The relationship between market expectations concerning when a firm might default and debt renegotiation frictions for US financial and non-financial institutions

(1)

Master’s Thesis:

The relationship between market expectations concerning when a

firm might default and debt renegotiation frictions for US financial

and non-financial institutions

MSc International Finance

Halim Banjar Perdana

Student number: 11000570

Supervisor: Prof. Tomislav Ladika

University of Amsterdam

September 2017

(2)

Acknowledgements

First and foremost, I would like to thank my thesis supervisor, Professor Ladika, for his great support in helping me to develop the thesis topic. I additionally thank him for his patience and helpful guidance whenever I needed questions answered, including in relation to obtaining missing data when necessary. I would also like to thank my son, Enrique; my wife, Maya; my parents; and my

close friends and colleagues for keeping me motivated and inspired throughout the thesis writing process.

(3)

Abstract

A credit default swap (CDS) is a protection or insurance that bondholders can buy to protect against a reference entity’s potential default. It is one of the many financial

derivative products that is actively traded. In contrast, debt renegotiation is an activity that firms may conduct when they face financial distress and wish to avoid default. This research thus assesses how debt renegotiation frictions correlate with default probabilities (among US financial and non-financial firms) derived from CDS price.

(4)

1.

Introduction

In this research the author’s specifically analyze the relationship between debt renegotiation frictions and CDS spread (bankruptcy probability). Frictions are triggered by several possible factors such as number of outstanding debts. Generally, it is originated from a financial distress. Relevant reasoning exists as to why looking closely examining debt renegotiating frictions would be very interesting. Since the introduction of the CDS Big Bang Protocol by International Swaps and Derivatives Association Inc. (ISDA) in 2009 as summarized by Markit (a global financial information and services company) on one of its publication, CDS contracting has been harmonized nationwide in the US. One key change resulting from this protocol was the removal of debt restructuring as an eligible credit event— which intuitively increases potential frictions (e.g. between lenders and borrowers).

This research finds that firms with more complex capital structures are more likely to default earlier. Further, it is plausible that firms with more complex capital structures that simultaneously faced with more distressed situations are likely to default earlier. How does it differ between US financial and non-financial institutions? This is one of the key questions the author wish to answer. This research also reveals that the Z-score is capable of identifying distress effectively and that the coefficient shows correctly the significant negative relation to default probability for more distressed firms i.e. firms with high chance for financial embarrassments but less for firms with good chance of default.

In addition, a quantitative empirical-based research is used as research methodology. Ordinary Least Square (OLS) regressions are conducted using statistical software: Eviews. A regression model is developed in order to be able to explain the variances of bankruptcy probabilities (obtained from two

(6)

Page 2

and one year CDS price of 237 US firms). In addition, number of outstanding debts is used as proxy of debt renegotiation frictions or complex capital structures (the main research variable). Leverage, return on equity (profitability), book value of total assets (size) and the Altman Z-score (accounting based distress measurement methodology) are used as control variables.

This remainder of this paper is structured as follows. Chapter 2 presents relevant studies on topics such as CDS pricing and debt renegotiation frictions. Thereafter Chapter 3 outlines the study’s hypotheses and methodology, including the development of a base model. Chapter 4 then explains the data and descriptive statistics, identifies the challenges experienced in obtaining accurate data and highlights the study’s potential biases. Finally, Chapter 5 presents the empirical results of the research and Chapter 6 offers conclusions and recommendations.

(7)

Page 3

2.

Literature Review

This chapter starts with explaining the Credit Default Swap (CDS) market followed by presenting relevant studies on topics such as CDS pricing and how probability of default is derived. Then moving on to sections explaining the potential sources of debt renegotiation frictions, determinants of firms’ health and liquidity risk in CDS market.

2.1 The CDS Market

According to the International Organization of Securities Commission (IOSC) (CDS Market Report, 2012), a credit default swap (CDS) contract is a binding bilateral agreement to transfer a reference entity’s credit exposure from a buyer to a seller (i.e. between market participants). The seller receives periodic payments in return, but the CDS contract guarantees a positive pay-off when a credit event is deemed to have occurred. Duffie (1999) explains that a credit event must be officially documented with a notice and announced by the firm (e.g. through the international press).

The above protection provides a bondholder (in this case, the buyer) with a hedging instrument against default risk. However, when such a buyer does not have credit exposure to the underlying entity, he or she is deemed to take a “naked” CDS position. Most CDS contracts are traded “over the counter (OTC)” by G14 dealers or major derivative trading banks (e.g. J.P. Morgan, Morgan Stanley and UBS). When a contract refers to a specific entity, it is called a “single name” CDS; when it concerns more than one reference entity, it is known as an “index” or “basket” CDS. According to data that the IOSC compiled from the BIS and DTCC (i.e. major global repositories), single name CDSs accounted for 60% of US$26,000 billion gross notional value as of the end of 2011.

