This article was downloaded by: [147.188.128.

74] On: 22 July 2020, At: 03:40

Publisher: Institute for Operations Research and the Management Sciences (INFORMS)
INFORMS is located in Maryland, USA

Management Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org

Financial Innovation and Financial Intermediation:

Evidence from Credit Default Swaps
Alexander W. Butler, Xiang Gao, Cihan Uzmanoglu

To cite this article:

Alexander W. Butler, Xiang Gao, Cihan Uzmanoglu (2020) Financial Innovation and Financial Intermediation: Evidence from
Credit Default Swaps. Management Science

Published online in Articles in Advance 21 Jul 2020

. https://doi.org/10.1287/mnsc.2019.3560

Full terms and conditions of use: https://pubsonline.informs.org/Publications/Librarians-Portal/PubsOnLine-Terms-and-

Conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.

Copyright © 2020, INFORMS

Please scroll down for article—it is on subsequent pages

With 12,500 members from nearly 90 countries, INFORMS is the largest international association of operations research (O.R.)
and analytics professionals and students. INFORMS provides unique networking and learning opportunities for individual
professionals, and organizations of all types and sizes, to better understand and use O.R. and analytics tools and methods to
transform strategic visions and achieve better outcomes.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
 MANAGEMENT SCIENCE
Articles in Advance, pp. 1–24
http://pubsonline.informs.org/journal/mnsc ISSN 0025-1909 (print), ISSN 1526-5501 (online)

Financial Innovation and Financial Intermediation: Evidence from

Credit Default Swaps
Alexander W. Butler,a Xiang Gao,b Cihan Uzmanogluc
a
Jones Graduate School of Business, Rice University, Houston, Texas 77005; b University of North Dakota, Grand Forks, North Dakota 58203;
c
Binghamton University, Binghamton, New York 13902
Contact: alex.butler@rice.edu, https://orcid.org/0000-0002-9278-9947 (AWB); xiang.gao.1@und.edu (XG); cuzmanog@binghamton.edu (CU)

Received: January 14, 2019
Revised: August 6, 2019; November 7, 2019
Accepted: November 29, 2019
Published Online in Articles in Advance:
July 21, 2020
https://doi.org/10.1287/mnsc.2019.3560
Copyright: © 2020 INFORMS
Abstract. We study the influence of credit default swaps (CDS) trading on the costs of
bond intermediation. After CDS initiation, CDS firms pay 12% to 28% (8 to 20 basis points)
lower underwriting fees than similar non-CDS firms do. Underwriting fees decline more
for riskier issuers and illiquid bonds for which the ability to hedge with CDS is more
valuable. In bond offerings, participation by investors facing risk-based regulatory re-
quirements increases after CDS initiation. Our evidence suggests that CDS-driven inno-
vations in risk sharing contribute to the transactional efficiency of the market by reducing
the financial intermediation costs of placing bonds.

History: Accepted by Karl Diether, finance.

Supplemental Material: The e-companion is available at https://doi.org/10.1287/mnsc.2019.3560.

Keywords: credit default swaps • underwriting fees • bond ownership • hedging credit risk

1. Introduction credit quality (e.g., Acharya and Johnson 2007). More-

During the past two decades, the bond market has
undergone a major transformation with the intro-
duction of credit default swaps (CDS)—derivative
contracts that protect the buyer against the credit
risk of the underlying debt—as new instruments for
trading credit risk. The CDS market grew signifi-
cantly after its inception in the early 1990s, reaching
its peak in 2007 with $61.2 trillion in outstanding
notional amounts.1 In this paper, we investigate whether
and how the initiation of CDS trading on a firm influ-
ences its bond underwriting fees to learn about both the
intermediation of the bond issuance process and the real
effects of CDS.
One of the major costs of a bond offering is the fee
paid to intermediaries to place the bonds. That un-
derwriting fee, often called the gross spread, is de-
termined in part by the difficulty and risk of placing
the bonds (e.g., Lee et al. 1996, Altinkilic and Hansen
2000, Chen and Ritter 2000, Butler et al. 2005, Yasuda
2005). Underwriting fees are transparent, are easily
measured, and offer a direct picture of the interaction
between CDS innovation and financial intermedia-
tion of bond offerings.
Theoretically, CDS can either decrease or increase
underwriting fees. Because trading corporate bonds
is costly, the availability of CDS enhances investors’
ability to manage credit risk (e.g., Blanco et al. 2005,
Bolton and Oehmke 2011). CDS trading can also pro-
vide useful market-based information about a firm’s
credit risk, reducing uncertainty about an issuer’s
over, the availability of CDS on issuers may expand
underwriters’ profit opportunities by allowing them
to cross-sell bonds together with the related CDS
contracts. These hedging, information, and cross-
selling benefits of CDS can make the bonds of un-
derlying firms more attractive to investors, making
them easier for underwriters to place, and thereby
lower underwriting costs. On the other hand, by
allowing investors to construct bond payoffs syn-
thetically, CDS can be a substitute for the underlying
bonds (e.g., Oehmke and Zawadowski 2015, 2017),
hence decreasing the demand for bond offerings and
increasing their underwriting costs. The availability
of CDS on issuers may also increase bond under-
writing fees by reducing the cross-monitoring bene-
fits of bank debt, as banks hedged with CDS would
have reduced incentives to monitor the underlying
issuers (e.g., Duffee and Zhou 2001, Morrison 2005,
Parlour and Winton 2013, Shan et al. 2019). Alter-
natively, a relationship between CDS trading and
underwriting fees could arise spuriously if firms with
CDS make use of underwriters that happen to charge
different fees. Therefore, whether and how the initi-
ation of CDS trading influences bond underwriting
fees is an empirical question.
In our first test, we examine the within-issuer changes
in underwriting fees for new bond issues from the
pre- to post-CDS initiation periods among firms
with CDS trading, that is, CDS firms. We find
that, controlling for numerous bond issue, issuer,

1
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
2 Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
and underwriter characteristics (e.g., bond maturity,
liquidity, credit rating, issuer size, leverage, underwriter-
issuer relationship, syndicate structure, and many
others), the initiation of CDS trading on an issuer is
associated with a reduction in its bond underwriting
fee of 12%, or eight basis points (bps) on average. Thus,
the availability of CDS contracts appears to reduce the
financial intermediation costs of placing bonds for the
underlying firms.
Because the initiation of CDS contracts is not ran-
domly assigned, we also employ other identification
approaches. We use three independent identification
approaches that have been used previously in the
CDS literature in various ways: (1) A matching pro-
cedure, (2) a natural experiment, and (3) an instru-
mental variable approach. The three approaches have
different strengths and weaknesses, but each leads us
to similar conclusions. We discuss these approaches
briefly here and in more detail below.
Our matching approach balances the ex ante ob-
servable characteristics that are related to under-
writing fees between firms with and without CDS.
Our matching procedure offers good external validity
but rests on a strong identifying assumption of
unconfoundedness (Rubin 1990). The results show that
relative to similar non-CDS firms, CDS firms experi-
ence a 17% (or 12 bps) reduction in their underwriting
fees following the inception of CDS trading.
Our second identification approach is a natural
experiment. Natural disasters can have major impacts
on the cash flows of insurance companies, forcing
them to liquidate bond holdings and reducing their
demand for subsequent bond offerings (Massa and
Zhang 2011). If the availability of CDS makes the
underlying bonds more attractive to investors, the
effect of CDS on underwriting fees should be stronger
during these periods of low investor demand for bond
offerings. We find that it is.
An instrumental variables approach (using as instru-
ments lenders’ foreign exchange hedging positions
and lenders’ Tier 1 capital ratios, as in Subrahmanyam
et al. 2014) puts the identification burden on the
portion of the variation in the endogenous variable
(CDS initiation) that arises through the influence of the
instruments, both of which reflect measures associ-
ated with hedging demand from lenders. We discuss
in detail below the instruments’ validity in our setting,
and we find that underwriting fees decline by about
20% to 28% (or 14 to 20 bps) with CDS initiation. The
results from these various approaches collectively
suggest that the decline in underwriting fees we
observe with CDS initiation is causal.
Next, we investigate whether this CDS-related de-
cline in underwriting fees is driven by hedging, infor-
mation, cross-selling, underwriter quality, or underwriter
competition channels. The hedging, information, and
cross-selling channels predict a more pronounced CDS
effect for, respectively, risky issuers, informationally
opaque issuers, and underwriters with greater CDS
activity. We find that CDS initiation reduces under-
writing fees more for riskier issuers but not for in-
formationally opaque issuers or for CDS-active un-
derwriters, consistent with a channel in which the
availability of CDS reduces bond underwriting costs
by enabling better risk sharing. The decline in under-
writing fees could also be due to a decrease in under-
writer quality or an increase in competition among
underwriters during our analysis period. If firms use
lower-quality underwriters following CDS initiations,
we would expect to see greater bond underpricing,
lower underwriter reputation, or higher offering yields.
We find no evidence of any of these outcomes nor evi-
dence that changes in the concentration of the under-
writing market affect the CDS effect we identify.
It is difficult to distinguish whether CDS are pre-
dominantly used for hedging or for speculation, as
detailed data on CDS holdings and trades are not pub-
licly available (e.g., Boyarchenko et al. 2018, Gunduz
2018). One approach to isolate evidence about the
hedging motive is to study who purchases newly
issued bonds referenced by CDS. We hypothesize that
insurance companies and banks, which are subject to
risk-based regulatory requirements, have strong in-
centives to hedge with CDS or use them for regulatory
arbitrage. Consistent with this notion, we find that the
participation of insurance companies and banks in
bond offerings increases with the initiation of CDS
trading relative to that of other investors.
Likewise, if hedging motives are important, then
bond liquidity should affect the relationship between
CDS initiation and underwriting fees because the
benefits of CDS in facilitating risk sharing are more
valuable for bonds that are costlier to trade in the
secondary market. Similarly, the CDS effect should be
stronger for bonds with larger negative CDS-bond
basis that are expected to have higher trading costs
and more disagreement about their default proba-
bilities (Oehmke and Zawadowski 2015). We indeed
find that underwriting fees decline with CDS initia-
tion more for bonds that are less liquid and have a
more negative CDS-bond basis.
Our findings provide evidence that CDS contracts
alter the financial intermediation costs of bond of-
ferings in a way consistent with the role of derivative
instruments in completing the market (e.g., Ross 1976,
Damodaran and Lim 1991, Figlewski and Webb 1993,
Kumar et al. 1998). In our setting, CDS improve out-
comes for issuers through enhancing investors’ hedg-
ing opportunities. Our finding that CDS introduction
benefits the underlying firms by reducing their bond
underwriting fees contributes to the empirical litera-
ture that examines the externalities associated with the
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
introduction of CDS contracts (e.g., Saretto and Tookes
2013, Das et al. 2014, Subrahmanyam et al. 2014,
Oehmke and Zawadowski 2015, Narayanan and
Uzmanoglu 2018a).2 Furthermore, our finding that
hedging is likely the channel driving this CDS-
related decline in underwriting fees supports the
hedging motives for trading CDS, thereby contrib-
uting to the emerging literature that tries to iden-
tify how investors use CDS contracts (e.g., Hasan
and Wu 2016, Gunduz et al. 2017, Oehmke and
Zawadowski 2017).
2. Data and Sample Characteristics
We identify all corporate bonds issued between 1996
and 2013 from the Mergent Fixed Income Securities
Database (FISD).3 Starting the sample with 1996 al-
lows us to analyze underwriting fees during the five-
year period before and after the earliest CDS initiation
date in our sample. Next, we exclude all but public
bonds (exclude Reg S, 144A, and other private of-
ferings) that are denominated in U.S. dollars and
issued by nonfinancial U.S. firms. To control for their
credit risk, we also require bonds to have at least one
credit rating on the offering date from S&P, Moody’s,
or Fitch. We then match the issuers to the CRSP and
Compustat databases to control for firm character-
istics. We match by the issuer CUSIP numbers from
the FISD database and then further investigate the
matching accuracy by comparing the names of the
matched firms across the databases. Merging the FISD
and CRSP/Compustat databases based on the issuer
CUSIP numbers and names results in an exact match
between the issuers in FISD and the firms in CRSP/
Compustat, thereby eliminating the bonds issued by
related entities from our final sample. This screening
results in an initial sample of 5,225 bonds offered
by 751 firms with available information in CRSP and
Compustat databases. From this sample, the infor-
mation on underwriting fees is available for 5,202
bonds issued by 739 firms.
CDS contracts are over-the-counter securities, and
there are no publicly available data sources for their
trading. Hence, identifying the precise date of CDS
initiation on a firm is challenging. To address this
issue, we construct a comprehensive data set of CDS
quotes and transactions from the Bloomberg, Credit
Market Analysis (CMA), and Depository Trust and
Clearing Corporation (DTCC) databases. The Bloom-
berg and CMA databases start covering the CDS
market in 2001 and 2004, respectively, and provide
daily consensus CDS quotes contributed by dealers.
DTCC provides aggregate positions data on the CDS
records registered in the Trade Information Warehouse.
The CDS positions data are available on a weekly basis
since 2008 for the 1,000 most actively traded CDS ref-
erence entities.
3
Most standard CDS contracts reference a firm’s
senior unsecured bonds. However, investors can use
these CDS to hedge against the credit risk on any of
the underlying firm’s bonds by adjusting their CDS
notional amounts for the differences in expected
default and recovery rates on the referenced and
hedged bonds. Therefore, we identify the CDS initi-
ation date for a firm as the earliest date it appears in
the Bloomberg and CMA combined database as a
reference entity to any single-name CDS contract (all
tenures and seniorities). To alleviate identification
concerns, we drop firms whose CDS initiation dates
fall in the first month that CDS coverage starts at
the Bloomberg and CMA databases.4 Because DTCC
reports CDS transactions for the most actively traded
contracts, the first time a firm appears in the DTCC’s
list may not be its CDS initiation date. Instead, we
utilize the DTCC’s data to improve the precision of
our identification. Accordingly, we exclude firms
from the sample if they have CDS data in DTCC
earlier than their CDS initiation dates. The CDS
initiation dates for these firms are inaccurate, as
they appear in the DTCC’s active CDS list in an
earlier period.5
After eliminating 109 firms with potentially inac-
curate CDS initiation dates from our sample, we have
a sample of 630 firms, of which 252 are CDS reference
entities.6 Within our CDS sample, 164 firms had their
CDS initiations between 2001 and 2003, 60 firms had
their CDS initiations between 2004 and 2006, and the
remaining 28 firms had their CDS initiations between
2006 and 2013.
To examine the evolution of underwriting fees
surrounding the CDS initiation date, we identify 211
CDS firms out of the 252 CDS firms in our initial
sample that issued at least one bond during the 10-year
period centered at the initiation of CDS trading.7
This 10-year event window in our baseline specifi-
cation matches the median maturity (rollover pe-
riod) of corporate bonds (e.g., Davydenko and
Strebulaev 2007). Using a longer event window
leads to a larger sample size and, hence, it is more
inclusive, but it may also lead to noisier estimates.
We show in Section 5.2 that our findings are similar
using shorter event windows.
Next, we select the benchmark firms without trad-
ing CDS for the 211 firms with traded CDS in our
sample. Within our initial sample of 630 bond issuers,
the average market value of equity is $23 billion for
the CDS firms, and it is $4.6 billion for the non-CDS
firms. Such a large difference in firm size can affect
the elasticity of underwriting fees with respect to firm
and bond characteristics—and hence bias estimations—
because firm size influences scale economies and
information environments. Furthermore, the litera-
ture (e.g., Narayanan and Uzmanoglu 2018a) shows
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
4 Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS

