Credit Default Swaps and Corporate Cash Holdings

Transcription

1 Credit Default Swaps and Corporate Cash Holdings Marti Subrahmanyam Dragon Yongjun Tang Sarah Qian Wang August 14, 2012 ABSTRACT Considerable attention has been devoted into the real effects of derivatives, particularly credit default swaps (CDS). In this paper we empirically estimate the effect of CDS on corporate cash holdings. Using a comprehensive sample of North American corporate CDS introductions between 1997 and 2009, we find that corporate cash holdings increase after the inception of CDS trading. The impact is significant after controlling for the endogeneity of CDS trading. Moreover, cash-to-assets ratios for firms with larger CDS contracts outstanding, and those with less access to financial market are more affected by CDS trading. The impact of CDS is beyond the direct effect of line of credit on cash holdings. Keywords: Credit default swaps, cash holdings, empty creditor We thank seminar participants at the University of Hong Kong. Stern School of Business, New York University; msubrahm@stern.nyu.edu School of Economics and Finance, University of Hong Kong; yjtang@hku.hk School of Economics and Finance, University of Hong Kong; sarawang@hku.hk

2 Credit Default Swaps and Corporate Cash Holdings ABSTRACT Considerable attention has been devoted into the real effects of derivatives, particularly credit default swaps (CDS). In this paper we empirically estimate the effect of CDS on corporate cash holdings. Using a comprehensive sample of North American corporate CDS introductions between 1997 and 2009, we find that corporate cash holdings increase after the inception of CDS trading. The impact is significant after controlling for the endogeneity of CDS trading. Moreover, cash-to-assets ratios for firms with larger CDS contracts outstanding, and those with less access to financial market are more affected by CDS trading. The impact of CDS is beyond the direct effect of line of credit on cash holdings. Keywords: Credit default swaps, cash holdings, empty creditor

3 I. Introduction Credit derivatives, especially credit default swaps (CDS), have attracted significant attention during the credit crisis. CDS are insurance-like derivative instruments that offer investors protection against default by a reference entity. While it was once labelled financial weapons of mass destruction, 1 CDS remain a robust and effective financial tool for hedging risk or taking on exposures, which was evidenced by global notional net open positions of $2,900 billion in 2011, up from $2,400 billion in However, the unfolding Eurozone crisis piles new challenge on the further development of the CDS market. In October 2011, European authorities banned the naked sovereign CDS trading. Investors are not allowed to buy CDS without holding the underlying debt. We need clarity from European leaders and regulators on the CDS market before it freezes. At the moment huge questions remain. 3 Therefore, understanding of the real effects of CDS is of crucial importance. Our paper contributes along this direction to present a careful empirical investigation about the impact of CDS on corporate liquidity policies. Why would CDS trading affect corporate cash holdings? Cash is an important liquidity management tool for corporations. CDS can affect corporate cash holdings in several ways, by affecting the availability of external financing. First, the introduction of CDS trading on the firm s debt increases credit supply from the capital suppliers. This is because CDS provide an effective tool for credit risk transfer. Given their CDS position, creditors bargaining power is also enhanced, which reduces firms incentive of strategic default and raises the debtors pledgeable assets. Due to the risk mitigation effects and enhanced bargaining power, creditors are more willing to lend after the introduction of CDS trading (Saretto and Tookes (2011), Bolton and Oehmke (2011)). Thus, firms financial constraint is relaxed after the inception of CDS contracts on their debt. Consequently, firms may hold less cash and rely more on the financial market to manage the liquidity needs. On the other hand, CDS can change the incentive of creditors. Due to the nature of CDS, lenders can over-insure their credit exposure. When the firm is in distress, CDS protected creditors are tougher in renegotiation. They even tend to push the firm into bankruptcy to get the payment from their CDS position. Therefore, after the introduction of CDS trading, it is harder for firms to get capital from creditors when liquidity is most needed. Given the tougher CDS protected creditors, firms may increase cash holdings to manage their potential liquidity needs. Moreover, CDS can affect corporate cash holdings by the feedback effects from CDS to cash holdings. CDS spread 1 Berkshire Hathaway Annual Report for Critics of Credit Default Swaps Have Got It Wrong, Financial Times, November 21, Eurozone Crisis Piles Pressure on Credit Default Swaps, Financial Times, November 6,