(8)

Page 4

Many studies have investigated the effect of CDS trading on firms. For example, Danis and Gamba (2016) found evidence that CDS trading slightly increases the likelihood of a firm’s liquidation (from 1.01% to 1.03%), which in effect reduces firm value. However, at the same time CDS trading could also create a better opportunity for firms to enter into cost-efficient debt renegotiation and equity financing. These researchers identified that on average, firm value ultimately experiences a 2.9% net increase after CDS trading is introduced to the market (i.e. 2.67% for large firms and 4% for small firms); higher results were found among financially constrained and lower productivity firms. Various explanations are possible for these findings. Firstly, it may be that bondholders are better protected or hedged with CDS trading and thus ask for a higher pay-off should there be debt renegotiation. This makes costly debt renegotiation a less attractive route for a firm to pursue, and it may instead choose to repay its debt. This is also in line with the findings of the qualitative research undertaken by legal scholars Hu and Black (2008a, b). Secondly, from an equity financing point of view, CDS trading empowers firms to rely less on expensive equity financing due to its positive effect on debt’s market value at the time of issuance. This high market value of debt (which intuitively further increase a firm’s value) enables firms to issue more debt with the same face value (i.e. to increase debt issuance capacity and lower the cost of debt financing). This is especially true for small and financially constrained US public firms.

In fact, when scholars and practitioners discuss the Credit Default Swap (CDS) topic, they also intensively analyze related subjects (e.g. how CDS spread connects with the probability of default and firms performance). Wang (2004) explains that the market’s view of firm credit risk (and of the embedded probability of default) is reflected in the CDS spread or premium paid by a buyer over the duration of the CDS contract. The CDS premium payment stops when the firm is in default. The yield spread over the firm’s debt (i.e. the bond spread) is similar, although the bond spread refers specifically to the yield of a corporate bond minus the yield of a risk-free bond (such as issued by the US Treasury)

(9)

Page 5

with an identical coupon rate and maturity date. Arbitrage trading does exist between the CDS and bond spread.

2.2 CDS Pricing and Probability of Default

CDS pricing can be determined using quantitative formulas. The CDS price is equal to a bond’s spread at inception, and many studies have tried to model it. Merton (1974) introduced the first quantitative CDS pricing model, which is known as the “structured approach.” This model entails building a formula that links credit risk with a firm’s characteristics, including its leverage and assets volatility. A second type of modeling, namely the “reduced form,” was later popularized by Duffie (1999). This formula includes building predefined/assumed parameters that define credit spread (including recovery rate and default probability). This model makes it possible to reflect the historical market price when the current pricing or ultimate importance of the assumed parameters is being determined. Wang (2004, p.20) defines the basic quantitative formula for determining off-the-market (i.e. absence of extrinsic factors such as liquidity) CDS spreads based on the standardized assumption that the present values of the CDS premium payment leg and the default leg (i.e. the bond’s spread) are equal. This can be described as follows:

(1)

Whereby:

S’: annual CDS spread

(10)

Page 6

q(tj)-q(tj-1): default probability in period j B(0,t): discount rate

∆ i-1,i: length of the payment period

1-q(tj): survival probability at time ti

q(ti): cumulative default probability up to time ti

Similarly, Duffie (1999) asserts that the basic quantitative formula for calculating the off-the-market CDS spread per period can be described as follows:

ai (h) = exp{ -[ h + y(i) ] T(i) } (2) bi (h) = exp[ -y(i) T(i) ] { exp[ -h T (i -1)] - exp[ -h T(i) ] } (3)

Whereby, h: hazard rate

ai (h): the present value of the receiving CDS premium payment at the i-th coupon date, assuming default takes place after this date.

bi (h): the present value of the bond’s receiving coupon payment at the i-th coupon date, assuming default between the (i - 1)th and i-th coupon dates.

T(i): time to maturity of the i-th coupon date.

(11)

Page 7

As such, spread (S’) equals the sum of bi (h) multiplied by loss given default (1-recovery rate) and then divided by the sum of ai (h). If four premium and coupon payments exist per year, S’ must be multiplied by four to arrive at the annualized CDS spread.

The default probabilities of a bond can also be obtained from on-the-market CDSs instead of using the historical default probabilities as per information from agencies such as S&P. In this case, the default probabilities are calibrated with the prevailing market situation at that time.

Hence, the default probabilities which are more widely referred to as hazard rates in the literature, are part of a triangulation that also includes the loss given default (i.e. 1-recovery rate) and the spread/premium.