that riskier firms are more likely to be referenced by
CDS. This credit risk effect creates a bias for finding a
positive association between CDS initiation and un-
derwriting fees. To address these concerns, for each
CDS firm, we select a similar benchmark firm without
CDS using a propensity score matching method. We
match at the firm level, rather than the bond level,
because CDS contracts reference all of a firm’s out-
standing bonds in a given seniority.
We construct a panel data set of quarterly firm
characteristics between 2001 and 2013 for the issuers
in our bond sample and estimate a probit regression
predicting the propensity of CDS initiation for each
firm-quarter, controlling for firm characteristics lag-
ged by a quarter (see Appendix A for details). The
coefficient estimates presented in column (1) of the
table in Appendix A show that larger, less profitable,
and riskier firms are more likely to have CDS trading.
These probit regression results are broadly consistent
with other results in the literature. For instance,
Subrahmanyam et al. (2014) show that larger firms
with higher leverage ratios and greater stock vola-
tility are more likely to have CDS traded on their debt,
and Saretto and Tookes (2013) document that firm
profitability is negatively associated with CDS trading.
We select, with replacement, a benchmark non-CDS
firm for each CDS firm using the nearest propensity
score approach within a 10% difference, conditional on
the firm issuing at least one bond during the five-year
period either before or after the CDS initiation date. If a
benchmark firm is referenced by a CDS contract in the
subsequent periods, we exclude its bonds with trad-
ing CDS from the benchmark sample. Following this
matching approach, we find a good match for 204 of
211 firms with CDS in our sample. Roberts and Whited
(2012) explain that using the best-match method to
select benchmark firms is the least biased method, but
it also generates the least precise estimates. There-
fore, we implement alternative matching methods in
Section 5.2.
Our final analysis sample includes 1,186 bonds
issued by 204 CDS firms and 735 bonds issued by
204 non-CDS firms. Table 1 reports the summary
statistics of firm characteristics for this sample mea-
sured in the quarter immediately before CDS initia-
tion, the statistical tests for the differences in char-
acteristics of CDS and matched non-CDS firms, and
the normalized differences in the means of their
covariates. The average probability of CDS trading is
44% for both CDS and non-CDS firms. On average,
CDS and matched non-CDS firms also have similar
size (market value of equity), leverage (long-term
debt/assets), stock volatility (standard deviation of
daily stock returns in a quarter), profitability (net
income/sales), asset tangibility,8 and credit ratings
(investment grade dummy based on the S&P long-
term credit ratings). In addition, the maximum ab-
solute value of normalized differences in covariate
means is 0.17, which is less than 0.25, suggesting that
the covariate distributions of CDS and benchmark
non-CDS firms are well balanced (Imbens and Rubin
2015). We conclude that CDS and matched non-CDS

Table 1. Summary Statistics on Firm Characteristics and Underwriting Fees

Test of Covariate
Sample: CDS firms Benchmark Non-CDS firms differences balance

Variables Mean Median St. dev. Mean Median St. dev. t-statistic z-statistic Norm. diff.

Propensity Score 0.442 0.417 0.251 0.441 0.421 0.248 0.01 0.00 0.00
Log(MVE) 9.115 9.102 1.132 8.946 8.978 1.055 1.55 1.14 0.15
Long-Term Debt/Assets 0.273 0.252 0.149 0.282 0.261 0.152 −0.57 −0.66 −0.06
Stock Volatility 0.022 0.020 0.010 0.021 0.019 0.011 0.26 1.11 0.03
Net Income/Sales 0.043 0.050 0.147 0.050 0.066 0.130 −0.49 −1.74* −0.05
Tangibility 0.407 0.416 0.111 0.389 0.404 0.126 1.48 1.05 0.15
Investment Grade Dummy 0.858 1.000 0.350 0.819 1.000 0.386 1.07 1.07 0.11
Underwriting Fee (%) 0.730 0.650 0.385 0.804 0.638 0.473 −1.54 0.43 −0.17

Notes. This table reports the summary statistics for the characteristics of 204 CDS firms and their one-to-one matched benchmark non-CDS firms
as of the quarter preceding CDS initiation. Appendix B and Table IA-1 provide the details on the sample selection criteria. Propensity Score is the
probability of CDS trading estimated using the probit model presented in column (1) of the table in Appendix A. Log(MVE) is the natural
logarithm of the market value of equity in millions. Long-Term Debt/Assets is the ratio of long-term debt to book value of assets. Stock Volatility
equals the standard deviation of daily stock returns measured in the quarter preceding CDS initiation. Net Income/Sales is the ratio of net income
to sales. Tangibility is the ratio of (Cash and Equivalents + 0.715 × Receivables + 0.547 × Inventories + 0.535 × PP&E) to Assets. Investment Grade
Dummy equals one if the S&P long-term credit rating is BBB– or above and zero otherwise. Underwriting Fee is the bond underwriting fee stated as
a percentage of the offering amount. For each firm, Underwriting Fee reported in the table equals the mean of underwriting fees on the firm’s
bonds issued during the pre-CDS initiation period. Appendix B reports the summary statistics on the remaining bond characteristics. The Test of
Differences column reports the statistics from the tests of differences in mean (t-statistic from a two-tailed Student’s t-test) and median (z-value
from a two-tailed Wilcoxon rank-sum test) firm characteristics across the CDS and benchmark non-CDS firms. The Covariate Balance column
reports the normalized differences in covariate means between the CDS and benchmark non-CDS firms. St. dev., standard deviation; Norm. diff.,
normalized difference.
*p < 0.10; **p < 0.05; ***p < 0.01.
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
firms in our sample are comparable with respect to
their size, profitability, asset tangibility, and credit
risk prior to the listing of CDS.9
The primary dependent variable in our study is the
bond underwriting fee from FISD stated as a per-
centage of the issue amount. We check and con-
firm the accuracy of underwriting fees using bond
offering documents from Bloomberg and the SEC’s
EDGAR website. The average underwriting fee in our
sample is 71 bps ($3.2 million) with a standard de-
viation of 44 bps ($2 million). Panel A of Figure 1
shows that almost half of the bonds in our sample
have an underwriting fee between 50 and 70 bps,
and panel B of Figure 1 shows that underwriting fees
are generally on an upward trend during our analysis
period.10
Table 1 reports the summary statistics of under-
writing fees averaged at the firm level during the
pre-CDS initiation period, and Appendix B provides
detailed summary statistics on the remaining bond
Figure 1. Histogram and Time Series of Underwriting Fees
Notes. This figure presents the distribution of underwriting fees
(panel A) and their time series variation (panel B) for our sample of
1,921 bonds issued by 204 firms with CDS and their one-on-one
matched benchmark non-CDS firms. The mean (median) under-
writing fee is 71 (65) bps with a standard deviation of 44 bps. See
Appendix B, Table 1, and Table IA-1 for details on sample selec-
tion procedures.
5
characteristics. To represent underwriting fees at the
firm level, we first average them by firm. The mean
(median) of average underwriting fee for CDS firms is
73 bps (65 bps), and it is 80 bps (64 bps) for non-CDS
firms, with the differences in underwriting fees being
statistically insignificant. Therefore, CDS and non-
CDS firms have similar underwriting fees before the
initiation of CDS trading. We also show some evi-
dence that CDS and non-CDS firms are on parallel
trends with respect to their underwriting fees before
the initiation of CDS trading by visually examining
the time-series trends in underwriting fees presented
in Figure 2. Additional statistical tests reported in
Table IA-2 of the online appendix confirm the simi-
larity in these time-series trends during the pre-CDS
initiation period.11
3. Main Analysis
In this section, we examine the relationship between
the availability of CDS on a firm and the firm’s bond
underwriting fees both visually and in a multivariate
setting, and we run additional tests to address the
endogeneity concerns associated with the initiation of
CDS trading on the firm.
3.1. Baseline Results
We begin our main analysis graphically. In panel A of
Figure 3, we plot the average underwriting fees of
bonds issued by our final sample of 204 CDS firms
during the pre- and post-CDS initiation periods. In
panel B of Figure 3, we plot the differences between
the changes in underwriting fees of CDS firms and
those of benchmark non-CDS firms following the
inception of CDS trading. We calculate the average
underwriting fees in both panels by controlling for
Figure 2. Time-Series Trends in Underwriting Fees
Notes. This figure plots the average difference in underwriting fees
for bonds issued by 204 CDS firms (1,186 bonds) and their one-to-one
matched benchmark non-CDS firms (735 bonds) by years relative to
CDS initiation dates. The shaded area represents the 90% confi-
dence interval for the average difference in underwriting fees. See
Appendix B, Table 1, and Table IA-1 for details on sample selec-
tion procedures.
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
6 Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
Figure 3. Underwriting Fees Across Rating Categories
Notes. Panel A plots the average underwriting fees of bonds issued
by our sample of 204 CDS firms (1,186 bonds) during the pre- and
post-CDS initiation periods by credit ratings of bonds. Panel B plots
the difference-in-differences of underwriting fees that these CDS and
their one-on-one matched non-CDS firms experience with CDS ini-
tiations, also by credit ratings. The average underwriting fees are
calculated after controlling for the year fixed effects to account for
the macrotrends in underwriting fees. See Appendix B, Table 1, and
Table IA-1 for details on sample selection procedures.
year fixed effects to account for the macrotrends in
underwriting fees. We present our findings by the
credit ratings of the bonds.
Panel A of Figure 3 illustrates that CDS firms ex-
perience lower underwriting fees following the ini-
tiation of CDS trading on their debt. Panel B of
Figure 3 shows that this effect is more pronounced
when compared with the underwriting fees of non-
CDS firms and that it increases monotonically with
lower credit ratings. These observations provide pre-
liminary evidence that the initiation of CDS trading is
associated with a decline in underwriting fees and
that this effect is more pronounced among risk-
ier bonds.
To investigate the influence of CDS initiation on
underwriting fees in a multivariate setting, we esti-
mate the following regression model:
 
