The effect of liquidation and continuation values on relative CDS
insurance
Taco Trimbach 6281052
University of Amsterdam Faculty of Economics and Business
Bachelor Thesis February 2014
Table of Content
1. Introduction 2
2. Literature Review 4
3. Hypothesis & Methodology 7
4. Data & Descriptive Statistics 10
5. Results 12
6. Conclusion 19
7. References 20
8. Appendix 21
1. Introduction
Credit derivatives were introduced in the beginning of the 1990s as a way of reducing the risk on credit. In 1997, JP Morgan introduced credit default swaps (CDS) and since then CDS have been the most common form of credit derivatives. The market for credit default swaps saw a tremendous growth and demise in the last decade. The International Swaps and Derivatives Association (ISDA) reported a market growth of $2 trillion in 2001 to $63 trillion in 2007. In 2008, the year that the Financial Crisis erupted, this amount had dropped to $38 trillion. This has led to an ongoing debate about whether or not CDS contributed to the recent financial crisis. Critics called for more transparency and regulation of the credit default swap market and CDS have been under investigation ever since. Even though the literature on credit default swaps is growing, still little is known about these relatively new financial products. A better understanding of these products would improve the financial system and enable more efficient regulation.
A credit default swap is an insurance contract that provides lenders protection against losses arising from a credit event. A credit event is defined as any event whereby a borrower defaults on its debt, whether its bankruptcy or delay of payment. The CDS buyer pays a periodic premium so that, in the event a borrower defaults on its debt, the CDS owner is compensated for its losses. Although seemingly beneficial at first sight, researchers have argued that CDS can give cause to the empty creditor problem (Hu & Black, 2008). The empty creditor problem arises when CDS protected lenders have low incentives to engage in an out-of-court restructuring and instead, force defaulting companies into bankruptcy. By forcing distressed companies into bankruptcy CDS protected lenders receive immediate compensation for the lending activities. This problem might be worse when lenders
“overinsure”, or have so-called negative net economic ownership. When lenders overinsure they take one CDS protection exceeding the amount that can be recovered in out-of-court restructuring (Campello & Matta, 2012). In the event that a company defaults lenders will trigger immediate bankruptcy, since bankruptcy allows them to receive compensation in the amount that exceeds the amount that lenders will receive in case of renegotiation. This comes at a cost of efficiency, where distressed firms are forced into bankruptcy while continuation might be optimal.
While the literature on CDS is consistently growing still many question are left to be answered. What firms benefit the most from CDS contracting? How do firm characteristics affect the level of CDS insurance? And how do liquidation values and continuation values
affect the level of CDS insurance. This paper focusses on finding a relationship between liquidation values and CDS overinsurance. Campello & Matta (2012) argue that
overinsurance is more likely to be found for firms with low continuation values and high liquidation values. Furthermore, Aspeli & Iden (2010) find that firms with higher going concern values have higher probability of restructuring out-of-court. This is related to less CDS insurance. The research question is therefore: What is the effect of liquidation values on the relative amount of CDS insurance for firms with low continuation values.
It adds to existing literature about credit default swaps by providing a framework that shows how firm characteristics affect CDS insurance. Further research on the matter has shown several implications about CDS contracting. Theoretical research by Campello & Matta (2012) as well as Arping (2012) shows that overinsurance is procyclical, that is more pronounced in economic upturns. Furthermore, they find that safer and larger firms are more prone to benefit from CDS contract. This is empirically confirmed by Ashcraft & Santos (2009) and Hirtle (2009). Research has also shown that CDS boost the availability of credit as well as allow firms to maintain higher leverage ratios and debt maturities (Campello & Matta (2012), Hirtle (2009), Saretto & Tookes (2010)). Empirical research about whether the empty creditor problem exists remains two-sided. Padding (2013) and Peristiani & Savino (2011) fail to show evidence in favor of the empty creditor problem, while research by Bendedo et al (2010) and Aspeli & Iden (2010) are unable to find evidence for the empty creditor problem. The research is conducted using a panel dataset over one time period. The time period ranges from 2008 – 2012 and contains 377 companies. Independent variables include
tangibility, and enterprise values as proxies for liquidation and continuation value, respectively. An OLS regression is run to see what effect liquidation values have on the relative amount of CDS insurance. This paper finds that lower continuation values increase the level of CDS insurance, in accordance with the hypothesis. However, it also shows that firms with higher liquidation values, or tangibility, should see less CDS insurance written on their debt. This was inconsistent with the hypothesis.
The remainder of this paper will look as follows. Section 2 will discuss the main literature related to the topic. In section 3 the main hypothesis is discussed as well as the research method used to conduct this research. Section 4 discussed the data and contains descriptive statistics. The empirical results will be reported in section 5. Section 6 will summarize the main results and contain the conclusion.
2. Literature Review
Since their existence in the early 1990s the literature on credit default swaps has been steadily growing. Especially in the last decade the influence on the financial markets has been noticed giving cause for the rising number of research papers on this subject. In this section the main literature on credit default swaps, “overinsurance”, and the empty creditor problem is
discussed. It is divided into three parts. The first part is used to define the main theory on CDS. The second part is dedicated to CDS overinsurance and the empty creditor problem. In the last section the remaining findings of CDS research are described.
I. Credit Default Swaps
Credit default swaps allow lenders to hedge themselves against potential losses incurred by them resulting from a credit event. By paying a premium to CDS sellers, CDS buyers receive compensation in the event that a borrower fails to repay the debt owed (Bolton & Oehmke, 2010). Because of CDS protection the bargaining position of lenders will be improved (Arping, 2012). Termination threats by CDS protected lenders can be credibly backed up which in turn solves the commitment problem. Borrowers are incentivized to exert high effort to increase the likelihood that the investment will succeed, and decrease the likelihood that lenders will exercise the termination threat. However, it could be that lenders abuse their bargaining power in order to increase their “piece of the pie” in debt
renegotiations. Borrowers might then refuse to exert any effort at all, in turn making the lender refusing to invest (Arping, 2012).
Hu & Black (2008) state that CDS insurance might lead to empty creditors. The definition of empty creditors stands for investors with control right but with negative economic ownership, or overinsurance. With the invention of CDS investors can separate their control rights from their cash-flow right. Not anymore do lenders depend on the
borrowers for their cash flows. Hu & Black (2008) argue that this can give cause to the empty creditor problem. A problem were solvent companies are forced into bankruptcy, where restructuring might be optimal. Bolton & Oehmke (2010) agree with this, as they argue that CDS have changed to relationship between borrowers and lender since the introduction of CDS. Lenders might refuse efficient renegotiation and force defaulting firms into bankruptcy.
II. Overinsurance & the Empty Creditor Problem
Overinsurance, or negative net economic ownership, occurs when lenders buy CDS protection beyond the level of renegotiation proceeds. Overinsurance increases the
probability that that defaulting firms will be forced into liquidation or bankruptcy, even when continuation would be efficient. Bolton & Oehmke (2010) say that lenders will generally overinsure when they choose the level of insurance. Their paper develops a model that proposes restrictions on CDS overinsurance. Campello & Matta (2012) show that by imposing restrictions on CDS insurance a firm credit capacity could shrink when the most need it. Another implication of their model is that overinsurance is procyclical.