Hazard rate = premium/(1-recovery rate) (4)

The standard market practice here is to use a recovery rate of 40%. The hazard rates can be identified if the last two parameters are known. In general, the above quantitative formulas are applicable when the hazard rate is constant.

2.3 Debt Renegotiation Frictions and Financial Distress

The potential causes of debt renegotiation frictions can intuitively be triggered by:

1. The higher number of creditors. In some instances, a syndicate of banks with a lead underwriter provides a committed credit line—which can make negotiation even more complex;

2. The higher number of outstanding debts; and 3. The longer duration of outstanding debts.

(12)

Page 8

In fact, the needs of debt renegotiation itself can originate when firms are in distress. According to Asquith et al (1994), when firms are in distress firms will seek solution such as by conducting activities such as: bank debt restructurings, public debt restructurings, asset sales and cutting capital expenditure. Most empirical research finds that the main source of financial distress is typically generated by high leverage. This financial distress could potentially result in a firm having difficulties in meeting its financial obligations, as Andrade and Kaplan (1998) confirm. Other factors that cause financial distress (e.g. poor firm or industry performance) contribute to a lesser degree. Andrade and Kaplan (1988) further conclude that the cost of financial distress equals 10–20% of firm value for the acquiring firm. This is due to a firm’s 1) inability to sustain capital expenditure activities, 2) necessity to perform a asset fire sale and 3) decision to delay restructuring or filing for Chapter 11 (i.e. the US bankruptcy protection code that allows firms to reorganize/restructure their business)—which appears on the surface as costly. Nevertheless, these reasons do not apply to firms that are not experiencing adverse economic shocks. In this context, it is helpful to note that these researchers define firms in distress as having the following characteristics:

1. An EBITDA/interest expense ratio < 1;

2. An announcement of a debt restructuring attempt that is due to an inability to meet debt obligations;

3. Defaulted debt; and

4. A completed Chapter 11 filing.

This definition is somehow more “practical” for determining whether firms are healthy or in distress than the accounting approach introduced by Altman (1968). The latter approach, which is known as the Altman Z-score, is described in section 2.5.

(13)

Page 9

It is important to note at this point that to the best of the author’s knowledge no specific literature tries to analyze the relationship between a market’s expectations of when a firm might default (i.e. the probability of default) and debt renegotiation frictions. In this research, the number of outstanding debts is used as proxy of debt renegotiation frictions. An index or basket type of CDS is not included in the current analysis.

2.4 Anatomy of Debt Renegotiation Frictions

As explained in 2.4 there are two type of debt restructuring: bank debt (private) and public debt restructuring. Firstly, as for the bank debt restructuring according to Asquith et al (1994) this involves “tightens the screw” activities: meaning by increasing of collateral requirement and decreasing lines of credit; and “loosens the screw” activities: meaning by supporting with additional funding, decreasing interest rate, postponing interest and principal payments and waiving covenants. Theoretically, according to them banks appetite to either tightens or loosens the screw in general depends on whether other debt-holders may reap the benefits (i.e. debt overhang) or whether the firms are in better position from economic and credit risk point of view in general. According to one of their findings banks tend to loosens when they are protected in case of bankruptcy.

Secondly, the public debt renegotiation process differs compared to the prior. It involves exchange offers vs. direct negotiation. The exchange offers provide debt holders opportunity to exchange old debts for new securities. Thus, it is obvious from these explanations that the higher the number of outstanding debts will result in higher renegotiation activities (hence frictions). To highlight an example of possible complication, Gilson et al (1990) gives an explanation of firms which decided to renegotiate debts outside of Chapter 11 (US bankruptcy procedure). The debt renegotiation can easily breaks if one of the creditors (depending on his/ her stake) are unhappy with the debt renegotiation

(14)

Page 10

offers. Debt renegotiation outside of Chapter 11 generally requires consent from all creditors holding defaulted securities.

2.5 The Altman Z-score

As proposed by Altman (1968) and explained by Hull (2015), the following key accounting ratios can be used to identify whether firms are healthy or in distress:

Ratio I: EBIT (operating income)1_{/total assets (5)}

Ratio II: Net sales/total assets (6)

Ratio III: Market value of equities/total liabilities (7)

Ratio IV: Working capital/total assets (8)

Ratio V: Retained earnings/total assets (9)

Using these ratios, an Altman Z-score can be calculated based on the following equation: Altman Z-score = 3.3*Ratio I + 0.999*Ratio II + 0.6*Ratio III +1.2*Ratio IV + 1.4* Ratio V (10)

When the result has been obtained, a firm’s healthiness level can be further explained as follows:

 A score > 3 means that the firm is “unlikely to default”;

 A score between 2.7 and 3.0 points to a reason to be “on alert” in relation to the firm’s health;

 A score between 1.8 and 2.7 indicates a “good chance of default”; and

 A score < 1.8 signifies a “high chance of financial embarrassment.”

