Do Credit Default Swaps trigger
Risk-Shifting?
University of Amsterdam, Amsterdam Business School
MSc Business Economics, Finance Track
Master Thesis
Student: QinYan
Supervisor: Dr. R. Matta
July 2015
Statement of Originality
This document is written by Student Qin Yan, who
declares to take full responsibility for the contents of this
document.
I declare that the text and the work presented in this
document is original and that no sources other than those
mentioned in the text and its references have been used in
creating it.
The Faculty of Economics and Business is responsible
solely for the supervision of completion of the work, not
for the contents.
Abstract
This paper offers empirical evidence of the impact of Credit Default Swaps (CDS) on firms’ risk choices. Study on data from 2001 to 2014 reveal significant positive
correlation between CDS and risk-shifting, solidifying the conception that CDS increases the probability of default (ex post and ex ante). However, further investigation, taking into account the financial status of the firm, shows that CDS refrains financially
constrained firms from taking on riskier investment profiles in volatile environments. To integrate CDS as an useful and beneficial instrument into the economy, these findings call for deeper understandings of CDS and mechanisms to improve regulating the CDS market.
Content
Introduction ...1
Section 1. Literature Review ... ３
Section 2. Empirical Design and Data ... ７
2.1 Empirical Foundation ... ７
2.2 Adjustments ... ８
2.3 Diff-in-Diff Analysis ... １０
2.4 Data ... １２
Section 3. Results ... １４
3.1 Baseline Regression and CDS ... １４
3.2 Firm Financial Status ... １５
3.3 Diff-in-Diff Analysis ... １６
Section 4. Robustness Test ... １７
Section 5. Concluding Remarks ... １９
References ... ２１
Tables ... ２３
Figures ... ３０
Introduction
Recent 2008-9 crisis has insinuated attention to a specific class of derivatives: Credit Default Swaps (CDS), as downfalls of various sectors in the economy are suspected to be driven by the use of this financial product. The first CDS contract was introduced by JP Morgan in 1997 and the market grew exponentially ever since. According to the
International Swaps and Derivatives Association (ISDA), the market value amounted to $58 trillion in its peak in 2008. This exuberance seems to go hand in hand with the
decline in the financial markets leading up to the eventual unraveling of the crisis in 2008. Instinctively, it appears prudent to associate these two and to suspect a certain affiliation. In fact, many concerns regarding the functioning of CDS contracts and its role in the financial market have been raised.
In its simplest form, CDS is best thought of as an insurance contract (Arora et al, 2012). The seller of the contract pays a prespecified compensation to the buyer on the event that an entity defaults on its debt. However, according to Stulz (2009), CDS and regular insurance contracts differ substantially in two important aspects. First, holding bonds of a company is not a prerequisite for buying a CDS of these bonds, whereas with insurance contracts, one is usually obligated to be directly exposed to certain economic risks to obtain an insurance. Second, most insurance contracts cannot be traded. In contrast, CDS contracts are frequently traded on over-the-counter markets. These implications form the basis of a CDS market, which by now is one of the largest financial products in the fixed-income category. The intended task of CDS contracts, as Angelini (2012) points out, is to allow credit risks to be separated from the underlying credit relationship and be traded independently. This leads to a broader distribution of credit risks and thus should improve the financial system’s overall ability to absorb risk and the allocation of capital, thereby fueling the growth of the economy.
However, skeptics question the proclaimed benefits of CDS and blame it as the most prominent villain contributing to the crisis. Hu and Black (2008) advanced the empty
２
might have low incentives to cooperate in out-of-court restructuring of financially distressed firms since in case of formal default they will receive an immediate
compensation. In the event of “overinsurance” (protection compensation is higher than the amount of debt that can be recovered in default), lenders might even have incentives to force the distressed entity into bankruptcy and collect a large payment. Facing tougher negotiators and less monitoring, equity holders are incentivized to take on excessive risks, as they benefit from the upside while bondholders bear the costs of downside risk. This risk-shifting behavior in turn amplifies the financial condition of firms and eventually leads to higher bankruptcy probabilities.
Contrasting arguments are brought up by Rocha (2014), asserting that CDS might
actually alleviate the risk-shifting problem. He builds his argument on two effects of CDS. On one hand, CDS boosts firm debt capacity. The perk of accessing external funds in more favorable terms materializes from the tougher hand in renegotiations CDS grant to insured creditors, and the consequent reduction in the probability of strategic defaults (Bolton and Oehmke, 2010; Campello and Matta, 2013). In particular, firms that ought to take on negative NPV projects due to debt overhang may benefit more when creditors purchase insurance. Therefore, increased debt capacity may weaken the risk-shifting incentive. On the other hand, financially healthy firms are incentivized to take excessive risks to avoid outcomes associated with overinsurance (Campello and Matta, 2013). As a consequence, to learn whether CDS provoke higher risk appetite, one should also
consider the financial status of the firm.
The aim of this paper is to investigate whether CDS contracting administers risk-shifting of the reference entities, taking into account the financial status of the firm. To
investigate this, we will largely adopt the methodology applied in Eisdorfer (2008), using the relationship between investment intensity and expected volatility as the main
indicator for management tendencies with regard to risk exposure. Adding to the model the net notional amounts of CDS, the financial status of the firm and several covariates such as firm size, tangibility, market to book ratio and leverage. Moreover, to further gauge the difference in effect of CDS on investment intensity under various levels of financial health of the firm facing either high or low expected volatilities, we conduct a
３
diff-in-diff analysis using subsamples constructed as suggested by Rocha (2014). To test for robustness, propensity score matching is applied with having CDS written against firms’ debts as the treatment effect. The dataset will cover financial specifics from 2001 to 2014.