In their theoretical paper, Campello & Matta (2012) develop a model for CDS contracting incorporating the state of the economy. They show that CDS overinsurance is procyclical, that is more pronounced during times of economic growth. During times of economic growth investments have a higher probability of succeeding, and the likelihood that borrowers are in distress decreases. Therefor the income of the lender will mostly consist of steady debt repayments. In order to maximize these payments lenders will take on negative net economic ownership (Campello & Matta, 2012). By doing this they incentivize borrowers to exert high effort and keep them from defaulting strategically. Strategic default is a process whereby borrowers try to renegotiate the debt repayments downwards because they can abscond part of the investment proceeds (Bolton & Oehmke, 2010). By exerting high effort borrowers increase the probability that an investment will succeed.
In economic downturns the probability that an investment will succeed is lower and the likelihood that a borrower is in distress increases. Whether or not investor overinsures depends on the continuation value and liquidation value. If continuation values exceed liquidation values then lenders are more likely to have zero net economic ownership
(Campello & Matta, 2012). With zero net economic ownership the amount of CDS insurance equals the amount of debt. Zero net economic ownership maximizes the debt repayments and increase bargaining power over the renegotiation value when investments fail. (Campello & Matta, 2012). If liquidation values exceed continuation values, renegotiation payoffs drop and lenders are more prone to overinsure.
Because overinsurance and the empty creditor problem have a positive relation this leads to the conclusion that the empty creditor problem is procyclical as well (Campello & Matta, 2012). The empty creditor problem arises when defaulting firms are forced into bankruptcy when continuation might be optimal, because bankruptcy payoffs for lenders are
higher than continuation payoffs (Hu & Black, 2008). Empirical research however is contradicting. Peristiani & Savino (2011) as well as Padding (2013) both found a rise in bankruptcy for firms with CDS written on their debt. Padding (2013) used new tax
regulations imposed by the Internal Revenue Service (IRS) as a natural experiment to test whether the empty creditor problems exists. The theory behind this that the new tax
regulations made out-of-court restructuring easier. This would increase continuation values for firms, thereby reducing the incentive to overinsure for lenders. Padding (2013) was able to show that relative amounts of CDS insurance dropped after the change in legislation, thereby showing that the empty creditor problem exists. However, after controlling for industry fixed effects evidence for the empty creditor problem weakened.
Empirical studies by Aspeli & Iden (2010), Bendedo et al. (2010), and Mengle (2009) are unable to show significant evidence that negative net economic ownership leads to more inefficient bankruptcies for defaulting firms. Bendedo et al. (2010) compares whether the filing bankruptcies increases for firms with CDS written on their debt relative to firms that don’t have CDS written on their debt. They fail to show a significant difference in
bankruptcies between the CDS entities and the non-CDS entities. Furthermore, they show that firms that restructure out-of-court have lower leverage, lower debt ratios and a lower proportion of secured debt. Aspeli & Iden (2010) study a sample of 218 debt restructurings in the United States between 1995 and 2010. They fail to show that the presence of CDS
significantly influences the choice between restructuring and bankruptcy in favor of bankruptcy. They did find that firms with higher going-concern values have a higher
probability that their debt will be restructured out-of-court. This would then lead to less CDS insurance and is in accordance with the hypothesis. Mengle (2009) argues that the empty creditor problem might well be a problem but that due to the lack transparency and compelling examples he fails to show this in practice.
III. Remaining Findings
Duffee & Zhou (2001) argue that CDS swaps have positive as well as negative effects on credit risk. In their theoretical paper they examine the effect CDS markets have on the financial sector. They find that CDS allow banks to reduce the risk on their loans, making it less likely that defaulting loans would cause distress. However, they also find that CDS can cause secondary markets to breakdown. Therefore they conclude that the risk-sharing
flexibility is not enough to claim that credit derivatives are beneficial to financial institutions. Further empirical research on the matter is suggested by them.
Saretto & Tookes (2010) find that firms with CDS written on their debt are able to keep higher corporate leverage as well as long debt maturities. They even find that the effect of CDS on leverage and maturity is highest in times where credit constraints are most
binding, for example during the recent financial crisis. They did this by comparing firm with CDS written on their debt and without CDS written on their debt. The results show that the introduction of CDS increases leverage by around 25 percent. When examining this effect using year fixed effects they find that this effect is greater in 2001 and 2008. These are periods when the economy was hit by recessions. Hirtle (2009) questions whether banks used financial derivatives as means to reduce risk or as a mean of increase lending. She finds that CDS did not reduce banks risk but rather increase the supply of credit by banks, but that it only holds to a limited extent. She finds that this holds mostly for larger term loans but that this increases the CDS spreads.
Ashcraft & Santos (2009) look at whether or not CDS have lowered the cost of debt. Their model finds that for risky and less transparent firms the cost of debt actually increases, but that for safer and more transparent firms the cost of debt does decrease. They come to the conclusion that, contrary to beliefs, CDS protection is more beneficial for safer and larger firms. This is confirmed by Arping (2012), Campello & Matta (2012), and Hirtle (2009). Arping (2012) argues that for firms where debt renegotiation is difficult, which are usually larger firms, profit the most from CDS protection. Furthermore Arping (2012) shows that firms with low levels of asset tangibility and lower levels of leverage tend to benefit more from CDS contracts. Hence, the conclusion is that safer and larger firms benefit the most from CDS contracting.
3. Methodology & Hypothesis
This section seeks to describe the hypothesis tested in part I and the methodology and statistical models used to conduct this empirical research in part II.
I. Hypotheses
Liquidation values can be explained as the value of the firm when all its assets are liquidated. In sense, this is what creditors would receive upon liquidation of a firm, or what they would recover upon liquidation (Bendedo et al, 2010). When a firms defaults creditors can renegotiate the debt out-of-court or in-court. CDS payment is only triggered when debt is renegotiated in-court. When debt is renegotiated out-of-court bondholders receive an equity
share in the firms continuation value (Garrido, 2012). When a debt is renegotiated in-court, the bondholders receive the liquidation value of assets. So whether or not lenders overinsure, thereby forcing defaulting companies into bankruptcy, depends on the continuation value and the liquidation value.
In their theoretical paper Campello & Matta (2012) predict that overinsurance is more likely to be found at firms with low continuation values, high liquidation values and high renegotiation costs. This paper will try to test what the effect is of liquidation values on the level of CDS insurance. In what way liquidation values affect the level of CDS insurance depends on the continuation value. If a firm has high continuation values, then a higher liquidation values increases the out-of-court payoffs for the lenders. Lenders benefit more from restructuring the debt privately and are therefore less likely to overinsure. If however, firms have a low continuation value, then an increase in liquidation values increases the payoffs resulting from bankruptcy. Now lenders benefit more from in-court restructuring and are therefore more likely to overinsure (Campello & Matta, 2012). The hypothesis is: Higher liquidation values lead to higher CDS insurance for firms with low continuation values. To test this hypothesis proxies are needed for liquidation values and continuation values. The proxy that will be used for liquidation value is tangibility (Braun (2002), Rajan & Zingales (1995)). Bendedo et al (2010) reports a positive relationship between tangibility and recovery rates, or liquidation value. Higher tangibility increases what lenders get upon liquidation of a firm and lenders will thus have a greater incentive to overinsure. However, tangibility can also have an effect on the continuation value of the firm. A higher tangibility can increase the continuation value of a firm, because it makes it harder for managers to abscond with resources. This increases the amount that can be bargained over between lenders and borrowers. Therefore it reduces the incentive of lenders to overinsure (Padding 2013). This effect is less likely to happen if firms have low continuation value (Campello & Matta, 2012). Thus, the question is whether higher tangibility increases the incentive to overinsure for firms with low continuation values.