1_{As Compustat North America does not provide EBIT financial figure’s quarterly data. Thus, Operating Income}

(15)

Page 11

The current research entails analyzing more and less distressed firms; Chapter 3 (Hypotheses and Methodology) provides more details. It should be noted here that due to its robustness, the Altman Z-score (as opposed to the “practical” approach) is used to differentiate between more and less distressed firms.

2.6 Liquidity Risk in the CDS Market

It is presumable that liquidity affects CDS prices, just as it affects the prices of all other asset and derivative products traded in the market. According to Junge and Trolle (2015), CDS market illiquidity was relatively stable around 2–3 bps or 0.01–0.02% at the end of 2006 (i.e. before the 2008 global financial crisis) before steadily increasing to 2–9 bps during Bear Stern credit events (namely the collapse of two Bear Stern- structured credit hedge funds in June 2007 and Bear Stern’s near bankruptcy in March 2008). It further significantly increased to 7–9 bps during the default of Lehman Brothers’ AIG bailout in September 2008. As of January 2010, it ranged from 3 to 10 bps. Liquidity risk is thus included in the quoted/traded CDS price: the higher the illiquidity, the higher the spread that the buyer must pay. Junge and Trolle (2015) also found that the excess return differential of CDS sellers between selling high credit quality (and high liquidity) and low credit quality (and low liquidity) equaled 5.45%. Of this percentage, 3.3% is attributed (annually) to the ordinary default risk, -0.06% is due to a pricing error and 2.21% actually results from liquidity risk.

Despite the above, liquidity risk was ignored in the scope of the current research. This is because the dependent (or Y) variable used is defined based on the price differences (hence bankruptcy probability) of two- and one-year CDSs. Presumably, liquidity risk should affect two- and one-year CDS prices on a similar scale in the event of idiosyncratic firms or market event. This is explained in more detail in Section 3.2 (Methodology).

(16)

Page 12

3.

Research Question, Hypotheses and

Methodology

This chapter explains the central research question which is used as a base to generate the main and additional hypotheses. Hence it further includes explanations of how the current findings connect to previous research when applicable. Followed by a section explaining the research methodology including the developed model.

3.1 Central Research Question and Hypotheses

The central question that this thesis strives to answer is “What expectations does a market have regarding when a firm might default related to renegotiation frictions?” This question is reflected in

Hypothesis 1: Firms with more complex capital structures are likely to default earlier than firms with less complex capital structures, as such firms may have less flexibility to re-negotiate

with creditors to delay default. The result is expected to be significant, especially after the CDS Big Bang Protocol was introduced (as explained earlier).

The subsequent sections of this thesis focus on testing the following hypotheses:

 Hypothesis 2: Firms with more complex capital structures are likely to default earlier

than firms with less complex capital structures and less distress. In this context,

distressed firms can be identified using the above-discussed combined accounting base performance ratio (i.e. the Altman z-score). It is assumed the lower the Z-score, the higher the probability that a firm defaults earlier. (NB: The author’s initial intention was to analyze how

(17)

Page 13

the results may differ before and after the 2008 crisis, to identify whether pre-2008 crisis firms that had complex capital structures and were in distress were more likely to default more easily. A possible assumption here is that pre-crisis lenders had more flexible lending approval criteria until the 2008 crisis and then became more risk averse. However, a lack of CDS price data meant that this could not be included in the scope of this thesis.)

 Hypothesis 3: Financial institutions with more complex capital structures and in distress are likely to default earlier than non-financial institutions with more complex capital structures and in distress. Thus, how the result differs between financial and

non-financial institution? This is quite interesting since non-financial institutions have been blamed for triggering the largest financial crisis in recent history, which is seen as starting in the summer of 2007. Financial institutions are widely blamed for providing mortgages and lending too easily, as well as for selling complex derivative products irresponsibly (e.g. mortgage-backed securities and collateralized debt obligations). It is necessary to clarify whether financial institutions have a relationship between default probability and debt renegotiation friction that deviates in comparison to the relationship that non-financial institutions enjoy. Especially, when banks and other types of financial institutions are subject to additional regulatory scrutiny post-2008 crisis (e.g. the requirement to comply with regulations such as Basel III: a global regulatory framework on market liquidity risk, bank capital adequacy and stress testing; and MIFID: Markets in Financial Instruments Directive) thus presumably less sensitive to frictions or stronger protected.