This research occurs relevant to me by virtue of the relatively young existence of CDS contracts. And as such, little work had been done in investigating the functioning of CDS and its role in the financial market. In particular, due to insufficient knowledge about the channels through which CDS exerts its up- and downside influences, and in fact what these up- and downsides are, no mechanisms are yet developed to guide CDS trading to be predominantly beneficial for the economy. The goal is to shed some light on the potential benefits and downsides of CDS which could contribute to improve the regulation of CDS markets.
The remainder of this paper proceeds as follows. Section 1 offers a brief summary of previous literature on this matter. In section 2, we provide details on the design of the empirical setup and data description. In section 3, the results are presented and discussed, section 4 tests the model for robustness and section 5 concludes.
Section 1. Literature Review
The most prominent theoretical framework addressing the effect of CDS on creditor debtor relations is the empty creditor problem proposed by Hu and Black (2008). They point out that creditors possessing CDS as insurance against default lose incentive in preserving the continuation of the firm. Instead, creditors might potentially even be better off forcing the firm into bankruptcy and collect the compensation causing inefficient liquidations of firms. In the context of our research, the empty creditor framework strongly suggests that CDS incentivizes the management of firms to risk-shift, as monitoring efforts are lowered due to the disinterest of creditors in firms financial
４
strategies and the increased enticement for firms to enlarge risk exposure as consequence of the strengthened bargaining position of creditors in possession of CDS.
Building on this framework, Bolton and Oehmke (2011) elaborated a limited
commitment model to analyze the behavior triggered by CDS. Based on this model, they argue that CDS adds value by acting as a commitment device for borrowers ex-ante. Firms with limited ability to commit to repay the debt gain an increase in pledgeable income, since creditors would be tougher in renegotiations and lowers the stimulus to strategically default. However, when creditors are free to choose the level of credit protection (Campello and Matta, 2012), they generally over-insure which leads to excessive incidents of bankruptcy and too little renegotiation. The lesson we take away from this study is when evaluating the pros and cons of CDS, one should balance the effects ex-post and ex-ante.
Despite interest in the impact of CDSs on creditor-borrower relations, for a long time literature lacked a theoretic model that examines important questions about these contracts. Campello and Matta (2012) constructed such a model with real life complexities (borrowers choose the riskiness of their investments and verification is imperfect) to address this issue. They show through this model that lenders choose the amount of CDS protection to inflect their bargaining position in case of distress which in turn has an impact on project choices of borrowers. The intuition is as follows. CDS strengthens the bargaining position of the creditor in distress renegotiations. In case of overinsurance, creditor liquidates the borrower when project fails which prevents
borrower from capturing rent in the bad state and maximizes debt repayment in the good state. This implies that the borrowers expected payoff is minimized. In case of just-insurance however, the lender does not commit to liquidation unconditionally. As a result the borrower can obtain a fraction of the restructuring value when funds verification is imperfect, implying a lower debt repayment in the good state. All in all, the creditor will prefer overinsurance in the case of high likelihood of success of the investment, whereas the borrower might profit from choosing a riskier investment profile so as to induce the lender to just-insure. Campello and Matta (2012) provide an innovative insight to the relationship between risk-shifting and CDS contracting. They point out that besides the
５
empty creditor problem, which creates inefficiencies ex post liquidation, the choice of a riskier profile caused by CDS forms the initial impulse leading to higher default
probabilities in the first place.
Rocha (2014) advertised the beneficial aspects of CDS by arguing that it alleviates the risk-shifting problem in financially constrained firms. The author points out that CDS boosts firm debt capacity which may weaken the risk-shifting incentive. This hypothesis is confirmed by the empirical analysis based on data around the 2007-2008 crisis. He concludes that the presence of CDS causes constrained firms to be more conservative when the economic environment is uncertain and help stabilizing the economy in
downturns. These findings lead us to believe that CDS affects firms with different levels of financial health quite differently. When examining the changes in risk profiles due to CDS, one should distinguish between distressed and financially unconstrained firms. Also, we learn from this study that firms’ behavior with regard to risk exposure might differ when economic prospects are stable versus uncertain when CDS is involved. These inferences should be borne in mind when constructing our empirical setting.
In another empirical study about the effects of CDS on credit risk, Subrahmanyam et al. (2011) argue that CDS is a two-edged sword. On the one hand, CDS allow creditors to hedge their credit risk; On the other hand, lenders may not be as vigilant in monitoring the borrowers once their credit exposures are hedged. Furthermore, CDS-protected
creditors are likely tougher during debt renegotiations, once the borrowers are in financial distress, by refusing debt workouts and making borrowers more vulnerable to bankruptcy. Their empirical results based on a comprehensive sample from June 1997 to April 2009 show that bankruptcy risk increase significantly after CDS inception. On the same grounds, they also speculate that the improved bargaining position of creditors instead of lacked monitoring was the main channel through which the empty creditor problem facilitates. Interestingly, they point out that the effect of CDS on credit risk is possibly related to the amount of CDS outstanding, presenting evidence showing that firms with relatively large amounts of CDS contracts outstanding are more likely to be adversely affected by CDS trading. This research solidifies the empty creditor theory, in particular stressing the important role played by the strong bargaining position of creditors as
６
consequence of CDS trading. Another useful point we take away is that the notional amount of CDS contracts outstanding is a quite relevant measure when it comes to quantifying the effects of CDS on risk-shifting, implying that we should include this variable in our analysis instead of using just a dummy.
Furthermore, Kim (2011) provides evidence that the introduction of CDS indeed alters the creditor debtor relationship. In the existence of CDS, when firms write incomplete debt contracts, its inability to commit to not default strategically will result into higher costs of borrowing that eventually will be borne by the firm. CDS reduces the cost of borrowing ex ante by reinforcing creditors’ bargaining position during distressed renegotiations. The author identified both theoretically and empirically this benefit of CDS. More specifically, he shows that the firms a priori most likely to face the limited commitment problem (i.e. firms with high incentive to strategically default) encounter a relatively larger reduction in their corporate bond spreads after the initiation of CDS. Inferences drawn from this research highlight the ex-ante benefits of CDS contracts and contradicts the view that empty creditors harm financial stability per se.
