Out-of-court restructuring costs and default timing : an examination of US firms

(1)

University of Amsterdam (UvA)

Out-of-court restructuring costs and default timing:

An examination of US firms

Name: Felix Snoeck Henkemans Supervisor: dr. T. Ladika

Identification: 10385320

Abstract

This paper observes firm-level factors that affect the likelihood of out-of-court debt restructuring and thereby time to default. First, a model is presented that derives bankruptcy probabilities from CDS spreads. Then, quantitative analysis is conducted on a sample of 304 US firms over a 2006-2016 sample period. Two different indicators of the restructuring likelihood are examined: CDS slope and short-term CDS spreads. With regards to the firm-level factors, this paper shows empirically that markets adjust the time to default for debt restructuring frictions. A positive influence of liquidity, intangible assets and bank loan dispersion on time to default is found. The effect of bondholder dispersion and secured bank debt is less evident. Although most of the results are in line with existing literature, the conclusions with respect to bondholder- and bank loan dispersion differ. Analysing variable measurement and recent findings on bank loan covenants, explanations for these contrasting findings are provided. The analysis is important in showing how firm liquidity and debt structure affect welfare during financial distress.

Keywords: Bankruptcy, debt renegotiation, credit default swaps, default timing JEL Classification: G32, G33

1th July 2017 Statement of Originality

(2)

This document is written by student Felix Snoeck Henkemans 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

Firms have different options when they find themselves in financial distress: they can restructure their debt out of court, file for a Chapter 7 liquidation or engage in a Chapter 11 debt reorganization. At stake here is the efficiency of financial distress resolution, as defaults can destroy a substantial portion of firm value (see e.g. Weiss (1990), and Bris et al. (2006)). Out-of-court renegotiation with creditors is believed to be advantageous as the relative costs are low. These low relative costs offer more debt relief, thus firm who restructure can avoid bankruptcy longer and ensure more firm value (see e.g. Gilson et al. (1990)). However, prior work identifies frictions that prevent efficient workouts. These frictions include e.g. creditor dispersion, which prevents efficient creditor coordination, a low intangible assets ratio, which reduces liquidation costs and a severe liquidity crisis, which reduces firms’ breathing time (Gilson (1997), and Chatterjee et al. (1996)). To examine the effects of restructuring frictions on the probability of default recent studies use a credit derivative; Credit Default Swaps (CDS).

A CDS contract is similar to an insurance contract that compensates the CDS buyer for losses which arise from firm- or sovereign default. The insurance buyer receives a specified amount at the time of the default. In return he pays an annuity called the CDS spread. CDS spreads only recently have become a widely-used indicator of default probability. Earlier work on corporate defaults often uses bond spreads as indicator of credit risk, however a growing body of literature has proven the inadequacy of bond spreads as credit risk indicator (see e.g. Collin-Dufresne et al. (2001), and Blanco et al. (2005)). Especially important for this study are the findings of Pan and Singleton (2008). They argue that CDS contracts are less complicated and more liquid than bond contracts. The development of the CDS market into a highly liquid and easy to access market allows for the use of new datasets and the specification of new regression models. Previous literature mainly uses the 5-year CDS spread to determine default probabilities. However, firms also trade CDS contracts with different maturities (e.g. 1-year, 2-year, 3-year). By using these contracts this study constructs a term structure for firm default probabilities. The term structure is the relationship between the default probability and CDS contract maturity and is analogous to the yield curve. This term structure is then used to construct a measure for the time to default.

(5)

More specifically, this paper tries to present new quantitative evidence for the influence of liquidity and renegotiation factors1 on the time to default. This highlights a new channel through which restructuring chances affect time to default. This research objective leads to the following research question:

(i) What is the effect of liquidity and renegotiation factors on the time to default?

To answer the abovementioned research question, this study utilizes a modified version of the implied default probability model presented by Duffie (1999). The modification is needed to extract default probabilities for longer maturities and follows the procedure of BNP Paribas (2015). The default probabilities are annualized, which is analogous to the yield curve and elaborated upon in Section 3.2. After derivation of default probabilities for all firms over all CDS maturities a proxy for default timing is constructed: the CDS slope (5-year default probability minus 1-year default probability). Logically, an increase in this proxy indicates a longer time to default as the 1-year default probability is relatively low compared to the 5-year default probability. Key to this study is the assumption that the restructuring likelihood positively affects time to default. Under this assumption firms engage in debt restructuring attempts first and thus debt restructuring likelihood and time to default are positively correlated. This assumption is motivated by existing work (Gilson et al. (1990)) and relates CDS slope as default timing proxy to the debt restructuring likelihood. In turn, this assumption is the justification for testing the effect of restructuring factors on the CDS slope.

The main analysis is conducted through an ordinary least squares (OLS) model. First, two liquidity factors are included besides a set of control variables found in existing work (Collin-Dufresne et al. (2001), and Campello et al. (2016)). Both liquidity factors have a significant positive effect on time to default. The most significant factor, the working capital ratio, is then included as a control in the next regression, which examines the influence of debt renegotiation factors on default timing. Thereafter, a second proxy for the restructuring

1

Throughout this paper the factors that affect the likelihood of debt restructuring are referred to as restructuring- or renegotiation factors. These two definitions are used interchangeable. The term factors is used since not all factors are restructuring frictions.

(6)

likelihood is introduced, this being 1-year CDS spreads. A logistic model is presented that examines the effect of the firm-level factors on changes in this likelihood. The sample used for the logistic model is unbalanced, therefore a probit model is presented as well to check for the validity of the logistic regression results. This study compares across non-distressed and distressed firms. The separation between these groups is made based on Z-score, a measure of financial health constructed (revised) by Altman in 1968 (2000). The observed effects should be stronger for distressed firms as they are most likely to end up in either renegotiations or bankruptcy procedures.

This study identifies five different renegotiation factors, these are: intangible assets, bondholder dispersion, bank loan dispersion, bank debt and the use of secured bank debt as credit line. These factors have proven to affect debt renegotiation and default (see e.g. Gilson et al. (1990), and Asquith et al. (1994)). For distressed firms, the empirical results show a significant effect for three of the factors examined. Intangible assets and bank loan dispersion both have a positive effect whereas bank debt has a negative effect on time to default. These results are similar across all regression models. The effect of a fourth factor, bondholder dispersion, is ambiguous. A small and positive effect on time to default is found in the OLS model. This positive effect is not found by the logistic and probit model. The fifth and last renegotiation factor examined, secured bank debt, does not significantly affect time to default in any of the models. For non-distressed firms, a marginal or even opposite effect is observed, which is in line with expectations. An additional set of checks is provided to account for limitations in the analysis. The first and most extensive check concerns the measure of financial distress. All regressions are run again using credit rating as measure of financial distress. Also, a different CDS slope definition and bondholder dispersion measure are tested. As a final check, the OLS regressions are run for different recovery rates. The results hold up to the abovementioned set of checks, which suggests a causal relationship between the variables examined and time to default.