4 has been used extensively as the measure of credit quality. When corporate liquidity is low, CDS market responds with soaring CDS prices (CDS spreads). This could undermine the market confidence and reinforce the negative view about the corporate. Anecdotal evidence indicates that high CDS spread could feedback into corporate cash holdings. For example, the CDS spread for Nokia soared to a record of 435 basis points in the first quarter of 2012, after the net cash dropped to 4.9 billion euros from 5.6 billion euros at the end of While Nokia was still rated as investment grade of Baa2, the soaring CDS spread implies junk grade rating of Ba2. This threatens the company s rating to be downgraded to junk grade. In response to this, Nokia vowed to take significant structural actions if and when necessary, including asset sales. 4 Therefore, it is valuable to keep more cash on hand especially after the introduction of CDS trading on their debt, since CDS spread is sensitive to the change of firms liquidity status. These potential channels suggest that CDS may affect corporate liquidity policy, in particular, cash holdings. But the direction and magnitude of the impact is ambiguous. To test the impact of CDS on cash holdings, we use a comprehensive real CDS transaction dataset for North American names. Given the over-the-counter nature of CDS market, it is hard to pin down the exact date for CDS introduction. We rely on multiple data sources to identify the CDS launch date, including GFI Inc., the largest global interdealer broker with the most extensive records of CDS trades and quotes, CreditTrade, a major intermediary especially in the early stages of the CDS market, and Markit, a data disseminator and vendor, which provides daily quotes from major institutions. In our final sample, there are 901 CDS introductions from 1997 to The corporate cash holding data are from Compustat database. We rely on the level of cash holding model to identify the CDS effect. However, the endogeneity of CDS introduction complicates the interpretation. CDS firms may be fundamentally different from non-cds firms regarding the decision of cash holdings. Besides of the fixed effect control, we address the endogeneity concern through a difference-in-difference comparison, a propensity score matching analysis, and instrument variable approach. Moreover, we employ a continuous measure of corporate CDS exposure, which is less affected by the selection problem. We find that the introduction of CDS trading on corporate debt increases the firm s cash holdings, after controlling for variables suggested by the cash holding models. The effect of CDS is both statistically significant and economically large. For our sample CDS firms, cash ratio increases from 8.2% to 10.2% once the CDS starts trading. Given the mean cash ratio of 4 Nokia Swaps Trade Like Junk as Cash Dwindles: Corporate Finance, Bloomberg, April 12,

5 8.2% for CDS firms, this represents 24.4% increase in cash holdings! The positive relationship is significant after controlling for the endogeneity of CDS trading. Moreover, the CDS effect is stronger for firms with larger CDS outstanding. We further find that cash holdings of unrated firms and firms with non-investment grade ratings are more affected by CDS. This is because unrated and non-investment grade firms have less access to financial markets, compared to investment grade firms, and thus fewer alternatives when their major debtors become tough CDS protected creditors. In addition, firms with more creditor relationship and line of credit might be more affected by the CDS introduction, since it is more likely for them to have tougher creditors. Our paper contributes to provide a new perspective on the real effect of the CDS market. Previous works have identified the impact of CDS on credit supply, creditors monitoring incentive, the reference entities borrowing cost and default risk (Bolton and Oehmke (2011), Parlour and Winton (2011), Beyhaghi and Massoud(2011), Saretto and Tookes (2011), Subrahmanyam, Tang, and Wang (2012), etc.). We show that CDS can affect corporate strategic cash holdings. Our paper also contribute to the growing literature on corporate cash holdings (Bates, Kahle, and Stulz (2009), Acharya, Davydenko, and Strebulaev (2012), Palazzo (2012), Gao, Harford, and Li (2012), etc.). U.S. firms are holding more cash than at any time in nearly half a century. 5 In addition to the identified factors that generate precautionary motives of cash holdings, CDS trading on firm s debt creates additional precautionary motives for cash holdings. These findings also have implications for policy makers. The CDS market regulators should understand and take into account the real effect of CDS in their policy actions. The outline of the paper is as follows. Section II presents the literature and hypothesis development. In section III, we describe our sample and empirical methods. Section IV presents the main empirical results about the effect of CDS on cash holdings. Firm characteristics and additional discussions regarding the channels of the effects are presented in section V. Section VI concludes. 5 Companies Shun Investment, Hoard Cash, Wall Street Journal, September 17,

6 II. Relevant Literature and Hypotheses Development A. Relevant Literature Given the potential role of CDS in destabilizing the market during the crisis period, concerns have been raised about the real effects of CDS. CDS can affect the relationship between creditors and firms by changing the behavior of the creditors. CDS market provides a venue for insider trading (Acharya and Johnson (2007, 2010), Hilscher, Pollet, and Wilson (2011)). Acharya and Johnson (2007, 2010) show that the intensity of insider trading is related to the number of lenders. CDS is also an effective way for credit risk transfer (Minton, Stulz, and Williamson (2009)). Credit risk can be transferred to those who are in the best position to bear them. Due to this risk mitigation effects, CDS can relax the credit supply constraint (Saretto and Tookes (2011)). However, by hedging credit exposure with CDS, creditors monitoring incentive is weakened (Duffee and Zhou (2001), Morrison (2005), Arping (2001), Thompsom (2007), Allen and Carletti (2006), Parlour and Winton (2012), Beyhaghi and Massoud (2011)). Parlour and Plantin (2008) show that if CDS market is liquid, lenders may initiate too many loans and reduce monitoring, ex post. Ashcraft and Santos (2009) also argue that such reduced monitoring may ultimately lead to a higher cost of debt. 6 Despite of the weakened monitoring incentive, CDS creates empty creditors who are creditors with an economic interest in the firm s claim, but no risk alignment with the other creditors who do not have CDS protection. Bolton and Oehmke (2011) formally model the empty creditor problem. Based on their model, empty creditors tend to be tougher in the process of renegotiation due to the potential gain from their CDS position. They even have incentive to push the firm into inefficient bankruptcy to get the compensation from the CDS contracts. Consistent with the theoretical prediction, Danis (2012) find that distressed firms with CDS trading are less successful in debt restructuring. Bedendo, Cathcart, and El-Jahel (2011) examine the distressed firms decisions regarding out-of-court restructuring and bankruptcy filing during the global financial crisis. They find that CDS contracts do not significantly increase the probability of bankruptcy when the firm is already in distress, although their relatively small sample spans a short time period. Similarly, Peristiani and Savino (2011) document the higher bankruptcy risk in the presence of CDS during Subrahmanyam, Tang, and Wang (2012) directly investigate the effect of CDS on firm s bankruptcy risk. They find CDS increase corporate bankruptcy risk. 6 Che and Sethi (2012) model the impact of naked CDS on economic fundamentals. They argue that CDS can crowd out debt investors, reduce the firm s debt capacity and increase its costs of debt. 5

