy than by reorganization (Soros, 2008). The Economist advanced the same observations, in an article stating that CDS-protected lenders can often make higher returns from CDS payouts than from out-of-court restructuring plans (The Economist, 2009).

As stated earlier, credit derivatives introduced the possibility of having contractual rights on an underlying security, without the annexed exposure to borrowers’ default (Hu and Black, 2008). Hence, the interests of CDS-holding creditors are not aligned with those of other debt-holders, who cannot rely on CDS protection. In particular, the Empty Creditor problem predicts that lenders with CDS insurance are less likely to help borrowers avoid bankruptcy. That is, creditors have a lower interest in monitoring the borrower, but are also less prone to sustain the debtor and sacrifice to avoid his default by rolling over debt, extending new financing, or proposing voluntary debt restructurings. This is particularly relevant, as failure of out-of-court debt restructurings can lead to future bankruptcy for the firms involved. Moreover, the knowledge of the Empty Creditor problem is also important for the present investment and financing decision-making of companies. In fact, despite the absence of current financial distress, firms need to consider its future probability and the consequent difficulty in renegotiations (Bolton and Oehmke, 2011).

Bolton and Oehmke (2011) formally model the Empty Creditor problem. They compare the ex-ante and ex-post outcomes of CDS presence, considering a hypothetical firm with a positive NPV investment project financed with debt, and assuming a limited commitment problem. From an ex-ante perspective, the stronger bargaining power of CDS-holding creditors allows for an increased debt capacity for borrowers, that is the total amount of debt a firm can incur. This means that a greater number of positive NPV projects will receive financing. However, Bolton and Oehmke (2011) find that in equilibrium lenders tend to over-insure their position, by buying CDS-protection exceeding the maximum amount they can receive in restructurings. Thus, creditor will have no desire to renegotiate with borrowers, preferring the CDS re-payment, and increasing the likelihood of costly bankruptcies, which are an inefficient burden also for the society. Bolton and Oehmke (2011) do not limit to the analysis of the Empty Creditor problem, but also suggest various solutions. The theoretical framework laid down by Arping (2014) supports the evidence of creditors over-insurance of debt positions.

(11)

Campello and Matta (2012) model reconsiders the negative role of CDS during the recent financial crisis, claiming that the Empty Creditor problem is procyclical. The authors note that over-insurance occurs more often during an economic upturn rather than during an economic downturn, so the Empty Creditor problem is particularly prominent during the times of economic boom. Nevertheless, Campello and Matta (2012) argue that banning CDS is not a solution, but rather detrimental to welfare.

Completely of a different opinion with respect to Hu and Black (2008) is the International Swaps and Derivatives Association (ISDA), which in a recent paper questions the validity of the Empty Creditor hypothesis. Observing the frequency of out-of-court restructurings in comparison with the number of default events before and after the introduction of CDS (between 1984 and 2009), the paper shows that there is no apparent relationship between the two variables. In fact, the proportion of out-of-court restructurings seems to increase following 2003, the year of the inception of CDS trading. Furthermore, this research challenges the soundness of the over-insurance theory (Bolton and Oehmke, 2011), arguing that the strategy would be prohibitively expensive and in any case could not influence the market (Mengle, 2009).

The aforementioned theoretical studies need to be confirmed by further empirical research, in order to verify the proposed models and apply them to actual economic data.

2.3 Empirical evidence on the Empty Creditor problem

This paragraph summarizes the empirical findings relative to the Empty Creditor problem. The studies related to the topic are relatively few and consist of mostly working papers; moreover, the majority of them state that their work is merely preliminary. With regard to the results, the following cited papers do not reach consistent conclusions. In fact, while some authors confirm, at least for a restricted period, the existence of the Empty Creditor problem, others reject the theory of Hu and Black (2008). Another observation is that some of the works are not reliable due to either a restricted sample or a limited data selection period. Finally, it is important to consider papers studying the relationship between the presence of CDS trading and the probability of bankruptcy, as the latter is strictly connected to the probability of loan renegotiations, the topic of this thesis.

(12)

Peristiani and Savino (2011) examined the Empty Creditor problem observing the probability of bankruptcy of US nonfinancial firms between 2001 and 2008 and using two different models. The first approach reports a higher default probability for CDS firms only in 2008; the second approach exhibits a greater expected default risk during 2004-08. However, their empirical analysis reveals no significant link between CDS and the default probability over the entire period. A possible explanation for their findings could be that the Empty Creditor hedging strategy surfaced gradually over a long period, as CDS market participants became more sophisticated (Peristiani and Savino, 2011 p.10).

Danis (2012), using a sample of 80 distressed exchange offers between 2006 and 2011, shows that the average participation rate in out-of-court restructurings by CDS-referenced entities decreases by 29 percentage points. In a similar research, Narayanan and Uzmanoglu (2012) observe a sample of 81 distressed exchange offers between 2004 and 2011 and show evidence of the Empty Creditor holdout. Moreover, they find that distressed debtors mitigate the Empty Creditor effect on their bankruptcy risk by restructuring their debt strategically.

Subrahmanyam, Tang and Wang (2011), using a sample of 901 US CDS transactions between 1997 and 2009, find evidence that the introduction of CDS led to a more than double increase in default probability, and a rise in the rating downgrade probability for the reference firms. However, the endogeneity control of the regression is doubtful, as the authors use two questionable instrumental variables in the IV regression. In fact, they use foreign exchange hedging positions of lenders and bond underwriters and lenders’ Tier One capital ratio as instruments for CDS trading. Nonetheless, an important implication of this study is that firms with contracts excluding restructuring clauses are more adversely subjected to CDS trading effect. In fact, the CDS contracts that include debt restructuring as a credit event will compensate their holders even when the debt of the reference company is restructured. (Subrahmanyam, Tang and Wang, 2011). This implication is relevant for this thesis because it gives insight on how to treat the initial data, as for each CDS contract, we know whether restructuring is covered as a credit event.

Contrary to the predictions of the empty creditor hypothesis, Bedendo, Cathcart, and El-Jahel (2012), do not find that CDS contracts are associated with a higher probability of bankruptcy.