This paper and research question fit in a large body of existing literature on the effect of renegotiation factors, with notable work done by e.g. Gilson (1990), Asquith et al. (1994) and more recently Bris et al. (2006). Some of the evidence is conflicting and since most of the research is done a considerable time ago, this provides room for new studies on this matter. Additionally, existing literature mostly relies on rather small samples (<100 firms) since corporate defaults are rare. This raises the question if the results obtained in prior work are applicable to other samples. The applicability to other samples is further impaired by the

(7)

fact that firm defaults are often unique cases where general conclusions are inadequate. Moreover, prior research has not yet examined the effect of firm characteristics on relative default timing using CDS slopes. Recently, Han and Zhou (2011) have used CDS slopes in their study but they examine the informative content for stock returns this being logically different from this paper. Also different from existing work is the treatment of CDS contracts as a payment trigger in case of default instead of a put option equivalent (e.g. Cao et al. (2010), and Han and Zhou (2015)). The key difference is that in the latter case, CDS contracts are related to the stock market so factors affecting the stock market are directly related to CDS spreads. In contrast, this study treats CDS contracts as unrelated to stock market events thereby isolating the variables of interest: liquidity factors and debt restructuring factors. Finally, this research differs from existing literature on CDS slopes in the use of implied default probabilities. Han and Zhou (2015) construct the CDS slope directly from CDS spreads whereas this study derives default probabilities before constructing the CDS slope. The analysis in this paper contributes to the abovementioned literature by showing how time to default and probability of default are affected by liquidity and renegotiation factors. Also, this study highlights the informative content of the term structure of default probabilities. Lastly, this paper presents results based on a recent sample that addresses the impact of the Global Financial Crisis. This analysis is important in showing how regulation on minimum liquidity ratios and debt structure can help distressed firms emerge from distress by efficient out-of-court workouts.

This paper proceeds as follows. Section 2 presents an overview of the relevant existing literature, a restructuring friction model and hypotheses. Then, Section 3 describes the empirical analysis of default timing. Thereafter, Section 4 provides additional checks for the validity of results. Section 5 concludes and provides a summary of results. Finally, Section 6 provides limitations of this paper and directions for further research.

2 Literature review

This section is structured as follows. It starts off with a brief introduction to CDS. Then, the justification of CDS spreads as indicator of default is elaborated upon. Thereafter, the effect of liquidity and renegotiation factors on restructuring likelihood is discussed. Finally, the link between renegotiation factors and successful debt restructuring is elaborated upon.

(8)

2.1 Credit default swaps as default probability indicator

Credit swaps are a form of protection in case of a credit event, such as a downgrade in credit rating or default. The protection buyer receives a specified amount (1 unit) at the time of the credit event. In return he pays an annuity at a certain rate, called the credit swap spread or premium. This premium is paid until either the maturity of the credit swap or the credit event, depending on which event occurs first. CDS are a special form of credit swaps. An important difference between CDS and other derivatives is that in the case of a CDS, the protection buyer receives money upon a documented and contractually defined default event (Duffie (1999, p. 73)). To avoid arbitrage opportunities, the present value of the premium leg should be equal to the present value of the default leg. In particular: a CDS transaction has a net present value of zero at the time of set-up. Thus, a simplified representation of a CDS contract at the time of set up is given by:

( ) ( ) (1) The elements in this equation are:

- Recovery rate ( ) - Hazard rates ( ) - Discount factor ( ) - CDS premium ( )

The recovery rate is defined as the extent to which the principal of a debt instrument that is in default can be recovered. The rate is expressed as a ratio of the debt instrument’s face value. The hazard rate is defined as the probability of default for a given entity in a period conditional on surviving all previous periods (BNP Paribas (2015, p.10)). The discount factor is the discount rate of future spread payments, equal to the prevailing risk-free rate. Lastly, the CDS premium is the abovementioned annuity.

CDS pricing models rely on several assumptions. First, they assume a constant recovery rate equal to a pre-specified percentage of par value. Second, there is no correlation between interest rates and the level of default (independence). Finally, hazard rates have a ‘piecewise constant form’, meaning they may change over different periods but are constant within periods. Both the premium and default leg in equation (1) are uncertain cash flows as the default payment of 1 minus recovery rate happens only upon default before maturity and the premiums are paid up until either default or maturity. Calculations that equalize both

(9)

uncertain cash flows are based on credit default probability. Thus, CDS spreads are an indicator of this probability. The term structure of CDS spreads is the path constructed by connecting the CDS spread levels over different time horizons. This implies that the term structure of CDS spreads reflects the probability of default for a given entity over different time horizons. Detailed CDS equations and the process of deriving default rates from CDS spreads are discussed more elaborately in subsection 3.2.1.

2.2 Advantages of CDS contracts over bond contracts

Previous studies mainly use firm-level differences in bond credit rating to forecast the term structure of credit spreads and thereby default probabilities. Some studies find relatively high explanative power for bond data (Avramov et al. (2007)). The growth in CDS markets since the beginning of this century allows more recent research to use CDS data instead of corporate bond data. Since then, several studies have documented the underperformance of bond data showing bonds suffer from bias by illiquidity factors and shocks in local supply and demand (see e.g. Collin-Dufresne et al. (2001), Blanco et al. (2005), Longstaff et al. (2005), and Ericsson et al. (2006)). Moreover, these studies show CDS spreads lead credit spreads in terms of price discovery.

Han and Zhou (2011) are one of the first to use the CDS term structure instead of a single maturity to assess the usefulness of different structural models. They identify a feature of the CDS contract that is key to this study: CDS contracts trade in different maturities while this is not always the case for bond contracts. Even if bonds do trade in different maturities for a given firm, the contract terms may differ. Studies that use corporate bond data to examine the credit term structure therefore must aggregate individual bonds into different portfolios. This makes bond data unable to assess firm-specific term structures. A final advantage of CDS spreads over bond yields is that CDS spreads are independent of the current risk-free rate by definition. Thus, comparison over time is facilitated by the independence of the prevailing interest rate. In line with recent work, this study uses CDS spreads instead of bond spreads to assess default timing. I now turn to the firm-level factors that drive differences in default timing.

2.3 Firm liquidity

A firm characteristic that should relate to default is liquidity as it affects the chance of being financially distressed. The general view of financial distress is a mismatch between current

(10)

obligations and current liquid assets. A higher level of liquid assets mitigates this mismatch problem. The degree of liquid assets differs both across firms and over time. Previous literature finds that liquidity ratios explain differences in the probability of default. The main theory is formulated by Altman (1968) and Altman et al.(1977). The former constructs a so-called Z-score model based on multiple financial ratios including liquidity ratios to predict corporate defaults. In 1977, Altman et al. update this Z-score model by constructing a ZETA-model consisting of seven factors. Their empirical results show that both ZETA-models are very accurate in prediction of firm bankruptcy, with accuracy ranging from 96 percent (70 percent) one period (five periods) prior to bankruptcy. The two liquidity measures with the highest informative content in these studies are the current ratio (in the ZETA-model) and the working capital ratio (in the Z-score model). Altman (1968) finds a significant difference in the working capital measure between non-bankrupt and bankrupt firms of 0.47. Similarly, the ZETA-model finds a significant difference in this ratio of 0.16. Unfortunately, Altman et al. (1977) do not state the difference in current ratio between non-bankrupt and bankrupt firms for proprietary reasons.