7 Different from previous literature which focuses on the behavior of creditors and its outcome, this paper investigates the strategic actions of firm after CDS introduction. Given the tougher CDS protected creditors in renegotiation when liquidity is most needed, firms may change the liquidity management strategy. Specifically, we examine whether firms rely more on cash holdings to manage the potential liquidity needs after the introduction of trading on CDS contracts referencing its default. Our paper also relates to the increasing literature about corporate cash holdings. 7 While it is accepted that the optimal level of cash holding depends on the trade-off between the cost and benefit of cash holding, the primary focus of this literature is on precautionary motives for holding cash. The precautionary motives stem from the expected financial constrains in the future. To protect themselves against the negative shocks in the future, firms tend to build cash holdings. motive. Previous papers have found evidence in support of precautionary They have also identified important factors/firm characteristics that induce this motive. Bates, Kahle, and Stulz (2009) find evidence in support of the precautionary motive for cash in a sample of US industrial firms from Opler et al. (1999) find that firm s cash holding negatively relates to size, B/M and positively relates to capital expenditure, payouts, R&D expenditure, and cash flow volatility. Consistent with Opler et al. (1999), Han and Qiu (2007) show that cash flow volatility increases precautionary motive. But the relationship is significant only for financially constrained firms. Almeida, Campello, and Weisbach (2004) show that cash flow of constrained firms is more sensitive to operating cash flow than unconstrained firms. Besides of conventional identified factors that affect precautionary motive, such as firm level risk as approximated by cash flow volatility, Palazzo (2012) further documents that firm s cash holdings positively relate to aggregate risk, which generates additional precautionary savings motive. Duchin (2010) also provides support for the precautionary motive. He explicitly emphasizes the diversification induced precautionary motive. While precautionary motive is of crucial importance in determining firm s cash holdings, a number of recent papers provide evidence for the agency problem induced cash holdings. 7 There are extensive literature on cash holdings, including Disatnik, Ran, and Schmidt (2012), Dittmar and Duchin (2011), McLean (2011), Denis and Sibilkov (2010), Lins, Servaes, and Tufano (2010), Nikolov and Whited (2010),Riddick and Whited (2009), Pinkowitz, Stulz, and Williamson (2006), Bertrand and Mullainathan (2003), Dittmar, and Mahrt-Smith (2003). 8 Kim, Mauer, and Sherman (1998) investigate the determinants of optimal cash holdings in a setting of three period framework (with the assumption that investors are risk neutral). They find that cash holding increase with external financing cost, variance of future cash flow and return on future investment opportunities. However, cash holdings decrease with the return difference between physical and liquid assets. Acharya, Davydenko, and Strebulaev (2012) emphasize the precautionary motives for firms cash saving. They explicitly document the positive relationship between firm s cash holding, credit spread and default probability. 6

8 The agency motive for holding cash dates back to Jensen (1986). Entrenched managers in company with poor investment opportunities tend to build excess cash holdings as evidenced in Dittmar et al. (2007) and Harford et al. (2008). Gao, Harford, and Li (2012) provide further insights of this channel by comparing the cash holding between public and private firms. In addition to the precautionary motive and agency motive of cash holdings, transaction cost motive and tax motive of cash holdings may also play a role in determining firm s cash level. It is less costly for large firms to convert non-cash asset to cash. Therefore, large firms tend to hold less cash due to these economies of scale (Baumol (1952)). However, given the increasing efficiency of market in dealing with transactions, the transaction based requirement of cash holding is expected to be less important (Bates et al. (2009)). Foley, Hartzell, Titman, and Twite (2007) find that firms with foreign operating subsidiaries have higher cash holdings which relates to the tax costs associated with repatriating foreign income. Bates et al. (2009), however, do not find tax effect in a sample covering longer time period. What is more, cash holdings may also be affected by the real effect of cash, such as that on firm s product market performance (Fresard (2010)). We contribute to the literature by documenting the real effect of CDS on corporate cash holdings. B. Hypotheses Development In this study, we focus on the effect of CDS on firm s strategic cash holdings. Firms hold cash strategically. The motives of corporate cash holdings have been discussed extensively in media and previous literatures. Most of the discussions focus on the precautionary motives of cash holdings. Firms want to hedge against the expected financial constraint in the future. To protect themselves against these shocks, they tend to build cash by themselves. CDS can change corporate cash holdings by affecting the availability of external financing, which generates additional precautionary motives of cash holdings. On the one hand, ex ante, CDS relax the credit supply constraint due to the risk mitigation effect and enhanced bargaining power for creditors (Saretto and Tookes (2011), Bolton and Oehmke (2011)). This induces lower precautionary motives of cash holdings. One the other hand, ex post, CDS protected creditors tend to be excessively tough in renegotiation (Bolton and Oehmke (2011)). They will accept the restructuring offer and rollover the debt only if the terms are really attractive. Then firm s financial constraint can be tightened after CDS introduction due to the tougher creditors. Thus CDS trading can generate additional precautionary motives for cash holdings. Firms rely more on cash to manage their liquidity need after CDS introduction. In addition, the effect of CDS on cash holdings may come from the feedback effects from CDS 7