Enterprise value is used as a proxy for continuation value. Enterprise value is calculated as the sum of market value of equity and market value of debt minus cash. It attempts to measure the value of the ongoing operations of a firm. It therefore incorporates all the intangible assets, such as goodwill and future revenues into the firm’s value. According to the literature, a higher continuation value should decrease the level of CDS insurance
(Campello & Matta, 2012).
Taking all previous discussed effect into account the following hypothesis is formulated:
Hypothesis: Higher tangibility increases the relative amount of CDS insurance for firms
with low enterprise values
II. Methodology
To investigate the effect of tangibility on the level of insurance an OLS regression is performed. The dependent variable will be the relative amount of CDS insurance denoted in RCDS. This is determined by taking the net notional amounts of CDS insurance over the firms continuation value (Campello & Matta, 2012). Since enterprise value is used as a proxy for continuation value the relative amount of CDS insurance is defined as net notional
amounts over enterprise value. Tangibility and enterprise value are added as explanatory variable. To test the effect of tangibility on relative amount of CDS insurance for firms with low enterprise values an interaction variable is added. This interaction variable is generated by multiplying tangibility with a dummy variable for the 10th percentile, 25th percentile and the 50th percentile of relative enterprise value, respectively. Furthermore, several control variables are added to the model.
This lead to the following OLS regression;
RCDSit = β0 + β1 Tangibility*X-percentile + β2 DummyX-percentile + β3 Tangibility + β4Enterprisevalue + βi Control Variables +
ε
tTangibility is used as a proxy for liquidation value and is added as an explanatory variable. Based on the literature and research the expected sign for this coefficient is expected to be positive holding constant the firm’s enterprise value. Higher tangibility would lead to higher relative CDS positions (Campello & Matta, 2012). Tangibility is defined as the ratio of net property, plant and equipment to total assets (Bendedo et al, 2010). As stated before
enterprise value is used as a proxy for continuation value. It is defined as the market value of equity plus the market value of debt minus cash. As predicted by the literature a higher enterprise value should lead to less CDS insurance (Campello & Matta, 2012). The coefficient of this variable is therefore expected to be negative.
Control variables are added in order to control for omitted variable bias. Empirical research conducted by Padding (2010) included several control variables and their expected effect on relative CDS positions. These variables based on the literature might have a
significant influence on the level of insurance with CDS. The first control variable that is added is profitability. Profitability is measured as EBIT over total assets (Padding, 2013). It is said to have a positive effect on liquidation values and therefore tangibility (Padding (2013), Bendedo et al (2010)). Higher profitability leads to higher continuation values. Investors are therefore less likely to overinsure. This variable is therefore expected to be negative.
The second control variable that is added is firm size. Research by Ashcraft & Santos (2009), (Arping (2012), Campello & Matta (2012), and Hirtle (2009) has shown that CDS insurance is more likely to benefit larger firms. However, firm size is expected to be positively related to continuation value. Firm’s size is also expected to have a negative relationship with bankruptcy (Danis, 2012). Higher firms should therefore see less CDS written on their debt. Hence, the coefficient is negative. Firm size is defined as the logarithm of the firm’s assets (Padding, 2013).
The third variable that is added is the Altman z-score. The Altman z-score is used to predict the likelihood that a firm will go bankrupt within two years (Altman, 2000). A z-score of below 1.81 means almost certain bankruptcy for a firm and a z-score of above 2.99 means that the firms are generally safe. A z-score between those limits is regarded as grey-area. Since firms with a lower z-score are more likely to go bankrupt the coefficient for the variable is expected to be negative, since lenders will tend to overinsure more. Another control variable that is added is Tobin’s Q. Bendedo et al (2010) argue that this should be taken into account for the effect of growth prospects of the firm’s assets. A higher Tobin’s Q means more growth prospects for a firm and increases the continuation value. The coefficient for Tobin’s Q is expected to be negative.
Lastly, the variable leverage is added. Aspeli & Iden (2010) argue that highly leveraged firms are more prone to bankruptcy. Bendedo et al (2010) also argue that higher leverage makes debt restructuring harder. Therefore, lenders are more likely to take on more CDS insurance. The coefficient is expected to be positive. Leverage is measured as total debt over total assets (Moor, 2013).
4. Data & Descriptive Statistics
The data is provided by dr. Mata, R. of the University of Amsterdam. It contains data on net notional amounts of CDS, total firms assets, total debt, liabilities as well as other firm characteristics. Using these data new variables are composed as describes in the previous part Net notional amounts of CDS can be found at the Deposit Trust and Clearing Corporation
(DTCC) website. They provide weekly data on gross and net notional amounts of CDS for the thousands largest reference entities. Data for these entities is provided for the time period 2008-2012. The dataset has been modified so that it does not contain state owned firms. Furthermore, financial institutions are omitted from the sample because their firm
characteristics are too different from that of other firms. This leaves a dataset of 377 firms. Tangibility, total assets, total debt and enterprise value tend to stay the same over the same fiscal year, therefore adjustments on the dataset are made. Since net notional amounts are weekly provided by the DTCC, net notional averages have to be computed per fiscal year per firms. The same is done for the Altman z-score, Tobin’s Q and market value of the firm, since these variables tend to change in the short run. Subsequently, duplicate observations are dropped for each year. This leaves one observation per fiscal year per firm in the dataset. When looking at descriptive statistics comparing the relative positions of CDS insurance to the fiscal year we can see that relative CDS insurance was highest in the year 2008. This was at the height of the recent financial crisis. Relative CDs insurance dropped every year since then.
Table 1: Summary Statistics RCDs per Fiscal Year
Fiscal Year RCDS
Fiscal Year Freq. Mean SD 2008 318 .10586465 .1481914 2009 358 .09343938 .1443178 2010 363 .07929306 .1273665 2011 365 .07331303 .1206284 2012 310 .06907288 .1218586
The table for summary statistics per industry can be found in the appendix. When looking at industry specific means we see that the sample contains the most observations for Communications-, Oil & Gas- and Electric Companies. Furthermore it shows that relative CDS insurance was highest for construction companies and home-furniture companies. Transportation companies and automotive repair companies had the highest levels of
tangibility and hospitality and eating & drinking companies had the highest relative enterprise value.