In line with existing work, I expect that firms with high liquidity ratios will be able to postpone bankruptcy longer relative to firms with low liquidity ratios. Even if the profitability of the more liquid firms declines, they can continue operating by selling some of these liquid assets. Moreover, liquidity may have a positive effect on the debt restructuring likelihood which in turn also lengthens time to default (Chatterjee et al. (1995)).

2.4 Bankruptcy procedures and renegotiation factors

This section discusses factors that affect time to default through their effect on restructuring chances. First, a brief overview of the different bankruptcy procedures is provided.

2.4.1 Chapter 7 liquidation

The first way of resolving bankruptcy is a Chapter 7 liquidation. Chapter 7 of the US Bankruptcy Code governs the process of liquidation under the bankruptcy laws of the United States. In a Chapter 7 liquidation, the court ensures priority of claims and governs the distribution of cash across claimholders.

(11)

2.4.2 Chapter 11 reorganization

The second way of distress resolution is a Chapter 11 reorganization. Chapter 11 of the US Bankruptcy Code defines the rules under which stakeholders of a financially distressed firm renegotiate their claims in court. Protection of the firm during the reorganization process is an important purpose of this bankruptcy procedure, especially protection from the seizure of assets by the firm’s creditors. Chapter 11 aims at securing continuation of viable firms by undergoing value-increasing restructurings (Weiss and Wruck, (1998)). For this sake, management has 120 days to file a reorganization plan. Ideally, Chapter 11 would force firms to not only restructure their debt, but also evaluate their current operations, change their ownership or choose to shut down.

In a Chapter 11 filing, the plan of reorganization affects firm value and the distribution of firm value across all claimholders. This incentivizes claimants to support a value-destroying reorganization plan, as long as the value of their claim exceeds the value of their claim in a value-maximizing plan. Wruck (1990) argues that conflicts which arise through concerns about firm value and distribution among claimholders result in a situation where all participants neither have the relevant information nor are willing to share this with others. Weiss (1990) examines 37 US firms that filed for bankruptcy in the 1979-1986 period and reports a violation of priority of claims (APR violation) in Chapter 11 filings. Secured creditors’ contracts are generally upheld but between unsecured creditors and equity holders and even among unsecured creditors APR violation occurs. Miller (1977, p.45) states: “[…] permitting stockholders to claim court protection and thereby retain control of a corporation in default would amount to giving them a call option at the expense of the creditors.”. The reorganization process leaves room for negotiations among the various parties and thus to the possibility of APR violation. In conclusion, a Chapter 11 filing potentially leads to inefficient outcomes due to the abovementioned problems.

2.4.3 Out-of-court debt restructuring

Instead of choosing a formal bankruptcy procedure by filing for a Chapter 11 or Chapter 7 firms may choose to restructure their debt out of court. This out-of-court debt restructuring theoretically leads to more efficient outcomes compared to the formal bankruptcy procedures (Gilson et al. (1990)). Ideally, firms in financial distress would renegotiate with their claimholders out of court thereby enhancing more firm value. They do so by either a tender offer or an exchange offer. In a tender offer debt is repurchased with a firm’s cash while in an exchange offer current debt is exchanged for (a combination of) equity and new debt. The

(12)

general view is that these offers based on mutual agreements are beneficial, thus firms will always try to renegotiate out of court first. The renegotiation of debt allows for more ‘breathing time’ to improve business processes and prevents asset sales in court. However, there are important factors that affect the chance of successful renegotiations. These factors are discussed in the following section.

2.4.3.1 Factors affecting out-of-court debt restructurings

In out-of-court debt restructurings, there is no formal supervisor (the court) to govern the renegotiation process. Thus, claimholders are on their own in determining the viability of the firm. Asymmetric information potentially leads to high costs as both information collection and inefficient outcomes are costly. Therefore, firms sometimes end up in court after a private restructuring, not being able to sufficiently reduce debt levels. Gilson (1997) examines 108 publicly-traded firms that restructured their debt either by a Chapter 11 reorganization or out of court. Gilson’s study mainly focusses on the transaction costs incurred during these procedures. He finds multiple reasons why transactions costs are lower during Chapter 11 reorganizations than during out-of-court debt restructurings including: (1) tax penalties for reducing debt are lower in a Chapter 11, (2) creditors have less power to block debt reorganization plans and (3) a Chapter 11 facilitates sales of assets (Gilson (1997, p. 5)). The tax penalty friction has been of less importance since the TD9599 regulation was enforced on September 12, 2012, significantly reducing out-of-court restructuring costs for distressed firms with high syndicated bank loans (see Campello et al. (2016)). However, creditors’ power to block debt reorganization plans is still an important friction. Debt renegotiation without an intervening court requires high levels of creditor coordination. To make sure renegotiation plans are not beneficial to only certain claimholders while others bear the cost, a reorganization plan can only pass when a majority of the claimholders approves. Furthermore, The Trust Indenture Act of 1939 states that only unanimous bondholder approval can lead to changes in a bond’s principal, maturity or interest. Intuitively, both bondholder approval and approval by majority are more likely when the number of creditors is small. Thus, creditor dispersion reduces the likelihood of out-of-court debt restructuring.

Another problem that arises during out-of-court restructurings is the holdout problem. In out-of-court restructurings public exchange offers are an important way to prevent bankruptcy. Asquith et al. (1994) find that firms who do not sell off a major part of assets or engage in an exchange offer have a high likelihood of bankruptcy. The holdout problem

(13)

occurs when bondholders won’t participate in an offer that potentially benefits non-tendering bondholders at the expense of tendering bondholders (Chatterjee et al. (1995, p. 334)). The holdout problem proves to be an important factor whereas it influences the type of security used during a workout. This problem is further exacerbated by creditor dispersion, as each creditor now accounts for only a small part of total liabilities. These small creditors are less eager to participate in renegotiation talks and tender, as they are not decisive in the outcome. They holdout until larger creditors take the so-called ‘haircut’. Consequently, if no one wants to tender debt restructuring fails.

The creditor dispersion problem as described above mainly concerns public creditor dispersion. While the same arguments potentially hold for bank creditor dispersion, recent work presents evidence that relates bank creditor dispersion to successful debt renegotiations. Saavedra Lux (2015) finds that the number of banks acting as a lender is negatively related to the level of restrictive covenants. Restrictive covenants are covenants that reduce a firm’s financial flexibility. Thus, having less restrictive covenants leaves more room for debt renegotiation and the number of bank lenders is a proxy for the restrictiveness of covenants. In this light, bank loan dispersion improves debt restructuring chances if less restrictive covenants substantially reduce restructuring costs. These recent findings highlight a new channel through which the number of bank lenders affects the restructuring likelihood as existing work mainly identifies a negative relation between number of creditors and restructuring chances (e.g. Gilson (1990), and James (1996)).