(13)

However, they use a relatively small sample of US firms that either filed for bankruptcy or incurred in out-of-court renegotiations during a short time period (2008-2009). This result could be due to the procyclical feature of the Empty Creditor hypothesis developed by Campello and Matta (2012), and described in the previous paragraph.

2.4 Further empirical studies

When studying the Empty Creditor problem, it is necessary to isolate all the other factors that can influence the behavior of firms, in particular the choice between out-of-court debt restructurings and formal bankruptcy proceedings. These factors include several financial aspects, as the complexity of a firm’s capital structure, asset tangibility and outstanding liabilities.

Nonetheless, CDS trading can affect some of the aforementioned financial aspects. Saretto and Tookes (2013) find that firm leverage increases significantly after the introduction of CDS trading. Hence, subsequent studies added a control for leverage in their regression analysis, both before and after the inception of CDS trading (Subrahmanyam, Tang and Wang, 2011). In a previous paper, Jostarndt and Sautner (2009) also show that higher leverage and debt levels can influence the restructuring success of firms. However, Jostarndt and Sautner (2009) do not distinguish between CDS and non-CDS companies. Their conclusions suggest that leverage should be included as a control variable in a regression analysis aiming at establishing the role of CDS on the likelihood of debt restructuring.

The present work tries to give a contribution to the understanding of the Empty Creditor problem, following a similar line of reasoning with respect to the above-mentioned empirical research. The fact that their results are not completely consistent, gives an incentive to investigate further on the topic.

3 Hypotheses and Methodology

This section presents the hypothesis and methodology resulting from the analysis of the literature review discussed in the previous section. The first paragraph describes the hypotheses related to the research questions. The second paragraph illustrates the methodology developed from the stated

(14)

hypotheses, and it gives a summary of the necessary data as well as the variables comprised in the regression model.

3.1 Hypotheses

Starting from the literature that asserts the existence of the Empty Creditor problem, this paper wants to contribute to the debate through further empirical research. Following the first descrip-tion of the Empty Creditor by Hu and Black (2008), the current literature did not reach consistent conclusions, as depicted earlier. Although confirmed in the papers by Peristiani and Savino (2011), Danis (2012), Narayanan and Uzmanoglu (2012) and Subrahmanyam et al. (2011), the theory was reconsidered by Bedendo et al. (2012), who studied in particular the relationship between CDS contracts and the probability of bankruptcy. Nevertheless, given the majority of positive signals about the existence of the Empty Creditor problem, there is an incentive to pursue a new study on the topic.

Following the example of Subrahmanyam et al. (2011), who analyzed the link between the intro-duction of CDS and the likelihood of bankruptcy for the reference entities, the present work tries to establish whether there is a positive causal relationship between CDS trading and the probability of loan renegotiations for firms. In fact, the probability of bankruptcy and the probability of loan renegotiations should be strictly connected, and of opposite sign. This could be an important step towards the comprehension of the mechanism and effects of CDS trading, as the valuable data relating to loan renegotiations are vast and easily available, and could potentially overcome some of the issues already mentioned about the study of Subrahmanyam et al. (2011). Moreover, an important reason to consider loan renegotiations rather than bankruptcy rates is that they can be agreed multiple times over the lifespan of a company, leading to a wider sample. Finally, loan renegotiations are a direct expression of the will of creditors and borrowers, therefore analyzing their probability and terms can highlight the presence of the Empty Creditor problem, defined as an alteration of lenders behavior. On the contrary, bankruptcy can depend on a variety of factors among which the willingness of creditors is important but less relevant.

Hence, according to the Empty Creditor theory, through studies that control for other factors in-fluencing the dependent variable, we should find a negative relationship between the presence of

(15)

CDS contracts and the probability of loan renegotiations.

Hypothesis 1 (Baseline): The likelihood of loan renegotiations decreases if firms have outstand-ing CDS contracts referencoutstand-ing their default.

In order to verify this hypothesis it is necessary to collect a dataset comprising the largest possible number of observations on loans made to US firms, the largest market for CDS, both with and without CDS contracts outstanding. The dataset will also need to include a set of control variables to account for firm-level differences. Moreover, data on the macroeconomic state of the economy should also be considered, as the current financial market conditions will naturally influence the probability of loan renegotiations. (Campello and Matta, 2012).

Furthermore, Bolton and Oehmke (2011) point to the fact that CDS-protected creditors are even less willing to restructure debt when the reference entity is already in distress. Therefore, the Empty Creditor problem is likely to be more frequent when the borrower is in this situation. This leads to the second hypothesis of the thesis.

Hypothesis 2 (Loan Renegotiations Conditional on Distress): The probability of loan renegotiation falls for firms with CDS contracts outstanding that are already in financial distress. The Altman Z-score is used to define whether a firm is in financial distress. The Z-score was first introduced by Altman (1968), and it results from a linear combination of five weighted business ratios, as follows:

Z-score = 1.2X1+ 1.4X2+ 3.3X3+ 0.6X4+ .999X5 1

The Z-score permits to identify three possible zones that allocate firms based on their state of distress. Companies with a score above 2.99 are considered to be safe; those with a Z-score ranging from 1.81 to 2.99 are in a so-called grey-area, and firms with a score lower than 1.81 are close to bankruptcy. Therefore, to answer the second research question, firms are divided into two subsam-ples, the first including firms with Altman Z-score below 1.81, and the other including firms with

1

X1 is calculated as Working Capital/Total Assets, and is a liquidity measure that gives insight on a companys

ability to cover its short-term financial obligations. X2 is Retained Earnings/Total Assets. This ratio measures

the profitability that in turn reflects the company's age and earning power. X3 is Earnings Before Interest and

Taxes/Total Assets, and is a an indicator of how effectively a company is using its assets to generate earnings. X4

is Market Value of Equity/Book Value of Total Liabilities, it shows how much the firms assets can decline in value before the liabilities exceed the assets and the firm becomes insolvent. Lastly, X5 is Sales/Total Assets, a standard

(16)

Altman Z-score above 2.99. These restricted chosen samples of firms are also useful in order to avoid a potential selection bias (Subrahmanyam et al., 2011).