Underwriting Feeij
α + αi + αy + Xit β + Wj γ
+ δCDS Tradingij + εij , (1)
where Underwriting Feeij is the percentage under-
writing fee of bond j issued by firm i, α is the intercept,
αi and αy are firm and year fixed effects, respectively,
Xit is a set of quarterly firm characteristics measured
before the bond offering date (t), Wj represents bond
characteristics, CDS Tradingij equals one for bonds
issued by CDS firms during the post-CDS initiation
period and zero for the remaining bonds, and εij is the
error term. We cluster standard errors at the firm level
to adjust the statistical significance of coefficient es-
timates for the correlation in errors within a firm. In
this regression, the coefficient on CDS Trading esti-
mates the within-firm evolution in underwriting fees
associated with the initiation of CDS trading.
Firm-level regressors include firm size, leverage,
stock volatility, profitability, and asset tangibility, all
described in Section 2. The bond-level controls in our
study are issue amount (normalized by firm size),
bond maturity, syndicate size (the number of lead
and comanagers), dummy variables indicating bond
features (callable, puttable, floating, convertible, and
global issue), credit ratings, and bond liquidity. We
measure credit ratings based on the median of bond
ratings received from Moody’s, S&P, and Fitch. Our
rating controls are at the bond—instead of the firm—
level to account for the variation in expected recov-
eries within a firm’s bonds related to their seniority
and security. We classify these ratings into six dummy
variables indicating AAA, AA, A, BBB, BB, and B
or below ratings. We control for post-issuance bond
(il)liquidity, measured as the percentage of non-
trading days in the month after the offering observed
in Bloomberg, because Davis et al. (2018) show that
bonds with greater post-issuance liquidity have lower
underwriting fees.12
We define two more bond-level variables using
FISD to account for the influence underwriters can
have on the underwriting fee. First, Underwriter Ex-
posure is the average ratio of the bond offering amount
to the book managers’ total underwritten amount
during the previous year. This variable controls for
the additional risk of unsold bonds when the in-
vestment bank underwrites a relatively large issue.
Second, Underwriter Relationship equals one if any of
the book managers in the bond syndicate has un-
derwritten the issuer’s bonds in the previous 10 years,
and zero otherwise. This 10-year period matches
the median rollover period for corporate bonds. The
Underwriter Relationship variable controls for the
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
influence of underwriting relationships on underwrit-
ing fees. Appendix B provides the statistics on all of
the bond-level variables in our study.
The firm- and bond-level control variables in our
regressions account for some of the CDS externalities
documented in the literature that may affect under-
writing fees, such as the influence of CDS initiation on
increasing the underlying firms’ leverage (Saretto and
Tookes 2013) and credit risk (Subrahmanyam et al.
2014), and reducing their bond liquidity (Das et al.
2014).13 Conditional on such externalities, our em-
pirical design sheds light on whether and how the
presence of CDS influences the difficulty of placing
bonds in the primary market as reflected in their
underwriting fees.
We first analyze the influence of CDS trading within
our sample of 204 firms with CDS. Columns (1), (2),
and (3) of Table 2 report the coefficient estimates from
these regressions of percentage underwriting fees.
Controlling for only year and firm fixed effects, re-
gression (1) shows that the coefficient estimate on
CDS Trading is –0.174 and is significant at the 1% level.
This finding suggests that underwriting fees of CDS
firms decline by 17.4 bps from pre- to post-CDS ini-
tiation periods. Regression (2), which includes firm
characteristics as additional control variables, shows
that the coefficient estimate on CDS Trading is –0.181,
and significant. Regression (3) includes firm and bond
characteristics as explanatory variables in addition
to firm and year fixed effects. The inclusion of bond
characteristics increases the model’s within R-squared
noticeably from 10.3% to 58.9%, showing that bond
characteristics are important determinants of the un-
derwriting fee. Controlling for firm and bond char-
acteristics, regression (3) shows that the coefficient
estimate on CDS Trading is –0.084 and is significant
at the 5% level, indicating an 8.4 bps decline in un-
derwriting fees for CDS firms following the initiation
of CDS trading. Thus, firms appear to benefit from
the initiation of CDS trading through a reduction in
their underwriting fees.
Next, we estimate the influence of CDS trading on
underwriting fees of CDS firms relative to those of
similar non-CDS firms to account for the possibility
that firms in general experience a reduction in their
underwriting fees around the time when CDS initi-
ations take place. We run the regression of under-
writing fees using our sample of 204 CDS firms and
their one-on-one matched non-CDS firms and report
the results in column (4) of Table 2. We find that the
coefficient estimate on CDS Trading is –0.116 and is
significant at the 1% level, suggesting a decline of
about 12 bps in underwriting fees associated with
CDS trading.14 Thus, the influence of CDS trading on
underwriting fees of CDS firms is even stronger
when estimated relative to non-CDS firms as opposed
7
to within CDS firms, perhaps because underwriting
fees exhibit a slight upward trend during our anal-
ysis period (see panel B of Figure 1 for details).
To put this CDS effect into perspective, the 12 bps
decline in underwriting fees following CDS initiation
is 17% of the average underwriting fee observed
during the pre-CDS initiation period, and it translates
into a $0.6 million reduction in issuance costs for the
average new bond issue in our sample. It is also
equivalent to 8% of the average bond risk premium
(offering yield minus the risk-free rate) of CDS firms
in our sample. Consequently, for a 10-year-to-maturity
bond, CDS initiation is associated with a 0.8% reduction
in the annual risk premium. Given the sheer size of the
corporate bond market, however, this seemingly small
reduction in issuance costs can lead to considerable
savings for the underlying firms in the aggregate.
According to the bond issuance data from Bloomberg,
the U.S. firms with CDS issued about $10 trillion in
face-value bonds during our analysis period of 2001 to
2013, and a 12 bps decline in issuance costs translates
into a $12 billion increase in the net proceeds these
firms receive from bond offerings.
A decline in underwriting fees following CDS initi-
ation implies that the competition among investment
banks is sufficient for the benefits of CDS to accrue to
the issuers. However, if the competition is imperfect,
investment banks might be splitting the gains from
the declining underwriting costs with—instead of
fully transferring them to—issuers. In this case, the
role of CDS in reducing underwriting costs can be
more economically significant than that measured by
a 12 bps reduction in underwriting fees. Our finding,
however, does not suggest that investment banks are
worse off following the initiation of CDS trading
on issuers. Instead, we conjecture, and later provide
suggestive evidence, that a decline in underwriting
fees represents a reduction in the costs investment
banks incur in underwriting the offer.
Overall, the evidence in this section suggests that
bond issuance costs decline with the development
of the CDS market. Next, we investigate the robust-
ness of our findings by addressing the endogeneity of
CDS initiation.
3.2. Addressing Endogeneity Concerns
The empirical tests we have implemented so far do
not address the possibility that the timing of CDS
initiation is endogenous. For instance, if the intro-
duction of CDS on a firm coincides with an increase
in overall interest in its securities, the underwrit-
ing fee would also decline with CDS initiation. An-
other possibility is that if investors demand CDS
contracts when they expect an increase in the firm’s
credit risk, the CDS initiation event would reflect
higher expected credit risk and, therefore, increase
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
8 Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS

Table 2. Baseline Regressions of Underwriting Fees

Variables (1) (2) (3) (4)

CDS Trading −0.174*** −0.181*** −0.084** −0.116***

(–2.72) (–2.96) (–2.00) (–3.87)
Log(MVE) −0.110** −0.033 −0.028
(–2.13) (–1.10) (–0.98)
Long-Term Debt/Assets 0.311* 0.125 0.200
(1.74) (0.89) (1.60)
Stock Volatility 7.860*** 3.574** 2.753**
(2.96) (2.02) (2.02)
Net Income/Sales −0.196 −0.074 −0.092
(–1.61) (–1.18) (–1.52)
Tangibility 0.033 −0.294 0.106
(0.10) (–1.44) (0.56)
Bond Offering Amount/MVE 0.571* 0.366
(1.75) (1.18)
Log(Bond Maturity in Years) 0.220*** 0.222***
(19.99) (20.98)
Bond Illiquidity 0.008 0.067**
(0.31) (2.25)
Callable Dummy 0.070** 0.067**
(2.06) (2.56)
Puttable Dummy −0.209*** −0.188***
(–2.86) (–2.66)
Global Issue Dummy −0.075*** −0.071***
(–3.09) (–3.25)
Floating Dummy 0.002 −0.031
(0.08) (–1.14)
Convertible Dummy 0.937*** 0.818***
(5.88) (5.91)
Underwriter Exposure 0.498** 0.843***
(2.51) (3.32)
Underwriter Relation Dummy 0.018 0.016
(0.99) (1.16)
Log(Number of Lead Underwriters) −0.013 0.024
(–0.59) (1.25)
Log(1+Number of Co-underwriters) −0.013 0.020
(–0.65) (1.09)
Intercept 1.204*** 2.121*** 1.450*** 1.006***
(7.13) (3.63) (4.10) (2.96)
Number of Observations 1186 1186 1186 1921
Adjusted R2 0.057 0.103 0.589 0.496
Firm Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Bond Credit Rating Fixed Effects No No Yes Yes

Notes. This table presents the results from the baseline regressions of underwriting fees. Columns (1),
(2), and (3) use the sample of 1,186 bond issues by 204 CDS firms, and column (4) contains an additional
735 bond issues from the matched benchmark non-CDS firms. Appendix B, Table 1, and Table IA-1
provide the details of sample selection criteria and variable definitions. Our regression equation is as
 
follows: UnderwritingFeeij
α + αi + αy + Xit β + Wj γ + δCDSTradingij + εij , where i, j, y, and t denote
firm, bond, year, and bond offering date, respectively, and X and W are controls for firm and bond level
characteristics, respectively. The variable of interest is CDS Trading that takes the value of one for bonds
with trading CDS and zero otherwise. Bond credit rating fixed effects are based on the median of ratings
from Moody’s, S&P, and Fitch. We classify the ratings into six categories: AAA, AA, A, BBB, BB, and B or
below. Reported in parentheses are t-statistics calculated using robust standard errors clustered at the
firm level.
*p < 0.10; **p < 0.05; ***p < 0.01.
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS 9

the underwriting fee. To address these concerns, we
exploit the period after natural disasters as a labo-
ratory to examine the CDS effect and implement an
instrumental variable approach.
3.2.1. Natural Experiment. We use natural disasters
as exogenous shocks to the demand for corporate
bonds from insurance companies and investigate
whether CDS availability reduces underwriting fees
more during these periods when placing bonds is
likely to be more difficult. It is important to use di-
sasters that are large enough that insurers will be
forced to change their balance sheets. Accordingly,
we obtain the 10 largest natural disasters in the U.S.
based on their costs to insurance companies from the
Swiss Re Institute’s website.15 Following the evidence
in Massa and Zhang (2011) that the negative impact
of Hurricane Katrina on the demand for bonds from
insurance companies lasted for several months, we
define a Post-Disaster (3-month) variable that equals
one for bond offerings announced within three months
after a natural disaster and zero otherwise. The Post-
Disaster (3-month) variable identifies 94 bond offer-
ings announced during the postdisaster periods.
Column (1) in Table 3 reports the coefficient esti-
mates on CDS Trading and its interaction with the
Post-Disaster (3-month) variable included in the base-
line regression of underwriting fees. The coefficient
estimate on the interaction term is negative and sig-
nificant, indicating that the effect of CDS on under-
writing fees is more pronounced during periods of
low investor demand for bonds. Because we select the
three-month period arbitrarily, we reestimate this
regression using Post-Disaster (2-month) and Post-
Disaster (1-month) variables that identify bond issues
announced during the two- and one-month periods
after the natural disasters, respectively.
Columns (2) and (3) in Table 3 report the coefficient
estimates on the interactions of CDS Trading with
Post-Disaster (2-month) and Post-Disaster (1-month)
variables, respectively. We find that the interaction
variables reported in both columns (2) and (3) are

Table 3. Alternative Identification Strategies

Model: Natural Experiment 2SLS

Variables (1) (2) (3) (4) (5) (6)

CDS Trading −0.110*** −0.110*** −0.110***

(–3.74) (–3.74) (–3.75)
Interactions
CDS Trading x Post Disaster (3-month) −0.085**
(–2.13)
CDS Trading x Post Disaster (2-month) −0.098**
(–2.12)
CDS Trading x Post Disaster (1-month) −0.145***
(–2.71)
Instrumented CDS Trading −0.157*** −0.202*** −0.144***
(–2.81) (–3.32) (–2.72)
Number of Observations 1921 1921 1921 2739 2739 2739
Adjusted R2 0.496 0.497 0.497 0.749 0.744 0.750
Firm Fixed Effects Yes Yes Yes No No No
Year Fixed Effects Yes Yes Yes Yes Yes Yes
Bond Credit Rating Fixed Effects Yes Yes Yes Yes Yes Yes
Firm and Bond Characteristics Yes Yes Yes Yes Yes Yes
Industry Fixed Effects No No No Yes Yes Yes