Apart from creditor dispersion, bank debt has also proven to affect the likelihood of debt restructuring. Previous literature presents conflicting evidence on the relation between bank debt and the ability to resolve distress out of court. Gilson et al. (1990) find that firms who rely more on bank financing have a higher likelihood of restructuring out of court. Gorton and Kahn (1993) support these findings by stating that bank debt forgiveness occurs during financial distress. If a secured creditor forgives debt this increases the likelihood of a junior for senior debt exchange that also involves bondholders. James (1996) argues that banks play an important role in successful out-of-court distress resolution, mainly by resolving hold out problems. Nonetheless, measuring the actual role of bank lenders in the reduction of hold out problems seems difficult. Note that secured bank loans are only expected to be renegotiated when borrowers are in severe financial distress (only in this case their claims are impaired). Simultaneously, severely distressed firms need high debt

(14)

reductions to prevent default. As a result, bank concessions may coincide with bondholder concessions even if holdout problems are absent.

In stark contrast with the abovementioned positive effects of bank debt on restructuring likelihood is work by Chatterjee et al. (1996). They find that firms filing for a Chapter 11 have higher levels of bank debt. Moreover, public creditors likely capture some of the gains created by bank participation in debt relief offers. This in turn limits the willingness of banks to bear the full costs of such offers. Asquith et al. (1994) and James (1995) support this argumentation, as they present evidence that banks rarely do concessions if a firm also has outstanding public debt. Also, the fact that banks are often secured lenders has proven to make them more likely to participate in renegotiation during early stages but more prone to push the firm into bankruptcy at later stages due to their claim protection. This is in line with John (1993) who states that fully secured lenders often prefer liquidations over debt restructurings.

The main conclusion I derive from the previous subsection is that there is a clear relationship between a company’s debt structure and the way of financial distress resolution. When public creditor dispersion is high, these small creditors are likely to free ride thereby reducing debt relief. The effect of bank creditor dispersion is ambiguous, as less restrictive covenants may actually reduce frictions for highly dispersed bank lenders. Moreover, public creditors likely benefit from banks who tender during renegotiation. Consequently, this reduces banks’ willingness to bear the full costs or even participate in restructuring offers. Lastly, bank lenders are mostly senior and sometimes even secured lenders. This protection reduces their willingness to renegotiate out of court.

Apart from debt structure, prior studies identify another factor that affects restructuring chances, intangible assets. Argument (3) by Gilson (1997) at the beginning of this subsection states a Chapter 11 facilitates asset sales and thereby reduces transaction costs. This argument only holds for assets with a high liquidation value, these being tangible assets. When asset liquidation value is low, as for intangible assets, asset sales in a Chapter 11 negatively affect firm value. Gilson et al. (1990) support this argument by stating that during a Chapter 11 the sales of assets is more likely and therefore a Chapter 11 is costlier when intangible assets are high. The sale of assets also makes it more difficult for firms to attract secured as well as unsecured debt, in turn forcing it into bankruptcy. In contrast with these findings is evidence presented by Asquith et al. (1994) which suggests sales of assets are an important way to prevent bankruptcy procedures. Looking at intangibles this way,

(15)

having a higher portion of intangibles rules out the possibility of these asset sales and therefore would shorten time to default. Arguably, the latter argument only holds in cases where intangible assets constitute a substantial portion of total assets.

In conclusion, prior work reports different effects of certain firm characteristics such as debt structure, creditor dispersion and intangible assets on the likelihood of successful debt restructuring. The effects seem to differ across different samples and levels of distress. Table 1 presents predicted effects of the different factors on time to default.

Table I

Firm-level factors and time to default

Variable Description Data source Predicted sign Liquidity Current Ratio/Working Capital Ratio Compustat + Intangible assets Intangible assets to total assets ratio Compustat + Bank loan dispersion Number of banks acting as lender LPC-Dealscan + Bank debt Fraction bank debt of total debt Compustat/LPC-Dealscan − Bondholder dispersion Number of bond issues Mergent FISD − Debt incl secured bank debt Credit line of secured bank debt LPC-Dealscan −

This table presents the key independent variables that explain differences in time to default. Mergent FISD is Mergent Fixed Income Securities, LPC-Dealscan is Loan Pricing Corporation-Dealscan.

2.5 Renegotiation frictions and bankruptcy outcome

When a firm enters financial distress, key to the probability of evolving out of this distress by renegotiation is the height of renegotiation costs (see e.g. Gilson (1990), Asquith et al. (1994), and more recently Campello et al. (2016)). Put differently, creditors only prefer debt renegotiation when the value of their outside option exceeds the value they obtain in court. This section describes the effect of renegotiation frictions on restructuring chances based on a framework by Campello et al. (2016).

Assume the following: a three-period economy, a single borrower, multiple creditors, a constant observable in-court recovery rate, no discounting and risk-neutrality for all participants. The borrower has issued debt securities at t = 0. Each debt security promises a payoff of 1 unit in = 2. If the firm is non-distressed, cash-flow exceeds debt obligations and bankruptcy or debt restructuring is prevented. If the firm is in distress, two possible scenarios arise: (1) bankruptcy or (2) out-of-court restructuring. In scenario 1, firm value is equal to the in-court recovery rate. In scenario 2 firm value is unobservable for all participants until = 2.

(16)

Campello et al. (2016) argue that firm value in out-of-court renegotiations is too complex or uncertain to contract upon and this uncertainty is resolved in = 2, leaving room for renegotiation at this time. Renegotiation is successful at time = 2 if all lenders accept the individual offer of the borrower. If renegotiation is unsuccessful, lenders turn to court and end up with the in-court recovery rate. All lenders have an exposure to the default risk of the borrower and can choose to buy protection against this risk in the form of CDS contracts. Each lender’s protection decision is made before actual firm value is known. As stated in Section 2.1 CDS contracts pay the protection buyer 1 unit if and only if default occurs.

Renegotiation only is successful if the borrower’s offer exceeds the value lenders obtain in-court. From a borrower’s perspective, the most efficient way of renegotiation is offering an amount equal to the in-court recovery rate for unprotected lenders and 1 for protected lenders. This makes lenders indifferent between going into bankruptcy and renegotiation and is called the optimal offer. This optimal offer is affected by renegotiation frictions. If frictions are sufficiently large then the optimal offer will always be lower than the in-court recovery rate. In this case, both protected and unprotected lenders will push the firm into bankruptcy, ensuring payment of the recovery rate. However, if both probability of the offer being lower than 1 (the pay-out for CDS buyers) and lender’s exposure are sufficiently large renegotiation can be successful when frictions are small. If frictions are small, the optimal offer can exceed the in-court recovery rate and in this case, protected lenders are willing to renegotiate because more value is enhanced. Logically, unprotected lenders are also willing to renegotiate in this situation as the maximum value they obtain in court is the recovery rate. Thus, the probability that out-of-court renegotiation succeeds is negatively related to total frictions.