Finally, the Empty Creditor theory could have impacts not only on the likelihood, but also on the terms of the potential loan renegotiations. Hence, it is expected that the renegotiations terms for CDS firms are tougher. This leads to the third hypothesis of this thesis.

Hypothesis 3 (Renegotiation terms favorability): The presence of CDS negatively affects loan renegotiation terms.

A decrease in the index caused by the presence of CDS would suggest the existence of the Empty Creditor problem. In fact, a CDS-protected creditor, through renegotiation terms that are partic-ularly unfavorable to borrowers, is expected to impose tougher conditions that could push debtors into bankruptcy. An index variable is computed in order to define the favorability of loan rene-gotiation terms for a firm. The Amendment Index is a simple index with values ranging from -4 to 4. The index is increased by 1 if the renegotiation decreases the loan markup, and an equal value is subtracted if the renegotiation increases the markup. Similarly, the index is increased by 1 if renegotiation increases loan principal or duration, and decreased by 1 if renegotiation decreases principal or duration. Further, the index is increased by 1 if the renegotiation removes a loan covenant, and decreased by 1 if it adds a covenant. In other words, the index is incremented by 1 for each change in the borrower's favor during renegotiation, and decremented by 1 for each change in the lender's favor. Hence, positive values indicate that renegotiation is overall good for the borrower.

3.2 Methodology

The elaboration of the data in order to test the first hypothesis is done through a logistic regression on US firms panel data, a regression model that allows the estimation of the probability of a binary response depending on one or more independent variables. Based on the aforementioned hypothesis, the dependent variable will reflect the probability of loan renegotiations for the selected sample of firms, conditional on the presence of outstanding CDS contracts referencing to their default. In the logistic regression stated below, this variable is represented in Amendment, an indicator variable

(17)

equal to 1 for loans that amend a previously issued loan, and which is able to determine whether the borrower renegotiated any loans in a specific year.

The key explanatory variable of interest is CDS Trading, a dummy variable that equals to one for firms that have CDS contracts outstanding in a given year, and zero for firms that do not. The corresponding coefficient is expected to be negative and significant, as the presence of CDS contracts should decrease the likelihood of loan renegotiations, due to the predicted Empty Creditor problem. Since this thesis uses a logistic regression model, there should also be a lower than one odds ratio corresponding to the CDS Trading variable. In addition, the same regression is run on two subgroups of firms based on their distressed status to capture the different CDS effects on their probability of loan renegotiations. In case of firms that are considered to be already in distress, that is with an Altman Z-score lower than 1.81, the odds ratio related to the variable is expected to be strongly negative, and lower with respect to the odds ratio relative to non-distressed CDS firms (with an Altman Z-score greater than 2.99). This is due to the fact that firms already in distress should face a more prominent Empty Creditor problem.

The following equation is at the base of the logistic regression:

Amendmentit = β0 + β1 CDS T radingit + β3 Zit + uit

Where Amendmentitis a binary variable that denotes whether the borrower renegotiated any loans

in a determined year and CDS T radingit is binary variable that determines whether the borrower

has CDS contracts outstanding in the same given year. The Zit variable represents the set of

firm-level and macroeconomic control variables included in the regression, which are derived from the studies of Subrahmanyam et al. (2011), Campello, Ladika and Matta (2015), Bedendo et al (2012) and Danis (2012). These control variables mitigate the omitted variable bias and include firm size, leverage ratio, credit rating, Tobins Q, tangibility ratio, and profitability ratio. Moreover, a set of dummy variables is included to control for the time fixed effects and industry fixed effects of the observed panel data. As Petersen (2009) suggests, an additional step is to cluster the standard errors in order to avoid the bias due to the within-firm autocorrelation of the sample. A complete table of the variables definitions is presented in Appendix B, while Appendix C reports a correlation matrix that verifies the non-existence of multicollinearity between the regression variables. In fact,

(18)

the coefficients in the correlation matrix are below the threshold that could indicate multicollinear-ity issues.

The primary coefficient of interest in our logistic model is exp(β1). An estimate lower than 1 for

exp(β1) indicates that the probability of loan renegotiations is lower for firms that have

outstand-ing CDS contracts referencoutstand-ing their default, supportoutstand-ing this thesis hypothesis. In fact, in order to interpret properly the logistic regression, a table showing the odds ratios for each variable is reported.

The first firm-level control variable included in the regression is the logarithm of market capitaliza-tion, a measure of the company’s size. Intuitively, smaller firms should have a lower probability of loan renegotiations, due to the harder access to credit and banks’ tight credit standards. Neverthe-less, the sample used for this thesis includes only data on syndicated loans, which are issued usually to larger-than-average firms. Hence, the related coefficient is not predictable, but it is reasonable to include it in the model. The second control variable is the credit rating of the company. This measure is an evaluation of the credit worthiness of a borrower, and gives a prediction of its default risk. The higher the rating, the lower the probability of financial distress. Loan renegotiations should be less complex for highly rated firms. Hence, the corresponding coefficient is expected to be positive. The variable Profitability, defined as Earnings Before Interest and Tax (EBIT) divided by total assets, is another important control variable. Bedendo et al. (2012) state that profitability and bankruptcy recovery rates have a positive relationship, incentivizing bankruptcy proceedings for creditors of highly profitable, and therefore highly valuable, firms. In turn, this means that the probability of loan renegotiations for these firms will be lower, and the Profitability coefficient is expected to be negative. The variable Tangibility, calculated as property, plant and equipment over total assets, measures the level of tangible assets of a company. Firms with higher levels of tangibility guarantee higher recovery rates to existing creditors in case of bankruptcy procedures, thus creditors are less favorable to pursue loan renegotiations (Jostarndt and Sautner, 2009).Con-sequently, the coefficient is expected to be positive. The other two firm-level control variables are leverage, computed as the sum of long term debt and current liabilities over total assets, and To-bins Q. The first assesses the ability of a company to meet financial obligations, while the latter measures the potential growth of the firm. These two variables must be taken into account because

(19)

they could influence the probability of loan renegotiations. Moreover, a set of dummy variables corresponding to the years in the observed panel data is added to the regression, in order to control for time fixed effects, that is, for variables that are different over time, but constant across entities. Finally, the macroeconomic control variable included in the regression is Term Slope, the difference between the yield on a 10-year US Treasury Bond and 2-year Treasury Note. This variable should account for the yearly variations in market conditions that could affect loan renegotiations. In order to test the third hypothesis of this thesis, a linear regression with the same independent variables as the one described earlier is executed on the sample of US firms. In this case, the dependent variable is Amendment Index, described in section 3.1.