Notes. This table presents the results from regressions that address the endogeneity of CDS availability. See
Appendix B, Table 1, and Table IA-1 for sample selection criteria and variable definitions. Regressions (1),
(2), and (3) explore the 10 natural disasters in the United States with the largest insured losses during the
sample period as exogenous shocks to the demand for bonds from insurance companies. The regression
specification is the same as in regression (4), Table 2, where the dependent variable is the underwriting fee.
Post-Disaster (3-month), Post-Disaster (2-month), and Post-Disaster (1-month) variables identify the bond of-
ferings announced within three months, two months, and one month after a natural disaster, respectively.
Regressions (4), (5), and (6) report the second-stage regression results of the two-stage least square (2SLS)
regressions using Lender FX, Lender Tier 1, and both, respectively, as instrumental variables of CDS Trading.
Refer to Appendix A for the definitions of Lender FX and Lender Tier 1. The second-stage estimates are from
OLS regressions of the underwriting fees. The sample in these instrumental regressions includes all firms in
our initial sample of public bond issuers with nonmissing information described in Appendix A. Reported in
parentheses are t-statistics computed using robust standard errors clustered at the firm level.
*p < 0.10; **p < 0.05; ***p < 0.01.
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
10
negative and significant. Furthermore, we observe
that the magnitude of the coefficient estimate on the
interaction term increases monotonically from –0.085
in the three-month specification to –0.15 in the one-
month specification, suggesting that the CDS effect is
more pronounced as offering announcement dates
approach the exogenous disaster dates.
A concern with studying bond offerings following
natural disasters is that firms self-select to issue bonds
during these periods of low investor demand. Hence,
firms expecting to pay lower underwriting fees may
be the ones issuing bonds, thereby leading us to
observe lower underwriting fees. We believe that this
self-selection issue does not influence our experiment
because we compare the changes in underwriting fees
of both CDS firms and non-CDS firms that choose to
issue bonds during the same period. Furthermore,
firms determine the maturity structure of their bonds
several years before these disasters. Therefore, firms
may have limited flexibility to postpone rolling over
their bonds or switch to other forms of financing in
response to the decline in demand for bonds.
The findings in this section show that, as predicted,
the effect of CDS on underwriting fees is more pro-
nounced following natural disasters that serve as
exogenous shocks to the demand for bond offerings.
This evidence provides additional support for the
hedging channel through which the availability of
CDS reduces underwriting fees.16
3.2.2. Instrumental Variable Approach. An instrumen-
tal variable approach gives a direct way to address the
endogeneity of the timing of CDS initiation. In esti-
mating the instrumental variable regressions, we use all
firms with public bonds in our sample, instead of the
matched sample in our baseline specification, which
increases the sample size and helps us generalize our
findings. Prior literature on CDS used two instruments
for the initiation of CDS trading: The average foreign
exchange hedging positions (normalized by assets) of
institutions that have lending or bond underwriting
relationships with a firm (e.g., Saretto and Tookes
2013, Subrahmanyam et al. 2014, 2017) and the av-
erage Tier 1 capital ratio of a firm’s lenders (e.g.,
Subrahmanyam et al. 2014, 2017, Narayanan and
Uzmanoglu 2018b). Both are good candidate in-
struments in our setting, and hence we use these
variables individually and together as instruments
for CDS Trading.
The intuition behind using foreign exchange hedg-
ing as an instrument is that institutions that hedge
their foreign exchange risk are also likely to hedge their
credit risk exposure to a firm, thereby leading to the
initiation of CDS trading on the firm’s debt. That is,
some portion of the variation in CDS trading comes
from a desire by investors to hedge in general, as
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
opposed to a desire to hedge bonds of a specific firm
or a firm with specific characteristics. To construct
the instrumental variable, for each firm-quarter we
identify the bank lenders that serve as a lead ar-
ranger to loans within a five-year period preceding
the quarter from Dealscan.17 Next, we obtain their
foreign exchange hedging positions and assets from the
consolidated financial statements for holding compa-
nies maintained by the Federal Reserve. Finally, we
construct the instrumental variable—Lender FX—as the
average of foreign exchange hedging positions di-
vided by assets of the lenders, measured in the quarter
before the CDS initiation date. The average Lender FX
in our sample is 3.45% for the firms with trading CDS
and 2.33% for the ones without trading CDS, indi-
cating that foreign exchange hedging positions of
banks are positively correlated with the availability
of CDS on their borrowers.
The intuition behind the average Tier 1 capital ratio
of lenders—Lender Tier 1—is that lenders that have
lower Tier 1 capital ratios would be more likely to
demand CDS contracts on their borrowers to hedge
their credit risk, thereby increasing the probability of
CDS initiation on their borrowers. Thus, Lender Tier 1
captures the variation in CDS trading that comes from
the lenders’ desire to hedge because of their own
balance sheets as opposed to a desire to hedge because
of specific firm characteristics. We construct this in-
strument based on Tier 1 capital ratios of lenders
similarly to the way we construct the Lender FX in-
strument. We find that the average Lender Tier 1 is
6.6% for CDS firms and 7.0% for non-CDS firms. It
appears that, relative to non-CDS firms, CDS firms
borrow from less-capitalized lenders.
We estimate a probit regression of CDS Trading
using the firm-quarter panel data for all firms with
public bonds in our sample. Columns (3), (5), and (7)
in Appendix A report the results from this probit
regression using, respectively, Lender FX, Lender Tier 1,
and both of these variables as instruments. Col-
umn (3) shows that the coefficient estimate on Lender
FX is positive and significant, consistent with lenders’
foreign exchange hedging positions increasing the
probability of CDS listing. Column (5) shows that, as
expected, Lender Tier 1 is a negative and significant
predictor of CDS trading, suggesting that borrowers
of banks with lower capitalizations are more likely to
have CDS trading. When Lender FX and Lender Tier 1
together enter into the probit regression, column (7)
shows that Lender FX becomes an insignificant pre-
dictor of CDS Trading. Nevertheless, the incremental
likelihood ratio and F-tests suggest that Lender FX
and Lender Tier 1 are—both jointly and individually—
unlikely to be weak instruments.
Next, we run the second-stage regressions of un-
derwriting fees following the method explained in
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
Wooldridge (2002) for endogenous dummy variables.
Accordingly, we predict the probability of CDS trading
based on the coefficient estimates from the probit re-
gressions discussed above and use it as an instrument
for CDS Trading in the second-stage least squares
(2SLS) regressions of underwriting fees. Columns (4),
(5), and (6) in Table 3 report the coefficient estimates
on instrumented CDS Trading using Lender FX, Lender
Tier 1, and both, respectively, as instruments. We find
that the coefficient estimate on instrumented CDS
Trading is negative and significant in all settings, and it
ranges between –0.14 and –0.20, indicating a 14 to 20 bps
decline in underwriting fees following CDS initiation.
The findings in this section suggest that our base-
line finding is robust to controlling for the endoge-
neity of CDS listing and studying all firms with public
bond issues—instead of the propensity score matched
firms—in our sample.
4. Why Do Underwriting Fees Decline with
CDS Initiation?
Having established that our baseline finding is ro-
bust to endogeneity concerns, we next investigate
whether the hedging, information, or cross-selling
motives explain the reduction in underwriting fees
following CDS initiation and whether changes in the
underwriting market confound our interpretations.
4.1. Hedging Channel
The availability of CDS contracts on a firm expands
the hedging opportunities for investors. These hedging
benefits would make the underlying bonds more at-
tractive to investors, thereby reducing underwriting
costs.18 If CDS initiation reduces underwriting fees by
enhancing hedging opportunities, this effect should
increase with the credit risk of the offer. The CDS
effect should also increase with bond illiquidity as the
transaction cost advantage of hedging in the CDS
market (relative to trading credit risk in the bond
market) increases with bond illiquidity. Therefore, if
the decline in underwriting fees we observe with CDS
initiation obtains through a hedging channel, then the
CDS effect should also be more pronounced for il-
liquid bonds.
We investigate the variation in the CDS effect based
on proxies of credit risk (stock volatility, financial
distress, and high-yield rating dummy) and bond
liquidity (post-issuance bond illiquidity). We define
all of these variables in the earlier sections except
the Distressed Dummy variable that equals one if
the Altman’s Z-score (Altman 1968) measured in the
quarter preceding CDS initiation is below 1.81 and
zero otherwise. If the hedging benefits of CDS drive
the results, then distressed firms, firms with higher
stock volatility, high-yield rated bonds, and more
11
illiquid bonds should experience a greater reduc-
tion in underwriting fees with the inception of
CDS trading.
Columns (1), (2), and (3) in Table 4 report the co-
efficient estimates on CDS Trading and its interactions
with Distressed Dummy, Stock Volatility, and Bond Il-
liquidity, respectively, included in the baseline re-
gression of underwriting fees (i.e., regression (4) in
Table 2). The coefficient estimates on these interac-
tion variables are negative and significant. The re-
gression results reported in Table IA-4 show that the
coefficient on the CDS Trading and High-Yield Dummy
interaction is also negative but is insignificant with
a p-value of 0.105.19 To put the hedging effect in
context, the coefficient estimate on the distressed
dummy interaction (–0.092) suggests that financially
distressed firms experience an additional 9.2 bps
decline in underwriting fees with CDS initiation, and
the stock volatility interaction (–4.516) suggests that
firms with a daily stock volatility that is one per-
centage point higher experience an additional 4.5 bps
decline in underwriting fees with CDS initiation. The
coefficient estimate on CDS Trading becomes insig-
nificant after controlling for the stock volatility in-
teraction in column (2) of Table 4, suggesting that the
decline in underwriting fees following CDS initia-
tions is concentrated among riskier issuers. Mean-
while the illiquidity interaction (–0.089) suggests that
bonds with no trading within a month after offer-
ing (Bond Illiquidity equals one) experience an addi-
tional 8.9 bps decline in underwriting fees with CDS
initiation. These findings show that, as the hedg-
ing channel predicts, the CDS-related reduction
in underwriting fees increases with credit risk and
bond illiquidity.
We also examine the variation in our baseline
findings by alternative bond risk characteristics. Bai
et al. (2019) show that downside risk (DRF), credit risk
(CRF), and liquidity risk (LRF) factors are strong
determinants of future bond returns. Accordingly, we
estimate the DRF, CRF, and LRF factor loadings for
the bonds in our sample and examine whether the
CDS effect on underwriting fees increases as these
risks increase.20 We compute monthly bond returns
using the TRACE Enhanced database. We clean the
bond prices for errors following the approach of Dick-
Nielsen (2014). Following Bai et al. (2019), we also
drop when-issued, locked-in, and special condition
trades, trades with more than two days of settlement,
trading volumes of less than $10,000, prices less than
$5 and greater than $1,000, floating coupon bonds,
and convertible bonds.
To compute monthly bond returns, we first com-
pute daily bond prices as the volume-weighted av-
erage of prices in a day. We then identify the end-of-
month prices as the latest daily price observed during
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
12 Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS

Table 4. Source of the CDS Effect: Hedging Channel

Dependent variable: Underwriting fee Insurer and bank ownership

Variables (1) (2) (3) (4)

CDS Trading −0.079** −0.026 −0.087*** 0.056***

(–2.49) (–0.50) (–3.17) (2.94)
Interaction Variables
CDS Trading x Distressed Dummy −0.092**
(–2.22)
CDS Trading x Stock Volatility −4.516*
(–1.78)
CDS Trading x Bond Illiquidity −0.089**
(–2.19)
Number of Observations 1921 1921 1921 1669
Adjusted R2 0.497 0.497 0.497 0.273
Firm Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Bond Credit Rating Fixed Effects Yes Yes Yes Yes
Firm and Bond Characteristics Yes Yes Yes Yes

Notes. This table presents the results from regressions that examine the influence of CDS on underwriting
fees through the hedging channel. See Appendix B, Table 1, and Table IA-1 for sample selection criteria
and variable definitions. The hedging channel predicts that the CDS effect is more pronounced for riskier
and less liquid bond issues. To test this prediction, regressions (1), (2), and (3) introduce the interaction of
CDS Trading and proxies of risk and illiquidity, respectively, as an additional regressor to the baseline
specification. This table reports the coefficient estimates on the interaction terms from the following
 
regression: UnderwritingFeeij
α + αi + αy + Xit β + Wj γ + δCDSTradingij + ζCDSTradingij xProxyi + εij ,
where i, j, y, and t denote firm, bond, year, and bond offering date, respectively, and X and W are controls
for firm and bond level characteristics, respectively. The proxies for credit risk are Distressed Dummy,
indicating whether a firm’s Altman’s Z-score is below 1.81, and Stock Volatility, measured as the standard
deviation of stock returns. Bond Illiquidity is the percentage of nontrading days in the month following
the offering date. Regressions (1) and (2) estimate the variation results based on firm characteristics
measured in the quarter before the CDS initiation date, and hence they control for the direct effects
through controlling for firm fixed effects. Regression (3) controls for the direct effects of the bond il-
liquidity by including illiquidity as an additional control. This table also presents the results from
regressions that investigate the influence of CDS on bond ownership. The hedging benefits of CDS
should be more valuable to insurance companies and banks that are subject to risk-based capital re-
quirements compared with other investors who do not face such requirements. To test this prediction,
we measure the ownership of insurance companies and banks in each bond at the end of the quarter in
which bonds are offered and examine whether their combined bond ownership increases with CDS
initiation. Regression (4) estimates the influence of CDS on the percentage ownership of insurance
companies and banks in bonds. The sample in all regressions includes CDS and benchmark firms that
have at least one bond with nonmissing information during either the pre- or the post-CDS initiation
period. The number of CDS (and non-CDS) firms is 204 in regressions (1), (2), and (3) and is 195 in
regression (4). Reported in parentheses are t-statistics computed using robust standard errors clustered
at the firm level.
*p < 0.10; **p < 0.05; ***p < 0.01.

the five-day period before the end of the month. Fi-
nally, we compute monthly bond returns as
( )
Pjt + AIjt + Cjt
rjt
( ) − 1,
Pjt−1 + AIjt−1
where P is bond price, AI is accrued interest, and C is
coupon payment. We find that the average monthly
bond return in our sample is 0.59%, in line with the
0.68% reported by Bai et al. (2019).
We estimate each factor loading for a bond by
regressing its monthly excess returns (bond return
minus the monthly Treasury bill rate) on the factor,
and we require at least 24 monthly observations to
estimate the factor loadings. Our final bond sample
with factor loadings includes 1,063 bonds that are
issued by CDS firms (661 bonds) and their one-on-one
matched non-CDS firms (402 bonds). As a final step,
(2)
we identify the bonds with the above median factor
loadings as being risky and test whether the CDS effect
is more pronounced among these risky bonds. The
regression results reported in Table IA-4 show that,
consistent with the hedging channel’s prediction, the
CDS effect is more pronounced among riskier bonds
that have higher downside, credit, and liquidity risks.
Next, we examine whether our findings are more
pronounced for CDS firms with a more negative
CDS-bond basis. CDS-bond basis is the difference in
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
maturity-matched CDS spread and the underlying
bond’s spread (i.e., yield minus the risk-free rate). When
this CDS-bond basis is negative, arbitrageurs may pur-
chase CDS protection and simultaneously buy the un-
derlying bonds to profit when the CDS-bond basis
narrows. Oehmke and Zawadowski (2017) provide
evidence that CDS contracts with more negative basis
are more actively traded. Furthermore, the theoretical
model of Oehmke and Zawadowski (2015) predicts
that bonds with more negative basis have higher trad-
ing costs and there is greater uncertainty about their
default probabilities. Therefore, the hedging benefits
of CDS should be greater for bonds with more neg-
ative basis.
To construct the CDS-bond basis, we first obtain the
offering yield spread for the bonds in our sample from
the FISD database. We then find the CDS spreads for
these bonds using the term structure of CDS spreads
(0.5-, 1-, 2-, 3-, 4-, 5-, 7-, and 10-year CDS spreads) data
from Bloomberg. We match the maturity of bonds
with the maturity of CDS contracts by linear inter-
polation. In our sample, we have 485 bonds issued by
CDS firms during the post-CDS initiation period. We
are able to construct the CDS-bond basis for 416 of
these 485 bonds, and we find that on average, the
CDS-bond basis is –76 bps. This level of CDS basis is
consistent with the statistics reported in the literature
(e.g., Bai and Collin-Dufresne 2019). We then identify
the bonds with high (low) negative basis as those with
below (above) median negative basis in our sample.
We report in Table IA-4 that CDS trading reduces
underwriting fees more for bonds with high negative
basis. In this regression, we include the interaction of
CDS Trading and Bond Illiquidity variables as an ad-
ditional regressor to control for the influence high
negative basis may have on bond trading and li-
quidity. This finding complements the predictions of
the hedging channel. It is also consistent with our
earlier finding that the CDS effect is more pronounced
among riskier bonds (specifically among BB rated
bonds) because Feldhutter et al. (2016) document that
the CDS-bond basis is larger for riskier firms.
As a final test for the hedging channel, we inves-
tigate how CDS initiations influence the bond own-
ership structure. The Financial Accounts of the United
States report published by the Federal Reserve in 2013
shows that the major investors in the U.S. corporate
bond market are insurance companies (life insurance
and property-casualty insurance), mutual funds, re-
tirement funds (private and public), and commercial
banks. Among them, insurance companies and banks
face risk-based regulatory capital requirements that
impose additional costs for holding risky securities.
Hence, relative to other investors, insurance companies
and banks should benefit more from the new hedging
opportunities CDS create for the underlying bonds.
13
For insurance companies, the National Association
of Insurance Commissioners (NAIC) develops the
risk-based capital adequacy standards as a tool for
state regulators to monitor the capital deficiency of
insurance companies. Federal-level regulations also
imposed capital requirements for large insurance
companies following the financial crisis of 2008. Ac-
cordingly, the risk-based capital that insurance com-
panies are required to maintain varies based on the
credit quality of their holdings (Ellul et al. 2011). For
banks, the Basel Committee on Banking Supervision
develops the risk-based capital standards that the
banking regulators in the United States adopt. Banks
are also required to maintain minimum capital levels
that increase with the riskiness of their assets.
The risk-based regulations thereby impose costs for
insurance companies and banks when the credit risk
of their bond holdings increases, and these costs do
not arise for other major investors in the bond market.
Thus, if CDS provide hedging (or regulatory arbi-
trage) benefits, the relative ownership of insurers and
banks in the bonds of reference firms should increase
with the inception of CDS trading. To test this pre-
diction, we obtain the ownership data for each bond
in our sample from Bloomberg as of the quarter-end
immediately following the offering date.
Bloomberg aggregates the holding data for each
institution using the 13F and Schedule D filings. The
13F form is an SEC form that must be filed quarterly
by institutional investors with $100 million or more
in equities. The Schedule D form is a filing made to
NAIC by U.S. insurance companies disclosing their
holdings and trades in their securities portfolios. We
are able to construct the ownership data for 1,669
bonds in our sample as of the end of the quarter in
which they were issued. We find that, on average,
insurance companies and banks combined hold 25%
of the bonds in our sample. To investigate the in-
fluence of CDS on bond ownership, we run a re-
gression of percentage insurer and bank holdings,
controlling for the firm and bond characteristics in
our baseline specification. Column (4) in Table 4 re-
ports that the coefficient estimate on CDS Trading is
0.056 and significant. This finding suggests that the
bond ownership of institutions with risk-based cap-
ital requirements increases by almost six percentage
points with the inception of CDS trading, consistent
with the role of CDS in creating valuable hedging
opportunities for investors.21
Overall, the findings in this section suggest that the
reduction in underwriting fees arises from the role
CDS play in improving risk sharing.
4.2. Information Channel
The literature shows that the CDS market reveals
new information on credit risk, which may reduce
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
14
adverse selection costs in bond offerings and thus
lower the underwriting fee. The information benefits
of CDS should be more pronounced for informationally
opaque firms and for firms with higher information
asymmetry because CDS spreads are more likely to
provide new information about these firms’ credit-
worthiness. To investigate this prediction, we study
the variation in the CDS effect by proxies of a firm’s
information environment.
We define five variables to proxy a firm’s infor-
mation environment: Three variables for informa-
tion opaqueness (number of analysts following, Rav-
enPack news, and Google search volumes) and two
variables for information asymmetry (analyst forecast
dispersion and relative bid-ask spread of stock price).
As analysts produce information through publishing
their research reports, firms with a greater number of
analysts following are likely to be less informationally
opaque. In a similar way, firms that have greater media
coverage—as proxied by their number of median
news and Google search volumes—are likely to be
less informationally opaque. Analyst forecast dis-
persion is a measure of disagreement among ana-
lysts regarding the firm’s prospects, which should
increase with higher information asymmetry. Finally,
bid-ask spread is a widely used proxy for informa-
tion asymmetry as dealers widen the spread when
the possibility of trading against informed investors
rises (e.g., Glosten and Milgrom 1985, Venkatesh and
Chiang 1986).
To compute the number of analysts following and
the analyst forecast dispersion variables, we obtain
the analyst data from the Institutional Brokers Esti-
mate System. As of the quarter before a firm’s CDS
initiation date, we define the number of analysts
following as the number of analysts providing annual
earnings estimates for the firm and calculate analyst
forecast dispersion as the standard deviation of an-
nual earnings estimates for the firm. We compute the
bid-ask spread as the average daily stock bid-ask
spread scaled by the closing price during the quar-
ter before the CDS initiation date. To construct the
news-based measures of information opaqueness, we
use news count data from RavenPack (i.e., Dow Jones
Edition and PR Edition) and Google search volume
data from Google Trends. We match our sample of
firms with RavenPack using their names and tickers
and obtain their historical news search volumes since
2000. We then define a RavenPack News variable as
the average number of times a firm is mentioned in the
news before the quarter of the CDS initiation date. The
Google search volume data from Google Trends are
available for years since 2004, which is after the
majority of CDS initiations take place. Thus, instead
of measuring the search volume in the quarter before
CDS initiations, we collect a firm’s monthly Google
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
search volume in Google Trends from the start of the
data until the end of our sample period. We then
assign the average number of searches for each firm as
its search volume. The RavenPack News variable is
available for 214 of 258 firms in our sample. The
average RavenPack News in our sample is 53, consis-
tent with the 51 news mentions reported in Dai et al.
(2015). The Google News variable (standardized based
on the IBM Google Search Volume) is available for all
258 firms in our sample with an average value of 25
(i.e., 25% of IBM’s Google search volume).
Columns (1) and (2) in Table 5 report the coefficient
estimates on CDS Trading and its interactions with
Log(1+Number of Analysts) and Analyst Forecast Dis-
persion, respectively, included in the baseline re-
gression of underwriting fees (i.e., regression (4) in
Table 2). We find that the coefficient estimates on
these interaction variables are insignificant. Regres-
sion estimates reported in Table IA-6 show that the
coefficients on the Log(RavenPack News), Log(1+Google
Search Volume), and Relative Bid-Ask Spread interac-
tions are also insignificant. These results suggest that,
within our firms with publicly traded bonds, CDS
initiation is unlikely to reduce underwriting fees
through revealing new information about firms.
4.3. Cross-Selling Channel
It is possible that underwriters get synergies between
providing underwriting services and initiating CDS
on the same firms. If so, underwriters may elicit
economic gains from cross-selling bonds and the
related CDS contracts to their clients, and accord-
ingly, offer lower underwriting fees to firms with
CDS. Because data on bank-firm specific CDS posi-
tions (i.e., positions of underwriters in issuers’ CDS)
that are needed to test this cross-selling channel di-
rectly are not available publicly, we test it indirectly
by examining the variation in the CDS effect by un-
derwriters’ overall CDS positions.
We first investigate whether the CDS effect is more
pronounced for underwriters who are also major CDS
dealers. To do so, we classify an underwriter as a CDS
dealer if it is listed as a CDS dealer on the website of
the International Swaps and Derivatives Association’s
Credit Derivatives Determinations Committee.22 We
find that about 40% of underwriters in our sample
are CDS dealers. These dealer-underwriters tend to
be larger than the average underwriter. We classify a
bond as underwritten by a CDS Dealer if any of its lead
underwriters is a CDS dealer. Column (3) in Table 5
reports that, in a regression of underwriting fees, the
coefficient on the interaction between CDS Trading
and CDS Dealer variables is insignificant, indicating
that whether an underwriter is a CDS dealer does not
influence the effect CDS has on underwriting fees.
In a similar way, we also examine whether the CDS
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS 15