9 spread. CDS spread represents the market view about the reference entities credit risk, which is sensitive to the corproate s liquidity status. CDS spread soars when corproate liquidity condition is poor. This could ruin market confidence in the corporate. Consequently, CDS spread may affect firm s access to external financing. It is valuable to keep more liquid asset on hand, especially after the introduction of CDS trading on the firm s debt. Because of these potential channels, CDS can affect firms external financing and change their strategic cash holdings. But the direction of the absolute impact is ambiguous. If the tough creditor and feedback effects channels dominate, we expect: Hypothesis 1: Corporate cash holdings increase after the introduction of trading on CDS contracts referencing its default. The effect of CDS on cash holdings may change with firm characteristics. CDS outstanding can be used as a proxy for CDS exposure. The larger of the CDS exposure, the more severe of the CDS impact. Moreover, credit ratings measure firm s financial constraint. Unrated firms and firms with non-investment grade ratings generally have less access to financial market. As a result, they have less alternatives when their major creditors become tougher CDS protected creditors. In addition, firms with more creditor relationship and line of credit might be more affected by the CDS introduction, since it is more likely for them to have tougher creditors. Therefore, we expect: Hypothesis 2: The increase in cash holdings after CDS introduction is more significant for firms with larger amount of CDS outstanding, those with unrated/non-investment grade ratings, and entities with multiple creditors and line of credit. In the rest of this paper, we are going to test each of these hypotheses. Regression controls are motivated by the transaction and precautionary explanations for cash holdings. III. Data and Empirical Specification A. Data We use real CDS transaction data to identify a sample of firms with CDS contracts referring to their debt. The CDS transaction data are from CreditTrade and GFI Group. Different from the CDS quotes data used in previous literature, our data contain the real trading records with complete contractual information. 9 Given the over-the-counter nature of CDS 9 See Subrahmanyam, Tang, and Wang (2012) for a detained discussion about the data set. Similar data sources are used in Acharya and Johnson (2007), Blanco, Brennan, and Marsh (2005), and Nashikkar, Sub- 8

10 contracts, we use the first CDS trading date in our sample as the CDS introduction date, and compare the changes in corporate cash holdings upon the onset of CDS trading. We further cross-checked this CDS sample with Markit database. In the analysis, we also utilize the detailed transaction information, and construct continuous measures of CDS exposures. The combined sample covers the period from June 1997 to April There are 901 North American corporates that have CDS initiated on them at any time in the sample period. Industry coverage of CDS firms are also quite diversified. Most CDS firms in our sample are in the manufacturing (SIC 2,3), transportation, communications, and utilities (SIC 4), and finance, insurance, and real estate (SIC 6) sectors. 10 Corporate cash holding and characteristic data are from Compustat. Following Bates, Kahle, and Stulz (2009), we measure cash holdings as the ratio of cash and marketable securities to total assets. While cash to assets is the most conventional measure of cash holdings, we also checked alternative measures of cash ratio, such as cash to net assets and cash to sales and cash to bond outstanding. We further obtain credit rating data from Compustat and FISD; bond issuance data from FISD; and, analyst coverage data from I/B/E/S. The bond trading data are from the Trade Reporting and Compliance Engine (TRACE) maintained by the Financial Industry Regulatory Authority (FINRA) over the period from Moreover, we obtain line of credit data from Dealscan 11 Table I presents the year-wise summary from for all firms in the Compustat database: the number of Compustat firms, the number of firms on which CDS are traded, and cash ratios for firms with and without CDS. As the table shows, the introduction of CDS is most pronounced from 2000 to The fifth and sixth columns show the cash ratios for Non-CDS firms and CDS firms respectively during each year. We measure cash ratio as cash and marketable securities divided by total assets. Similar to the finding in Bates, Kahle, and Stulz (2009), there is an increasing trend in cash ratio for both Non-CDS and CDS firms in our sample. The cash ratio for Non-CDS firms in 2009 is 116% of the cash ratio in 1997, while it is 143% of its value for CDS firms. Moreover, the average cash ratio for Non-CDS firms is more than doubled compared with that of CDS firms, since only relatively large firms, by asset size and debt outstanding, are selected for CDS trading. Large firms generally hold less cash. rahmanyam, and Mahanti (2011) etc. 10 In our main analysis, we do not exclude financial firms. We also drop financial firms from the sample as robustness checks. The results are quite similar. 11 The line of credit data are detailed by Achaya, Almeida, and Campello (2012). 9

