16
3.4. Descriptive statistics
Table 3 reports descriptive statistics of the variables used in our analyses separately for CDS and non-CDS firms. Characteristics are presented separately for periods prior to and after the
initiation of CDS trading for the CDS firms. Corporate investments in both intangible
RDEXP
and tangible
CAPEX
assets decrease following CDS inception. Hence, we do not find preliminary support for H1 on a univariate basis. CEO total and excess compensation as well as vega
significantly increase after CDS inception, which is consistent with H2A. While sales revenue increases, growth opportunities, stock return, and profitability decrease subsequent to CDS trading
initiation. These results are consistent with prior studies finding that CDS firms can experience negative shock to their operational performance Subrahmanyam, Tang, and Wang, 2014.
[Insert Table 3 near here]
4. Tests of hypotheses
This section presents the tests of hypotheses.
4.1. Tests of H1: substitution of safe with risky assets upon CDS inception
H1 states that firms shift a part of their safe assets to risky ones after CDS inception. We estimate the following regression to test this hypothesis:
RDEXP
i,t
or CAPEX
i,t
=β + β
1
CDS_Trade
i,t
+ β
2
CDS_Firm
i
+ β
3
Tenure
i,t
+ β
4
Cash_Comp
i,t
+ β
5
MTB
i,t
+ β
6
SALES
i,t
+ β
7
SurplusCash
i,t
+ β
8
SalesGrowth
i,t
+ β
9
STRET
i,t
+ β
10
LEV
i,t
+ε
i,t
, 2 where the dependent variable is RD expenditures
RDEXP
or capital expenditures
CAPEX
.
RDEXP
is set to zero when the value is missing from Compustat and both
RDEXP
and
CAPEX
are divided by assets at the end of the fiscal year
.
Dummy variable
CDS_Trade
takes a value of one after CDS inception for CDS firms and zero otherwise. Dummy variable
CDS_Firm
takes a value
17
of one for firms that have their CDSs traded during our study period. CDS-traded firms are considered the treatment group after CDS inception.
Including both
CDS_Trade
and
CDS_Firm
provides a difference-in-differences research design to isolate the impact of CDS inception relative to contemporaneous changes for non-CDS
firms. Hence, the variable
CDS_Trade
captures the marginal impact of CDS introduction on corporate investments relative to the impact on non-CDS firms at the same time. If firms substitute
a part of their safe assets with risky ones following the onset of CDS trading H1, then β
1
would be significantly positive when
RDEXP
is the dependent variable and significantly negative when
CAPEX
is the dependent variable. To control for the determinants and managerial incentive of investments, we include firm size
SALES
and financial leverage
LEV
, as well as growth opportunity
MTB
, cash availability
SurplusCash
, sales growth
SalesGrowth
, and stock return
STRET
, following Coles, Daniel, and Naveen 2006. Also following Coles, Daniel, and Naveen 2006, we use two proxies for CEO level of risk aversion: CEO tenure
Tenure
and salary plus bonus
Cash_Comp
. We control for year and industry idiosyncratic characteristics by including their fixed effects in all regressions. Detailed variable definitions are provided in Appendix A.
In Table 4, the first two columns report results of Eq. 2 with
RDEXP
as the dependent variable; the last two columns,
CAPEX
as the dependent variable. The first and third columns use only CDS firms, and the second and fourth columns include both CDS firms and non-CDS firms.
The coefficient on
CDS_Trade
is not significant in any test at a conventional level. Thus, we do not find any significant change, on average, in either risky or safe assets upon CDS inception. H1,
which posits asset substitution upon CDS inception, is rejected. This null result can be consistent with the empty creditor theory; that is, after diversifying its credit risk against the borrower, an
empty lender could let renegotiations with the distressed borrower fail and force it into bankruptcy.