Table 5. Source of the CDS Effect: Information and Cross-Selling Channels

Model: Information Cross-Selling

Variables (1) (2) (3) (4)

CDS Trading −0.113* −0.120*** −0.111*** −0.091**

(–1.92) (–3.54) (–3.04) (–2.45)
Interaction Variables
CDS Trading x Log(1+Number of Analysts) −0.000
(–0.06)
CDS Trading x Analyst Forecast Dispersion 0.054
(0.32)
CDS Trading x CDS Dealer −0.005
(–0.16)
CDS Trading x CDS Active Underwriter −0.014
(–0.38)
Number of Observations 1921 1921 1921 1692
Adjusted R2 0.496 0.496 0.495 0.490
Firm Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Bond Credit Rating Fixed Effects Yes Yes Yes Yes
Firm and Bond Characteristics Yes Yes Yes Yes

Notes. This table presents the results from regressions that examine the influence of CDS on underwriting
fees through the information and cross-selling channels. See Appendix B, Table 1, and Table IA-1 for
sample selection criteria and variable definitions. The information and cross-selling channels predict that
the CDS effect is more pronounced for less transparent firms and bonds underwritten by CDS-active
institutions, respectively. This table reports the coefficient estimates on CDS Trading and its interactions
with proxies for transparency and CDS activity from the following regression: UnderwritingFeeij

 
α + αi + αy + Xit β + Wj γ + δCDSTradingij + ζCDSTradingij xProxyi + εij , where i, j, y, and t denote firm,
bond, year, and bond offering date, respectively, and X and W are controls for firm- and bond-level
characteristics, respectively. The proxies for information transparency are Number of Analysts, providing
annual earnings estimates on the firm, and Analyst Forecast Dispersion, measured as the standard deviation of
annual earnings estimates. The proxies for an underwriter’s CDS positions are a CDS Dealer dummy,
indicating whether an underwriter is a major CDS dealer, and a CDS Active Underwriter dummy, indicating
whether an underwriter—on average—writes more CDS contracts than the median underwriter does
during our analysis period. Regressions (1) and (2) estimate the variation in results based on firm char-
acteristics measured in the quarter before the CDS initiation date, and hence they control for the direct effects
through controlling for firm fixed effects. Regressions (3) and (4) control for the direct effects of underwriters’
CDS positions by including those proxies as additional controls. Reported in parentheses are t-statistics
computed using robust standard errors clustered at the firm level.
*p < 0.10; **p < 0.05; ***p < 0.01.

effect is larger for underwriters who actively write
CDS contracts than for those who do not. We obtain
the notional amount of CDS contracts sold by un-
derwriters from their Y9C reports maintained by the
Federal Reserve and compute an underwriter’s CDS
activity as the average quarterly notional amount of
CDS contracts it sells during our analysis period. The
mean (median) quarterly notional amount of CDS
contracts sold is $1.2 billion ($1.4 billion) with a
standard deviation of $1.1 billion. We then classify
CDS active underwriters as those who sell more CDS
contracts than the median underwriter in our sample
does. Finally, we construct a dummy variable—CDS
Active Underwriter—that indicates whether any of a
bond’s lead underwriters is CDS active. Column (4)
in Table 5 shows that the coefficient estimate on the
interaction of CDS Trading and CDS Active Under-
writer is insignificant. These regression results show
that there is no meaningful difference in the CDS
effect across underwriters who engage in CDS-related
activities and who do not. This finding is inconsistent
with the cross-selling channel.
As an additional test, we investigate whether CDS
firms are more likely to switch underwriters fol-
lowing CDS initiation. If certain types of under-
writers offer lower underwriting fees for issuers with
CDS, these issuers would be more likely to switch
underwriters following the inception of CDS trading.
We find that 67% of issuers switch to new under-
writers during the post-CDS initiation period, but
this switching propensity is not statistically differ-
ent for CDS and non-CDS firms (see Table IA-7).23
This finding indicates that CDS initiation does not
affect the probability of an issuer switching to a new
underwriter, which is inconsistent with the cross-
selling channel.
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
16
4.4. Underwriter Quality and Underwriter
Competition Channels
A bond issuer’s choice of underwriters may relate
to the fees that the underwriter charges, and under-
writer characteristics are likely to be important de-
terminants of underwriting fees. If underwriter char-
acteristics are correlated with CDS initiations, our
inferences about the effect of CDS initiations may be
confounded by an omitted variable. Our interest here is
not whether underwriter characteristics affect under-
writing fees, per se, but rather whether underwriter
characteristics are correlated with CDS initiations
and hence create an omitted variable bias. We ex-
amine the association between CDS trading and
underwriter quality using bond underpricing, offering
yield spreads, and underwriter reputation as proxies
for underwriter quality.
Bond underpricing is the bond’s first trading day
price run-up relative to its offering price. Cai et al.
(2007) report a significantly positive first-day return
in bond offerings, suggesting that, on average, bonds
could be offered to the market at higher prices. Although
not a traditional measure of underwriter quality, a
higher-quality underwriter acting in the issuer’s best
interests might be able to get better pricing of the bond
issue. Therefore, if the underwriter quality declines,
the issuer would experience a greater underpricing.
A decline in underwriter quality would also increase
the offering yield spread (offering yield minus the
maturity-matched risk-free rate), which represents the
cost of bond financing.
We obtain the offering yields from FISD and adjust
them for the maturity-matched (through linear in-
terpolation) risk-free rates from Bloomberg. The av-
erage offering yield spread in our sample is 1.58%. To
estimate bond underpricing, we use the transaction
prices from the TRACE database. We estimate the
end-of-day price as the weighted average transaction
prices in a day where the trading amounts represent
the weights.
To minimize the number of missing observations,
we use the first closing day price available within seven
calendar days of the offering date to compute bond
returns.24 When they are not available in TRACE, we
obtain the first closing day prices from Bloomberg.
The mean (median) number of days between the date
of the first closing day price and the offering date is
1.73 (1.00) in our sample. We find that the average
first-day return—adjusted for the return on credit
rating matched bond indices from Bloomberg—is
0.52% in our sample.
We also proxy for underwriter quality with a tra-
ditional measure of underwriter reputation. We expect
more reputable underwriters to provide higher quality
underwriting services (e.g., Carter et al. 1998). There-
fore, if firms switch to less-reputable underwriters
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
following the inception of CDS trading, this would
reduce the underwriter quality. To test this predic-
tion, we estimate underwriter reputation in each
quarter as the underwriter’s market share in the U.S.
corporate bond market measured using the bond
issuance activity in the preceding year from FISD. For
each bond, we take the average of its lead under-
writers’ (also identified from FISD) market share,
which equals 10% for the average bond in our sample.
We then examine whether firms use lower-reputation
underwriters after CDS trading starts.
Columns (1), (2), and (3) in Table 6 report the results
from regressions of first-day returns, offering yield
spreads and underwriter reputations, respectively.
Controlling for the firm and bond characteristics, the
coefficient estimate on CDS Trading is insignificant in all
of these regression models, suggesting that CDS Trading
is not codetermined with underwriter quality.
We then ask whether controlling for underwriter
quality affects our estimates of CDS Trading on un-
derwriting fees, that is, “the underwriter quality
channel.” In each quarter, we rank underwriters into
reputation deciles and control for reputation fixed
effects in our baseline regressions of underwriting
fees to account for underwriter quality. Column (4) in
Table 6 shows that underwriting fees decline with
CDS initiation even after controlling for underwriter
reputation fixed effects.25 Column (5) in Table 6 re-
ports similar results from a regression that includes
fixed effects for each underwriter, thereby holding
time-invariant aspects of the underwriters constant.
Finally, we examine whether controlling for the com-
petition among underwriters influences our find-
ings, that is, the underwriter competition channel.
Columns (6) and (7) control changes in the com-
petitiveness of the bond underwriting market. Col-
umn (6) adds to our main tests a control for the
combined market share of the top five underwriters in
the market. Column (7) adds a control for the con-
centration of the market (logged Herfindahl index of
underwriter market shares). In addition, we use a
within-underwriter matching approach to select our
control sample of non-CDS firms We relegate these
results to Table 8, which describes several robustness
tests. Our conclusion, that CDS Trading is negatively
related to underwriting fees, remains unchanged.
5. Additional Analyses
5.1. Initiation of CDS Trading and Loan Fees
In this section, we investigate the influence of CDS
on loan fees because syndicated loans are also an
important source of corporate financing. Identifying
the initiation date of CDS contracts referencing cor-
porate loans (LCDS) is challenging because the data
on LCDS transactions are scarce. As bank lenders
can also purchase CDS contracts referencing their
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS 17

Table 6. Underwriter Quality and Competition Channels

First-Day Offering Yield

Dependent Variable: Return Spread Underwriter Reputation Underwriting Fee

Variables (1) (2) (3) (4) (5) (6) (7)

CDS Trading 0.042 −0.092 0.004 −0.108*** −0.077*** −0.115*** −0.116***

(0.15) (–1.04) (0.71) (–3.49) (–2.65) (–3.88) (–3.89)
Number of Observations 1312 1609 1921 1921 1921 1921 1921
Adjusted R2 0.092 0.594 0.344 0.517 0.739 0.498 0.498
Firm Fixed Effects Yes Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes
Bond Credit Rating Fixed Effects Yes Yes Yes Yes Yes Yes Yes
Firm and Bond Characteristics Yes Yes Yes Yes Yes Yes Yes
Underwriter Quality Fixed Effects No No No Yes No No No
Underwriter Fixed Effects No No No No Yes No No