2.6 Testable Hypotheses

One can derive several conclusions from the previously stated debt renegotiation theory. The main conclusion I derive is: firms with high renegotiation frictions have a higher relative short-term probability of default. The argument is that firms with low frictions are more likely to restructure out of court, thereby preserving more firm value and preventing default. The influence of renegotiation factors should be smaller for non-distressed firms as an inverse relation between level of distress and the impact of these factors on time to default is expected. Furthermore, a simple comparison of time to default across firms might be short-sighted. Recall the proxy for time to default is the CDS slope defined as the 5-year default

(17)

probability minus the 1-year default probability. It is possible that CDS slopes are equal for two firms with very different 1-year and 5-year probabilities of default (under positive correlation). Therefore, throughout this paper this study compares across firms with the same 5-year probability of default when analysing differences in default timing. Guided by the conclusions in Section 2.4 and 2.5, the following hypotheses are developed:

Hypothesis 1: Liquidity has a positive influence on time to default for distressed firms.

Hypothesis 2: Intangible assets have a positive influence on time to default for distressed firms.

Hypothesis 3: Bank loan dispersion has a positive influence on time to default for distressed firms.

Hypothesis 4: Bank debt has a negative influence on time to default for distressed firms.

Hypothesis 5: Bondholder dispersion has a negative influence on time to default for distressed firms.

Hypothesis 6: Secured bank debt has a negative influence on time to default for distressed firms.

3 Empirical Tests

This section describes the empirical process of testing the hypotheses by CDS spread analysis. First, I describe my data. Then, I present the CDS default probability model which extracts default probabilities from CDS spreads. Thereafter, I present my ordinary least squares regression model. Finally, the results section presents empirical results and implications of these results.

(18)

3.1 Data

This study uses daily CDS data obtained through Thomson Reuters’s Datastream. The raw data sample consists of daily CDS dealer quotes on 1743 public firms in the US, UK and Japan ranging from January 2004 to December 2016. I convert daily CDS data to monthly data by keeping the last observation per month based on trading day. As I show subsequently in subsection 3.2.1, all contracts with maturities ≥ 1-year and ≤ 5-year are required for each firm at time to extract the 5-year default probability. Therefore, I exclude all observations with missing data on these contracts. Moreover, I exclude all CDS contracts with clauses other than ‘No restructuring’. Apart from differences in pricing, (see O’Kane, Pedersen and Turnbull (2003)) a debt restructuring triggers CDS payment for all clauses other than ‘No restructuring’. This troubles results as an increase in CDS spread now either indicates higher probability of default or higher probability of restructuring.

The CDS sample from Datastream is merged with multiple other databases. I collect detailed bond data through Mergent FISD. This database contains issue- and issuer details on US corporate bonds and US corporate medium term notes. Data on syndicated bank loans is from LPC-Dealscan. The data includes loan principle, type and number of loan participants, signing dates and maturity dates. Data on firm fundamentals is from Compustat’s Quarterly Update. I exclude 102 financial institutions (SIC code 6000-6300) and 537 firms that are foreign-based (UK/Japan). Thereafter, I drop 526 firms with missing data for the major part of the sample period on either bond issues or firm fundamentals. Next, I convert quarterly Compustat data into monthly data by simply assigning the quarterly value to the corresponding three months. In line with the procedure of Collin-Dufresne et al. (2001), Galil et al. (2007), and Avramov et al. (2007) I exclude all firms with less than 25 monthly CDS quotes. To prevent bias of results by an illiquidity factor I conduct the analysis using only contracts that are at least traded once a month. I exclude all contracts where CDS quotes are equal in subsequent month ( ) which indicates no trade has occurred. This restriction is imposed for all maturities to prevent contamination of 5-year default probabilities. Approximately 84 percent of the original sample matches the constraint. While this shrinks my sample, Table IX in the appendix shows that results are indeed severely biased without exclusion of illiquid contracts (see e.g. and estimates), thereby proving the existence of the illiquidity factor. My final data sample contains 304 US firms.

(19)

3.2 Methodology

The next section carefully describes the methodology this study uses to asses default timing using CDS contracts. First, I state the implied default probability model. Then, I present the ordinary least squares regression model. Lastly, a detailed variable description is provided.

3.2.1 CDS spread implied default probability

To test the hypotheses stated in Section 2.6, I need to construct a model that explains variation in default timing based on firm-level factors. First, I need to operationalize the dependent variable which is the measure of default timing. Therefore, implied default probabilities are derived from CDS spreads. The pricing of CDS contracts follows a so-called Poisson process. Duffie (1999, p. 79) constructs a model for CDS pricing under Poisson processes which is represented by the following equations. Suppose a constant hazard rate of and constant interest rate of per period. In this case:

( ) [ ( )] ( ) (2)

( ) [ ( ) ( )] [ ( )] [ ( )] (3) Where: ( ) is the present value of receiving 1 unit at coupon date if default occurs after that date; ( ) is the present value of receiving 1 unit at coupon date if default occurs between the ( )th and the th coupon date; ( ) is the time to maturity of the th coupon date and ( ) is the default-free zero-coupon yield with continuously compounding to the th coupon date. As ( ) and ( ) must be equal to prevent arbitrage opportunities, I can derive the hazard rate in year 1 from the 1-year CDS spread . This is done by solving equation (4) for while assuming a spread payment frequency of and constant recovery rate of : [∑ ( ) (∑ ( ) ) ] (4)

The recovery rate is set to 0.6 which is arbitrary, however in line with Moody’s Ultimate Recovery Database. This database shows historical recovery rates ranging from 38 to 82 percent depending on the type of debt security. As CDS spreads consist of a default probability and recovery component, a higher recovery rate implies a higher probability of

(20)

default for a given CDS spread. Moreover, a risk-free rate of 0 and quarterly spread payment are assumed. Now, the implied 1-year default probability is equal to:

( [ ]) (5)

To derive default probabilities for longer maturities an adjustment is made for each period following BNP Paribas (2015). For the 2-year default probability, the 2-year CDS spread is used in addition to the 1-year hazard rate to calculate the hazard rate in year 2 (equation (6)). Subsequently, the 2-year annualized probability of default is calculated using equation (7). This calculation of default probabilities over different maturities is similar to the calculation of the yield curve using bond yields. This process is called ‘bootstrapping’ of interest rates (Munk (2011)). The main difference here is that this paper uses CDS contracts in different maturities to calculate hazard rates. So, in this procedure the bond is replaced by the CDS contract and the interest rate is replaced by the hazard rate. Similar to the yield curve, the bootstrapped probabilities of default are annualized. One example is provided below, the formulae used for deriving the default probability in years 3-5 can be found in the Appendix, as they are an iteration of this process.