Contrary to what has been claimed by Kim, Ashcraft and Santos (2009) found that the average debtor has not benefited from CDS contracts in the form of a reduction in the cost of bond or loan funding. Even significantly adverse effects were uncovered. Notably, the sort of firms that are believed to benefit the most (risky and informationally opaque firms) appear to have been adversely affected by the CDS market, while the types of firms expected to benefit the least (safe and transparent firms) have benefited from a small reduction in both bond and loan spreads. The authors argue that the adverse outcomes could be explained by another implication of the CDS market. Through the CDS market, banks could cut off their credit links to borrowers in a way that is unnoticed by firms and investors. Without direct exposure to the borrowers, banks have low to no incentives to monitor their behavior. In anticipation of this mechanism, institutional and private creditors may demand higher compensation for their loans. This effect should be especially eminent for firms where monitoring is most valuable, risky and
informationally opaque firms. Although not cemented with empirical evidence, the authors logically deduced the implications of the CDS market that causes less monitoring
７
by creditors. With respect to our research, this would suggest a higher likelihood of risk-shifting by firms with trading CDS.
Overall, the literature seems still inconclusive about the true effects of CDS on management behavior regarding their propensity to risk-shift in particular and on the economy as a whole in general. The ongoing debate in the US and Europe mainly concern the regulation of the CDS market. Should speculative use of CDS be banned completely? Should there be an exchange instead of over-the counter trading of CDS? Coinciding with our assessment of the empirical and theoretical literature, no consensus is yet reached. As for our research, we can expect CDS to affect the risk appetite of firm equity holders which is reflected in their decision making. Whether this relationship is positive or negative is left to be investigated. We have also learned a few fundamental components in the functioning of CDS that should be considered when doing analyses. The means of empty creditors is not the only channel through which CDS is affecting risk-shifting and default probability. Moreover, the financial state of the firm should also be part of the consideration, hypothesizing that distressed and healthy firms behave differently with respect to risk exposure. When possible, notional CDS amount
outstanding should be used to inspect the effects of CDS on firms’ risk choices instead of a simple dummy for CDS, as the magnitude of this measure might be relevant in the impact quantifying process.
Section 2. Empirical Design and Data
2.1 Empirical Foundation
Our research in great lines follows Eisdorfer (2008). In his study, the author examined the risk-shifting behavior in financially distressed firms by studying the relationship between investment and expected volatility. Under the real options framework, the investment decisions of firms involves a trade-off between the immediate cash flows that can be obtained by exercising the investment at short notice and the extra information regarding
８
the true value of the investment project that can be gathered by delaying the investment. As with options, the value of delaying an investment decision increases with the
uncertainty about the projects cash flows (expected volatility). Therefore a negative relationship between expected volatility and investment is expected. This view is
generally supported by empirical literature (e.a. Pindyck and Solimano, 1993; Episcopos, 1995; Leahy and Whited, 1996 and Bulan, 2003). However, when firms engage in risk-shifting, the opposite effect is expected. Since high risk projects are likely to benefit the equity holders of a distressed firm who control the investment policy, higher volatility might increase investment. Eisdorfer suggested the following hypothesis: There is a less negative or even positive relationship between risk-shifting and volatility in distressed firms. To test this hypothesis, he used the following regression model for four separate groups: a) Actual investment+ healthy firms, b) actual investment+ distressed firms, c) industry adjusted investment+ healthy firms and d) industry adjusted investment+ distressed firms. 0 1 2 3 4 5 6 7 8 _ Re
Investment ExpVol Log size Market to book Leverage LaggedCF cessionDummy DefaultSpread InterestRate 2.2 Adjustments
The main aim of our study is to test whether CDS is interacting with the firms’ incentives to alter its risk profile. To be more specific, we want to examine whether the existence of CDS causes the relationship between investment and financial status given various levels of uncertainty to alter. Hence, the CDS amounts outstanding and the Altman z-score as proxy for the financial status of the firm are relevant additions to the model. Furthermore, instead of using the Altman z-score as an indicator to create subsamples in which firms are either distressed or healthy (with threshold value 1.81) as in Eisdorfer, we use the value of the z-core itself as a continuous variable for financial health. Since, first of all the threshold value of 1.81 might be outdated and inaccurate. Secondly, different industries have different financial structures, and as such, adopting one threshold value for all firms might be too rigid. Another point worth mentioning is how to calculate the
９
expected volatility. This is obtained using a Generalized Autoregressive Conditional Heteroskedasticity model or GARCH(1,1), which is a common procedure in the financial literature (e. a. Rocha, 2014; Saretto and Tookes, 2011). Having retrieved these variables, we can then proceed to create interaction terms between CDS, z-score and expected volatility to be included in the model. Our investigation therefore is built on the following specification:
0 1 2 3 _
Investment ExpVol CDS z score Interactions Controls
In this model, Investment is the investment intensity measured as the ratio of capital expenditures to net property plant and equipment (PP&E). ExpVol is the expected volatility calculated from the S&P500 index returns using GARCH (1,1). CDS is the notional amounts outstanding as percentage of total long term debt. Z_score is the
Altman z-score as calculated in Altman (1968). The controls include leverage, tangibility, market-to-book ratio and firm size. Leverage is measured as the ratio of total of long-term debt to total assets, tangibility as PP&E divided by total assets, market-to-book ratio as market value of equity divided by book value of equity and firm size is approximated by summing up total assets and liabilities.