Notes. This table presents the results from regressions that investigate the influence of CDS trading on underwriting quality and examines
whether the decline in underwriting fees associated with CDS trading is caused by changes in underwriter quality and underwriter competition.
See Appendix B, Table 1, and Table IA-1 for sample selection criteria and variable definitions. Our proxies for underwriting quality are bond
underpricing, offering yield spread, and underwriter reputation. Bond underpricing is the increase in the bond price from the offering price
measured at the end of the first day of trading. A greater underpricing suggests that bonds could have been sold at a higher offering price. To
reduce the number of missing observations, we use the first available closing day price in TRACE within seven calendar days of the offering date
to compute the first-day return and substitute the pricing data from Bloomberg when it is missing from TRACE. We adjust the first-day return by
the market return (return on the credit rating matched bond indices from Bloomberg) to control for the market conditions. Offering yield spread
equals the offering yield minus the risk-free rate. As a measure of the cost of debt, offering yield spread would increase with a decline in
underwriting quality. Underwriter reputation is the market share of a bond underwriter measured quarterly using its activity in the U.S.
corporate new issues market during the previous year. For each bond, underwriter reputation equals the average reputation of its lead un-
derwriters. The intuition is that a more reputable underwriter provides higher-quality underwriting services. Columns (1) through (7) report the
 
coefficient estimate on CDS Trading from the following regression: DependentVariableij
α + αi + αy + Xit β + Wj γ + δCDSTradingij + εij , where i,
j, y, and t denote firm, bond, year, and bond offering date, respectively, and X and W are controls for firm- and bond-level characteristics,
respectively. The dependent variables in regressions (1), (2), and (3) are First Day Return, Offering Yield Spread, and Underwriter Reputation,
respectively. The dependent variable in regressions (4) through (7) is Underwriting Fee. Regressions (4) and (5) include underwriter reputation
fixed effects (based on deciles of quarterly reputation rankings) and underwriter-level fixed effects, respectively, as additional controls for
underwriter ability. Regressions (6) and (7) include Top Five Underwriters’ Market Share and Log (Herfindahl Index of Underwriters), respectively, as
additional controls for the degree of underwriter competition. The sample in all regressions includes CDS and benchmark non-CDS firms that
have at least one bond with nonmissing information during either the pre- or the post-CDS initiation period. The number of CDS (and also non-
CDS) firms is 174 in regression (1), 184 in regression (2), and 204 in regressions (3) through (7). Reported in parentheses are t-statistics computed
using robust standard errors clustered at the firm level.
*p < 0.10; **p < 0.05; ***p < 0.01.

borrowers’ bonds to hedge their credit risk, we study
whether CDS initiation on bonds affects loan fees.
We obtain the loan data for our sample firms from
Dealscan (see Section 3.2.2 for details). The typical
fees for a revolver loan are facility fee, commitment
fee, utilization fee, and upfront fee, and those for a
term loan are upfront fee and cancellation fee. We first
estimate a regression of all types of fees, controlling
for firm characteristics, loan characteristics, and fee
types. We also estimate a regression of loan-level fees
(total loan fee) constructed following the method of
Berg et al. (2016).26 The average total loan fee is 28 bps
in our sample.
Columns (1) and (2) in Table 7 report the results
from regressions of individual loan fees and the total
loan fees, respectively. We find that the coefficient
estimate on CDS Trading is insignificant in both re-
gressions, suggesting that CDS initiation on a firm’s
bonds does not significantly influence its loan fees.
This result is consistent with the finding of Saretto
and Tookes (2013) that the capital structure influences
of CDS trading arises through corporate bonds rather
than through bank loans. Keeping in mind the limi-
tations of studying a sample of bond issuers, this
finding implies that CDS initiation primarily reduces
the issuance costs for the underlying debt class
(bonds), and hence the benefits of CDS in reducing
debt issuance costs accrue to the firms active in the
bond market. 27
5.2. Robustness Tests
We now examine whether our estimate of the CDS
effect is robust to alternative specifications and report
the findings in Table 8. We first examine the sensi-
tivity of our findings to the discretionary choices we
make in implementing the propensity score matching
method. In our baseline approach, we require the
absolute difference in the propensity scores of a CDS
and its one-to-one matched benchmark non-CDS firm
to be within a 10% distance. Alternatively, we con-
struct the benchmark sample by matching each CDS
firm with one and two non-CDS firms with the nearest
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
18 Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS

Table 7. CDS Initiation and Syndicated Loan Fees

Dependent Variable: Individual Loan Fee Total Loan Fee

Variables (1) (2)

CDS Trading −0.005 0.031

(–0.22) (0.01)
Number of Observations 4370 3415
Adjusted R2 0.377 0.360
Firm Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
Firm Credit Rating Fixed Effects Yes Yes
Firm and Loan Characteristics Yes Yes
Fee Type Fixed Effects Yes No

Notes. This table presents the results from the following regression model that investigates the influence
 
of CDS on loan fees: DependentVariableij
α + αi + αy + Xit β + Wj γ + δCDSTradingij + εij , where i, j, y, and
t denote firm, bond, year, and bond offering date, respectively, and X and W are controls for firm- and
bond-level characteristics, respectively. See Appendix B, Table 1, and Table IA-1 for sample selec-
tion criteria and variable definitions. The typical fees for a revolver loan are facility fee, commitment fee,
utilization fee, and upfront fee, and those for a term loan are upfront fee and cancellation fee. Each loan
can have multiple types of fees and the fee structure is not uniform across loans. In regression (1), the
dependent variable is the individual loan fee, and the control variables include firm characteristics, loan
characteristics, and dummy variables identifying the fee types. The loan characteristics are the log of loan
maturity in years, loan amount adjusted by firm size, and loan type (revolver or secured). In regression
(2), the dependent variable is a loan-level aggregate fee (total loan fee) estimated using the method of
Berg et al. (2016). The sample in all regressions includes CDS and benchmark non-CDS firms that have
at least one loan with nonmissing information during either the pre- or the post-CDS initiation period.
There are 200 CDS (and also non-CDS) firms in both regressions (1) and (2). Reported in parentheses are
t-statistics computed using robust standard errors clustered at the firm level.
*p < 0.10; **p < 0.05; ***p < 0.01.

propensity scores, respectively. In addition, we use a
“within underwriter” matching approach to select
our control sample of non-CDS firms. For each firm
CDS firm, we identify its lead underwriters. We then
find all the bonds underwritten by these underwriters
in the same quarter. Among these bonds, we match
the bond issued by the CDS firm with a bond issued
by a non-CDS firm that has the closest propensity
score to that of the CDS firm. Furthermore, we test
whether our findings are robust to implementing the
overlap weighting approach of Li et al. (2018), using
industry controls based on the two-digit SIC codes,
and double clustering standard errors by firm and year.
We also study the changes in underwriting fees within
a one-year and a three-year event window—instead of
the five-year event window—surrounding CDS ini-
tiation to investigate the robustness of our findings
to the choice of event horizon.
Next, we test the sensitivity of our baseline finding
to over- or underrepresentation of firms in the sample.
We first study the effect of CDS on underwriting fees
within a balanced sample, instead of the unbalanced
sample in our baseline specification, constructed by
randomly selecting one bond for each firm during the
pre- and post-CDS initiation periods. We then in-
vestigate the CDS effect using a sample of firms with
at least one bond offering during the five-year pe-
riod both before and after the CDS initiation date.
This sample ensures that all issuers in our sample
are represented during the pre- and post-CDS initi-
ation periods.
We also investigate whether our finding is driven
by outliers in the sample. We reestimate the baseline
model using a sample of bonds excluding those of-
fered during the financial crisis of 2008 (between
August 2007 and December 2009). Next, we eliminate
CDS firms (and their matched non-CDS firms) from
the sample if they have a propensity score less than
10% or greater than 90% to reduce the influence of
potential matching errors on our finding. We also
drop convertible bonds when estimating the CDS
effect because these hybrid securities may have dif-
ferent fee structures. As additional tests, we examine
the robustness of our finding to excluding puttable
bonds and including unrated bonds in our sample.
Finally, we investigate whether our baseline find-
ing holds when using alternative control variables.
We control for the bond offering amount instead of
the offering amount normalized by firm size as a
proxy for scale economies. We also control for the
liquidity of a firm’s stock as an additional regressor
because Butler and Wan (2010) show that stock li-
quidity of an issuer can also affect its bond under-
writing fees. Furthermore, we test whether including
the following variables as additional controls in our
baseline regressions influences our findings: Firm
characteristics measured at the end of the bond is-
suance quarter, quarter fixed effects, year-quarter
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS 19

Table 8. Robustness Tests

Coefficient estimate Number of Adjusted

Robustness tests on CDS trading observations R2

A. Alternative PSM approaches

Nearest One −0116*** (–3.87) 1927 0.496
Nearest Two −0.106*** (–3.96) 2725 0.507
Within Underwriters Matching −0.100** (–2.23) 1306 0.640
Overlap Weighting Matching −0.111*** (–3.57) 1921 0.503
Controlling SIC Code-Based Industry Dummies −0.075*** (–2.61) 1913 0.533
Double Clustering Standard Errors at the Firm-Year Level −0.116** (–2.46) 1921 0.633
B. Shorter event windows
One-Year Period −0.196*** (–3.14) 501 0.378
Three-Year Period −0.110*** (–2.80) 1207 0.493
C. Over- or underrepresentation of firms
Balanced Panel −0.117** (–2.02) 573 0.586
Non-Missing during both Pre- and Post-Periods −0.108*** (–3.48) 1487 0.548
D. Alternative samples
Excluding the Financial Crisis Period −0.110*** (–3.34) 1793 0.493
Excluding High and Low Propensity Scores −0.116*** (–3.34) 1590 0.523
Excluding Convertible Bonds −0.115*** (–3.72) 1812 0.443
Including Unrated Bonds −0.105*** (–3.45) 2055 0.543
Excluding Puttable Bonds −0.116*** (–3.89) 1844 0.444
E. Alternative control variables
Controlling Log(Bond Offering Amount) −0.123*** (–4.03) 1921 0.502
Controlling Stock Liquidity −0.129*** (–4.15) 1921 0.498
Controlling Firm Characteristics after Bond Offerings −0.107*** (–3.42) 1921 0.499
Controlling Quarter Dummies −0.107*** (–3.62) 1921 0.499
Controlling Year-Quarter Dummies −0.085** (–2.52) 1921 0.519
Controlling Year-Credit Rating Dummies −0.075** (–2.48) 1921 0.549
Controlling Covenant Dummies −0.068** (–2.36) 1921 0.578
Controlling Offering Yield Spread −0.113*** (–3.85) 1609 0.414
Controlling First Day Return −0.155*** (–5.20) 1312 0.503

Notes. This table presents the results from regressions that examine the robustness of our baseline finding. Reported in this table is the coefficient
 