Amendment Indexit = β0 + β1 CDS T radingit + β3 Zit + uit

The coefficient of interest is β1, and it is expected to be negative and significant, as the Empty

Creditor hypothesis predicts that the presence of CDS should impact negatively the loan renegoti-ation terms.

As previously stated, the regression model relies on a CDS dummy variable. This indicator variable is a useful tool in order to determine whether a firm had outstanding CDS contracts in a given year, but unfortunately, it does not distinguish between covered and naked CDS positions. This could bias the regression results, as the Empty Creditor problem manifests itself only when the CDS-holder is an existing bondholder of the reference entity (Subrahmanyam, Tang and Wang, 2011).

Most of the past research studying the effect of CDS trading on the likelihood of bankruptcy state a potential endogeneity problem. In fact, as the probability of an imminent bankruptcy for the reference entity increases, creditors might be willing to cover their positions through CDS. This could also be valid for an increase in loan restructuring likelihood, though probably in a lesser magnitude. Nonetheless, this potential bias is not likely to affect this thesis, as it includes only contracts with no-restructuring clause, where loan restructurings are not considered as a credit event.

(20)

4 Data and Descriptive Statistics

This section presents an explanation of the dataset, both as regards to its origin and its sum-mary statistics. The first paragraph describes from which sources the data are obtained, and the choices made to create the final sample. The second paragraph reports the specific characteristics attributable to the dataset.

4.1 Data

The dataset includes observations for all US firms that had outstanding syndicated loans in the period between 2004 and 2013. For each borrower and each year, the key characteristics of the outstanding loan are available, including a binary variable that indicates whether the loan was renegotiated and an index that indicates the loan renegotiation terms favorability. Moreover, other variables specify whether it had any outstanding CDS contract during the year and account for different firm characteristics. Due to some missing values, the sample consists of an unbalanced panel dataset with annual data.

The aforementioned dataset is collected from two main data sources: Thompson One and Datas-tream. The former covers all the syndicated loans issued to US firms, allowing the download of the necessary Loan Renegotiation Data for this thesis. The latter contains CDS data. From this database, the spreads of CDS with no-restructuring clauses and 5 years maturity are collected. As described in the literature, these kind of contracts are defined as the most liquid (Oehmke and Za-wadowski, 2012). Finally, the various firm specific control variables are collected from the Standard & Poor's Compustat database, whereas the macroeconomic control variable is available at the US treasury database. Since the data are taken from these different sources and need to be merged to form a final dataset, only firms with a given Global Company Key (the GVKEY company identifier in Compustat) are collected. Since Datastream does not contain this variable, this passage was computed manually.

Although similar studies, including Peristiani and Savino (2011), avoid the inclusion of financial firms in their datasets, important empirical papers that questioned the Empty Creditor problem did not exclude them (Subrahmanyam et al., 2011 and Bolton and Oehmke, 2011). Hence,

(21)

finan-cial firms are considered part of the sample. In addition, the choice of considering only syndicated loans, usually granted to larger firms, allows for a better match in terms of size between firms with outstanding CDS contracts and the remaining ones. This, in turn, can potentially reduce endogeneity issues.

The final sample contains 2140 unique reference entities that received a syndicated loan between 2004 and 2013. Among them, 422 had CDS trading on their default at any point in time during the same period. At a later stage, firms are distinguished in two groups depending on their distressed state.

The number of companies by year in the analyzed group varies from a minimum of 1031 to a maxi-mum of 1360 companies. Firms with outstanding CDS contacts are a smaller fraction with respect to the firms without CDS throughout the sample period. A clear tendency in both the amount of renegotiations and the its yearly percentage is not noticeable, however, the percentage of renegoti-ations falls during the two years following the 2008 crisis, before rising again. Table 1 presents the overall picture: for each year, the total number of firms, the number of loan renegotiations, and the percentage of loan renegotiations is listed.

Table 1: Loan Renegotiation rate for firms with and without CDS

This table shows a summary of the companies with and without CDS that received a syndicated loan in the period from 2004 to 2013. For each year, the total number of firms, the number of loan renegotiations, and the percentage of loan renegotiations is listed. The data are collected from Thompson One and Datastream.

Non-CDS firms CDS firms Year Total number

of firms Number of firms Number of Renegotiations % Number of firms Number of Renegotiations % 2004 1194 883 87 9.85 % 311 28 9.00 % 2005 1225 905 96 10.61 % 320 36 11.25 % 2006 1269 938 92 9.81 % 331 38 11.48 % 2007 1282 928 57 6.14 % 354 33 9.32 % 2008 1347 991 82 8.27 % 356 35 9.83 % 2009 1360 1016 43 4.23 % 344 23 6.68 % 2010 1312 977 22 2.25 % 335 11 3.28 % 2011 1292 959 40 4.17 % 333 21 6.31 % 2012 1227 911 73 8.01 % 316 23 7.28 % 2013 1031 767 37 4.82 % 264 24 9.09 %

(22)

4.2 Descriptive Statistics

This paragraph is dedicated to the understanding of the final sample from a statistical perspec-tive. Table 2 shows the summary statistics for the control variables included in the regression, distinguishing between firms with and without outstanding CDS contracts. For each variable, the difference between the mean of the two groups is then tested with a t-test in order to grasp their dissimilarity. A comparable table is shown in Appendix D, where the statistical variables are com-puted regarding distressed and healthy firms separately.