11 B. Empirical Specification To investigate the effect of CDS on corporate cash holdings, we employ a model for levels of cash holdings based on Bates, Kahle, and Stulz (2009). The level of cash holding model has been used extensively in literature, such as Kim, Mauer, and Sherman (1998), Opler, Pinkowitz, Stulz, and Williamson (1999), Dittmar and Mahrt-Smith (2007), Foley, Hartzell, Titman, and Twite (2007), and Harford, Mansi, and Maxwell (2008) etc. Regression controls are motivated by the transaction and precautionary explanations for cash holdings. Following Bates, Kahle, and Stulz (2009), we assume corporate cash holdings are determined by: Cash = Xβ + β 1 CDS Trading + ɛ (1) where Cash is the cash ratio measured as the ratio of cash to total assets, and X is a vector of determinants of cash holdings. We follow Bates, Kahle, and Stulz (2009) and include a set of fundamental determinants of cash holdings, including industry cash flow risk (Industry Sigma), the ratio of cash flow to total assets (Cash Flow/Assets), Market to Book, a measure of its investment opportunity, the logarithm of total assets (Size), the firm s ratio of Net Working Capital/Assets, Capital Expenditure, Leverage, the ratio of R&D to sales (R&D/Sales), the firm s dividend payment (Dividend Dummy), and the ratio of acquisition to total assets (Acquisition Activity). To estimate the impact of CDS trading on the corporate cash holdings, we include credit default swaps variables in the model specification, similar to Ashcraft and Santos (2009), Saretto and Tookes (2011), Subrahmanyam, Tang, and Wang (2012). CDS Trading is a dummy variable that equals one after the inception of the firm s CDS trading. So for a CDS firm, CDS Trading equals zero before CDS introduction, and one thereafter. Therefore, the coefficient of interest is that of CDS Trading, which captures the impact of CDS on cash holdings after the inception of CDS trading. The analysis is conducted in the sample including CDS firms and Non-CDS firms. Given the unobservable difference between CDS and non-cds firms, we control the fixed effects in the panel data analysis. To establish the relationship between CDS trading and corporate cash holdings, our main challenge is the potential endogeneity of CDS trading. It is possible that there is a third factor affecting both the introduction of CDS trading and corporate cash holdings. Then the observed effects may not be caused by the CDS contracts, but the impact of the third factor. We use four methods to address the endogeneity concern: the 10

12 difference-in-difference estimate, propensity score matching, instrumental variable approach, as well as the continuous CDS exposure measures which are less affected by the selection issue. We expect firms with larger CDS exposure are more affected by the CDS trading, and therefore has greater precautionary motives for cash holdings. In addition, to assess whether the increase in cash is related to firm characteristics such as credit ratings, we divide the sample firms into Unrated and Rated/Non-Investment grade and Investment grade groups. We also control for the direct effect of credit line, and investigate the effect of CDS on change in cash. IV. CDS Trading and Cash Holdings: Empirical Results In this section, we focus on the effects of CDS trading on corporate cash holdings. We first report the baseline results regarding the effects of CDS on cash holdings with fixed effects in the sample of CDS firms and all Non-CDS Compustat firms as control group. Then we investigate if the effects are robust to controlling for the endogeneity of CDS trading. A. Changes in Cash Ratios around CDS Introduction The summary statistics in Table I illustrate that there is an increase in the cash ratio for both CDS and Non-CDS firms. To illustrate that CDS firms experience a more significant increase in cash ratio, we focus on the changes in cash ratio around CDS introduction. Figure 1 shows the changes in cash ratios for CDS and Non-CDS firms from one year before the inception of CDS trading to zero (-1,0), one (-1,1), two (-1,2) or three (-1,3) years after the inception of CDS trading. Non-CDS matching firms are selected in a sample of firms that do not have CDS trading throughout the sample period. For each CDS firm, we find a Non-CDS matching firm that in the same industry (measured by 4 digit SIC code) and with the closest size (measured by total assets) as the CDS firm. The average cash ratio increases for both CDS and Non-CDS firms, but the increase is more pronounced for CDS firms. We observe 6% increase in cash ratio for both CDS firms and Non-CDS matching firms from year t-1 to year t. However, from year t-1 to year t+3, the increase in cash holdings for CDS firms is 0.7% more than that of Non-CDS matching firms. Given the mean cash ratio of 8% for CDS and non-cds firms matched on industry and size, the 0.7% additional increase in cash ratio is economically significant. Therefore from this figure, we get preliminary evidence that the increase in cash ratios is more significant for CDS firms across years. 11

13 B. Impact of CDS on Cash Holdings: Baseline Results We next estimate the CDS effects on cash holdings with appropriate control variables. The baseline analysis is conducted in the sample including both firms with CDS and without CDS traded with quarterly observations. We control time and firm fixed effects in all regressions. As discussed in section III.B, we follow Bates, Kahle, and Stulz (2009) and estimate the model of cash holdings. The variable of interest is on CDS Trading, which equals one after the introduction of CDS trading on the firm s debt. Therefore the coefficient of CDS Trading captures the effect of CDS introduction on corporate cash holdings. The baseline regression results are reported in Panel A of Table II. The first column in Panel A lists the independent variables in model specifications. The dependent variable for all specification is the cash ratio measured as the ratio of cash and marketable securities to total assets. As discussed in Bates, Kahle, and Stulz (2009), the most traditional measure of cash holding is cash to total assets. Other alternative measures of cash holdings include cash to net assets, where net assets is measured as total assets minus cash, logarithm of cash to net assets, cash to sales, and cash to bond outstanding. As robustness checks, in unreported results, we employ all alternative definitions of cash ratio. Our results are not affected by the different definitions. In the baseline results, we stick to the conventional definition of cash to total assets as the measure of corporate cash holdings. Specification 1 of Panel A estimates the effects of CDS with only cash flow related controls, including cash flow risk and the ratio of cash flow to total assets. Firms may determine leverage, cash holdings, payout policy, and investment policy simultaneously. Following Opler et al. (1999), Dittmar et al. (2003) and Gao, Harford, and Li (2012), in specification 2, we estimate a model of cash by including other determinants of cash holdings but excluding leverage, dividends, capital expenditures, acquisition and R&D from the set of explanatory variables. Specification 3 include more firm characteristics as determinants of cash holdings. CDS Trading has positive coefficient in specification 1, suggesting that CDS contracts lead to higher cash ratios. Moreover, the effect of CDS Trading is significant at 1% level. The economic magnitude is also large: compared to sample mean cash ratio of 8.2% for CDS firms, the 1.7% change in cash ratio after CDS introduction in specification 1 represents 20.7% increase in cash ratio. 12 Specification 2 and 3 reports similar findings after controlling for more firm characteristics. The magnitude of the coefficient for CDS Trading is even higher. CDS 12 A related question is that whether the 1.7% change in cash ratio is enough for the firm to tackle the potential liquidity need generated by the tougher CDS protected creditors. However, it is difficult to measure this potential liquidity need directly due to the data limitation. One possible approach is to use the amount of debt due in one year as a proxy. 12