As hypothesized, we would expect a negative estimate of the parameter for expected volatility, because of the assumption that the value of delaying an investment decision increases with the uncertainty about the projects cash flows (expected volatility). Based on the literature review we are yet unable to predict he sign of the CDS coefficient estimates, however, it is quite likely to be significant. Estimates of the parameters for the interaction terms are of particular interest, as they convey information about how CDS affects firms’ decisions. For example, if the interaction term between CDS and expected volatility were to be negative, it would suggest that CDS contracts prevents firms to increase investments during more volatile periods. Hence, much attention shall be paid to these parameters when analyzing the results. For the control variables, we would expect positive estimates for firm size, as it seems sensible to assume higher investment
１０
that the firms with the highest investment intensities are those with high R&D expenses which tend to have lower PP&E to asset ratios.
2.3 Diff-in-Diff Analysis
Implementing the above approach should give us a general picture of how CDS impacts the relationship between investment and expected volatility. However, when
incorporating financial status of the firm into the analysis, the specification above is likely to face complications. First of all, including a firm financial status indicator would create a three way interaction term. Although there are ways to make sense of the results, the interpretation of such coefficients remains far from straightforward and a tedious task. Secondly, as we want to determine whether the interaction between risk exposure and expected volatility given the financial status is influenced by CDS, and CDS alone, it would be plausible to isolate the role played by CDS in the process. To achieve this objective, we can broadly emulate the procedure employed by Rocha (2014). Contrasting to common prior literature, where differences in characteristics of CDS and non-CDS firms are compared throughout a panel dataset, Rocha (2014) elected to carry out an event study based on Diff-in-Diff regressions of investment intensity on financial status around the financial crisis of 2007-2008 for samples of CDS firms only. Again, the empirical setting of Eisdorfer (2008) is adapted as a basis for the tests, with one major modification. Instead of Altman z-score, the maturing amount of long-term debt in the year following the beginning of the financial crisis is used as the indicator for financial status. This measure is derived from Almeida et al (2011) and suggests that the more financially constrained firms (more distressed) are firms with higher amounts of debt maturing right after the commencement of the crisis. A common caveat for the
conventional measure of financial health is that it would simultaneously have an impact on investment and the risk exposure decisions without illustrating a clear causal direction or how much of the effect can be traced back to the differences in financial status. Using
１１
the maturing loan measure would alleviate this issue since it allows for an exogenous change on firms’ financial health and helps detaching its role from other factors. More specifically, adapting this variable allows us to study the role played by CDS and make direct comparisons between firms that were affected the most by the crisis (large amount of debt due, without new sources of funding and become financially constrained) and the firms that were financially scarcely affected by the crisis (smaller amounts due, no
significant impact on investment policies). Another benefit that arises from the use of this measure is that the processing of CDS information is restricted to a time when the CDS market is more mature for the majority of the firms. Unlike previous studies where the empirical setting covers the whole time frame since the inception of CDS trading (2002), this specification takes into account the learning process of the firms’ management and its eventual reaction regarding risk choices to the consequence of having CDS written against the firms debts, which is exactly what we are interested in. Downside of this approach, however, is the limited time frame and the reduced number of usable observations.
In Rocha (2014), the 2007-2008 sample is split into two subsamples depending on the expected volatility faced by every firm. Setting 2007Q3 as the beginning of the crisis and the event date, the period before the shock is represented by 2007Q2 and the period after the shock by 2008Q2. The reason for choosing the same quarter is to avoid seasonal frictions and characteristics of the cycle which might bias the outcomes. Notice that the period right before the shock has relatively low volatility and vice versa for the period after the shock. Unfortunately, calculating firm specific expected volatilities based on Black and Scholes and GARCH(1,1) is very time consuming and given the time scope of this research hardly attainable. Furthermore, our data on CDS within this given time frame is markedly limited making the emulation of this exact procedure infeasible. Hence, we have to create an alternative simplified setting where the essential components of Rocha (2014) are still integrated while the obstacles can be circumvented. When studying financial events in the recent years, the Greek bailout of 2012 seems to be a suitable replacement as a ‘shock’, as sufficient information regarding CDS is available. Our periods of interest would then be 2011Q4 and 2012Q4. As for the expected volatilities,
１２
we can rely on the industry expected volatilities to construct subsamples. In addition to Rocha’s (2014) method, we further divide the subsamples into an upper and lower half based on the notional amounts of CDS outstanding to investigate whether having more CDS written against their debt causes firms to act differently regarding their choice of risk exposure. All subsamples undergo the following regression specification:
0 1
Investment LTDdue Controls
Here, ΔInvestment represents the difference investment intensity in the given quarter with the last quarter, LTDdue is the amount of long-term debt due in one year as a
percentage of total long-term debt outstanding, taking on value 1 for the upper half of the sample and 0 for the bottom part of the sample. The controls include tangibility, leverage, logarithm of firm size and the Altman z-score as additional covariate.
Inferring from the results obtained by Rocha (2014), we would expect the estimate of the parameter for LTDdue to be negative and significant for the high expected volatility sample and not significantly different from zero in the low expected volatility sample, which would imply that CDS contracts refrain more distressed firms from engaging in risky activities when faced with uncertainty, while it has no significant impact in periods with low expected volatilities. Moreover, the intercept is expected to have no statistical significance, as it represents the difference in investment intensities after-before for firms in the bottom half of LTDdue (financially unconstrained firms). This would indicate that these financially healthier firms are likely to navigate through the shock without altering their initial risk profiles.