18
Faced with a more credible threat of foreclosure, the borrower might avoid actions detrimental to the
lender’s interests Arping 2014. Another possible explanation is that the interests of risk- averse managers are largely aligned with those of lenders as far as the effects of asset volatility are
concerned. Managers thus are unlikely to support the opportunistic behavior of shareholders by increasing asset volatility or dividend payouts, even when the lender reduces monitoring, because
those actions hurt their interests. [Insert Table 4 near here]
4.2. Tests of H2A: changes in managerial compensation after CDS inception
We hypothesize that the onset of CDS trading reduces lender banks’ monitoring efforts, thus enabling borrower firms to substitute safe assets with risky ones, thereby benefiting
shareholders. We also hypothesize that managers would facilitate such actions when they are consistent with their personal interests. We examine whether shareholders offer or managers
demand higher vega and higher compensation to increase asset volatility post –CDS inception. We
estimate the following regression to test H2A: Total_Comp
i,t
or Vega
i,t
=β + β
1
CDS_Trade
i,t
+ β
2
CDS_Firm
i
+ ∑β
n
Controls
i,t
+ε
i,t
, 3
where the dependent variable is either total compensation
Total_Comp
or the sensitivity of
CEO ’s compensation to stock price volatility
Vega
. That is,
Vega
is the change in the dollar value of the
CEO’s wealth for a 0.01 change in the standard deviation of stock returns per year. The vega measure is derived from the Black-Scholes option valuation model e.g., Yermack, 1995; Hall and
Leibman, 1998; Aggarwal and Samwick, 1999; Guay, 1999; Cohen, Hall, and Viceira, 2000; Datta, Iskandar-Datta, and Raman, 2001; Rajgopal and Shevlin, 2002; Core and Guay, 2002.
Total_Comp
is a natural logarithm of the sum of salary, bonus, long-term incentive plan payouts,
19
the value of restricted stock grants, the value of options granted during the year, and any other annual pay for the CEO in the fiscal year. Of interest is the coefficient on
CDS_Trade
β
1
. If shareholders increase CEO total value or convexity of managerial compensation following the
onset of CDS trading, then β
1
is predicted to be positive and significant. We also follow prior literature to employ a vector of control variables in Eq. 3 Core,
Holthausen, and Lacker, 1999. To control a determinant and an incentive of CEO compensation, we include firm size
SALES
, firm reputation
SP500
, growth opportunity
MTB
, stock return
STRET
, and profit ratio
ROA
. We also use salary plus bonus
Cash_Comp
and CEO tenure
Tenure
to control for CEO risk aversion. With
Vega
as the dependent variable, we follow Richardson 2002 and Coles, Daniel, and Naveen 2006 to select control variables. We include
firm size
SALES
, financial leverage
LEV
, growth opportunity
MTB
, and cash
CashSize
, as well as sales growth
SalesGrowth
, CEO tenure, stock return
STRET
, and stock return volatility
STRETVOL
, following Coles, Daniel, and Naveen 2006. Panel A and Panel B of Table 5 present results with
Total_Comp
and
Vega
as the dependent variable, respectively. The coefficient on
CDS_Trade
is consistently positive and significant in all models
p
-value 0.01, providing strong support for H2A that CDS inception is positively associated with CEO vega and compensation. The economic significance of this increase is
estimated by dividing the regression coefficient by the mean value pre –CDS inception, which
amounts to 39 for
Vega
but only 1.3 for
Total_Comp
. So, the principal change in managers’
compensation structure appears to be an increase in convexity, not the total value. [Insert Table 5 near here]
20
4.3. Test of H2B: substitution of assets and increases in excessive dividend payouts, conditional on managerial compensation
In the following regression, we examine whether managers holding incentives consistent with shareholder interests increase the asset volatility post
–CDS inception H2B: RDEXP
i,t
or CAPEX
i,t
=β + β
1
CDS_Trade
i,t
+ β
2
Vega
i,t
or Total_Comp
i,t
+ β
3
CDS_Trade
i,t
× Vega
i,t
or Total_Comp
i,t
+ β
4
CDS_Firm
i
+∑β
n
Controls
i,t
+ε
i,t
. 4
In Eq. 4, the dependent variable is
RDEXP
or
CAPEX
. We use one of the two proxies of managers’ incentives in the model at a time
Total_Comp
or
Vega
. The coefficient of interest is on the interaction term
CDS_Trade
×
Vega
or
Total_Comp
. If managers have the right incentives to increase operating risk-taking following the onset of CDS trading, then
β
3
should be significantly positive for
RDEXP
and significantly negative for
CAPEX
. We follow Coles, Daniel, and Naveen 2006 in the selection of control variables.
