or withdraw financing to the firm. Contrary to the related empirical literature that has been investigating firms’ aggregated financing using information on the availability of CDS contracts, this paper will, due to the use of regulatory data, focus on the impact of creditors’ CDS activity, conditional on the availability of CDS contracts.
specific CDS position information such as the number of contracts outstanding and the notional amount bought and/or sold by the individual bank.9
Next, I match the credit registry and CDS position data on the firm-bank level. The sample period is 2008-Q1 to 2016-Q2 and is limited by the availability of the CDS position data. Based upon the credit registry and CDS position data I first identify all firms on which there are, respectively, are no, traded CDS contracts, as well as the credit exposure of each firm-bank pair. Secondly, I use the combined data sample to reveal whether the individual creditor holds a CDS contract that is written on the respective firm’s debt. The unique data allows me for each firm (borrower) to distinguish between the creditors that hold CDS contracts (CDS creditors), and the creditors that do not hold CDS contracts (non-CDS creditors). As this paper focuses on the role of creditors’ use of CDSs, the identification of CDS creditors is key for the empirical investigation. Thirdly, I exploit the comprehensive CDS position data by determining whether the individual creditors are net protection buyers or sellers, and the magnitude of their CDS holdings. This is of particular interest in order to differentiate between the general impact on credit markets due to the prevalence of CDS contracts and the firm-specific impact on debt financing due to creditors’ holdings of CDSs.
Figure 1
-Figure 1 shows the time series of aggregated CDS holdings by German banks on German non-financial firms. While the banks during most of that sample period have been buyers of credit protection (indicated by a positive CDS net notional value), the figure outlines most interestingly that the same banks in recent years have been net sellers of credit protection.10 In gross terms, the German CDS market reached its peak in late 2010 accounting for about 11% of the total credit amount outstanding and since then, come down to a value of about 4% of firm-level credit exposures. Accordingly, the figure highlights the necessity for studying the impact of both buy and sell CDS positions of banks, as well as the impact of creditors’ CDS holdings supplementary to the general effect of the existence of these markets.
Table 1
-In addition, Table 1 summarizes credit and CDS statistics for firms with available CDS and credit registry. As outlined in the table, the CDS reference firms in my baseline
also trade CDS contracts on the firm’s debt. The average firm-level credit exposure is approximately 1,025 million euro of which about one-third is held by CDS creditors.11 The aggregated firm-level credit exposures have been fairly stable over time while the share of credit held by CDS creditors, similar to the size of the CDS market, has been decreasing over time. Further, the table outlines that not only the amount of credit protection bought relative to credit protection sold has been decreasing, but also that the number of contracts that were bought compared to the number of contracts that were sold has been decreasing. This confirms that the liquidity of the German CDS market has been decreasing, but also that creditors in recent years to a larger extent have become net sellers of CDS contracts.
In addition to the detailed CDS position and credit registry data, I collect other firm fundamental data from the Compustat Global database.12 From this source, I collect all yearly and quarterly corporate financial and stock price data for the period 2008-Q1 to 2016-Q2. Further, I supplement the data from Compustat with firm-level data from the Capital IQ and the Mergent FISD databases. In contrast to Compustat, Capital IQ compiles, inter alia, detailed information on corporate debt structure, using footnotes contained in the firm’s financial reports. Specifically, I obtain data on firms’ debt maturity structure from this source. In addition, I collect bond-related information on the firm level from Mergent FISD. To mitigate the effect of outliers, I winsorize data from these sources at the 1st and 99th percentiles. As this study investigates the impact of creditors’
CDS holdings on firm-level debt financing measures which are very different for financial and utility firms, I follow the literature and exclude financial firms (SIC codes 6000 to 6999), utility firms (SIC codes 4900 to 4999) and firms for which no SIC code is available. Further, as firms that act as CDS reference entities are relatively large firms compared to the average firm covered in the credit registry data and in order to provide a more unbiased sample, I exclude all small and medium-size firms from the CDS position and credit registry sample by restricting the sample to firms with an aggregated credit exposure that on average is within the 1st and 99th percentiles of that of the identified CDS firms. Finally, I restrict the sample to only include firms for which credit registry data from the MiMik database as well as firm fundamental data from Compustat and Capital IQ is available.13 My final sample consists of 78 non-financial German firms, of
11For comparison, the average non-CDS reference firm in the credit registry data has 14 credit relations and the total firm-level credit exposure is approximately 570 million euro. These values refer to quarterly observations of 2,267 non-financial firms operating in the same industry as the identified CDS reference firms.