14 trading increases the corporate cash holdings; against the average cash ratio of 8.2% for CDS firms, the 2% change in cash ratio after CDS introduction in specification 3 represents 24.4% increase in cash ratio. Since firms can have different unobservable characteristics that affect corporate cash holding policy, we control for firm fixed effect in all model specifications. CDS Firms can be different from non-cds firms on various dimensions. As robustness checks, instead of using firm fixed effect, we also replace the firm fixed effect by adding CDS Firm as an additional control. Similar to Ashcraft and Santos (2009), Saretto and Tookes (2011) and Subrahmanyam, Tang, and Wang (2012), CDS Firm is a dummy variable that equals one for firms with CDS trading at any point during our sample period. So CDS Firm can be used to control for the unobservable difference between CDS and Non-CDS firms. Since firms in different industries can vary from one another on many dimensions, we control for industry fixed effects in all model specification with CDS Firm dummy. Our results are not materially affected by replacing firm fixed effect with CDS Firm and industry fixed effects. The coefficients on control variables are consistent with prior findings. Firms with high cash flow risk, as measured by Industry Sigma, hold more precautionary cash. The coefficient for Cash Flow/Assets is not significant in specification 1. But it is positive and significant in all other model specifications, i.e. firms with higher cash flow build more cash holdings. Market to Book has positive and significant coefficients. It indicates that firms with higher investment opportunity have higher cost to be financially constrained, and therefore they tend to hold more cash. The negative sign for Size relates to the economic scale to holding cash: large firms hold less cash. Net Working Capital/Assets decreases cash holdings, since Net Working Capital represents assets that substitute for cash. The coefficient for Capital Expenditure is negative and significant, because capital expenditures create asset that can be used as collateral, reducing precautionary need of cash holding. The sign on Leverage is negative since firms may use cash to reduce leverage. R&D/Sales is another measure for growth opportunity. Firms with high R&D have greater cost to be financially constrained, and therefore hold more cash. Acquisition Activity is expected to have the same sign as Capital Expenditure since acquisitions and capital expenditures would seems to be substitutes. The only coefficient differs from expectation is for Dividend Dummy. Firms with dividend payment were expected to hold less cash because they are likely to be less risky firms and less financially constrained. However, in our sample, we find positive sign for Dividend. This may be due to the changes in the relationship between cash holdings and firm characteristics over time as discussed in Bates, Kahle, and Stulz (2009). Our main analysis focus on determinants of the level of cash. We find that firms with CDS trading on their debt tend to hold more cash than non-cds firms. In panel B, we focus 13

15 on the changes in cash ( Cash/Assets), which can help us understand how the difference in cash occurs. Following Almedia, Campello, and Weisbach (2004), Bates, Kahle, and Stulz (2009) etc., we estimate a model of annual change in cash. 13 In addition to the variables in the baseline model, we also include the lagged cash ratio to allow for partial adjustment of the cash ratio to the optimal level. Panel B reports the results. Similar to the baseline results in Panel A, specification 1 only includes cash flow related variables, while specification 2 and 3 include more firm characteristics related to cash holdings. The coefficient for CDS Trading is positive and significant in all model specifications. This indicates that, in a given year, firms with CDS trading on their debt add to their cash reserves by more than do non-cds firms. Moreover, lagged cash ratio decreases the change in cash holdings. Similar to the findings in previous literature, Size is positively related to changes in cash holdings. The coefficient on other factors are generally consistent with the baseline model for level of cash in Panel A. In the rest of the analysis, we focus on the level of cash model to identify the CDS effect. In summary, Table II presents our baseline analysis of determinants of corporate cash holdings. Consistent with expectation, the results indicate that corporate cash holding increases after CDS introduction. However, to establish the causal relationship between CDS trading and cash holdings, it is critical to appropriately control the endogeneity of CDS trading. In the following subsections, we formally address the endogeneity issue, using three alternative econometric approaches. C. Propensity of CDS Introduction The endogeneity of CDS trading complicates the interpretation of CDS impact on cash holdings. Investors may expect the increase in cash holdings, and initiate CDS trading on this firm. We control for firm fixed effects in all model specifications, which accounts for the time invariant difference in characteristics between CDS and Non-CDS firms. Moreover, Li and Prabhala (2007) and Roberts and Whited (2012) discuss the econometric approach for endogeneity, including difference-in-difference estimator, propensity score matching, and instrumental variable approach. We follow Ashcraft and Santos (2009), Saretto and Tookes (2011), Subrahmanyam, Tang, and Wang (2012) and other studies with similar endogeneity concerns for CDS introduction, and re-estimate the CDS effect after controlling for the selection into CDS sample. To implement these approaches, we first estimate a model for the selection of firms into 13 It is difficult to model change in cash with quarterly data due to the seasonality effect. 14