2.4 Data
We draw on mainly three sources to construct our dataset. Market index return data of the S&P500 from CRSP, which is used to calculate the expected volatilities. The resulting line plot of the expected volatilities from 2001 to 2014 retrieved using GARCH (1,1) is shown in figure 1. Considering the time horizon of the firm specific accounting details (which is per quarter), The monthly expected volatilities are converted into quarterly
１３
volatilities by taking geometric averages of the months that are part of the quarters in question. The CDS data are from DTCC, most relevantly the net notional amounts outstanding are what we look for. Unfortunately, due to limited resources, databases like Bloomberg could not be latched on to, where more detailed specifications like initiation dates are accessible. In total, the sample consists of 891 distinct firms with CDS activity. For firm specific accounting data, COMPUSTAT’s quarterly fundamentals were used. All firms with non-missing values for capital expenditure, z-score and total assets that were listed on the S&P500 since 2001 were included in the dataset. The reason why only S&P500 firms were elected is primarily for the sake of homogeneity. Within the dataset, further modification to the sample was made by dropping all observations that can be considered penny stocks. The final sample is comprised of 2.467 different firms and 93.307 firm-quarter observations. The descriptive statistics of the dataset are summarized in Table 1. The table contains three different panels. Panel A shows firm characteristics for all firms enclosed by the sample. Panel B and C show descriptive statistics for CDS and non-CDS firms respectively. It reveals information regarding the mean, standard deviation and the first to third quantiles for each variable. These variables are investment, size, leverage, market to book ratio, tangibility and z-score. Interestingly, we see that, on the whole, CDS firms have lower investment intensities even though they are larger in size on average. Furthermore, we observe that CDS firms are generally more leveraged, which is a possible reflection of the fact that CDS contracts enhance accessibility to external capital. Finally, it is worth noticing that the mean z-score appears to be much higher for non-CDS firms. This might be indicative for more solvent firms to not engage in CDS activities or simply are less inclined to issue debt in general.
For the Diff-in-Diff sample, a separate set of descriptive statistics is generated, as reported in Table 2. Panel A and B present the firm characteristics for CDS firms and non-CDS firms around the event in 2012 respectively. Similar to Table 1, the mean, standard deviation and the first to third quantile for each variable are disclosed. In
addition to the characteristics shown in Table 1, the variable percentage of long-term debt due in one year (LTDdue) is also included in the summary. Again, we see that CDS firms are generally larger in size compared to non-CDS firms and the main sample. We also
１４
detect the same differences in financial characteristics like leverage, tangibility and z-score, although the values are much closer to each other in the restricted sample. Given the limited sample size, it is hard to make inferences about which sample is more constrained. Lastly, looking at the LTDdue stats, we conclude that both CDS and non-CDS firms seem to be similar in the proportion of debt due.
Section 3. Results
Building on the methodology of Eisdorfer (2008) and the modifications assumed in section 2.2, we perform a series of regressions in order to gauge the effect of CDS on investment intensity under various economic circumstances. More specifically, we want to test whether the presence of CDS significantly affects the relationship between expected volatility and investment (risk-shifting).
3.1 Baseline Regression and CDS
In column (1) of Table 3, we explore the correlation between expected volatility and investment. Not surprisingly, the regression outcomes indicate a negative relationship, in line with economic theory and prior literature (ea. Eisdorfer, 2008). With the clustered standard error being significant at the 5% level, this result confirms the conjecture that firms prudently cut back investments whenever the economic environment becomes more uncertain (higher expected volatility) and delaying investment decisions becomes more valuable. Interestingly, we further observe that the market-to-book does not seem to have any impact on investment. This differs from the results obtained by Eisdorfer (2008). Possible explanation for this might be that a considerable number of quarterly
shareholders’ equity were missing in the initial dataset and thus the market-to-book ratio could not be calculated in several cases.
１５
In addition to the baseline regression depicted in column (1), the variable CDS, the notional amount of CDS outstanding as a percentage of total liabilities, and the
interaction term between CDS and expected volatility are incorporated in the regression shown in column (2). While the coefficients and standard errors for the control variables and expected volatility stayed roughly the same, CDS is reported to have a negative coefficient, which is quite notable at first. This suggests that having more CDS notional amount outstanding would lead to less investment, contradicting the idea that CDS lowers borrowing costs and therefore should boost investment. However, this finding is in accordance with the results obtained by Rocha (2014), where he supported the view with empirical evidence that the presence of CDS gives constrained firms incentives to be more conservative when the economic environment is uncertain and in general help alleviate the weight that the empty-creditor imposes on firms with CDS written against their debt. Moreover, Ashcraft and Santos (2009) failed to find empirical evidence of CDS having any significant effect on cost of debt. Hence, this outcome appears to lie within range of reason after all. Regression (2) further documented a positive coefficient for the interaction term between CDS and expected volatility, significant at the 1% level. This suggests that on average, firms with more notional amounts of CDS outstanding tend to invest more in volatile times, hence, to risk-shift. This outcome supports the empty creditor theory, as the result that CDS leads to risk-shifting behavior seems to be in consonance with Hu and Black (2008). However, whether this effect happened ex-ante or ex-post and through which channel it exerted influence remains unclear.
3.2 Firm Financial Status
To further examine the effect of CDS on risk-shifting, we complement the regression specification used in Table 3 with an additional variable z_score, a proxy for firm financial status measured with the Altman score, and the interaction terms between z-score and expected volatility, as well as the interaction term between z-z-score, CDS and expected volatility. Recall that lower z-scores represent more distress, with the cut-off level between healthy and distressed firms being at 1.81 (the cut-off level is not of particular importance in this analysis). The estimates are reported in Table 4.