5. Results
This section contains the main results from the research. It is divided up into two parts. The first part describes univariate analysis and the variable correlation. The second part describes and explains the multivariate analysis. In this part the results are also checked for robustness by controlling for industry fixed and time fixed effects. The White test for heteroscedasticity came out significant. Therefore all further regressions are done using robust standard errors.
I. Univariate Analysis & Correlations
This subsection describes the univariate analysis of the variables used in the regression discussed in 3.II. It measures the effect of the single variables on the relative amount of CDS insurance. Table 2 shows the results of the univariate analysis for each variable. This subsection also contains the correlations between the variables used in the regression.
Table 2: Univariate Analysis on Relative CDS Positions
*, **, *** denote statistical significance difference at the 10%, 5% and 1% levels, respectively.
In the first three regressions it is shown that tangibility has a positive effect on relative CDS positions for the lowest 10 percentile, 25 percentile and 50 percentile of relative
Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) RCDS RCDS RCDS RCDS RCDS RCDS RCDS RCDS RCDS RCDS Tangible*10pctile REV 0.1045* (0.060) Tangible*25pctile REV 0.0673*** (0.018) Tangible*50pctile REV 0.0567*** (0.011) Tangibility -0.0184* (0.011)
Relative Enterprise value -0.0648*** (0.007) Profitability -0.6128*** (0.084) Firm Size -0.0861*** (0.006) Altman z-score -0.0050 (0.003) Tobin's Q -0.0744*** (0.007) Leverage 0.0992*** (0.024) Constant 0.1098*** 0.1037*** 0.0956*** 0.1263*** 0.1784*** 0.1707*** 0.9600*** 0.1198*** 0.2012*** 0.0853*** (0.005) (0.005) (0.005) (0.009) (0.010) (0.010) (0.058) (0.009) (0.011) (0.009) Observations 1,429 1,429 1,429 1,429 1,445 1,445 1,445 1,34 1,264 1,445 R-squared 0.008 0.016 0.019 0.002 0.037 0.063 0.306 0.001 0.032 0.007 12
enterprise value, respectively. While the effect of tangibility decreases as a larger percentile is used all three coefficients are statistically significant, the 10th percentile at a 10% level and both the 25th and 50th percentile at a 1% level. The effect of the coefficients of the first three regressions is in line with the hypothesis. However in the fourth regression, the coefficient for tangibility is negative at a 10% significance level. This means that an increase in ratio of tangibility of 1% leads to a decrease of in relative CDS positions of 0.0184%. This result is inconsistent with the hypothesis. One explanation for the negative coefficient could be that tangibility increases what lenders receive in bankruptcy. Therefor there is less of an incentive to insure themselves with CDS (Padding, 2013). However, since this is univariate regressions we ignore the fact that the coefficients of the percentiles and that of tangibility are affected by other independent variables as well
The fifth regression show the effect of relative enterprise value on relative CDS positions. This coefficient is negative and significant at a 1% level. Relative CDS insurance increases with 0.0648% for an increase of 1% of relative enterprise value. Enterprise value can be considered as continuation value so this result is consistent with the hypothesis. A lower continuation value for a firm increases the lenders incentive to overinsure. Profitability is significant negative as well. Higher profitability increases the value of the firms and therefore the continuation value, hence lower levels of CDS insurance.
The negative coefficient for firm size is consistent with the expectations described in the methodology and consistent with the findings of Padding (2013). Higher firms size increase continuation value and therefore decreases the incentive for lenders to overinsure. It is significant at a 1% level. The Altman Z-score is negative as expected but this result is not significant. Therefore no conclusion about the Altman z-score can be made.
Tobin’s Q has a negative sign for its coefficient. This is consistent with the
methodology. Bendedo et al (2010) find that this coefficient has a positive effect on recovery rates and a positive effect on continuation value. A higher Tobin’s Q increases a firm’s continuation values so lenders payoffs are less likely to come from bankruptcy. CDS
insurance decreases. Lastly, leverage has a positive coefficient. This is expected since higher leverage increases the probability of bankruptcy and reduces the continuation value (Aspeli & Iden, 2010). Therefore, lenders take on more CDS insurance.
Table 3 shows the correlation matrix of the variables. Interesting to see is that the three percentile variables have highly significant negative correlation with relative enterprise value. Profitability and Altman z-score have high correlation with relative enterprise value. This is consistent with the literature both variables increase the continuation value of the
firm. But in order to prevent multicollinearity these explanatory variables are dropped in multivariate analysis. Firm size is negatively correlated to all percentile variables, tangibility and relative enterprise value. This could be due to the fact that the relative amount of tangible assets and enterprise value decrease as firm size increases. Z-score is negatively correlated since with tangibility and the percentile variables since firms with a higher z-score have a decreased chance of bankruptcy. This is mostly found at firm with low tangible but high intangible assets (Aspeli & Iden (2010).
Table 3: Correlations Matrix
*, **, *** denote statistical significance difference at the 10%, 5% and 1% levels, respectively.
II. Multivariate Analysis
The previous section showed the results of the univariate regressions. Some of the findings were consistent with the predictions and others were not. This subsection discussed the multivariate analysis of the regression described in section 3.II. This is done since univariate doesn’t show the influence of a variable, given other firm characteristics. In order to test the effect of the variables, ceteris paribus, a multivariate regression is performed. Table 4 shows the results of the multivariate analysis. The explanatory variables profitability and Altman z-score are omitted from the analysis in order to prevent multicollinearity. In the first regression the interaction variable of the 10th percentile, the dummy, tangibility and relative enterprise value are regressed on the dependent variable. The dummy variable for the 10th percentile of relative enterprise value is added to prevent omitted variable bias. In the first regression the interaction variable is positive and statistically significant at a 5% level, stating that higher tangibility leads to higher relative CDS insurance for the lowest
Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (1) Tangible*10pctile REV 1.000 (2) Tangible*25pctile REV 0.253 1.000 (0.000)*** (3) Tangible*50pctile REV 0.114 0.493 1.000 (0.000)*** (0.000)*** (4) Tangibility 0.001 0.140 0.352 1.000 (0.977) (0.000)*** (0.000)***
(5) Relative Enterprise value -0.168 -0.392 -0.523 -0.059 1.000 (0.000)*** (0.000)*** (0.000)*** (0.046)** (6) Profitability -0.143 -0.274 -0.396 -0.278 0.652 1.000 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (7) Firm Size -0.077 -0.081 -0.092 -0.156 0.039 0.080 1.000 (0.009)*** (0.005)*** (0.002)*** (0.000)*** (0.179) (0.007)*** (8) Altman z-score -0.056 -0.205 -0.369 -0.188 0.611 0.619 -0.108 1.000 (0.058)* (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (9) Tobin's Q -0.078 -0.154 -0.211 -0.228 0.303 0.218 0.490 0.078 1.000 (0.008)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.008)*** (10) Leverage -0.063 -0.079 0.062 0.293 0.074 -0.140 -0.239 -0.387 -0.201 1.000 (0.031)** (0.007)*** (0.035)** (0.000)*** (0.012)** (0.000)*** (0.000)*** (0.000)*** (0.000)*** 14
10% firms of relative enterprise value. This is in accordance with the hypothesis that higher liquidation values lead to higher CDS insurance for firms with low continuation values. However we neglected the effect of the control variables in the model.