Observing the second column of table 2, which reports the standard deviation for each control vari-able, a difference in the heterogeneity of the two groups is noticeable. In particular, the subsample of non-CDS firms has a wider distribution in most of the listed variables.

The most evident asymmetry between the two groups is the size of the company, captured by the market capitalization variable. In fact, the size of companies without CDS trading on their default is much smaller than the second group. The average CDS firm is more than 7 times larger than the average non-CDS firm during the period under observation. This is because smaller firms are un-likely to attract any significant interest from CDS buyers (Peristiani and Savino, 2011). Moreover, firms without CDS are mostly unrated by Standards and Poor's, only 34% of companies are rated, in contrast to the CDS group (94%). Nevertheless, companies that belong to CDS group show an important heterogeneity.

A more straightforward difference is in the Altman Z-score variable. Firms included in the CDS group show a lower mean value (2.84) than the other subset (3.30). As said earlier, a lower Z-score means that a firm is more subjected to face distress in the near future. Even if firms are not yet in a full distress condition, the market is able to understand and take advantage of their financial weaknesses. Hence, investors and creditors are more prone to buy CDS on the default of lower Z-score firms. This is reflected in the sample under observation.

The subsamples are however comparable in terms of leverage and profitability, two variables usu-ally related. With a marginal but statisticusu-ally significant difference, firms with CDS show a higher leverage ratio and higher profitability. Therefore, CDS companies seem to be slightly riskier, with annexed higher earnings relative to total assets.

(23)

Table 2: Descriptive Statistics for firms with and without CDS

This table shows the summary statistics for the control variables included in the regression (and described in Appendix B), for firms with and without CDS. The variable Market Cap. is the logarithm of market capitalization. Variables are winsorized at the 1-99% level. The financial ratios are computed using data from Compustat database. The difference between the mean of the two groups is tested with a t-test.

Non-CDS firms CDS firms Variable Mean St. Dev. 25th Perc. Median 75th Perc. Mean St. Dev. 25th Perc. Median 75th Perc. Diff. Market Cap. 6.64 1.54 5.73 6.77 7.65 8.88 1.44 8.02 8.96 9.81 -2.24*** Leverage 0.47 0.22 0.33 0.44 0.58 0.51 0.19 0.39 0.48 0.59 -0.03*** Profitability 0.08 0.08 0.04 0.08 0.12 0.10 0.07 0.06 0.09 0.13 -0.01*** Tangibility 0.58 0.42 0.24 0.49 0.86 0.63 0.38 0.31 0.60 0.90 -0.05*** Tobin's Q 1.65 0.81 1.12 1.41 1.90 1.63 0.66 1.18 1.44 1.87 0.02 Z-score 3.30 2.53 1.69 2.91 4.46 2.84 1.87 1.57 2.62 3.80 0.46*** Amend Index 1.22 1.53 0 1 2 1.11 1.62 0 1 2 0.11

In order to test the second hypothesis of this research, the sample is split in two subgroups, one including firms already in distress (with an Altman Z-score<1.81) and the other containing so-called safe firms (Altman Z-score>2.99). These two groups still distinguish between firms with and without CDS. As regards to presence of CDS, by observing the newly separated groups, there is no anomaly from the previously stated differences.

Unsurprisingly, distressed firms are generally smaller, as the market capitalization variable indi-cates, have a much smaller annual profitability (3% for distressed firms versus 12% for safe firms), and are riskier. Moreover, distressed firms have a higher tangibility ratio, probably due to their lack of liquidity, and a lower Tobins Q value. This last variable suggests that the firms in the sample are mostly overvalued, especially safe firms.

Finally, an important remark is that the different subgroups under consideration (first between CDS and non-CDS firms and then subdividing the sample between firms already in distress and safe firms), created from the total available data, are overall comparable in terms of numbers and consistency in values. These are the right premises to perform a significant empirical analysis, which is outlined in the next section. The only questionable characteristic is the mean size of firms. In fact, comparing much larger CDS firms with non-CDS ones could bring endogeneity issues to the regression.

(24)

5 Results

This section displays and discusses the main results of this thesis. The first part presents the results of the regressions used to test the hypotheses described in section 3.1. In particular, the first paragraph describes the regressions on the likelihood of loan renegotiations. Thereafter follows the regressions on the loan renegotiations term index.

5.1 Regressions on the likelihood of loan renegotiations

The results of the baseline logistic regressions are presented in table 3. These regressions measure the impact of outstanding CDS contract on a company’s debt on the likelihood of loan renegotia-tions. The dependent variable is Amendment, a binary variable that denotes whether the borrower renegotiated any loans in a determined year. The independent variable of interest, CDS Trading, is a binary variable that equals 1 for firms with XR (non-restructuring) CDS contracts outstand-ing and 0 otherwise. In order to have clearly interpretable results, the table shows the effects of CDS on the probability of loan renegotiations in terms of odds ratios. Mathematically, the odds ratio is obtained dividing the probability that a CDS firm will undergo a loan renegotiation by the probability that a non-CDS firm will restructure a loan.

Loan renegotiation Odds Ratio = Probability CDS Firm has loan renegotiation Probability non-CDS Firm has loan renegotiation

For example, a loan renegotiation odds ratio equal to 0.9 means that CDS firms are 10% lesser likely to renegotiate a loan. If the odds ratio is not significantly different from one, the hypothesis that the presence of CDS affects firm loan renegotiations cannot be confirmed with sufficient certainty. The first three columns of table 3 reports the results of the logistic regression with panel data on the complete sample of US firms. The odds ratio of interest is the one related to the variable CDS Trading. In order to confirm the Empty Creditor hypothesis, the odds ratio is expected to be lower than one. The first column shows results of a logistic regression conducted without any control variable. In this case, despite the significance of the CDS Trading odds ratio, the results are surely subject to omitted variable bias. The second column includes various firm-level control variables: firm size, leverage, profitability, credit rating, tangibility and Tobin's Q. The odds ratio

(25)

of CDS Trading is 0.90, which means that firms with outstanding CDS contracts are 10% less likely to renegotiate a loan. This result is economically significant and is consistent with previous considerations; however, from a statistical point of view it is not significant. Moreover, all the control variables have the expected sign for their odds ratios. The third column reports the logistic model with both firm-level and a macro-level control variables and includes controls for year fixed effects and industry fixed effects. Controlling for year fixed effects and industry fixed effects does not vary the CDS Trading odds ratio (0.89), which remains statistically insignificant.