Results of Eq. 4 are reported in Panel A and Panel B of Table 6 for
Vega
and
Total_Comp
, respectively. The first two columns in each panel report results with
RDEXP
as the dependent variable; the last two columns, with
CAPEX
as the dependent variable. The coefficients on the interaction term
CDS_Trade
×
Vega
or
Total_Comp
are consistently positive and significant with
RDEXP
as the dependent variable and negative with
CAPEX
as the dependent variable in one model. Results of the two panels support the hypothesis that when managers have high vega
incentives, they engage in asset substitution, reallocating firm resources from less risky capital expenditures to more risky RD. This resource reallocation is consistent with shareholder interests
and contrary to lender interests. It is noteworthy that the coefficient on
CDS_Trade
is negative and
21
significant with
RDEXP
as the dependent variable. It shows that, without taking the convexity into account, borrowers follow more conservative investment policies anticipating more credible threat
of foreclosure Arping 2014; Subrahmanyam, Tang, and Wang, 2017. [Insert Table 6 near here]
Based on the coefficient estimates from the models in Table 6, Panel A, the effect of a one standard deviation increase in vega is associated with a 3.48 increase in the mean value of RD
expenditure. This effect of
Vega
on corporate investment policy subsequent to CDS trading appears to be large and economically significant relative to that of total compensation.
4.4. Frequency of mergers and acquisitions after CDS inception, conditional on CEO incentives
We examine mergers and acquisitions as another proxy for firms’ risky investment policy, which can shift corporate wealth from lenders to shareholders. As MA activities can increase
asset volatility to the detriment of lender interests, debt covenants typically include constraints on MA activities Nash, Netter, and Poulsen, 2003. We estimate the regression
MA_Freq
i,t
=β +β
1
CDS_Trade
i,t
+ β
2
Vega
i,t
or Total_Comp
i,t
+ β
3
CDS_Trade
i,t
× Vega
i,t
or Total_Comp
i,t
+ β
4
CDS_Firm
i
+∑β
n
Controls
i,t
+ε
i,t
, 5
where the dependent variable,
MA_Freq
, is the number of MA transactions conducted in a given firm-year. We employ a negative binomial regression model because our dependent variable is
count data Long 1997.
7
We use the same set of control variables as in Eq. 4 and use one of the two proxies for
managers’ incentives
Vega
or
Total_Comp
in the model at a time. The coefficient
7
We find similar results using a Poisson regression model, another widely accepted method for count data.
22
of interest is on the interaction term
CDS_Trade
×
Vega
or
Total_Comp
. If managers have the right incentives to undertake riskier investment activities following the onset of CDS trading, then
β
3
should be significantly positive. Table 7 shows that the coefficients on the interaction term
CDS_Trade
×
Vega
are positive and significant
p
-value 0.05. Thus, borrowers engage in more frequent MA transactions post –
CDS inception when their managers have more convex compensation arrangements. As with Equation 4 results, the coefficient on
CDS_Trade
is negative and significant. It is consistent with the idea that borrowers pursue more conservative investment policies post CDS inception, if managers’
vega incentives are not taken into account. [Insert Table 7 near here]
Changes in borrowers’ investment policies, described in Tables 6 and 7, should also cause an increase in earnings volatility. We also examine the effect on earnings volatility by estimating the
regression CEVOL
i,t
=β +β
1
CDS_Trade
i,t
+ β
2
Vega
i,t
or Total_Comp
i,t
+ β
3
CDS_Trade
i,t
× Vega
i,t
or Total_Comp
i,t
+ β
4
CDS_Firm
i
+∑β
n
Controls
i,t
+ε
i,t
, 6
where the dependent variable,
CVOL
, is the standard deviation of a firm’s quarterly return on assets
in a given year Brown, Christensen, Elliott, and Mergenthaler, 2012; Choy, Lin, and Officer, 2014. Following Choi, Mao, and Upadhyay 2014, we use similar control variables as in Eqs. 4
and 5. The coefficient of interest is on the interaction term
CDS_Trade
×
Vega
or
Total_Comp
. If managers have the right incentives to undertake riskier activities following the onset of CDS
23
trading, then β
3
should be significantly positive. Table 8 presents results consistent with this idea.
CEVOL
increases post –CDS inception for managers with highly convex payoffs.