Table 4: Multivariate Analysis on Relative CDS Positions
*, **, *** denote statistical significance difference at the 10%, 5% and 1% levels, respectively.
In the second multivariate regression control variables are added to the four variables in the first regression. Interesting to see is that the interaction variable is still positive but no longer statistically significant. Therefore no conclusions can be made about the variable. It could be that there is no direct relationship between liquidation and continuation values on relative CDS positions.Notice that the variable tangibility is negative and statistically significant. It tells us that higher tangibility decreases the level of CDS insurance. Padding (2013) and Bendedo et al (2010) give as an explanation that higher tangibility increase what lenders get in bankruptcy. Since lender would receive more in bankruptcy, less CDS
Variables (1) (2) (3) (4) (5) (6)
RCDS RCDS RCDS RCDS RCDS RCDS
Tangible*10pctile REV 0.2139** 0.2395 (0.098) (0.268) Dummy 10th percentile REV -0.1288*** 0.0904 (0.033) (0.085)
Tangible*25pctile REV 0.0885*** -0.0655
(0.029) (0.047)
Dummy 25th percentile REV -0.0552** 0.1015***
(0.023) (0.036)
Tangible*50pctile REV 0.0321 -0.0240
(0.022) (0.021)
Dummy 50th percentile REV 0.0121 0.0520***
(0.022) (0.019) Tangibility -0.0309*** -0.0444*** -0.0367*** -0.0389*** -0.0303** -0.0372***
(0.011) (0.010) (0.012) (0.009) (0.012) (0.009) Relative Enterprise value -0.0752*** -0.0928*** -0.0668*** -0.0865*** -0.0450*** -0.0820***
(0.007) (0.007) (0.007) (0.007) (0.008) (0.008) Firm Size -0.1166*** -0.1186*** -0.1174*** (0.008) (0.008) (0.008) Tobin's Q 0.0832*** 0.0848*** 0.0813*** (0.009) (0.009) (0.009) Leverage -0.0453* -0.0472* -0.0616** (0.025) (0.025) (0.026) Constant 0.2112*** 1.2901*** 0.2030*** 1.2927*** 0.1611*** 1.2778*** (0.015) (0.081) (0.016) (0.082) (0.019) (0.086) Observations 1,429 1,253 1,429 1,253 1,429 1,253 R-squared 0.054 0.450 0.048 0.435 0.047 0.430 15
insurance is needed to be fully reimbursed for their investment. The result of relative
enterprise value, however, is consistent with the hypothesis and the literature. Its coefficient is highly significant and the relationship with relative amount of CDS insurance in negative. This means that firms with low enterprise values should expect more CDS written on their debt. This is consistent with the research of Campello & Matta (2012), stating higher continuation values lead to less CDS insurance.
Other remarkable results include Tobin’s Q and leverage. Tobin’s Q was expected to have a negative coefficient since it would increase the continuation value of a firm (Padding, 2013). The positive coefficient of Tobin’s Q is hard to interpret because of limited literature on CDS. The other variable that is remarkable is leverage. Leverage is expected to have a positive effect on bankruptcy and therefore a positive effect on CDS insurance (Aspeli & Iden, 2010). Nevertheless, the coefficient for leverage is significantly negative, meaning that higher leverage will lead to less CDS insurance. On explanation that could be given is by Jensen (1989). He argues that highly levered firms have high continuation values compared to their liquidation values. These firms will have to lose more in bankruptcy and bankruptcy is therefore less likely. This could be an explanation for why firms with higher leverage have less CDS insurance.
In the third regression the interaction variable of the 25th percentile is regressed on the relative CDS positions. This effect is, although decreasing, still positive and significant at a 1% level. This result is in line with the hypothesis. However, once the control variables are added this result changes. The effect of the interaction variable is no longer significant and its sign is negative. This would mean that for the 25% lowest relative enterprise value firms, a higher tangibility would lead to less CDS insurance. However, since this result is not significant no conclusions can be made about this. The control variables in the fourth regressions remain fairly unchanged.
In the fifth and sixth regressions the interaction variable for the 50% lowest relative enterprise value firms is used. In both regressions the interaction variable is insignificant. It shows that the effect of tangibility on relative CDS positions diminishes as relative enterprise values increase. This is in accordance with the hypothesis since it would mean that for firms with higher continuation values the effect of tangibility on relative CDS positions decreases. In the fifth and sixth regression the control variables remain relatively similar to those in the previous regressions.
While the results provide some evidence for the hypothesis, showing that the effect after of tangibility is higher and significant for firms that have low enterprise values, or continuation values. However, after adding the control variables this evidence seems to diminish. The results fail to provide evidence for the hypothesis of this research. It could be that there is simply no relationship between liquidation values and low continuation values on relative CDS insurance. Another explanation could be that the effects are more industry specific. For example, tangibility is expected to be higher in the manufacturing industry than in the services industry. This research was able to show that there is a relationship between a firm’s continuation value and relative amount of CDS insurance. Namely, firms with higher continuation values should see less CDS insurance written on their debt. This result is consistent with the research of Campello & Matta (2012). It is also consistent with the research of Aspeli & Iden, who show that firms with higher going concern values are more likely to renegotiate debt out-of-court.
III. Robustness
To check the robustness of the results of this research industry fixed effects and time fixed effects are added. This is done since the regressions controls for random effects between industries and time. Robustness checks are performed on all three interaction
variables and the control variables. Table 5 shows the regressions results controlling industry fixed and time effects.
Industry are divided up into four parts according to their SIC codes. SIC codes 01-09, 10-19, 20-30, 40-99 are given the values 1, 2, 3, 4 respectively. They represent the
agriculture, the mining, the manufacturing and the services industry respectively. Since no observation were made in the agriculture industry this is not relevant for the regression. Industry 2 is omitted from the regression to prevent perfect multicollinearity. The same is done for the time effects, where the year 2008 is omitted.
While controlling for industry fixed effects and time fixed effects the output of the regression shows little difference with the multivariate analysis. All three interaction variables remain insignificant. It makes the evidence for the hypothesis weaker. It could be that there is no direct relationship between liquidation values and continuation values on relative amounts of CDS insurance.
Table 5: Regression controlling for Industry and Time Fixed Effects
*, **, *** denote statistical significance difference at the 10%, 5% and 1% levels, respectively.