１６
Dealing with a three way interaction term of all continuous variables instigates
complications in the interpretation of its coefficient. Therefore, a Dawson and Richter (2006) test for differences between slopes is performed to clarify matters. In essence, this method documents the simple slopes between the dependent and an independent variable covered by the interaction term when the other two variables are held constant at different combinations of high and low values. After the simple slopes are computed, the
differences among all pairs of the slopes will then be gauged and tested. The Dawson and Richter method uses a number of moderately-complex formulas while measuring the slopes and differences, a more detailed deduction of the mechanisms goes beyond the scope of this paper, and the interested reader is referred to study Dawson and Richter (2006). As was partly expected, the slope difference tests failed to provide significant results, implying no significant correlation between CDS, financial status of the firm and the firms’ investment intensity. This in turn means that no evidence could be established that indicates a certain impact of CDS on firms risk appetite taken into account various levels of financial status of the firm. For further investigation, the aforementioned Diff-in-Diff study based on the methodology derived from Rocha (2014) should provide more useful insights. Moreover, the coefficient of the interaction between z-score and expected volatility is statistically significant but positive, which differs from the negative
coefficient estimated by Eisdorfer (2008). As a lower z-score implies a higher
probability of default, a positive coefficient on the interactive variable issuggesting that constrained firms are likely to invest less when expected volatility is high.
3.3 Diff-in-Diff Analysis
Table 5 presents the coefficients estimated given the empirical specification described in section 2.3. The same regression is ran for four subsamples: low CDS notional amount-low expected volatility, amount-low CDS notional amounts-high expected volatility, high CDS notional amounts-low expected volatility and high CDS notional amounts-high expected volatility, estimates of the parameters of which are reported in the panels A, B, C and D respectively. This test is conducted complementary to the general model disclosed in the previous sections. Using percentage of long-term debt due as measure for financial health, we find similar results as Rocha (2014). First of all, the constant term appears to be
１７
statistically insignificant, leading us to believe that financially less constrained firms are likely to preserve their investment policies throughout the event period, as the intercept measures the change in investment intensity for firms with low to no debt due. Our parameter of interest is the coefficient reported for percentage of long-term debt due, it measures the elasticity of changes in investment intensity to financial health. We observe that the parameters in question are statistically significant and negative in the high
volatility cases with both low and high CDS notional amounts outstanding. Whereas the same parameters in the low expected volatility samples have no statistical significance. This suggests that CDS has a restraining effect on risk exposure in association with financially more constrained firms when facing an uncertain environment. In case of less volatile expectations, the presence of CDS does not seem to play a significant role. This result is in accordance with the empirical outcomes obtained by Rocha (2014), hence, supporting his conclusion that CDS makes financially constrained firms conservative when the future is uncertain but generally has no impact on investment intensity for both distressed and healthy firms when less volatility is predicted. When comparing high and low CDS notional amounts outstanding, the results are no different. However, as
mentioned before, the small sample size restricts the time frame being studied (only two quarters) and does not warrant external validity. Nonetheless, given the fact that this result coincides with Rocha (2014) and partly with the inferences drawn from section 3.1, one should not discard it as evidence.
Section 4. Robustness Test
Examining core parameters obtained from regressions for structural validity is a common exercise in empirical literature. The main concern in our empirical model arises from potential selection bias. Firms engaging in CDS activity might have specific
characteristics which in turn could have an impact on their investment decisions. To state an extreme example, it could be the case that CDS is only written against the debts of distressed firms, implying that the changes in investment intensity we observe are due to
１８
this selection process instead of being a consequence of CDS itself. To mitigate this problem, propensity score matching could be applied. This statistical technique attempts to estimate the effect of a treatment by accounting for predictor variables that forecast the probability of receiving the treatment in the first place. In our case, CDS is the treatment and we need to establish a set of covariates that predict the propensity score of being selected for CDS trading. Using the probit model suggested by Subrahmanyam et al (2012) a propensity score for CDS trading is calculated. The probit model and the estimated parameters are reported in Table 6. Examining the estimated coefficients, we observe that firms with higher asset values and returns on assets are more likely to have CDS written against their debt. Conventional measures for profitability like EBIT or retained sales per asset show positive outcomes, suggesting higher CDS likelihood by financially healthier firms. While higher leverage ratio also correlates with higher CDS probability, indicative of more risky firms being associated with CDS. Generally speaking, no obvious pattern of CDS self-selection could be detected, allaying our concern of sample selection bias.
Based on the propensity scores obtained from the specification in Table 6, we proceed to match each treatment firm (CDS=1), with firms in the control group (CDS=0). This is achieved by using the nearest neighbor approach. With the newly constructed sample of propensity score matched firms, the same regressions as in section 3.1 and 3.2 are conducted. The results are summarized in Table 7. Column (1) reports the baseline regression, showing largely similar estimates for all parameters as in the unmodified sample. In particular, the signs and magnitudes of the expected volatility coefficients are almost identical. Column (2) and (3) document the regressions involving CDS, z-score and their interaction with expected volatility. Again, the outcomes were mostly
coinciding with the results obtained in Table 4 and 3. Overall, the propensity score matching procedure consolidates our earlier findings.
１９
Section 5. Concluding Remarks
Credit Default Swaps, being the most recent and influential innovation in financial derivatives, is facing resilient criticism in the aftermath of the global financial crisis. Many argue that it harmed financial stability and stress the role it played aggravating the chains of liquidations following the economic demise. Much need to be learned about its functioning, the incentives it provides and the way it affects the availability of credits.
This paper addresses the issue of whether CDS incentivizes risk-shifting behaviors by firms. We find empirical evidence of increasing risk appetite when CDS is written against firms debts, studying data from 2001-2014. The results remained consistent after testing for robustness using propensity score matching. This supports the view that CDS reduces monitoring efforts by creditors, once they were sufficiently insured against defaulting of the firm and enjoy the improved bargaining position in the renegotiation process, making management of the firms more prone to risk-shift. However, our results also suggest that CDS does not affect all firms equally. More specifically, we find evidence that
financially constrained firms might in fact be more cautious with regard to risk exposure when facing uncertain future prospects in the presence of CDS. This result is in
accordance with Rocha (2014), consolidating the idea that the discriminating effect of CDS partly alleviates the aforementioned empty creditor problem and could help stabilize the economy in downturns.