12The advantage of using data from Compustat rather than, for instance, Amadeus or USTAN, is that I have quarterly rather than only annual data, which allows for greater granularity in my analysis.
13I follow the conventional approach in related empirical research (e.g., Bates, Kahle, and Stulz (2009)) and assume that a firm has no merger and acquisition activities in a given quarter if it is reported as
“missing” by Compustat. I use the same argument for observations of capital expenditures.
which 31 act as reference entities in CDS contracts. In the latter, I will refer to CDS reference entities as CDS Firms, and to the remaining firms as Non-CDS Firms.
3.2 Empirical Specification
In order to investigate the role of creditors’ CDS holdings for firms’ debt financing I first analyze the implications for firms’ debt financing. Specifically, I use the firm-specific credit exposures which I obtain using the quarterly bank-firm credit registry data from MiMik and by aggregating individual banks’ credit exposures to the firm level. As the credit measure in MiMik covers firms’ total credit amounts, i.e., both loans and corporate bonds, a change in this measure will reflect the extension or withdrawal of debt financing to the firm. Based upon these data I use Total Credit Ratio and Total Credit Amount as my key measures. Total Credit Amount is the natural logarithm of the firm-specific credit exposure at the end of a quarter, while Total Credit Ratio is the firm-specific credit exposure at the end of a quarter, scaled by total assets. In the empirical analyses I use various versions of both measures. Second, I investigate the role of creditors’ CDS holdings for firms’ debt refinancing by analyzing the type of credit provided to the firms.
As a proxy for firms’ exposure to debt refinancing I mainly use the variableDebt Maturity which is the principal-weighted maturity of all debt as reported in Capital IQ and given in years. To proxy the direct cost of debt I use information on the firm’s total interest and related payments and define the variable Interests To Credit as the total amount of interests, scaled by total credit outstanding. To further test for the changes in the type of debt I use Bonds To Debt which I define as the ratio of the principal of the firm’s outstanding bonds to total bank debt.
For the choice of potential determinants of the outlined firm-level debt financing measures and the empirical design I largely follow Saretto and Tookes (2013), Ashcraft and Santos (2009), and Subrahmanyam, Tang, and Wang (2017) and relate the measures to a set of explanatory variables obtained from Compustat.
Table 2
-Table 2, Panel A, presents base firm characteristics for the CDS and non-CDS firm sample.
As seen from the table, the average CDS firm has slightly more credit outstanding and a higher leverage ratio than non-CDS firms. In particular, the average leverage ratio is 27.7% and 25.0% for CDS and non-CDS firms, respectively, implying that CDS firms,
the debt financing across both CDS and non-CDS firms. With respect to other base firm characteristics Table 2, Panel A, outlines that the average CDS firm in my sample is larger, uses more commercial papers and has a slightly lower Z-score, but else wise is similar to the average non-CDS firm.