Variables (1) (2) (3)
RCDS RCDS RCDS Tangible*10pctile REV 0.2286
(0.266) Bottom 10% Relative Enterprise Value 0.0868 (0.085)
Tangible*25pctile REV -0.0726 (0.046) Bottom 25% Relative Enterprise Value 0.0995***
(0.035)
Tangible*50pctile REV -0.0300
(0.021) Bottom 50% Relative Enterprise Value 0.0511**
(0.020) Tangibility -0.0411*** -0.0346*** -0.0309**
(0.011) (0.011) (0.012) Relative Enterprise value -0.0904*** -0.0857*** -0.0821***
(0.008) (0.007) (0.009) Firm Size -0.1151*** -0.1171*** -0.1159*** (0.008) (0.008) (0.008) Tobin's Q 0.0827*** 0.0843*** 0.0811*** (0.009) (0.009) (0.009) Leverage -0.0388 -0.0416* -0.0537** (0.025) (0.025) (0.026) 3. Manufacturing Industry 0.0124 0.0122 0.0152 (0.012) (0.012) (0.012) 4. Services Industry 0.0002 -0.0012 0.0013 (0.010) (0.010) (0.010) 2009.Year -0.0074 -0.0076 -0.0063 (0.016) (0.016) (0.016) 2010.Year -0.0325** -0.0351*** -0.0353*** (0.013) (0.013) (0.013) 2011.Year -0.0541*** -0.0557*** -0.0553*** (0.012) (0.012) (0.013) 2012.Year -0.0380*** -0.0373*** -0.0402*** (0.013) (0.013) (0.013) Constant 1.2918*** 1.2979*** 1.2798*** (0.079) (0.080) (0.085) Observations 1,253 1,253 1,253 R-squared 0.462 0.448 0.443 18
6. Conclusion
This study analyzes the effect of recovery rates and firm continuation values on the level of CDS insurance using a dataset of 377 companies over the timeframe 2008-2010. Previous literature predicted that the presence of CDS overinsurance is more likely to occur for firms with high liquidation values and low continuation values (Campello & Matta, 2012). The proxies for liquidation-, and continuation value that are used are tangibility and enterprise value, respectively. According to the literature when lenders overinsure they pre-commit to forcing defaulting firms into bankruptcy. This problem is labeled as the empty credit problem (Hu & Black, 2008). The incentive to overinsure is enhanced when bankruptcy payoffs are high and going-concern values are low.
The results of the research are presented in section 5. This research was able to show that higher liquidation values lead to higher CDS insurance for firms with low continuation values. However, after adding control variables this result was no longer significant. It did however show that firms with lower continuation values should see more CDS insurance written on their debt, regardless of their liquidation values. This is consistent with the literature on the matter. After controlling for industry fixed and time fixed effect the results remained insignificant. Therefor this research was unable that draw clear conclusions about the effect of liquidation values and continuation values on relative CDS insurance.
While the results were not completely consistent with the hypothesis further research on CDS insurance and firm characteristics could provide more insight on why people take on CDS insurance. This research could be done by increasing the sample size or testing the effect on CDS insurance per industry on CDS insurance.
7. References
Altman, E. I., 2000, “Predicting Financial Distress of Companies: Revisiting the Z-score and Zeta models”, Working Paper, Stern School of Business, 1-54.
Arping, S., 2004, “Credit Protection and Lending Relationships," Working Paper, University of Amsterdam.
Ashcraft, A., and J. Santos, 2009, “Has the CDS Market Lowered the Cost of Corporate Debt?" Journal of Monetary Economics 56, 514 – 523.
Aspeli, N., and K. Iden, 2010, “The Empty Creditor Hypothesis: An Empirical Study of the Effects of Credit Insurance on the Choice between Bankruptcy and Private Restructuring," Master Thesis, Norwegian School of Economics and Business Administration.
Bedendo, M., L. Cathcart, and L. El-Jahel, 2010, “In- and Out-of-Court Debt Restructuring in the Presence of Credit Default Swaps," Working Paper, Imperial College.
Bolton, P., and M. Oehmke, 2011, “Credit Default Swaps and The Empty Creditor Problem," Review of Financial Studies 24, 2617 – 2655.
Braun, M., 2002, “Financial Contractibility and Asset Hardness: Industrial Composition and Growth”, Working paper, University of California, Los Angeles.
Campello, M. and R. Matta, 2012, “Credit Default Swaps, Firm Financing and the Economy," Working Paper, Cornell University.
Danis, A., 2012, “Do Empty Creditors Matter? Evidence from distressed exchange offers”, Working paper, Vienna Graduate School of Finance.
Duffee, G. R., and C. Zhou, 2001, “Credit Derivatives in Banking: Useful Tools for Managing Risk?" Journal of Monetary Economics 48, 25-54.
Garrido, J.M., 2012, “Out-of-Court Debt Restructuring”, World Bank Studies.
Hirtle, B., 2009, “Credit Derivatives and Bank Credit Supply," Journal of Financial Intermediation 18, 125 – 150.
Hu, H., and B. Black, 2008, “Debt, Equity, and Hybrid Decoupling: Governance and Systemic Risk Implications," European Financial Management 14, 663-709.
Jensen, M.C., 1989, “Eclipse of the Public Corporation”, Harvard Business Review, 61-74.
Mengle, D., 2009, “The Empty Creditor Hypothesis," International Swaps and Derivatives Association (ISDA) Research Notes No. 3.
Moor, J. H., 2013, “The Effect of the Business Cycle on the Empty Creditor Problem”, Master Thesis, University of Amsterdam
Padding, E., 2013, “The Empty Creditor Problem; Evidence from the United States”, Master Thesis, University of Amsterdam.
Peristiani, S., and V. Savino, 2011, “Are Credit Default Swaps associated with Higher Corporate Defaults?” Working Paper, Federal Reserve Bank of New York.
Rajan, R., and L. Zingales, 1995, “What Do We Know About Capital Structure? Some Evidence from International Data”, Journal of Finance 50, 1421 – 1460.
Saretto, A., and H. Tookes, 2013, “Corporate Leverage, Debt Maturity and Credit Default Swaps: The Role of Credit Supply," Review of Financial Studies 26, 1190 – 1247.