 Hypothesis 4: Larger firms with more complex capital structures are likely to default

earlier than smaller firms with more complex capital structures. It can be assumed that

(18)

Page 14

As indicated earlier, Danis and Gamba (2016) found evidence that CDS trading slightly increase the likelihood of a firm’s liquidation.

3.2 Methodology

A quantitative empirical-based approach is used for this research and regressions are conducted using statistical software: Eviews. Based on the previously explained central question (hypothesis 1), the following ordinary least square (OLS) regression model is defined as follows (and used as a base for the further models development for hypothesis 2, 3 and 4):

Y = a + b*friction + c*profitability + d*leverage + e*size +f*zscore + e (11)

In this model, “friction” represents capital structure complexity and the number of outstanding debts is used as the proxy. A higher number of outstanding debts presumably increases potential friction over time in relation to potential debt renegotiation. “Profitability” can be represented by the return on equity (RoE) ratio, which can be derived from: (net income/assets) * (assets/equity). “Leverage” is derived from the leverage ratio: debt/(debt + book value of shareholders’ equity). The inclusion of leverage, profitability and Z-score are important in the model as Asquith et al (1994) and Andrade and Kaplan (1998) mentioned industry downturn, poor firms’ performance and high interest expenses as possible origins of financial distress.

Furthermore, a firm’s “size” can be identified from its book value of total assets; it can alternatively be identified from the market value of its total equities. Firms’ “Z-scores” (which are explained in Section 2.3) are further divided into three of the aforementioned categories for usage in one of the regressions (the category of healthy firms with a Z-score > 3 is excluded). Moreover, profitability, leverage, size and z-score act as controls and along with friction these are the independent or explanatory variables. Empirical

(19)

Page 15

On the other side of the equation, the dependent variable refers to the bankruptcy probability, whereby Y = 1 if bankruptcy probability in year 2 is higher than in year 1 and Y = 0 if the opposite is true (i.e. dummy variables are used to identify if bankruptcy probability increases over time). The bankruptcy probability can be obtained from the reduced form model or Duffie’s (1999) basic quantitative formula (i.e. equations 2 and 3 above), as explained earlier. An F/Wald test is used for any joint hypotheses (e.g. β2 = β5 = 0) and a significance level of 5% is used through-out all the regressions. A test significance level below 5% meaning an alternative hypothesis can be accepted.

(20)

Page 16

4. Data and Descriptive Statistics

This chapter explains the data sources and gathering process, details of the observation period and descriptive statistics. It also identifies the challenges experienced in obtaining accurate data and highlights the study’s potential biases.

4.1 Data

All of the monthly CDS prices (dated 2009 through 2015) used to produce the dependent variable were obtained from the Thompson Reuters Datastream. In the case of a firm issuing a bond through several group entities, the ultimate parent entity’s monthly CDS price was used. Further, from visual validation of the data only monthly CDS prices for years one and two were consistently available for the entire observation period; as such, the monthly CDS prices for years three, four and five were excluded from the analysis. Any firm that has monthly CDS prices missing from the observation period was also excluded. Chapter 5 provides a more detailed explanation of these issues.

On the independent variable side, all financial data to produce the quarterly information on return on equity, leverage, company size and Altman Z-score was sourced from the Wharton Research Data Services (WRDS) Compustat database. The quarterly number of outstanding debts, which previously explained as the proxy of debt renegotiation frictions used in this research, were obtained from S&P Capital IQ. This includes 33 categories of debt, including bank loans and market debts (e.g. term loan, revolver loan, corporate MTN and asset-backed securities; Appendix Icontains a complete list). It is notable that as described by Pagano (2005), the modigliani-miller theorem dictates that the amount, choice and structure of a firm’s debt do not affect that firm’s enterprise value in the following scenario:

(21)

Page 17

1) an absence of taxes, 2) a lack of bankruptcy costs (including reputational costs for its directors) and 3) financial markets that are competitive, frictionless and free from any informational asymmetry. Further, according to Crouzet (2016), a firm can in practice adjust its funding mix of bank loans and market debts depending on its appetite and capacity. In this context, bank debts offer more flexibility in terms of covenant and renegotiation possibilities—especially in distress situations; however, they are generally more expensive than market debts. In contrast, market debts hardly offer flexibility in relation to renegotiation. Firms can gradually opt to use more markets debts as their credit ratings increase.

Hence, it is notable that the number of outstanding debts may reflect a certain degree of bias (difficult to quantify). This is because an identified limitation during data sourcing was the exclusion of debts that were recalled/terminated before their original expiration dates. Lastly, financial and non-financial institutions are differentiated based on their Standard Industrial Classification (SIC) code, as available in Compustat.