[Insert Table 8 near here]
4.5. Excess dividends after CDS inception, conditional on CEO incentives
We examine corporate dividend payouts because shareholders can most directly transfer wealth from creditors by paying themselves cash dividends. We calculate dividend payout ratio by
dividing common dividends by market value of equity Grullon and Michaely, 2002. We then create an indicator for firm-year observations in the highest quartile of dividend payouts for our sample
EXCESS_DIV
. This variable is effectively year- and industry-adjusted because we control for their fixed effects. We then estimate the regression
EXCESS_DIV
i,t
=β + β
1
CDS_Trade
i,t
+ β
2
Vega
i,t
or Total_Comp
i,t
+ β
3
CDS_Trade
i,t
× Vega
i,t
or Total_Comp
i,t
+ β
4
CDS_Firm
i
+ ∑β
n
Controls
i,t
+ε
i,t
. 7
As in earlier tests, we use one of the two proxies of managers’ incentives in the model at a
time:
Total_Comp
or
Vega
. The coefficient of interest is the interaction term
CDS_Trade
×
Vega
or
Total_Comp
. If managers have incentives consistent with those of shareholders, they can pay shareholders larger dividends following the onset of CDS trading. Table 9 reports results of this
analysis and shows that the coefficient on the interaction term is positive and significant
p
-value 0.10. Thus, borrowers appear to change their financing policy to benefit shareholders subsequent to
CDS inception, provided managers’ and shareholder interests are aligned through managerial
compensation contracts. The coefficient on
CDS_Trade
is negative, indicating that borrowers reduce
24
dividend payouts, if managers’ interests are not taken into account Subrahmanyam, Tang, and Wang,
2017.
[Insert Table 9 near here]
4.6. Tests of H3: The joint effect of CDS inception and managerial interests on default risk
H2A and H2B tests indicate that when managers have convex payoffs, they substitute safe assets with risky ones post
–CDS inception. The resulting increase in operating risk should enhance the bankruptcy risk. We test in H3 whether CDS inception increases bankruptcy risk for firms with
managers holding high vega incentives. We estimate Bankruptcy
i,t
= β
+ β
1
CDS_Trade
i,t
+ β
2
Vega
i,t
or Total_Comp
i,t
+ β
3
CDS_Trade
i,t
× Vega
i,t
or Total_Comp
i,t
+ β
4
CDS_Firm
i
+ ∑β
n
Controls
i,t
+ε
i,t
, 8
where
Bankruptcy
is an indicator variable that takes a value of one if a firm files for bankruptcy in any of the five years after a give year
t
. As in earlier tests, our main interest is the coefficient on the interaction term
CDS_Trade
×
Vega
or
Total_Comp
, which we expect to be positive based on H3. We follow Subrahmanyam, Tang, and Wang 2014 and include a vector of control
variables, which are known to affect corporate bankruptcy risk: firm size
MKV
, debt size
LNDEBT
, stock return
STRET
, stock return volatility
STRETVOL
, and profitability
ROA
. The results are presented in Table 10. The coefficient on
CDS_Trade
is significant and positive, showing
that firms’ default risk increases subsequent to CDS trading. This finding is consistent with Subrahmanyam, Tang, and Wang 2014. More importantly, the coefficients on
the interaction term
CDS_Trade
×
Vega
or
Total_Comp
are positive and significant
p
-value
25
0.05, providing strong support for the view that when managers have high vega incentives, the onset of CDS trading is followed by managerial actions that increase the likelihood of corporate
default risk. [Insert Table 10 near here]
Tables 6−10 show that borrowing firms shift their investment policies post–CDS inception from safe to risky one when managers have high vega incentives. Furthermore, the onset of CDS
trading is followed by changes in managerial compensation structure toward higher vega incentives. Results are consistent with the idea that shareholders capitalize on the opportunity
offered by reduced lender monitoring by changing the structure of managerial compensation toward higher convex payoffs. Nevertheless, our findings also demonstrate that high vega
managers increase the bankruptcy risk following CDS inception Subrahmanyam, Tang, and Wang, 2014.
In sum, our study indicates that CDS inception is not always a value-decreasing proposition for shareholders, as indicated in prior studies. For example, an increased threat of foreclosure on
account of lost lender interest Subrahmanyam, Tang, and Wang, 2014. For firms with convex managerial payoffs, shareholders could benefit because the firm enhances activities that were
previously constrained by lender monitoring. Furthermore, while managers could lose the value of their firm-specific investments from the increased risk of bankruptcy, they could also benefit from
the increased value of their vega incentives. We thus present a fuller picture of shifts in the rival lender, shareholder, and managerial forces that determine firms’ investment and financing policies
post –CDS inception. We respond to the Augustin et al.’s 2014 call for a thorough examination
of changes in corporate policy and stakeholder interests upon CDS inception.
26
5. Robustness checks