16 CDS trading. In addition, we need to identify factors that affect the introduction of CDS trading, but not directly relate to the cash holdings of firms. We use Bond Turnover and Analyst Coverage as instruments. Bond turnover measures the trading activities in firms bond, which is closely related to the trading demand in firms credit (Saretto and Tookes (2011)). Analyst Coverage is a proxy for the information environment for the firm, which also relates to the propensity of CDS introduction (Subrahmanyam, Tang, and Wang (2012)). Both of these two variables, however, are not expected to affect firms liquidity policy directly. We further include firm characteristics such as size, leverage and other cash related controls in CDS predicting models. The CDS predicting models are estimated with probit model with dependent variable equals one after the introduction of CDS trading, and zero otherwise. For CDS firms, we include all firm observations from 1997 until the first quarter of CDS introduction. Then we add the observations for all non-cds firms in Compustat during the sample period The model is estimated with quarterly observations. The CDS trading prediction results are reported in Table III. CDS prediction model 1 only include Size as the predictor. The positive significant sign for size indicates that larger firms are more likely to have CDS trading. The pseudo-r 2 is 30.58% with only size and fixed effects. Model specification 2 and 3 add instrumental variables Bond Tunover and Analyst Coverage as determinants. Besides of size, Bond Tunover and Analyst Coverage are also significant determinants of CDS trading. This validates the relevants criteria for the instrumental variables. We add a number of cash holdings related factors in Model 4. The results show that CDS trading is more likely for firms with high leverage and high acquisition activity, but with less R&D activities. Overall, it appears that CDS trading is more likely for large firms and firms with better visibility at the time of CDS inception. In later analysis, we use bond turnover and analyst coverage as instrumental variables. The results in Table III show that these instruments satisfy the relevance condition as they jointly predict CDS trading, even after controlling for other variables. In the following three subsections, we apply these CDS selection models to our difference-in-difference analysis, propensity score matching, and instrumental variable approach to address endogeneity concerns. D. Difference-in-Difference Analysis After constructing the propensity score matching sample, we first conduct difference-indifference analysis to identify the treatment effect, i.e. the introduction of CDS trading. In constructing propensity score matching sample, as robustness checks, we use all four prediction models for CDS trading. Additionally, we use three different propensity score matching criteria to choose matching firms: (1) the one Non-CDS firm with the nearest distance, in 15

17 terms of propensity score, to the CDS firm; (2) the one firm with nearest propensity score but within a difference of 1%, and (3) the two firms with propensity scores closest to the CDS firm. In the analysis, we investigate two windows for the CDS effect: year t 1 to year t + 2 and year t 1 to year t + 3. We focus on the change in cash to total asset ratio, which is a traditional measure of cash holdings. The results are shown in Table IV. It indicates that the cash ratio difference-indifference estimates are both statistically and economically significant for the (t 1, t + 2) and (t 1, t+3) event windows using all model specifications, with one or two matching firms. For example, when we use Model 4 to choose the nearest-one propensity score matching firm, cash ratio is 1.8% higher after CDS introduction relative to Non-CDS matching firm in (t 1, t+3) event window. Recall that the average CDS firms has a 8.2% cash ratio. Such an increase in cash holding is substantial. E. Propensity Score Matching Propensity score matching is a simple approach to address the endogenetiy issue. For each CDS firm, we find one non-cds matching firm with the similar propensity score for CDS trading. We then run the cash holding analysis on this matched sample. Different from the baseline model with all Compustat non-cds firms as control group, firms in the restricted sample are more comparable with each other. We use different CDS prediction models to assess the robustness of my propensity score matching results. These models produce different matching samples due to data availability for each prediction model to calculate propensity scores. Table V presents the regression results. In all specifications, the coefficient estimates for CDS Trading are significantly positive. This indicates that corporate cash holdings increase after CDS trading. The economic magnitude is also large. : compared to sample mean cash ratio of 8% for this restricted sample, the 2.7% change in cash after CDS introduction in CDS prediction model 3 represents 33.75% increase in cash ratio. However, propensity score matching approach is only effective in controlling for the observable difference in characteristics between treatment and control groups. It is possible that there is an unobservable variable that drives both CDS introduction and corporate cash holdings. Then propensity score matching cannot effectively address the endogeneity in this setting. In next section, we use the instrumental variable approach to further address the endogeneity issue. 16