4.2 Descriptive Statistics

Descriptive statistics were calculated for the data once it was collected. An overview of these statistics by variable is presented in Table 1 below:

(22)

Page 18

In relation to data validation, the first step was to correct for any outliers and extreme values by replacing them with the corresponding upper or lower quantile. This is especially true in relation to the profitability (RoE) and leverage ratios. Regarding friction (or the number of outstanding debts) and size, it is assumed that the extreme values are valid under the central limit theorem.

As proxy of friction, the highest number of outstanding debts among the observed companies is 30,826; the lowest is zero. The higher numbers are typically seen in relation to financial institutions such as banks, as lending is part of their primary business. As such, it is not appropriate to correct for extreme values. Moreover, the extreme values relate to the fact that the non-financial institutions greatly outnumber the financial institutions, which is also demonstrated by the low median number of 19. Similarly, concerning the size variable the highest asset book value is US$2.6 billion and the lowest is US$536,000. However, while many small and medium-size firms exist, only a handful of large firms can be found. The high presence of small and medium-size firms is identified by the low median number at US$14.4 million. Again, correcting extreme values are not appropriate for this variable. Thus, the high positive skew in the two variables can be accepted.

(23)

Page 19

5. Robustness Check and Results

This chapter starts with an explanation of the robustness check concluded in order to obtain reliable results. It follows by explanations of the empirical results from the research. Each hypothesis is analyzed individually.

5.1 Robustness Check

In fact, within the observation period of January 2009 through December 2015, the number of firms with complete monthly data ranges on average annually from 290 to 335 firms (290 in 2009, 305 in 2010, 325 in 2011, 335 in 2012, 335 in 2013, 330 in 2014 and 320 in 2015). As an equal number of monthly observations for each firm did not exist over the period, a cross-sectional OLS pooled regression was run on the panel data. However, although the results are significant for the independent variable (except for RoE), the results cannot be accepted since it cannot be assumed that all firms are the same (i.e. the firms’ individuality and heterogeneity cannot be denied). As strong autocorrelations were also found, thus should it proceed with such regression approach it is plausible that the results are unreliable. An obvious solution would be to convert all variables into log(). However, this does not work with negative numbers such as on RoE (i.e. profitability), leverage and the Z-score.

To address the above situation, the cross-sectional panel data needs to be properly dated (i.e. transformed to panel data with a time series). Moreover, it should only include firms that have complete monthly data for the entire observation period (i.e. 84 observations, representing 12 months x 7 years). A total of 237 firms were ultimately included in the research. It was only at this point that it could be determined whether it was best to accept the fixed effect (i.e. allowing change in intercept value but intercept is time invariant or random effect (i.e. with common mean value of intercept) model.

(24)

Page 20

5.2 Results

In this section, regression results from hypothesis testing are presented and analyzed. The results significance is explained and the validity of previous research is commented when applicable. As earlier indicated, each hypothesis is analyzed individually.

Hypothesis 1: Firms with more complex capital structures are likely to default earlier than firms with less complex capital structures

Using the base model of Y = a + b*friction + c*profitability + d*leverage + e*size +f*zscore + e, several regressions were run to determine the best linear unbiased estimator (BLUE); the results are shown in Figure 1 below. Firstly, when Y is equal to the absolute difference between the bankruptcy probabilities of years 2 and 1, it can be concluded that the random effect model is appropriate. Under the Hausmen test, the result is not significant (i.e. it is below the 5% significance level, which is the level of significance used for all regression tests hereafter). As such, the null hypothesis (namely that the random effect model is appropriate) has a probability of 61.18% and cannot be rejected. According to the concluded random effect model, the proxy of debt renegotiation frictions variable shows a probability of 72.64%. The alternative hypothesis that firms with more complex capital structures are likely to default earlier than firms with less complex capital structures therefore cannot be accepted. The adjusted R-squared is 0.00898.

(25)

Page 21

Secondly, when Y uses a dummy variable it can be concluded that the fixed effect model is appropriate. As the result is significant according to the Hausmen test, the null hypothesis (i.e. the random effect model is appropriate) can be rejected and the alternative hypothesis (i.e. the fixed effect model is appropriate) can be accepted at a probability of 0.00%; Figure 2 provides further details. Furthermore, under the concluded fixed effect model proxy of debt renegotiation, the frictions variable shows a probability of 0.00% or statistically significant (i.e. p-value < 0.05). As such, the alternative hypothesis that firms with more complex capital structure are likely to default earlier than firms with less complex capital structures can be accepted. The adjusted R-squared improves to 0.282493.