8. Appendix
I. Summary Statistics
Table 2: Summary Statistics
SIC Code Freq. RCDS Tangibility Relative Enterprise
Mean SD Mean SD Mean SD
10 31 0.029 0.028 0.669 0.208 0.939 0.669 12 10 0.054 0.061 0.875 0.107 1.072 0.407 13 99 0.053 0.074 1.177 0.575 1.043 0.447 14 7 0.049 0.011 0.965 0.201 1.240 0.331 15 48 0.399 0.155 0.038 0.017 0.883 0.150 20 68 0.069 0.066 0.515 0.211 1.391 0.684 21 19 0.048 0.037 0.197 0.089 1.617 1.138 22 5 0.197 0.043 0.566 0.029 0.828 0.054 23 11 0.610 0.612 0.318 0.123 1.119 0.497 24 10 0.671 0.699 1.090 0.297 0.831 0.292 25 4 0.057 0.027 0.390 0.015 0.745 0.126 26 35 0.145 0.128 1.019 0.340 1.102 0.522 27 33 0.349 0.198 0.477 0.266 0.777 0.321 28 120 0.116 0.198 0.565 0.324 1.308 0.815 29 66 0.041 0.044 0.629 0.370 0.562 0.386 30 15 0.239 0.181 0.748 0.307 0.700 0.271 32 7 0.039 0.015 0.685 0.078 0.892 0.115 33 14 0.063 0.020 0.844 0.176 0.780 0.406 34 8 0.186 0.116 0.384 0.143 0.979 0.160 35 55 0.109 0.211 0.398 0.198 0.952 0.398 36 54 0.107 0.116 0.532 0.425 0.759 0.607 37 47 0.053 0.056 0.392 0.203 0.727 0.363 21
38 33 0.082 0.113 0.325 0.296 0.989 0.429 39 10 0.091 0.027 0.263 0.099 1.526 0.214 40 20 0.034 0.018 1.189 0.056 1.096 0.187 42 18 0.130 0.128 0.945 0.217 1.121 0.420 44 10 0.108 0.060 1.083 0.054 0.751 0.115 45 9 0.127 0.102 1.099 0.093 0.808 0.146 47 10 0.221 0.098 0.628 0.538 0.864 0.208 48 138 0.074 0.094 0.838 0.520 0.729 0.414 49 158 0.041 0.029 0.868 0.264 0.794 0.200 50 8 0.090 0.087 0.451 0.202 0.586 0.243 51 16 0.155 0.148 0.163 0.118 0.648 0.121 52 10 0.031 0.011 0.842 0.210 1.331 0.302 53 44 0.114 0.105 0.697 0.254 0.890 0.340 54 19 0.128 0.044 0.974 0.249 0.719 0.273 56 15 0.149 0.102 0.858 0.132 1.612 0.500 57 5 0.675 0.166 0.482 0.023 1.100 0.290 58 19 0.165 0.220 0.865 0.304 1.901 1.001 59 25 0.094 0.123 0.463 0.265 0.973 0.440 70 10 0.152 0.061 0.478 0.162 1.423 0.645 73 49 0.216 0.354 0.395 0.371 0.932 0.459 75 10 0.194 0.169 1.055 0.290 0.820 0.082 79 14 0.083 0.035 0.740 0.264 0.992 0.385 80 25 0.104 0.120 0.731 0.257 0.952 0.283 99 4 0.091 0.005 0.271 0.029 0.224 0.008 Total 1,445 0.116 0.184 0.678 0.424 0.959 0.547
II. Univariate Regressions
_cons .109829 .0048283 22.75 0.000 .1003576 .1193003 tangible10 .1044918 .0598466 1.75 0.081 -.012905 .2218886 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .1808 R-squared = 0.0075 Prob > F = 0.0810 F( 1, 1427) = 3.05 Linear regression Number of obs = 1429
_cons .1036977 .0047407 21.87 0.000 .0943982 .1129972 tangible25 .0673457 .0182136 3.70 0.000 .0316175 .103074 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .18006 R-squared = 0.0157 Prob > F = 0.0002 F( 1, 1427) = 13.67 Linear regression Number of obs = 1429
_cons .0956095 .0053061 18.02 0.000 .0852008 .1060181 tangible50 .0566822 .011134 5.09 0.000 .0348414 .078523 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .17978 R-squared = 0.0187 Prob > F = 0.0000 F( 1, 1427) = 25.92 Linear regression Number of obs = 1429
_cons .1263321 .0092468 13.66 0.000 .1081934 .1444708 tangible -.0184293 .0105675 -1.74 0.081 -.0391587 .0023001 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .18131 R-squared = 0.0019 Prob > F = 0.0814 F( 1, 1427) = 3.04 Linear regression Number of obs = 1429
_cons .1784004 .0097557 18.29 0.000 .1592635 .1975373 Rel_Enterprise -.0648414 .0065256 -9.94 0.000 -.0776421 -.0520408 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .18084 R-squared = 0.0370 Prob > F = 0.0000 F( 1, 1443) = 98.73 Linear regression Number of obs = 1445
_cons .1707097 .0103976 16.42 0.000 .1503136 .1911058 profitability -.6128431 .0844119 -7.26 0.000 -.7784262 -.44726 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .17842 R-squared = 0.0625 Prob > F = 0.0000 F( 1, 1443) = 52.71 Linear regression Number of obs = 1445
_cons .9599821 .0577327 16.63 0.000 .846733 1.073231 size -.0860917 .0055691 -15.46 0.000 -.0970161 -.0751674 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .15355 R-squared = 0.3057 Prob > F = 0.0000 F( 1, 1443) = 238.98 Linear regression Number of obs = 1445
_cons .1197747 .0088124 13.59 0.000 .1024871 .1370624 Mean_z_score -.0049709 .0030339 -1.64 0.102 -.0109227 .0009809 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .18036 R-squared = 0.0014 Prob > F = 0.1016 F( 1, 1338) = 2.68 Linear regression Number of obs = 1340
_cons .2011702 .0109438 18.38 0.000 .1797002 .2226402 Mean_Tobins_q -.0744473 .0065287 -11.40 0.000 -.0872555 -.0616391 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .19018 R-squared = 0.0318 Prob > F = 0.0000 F( 1, 1262) = 130.03 Linear regression Number of obs = 1264
III. Multivariate Regressions
_cons .0853092 .0090561 9.42 0.000 .0675447 .1030737 leverage .0992139 .0242035 4.10 0.000 .051736 .1466917 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .18364 R-squared = 0.0069 Prob > F = 0.0000 F( 1, 1443) = 16.80 Linear regression Number of obs = 1445
_cons .211158 .0148768 14.19 0.000 .1819752 .2403407 Rel_Enterprise -.0751777 .0070947 -10.60 0.000 -.089095 -.0612605 tangible -.0308885 .0111255 -2.78 0.006 -.0527125 -.0090644 REV10 -.12878 .0332779 -3.87 0.000 -.194059 -.0635011 tangible10 .2138998 .0980378 2.18 0.029 .0215859 .4062138 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .17667 R-squared = 0.0544 Prob > F = 0.0000 F( 4, 1424) = 30.98 Linear regression Number of obs = 1429
_cons 1.290056 .0808936 15.95 0.000 1.131354 1.448759 leverage -.0453249 .0247086 -1.83 0.067 -.0937999 .0031502 Mean_Tobins_q .0831694 .008704 9.56 0.000 .0660932 .1002455 size -.1166023 .0076829 -15.18 0.000 -.1316752 -.1015295 Rel_Enterprise -.0927721 .0073343 -12.65 0.000 -.1071611 -.0783831 tangible -.044409 .0095598 -4.65 0.000 -.0631641 -.0256539 REV10 .0904246 .0852052 1.06 0.289 -.076737 .2575862 tangible10 .239534 .26777 0.89 0.371 -.2857964 .7648643 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .14129 R-squared = 0.4504 Prob > F = 0.0000 F( 7, 1245) = 56.64 Linear regression Number of obs = 1253