estimate on CDS Trading from the following regression: UnderwritingFeeij
α + αi + αy + Xit β + Wj γ + δCDSTradingij + εij , where i, j, y, and t
denote firm, bond, year, and bond offering date, respectively, and X and W are controls for firm- and bond-level characteristics, respectively.
The sample in all regressions includes CDS and benchmark non-CDS firms that have at least one loan with nonmissing information during either
the pre- or the post-CDS initiation period. See Appendix B, Table 1, and Table IA-1 for sample selection criteria and variable definitions. Section A
reestimates the baseline regression using a sample of benchmark non-CDS firms selected based on the nearest match method, the nearest two
matches method, the within underwriter match method, the overlap weighting match method proposed by Li et al. (2018), and SIC Code-based
industry definitions and by double clustering standard errors at the firm and the year level. Section B reestimates the baseline regression
including firms that issued bonds in the one-year and three-year periods—instead of the five-year period—either before or after the CDS
initiation date. Section C uses alternative sample selection methods to address the potential biases caused by over or under representation of
sample firms. Specifically, we estimate the CDS effect using a balanced panel data set constructed by randomly selecting a bond for each firm
during the five-year period pre- and post-CDS initiation. We also analyze the CDS effect using firms with at least one bond issuance during the
five-year period both before and after the CDS initiation date. Section D estimates the CDS effect using (a) a sample that excludes bonds issued
during the financial crisis of 2008 (between August 2007 and December 2009), (b) a sample that excludes CDS firms (and their matched non-CDS
firms) if their probability of CDS initiation (propensity score) estimated in regression (1) of Appendix A is less than 10% or greater than 90%, or
(c) a sample that excludes convertible bonds, (d) a sample that includes both rated and unrated bonds, and (e) a sample that excludes puttable
bonds. Section E controls for issue size using Log(Bond Offering Amount) instead of Bond Offering Amount/MVE, Stock Liquidity, Underwriter
Competition, Firm Characteristics after Bond Offerings, Quarter Dummies, Year-Quarter Dummies, Year-Credit Rating Dummies, Covenant Dummies,
Offering Yield Spread, and First Day Return. Stock Liquidity is Amihud’s (2002) illiquidity measure estimated during the one-month period before
the offering date. Year-Credit Rating Dummies are a set of dummy variables that are created based on the interactions of the year dummies and
credit rating dummies. Covenant Dummies indicate whether bonds have the following covenants using the FISD database: Asset sale clause,
change control put provision, consolidation merger, cross acceleration, cross-default declining net worth, dividends related payments, economic
covenant defeasance, fixed charge coverage, funded debt, indebtedness, investments, subsidiary investments unrestricted, liens, maintenance
net worth, negative pledge covenant, rating decline trigger put, restricted payments, sale assets, sale leaseback, stock issuance, subsidiary
dividend related payments, subsidiary fixed charge coverage, subsidiary funded debt, subsidiary indebtedness, subsidiary liens, subsidiary
preferred stock issuance, subsidiary sale assets unrestricted, subsidiary sales leaseback, subsidiary stock issuance, subordinated debt issuance,
subsidiary guarantee, subsidiary redesignation, and transaction affiliates. Refer to Mansi et al. (2013) for definitions of these covenants and
Table 6 for the definitions of Offering Yield Spread and First Day Return. All regressions control for firm, bond, and underwriter characteristics,
and include firm, year, and bond credit rating fixed effects. Reported in parentheses are t-statistics computed using robust standard errors
clustered at the firm level.
*p < 0.1; **p < 0.05; ***p < 0.01.
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
20
fixed effects, year-credit rating fixed effects, bond
covenants, offering yield spread, and first-day return.
Table 8 reports the coefficient estimates on CDS
Trading from regressions that implement the afore-
mentioned alternative specifications and shows that
the onset of CDS trading is associated with a decline of
about 12 bps in underwriting fees in most regression
estimates. These findings demonstrate that the negative
association between CDS initiation and underwriting
fees is not sensitive to the empirical specification.
6. Conclusion
CDS contracts are derivative instruments that allow
for swapping the credit risk of a reference obligation
from one party to another. In a perfect world, CDS are
redundant securities because a combination of the
underlying bond and a risk-free asset replicates their
payoffs. In a world with transaction costs and in-
formation asymmetries, however, the introduction of
CDS can have real effects on the underlying bond
market. In this paper, we investigated the influence
CDS have on the financial intermediation of the bond
offering process.
We provided evidence that CDS contracts can affect
the intermediation of bond offerings by providing
new hedging opportunities to investors. Other papers
have examined this hedging channel indirectly by
looking at how CDS affect bond offering yields (e.g.,
Ashcraft and Santos 2009). However, yields reflect,
among other things, ex post monitoring by banks.
Recent research, both theoretical (Parlour and Winton
2013) and empirical (Shan et al. 2019), shows that the
advent of CDS can lead to reduced monitoring in-
centives by banks because they can lay off default risk
through CDS contracts, and that reduction in moni-
toring is priced in bond offerings (Ashcraft and Santos
2009), creating a confounding effect that makes it
difficult to estimate the beneficial effect CDS might
have through improving risk sharing. Our approach
is, arguably, better for studying this hedging channel
because the underwriting process ends at the issuance
and placement of bonds; there is, in general, no ex
post monitoring by the lead underwriter.
We find that CDS initiation is associated with a 17%
(12 bps) decline in bond underwriting fees on aver-
age. Although underwriting fees decline, the quality
of underwriting services, as proxied by bond under-
pricing, offering yield spread, and underwriter repu-
tation, is not affected by CDS initiation. The reduction
in underwriting fees is more pronounced among risk-
ier firms and illiquid bonds for which the hedging
benefits that CDS provide are greater. Furthermore,
investors with risk-based regulatory requirements—
insurance companies and banks—who can use CDS
to manage such requirements increase their partici-
pation in bond offerings following CDS inception on
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
the bonds. These findings suggest that the hedging
benefits of CDS lead to the decline in the cost of un-
derwriting bonds.
Overall, our findings show that financial innova-
tion in the form of credit insurance contributes to the
transactional efficiency of the bond market by re-
ducing financial intermediation costs. For the un-
derlying firms, this CDS-related decline in interme-
diation costs increases their net proceeds from bond
offerings and improves their financial flexibility.
Acknowledgments
The authors thank Lee Ann Butler, Pedro Cuadros-Solas,
David Denis, David Hwang, Murali Jagannathan, Hyunseob
Kim, Srini Krishnamurthy, David Maslar, Rajesh Narayanan,
Kristian Rydqvist, Joao Santos, Alessio Saretto, Mobina Shafaati,
Tom Shohfi, Heather Tookes, and the seminar participants
at Binghamton University and Rensselaer Polytechnic Institute,
as well as the conference participants at the 2017 Annual Global
Finance Conference and the 2017 Financial Management As-
sociation Annual Meeting for valuable comments. Elizabeth
Rasich provided excellent editorial suggestions. All remaining
errors are the authors’.
Appendix A. Estimating the Probability of
CDS Trading
This table presents the coefficient estimates and their
marginal effects from probit regressions that predict the
initiation of CDS trading. Our sample includes 630 nonfi-
nancial public U.S. firms (of which 252 have CDS) that issued
U.S. dollar denominated public bonds with nonmissing in-
formation between 1996 and 2013. Using these firms, we
construct a firm-quarter panel data set from 2001 to 2013. The
panel data starts in 2001 because it is the earliest year CDS
initiation is observed in our sample. We define a CDS Trading
variable that equals one for quarters following the CDS
initiation day and zero otherwise, and estimate probit re-
gressions of CDS Trading controlling for the firm charac-
teristics lagged by a quarter. Therefore, we estimate the
probability of CDS trading in a quarter based on observed
firm characteristics in the previous quarter. Lender FX and
Lender Tier 1 variables serve as the instruments for CDS
Trading. In a quarter, Lender FX for a firm equals the average
of foreign exchange derivatives (normalized by assets) used
for hedging purposes by banks that serve as the lead ar-
ranger in the firm’s syndicated loans within the five-year
period before the quarter, and Lender Tier 1 for a firm equals
the average of Tier 1 capital ratio of banks that serve as the
lead arranger in the firm’s syndicated loans within the five-
year period before the CDS initiation quarter. Credit rating
fixed effects are based on the S&P long-term ratings
grouped into six categories: AAA, AA, A, BBB, BB, and B or
below. Industry fixed effects are based on Fama and French
12-industry classifications. See Table 1 for the remaining
variable definitions. Column (1) reports the coefficient esti-
mates from probit regressions used for implementing the
propensity score matching. Columns (3), (5), and (7) report the
coefficient estimates from probit regression for implementing
instrumental variable methods. Columns (2), (4), (6), and (8)
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS 21

Table A.1. Estimating the Probability of CDS Trading

Method: Propensity Score Matching Instrumental Variable

Probit Marginal Probit Marginal Probit Marginal Probit Marginal

Model: Regression Effects Regression Effects Regression Effects Regression Effects

Variables (1) (2) (3) (4) (5) (6) (7) (8)

Log(MVE) 0.835*** 0.281*** 0.903*** 0.337*** 0.891*** 0.332*** 0.890*** 0.332***

(11.14) (12.02) (12.65) (13.24) (12.55) (13.19) (12.52) (13.16)
LT Debt/Assets 1.658*** 0.558*** 1.661*** 0.619*** 1.623*** 0.605*** 1.617*** 0.603***
(4.22) (4.30) (4.01) (4.05) (3.90) (3.94) (3.89) (3.93)
Stock Volatility 9.476*** 3.188*** 10.156*** 3.786*** 11.034*** 4.112*** 10.964*** 4.087***
(4.46) (4.36) (4.59) (4.54) (4.94) (4.90) (4.91) (4.87)
Net Income/Sales −0.123** −0.041** −0.173*** −0.065*** −0.171*** −0.064*** −0.170*** −0.063***
(–2.48) (–2.48) (–2.68) (–2.68) (–2.65) (–2.65) (–2.64) (–2.64)
Tangibility −0.054 −0.018 0.465 0.173 0.448 0.167 0.451 0.168
(–0.11) (–0.11) (0.91) (0.91) (0.87) (0.87) (0.88) (0.87)
Lender FX 3.971** 1.481** 1.500 0.559
(2.39) (2.41) (0.96) (0.97)
Lender Tier 1 −22.121*** −8.245*** −20.778*** −7.745***
(–3.04) (–3.08) (–2.87) (–2.91)
Intercept −9.365*** −10.059*** −7.862*** −8.042***
(–9.43) (–10.79) (–7.64) (–7.68)

Number of obs. 21919 19256 19246 19246

Pseudo R2 0.414 0.414 0.419 0.419

Tests for the IV

Incremental LR Test 130.87*** 255.39*** 261.10***
F-Test 247.29*** 233.51*** 232.54***

Year FE Yes Yes Yes Yes Yes Yes Yes Yes

Credit rating FE Yes Yes Yes Yes Yes Yes Yes Yes
Industry FE Yes Yes Yes Yes Yes Yes Yes Yes

Note. LT, long term; IV, instrument; obs., observations; FE, fixed effects.
*p < 0.10; **p < 0.05; ***p < 0.01.

report the marginal effects at the mean values of the ex-
planatory variables presented in columns (1), (3), (5),
and (7), respectively. Reported in parentheses are z-values
calculated using robust standard errors clustered at the
firm level.
Appendix B. Summary Statistics on
Bond Characteristics
This table compares the bond characteristics across the CDS
and matched non-CDS firms in our sample during the pre-
CDS initiation period. As described in Appendix A, our
initial sample includes 630 nonfinancial public U.S. firms
that have complete information and issued at least one U.S.
dollar denominated public bond during the 1996–2013
period. Among these firms, 211 with CDS issued bonds
during the 10-year period centered at the CDS initiation
date. For each of these CDS firms, we select (with re-
placement) a benchmark non-CDS firm with the closest
probability of CDS trading within a 10% absolute difference
using the probit estimates in column (1) of the table in
Appendix A. We also require the benchmark firms to have
bond offerings during the 10-year period centered at the
CDS initiation date. If a benchmark firm is referenced by a
CDS contract in the subsequent periods, we exclude its
bonds with trading CDS from the benchmark sample. This
table reports the summary statistics for the bond charac-
teristics issued by 204 CDS firms and their one-to-one
matched benchmark non-CDS firms. For each firm, bond
characteristics reported in the table equal the mean of each
characteristic on the firm’s bonds issued during the pre-
CDS initiation period. Offering Yield Spread equals the of-
fering yield minus the risk-free rate. Underwriter Exposure
is the ratio of bond offering amount to the book manager’s
total underwritten amount during the previous year. Un-
derwriter Relation Dummy equals one if any book manager in
the bond syndicate has underwritten the issuer’s bonds in
the previous 10 years and zero otherwise. Number of Lead
Underwriters and Number of Co-underwriters are the number
of lead managers and comanagers in a bond underwriting
syndicate, respectively. Offering Amount/MVE is a ratio of a
bond’s offering amount to the issuer’s market value of
equity. Log(Maturity in Years) is the natural logarithm of a
bond’s maturity measured at the offering date. Bond Illi-
quidity is the percentage of nontrading days in the month
following the offering date. Callable, Puttable, Global Issue,
Floating, and Convertible dummy variables indicate whether
a bond is a callable, puttable, global issue, floating coupon,
or convertible bond, respectively. The “Test of Differences”
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
22 Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS

Table B.1. Summary Statistics on Bond Characteristics

Benchmark
Sample: CDS firms non-CDS firms Test of differences Covariate balance

Variables Mean Median SD Mean Median SD t-statistic z-statistic Norm. diff.

Offering Yield Spread (%) 1.668 1.549 0.826 1.695 1.413 0.941 −0.27 0.44 −0.03
Underwriter Exposure 0.014 0.005 0.036 0.014 0.004 0.048 −0.02 3.32*** 0.00
Underwriter Relation Dummy 0.519 0.600 0.414 0.475 0.500 0.418 0.94 0.90 0.11
Number of Lead Underwriters 1.629 1.571 0.543 1.558 1.429 0.608 1.10 1.71* 0.12
Number of Co-underwriters 2.848 2.889 1.531 2.539 2.500 1.751 1.68* 1.99** 0.19
Offering Amount/MVE 0.064 0.047 0.060 0.059 0.035 0.062 0.61 1.53 0.07
Log(Maturity in Years) 2.214 2.301 0.513 2.236 2.135 0.565 −0.37 0.25 −0.04
Bond Illiquidity 0.455 0.238 0.441 0.467 0.318 0.429 −0.48 −0.10 −0.03
Callable Dummy 0.750 1.000 0.336 0.755 1.000 0.361 −0.11 −0.56 −0.01
Puttable Dummy 0.052 0.000 0.159 0.044 0.000 0.182 0.40 1.93** 0.05
Global Dummy 0.131 0.000 0.287 0.105 0.000 0.287 0.81 1.79* 0.09
Floating Dummy 0.089 0.000 0.195 0.040 0.000 0.132 2.62*** 3.01*** 0.29
Convertible Dummy 0.043 0.000 0.177 0.058 0.000 0.214 −0.68 −0.42 −0.08

SD, standard deviation; Norm. diff, normalized difference.

*p < 0.10; **p < 0.05; ***p < 0.01.

column reports the statistics from the tests of differences in
mean (t-statistic from a two-tailed Student’s t-test) and
median (z-statistic from a two-tailed Wilcoxon rank-sum
test) bond characteristics across the CDS and benchmark
non-CDS firms. The “Covariate Balance” column reports
the normalized differences in covariate means between the
CDS and benchmark non-CDS firms.
Endnotes
1
See Aldasoro and Ehlers (2018) for these statistics and more details
on the evolution of the CDS market.
2 10
See Augustin, Subrahmanyam, Tang, and Wang (2014, 2016) for a
detailed survey of the CDS literature.
3
Table IA-1 provides a summary of our sample selection criteria and
the number of observations in our sample after each screening step.
4
Instead of dropping these CDS firms with potentially inaccurate
CDS initiation dates, we could use them as controls for the treated
firms whose CDS initiation dates are potentially accurate. For two
reasons, we choose to drop these CDS firms from our sample. First,
because we cannot identify the CDS status of these firms before the
database inception dates (2001 for Bloomberg and 2004 for CMA), we
can only study their bonds issued after 2001 or 2004. Such a limitation
on the event window does not exist for the treated sample. This event
window mismatch across the treatment and control firms may in-
fluence our estimates of the treatment (CDS) effect. Second, including
the firms with potentially inaccurate CDS initiation dates in the
control sample complicates the interpretation of the effect CDS has on
underwriting fees.
5
This DTCC filter eliminates about 1% of the CDS observations,
suggesting that using the earliest CDS trading date from CMA and
Bloomberg databases is a reasonable approach to identify CDS ini-
tiation dates.
6
We find that a large number of firms with CDS have potentially
inaccurate CDS initiation dates. This could be because the earliest
CDS trading in our CDS databases (Bloomberg, CMA, and DTCC) is
observed in 2001 although the CDS market has existed since 1990s.
For instance, using the CDS data from the GFI Group, CreditTrade,
and Markit, Subrahmanyam, Tang, and Wang (2014) document that
237 firms had their CDS initiated between 1997 and 2000.
7 15
Section 3.2.2 reports the results from the analysis of underwriting
fees using our entire sample of 252 CDS firms and 378 non-CDS firms.
8
We follow Almeida and Campello (2007) and define asset tangi-
bility as (Cash + 0.715 × Receivables + 0.547 × Inventory + 0.535 ×
Capital)/Assets.
9
Figure IA-1 compares the number of bond issues between CDS and
non-CDS firms during the 10-year period surrounding CDS initiation.
On average, CDS firms issue more bonds than non-CDS firms do.
However, the change in the number of bond issues of the CDS and
non-CDS firms follows a similar trend. We observe that the number
of bond issues of both CDS and non-CDS firms declines with CDS
initiation. This pattern could be due to the influence of the 2008 fi-
nancial crisis, which overlaps with the most of the firms’ post-CDS
initiation period in our sample.
The average underwriting fee in our sample is slightly lower than
what has been reported elsewhere in the literature. For instance, Lee
et al. (1996) report that the average underwriting fee for the period
between 1990 and 1994 is 162 bps, Altinkilic and Hansen (2000) report
that it is 109 bps between 1990 and 1997, Livingston and Zhou (2002)
report that it is 72 bps between 1997 and 1999, and Sufi (2004) reports
that it is 76 bps from 1990 to 2003. This could be because our sample
of large firms (CDS-referenced issuers and their matched non-CDS
firms) pay lower underwriting fees than does the average firm.
11
We compute the differences in underwriting fees of CDS and non-
CDS firms in each year relative to CDS initiation dates and perform a
joint F-test for the null hypothesis that these differences in under-
writing fees are equal during the pre-CDS initiation period. In ad-
dition, we perform a similar test in a multivariate setting while con-
trolling for firm characteristics. We fail to reject the null hypothesis that
the time series of yearly differences in underwriting fees are equal.
12
Constructing bond liquidity measures using transaction vol-
umes and prices is not feasible in our study because of the limited
availability of these data from the TRACE database during our
analysis period.
13
Recent work shows that the influence of CDS introduction on the
bond market liquidity may be positive in the long run (Sambalaibat,
2014) and that it may differ based on the liquidity proxy used
(Oehmke and Zawadowski, 2015).
14
The effect is not transitory. We report in Table IA-3 that the first
bond issued after CDS initiation and the subsequent bond issues
during the analysis period experience a similar decline in under-
writing fees.
The disasters in our sample, with their dates and insured losses
(in billions of 2013 dollars) in parentheses, are Hurricane Charley
Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS 23