2-year default probability =

[(∑ ( ) (∑ ( ) ) ) (∑ ( ) (∑ ( ) ) ) ] (6) ( [ ]) (7)

The choice for the 5-year CDS contract in this analysis is arbitrary, however in line with prior work (see e.g. Hull et al. (2004), Blanco et al. (2005), and Han and Zhou (2015)). Trading volume of this contract is high and therefore liquidity. This mitigates bias by illiquidity as a non-default factor. To assess differences in default timing I construct the CDS slope by subtracting the 1-year default probability:

(8)

The dependent variable for this study has now been operationalized and requests a regression model for further analysis. The next subsection presents this model.

(21)

3.2.2 Regression models

The model presented in this section predicts that higher liquidity and lower debt renegotiation frictions are associated with a lower short-term default probability. Arguably, this effect should be more pronounced for financially distressed firms as those firms are most likely to end up in bankruptcy procedures or debt renegotiations. The model therefore predicts a more severe response to the firm-level factors for distressed firms. The sample period allows for examination of default timing in two separate ways: (1) distressed firms versus non-distressed firms and (2) crisis versus non-crisis periods. I conduct the two-dimensional comparison by an ordinary least squares regression of firm in month . Two regression models are used which are represented by the following equations:

( ) (9) ( ) ( ) ( ) ( ) ( ) ( ) (10)

In equation (9), the least squares method is used to estimate the effect of firm liquidity on CDS slope. Hypothesis 1 is accepted if is positive and significant at the 10 percent level. In equation (10), five renegotiation factors are included in the regression model besides liquidity. The main coefficients of interest in equation (10) are the interaction coefficients , , , , and . A negative coefficient for these variables means a lower CDS

slope, a lower likelihood of debt restructuring and thus shorter time to default. Correspondingly, a negative coefficient suggests the factor is a friction during out-of-court debt restructuring while a positive coefficient suggests the factor facilitates debt restructuring. In terms of interpretation, hypothesis 1 is accepted if is positive and significant at the 10 percent level. Likewise, hypothesis 2 and 3 are accepted if and are positive and significant at the 10 percent level. Hypothesis 4, 5 and 6 are accepted if the coefficient on the

(22)

corresponding factor ( , or ) is negative and significant at the 10 percent level.

These three factors are expected to be renegotiation frictions.

As Brooks (2014) states, a plausible goodness of fit measure for OLS models is . This measure observes the proportion of variability in the dependent variable explained by the independent variables. Therefore, is reported for all OLS regressions.

3.2.3 Variable description

The following section carefully describes the variables used during this study. The variables are separated in two categories: liquidity factors and renegotiation factors.

3.2.3.1 Liquidity measures

Firm liquidity can be represented by different ratios or even models. This paper uses the liquidity measures as defined in studies by Altman (1968) and Altman et al. (1977). equals the working capital ratio whereas represents the current

ratio: 3.2.3.2 Renegotiation factors

Besides liquidity this paper tests the effect of five different renegotiation factors, described in the following subsection. represents the portion of intangible assets to total

assets. It is defined as:

is a proxy for bondholder dispersion at firm-level. Bondholder dispersion is difficult to measure precisely because the number of bonds each bondholder holds for a given bond issue is unknown. Therefore, a proxy is needed to estimate bondholder dispersion. In line with prior research (Gilson et al. (1990), and Asquith et al. (1994)) this proxy is defined as the total number of bond issues outstanding for firm at time . is a proxy for bank loan dispersion. The LPC-Dealscan database specifies the number of bank loans outstanding for each firm as well as the number of different banks participating in each loan. The bank loan dispersion proxy is defined as the

(23)

total number of different banks participating in loans for each firm at time . is a dummy variable equal to 1 if firm debt includes secured bank debt and 0

otherwise. is the part of total liabilities consisting of bank debt and is a dummy variable equal to 1 if a firm is in distress and 0 otherwise.

equals 1 for firms with a Z-score < 1.9 and 0 for firms Z-score ≥ 2.8. (firms

with 1.9 ≤ Z-score < 2.8 are omitted). This separation follows the procedure of Altman (2000) and excludes firms for which the Z-score is ambiguous in the determination of financial distress. To compare effects across non-distressed and distressed firms, an interaction variable is created for each factor, , with being one of the five renegotiation factors.

3.2.3.3 Control variables

contains a set of firm-level control variables based on existing work (see e.g. Ericsson et

al. (2009), and Campello et al. (2016)). These include , , and . For a positive effect is expected as it is a profitability measure and more profitable firms should have a lower short-term default probability. For , a positive coefficient is expected as Gilson et al. (1990) state firms that restructure out of court are generally larger. A negative effect is expected for as high leverage should increase the short-term default likelihood. is an indicator of the state of the economy and following Han and Zhou (2015), a negative effect is expected for as a steeper yield curve reduces the default probability more over a longer horizon. I also include the 5-year default probability, , as a control variable. The inclusion of this control is key to this study, as controlling for the 5-year default probability leads to testing the differences in time to default for a given 5-year default probability. In other words, it leads to testing the unique relationship between the variables of interest and CDS slope. A positive effect of this variable on CDS slope is expected as CDS slope should increase for an increase in by the analogy to the yield curve ( is affected less than ). A control, , is included as well to account for differences within CDS contracts. Explanations for the control variables are given in the Data Appendix.

Han and Zhou (2015) find that credit rating affects the probability of default in the short-term and thereby CDS spreads. To account for credit rating differences, fixed effects for credit rating are used at firm level. Year fixed effects are also included to account for

(24)

shocks in average CDS spreads over time. Finally, industry fixed effect based on the Fama-French 12-industry classification are included to account for industry differences.

3.3 Descriptive statistics

Table II presents summary statistics at firm level for the variables used in the previously stated regression model. Panel A summarizes firm characteristics for the distressed subsample whereas Panel B summarizes firm characteristics for non-distressed firms. The separation is made based on Z-score. Panel C presents t-statistics of a test for difference in means between the two subsamples. In line with expectations distressed firms on average have higher CDS spreads, leverage and probabilities of default. Also unsurprising are the lower average return on assets and liquidity ratios for distressed firms. In addition, distressed firms have more dispersed creditors and rely more on bank debt although the difference in the latter is rather small. Lastly, firm size is roughly the same for distressed and non-distressed firms. These conclusions are based on the significant mean differences for all variables except total assets shown in Panel C.