(26)

Page 22

The acceptance of the alternative hypothesis is in line with expectations, especially due to the introduction of CDS Big Bang Protocol in the US it is plausible that debt renegotiation frictions have increased to a new standard. The coefficients also show that the correct relation (e.g. when friction increases, bankruptcy probability increases); the exception is the Z-score, which is supposed to be negative (i.e. a higher Z-score or healthier firm should lead to a lower bankruptcy probability over time). The later variable may be endogenous. It would be interesting for future research to explore how the relationship differs prior to the introduction of the CDS Big Bang Protocol.

(27)

Page 23

It thus seems that using the dummy variable changes the results significantly. It is assumed that the monthly unexplained CDS price variance differs across the observed firms and that the dummy variable is somehow able to control or minimize it.

Last but not least, it is important to note based on the coefficient of friction we can interpret that every one unit increase in friction (i.e. number of outstanding debts) leads to a 0.0000675 unit increase in the Y variable. This is a small effect from economic stand point because the mean value of the Y variable is 0.951728.

Hypothesis 2: Firms with more complex capital structures are likely to default earlier than firms with less complex capital structures and less distress

The following slightly adjusted model is used to prove hypothesis 2: Y = a + b*friction + c*profitability

+ d*leverage + e*high chance for financial embarrassment + f*good chance of default + g*on alert + e. As shown in

Figure 3, the last three variables are dummy variables: “1” signifies that a firm z-score falls into one of the three specified categories (i.e. < 1.8; > 1.8 and < 2.7; or > 2.7 <3) whereas “0” signifies that it does not. According to the expectations, it is plausible that firms with more complex capital structures that simultaneously faced with more distressed situations are likely to default earlier. The F/Wald-test (β1 = β2 = β3 = β4 = β5 = β6 = β7) is showing a probability of 0.00% or statistically significant with p-value <0.05. The results also reveal that the Z-score is capable of identifying distress effectively and that the coefficient shows correctly the significant negative relation to default probability for more distressed firms i.e. firms with high chance for financial embarrassments but less for firms with good chance of default.

(28)

Page 24

Hypothesis 3: Financial institutions with more complex capital structures and in distress are likely to default earlier than non-financial institutions with more complex capital structures and in distress

To test this hypothesis, two separate regressions were run using the model from the hypothesis 2: Y

= a + b*friction + c*profitability + d*leverage + e*high chance for financial embarrassment + f*good chance of default + g*on alert + e. One for financial institutions and another for non-financial institutions. Figure

(29)

Page 25

Frictions thus indeed have a significant impact on financial firms post-crisis, which means that banks are still not better protected despite being under regulatory scrutiny than non-financial institutions. The coefficient is unexpectedly close to 14 times higher for financial institutions than non-financial firms (0.001315/0.000094 = 13.99). Looking at financial institution specifically (i.e. firms with the higher friction coefficient), the coefficient can be interpreted that every one unit increase in friction (i.e. number of outstanding debts) leads to a 0.001315 unit increase in the Y variable. This is a small effect from economic stand point because the mean value of the Y variable is 0.978291.

(30)

Page 26

In addition, it is noteworthy that the regression for US financial firms is potentially exposed to bias, given that only 17 firms were included. It would be very interesting for future research to analyze how the coefficient differs pre-crisis.

Hypothesis 4: Larger firms with more complex capital structures are likely to default earlier than smaller firms with more complex capital structures

Finally, hypothesis 4 leverages the following slightly adjusted model: Y = a + b*friction + c*profitability

+ d*leverage + e*size + e. The results are presented in Figure 5 below.

The coefficient of size reveals that this variable has a positive relation to bankruptcy probability. As such, size does matter and larger firms with complex capital structures have a greater probability of bankruptcy. This is in line with the findings of Danis and Gamba (2016). In which as explained, larger firms have greater access to debt market larger traded CDS market share.

(31)

Page 27

6. Conclusion and Recommendation

This research finds that firms with more complex capital structures are more likely to default earlier. This is in line with the author’s expectation, especially due to the 2009 introduction of the CDS Big Bang Protocol in the US it is plausible that debt renegotiation frictions have increased to a new standard. Although from economic stand point the effect is small. Hence, it is plausible that firms with more complex capital structures that simultaneously faced with more distressed situations are likely to default earlier. The results are even more pronounce (14 times larger) for financial institutions. It would be very interesting for future research to analyze how the relations and coefficients differs pre-crisis.

The results also reveal that the Z-score is capable of identifying distress effectively and that the coefficient shows correctly the significant negative relation to default probability for more distressed firms i.e. firms with high chance for financial embarrassments but less for firms with good chance of default.