18 F. Instrumental Variable Approach The introduction of CDS trading may not be exogenous. It is possible that firms with greater future cash holdings are selected for CDS trading. There could also be unobserved omitted variables that drive both the selection of firms for CDS trading and cash holdings. To allow for time-varying unobserved heterogeneity across firms, we resort to instrumental variable approach, where the indicator variable CDS Trading is treated as endogeneous. Specifically, the process of cash holdings and the CDS contract status of a firm can be modeled as follows: Cash = Xβ + β 1 CDS Trading + ɛ, (2) CDS Trading = Zγ 2 + ω, CDS Trading = 1, if CDS Trading > 0; CDS Trading = 0, otherwise. Similar to the baseline model, the dependent variable is the cash ratio, measured by cash and marketable securities to total assets. X is a list of control variables. The coefficient of interest is β 1 which captures the impact of CDS on corporate cash holdings. The variable CDS Trading represents the latent propensity of a firm having CDS trading on their debt. CDS Trading is allowed to be endogenous due to corr(ɛ, ω) 0. For identification, we include instrumental variables that affect a firm s propensity of CDS introduction, but do not affect its cash holdings directly other than through the impact of CDS introduction. Therefore, Z in equation (9) include instrumental variables in addition to a full overlap with vector X. Our choice of the instrumental variables is econometrically and economically motivated. In the analysis, we use both Bond Turnover and Analyst Coverage as instruments. Econometrically, the instruments need to satisfy the relevance and exclusion restrictions. The relevance condition is met as Table III shows that CDS trading is significantly associated with Bond Turnover and Analyst Coverage. While it is impossible to test the exclusion restriction formally (Roberts and Whited (2012)), in unreported results, we find that Bond Turnover and Analyst Coverage cannot predict cash holdings by the sample firms. Therefore, they seem to satisfy the exclusion restriction. Bond Turnover and Analyst Coverage are economically sound instruments, because they are associated with the credit trading demand in, and with the information on, the reference entities. Beyhaghi and Massoud (2012) suggest that the selection of a firm for CDS trading may be contingent on monitoring costs. Thus, firms with more active bond trading and analyst coverage (which is also associated with equity trading) are more likely to trigger CDS trading. On the other hand, bond trading and analyst coverage 17

19 cannot influence corporate decisions about cash holding in a direct manner. Table VI presents the results. In the table, we show the second stage results, and refer to those in the the first stage in Table III. Similar to the baseline results in Table II, we use three different model specifications. Specification 1 only includes cash flow variables: Industry Sigma and Cash Flow/Assets. Therefore, in the first stage for the probability of CDS Trading, we only use cash flow related variables and the instrumental variables. Then the fitted value of CDS Trading is included in the second stage analysis for the determinants of cash holdings. In this case, the coefficient of CDS Trading is positive but not statistically significant. This may be due to the imprecise estimation of the CDS Trading. As we can see from Table III, Size is the most important determinant of the propensity of CDS trading. In model specification 2 and 3, we include size and other firm characteristics in predicting CDS introduction. The second stage results are presented in Table VI. We find that CDS Trading has a positive and significant coefficient estimates in these two model specifications. In other words, firms are holding more cash after the introduction of CDS trading. The results so far suggest significant effects of CDS on corporate cash holdings. The CDS effect is robust to controlling for the endogeneity of CDS trading, including differencein-difference analysis, propensity score matching and instrumental variable approach. In addition to these econometric approach, in next section, we further construct continuous economic measure of CDS exposure, which is less affected by the selection issue. We also investigate cross-sectional difference in CDS effect. V. Firm Characteristics and CDS Effect: Implications for Mechanisms The effect of CDS on corporate cash holdings may change with firm characteristics. In this section, we investigate whether CDS impact is more significant for firms with larger CDS outstanding, those with unrated/non-investment grade ratings, and entities with multiple creditor relationship and line of credit. Implications for the mechanisms are also discussed. A. Outstanding CDS Positions Instead of using a regime variable CDS Trading, which equals one after the introduction of CDS trading, we utilize the detailed information about the notional amount of CDS contracts to construct continuous CDS exposure measure. The continuous economic variables also 18

20 help further address the self-selection of CDS trading. As pointed out by Li and Prabhala (2007), the magnitude of the selection variable (selection for CDS trading) introduces an independent source of variation and helps with the identification of the treatment effect, while ameliorating the self-selection concern. In addition, the continuous CDS outstanding measure is also a proxy for the severity of CDS effect. The larger the CDS outstanding, the greater the benefits to the CDS protected creditors, and therefore the tougher of these creditors in the process of re-negotiation. Moreover, CDS outstanding relates to CDS market liquidity. CDS spread for firms with liquid CDS market is more sensitive to new information, such as the firm s liquidity status. Therefore, the feedback effect is more severe for firms with larger CDS outstanding. Thus, both the tougher creditor mechanism and the CDS feedback mechanism predict firms have greater incentive to hold cash reserves when there are more CDS contracts trading on their debt. Specifically, we measure the corporate CDS outstanding in two ways. First, we use the Number of Live CDS Contracts, measured by the number of CDS contracts initiated, but yet to mature, as of the observation quarter. 14 This variable measures the breadth and consistency of CDS trading activity and should be positively correlated with the extent of CDS effect. Second, we calculate the ratio of the notional dollar amount of CDS contracts outstanding to the total dollar amount debt outstanding at the same time, CDS Notional Outstanding/Total Debt. We scale the CDS position by total debt to relate the dollar amount of CDS outstanding to creditors demand. We conjecture that firms with greater relative proportions of CDS outstanding are more exposed to CDS effect. The estimation results are presented in Table VII. The cash analysis is conducted with the two continuous CDS exposure measures. Specification 1 use the Number of Live CDS Contracts as the measure of CDS impact on cash holdings. The positive and significant coefficient indicate that CDS trading increases cash holdings. Specification 2 employs the ratio of CDS Notional Outstanding/Total Debt as the CDS exposure measure. We continue to find a significant positive coefficient for the CDS effect measure. These findings suggest that the greater the CDS exposure is, the higher the corporate cash holdings will be. B. Credit Ratings and CDS Effect The effect of CDS on corporate cash holdings may change with firm credit ratings. Almedia, Campello, and Weisbach (2004) find that firms with capital market frictions, i.e. financially constrained firms, are more likely to save cash out of cash flow. Similarly, financially con- 14 Since CDS contracts are defined by their maturity, rather than their maturity date, new contracts are created potentially each trading day, depending on the level of trading activity. 19