The same logistic regression is ran on a smaller subsample in order to test the second hypothesis of this research, which states that for firms already in distress the Empty Creditor problem is more prominent. Specifically, the sample was divided into two subgroups, one including firms already in distress (with an Altman Z-score<1.81) and the other containing so-called safe firms (Z-score>2.99). The results are shown in the right columns of table 3. The CDS Trading odds ratios are lower than one; with distressed CDS firms that are 11% less likely to have loan renegotiations compared with non-distressed CDS firms. Although the odds ratios are again statistically insignificant, they are coherent and similar in terms of value across the whole table. In fact, all the odds ratios in the different subgroups gravitate around the same numbers, for example ranging from 0.80 to 0.93 for CDS Trading, and providing favorable signs in support of the stated hypothesis.

Nevertheless, the lack of significant evidence linking CDS to the likelihood of loan renegotiations in the period between 2004 and 2013 cannot confirm the hypotheses presented in this thesis. Hence, despite the probable existence of the Empty Creditor problem in some cases, this seems not to be a regular occurrence for the generality of companies in the sample under observation. This part of the present work is in line with Bedendo et al. (2012), who did not find evidence of the Empty Creditor problem. As the authors suggest, the limited presence of Empty Creditors could be due to two potential explanations. First, the proportion of over-insured bondholders, who would benefit the most from the bankruptcy of borrowers, may not be sufficiently large to affect the loan renegotiations process. Second, even though the proportion of insured creditors is large enough, creditors may choose to support it. This behavior could depend on different factors, such as reputational issues for the creditors, or the insecurity related to the bankruptcy process (Bedendo et al., 2012).

(26)

T able 3: Impact of outstanding CDS con tract s on the lik eliho o d of loan renegotiations This table lo oks at the CDS impact on the lik eliho o d of loan renegotiations for three differen t grou ps of firms: the complete sample (column 1-3) , only Distressed firms (column 4-5), and only n on-Distres sed firms (column 6-7) .The re gr e ssion is form ulated in section 3.2. All the con trol v ariables are defined in App endix B. The regressions use ann ual data from 2004 to 2013. The sample con tains all the firms resulting from the selection pro cedure of section 4. 1 . Con trol v ariables are winsorized at the 1-99% lev el. The z-statistics are presen ted b et w een paren th e ses. Constan ts w ere included in the regressions but are not rep orted. *, **, and *** indicate significance at 10%, 5%, and 1%, resp ectiv ely . T yp e of firm: Complete sample Distressed Non-Distressed (1) (2) (3) (1) (2) (1) (2) CDS T rading 1.26*** 0.90 0.89 0. 80 0.81 0.93 0.91 (2.83) (-0.97) (-1.07) (-1.17) (-1.04) (-0.41) (-0.52) Log(Mark et Capitalization) 1.10*** 1.13*** 1.23*** 1.26*** 1.11* 1.13** (2.73) (3.38) (3.39) (3.65) (1.80) (2.01) Lev erage 1.52** 1.54** 1.26 1.19 2.82*** 3.31*** (2.04) (2.00) (0.53) (0.39) (3.14) (3.36) Profitabilit y 0.47 0.34 0.08*** 0.15* 0.25 0.18 (-1.32) (-1.84) (-2.45) (-1. 78) (-1.43) (-1.72) Rating 1.02 1.03 1. 03 1.03 1.04 1.04 (1.14) (1.56) (1.03) (0.96) (1.22) (1.30) T angibili ty 1.15 1.08 1.12 1.11 0. 98 0.96 (1.57) (0.75) (0.72) (0.22) (-0.13) (-0.24) T obin ' s Q 0.90* 0.88* 0.72 0.73 0.95 0.95 (-1.70) (-1.90) (-1.49) (-1. 39) (-0.63) (-0.58) T ermslop e 0.67*** 0.83 0.69** (-3.50) (-0.89) (-2.09) Industry Fixed Effec ts No No Y es No Y es No Y es Time Fixed Effects No No Y es No Y es No Y es Num b er of Ob serv ations 12539 11825 16636 3221 3172 5696 5599 P-v alue Chi-square statistic 0.005 0.000 0.000 0.000 0.000 0.000 0.000

(27)

In order to obtain a greater precision in the outcome, an additional analysis with the purpose of highlighting the time-varying effect of CDS on the probability of loan renegotiations has been conducted. The resulting table 4 decomposes the effect across years 2004 to 2013. Looking at the loan renegotiation odds ratios in the table, there is no evidence in favor of the Empty Creditor problem in the later years of the sample. However, there is a lower than one (0.62) and significant odds ratio for CDS in 2004, indicating a lower probability of loan renegotiation for CDS firms relative to non-CDS firms for that year. Hence, between the many non-significant results in this part of the study, for this year and for this specific sample, the Empty Creditor hypothesis is effective. This study could be continued by fractioning time into longer periods of two or three years, as it is quite plausible that the Empty Creditor problem manifests itself gradually over longer periods (Peristiani and Savino, 2011).

Table 4: Impact of outstanding CDS contracts on the probability of loan renegotiations

The odds ratio estimates were derived from the logistic regression presented in the third column of table 3, with the addition of the interaction variable CDS Trading × year. A loan renegotiation odds ratio equal to 1.5 indicates that the CDS firm has a 50% greater chance of undergoing a loan renegotiation. The symbols ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.