Next, graphical evidence is presented to get a grasp on default probabilities and CDS slopes. Figure 1 presents average default probabilities derived from 5-year and 1-year CDS contracts respectively. The 5-year probability of default is substantially higher than the 1-year probability of default, thus the market seems to account for differences in default risk associated with maturity. The graphical evidence in Figure 1 further suggests that 5-year and year default probabilities are positively correlated as distressed firms have both higher 1-year and 5-1-year default probabilities. The correlation coefficient of 0.57 between these two variables supports this finding. Figure II presents the average CDS slope measured as the vertical distance between 5-year and 1-year default probabilities in Figure 1. The high peak during 2008 and 2009 in Figure 1 (low CDS slope in Figure II) coincides with the Global Financial Crisis. Average CDS slopes are substantially higher for distressed firms. Moreover, CDS slopes are overall positive meaning that annualized 1-year probability of default is smaller than its 5-year counterpart. This implies a higher likelihood of survival on the short-term relative to the long-short-term. The CDS slope is usually upward sloping because longer maturity securities are riskier in the sense that they are more sensitive to uncertain shocks in the future (Palhares, (2013)). However, a negative CDS slope is also possible meaning that short-term default is more likely. This phenomenon is observed during the first months of

(25)

Table II Summary statistics

This table presents summary statistics for my sample over time period 2006-2016. The sample contains only non-financial firms with matching observations from the Compustat, LPC-Dealscan and Mergent FISD databases. Observations are monthly and at firm level. Distressed firms have Z-score < 1.9 and non-distressed firms have Z-score ≥ 2.8. (firms with 1.9 ≤ Z-score < 2.8 are omitted). is the sum of debt in current liabilities and long-term debt divided by the market capitalization. is defined as current assets minus current liabilities divided by assets. is the ratio of current assets to total assets. is the ratio of intangibles to total assets. is the ratio of bank debt to total debt.

equals the number of bond issues. is the number of banks acting as lender. is interest expenses plus net income,

divided by total assets. All variables are winsorized at the 1-99 level. *, **, *** represent statistical significance at the 10, 5 and 1% level.

Panel A: Distressed firms

N Mean Std. dev 25th Perc. Median 75th Perc

CDS spread 1-year 7922 176.18 371.19 23.60 66.08 172.10 CDS spread 5-year 7922 310.87 385.34 83.50 194.14 402.50 Default 1 (%) 7922 3.92 7.03 0.59 1.64 4.19 Default 5 (%) 7922 11.67 10.45 3.36 7.92 16.84 CDS slope 7922 7.75 8.83 1.82 4.64 10.94 Z-score 7922 0.88 0.76 0.53 0.96 1.43 Leverage 7922 1.12 1.46 0.39 0.67 1.23 Liquidity 7922 0.07 0.11 -0.01 0.06 0.15 Liquidity 2 7922 1.40 0.63 0.92 1.31 1.78 Intangible 7922 0.20 0.20 0.04 0.14 0.32 BankCreditor 7922 0.56 0.33 0.27 0.56 0.90 BondDispersion 7922 5.37 6.16 1.00 4.00 7.00 BankDispersion 7922 135.17 165.57 35.00 78.00 180.00 Return on assets 7922 0.01 0.02 0.01 0.01 0.02 Total Assets 7922 17,687 27,043 5,142 9,340 19,429 Panel B: Non-distressed firms

N Mean Std. dev 25th Perc. Median 75th Perc

CDS spread 1-year 2217 51.35 159.62 8.50 19.90 47.41 CDS spread 5-year 2217 101.21 145.90 32.00 54.00 120.00 Default 1 (%) 2217 1.20 2.86 0.21 0.50 1.18 Default 5 (%) 2217 4.20 5.00 1.36 2.38 4.82 CDS slope 2217 3.00 4.51 0.91 1.6 3.04 Z-score 2217 3.78 0.92 3.06 3.46 4.29 Leverage 2217 0.26 0.28 0.09 0.16 0.33 Liquidity 2217 0.13 0.14 0.01 0.12 0.23 Liquidity 2 2217 1.70 0.86 1.05 1.56 2.10 Intangible 2217 0.18 0.16 0.06 0.14 0.27 BankCreditor 2217 0.50 0.31 0.22 0.45 0.79 BondDispersion 2217 4.15 4.74 1.00 3.00 6.00 BankDispersion 2217 74.49 91.27 25.00 52.00 92.00 Return on assets 2217 0.03 0.02 0.02 0.03 0.04 Total Assets 2217 17,231 20,087 4,667 10,959 20,586 Panel C:

Test of equal means

Distressed ─ Non-distressed t-stat

CDS spread 1-year 124.84 15.44*** CDS spread 5-year 209.66 25.12*** Default 1 (%) 2.72 17.78*** Default 5 (%) 7.47 32.61*** CDS slope -4.75 -24.44*** Z-score -2.90 -130.12*** Leverage 0.86 27.62*** Liquidity -0.06 -20.24*** Liquidity 2 -0.30 -18.42*** Intangible 0.02 4.42*** BankCreditor 0.07 8.52*** BondDispersion 1.22 8.65*** BankDispersion 60.68 16.57*** Return on assets -0.02 -31.37*** Total Assets 455.81 0.74

(26)

2009. Figure II further suggests that absolute CDS slope increases with CDS spread level, as distressed firms have both higher CDS slopes and CDS spread levels. To check whether the market expects differences in default timing across non-distressed and distressed firms, Figure III presents average CDS slopes scaled by 5-year default probability. These adjusted slopes thus represent the slope as a fraction of this probability. The scaling is necessary for comparison across the two subsamples given a 5-year default probability. If the market expects distressed firms to default sooner the CDS slopes would be lower for distressed firms for the same 5-year default probability. Evidence in Figure III shows adjusted CDS slopes are comparable for both subsamples, suggesting that although distressed firms have higher 1-year and 5-year default probabilities, the time to default is similar given a 5-year default probability.

Figure I

Average default probability for distressed versus non-distressed firms

Default probabilities are averaged across firms for each month. Distressed firms have Z-score < 1.9 and non-distressed firms have Z-score ≥ 2.8 (firms with 1.9 ≤ Z-score < 2.8 are omitted).

0 5 10 15 20 25 30 D e fa u lt p ro b a b ili ty (% ) 01/01/06 01/01/08 01/01/10 01/01/12 01/01/14 01/01/16 Date

Default probability 1-year (dis) Default probability 5-year (dis) Default probability 1-year (non-dis) Default probability 5-year (non-dis)

US firms 2006-2016

(27)

Figure II

Average unadjusted CDS slope for distressed versus non-distressed firms

CDS slopes (defined as 5-year default probability minus 1-year default probability) are averaged across firms for each month. Distressed firms have Z-score < 1.9 and non-distressed firms have Z-score ≥ 2.8 (firms with 1.9 ≤ Z-score < 2.8 are omitted).