(August 9, 2004; $11), Hurricane Frances (August 25, 2004; $7), References
Hurricane Ivan (September 2, 2004; $18), Hurricane Katrina (August
Acharya VV, Johnson TC (2007) Insider trading in credit derivatives.
23, 2005; $75), Hurricane Rita (September 18, 2005; $13), Hurricane
J. Financial Econom. 84:110–141.
Wilma (October 15, 2005; $15), Hurricane Ike (Septemebr 1, 2008; $21),
Aldasoro I, Ehlers T (2018) The credit default swap market: What a
Tornado Super Outbreak (April 25, 2011; $8), Hurricane Irene (August
difference a decade makes. BIS Quart. Rev., ePub ahead of print
20, 2011; $7), and Hurricane Sandy (October 22, 2012; $31). Our results
June 5, https://www.bis.org/publ/qtrpdf/r_qt1806b.htm.
are similar when we study the natural disasters with insured losses
Almeida H, Campello M (2007) Financial constraints, asset tangibility,
greater than $10 billion.
16
and corporate investment. Rev. Financial Stud. 20:1429–1460.
We also consider using the implementation of the Big Bang Protocol Altinkilic O, Hansen RS (2000) Are there economies of scale in un-
(BBP) on April 8, 2009, as a laboratory to test the CDS effect. In brief, derwriting fees? Evidence of rising external financing costs. Rev.
the BBP induced upfront fees for trading CDS. Wang et al. (2019) Financial Stud. 13:191–218.
show that this fee is a function of a fixed coupon rate and CDS spread
Altman E (1968) Financial ratios, discriminant analysis, and the
levels, making CDS contracts with lower funding fees more liquid
prediction of corporate bankruptcy. J. Finance 4:589–609.
than those with higher funding fees. Therefore, the hedging benefits
Amihud Y (2002) Illiquidity and stock returns: Cross-section and
of CDS may be more pronounced for reference entities traded at
time-series effects. J. Financial Markets 5:31–56.
certain spread levels (i.e., lower funding fees) following the BBP.
Ashcraft AB, Santos JAC (2009) Has the CDS market lowered the cost
However, we do not find evidence of this, perhaps because the fi-
of corporate debt? J. Monetary Econom. 56:514–523.
nancial crisis of 2008 (which coincides with the pre-BBP period) or the
Augustin P, Subrahmanyam MG, Tang DY, and Wang SQ (2014)
other changes driven by the BBP (e.g., changes to the credit event
Credit default swaps: A survey. Found Trends Finance 9(1–2):
definitions) contaminate the experiment.
17
1-196.
We construct this instrumental variable using only the bank lenders
Augustin P, Subrahmanyam MG, Tang DY, Wang SQ (2016) Credit
to eliminate the possibility that the instrument predicts underwriting
default swaps: Past, present, and future. Annual Rev. Financial
fees based on its correlation with bond underwriter characteristics.
Econom. 8:175–196.
Our findings are nevertheless similar when we instrument CDS
Bai J, Collin-Dufresne P (2019) The CDS-bond basis. Financial Man-
initiation with the combined foreign exchange hedging positions of
agment 48(2):417–439.
bank lenders and bond underwriters. We obtain the Dealscan-
Bai J, Bali TG, Wen Q (2019) Common risk factors in the cross-section
Compustat link file from Michael R. Robert’s website. See Chava
of corporate bond returns. J. Financial Econom. 131(3):619–642.
and Roberts (2008) for details on constructing the data. When the
Berg T, Saunders A, Steffen S (2016) The total cost of corporate
links are not available, we manually match firms with the loan in-
borrowing in the loan market: Don’t ignore the fees. J. Finance 71:
formation from Dealscan.
18
1357–1392.
Boyarchenko et al. (2018) study the CDS and bond transactions of Blanco R, Brennan S, Marsh IW (2005) An empirical analysis of the
large financial institutions and show that these institutions rarely dynamic relation between investment-grade bonds and credit
change their positions in an entity’s bonds and CDS contracts si- default swaps. J. Finance 60:2255–2281.
multaneously in a week. This finding implies that hedging may not
Bolton P, Oehmke M (2011) Credit default swaps and the empty
be the only motivation for institutions to trade CDS contracts and
creditor problem. Rev. Financial Stud. 24(8):2617–2655.
that our identification strategy based on CDS introduction is a
Boyarchenko N, Costello AM, Shachar O (2018) Credit market choice.
noisy approach for estimating the hedging benefits associated
Preprint, submitted August 7, https://ssrn.com/abstract=3226951.
with CDS availability.
19
Butler AW, Wan H (2010) Stock market liquidity and the long-run
The greater benefits CDS provide for high-yield bonds arise from stock performance of debt issuers. Rev. Financial Stud. 23:
BB rated bonds. This finding suggests that marginally distressed 3966–3995.
bonds benefit the most from the availability of CDS as opposed to Butler AW, Grullon G, Weston JP (2005) Stock market liquidity and
severely distressed bonds. the cost of issuing equity. J. Financial Quant. Anal. 40:331–348.
20
These factors are available at Turan Bali’s website in monthly Cai NK, Helwege J, Warga A (2007) Underpricing in the corporate
frequency between July 2004 and December 2016. Refer to http:// bond market. Rev. Financial Stud. 20:2021–2046.
faculty.msb.edu/tgb27/ for more details. Carter RB, Dark FH, Singh AK (1998) Underwriter reputation, initial
21
We also examine whether the effect of CDS on bond ownership is more returns, and the long-run performance of IPO stocks. J. Finance
pronounced among riskier and more illiquid bonds. We find weak 53:285–311.
evidence supporting the credit risk prediction and no evidence sup- Chava S, Roberts MR (2008) How does financing impact investment?
porting the illiquidity prediction. Smaller sample sizes in these owner- The role of debt covenants. J. Finance 63:2085–2121.
ship tests may explain these weaker findings. See Table IA-5 for details. Chen H-C, Ritter JR (2000) The seven percent solution. J. Finance 55:
22 1105–1131.
See https://dc.isda.org/about-dc-committees/ for more details.
23
Table IA-7 also shows that CDS initiation does not significantly Dai L, Parwada JT, Zhang B (2015) The governance effect of the
influence the number of lead- and co-underwriters in bond syndicates. media’s news dissemination role: Evidence from insider trading.
24 J. Accounting Res. 53(2):331–366.
Following Cai et al. (2007), we also compute bond returns using the
Damodaran A, Lim J (1991) The effects of option listing on the un-
first trading price within 14 days of the offering date and find that our
derlying stocks’ return processes. J. Banking Finance 15:647–664.
results remain unchanged.
Das S, Kalimipalli M, Nayak S (2014) Did CDS trading improve the
25
We report in Table 8 that our findings are similar using offering market for corporate bonds? J. Financial Econom. 111:495–525.
yield spread and bond underpricing to proxy for underwriter quality. Davis RL, Maslar DA, Roseman BS (2018) Secondary market trading
26
See Section 4 (pp. 24–36) in the online appendix of Berg et al. (2016) and the cost of new debt issuance. Preprint, submitted March 28,
for details on constructing loan-level fees. https://ssrn.com/abstract=2954857.
27
Given that CDS initiation reduces issuance costs for bonds but Davydenko SA, Strebulaev I (2007) Strategic actions and credit
not for loans, firms may prefer bonds over loans to raise capital spreads: An empirical investigation. J. Finance 62:2633–2671.
following CDS initiation. We do not find evidence in support of Dick-Nielsen J (2014) How to clean enhanced TRACE data. Preprint,
this prediction. submitted December 3, https://ssrn.com/abstract=2337908.
 Butler, Gao, and Uzmanoglu: Financial Innovation and Financial Intermediation
24
Duffee GR, Zhou C (2001) Credit derivatives in banking: Useful tools
for managing risk? J. Monet. Econom. 48:25–54.
Ellul A, Jotikasthira C, Lundblad CT (2011) Regulatory pressure and
fire sales in the corporate bond market. J. Financial Econom. 101:
596–620.
Feldhutter P, Hotchkiss E, Karakas O (2016) The value of creditor
control in corporate bonds. J. Financial Econom. 121:1–27.
Figlewski S, Webb GP (1993) Options, short sales, and market
completeness. J. Finance 48:761–777.
Glosten LR, Milgrom PR (1985) Bid, ask and transaction prices in
a specialist market with heterogeneously informed traders.
J. Financial Econom. 14(1):71–100.
Gunduz Y (2018) Mitigating counterparty risk. Preprint, submitted
December 1, https://ssrn.com/abstract=2654591.
Gunduz Y, Ongena S, Tumer-Alkan G, Yu Y (2017) CDS and credit:
Testing the small bang theory of the financial universe with
micro data. Preprint, submitted December 20, https://ssrn.com/
abstract=2607909.
Hasan I, Wu D (2016) How large banks use CDS to manage risks:
Bank-firm-level evidence. Preprint, submitted May 2, https://
ssrn.com/abstract=2772965.
Imbens GW, Rubin DB (2015) Causal Inference for Statistics, Social, and
Biomedical Sciences (Cambridge University Press, Cambridge, MA).
Kumar R, Sarin A, Shastri K (1998) The impact of options trading on
the market quality of the underlying security: An empirical
analysis. J. Finance 53:717–732.
Lee I, Lochhead S, Ritter JR, Zhao Q (1996) The costs of raising capital.
J. Financial Res. 19:59–74.
Li F, Morgan KL, Zaslavsky AM (2018) Balancing covariates via pro-
pensity score weighting. J. Amer. Statist. Assoc. 113(521):390–400.
Livingston M, Zhou L (2002) The impact of Rule 144A debt offerings
upon bond yields and underwriter fees. Financial Management 31:
5–27.
Mansi S, Qi Y, Wald JK (2013) Debt covenants, bankruptcy risk, and
issuance costs. Preprint, submitted August 22, https://ssrn.com/
abstract=1786709.
Massa M, Zhang L (2011) The spillover effects of Hurricane Katrina
on corporate bonds and the choice between bank and bond
financing. Preprint, submitted March 11, https://ssrn.com/
abstract=1782791.
Morrison AD (2005) Credit derivatives, disintermediation, and in-
vestment decisions. J. Bus. 78:621–648.
Narayanan R, Uzmanoglu C (2018a) Credit default swaps and firm
value. J. Financial Quant. Anal. 53:1227–1259.
Management Science, Articles in Advance, pp. 1–24, © 2020 INFORMS
Narayanan R, Uzmanoglu C (2018b) Credit insurance, distress reso-
lution costs, and bond spreads. Financial Management 47:931–951.
Oehmke M, Zawadowski A (2015) Synthetic or real? The equilibrium
effects of credit default swaps on bond markets. Rev. Financial
Stud. 28:3303–3337.
Oehmke M, Zawadowski A (2017) The anatomy of the CDS market.
Rev. Financial Stud. 30:80–119.
Parlour CA, Winton A (2013) Laying off credit risk: Loan sales vs.
credit default swaps. J. Financial Econom. 107:25–45.
Roberts MR, Whited TM (2012) Endogeneity in empirical corporate
finance. Constantinides GM, Harris M, Stulz RM, eds. Handbook
of the Economics of Finance, vol. 2, Part A (Elsevier, Amsterdam),
493–572.
Ross AS (1976) Options and efficiency. Quart. J. Econom. 90:75–89.
Rubin DB (1990) Formal modes of statistical inference for causal
effects. J. Statist. Planning Inference 25:279–292.
Sambalaibat B (2014) CDS and sovereign bond market liquidity.
Working paper, Carnegie Mellon University, Pittsburgh.
Saretto A, Tookes HE (2013) Corporate leverage, debt maturity and
credit supply: The role of credit default swaps. Rev. Financial
Stud. 26:1190–1247.
Shan C, Tang DY, Winton A (2019) Do banks still monitor when there
is a market for credit protection? J. Accounting Econom. 68(2-3):
101241.
Subrahmanyam MG, Tang DY, Wang SQ (2014) Does the tail wag the
dog? The effect of credit default swaps on credit risk. Rev. Fi-
nancial Stud. 27:2927–2960.
Subrahmanyam MG, Tang DY, Wang SQ (2017) Credit default
swaps, exacting creditors and corporate liquidity management.
J. Financial Econom. 124:395–414.
Sufi A (2004) Does joint production of lending and underwriting
help or hurt firms? A fixed effects approach. Preprint, submitted
September 28, https://ssrn.com/abstract=596201.
Venkatesh PC, Chiang R (1986) Information asymmetry and the
dealer’s bid-ask spread: A case study of earnings and dividend
announcements. J. Finance 41(5):1089–1102.
Wang X, Wu Y, Yan H, Zhong Z (2019) Funding liquidity shocks in a
quasi-experiment: Evidence from the CDS Big Bang. Preprint,
submitted November 30, https://ssrn.com/abstract=2730877.
Wooldridge J (2002) Econometric Analysis of Cross Section and Panel
Data (MIT Press, Cambridge, MA).
Yasuda A (2005) Do bank relationships affect the firm’s underwriter
choice in the corporate bond underwriting market? J. Finance 60:
1259–1292.