Overall, CDS still seems to create contracting inefficiencies and provide peculiar incentives, but one should not discard the potential benefits it introduces in the interim. Reducing transaction costs, increased debt capacity due to lower cost of debt and refraining constrained firms from risky investments in environments of uncertainty are just a few of them. Further research with the intention of identifying the channels through which CDS exerts its negative influences and designing mechanisms in the regulation process to prevent these consequences from occurring is key in the near future. Only then, CDS could be proofed to be truly beneficial for the economy.
２０
At last, I want to address some shortcomings of the empirical setting in this paper. Most prominently, the causal direction of the variables in the applied models could not be established with certainty. Regressions can only imply correlation but not causality. Unfortunately, due to my inexperience in data handling and the limited time span this part was done rather poorly. Further improvement could contain a reverse causality test to alleviate this concern. Moreover, the dataset on financial measures was missing quite some observations of certain variables. Due to limited resources, I did not know where to retrieve them, and as a consequence, lots of observations were omitted, resulting in smaller sample sizes.
２１
References
Altman, E., (1968), Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy, Journal of Finance 23(4), 589-609.
Angelini, E., (2012), Credit Default Swaps (CDS) and their Role in the Credit Risk Market, International Journal of Academic Research in Business and Social Sciences 2(1), 584-594.
Arora, N., Gandhi, P., Longstaff, F. A., (2012), Counterparty Credit Risk and the Credit Default Swap Market, Journal of Financial Economics 103(2), 280-293.
Arping, S., (2004), Credit protection and lending relationships, In EFA 2004 Maastricht
Meetings Paper (No. 4551).
Ashcraft, A. B., Santos, J. A., (2009), Has the CDS market lowered the cost of corporate debt?, Journal of Monetary Economics 56(4), 514-523.
Bolton, P., Oehmke, M., (2011), Credit default swaps and the empty creditor problem, Review of Financial Studies 24(8), 2617-2655.
Bulan, L. T., (2003), Real Options, Irreversible Investment and Firm Uncertainty: New Evidence from US Firms, working paper, Colombia University.
Campello, M., Matta, R., (2012), Credit default swaps and risk-shifting, Economics
Letters 117(3), 639-641.
Eisdorfer, A., (2008), Empirical Evidence of Risk Shifting in Financially Distressed Firms, Journal of Finance 63, 609-637.
Episcopos, A., (1995), Evidence on the Relationship Between Uncertainty and Irreversible Investment, Quarterly Review of Economics and Finance 35, 41-52.
Hu, H. T., Black, B. (2008), Debt, equity and hybrid decoupling: Governance and systemic risk implications, European Financial Management 14(4), 663-709.
２２
Kim, G. H., (2011), CDS, Strategic Default and the Cost of Corporate Debt, working
paper, University of Warwick.
Leahy, J. V., Whited, T. M., (1996), The Effect of Uncertainty on Investment: Some Stylized Facts, Journal of Money, Credit and Banking 28, 64-83.
Pindyck, R. S., Solimano, A., (1993), Economic Instability and Aggregate Investment,
NBER Macroeconomics Annual 8, 259-303.
Rocha, M. A. (2014), Essays on Risk Assumption and Liquidity Management (Unpublished Doctoral Dissertation), University of Illinois, Urbana-Champaign.
Saretto, A., Tookes, H., (2011), Corporate Leverage, Debt Maturity and Credit Default Swaps: The Role of Credit Supply, working paper, University of Texas.
Stulz, R. M., (2010), Credit Default Swaps and the Credit Crisis, Cambridge, MA : National Bureau of Economic Research Sep. 2009
Subrahmanyam, M. G., Tang, D. Y., Wang, S. Q., (2012), Does the Tail Wag the Dog? The Effect of Credit Default Swaps on Credit Risk, working paper, Hong Kong Institute for Monetary Research.
２３
Tables
TABLE 1. Descriptive Statistics Main Sample
Panel A. All firms Mean SD Q1 Q2 Q3
Investment 0.13 0.12 0.05 0.1 0.18 Size 9.01 1.87 7.6 9.24 10.41 leverage 0.57 0.23 0.44 0.58 0.71 Market to Book 2.71 47.9 1.26 1.98 3.21 Tangibility 0.31 0.23 0.12 0.25 0.47 Z_score 3.02 6.15 1.1 1.96 3.27
Panel B. CDS firms Mean SD Q1 Q2 Q3
Investment 0.11 0.09 0.05 0.09 0.16 Size 10.28 1.16 9.52 10.27 11.07 leverage 0.66 0.17 0.54 0.64 0.75 Market to Book 2.51 34.24 1.12 1.75 2.79 Tangibility 0.34 0.23 0.14 0.3 0.53 Z_score 1.71 1.21 0.89 1.57 2.31
Panel C. Non CDS firms Mean SD Q1 Q2 Q3