95% Odds Ratio Confidence Limits Loan renegotiation

odds ratio Lower Upper CDS Trading × 2004 0.62* 0.38 1.03 CDS Trading × 2005 0.75 0.48 1.18 CDS Trading × 2006 0.84 0.54 1.32 CDS Trading × 2007 1.13 0.69 1.85 CDS Trading × 2008 0.75 0.47 1.21 CDS Trading × 2009 1.15 0.66 2.02 CDS Trading × 2010 1.07 0.50 2.28 CDS Trading × 2011 1.15 0.65 2.03 CDS Trading × 2012 0.70 0.42 1.18 CDS Trading × 2013 1.45 0.82 2.57

(28)

A possible concern that could lead to the non-significance of the odds ratios is sample selection bias. Only firms that had an outstanding syndicated loan in a determined year are included in the sample. Although this can provide uniformity in term of size between CDS and non-CDS firms, the exclusion of other types of loans could limit the heterogeneity and the size of the sample. Therefore, a solution would be to use a wider sample, including also different kind of loans. Moreover, another concern is a possible omitted variable bias. As explained by Mengle (2009), many factors influence the likelihood of restructuring before considering CDS: the author names, among others, the amount of pension liabilities, burdensome labor contracts, complexity of capital structure and different kinds of legal obligations. It would be useful to control for all the aforementioned factors, as they could affect the regression results. However, such analysis would require extensive data collection beyond the scope of this thesis. In addition, Subrahmanyam et al. (2011) note that the effects of CDS trading may be less pronounced for firms with thin or illiquid CDS trading. In order to test this hypothesis, Subrahmanyam et al. (2011) include the CDS Notional Outstanding/Total Debt ratio, a measure that should be included for a deeper understanding of the Empty Creditor problem.

5.2 Regressions on the loan renegotiation terms index

A further step is to check the impact of CDS on a different dependent variable, which can confirm or contradict the previous analysis. The effect of CDS on loan renegotiations can be analyzed through a similar method, where a regression is performed on Amendment Index. This variable is a simple index that determines the loan renegotiation terms favorability for borrowers, with values ranging from -4 to 4, and positive values indicating that renegotiation is overall advantageous for the debtor. If the coefficient of the variable were significantly negative, it would indicate that the terms of loan renegotiations are less favorable for firms with outstanding CDS contracts. In turn, this would suggest the presence of Empty Creditors.

Table 5 has the same format as table 3, since, although the dependent variable is different and the regression is linear, the other variables chosen are the same. Considering panel data ran on the complete sample of US firms, focusing on the regression with both industry and time fixed ef-fects; for a firm with outstanding CDS, the mean amendment index is lower than that of non-CDS firms by about 0.47. This result is statistically significant at the 1% level. Moreover, when the

(29)

dataset is split between distressed and non-distressed firms, the CDS coefficient for distressed firms decreases further to -0.61, whereas the coefficient for non-distressed firms becomes non-significant. This means that for a distressed firm with outstanding CDS, the mean amendment index is lower than that of non-CDS firms by about 0.61. This result is significant at the 5% level. The table also includes the p-value of the Wald chi-square for each regressions, which indicates the goodness of fit of the overall model. Nevertheless, although the results are statistically significant, the regression could still be subject to sample section bias and omitted variable bias, as stated in the previous paragraph.

The decrease in the index caused by the presence of CDS is important, and suggests the existence of the Empty Creditor problem. In fact, as loan renegotiations terms are particularly unfavorable to borrowers with CDS, especially for firms already in distress, it seems like creditors prefer to im-pose conditions that can push debtors into bankruptcy. These considerations fit in the theoretical framework provided in section 2. As may be inferred from the previously described different posi-tions in the literature with respect to the Empty Creditor, these results are in line with the theory delineated by Hu and Black (2008). This thesis confirms the theoretical and empirical works of Bolton and Oehmke (2011), Danis (2012) and Subrahmanyam et al. (2011), even though a different sample and different regression methods were used to answer a similar research question. Hence, the outcome of this part of the present work does not coincide with the results of Bedendo et al. (2012), who did not find any evidence of the Empty Creditor problem.

The examination of the impact of CDS on the loan renegotiations terms is a new approach to the Empty Creditor hypothesis that, instead of looking simply at the rate of bankruptcies or loan renegotiations, digs deeper into the behavior of creditors. Creditors are less prone to help borrow-ers avoid bankruptcy, especially if those debtors are already in distress. This could be a serious distortion of the regular debtor-creditor relationship with negative implications for the overall econ-omy. That is the reason why many financial observers ask for a more complex regulatory system concerning CDS derivatives.

(30)

T able 5: Impact of outstanding CDS con tract s on loan rene got iation terms This table lo oks at the CDS impact on loan renegotiations terms for three di ff eren t groups of firms: th e complete sample (column 1-2), only Distressed firms (column 3-4), and only n on-Distres sed firms (column 5-6). The fa v orabili ty of loan renegotiation terms is defined through the Amendmen t Index, describ ed in se ction 3.1. The regression is form ulated in sec ti on 3.2. All the con trol v ariables are defined in App endix B. The regressions u se ann ual data from 2004 to 2013. The sample con tains all the firms resulting from the selection pro cedure of section 4.1. Con trol v ariables are winsorized at the 1-99% le v el. The z-statistics are presen ted b et w een paren theses. Constan ts w ere in c lu ded in the re gressions but are not rep orted. *, **, and *** indicate significance at 10%, 5%, and 1%, resp ectiv ely . Dep enden t v ariable: Amendmen t Index Complete Sample Distressed firms Non -d is tre ssed firms (1) (2) (1) (2) (1) (2) CDS T rading -0.58*** -0.47*** -0.69** -0.61** -0.37 -0.32 (-3.67) (-3.08) (-2.51) (-2.30) (-1.49) (-1.27) Log(Mark et Capitalization) 0.14*** 0.09* 0.20*** 0.16* 0. 13* 0.06 (2.98) (1.80) (2.57) (1.91) (1. 66) (0.74) Lev erage -0. 48 -0.72** 0.08 -0.15 -0.15 -0.47 (-1.62) (-2.48) (0.17) (-0.31) (-0.32) (-0.95) Profitabilit y 3.43*** 1.89** 1.99 0.47 3.49*** 1.51 (4.61) (2.53) (1.37) (0.30) (3. 13) (1.31) Rating -0.02 -0.01 -0.04 -0.01 0.02 0.02 (-0.94) (-0.59) (-1.03) (-0.36) (0.47) (0.54) T angibili ty 0.08 0. 11 0.26 0.25 -0.30 0.03 (0.59) (0.75) (1.10) (0.87) (-1.27) (0.10) T ermslop e 0.04 -0.20 0.16 (0.27) (-0.73) (0.67) Industry Fixed Effec ts No Y es No Y es No Y es Time Fixed Effects No Y es No Y es No Y es Num b er of Ob serv ations 786 775 239 238 351 345 P-v alue Chi-square statistic 0.000 0.000 0.031 0.000 0.001 0.000