In order to analyze the implications of creditor’ use of CDS contracts I relate various proxies to the outlined firm-level debt financing measures. To control for fundamental differences between CDS and non-CDS firms I use CDS Firm which is a dummy variable equal to one for firms that at some point during the sample period act as a reference firm in a CDS contract. To capture the impact of the presence and magnitude of creditors’
actual CDS holdings I first create quarterly firm-bank-specific CDS measures based on the weekly CDS position data. To this end, I follow the approach in Caglio, Darst, and Parolin (2017) and use the average of the weekly CDS positions of the last three weeks before the final week in a given quarter.14 Then, similar to banks’ credit exposures I calculate firm-specific CDS measures in a given quarter by aggregating creditors’ CDS position on the firm level. Following this methodology I obtain quarterly data on the number of CDS contracts bought and sold, the net notional values of these contracts, as well as how many of a firms’ creditors hold a CDS contract in a given month. Based upon this quarterly information I then define several variables to measure the extent to which the firm’s creditors hold CDS contracts on the firm. As a proxy for the presence of CDS creditors in a given quarter I use CDS Outstanding which is a firm-quarter-level dummy that is equal to one in quarters where at least one of the firm’s creditors holds a CDS contract on the firm, and is zero otherwise.15 As a proxy for the significance of creditors’
CDS positions I use the comprehensive firm-level information on the outstanding CDS contracts. Specifically, I make use of various versions of the variableCDS Coverage, which is defined as the total net notional amount of all creditors’ outstanding CDS contracts, scaled by the aggregated credit exposure to the firm. That is,
CDS Coverageit= Total CDS Net Notionalit
Total Credit Exposureit (1)
where t indicates the year-quarter and i refers to the firm, and the CDS net notional value is the CDS notional amount bought net of the CDS notional amount sold.16 While the size of CDS Coverage measures the fraction of the firm-specific credit exposure that is covered by outstanding CDS held by their creditors, the sign of the ratio outlines whether the creditors are net buyers or net sellers of credit protection. Specifically, the
14In robustness tests I use end of quarter observations and obtain similar results.
15In contrast to the measures used in the related literature, CDS Outstanding reflectsboth the CDS trading and outstanding CDS positions of the reference firm’screditors and on aquarterly basis.
16In robustness tests I use the weighted average of each individual CDS net notional to credit exposure ratios and find that the main results are robust to this alternative specification.
ratio will be positive when the creditors in sum are net protection buyers and, thus, hedge their credit exposures. In contrast, the ratio will be negative when the creditors in sum are net protection sellers and, in fact, implicitly increase their credit exposures.
Further, the ratio will be equal to zero when there are (in net terms) no outstanding CDS contracts in the respective quarter. As outlined in Table 2, Panel B, about 9 percent of a CDS firm’s creditors hold CDS contracts on the firm in a given quarter. Accounting for the credit supply of these creditors, the average CDS firm has a CDS coverage ratio of approximately 7 percent implying that the creditors on average are net protection buyers.
As I find that the number of CDS contracts bought is slightly lower than the number of CDS contracts sold, the notional value of credit protection bought by creditors is in fact larger than the notional value of credit protection sold by creditors. However, the coverage of creditors’ CDS contracts shows a substantial variation both in terms of sign and size and will, in line with the measures’ definition, vary both across firms and across time. For the more formal study of the firm and quarter-specific effect of creditors’ CDS holdings on firms’ debt financing I mainly use firm-level panel regressions where I regress the measures for firms’ debt financing on proxies for the CDS activity of the respective firms’ creditors, as well as a set of control variables. In order to allow for a general lagged effect of the time-varying measures of creditors’ CDS holdings, as well as to address concerns of reversed causality I use the one-quarter lagged versions of the CDS-related in all empirical specifications. Specifically, I use various versions of the following panel regression model specification, i.e.,
Debt Financing Proxyit =α+βCDS Activity Proxyit−1+θXit+it (2) where t indicates the year-quarter and i refers to the firm. In terms of the controls in the model,Xit, I follow the literature and include firm fundamentals, as well as industry, firm, time and rating fixed effects. The set of controls used in the specific model depends on the choice of the debt financing proxy used as dependent variable. Definitions of all variables included in the regression models are presented in Appendix Table A1.