Investment 0.15 0.15 0.06 0.11 0.2 Size 7.94 1.68 6.74 7.81 9.06 leverage 0.5 0.25 0.33 0.5 0.65 Market to Book 2.88 56.99 1.4 2.2 3.56 Tangibility 0.28 0.23 0.1 0.21 0.42 Z_score 4.14 8.13 1.43 2.54 4.6
２４
TABLE 2. Descriptive Statistics Diff-in-Diff Sample Panel A. CDS firms (restricted around event)
' Mean SD Q1 Q2 Q3 Investment 0.2 0.1 0.1 0.13 0.18 Size 10.41 1.18 9.63 10.41 11.31 leverage 0.67 0.17 0.55 0.64 0.76 Market to Book 1.74 8.12 1.18 1.84 2.89 Tangibility 0.29 0.21 0.12 0.24 0.43 Z_score 1.63 1.13 0.99 1.53 2.29 LTDdue 0.11 0.11 0.01 0.1 0.18
Panel B. Non CDS firms (restricted around event)
' Mean SD Q1 Q2 Q3 Investment 0.24 0.15 0.14 0.2 0.29 Size 8.22 1.47 7.12 8.16 9.18 leverage 0.54 0.25 0.39 0.5 0.64 Market to Book 2.18 29.45 1.23 2.02 3.25 Tangibility 0.28 0.24 0.09 0.2 0.42 Z_score 2.9 3.47 1.44 2.39 3.79 LTDdue 0.11 0.2 0 0.03 0.11
２５
TABLE 3. Effect of CDS on reationship expected volatility and investment standard errors in parenthesis, significance level: *** 1%, **5%, *10%. Dependent Variable: Investment [1] [2]
Expected Volatility -0.0482 -0.0543 (0.0219)** (0.0231)** CDS -0.0588 (0.0049)*** CDS*Expected Volatility 0.0089 (0.0011)*** tangibility -0.128 -0.128 (0.002)*** (0.002)*** leverage -0.065 -0.064 (0.002)*** (0.002)*** log(size) -0.005 -0.005 (0.000)*** (0.000)*** Market-to-Book Ratio 0 0 (0.000)** (0.000)** constant 0.258 0.259 (0.002)*** (0.002)*** R-squared 0.0966 0.0985 #observations 93307 93307
２６
TABLE 4. Financial Status, CDS and interaction
standard errors in parenthesis, significance level: *** 1%, **5%, *10%. Dependent Variable: Investment [1]
Expected Volatility -0.108 (0.04249)** CDS -0.0131 (0.00083)*** z_score -0.01255 (0.00114)*** CDS*Expected Volatility -0.00265 (0.00079)*** Z_score*Expected Volatility 0.22316 (0.02078)*** CDS*z_score*Expected Volatility 0.00145 (0.00018)*** tangibility -0.09894 (0.00184)*** leverage -0.02679 (0.00285)*** Market-to-Book Ratio 0.00005 (0.00001)*** Constant 0.14297 (0.00580)*** R-squared 0.1066 #observations 93307
２７
TABLE 5. Diff-inDiff Regressions
Robust standard errors in parenthesis, significance level: *** 1%, **5%, *10%. Dependent Variable: ΔInvestment
Low CDS Notional
Panel A. Low Expected Volatility Panel B. High Expected Volatility
LTDdue -0.0289 -0.0918 (0.0391) (0.0338)*** log(size) 0.0004 0.003 (0.0052) (0.0066) tangibility -0.047 -0.0525 (0.0265)* (0.0203)** leverage -0.0383 -0.0669 (0.0399) (0.0363)* z-score -0.0047 -0.0052 (0.0074) (0.0065) Constant 0.0988 0.0858 (0.0786) (0.0842) R-squared 0.1578 0.2579 #observation s 52 49 High CDS Notional
Panel C. Low Expected Volatility Panel D. High Expected Volatility
LTDdue -0.0579 -0.0773 (0.0618) (0.0520)* log(size) -0.0027 0.0018 (0.006) (0.0046) tangibility -0.0418 -0.0732 (0.0174)** (0.0179)*** leverage 0.0207 0.0119 (0.0330) (0.0290) z-score 0.0111 0.0049 (0.0067) (0.0044) Constant 0.0619 0.0432 (0.0514) (0.0508) R-squared 0.1613 0.3005 #observations 52 49
２８
TABLE 6. Probit Model Propensity Score Matching
standard errors in parenthesis, significance level: *** 1%, **5%, *10%.
Dependent Variable: CDSdummy
Explanatory Variables [1] ln(asset) 0.1303 (0.0009)*** leverage 0.2828 (0.0066)*** ROA 0.1126 (0.0165)*** PPENT/assets 0.0932 (0.0092)*** Sales/assets 0.0584 (0.0083)*** EBIT/assets -0.1677 (0.0145)*** WCAP/assets 0.01 (0.012) Re/assets -0.0044 (0.0013)*** CAPX/assets -0.1489 (0.0419)*** Constant -0.0703 (0.0130)***
Time fixed effect YES
Industry fixed effect YES
R-squred 0.4813
２９
TABLE 7. Regressions Using Propensity Score Matched Sample standard errors in parenthesis, significance level: *** 1%, **5%, *10%.
Dependent Variable: Investment [1] [2] [3]
Expected Volatility -0.0488 -0.0319 -0.2459 (0.0227)** (0.0232)** (0.0268)*** CDS -0.0636 -0.0182 (0.0053)*** (0.0125)** z_score -0.0096 (0.0037)** CDS*Expected Volatility 0.0073 -0.0082 (0.0007)*** (0.0018)*** Z_score*Expected Volatility 0.2237 (0.0209)*** CDS*z_score*Expected Volatility 0.0014 (0.0002)*** tangibility -0.1191 -0.1188 -0.0986 (0.0017)*** (0.0017)*** (0.0018)*** leverage -0.0528 -0.051 -0.2619 (0.0019)*** (0.0019)*** (0.0028)*** log(size) -0.0036 -0.0038 0.003 (0.0002)*** (0.0002)*** (0.0004)*** Market-to-Book Ratio 0 0 0 (0.0000) (0.0000) (0.000) Constant 0.2269 0.2309 0.1436 (0.0025)*** (0.0025)*** (0.0058)*** R-squared 0.0915 0.0939 0.1022 #observations 68020 68020 68020
３０
Figures
FIGURE 1. S&P500 Expected Volatilities
.0 2 .0 4 .0 6 .0 8 .1 Ex pV o l
01jan2000 01jan2005 01jan2010 01jan2015
Calendar Date