(31)

With an additional regression on the loan renegotiation terms, this thesis also focuses on two specific periods in order to determine whether the Empty Creditor problem is more prominent during years of financial crisis. In particular, the years 2004-2005, showing the highest rates of real GDP growth (Worldbank), were chosen among the period under observation due to their relatively good economic conditions. On the other hand, the years 2008-2009 represent a period of strong recession, also shown by the low or negative real GDP growth rates. Table 6 displays the results of these regressions on the complete sample of firms. It can be noticed that, during a period of recession, a firm with outstanding CDS has a mean amendment index lower than that of non-CDS firms by about 0.71. This result is statistically significant at the 5% level. Regarding the economic stable period, the difference remains negative (-0.37) but statistically non-significant. This could result from the low amount of observations available. Therefore, this supplementary result confirms the previously stated affirmations on the effective existence of the Empty Creditor problem, and adds that economic conditions can amplify its effects.

Table 6: Impact of outstanding CDS contracts on loan renegotiations terms - difference between periods of economic stability and recession

This table looks at the CDS impact on loan renegotiations terms for the complete sample of firms considering two different economic periods: years of stability (2004-2005, column 1), and years of recession (2008-2009, column 2). The favorability of loan renegotiation terms is defined through the Amendment Index, described in section 3.1. All the control variables are defined in Appendix B. The z-statistics are presented between parentheses. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Dependent variable: Amendment Index Economic stability (2004-2005) Recession (2008-2009) CDS Trading -0.37 -0.72** (-1.09) (-2.26)

Control variables Yes Yes

Number of Observations 214 160 P-value Chi-square statistic 0.000 0.001

(32)

6 Robustness Checks

This paragraph reports the robustness checks performed on the analysis. The regressions presented in the previous paragraph are already controlled for time fixed effects and industry fixed effects. In order to control for the latter, the 1-digit SIC (Standard Industrial Classification of economic ac-tivities) variable from Compustat is used. The fist paragraph presents a propensity score matching exercise. The second paragraph presents another important robustness check that consists in the change of the amendment index definition.

6.1 Propensity score matching

Propensity score matching is a useful method to compensate for imbalances in the data and en-dogeneity. As can be seen from Table 2, the two groups of CDS-firms and non-CDS firms do not have substantial overlap in most of the control variables. This, in turn, can bias the regression outcomes. The purpose of propensity score matching is to detect a set of non-CDS firms that share comparable features with the reference entities in the observed sample. The procedure used to create the matched sample is based on the propensity score, defined as the probability for each firm of having CDS traded on its default. The propensity score-matching estimator minimizes the distance between a vector of observed covariates across CDS and non-CDS firms, and selects a corresponding group of control firms based on the matches with the smallest distance (one-to-one matching). In this case, the sample is matched in terms of size, tangibility ratio, and leverage ratio. The sample of the CDS firms and the propensity score-matched non-CDS firms is then compared in order to check whether there are significant differences in the mean covariates. Using the one-to-one matching, the CDS firms are still significantly larger than their matched counterparts. Therefore, a further limitation is introduced, which limits the propensity score difference to less than 1%. The regressions on loan renegotiations likelihood and terms are re-estimated using both the propensity score-matched samples.

Table 7 presents the regression results for the propensity-matched samples. It can be noticed that the results are similar to the ones proposed earlier in this thesis. In fact, the loan rene-gotiation likelihood odds ratio is again statistically non-significant and around 0.94, whereas the

(33)

mean Amendment Index still decreases significantly due to the presence of CDS (-0.59). When the matching criterion is modified to one-to-one matching with propensity score difference within 1%, the results do not show sensible variation.

Table 7: CDS impact on loan renegotiations: propensity score matching

This table presents the results of the regressions on the probability of loan renegotiations (expressed in odds ratios) and the Amendment Index, using a sample including firms with CDS and non-CDS propensity score-matched firms. Columns 2 and 4 present the analysis on the nearest one propensity score-matched firms. Columns 3 and 5 present the same analysis, but on the nearest one firms with a propensity score difference within 1%. All the control variables are defined in Appendix B. The z-statistics are presented between parentheses. Constants were included in the regressions but are not reported. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Probability of loan renegotiations Amendment Index

Year Nearest One

Matching Nearest One PS Diff<1% Nearest One Matching Nearest One PS Diff<1% CDS Trading 0.94 0.97 -0.59*** -0.49** (-0.55) (-0.17) (-3,31) (-2.04) Log(Market Capitalization) 1.06 1.15** 0.08 0.17 (1.21) (2.21) (2.57) (1.50) Leverage 1.37 2.74*** -0.29 -0.13 (1.12) (2.96) (-0.66) (-0.23) Profitability 0.23* 0.18 1.49 1.70 (-1.69) (-1.57) (1.05) (0.92) Rating 1.03 1.02 -0.03 -0.07 (1.38) (0.61) (-0.86) (-1.57) Tangibility 0.95 1.09 0.17 0.26 (-0.41) (0.50) (0.86) (0.93)

Time Fixed Effects Yes Yes Yes Yes

Number of Observations 6886 4046 351 212 P-value Chi-square statistic 0.000 0.000 0.000 0.000

6.2 Redefinition of the loan renegotiation terms index

Two new definitions for the loan renegotitaion terms index are presented in this paragraph. The regressions on these variables could confirm the earlier results and increase their significance. Amendment Index 2 is similar to Amendment Index, but it adds up the fractional change in each loan term and then divides by 4. For example, big positive values mean that markups decrease

The empty creditor problem : the impact of Credit Default Swaps on loan renegotiations