_cons .2029546 .01573 12.90 0.000 .1720981 .2338111 Rel_Enterprise -.0667919 .0066175 -10.09 0.000 -.079773 -.0538109 tangible -.0366572 .0117405 -3.12 0.002 -.0596877 -.0136268 REV25 -.0551927 .0233039 -2.37 0.018 -.1009063 -.0094791 tangible25 .0885134 .0288651 3.07 0.002 .0318907 .1451362 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .17722 R-squared = 0.0484 Prob > F = 0.0000 F( 4, 1424) = 34.88 Linear regression Number of obs = 1429
_cons 1.292713 .0817282 15.82 0.000 1.132373 1.453054 leverage -.0472308 .0246938 -1.91 0.056 -.0956768 .0012152 Mean_Tobins_q .0847621 .0086622 9.79 0.000 .0677681 .1017562 size -.1185502 .0079487 -14.91 0.000 -.1341444 -.1029559 Rel_Enterprise -.0865161 .0066846 -12.94 0.000 -.0996304 -.0734017 tangible -.0388913 .0089899 -4.33 0.000 -.0565283 -.0212543 REV25 .1014927 .0355494 2.85 0.004 .0317493 .1712361 tangible25 -.0654678 .0466711 -1.40 0.161 -.1570305 .0260948 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .1432 R-squared = 0.4354 Prob > F = 0.0000 F( 7, 1245) = 54.95 Linear regression Number of obs = 1253
_cons .1611061 .0188808 8.53 0.000 .124069 .1981432 Rel_Enterprise -.045022 .0078247 -5.75 0.000 -.0603712 -.0296729 tangible -.0302765 .012315 -2.46 0.014 -.054434 -.006119 REV50 .0120502 .0219673 0.55 0.583 -.0310417 .055142 tangible50 .0321437 .0221583 1.45 0.147 -.0113228 .0756102 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .17737 R-squared = 0.0469 Prob > F = 0.0000 F( 4, 1424) = 29.09 Linear regression Number of obs = 1429
IV. Robustness _cons 1.277754 .0861931 14.82 0.000 1.108654 1.446853 leverage -.0615982 .0264467 -2.33 0.020 -.1134832 -.0097132 Mean_Tobins_q .0813375 .0085632 9.50 0.000 .0645377 .0981374 size -.1174071 .0080356 -14.61 0.000 -.1331719 -.1016424 Rel_Enterprise -.0820257 .0084317 -9.73 0.000 -.0985676 -.0654837 tangible -.0371596 .0094558 -3.93 0.000 -.0557107 -.0186085 REV50 .0520235 .0193526 2.69 0.007 .0140561 .0899908 tangible50 -.0240142 .0212773 -1.13 0.259 -.0657575 .017729 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .14391 R-squared = 0.4298 Prob > F = 0.0000 F( 7, 1245) = 52.97 Linear regression Number of obs = 1253
_cons 1.255662 .0838024 14.98 0.000 1.091252 1.420072 4 .0012707 .0097433 0.13 0.896 -.0178444 .0203858 3 .0162682 .0121871 1.33 0.182 -.0076413 .0401778 Industry leverage -.0511932 .0257297 -1.99 0.047 -.1016715 -.0007148 Mean_Tobins_q .0810051 .0086375 9.38 0.000 .0640594 .0979509 size -.1163833 .0081102 -14.35 0.000 -.1322944 -.1004722 Rel_Enterprise -.0841973 .0088387 -9.53 0.000 -.1015376 -.066857 tangible -.0308738 .0121088 -2.55 0.011 -.0546297 -.0071178 REV50 .0557694 .0201846 2.76 0.006 .0161697 .0953691 tangible50 -.0292464 .0216663 -1.35 0.177 -.071753 .0132602 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .14386 R-squared = 0.4311 Prob > F = 0.0000 F( 9, 1243) = 44.10 Linear regression Number of obs = 1253
_cons 1.291751 .0793686 16.28 0.000 1.136039 1.447463 2012 -.0380104 .0130046 -2.92 0.004 -.0635239 -.0124968 2011 -.0540846 .0123931 -4.36 0.000 -.0783983 -.0297709 2010 -.0325302 .0132616 -2.45 0.014 -.058548 -.0065125 2009 -.0073884 .0157383 -0.47 0.639 -.0382651 .0234884 fyear 4 .0002257 .0098858 0.02 0.982 -.019169 .0196204 3 .0123885 .0117883 1.05 0.294 -.0107387 .0355157 Industry leverage -.0387827 .0245129 -1.58 0.114 -.0868741 .0093088 Mean_Tobins_q .0826918 .008914 9.28 0.000 .0652036 .10018 size -.1151479 .0077804 -14.80 0.000 -.1304121 -.0998837 Rel_Enterprise -.0903541 .0075614 -11.95 0.000 -.1051886 -.0755195 tangible -.0410519 .0108137 -3.80 0.000 -.0622671 -.0198367 REV10 .0868093 .0852492 1.02 0.309 -.0804394 .254058 tangible10 .2285823 .2655737 0.86 0.390 -.2924415 .7496062 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .14011 R-squared = 0.4621 Prob > F = 0.0000 F( 13, 1239) = 32.38 Linear regression Number of obs = 1253
_cons 1.297925 .0801399 16.20 0.000 1.1407 1.45515 2012 -.0373422 .0132256 -2.82 0.005 -.0632894 -.0113951 2011 -.0557484 .0124232 -4.49 0.000 -.0801212 -.0313756 2010 -.0351209 .013373 -2.63 0.009 -.0613572 -.0088846 2009 -.0076427 .0160997 -0.47 0.635 -.0392283 .0239429 fyear 4 -.0011604 .0098383 -0.12 0.906 -.020462 .0181413 3 .0122235 .0115506 1.06 0.290 -.0104373 .0348844 Industry leverage -.0416018 .0248213 -1.68 0.094 -.0902981 .0070946 Mean_Tobins_q .0843214 .0090129 9.36 0.000 .0666391 .1020036 size -.1170806 .0080017 -14.63 0.000 -.132779 -.1013821 Rel_Enterprise -.0856839 .0070764 -12.11 0.000 -.099567 -.0718009 tangible -.0346078 .0106862 -3.24 0.001 -.0555729 -.0136427 REV25 .0995326 .0352393 2.82 0.005 .0303973 .168668 tangible25 -.0725626 .0457137 -1.59 0.113 -.1622475 .0171223 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .14197 R-squared = 0.4478 Prob > F = 0.0000 F( 13, 1239) = 31.72 Linear regression Number of obs = 1253
_cons 1.279844 .0846715 15.12 0.000 1.113729 1.445959 2012 -.0402102 .0132987 -3.02 0.003 -.0663005 -.0141198 2011 -.0552855 .0125803 -4.39 0.000 -.0799665 -.0306044 2010 -.0352572 .0134513 -2.62 0.009 -.061647 -.0088673 2009 -.0062508 .0162908 -0.38 0.701 -.0382114 .0257099 fyear 4 .0012602 .0096886 0.13 0.897 -.0177477 .0202682 3 .0152084 .0120296 1.26 0.206 -.0083922 .0388091 Industry leverage -.0537325 .0258324 -2.08 0.038 -.1044126 -.0030524 Mean_Tobins_q .0810923 .0089173 9.09 0.000 .0635976 .098587 size -.1158531 .0080791 -14.34 0.000 -.1317033 -.1000029 Rel_Enterprise -.0820965 .0087908 -9.34 0.000 -.0993431 -.0648499 tangible -.0309342 .0122403 -2.53 0.012 -.0549481 -.0069202 REV50 .0511027 .0200139 2.55 0.011 .0118378 .0903677 tangible50 -.0300136 .0214589 -1.40 0.162 -.0721135 .0120862 RCDS Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .1426 R-squared = 0.4428 Prob > F = 0.0000 F( 13, 1239) = 31.76 Linear regression Number of obs = 1253