Figure III

Average adjusted CDS slope for distressed versus non-distressed firms

CDS slopes (defined as 5-year default probability minus 1-year default probability) are averaged across firms for each month and scaled by 5-year default probability. Distressed firms have Z-score < 1.9 and non-distressed firms have Z-score ≥ 2.8 (firms with 1.9 ≤ Z-score < 2.8 are omitted). 0 5 10 15 20 25 30 C D S sl op e (% ) 01/01/06 01/01/08 01/01/10 01/01/12 01/01/14 01/01/16 Date

Avg. CDS slope (dis) Avg. CDS slope (non-dis)

US firms 2006-2016 0 .2 .4 .6 .8 1 C D S sl op e 01/01/06 01/01/08 01/01/10 01/01/12 01/01/14 01/01/16 Date

Avg. CDS slope (dis) Avg. CDS slope (non-dis)

US firms 2006-2016

(28)

3.4 Regression results

The next section carefully describes the main regressions results of this study for three different models. First, the results of the OLS models that test the effect of liquidity and renegotiation factors are presented. Thereafter, the results of a logistic model and probit model testing the effect of these factors on CDS spread changes are shown.

3.4.1 Liquidity factors and default timing

First, I present results for the regressions including liquidity measures and a set of control variables (equation (9)). Column (3) and (4) differ from column (1) and (2) in the exclusion of crisis years (2008-2009). The main variables of interest are , and to a lesser extent and . For the abovementioned interaction variables, a positive effect is expected on CDS slope as higher liquidity ratios should lengthen time to default for distressed firms. For and , the estimates for non-distressed firms, I expect the effect to be marginal or insignificant.

The results are roughly in line with expectations. For distressed firms, a significant positive effect of liquidity factors on the time to default is observed, ranging from 0.172 to 1.724 percentage points for a one unit increase across different liquidity measures and time windows. As expected, liquidity ratios do not matter as much for non-distressed firms: liquidity ratios even have a slightly negative effect over the whole sample period. After exclusion of crisis years increases with more than 15 percent, this being in line with Galil et al. (2014) who find that the ability of structural models to predict CDS spreads was lower during the Global Financial Crisis. Over the non-crisis period, an increase of one standard deviation in ( ) increases CDS slope by 0.13 (0.12) percentage points which is about 1.7 (1.5) percent of average CDS slope. The estimated coefficients are significant at the 10 percent level over both sample periods thus hypothesis 1 is accepted.

Striking at first sight are both the positive coefficient for in column (1) and (2) and the difference in impact between the two liquidity ratios. Apparently, distressed firms on average have slightly higher CDS slopes over the full period. This effect is not found during non-crisis years, as can be seen from column (3) and (4). The difference in

(29)

Table III

Effect of liquidity factors on default timing

This table examines the effect of liquidity factors on CDS slope (defined as 5-year default probability minus 1-year default probability). The sample contains only non-financial firms with matching observations from the Compustat, LPC–Dealscan and Mergent FISD databases. Each column is an OLS regression, with monthly observations at the firm-level. Column (1)-(3) address the full 2006-2016 period. In column (4)-(6) the Global Financial Crisis is excluded. The dependent variable in each regression is the monthly CDS slope. is a dummy variable equal to 1 for firms with Z-score < 1.9 and 0 for firms with Z-score ≥ 2.8. (firms with 1.9 ≤ Z-score < 2.8 are omitted). is defined as current assets minus current liabilities divided by total assets. is the ratio of current assets to total assets. is the sum of debt in current liabilities and long-term debt divided by the market capitalization. is interest expenses plus net income, divided by total assets. is the natural logarithm of total assets. is the 5-year probability of default. is the difference in yield between a 10-year US Treasury Bill and 2-year Treasury Note. Contract control variables include CDS class. All variables except CDS class (ordinal) are winsorized at the 1-99 level. All regressions include year fixed effects, Fama French 12-industry fixed effects and credit rating fixed effects based on Standard&Poor’s long-term ratings. T-statistics are in parentheses. Robust standard errors are used. *, ** and *** represent statistical significance at the 10, 5, and 1% level.

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

CDS slope CDS slope CDS slope CDS slope

Distressed 0.729*** 0.626*** 0.024 -0.144 (7.14) (3.81) (0.42) (-1.26) Liquidity -1.047*** -0.143 (-2.64) (-0.58) Liquidity 2 -0.130* -0.051 (-1.83) (-1.11) Distressed x Liquidity 1.724*** 1.246** (2.68) (2.36) Distressed x Liquidity 2 0.172* 0.180** (1.66) (2.26) Return on assets 27.869*** 27.776*** 7.669*** 7.644*** (7.57) (7.54) (3.74) (3.72) Log Assets 0.013 0.012 0.227*** 0.225*** (0.26) (0.23) (6.51) (6.34) Leverage -1.021*** -1.024*** -0.650*** -0.654*** (-10.15) (-10.18) (-7.22) (-7.22) Default 5 0.781*** 0.781*** 0.922*** 0.922*** (51.69) (51.72) (73.59) (73.50) Yield curve 0.085 0.084 -0.019 -0.022 (0.55) (0.54) (-0.20) (-0.23)

Year fixed effects Yes Yes Yes Yes

Rating fixed effects Yes Yes Yes Yes

Industry fixed effects Yes Yes Yes Yes

Contract control variables Yes Yes Yes Yes

Observations 10102 10102 6952 6952

(30)

liquidity ratios has a logical explanation because current ratio does not take into account current liabilities. If firms have high current assets and high current liabilities, working capital is low and simultaneously current ratio is high.

3.4.2 Renegotiation factors and default timing

This section presents empirical evidence on the effect of renegotiation factors on default timing. The working capital ratio is included as control variable due to its relation to default timing, inclusion of current liabilities and higher significance in Table III. Again, I present results including, column (1)-(3), and excluding, column (4)-(6), crisis years. The main variables of interest are the interaction variables , with being one of the five factors discussed in subsection 3.2.3.2. As shown in Table 1, a negative effect for , and is expected. In contrast, a positive effect for and is predicted. The factors will be discussed in descending order of significance and economic importance.

The most important factor is , column (1) shows a positive coefficient of 2.457 for , meaning an increase of one unit in this variable increases CDS slope by 2.457 percentage points. This effect is significant at the 1 percent level. Effects are similar in other columns, these showing results after exclusion of crisis years and inclusion of other restructuring factors. A one standard deviation increase in this variable increases CDS slope by 0.47 (0.17) over the full sample period (non-crisis years) which is over 6 (1.8) percent of average CDS slope. In contrast, does not affect time to default for non-distressed firms regardless of the sample period or the inclusion of other factors. Apparently, the high relative costs of in-court procedures when intangibles are high increase the likelihood of debt restructuring which is in line with Gilson et al. (1990). Another explanation for the positive coefficient, which contrasts Asquith et al. (1994), is that must be sufficiently high to obstruct asset sales. These asset sales prevent bankruptcy and thus obstruction would lead to a shorter time to default. However, the average fraction intangible assets to total assets is only 0.2 for distressed firms, suggesting that the obstruction argument does not play an important role here. Hypothesis 2 is accepted as the coefficient for is positive and significant at the 1 percent level.

I now turn to the results of the second factor, . Column (2) shows a positive effect of on CDS slope for distressed firms, although insignificant. The results are similar after the inclusion of , as can be seen from column (3). The

Out-of-court restructuring costs and default timing : an examination of US firms

University of Amsterdam (UvA)