Finally, the findings indicate that size has a positive relation to bankruptcy probability and it is statistically significant. This demonstrates that size does matter and that larger firms with complex capital structures have a greater probability of bankruptcy, which is in line with the findings of Danis and Gamba (2016).

(32)

Page 28

References

Altman, E.I. (1968), Financial Ratios, Discriminant Analysis, and the Prediction of Corporate Bankruptcy, Journal of Finance 23, 589-609.

Andrade, G. and Kaplan, S. N. (1998), How Costly is Financial (Not Economic) Distress? Evidence from Highly Leveraged Transactions that Became Distressed, Journal of Finance.

Asquith, P., Gertner, R. and Scharfstein, D. (1994), Anatomy of Financial Distress an Examination of Junk-bond Issuers, The Quarterly Journal of Economics 3.

Aurora, N., Bohn, J.R, Zhu F. (2005), Reduced Form vs. Structural Models of Credit Risk: A Case Study of Three Models. Mood’s KMV Company.

Compustat North America, Company Financials, Web, 12 June 2017, < https://wrds-web.wharton.upenn.edu/wrds/analytics.cfm>

Crouzet, N. (2016), Aggregate Implication of Corporate Debt Choices, Kellogg School of Management -

Northwestern University.

Danis, A. and Gamba, A. (2016), The Real Effects of Credit Default Swaps, Georgia Institute of

Technology.

Duffie, D. (1999), Credit Swap Valuation, Association for Investment Management and Research.

ESMA, Short Selling, 1 November 2012, Web, 17 August 2017,

<https://www.esma.europa.eu/regulation/trading/short-selling>

Gilson, S. C. (1997), Transaction Costs and Capital Structure Choice: Evidence from Financially Distressed Firms, Wiley and American Finance Association.

(33)

Page 29

Gilson, Stuart C., Kose, J., Lang, L. H. P. (1990), Troubled Debt Restructurings, Journal of Financial

Economics 27, 315-353.

Hu, Henry T.C., and Bernard Black (2008a), Debt, Equity, and Hybrid Decoupling: Governance and Systemic Risk Implications, European Financial Management 14, 663-709.

Hu, Henry T.C. and Bernard Black (2008b), Equity and Debt Decoupling and Empty Voting II: Importance and Extensions, University of Pennsylvania Law Review 156, 625-763.

Hull, John C. (2015), Risk Management and Financial Institutions 4th_{Ed., Wiley, 400-418.}

Junge, B. and Trolle, A.B. (2015), Liquidity Risk in Credit Default SWAP Markets, EPFL and Swiss

Finance Institute.

Longstaff, F.A., Mithal, S. and Neis, E. (2004), Corporate Yield Spreads: Default Risk or Liquidity? New Evidence from the Credit Default Swap Market, National Bureau of Economic Research.

Markit, CDS Big Bang, 13 March 2009, Web, 1 September 2017, <http://www.markit.com/cds/announcements/resource/cds_big_bang.pdf>

Not specified (2012), the Credit Default Swap Market, International Organization of Securities Commissions. Pagano, M. (2005), the Modigliani-Miller Theorems: a Cornerstone of Finance, Working Paper. S and P Capital IQ, Fixed Income, Web, 19 July 2017, <https://www.capitaliq.com/ciqdotnet/login>

Thomson Reuters Datastream, CDS data, Web, 15 May 2017, <

https://financial.thomsonreuters.com/en/products/tools-applications/trading-investment-tools/datastream-macroeconomic-analysis.html>

(34)

Page 30

Appendices

Appendix I: Debt Types

1. Corporate MTN 2. Corporate Debentures 3. Corporate Convertible 4. Foreign Currency Debenture 5. Asset-Backed Security 6. Corporate Strip 7. Collateralized Mortgage Obligation 8. Collateralized Debt Obligation

9. Foreign Governments and Agencies 10. Corporate MTN Zero 11. Preferred Security 12. Corporate Insured Debenture 13. Retail Note 14. Term Loan

15. Corporate Bank Note 16. Collateralized Loan

Obligation 17. Revolving Credit 18. Municipal

19. Corporate Pass Thru Trt 20. Euro MTN

21. Preferred Stock

22. Foreign Government Strip 23. Letter of Credit

24. Trust Preferred Capital Security

25. Corporate Zero 26. Eurobond

27. Canadian Treasury Bond 28. Agency Debenture

(35)

Page 31 29. Bridge Loan

30. Corporate PIK Bond 31. Revolving Credit/Term

Loan

32. Inflation-Indexed Security 33. Agency MTN

The relationship between market expectations concerning when a firm might default and debt renegotiation frictions for US financial and non-financial institutions

Master’s Thesis: