Description
This paper aims to analyze how banks transmit shocks that hit the debt market to their
borrowers. Recent financial crisis demonstrated that the banking system can be a pathway for shock
transmission.
Journal of Financial Economic Policy
Do banks propagate debt market shocks?
Galina Hale J oão A.C. Santos
Article information:
To cite this document:
Galina Hale J oão A.C. Santos , (2014),"Do banks propagate debt market shocks?", J ournal of Financial
Economic Policy, Vol. 6 Iss 3 pp. 270 - 310
Permanent link to this document:http://dx.doi.org/10.1108/J FEP-03-2014-0023
Downloaded on: 24 January 2016, At: 21:49 (PT)
References: this document contains references to 31 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 119 times since 2014*
Users who downloaded this article also downloaded:
Faten Ben Bouheni, (2014),"Banking regulation and supervision: can it enhance stability in Europe?",
J ournal of Financial Economic Policy, Vol. 6 Iss 3 pp. 244-269http://dx.doi.org/10.1108/J FEP-11-2013-0059
Tong Li, (2014),"Shadow banking in China: expanding scale, evolving structure", J ournal of Financial
Economic Policy, Vol. 6 Iss 3 pp. 198-211http://dx.doi.org/10.1108/J FEP-11-2013-0061
Norbert Gaillard, (2014),"Assessing sovereign risk: the case of rich countries", J ournal of Financial
Economic Policy, Vol. 6 Iss 3 pp. 212-225http://dx.doi.org/10.1108/J FEP-03-2014-0017
Access to this document was granted through an Emerald subscription provided by emerald-srm:115632 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.
*Related content and download information correct at time of download.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Do banks propagate debt market
shocks?
Galina Hale
Federal Reserve Bank of San Francisco, San Francisco,
California, USA, and
João A.C. Santos
Federal Reserve Bank of New York, New York, NY, USA and Nova School
of Business and Economics, Lisbon, Portugal
Abstract
Purpose – This paper aims to analyze how banks transmit shocks that hit the debt market to their
borrowers. Recent fnancial crisis demonstrated that the banking system can be a pathway for shock
transmission.
Design/methodology/approach – Bank-level panel regressions.
Findings – This paper shows that when banks experience a shock to the cost of their bond fnancing,
they pass a portion of their extra costs or savings to their corporate borrowers. While banks do not offer
special protection from bond market shocks to their relationship borrowers, they also do not treat all of
themequally. Relationship borrowers that are not bank-dependent are the least exposed to bond market
shocks via their bank loans. In contrast, banks pass the highest portion of the increase in their cost of
bond fnancing to their relationship borrowers that rely exclusively on banks for external funding.
Research limitations/implications – These fndings show that banks put more weight on the
informational advantage they have over their relationship borrowers than on the prospects of future
business with these borrowers. They also show a potential side effect of the recent proposals to require
banks to use CoCos or other long-term funding.
Originality/value – The fndings are timely, given the ongoing debates on the proposals to introduce
bail-in programs and proposals to require banks to use CoCos or other long-term funding.
Keywords Banks, Debt, Credit, Bank subordinate debt, Bond spreads, Lending channel, Loan
spreads
Paper type Research paper
1. Introduction
Recent fnancial crisis demonstrated that the banking system can be a pathway for
shock transmission. In this paper, we analyze how banks transmit shocks that hit the
JEL classifcation – E51, G21, G32
The authors thank David Marqués, Evren Damar, Filipa Sá, Julio Rotemberg, Mark Flannery,
and seminar participants at Tilburg University, Paris School of Economics, Federal Reserve Bank
of San Francisco, Bank of Brazil, Bank of Canada workshop on Financial Institution Behaviors
and Regulations, ECB conference “The bank lending channel in the euro area: New models and
empirical analysis,” the 2010 University of Cambridge Conference on Networks, and the 2009
Gersenzee summer workshop for useful comments. The views stated herein are those of the
authors and are not necessarily the views of the Federal Reserve Banks of San Francisco or New
York, or the Federal Reserve System.
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1757-6385.htm
JFEP
6,3
270
Journal of Financial Economic Policy
Vol. 6 No. 3, 2014
pp. 270-310
© Emerald Group Publishing Limited
1757-6385
DOI 10.1108/JFEP-03-2014-0023
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
debt market to their borrowers. Traditionally, banks have funded their business with
deposits. Because deposits are largely inelastic with respect to the interest rate, this
source of funding gave banks an opportunity to shield their corporate borrowers from
economy-wide shocks. Berger and Udell (1992), for example, document that bank loan
rates move in a smoother fashion than the market interest rate, and Berlin and Mester
(1999) fnd that banks with more core deposits smooth loan interest rates in response to
adverse economic shocks.
In recent years, banks have been increasingly relying on the bond market to fnance
their business. At the end of 1988, the ratio of bond fnancing to deposit funding was 3.5
per cent among the top 100 US banks. By the end of 2007, this ratio had gone up to 9 per
cent. This change in banks’ funding choices is important because it is likely to make it
more diffcult for banks to shield corporate borrowers from economy-wide shocks. In
addition, it is likely to create an additional indirect link between the bond market and the
corporate sector, as shocks to the bond market may nowget propagated to the corporate
sector via banks’ loan pricing policies. We test for evidence of this link and identify
which corporate borrowers have become more exposed to bond market shocks as a
result of banks’ growing reliance on bond fnancing.
We consider two hypotheses. Our frst hypothesis links the interest rates banks
charge on their corporate loans to banks’ costs of issuing in the bond market.
Specifcally, we hypothesize that shocks to the costs banks pay to issue in the bond
market are transmitted to the corporate sector via banks’ loan pricing policies. The
alternative to this hypothesis is that banks perfectly shield their corporate borrowers
from the bond market shocks, absorbing all fuctuations in the cost of bond issuance.
Our second hypothesis is about the borrowers who are more likely to be affected by
shocks to banks’ cost of bond fnancing. Following Berlin and Mester (1999), we
hypothesize that the prospects of future business with their relationship borrowers
leads banks to protect these borrowers from shocks to banks’ cost of bond fnancing,
possibly passing these costs onto relationship borrowers over a longer period. Under
this condition, we expect banks to pass a smaller portion of the shocks to their bond
funding costs onto their relationship borrowers than onto their non-relationship
borrowers. An alternative hypothesis is following the hold-up theories of Sharpe (1990)
and Rajan (1992). It states that banks will pass a larger portion of the shocks to their
bond funding costs onto their relationship borrowers than onto their non-relationship
borrowers, as banks are more likely to have an informational advantage over their
relationship borrowers[1].
The effects of bond market shocks on loan rates under our second hypothesis or
under its alternative are likely to be more pronounced for relationship borrowers that do
not have access to the bond market. On the one hand, these borrowers will likely be more
dependent on banks for external funding, increasing banks’ prospects of future business
with them. On the other hand, banks will likely have a bigger informational advantage
over these borrowers, as these borrowers do not beneft fromthe information that comes
with the issues in the bond market. We use these differences to refne our second
hypothesis, by hypothesizing that if banks shield their relationship borrowers from
bond market shocks, then relationship borrowers that depend on themfor funding stand
to be the least affected by shocks to banks’ cost of bond funding. In contrast, if banks
take advantage of their informational monopoly over their borrowers, then these
bank-dependent relationship borrowers stand to be affected the most by such shocks.
271
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
To test our hypotheses, we start by investigating whether banks’ loan pricing varies
with the cost they paid to raise funding in the bond market the last time they issued a
bond prior to any given loan. We next test whether this link is stronger or weaker for the
borrowers that have a relationship with their bank. Finally, we investigate whether this
link is stronger for relationship borrowers that are more dependent on banks because
they do not have access to the bond market. To help disentangle our hypotheses, we
compare how banks’ loan pricing policies vary across borrowers when the bank
experiences a shock that increases its cost of bond fnancing versus when it benefts
from a shock that lowers its cost of bond fnancing.
We fnd, consistent with our frst hypothesis, that when banks experience a shock to
the cost they pay to issue in the bond market, they respond by passing a portion of the
shock onto their corporate borrowers. Importantly, we fnd that on these occasions,
banks do not offer special protection to their relationship borrowers: relationship
borrowers are less exposed to shocks to the cost of their banks’ bond fnancing than
non-relationship borrowers, but the difference is not statistically signifcant. On closer
inspection, we fnd that relationship borrowers that are not dependent on banks are less
exposed to shocks to the cost their banks pay to raise bond fnancing compared to
relationships borrowers that are dependent on banks for funding. Moreover, our results
show that when banks are able to raise bond fnancing at very low cost, they pass a
small portion of the resulting savings to all of their relationship borrowers. In contrast,
when banks experience a shock that substantially increases their cost of bond fnancing,
they pass the bulk of this cost onto their relationship borrowers that are dependent on
them, while fully protecting the relationship borrowers that have access to the bond
market. In other words, dependent borrowers are more exposed to shocks to banks
funding costs than non-dependent borrowers. Further, our evidence shows that
dependent borrowers are more exposed to shocks that raise the cost of banks’ bond
fnancing than to shocks that lower the cost of this funding source for banks.
These fndings do not support the hypothesis that the prospects of future business with
relationship borrowers leads banks to smooth, over time, the interest rates they charge on
loans to these borrowers. In contrast, our fndings are consistent with the idea that banks
take into account the informational advantage they have over their relationship borrowers
when they decide on their loan rates. Even though banks do not pass the entirety of the
shocks to their cost of bond fnancing onto their relationship borrowers that do not have
access to the bond market, possibly because these borrowers are not fully dependent on
them, banks do expose these borrowers the most to these shocks while protecting their
relationship borrowers that do have access to the bond market[2].
Our paper is most closely related to that of Berlin and Mester (1999) and complements
their work in at least three important respects[3]. Berlin and Mester focus on a period
when banks funded themselves almost entirely with deposits, the 1970s and 1980s.
Their key fnding is that banks’ use of core deposits makes it possible for them to shield
borrowers from economy-wide shocks, possibly because these deposits are
interest-inelastic. Our focus is on the two decades that follow their sample period
(1988-2007), and while we fnd evidence similar to theirs in the frst half of our sample
period, we also fnd that the effect of core deposit fnancing weakens in the second half
of our sample period. More importantly, we fnd that after controlling for the role of
deposit funding, banks still adjust their loan pricing policies in response to shocks they
experience when raising funding in the bond market. Consistent with increasing
JFEP
6,3
272
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
importance of bond fnancing for banks, this bond effect is only evident in the second
half of our sample period. It appears, therefore, that the absorbing effect of deposit
funding weakened over time, while the bond-funding effect grew in importance.
Our paper also extends Berlin and Mester’s analysis by investigating how the
propagation of the bond market shocks by banks differs across different sets of
borrowers, depending on whether they have a lending relationship with the bank and
whether they are dependent on the bank for external funding. In doing so, we also
investigate how banks pass the savings or the extra costs they incur when raising
funding in the bond market onto these sets of borrowers.
Finally, like Berlin and Mester, we have detailed information about bank lenders and
their loans[4]. In contrast to them, however, we also have information on the identity of
borrowers, which gives us the opportunity to control for frm-specifc factors known to
explain loan interest rates and to distinguish whether the borrower is a relationship
borrower and whether it is likely to be bank-dependent or not.
Our fndings are timely, given the ongoing debates on the proposals to introduce bail-in
programs and proposals to require banks to use CoCos or other long-term funding. These
debates have focused on the effects these programs would have on banks’ risk-taking
incentives and their contribution to solve problems of fnancial distress and have paid little
attention to the effects these programs might have on corporate borrowers relying on these
banks for funding. Because these proposals will require banks to access the bond market
periodically, our fndings suggest that they will further expose corporate borrowers, in
particular those that are dependent on banks, to the conditions in the bond market.
Our fndings are also important because they showthat as banks increasingly rely on
bond fnancing, they will fnd it more diffcult to promote relationship lending, which
remains a distinctive feature of banks[5]. Finally, our fndings show a new mechanism
that interlinks the fnancial intermediation done through banks with the intermediation
done through the debt market[6]. A common view in the fnancial architecture literature
is that banks and debt markets operate independently from each other[7]. Holmstrom
and Tirole (1997), Allen and Gale (2000) and Song and Thakor (2010) develop models in
which banks and fnancial markets complement each other, but none of them consider
the complementarity that we identify in this paper. In the study by Holmstrom and
Tirole (1997), the complementarity arises because access to bank funding allows some
borrowers to tap debt markets for additional funding. In the study by Allen and Gale
(2000), intermediaries provide individuals with insurance against unforeseen
contingencies in some states of nature, thereby eliminating the need for individuals to
acquire costly information. Analysis in Song and Thakor (2010) is the closest to the
complementarity we identify, but in their setting, banks rely on the equity market, not
the bond market, to raise the equity capital they need for regulatory reasons.
The remainder of our paper is organized as follows. The next section presents our
methodology and data, and characterizes our sample. Section 3 investigates whether the
interest rates banks charge their borrowers vary with the cost banks pay to issue in the
bond market. Section 4 investigates whether banks pass on the shocks to the bond
market equally to all corporate borrowers. This section also investigates whether banks
adjust their loan pricing policies differently when they experience shocks that
substantially increase their cost of bond fnancing as opposed to shocks that allowthem
to raise bond fnancing at a very low cost. Section 5 concludes the paper.
273
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
2. Methodology, data and sample characterization
2.1 Methodology
Our methodology has two parts corresponding to our two main hypotheses. Part I
investigates whether the spreads banks charge on their corporate loans vary with the
cost they pay to issue in the bond market. Part II investigates whether banks pass the
cost they pay to raise bond fnancing to a greater or lesser extent to borrowers with
whom they have a lending relationship.
2.1.1 Loan spreads and cost of banks’ access to the bond market. We start by
investigating whether the spreads banks charge on their corporate loans vary with the
cost they pay to raise funding in the bond market. To this end, we estimate the following
model of loan spreads:
LLOANSPD
b, f, l, t
? c ? ? LBK BOND COST
b
? ? LBBBSPD
t
? ?LIBOR
t
?
?
i?1
I
?
i
B
i, b, t?1
?
?
j?1
J
?
j
F
j, b, t?1
?
?
k?1
k
?
k
L
k, l
? ?
f, t
where, LLOAN SP D
b,f,l,t
is the natural log of the all-in-drawn spread over Libor of loan l
extended by bank b to frmf at date t[8]. According to Dealscan, our source of loan data, the
all-in-drawnspreadis ameasure of the overall cost of the loan, expressedas aspreadover the
benchmark Libor because it takes into account both one-time and recurring fees associated
with the loan; LBKBONDCOST
b
is the natural log of our measure of the cost bank b paid
the last time it issued in the bond market. The coeffcient on this variable, ?, measures the
elasticityof loanspreads withrespect to the cost banks paidto issue inthe bondmarket, and
so we wouldexpect it to be positive. Because bonds are fxedrate securities, we measure this
cost at the time of the bank’s most recent public bond issue (prior to the loan).
We consider the cost to issue in the bond market only if the bank issued bonds within
three years prior to the loan. If the bank issued bonds a long time ago, the cost it incurred
back then is less likely to affect its current loan pricing policy. To isolate the effect of the
bank cost of bond fnancing from the effect of a change in the overall interest rates or in
the overall “price” of risk at the time of the loan, we add the following two controls to our
model of loan spreads. We control for the cost to issue in the bond market at the time of
the loan by including in our model the log of the spread between triple-B and triple-A
primary yields on new bonds issued at the time of the loan, LBBBSP D
t
. Because this
spread tends to be correlated with the overall price of risk, we expect ? ?0. In addition,
we control for the level of interest rates at the time of the loan, byincludingLIBOR
t
inour
model. Because loan spreads are computed over Libor, the two variables tend to move in
opposite directions and so we expect ? ?0.[9]
We complement these controls for the overall interest rates in the economy, with
three sets of bank-, frm- and loan-specifc controls, B, F and L, which we describe next.
We frst discuss the set of frm-specifc variables that we use. A subset of these
variables, which includes LAGE, the log of the frm’s age in years (we compute the frm’s
age by subtracting the date the frm frst appeared in Compustat from the date of each
observation in the sample), and LSALES, the log of the frm’s sales in hundreds of
millions of dollars, control for the frm’s overall risk. Older frms are typically better
established and so less risky. Similarly, larger frms are usually better diversifed across
customers, suppliers and regions.
JFEP
6,3
274
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Asubset of frmvariables controls for the risk of the frm’s debt. It includes the frm’s
proft margin, P ROF M ARGIN (net income divided by sales); interest coverage, IN T
EREST COV (EBITDA divided by interest expense); the leverage ratio, LEVERAGE
(debt over assets); and its earnings volatility, EARNINGS VOL (the standard deviation
of the frm’s quarterly return on assets over the past three years). More proftable frms,
as well as frms with higher interest coverage, have a greater cushion for servicing debt
and so should pay lower spreads on their loans. In contrast, frms with higher
leverage and those with higher earnings volatility will likely have a higher probability
of default and so should pay higher spreads on their loans.
The next set of frm variables attempts to control for another aspect of credit risk – the
losses that debt holders incur in the event of default. To capture this, we consider several
variables that measure the size andqualityof the asset base that debt holders candrawonin
default, including the frm’s tangible assets, TANGIBLES (inventories plus plant, property
andequipment over assets); its advertisingexpenses, ADVERTISING(advertisingexpense
divided by sales); and its expenses with research and development, R&D (research and
development expense dividedbysales)[10]. Tangible assets lose less of their value indefault
than do intangible assets, so we expect this variable to have a negative effect on spreads. In
contrast, advertising expenses and R&Dexpenses, which proxy for the frm’s brand equity
and intellectual capital, respectively, are intangible, and so we also expect them to have a
positive effect onspreads. We also control for the value the frmis expectedto gainbyfuture
growth, MKTTOBOOK (frm’s market to book ratio), and the frm’s net working capital,
NWC (current assets less current liabilities over debt)[11]. Although growth opportunities
are vulnerable tofnancial distress, we alreadycontrol for the portionof the frm’s assets that
are tangible. Thus, this variable could have a negative effect on spreads if it represents
additional value (over and above the book value) that debt holders can, in part, access in the
event of default. Withregards tothe frm’s net workingcapital, as the frm’s liquidasset base
is less likely to lose value in default, we expect this variable to have a negative effect on
spreads.
We complement this set of frm controls with RELATIONSHIP, which is a dummy
variable equal to one if the frmborrowed fromthe same lead arranger in the three years
prior to the current loan[12]. In addition, we include dummy variables for single-digit
Standard Industrial Classifcation industry groups, as each industry may face
additional risk factors that are not captured by our controls, and include a time trend,
TREN D, to account for a potential secular trend in loan interest rates.
The next set of variables controls for aspects related to the loan that are likely to
affect loan spreads. It includes the log of loan amount in dollars, LAM OU N T; and the
log of the loan maturity in years, LM AT U RIT Y. Larger loans may represent more
credit risk, but they may also allow economies of scale in processing and monitoring
the loan. Similarly, loans with longer maturities may face greater credit risk, but they are
more likely to be granted to frms that are thought to be more creditworthy. So, the
effects of these variables on the spread are ambiguous. This set also includes dummy
variables equal to one if the loan has restrictions on paying dividends (DIV IDEN D
REST), is senior (SEN IOR) or is secured (SECU RED). All else equal, any of these
features should make the loan safer, decreasing the spread, but it is well known that
lenders are more likely to require these features if they think the frm is riskier (Berger
and Udell, 1990), so the relationship may be reversed. Because the purpose of the loan is
likely to affect its credit spread, we include dummy variables for loans taken out for
275
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
corporate purposes (CORP P U RP OSES), to refnance a loan (REF IN AN CE) and for
working capital purposes (WORKCAP ITAL). Similarly, we include dummy variables
to account for the type of the loan, in particular for lines of credit (CREDIT LIN E) and
for term loans (T ERM LOAN). Some of the loan controls are likely to be jointly
determined with loan spreads[13]. Because we do not have instruments for the various
loan controls, we address the concerns that may arise with this endogeneity by
estimating our models both with and without the set of loan controls. Using either
approach does not affect our key fndings.
Finally, we control for a set of bank-specifc variables. These variables aim at
controlling for aspects related to banks that are likely to play a role in their loan
pricing policies. In addition to potentially having a direct effect on the pricing of the
loan, these variables are also meant to proxy for unobserved characteristics of
banks, including their management quality that may affect their loan pricing
policies at a given time. LASSET S, the log of the bank’s total assets in hundreds of
millions of dollars, controls for bank size. Larger banks are likely to be better
diversifed or to have access to funding under better terms giving them the
opportunity to charge lower loan spreads. If safer banks are able to access funding
under better terms, then we also expect other measures of bank risk, such as the
return on assets, ROA; the volatility of return on assets, ROAVOL; and net loan
charge-offs as a fraction of assets, CHARGEOF F S0 to be correlated with the
interest rates banks charge on their corporate loans[14]. For the same reason we
expect the bank’s capital-to-assets ratio, CAPITAL, to be negatively related to loan
interest rates. This relationship may also arise because, according to Boot et al.
(1993), banks with low capital are more willing to consume reputational capital to
build up fnancial capital and thus are more likely to renege on implicit guarantees,
including the guarantee not to explore their informational monopoly.
We include the bank’s holdings of cash and marketable securities as a fraction of total
assets, LIQUIDITY, because banks with more liquid assets may fnd it easier to fund loans
onthe margin, leadingto lower loanspreads. We include the fractionof subdebt over assets,
SUBDEBT, to control for the importance of bond fnancing for the bank. Again, banks that
rely on this funding source are likely to fnd it easier to fund loans on the margin and so we
should expect them to charge lower spreads. This variable is also likely to have a negative
effect on loan spreads because bond fnancing is predominantly used by larger banks.
Finally, we include the fraction of total deposits over assets, DEPOSITS, to control for the
importance of deposit funding for the bank. Insured deposits are believed to be the least
expensive source of fundingfor banks. Our control for the importance of this fundingsource,
however, also includes uninsured deposits which tend to be more expensive[15].
2.1.2 Are relationship borrowers less exposed to the cost of banks’ bond funding? We
next investigate whether banks pass the shocks to their funding costs that arise with
their use of bond fnancing to all of their borrowers equally. Specifcally, we are
interested in learning whether borrowers that have a lending relationship with
banks are exposed to these shocks to a lesser or greater extent. If banks expect to
continue doing business with their relationship borrowers, they may be willing to
shield them from bond market shocks, thereby smoothing the interest rates they
charge their relationship borrowers over time. On the other hand, the pressure to
maintain their fnancial performance may lead banks to renege on implicit
guarantees, including the guarantee not to explore their informational monopoly
JFEP
6,3
276
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
over their relationship borrowers, and pass onto these frms a larger portion of the
cost increase they face in raising funding in the bond market[16].
Thus, we investigate whether banks shield their relationship borrowers from the
shocks they face to their bond funding costs or whether they exploit their informational
advantage over such borrowers. To that end, we estimate the following model:
LLOANSPD
b, f, l, t
? c ? ? LBK BOND COST
b
? ? RELATIONSHIP
f, t?1
? ? RELATIONSHIP
f, t?1
LBK BOND COST
b
? ?LBBBSPD
t
? ? LIBOR
t
?
?
i?1
I
?
i
B
i, b, t?1
?
?
j?1
J
?
j
F
j, f, t?1
?
?
k?1
K
?
k
L
k, l
? ?
f, t
.
where, all the variables are as defned in Part I. If banks shield their relationship
borrowers fromthe shocks to their bond fnancing costs, then we would expect ??0. If,
on the other hand, banks pass these shocks onto their relationship borrowers to a greater
extent than on other borrowers, then we would expect ? ?0.
These effects are likely to vary within the sample of relationship borrowers
depending on how dependent the borrower is on the bank for funding. For example,
banks will likely fnd it diffcult to pass shocks to their cost of funding onto borrowers
that have alternative funding sources, as these borrowers will respond by ending their
relationship with the bank. Banks will fnd it easier to pass these shocks onto borrowers
that depend on them for their funding, but the greater prospects of future business with
these borrowers may lead banks to shield them form shocks to their funding costs.
To investigate these possibilities, we need to distinguish borrowers that are
dependent on banks fromthose that are not. We assume that borrowers that have access
to the bond market are not bank dependent. Besides having access to an alternative
source of funding, there will also be more information available on these borrowers
coming from the opinions of bond analysts and ratings of rating agencies and the
spreads on their bonds. This additional information will reduce banks ability to hold
these borrowers up for higher interest rates Sharpe (1990) and Rajan (1992) when their
cost of funding goes up.
Our loan pricing model (2) distinguishes between borrowers that have and those that
do not have a lending relationship with their bank, and it compares howthese borrowers
are exposed to changes in the cost of bond fnancing of their banks. To avoid adding a
third level of interaction terms that would distinguish borrowers that have access to the
bond market among those that have relationship with their bank, we opted for
estimating model (2) separately for borrowers with a lending relationship with their
bank and those borrowers without such a relationship. Within these subsamples of
loans, we distinguish between borrowers with and without access to the bond market. In
other words, we replace RELAT ION SHIP
f,t?1
, in model (2) with M ACCESS
f,t?1
, our
proxy for borrowers’ access to the bond market, and estimate the model separately for
relationship borrowers and non-relationship borrowers.
We assume that borrowers have access to the bond market if they issued a public
bond in the recent past. For the purpose of our tests, we defne the recent past as the
three-year period prior to the loan. We get similar results if we assume that any borrower
that has issued at least one bond in the past has access to the bond market[17]. Further,
277
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
as according to Rajan (1992), the holdup problem is more acute for risky frms than safe
frms, among the relationship borrowers with access to the bond market, we would
expect those that are rated investment grade to be the least exposed to the bond market
shocks that affect their bank’s cost of funding. For this reason, we also distinguish
among the frms that have access to the bond market those that are rated investment
grade from those that are rated below investment grade. We use the rating of the
borrower’s most recent public bond (prior to a given loan) to determine whether it is
rated investment grade or below investment grade.
2.2 Data
The data for this project come from several data sources, including the Loan Pricing
Corporation’s Dealscan database (LPC), the Securities Data Corporation’s Domestic New
Bond Issuances database (SDC), the Center for Research on Securities Prices’s stock
prices database (CRSP), the Salomon Brother’s bond yields indices, Compustat and from
the Federal Reserve’s Call Reports.
We use LPC’s Dealscan database of business loans to identify frms that borrowed
from banks and to gather information on their loans. This database goes as far back as
the beginning of the 1980s, but in the frst part of that decade, it has a somewhat limited
number of entries. It is for this reason that we begin our sample in 1987. Our sample ends
in December 2007 to avoid any effects arising from the subprime fnancial crisis.
We rely on SDC’s Domestic New Bond Issuances database to identify which frms in
our sample issued bonds prior to borrowing in the syndicated loan market and to gather
information on banks’ bond issuance activity. We also rely on this database to gather
information on the bond issuance activity of banks.
We use Compustat to get information on frms’ balance sheets. Even though LPC
contains loans from both privately held and publicly listed frms, given that Compustat
is dominated by the latter, we have to exclude from our sample the loans borrowed by
privately held frms.
We rely on the CRSP database to link companies and subsidiaries that are part of the
same frm, and to link companies over time that went through mergers, acquisitions or
name changes[18]. We then use these links to merge the LPC, SDC and Compustat
databases to fnd out the fnancial condition of the frm at the time it borrowed from
banks and if, by that date, the frm had already issued bonds.
We use Salomon Brothers’ indices on the yields of new industrial long-term triple-A
and triple-B-rated bonds to control for the conditions in the bond market at the time
frms take out loans from banks.
Finally, we use the Reports of Condition and Income (Call Reports) compiled by the
FDIC, the Comptroller of the Currency, and the Federal Reserve System to obtain
bank-level data for the lead bank(s) in each loan syndicate.
2.3 Sample characterization
Table I characterizes our sample of 19,930 loans. These loans are extended by 381 banks
over the years 1987-2007 to 4,222 borrowers. The top panel compares the 150 banks that
had access to the bond market at the time of their loans with the 335 banks that relied on
deposit funding[19]. We classify a bank as having access to bond fnancing, if it issued
at least once in the bond market in the three years prior to the loan and it still had public
debt on its balance sheet at the time of the loan. Otherwise, we assume the bank funds its
JFEP
6,3
278
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
operations mainly with deposits. The middle panel compares the 16,212 loans in the
sample that “bond-fnancing” banks extended with the 3,718 loans extended by banks
that do not use bond fnancing. Finally, the bottompanel of the table compares the 3,529
borrowers that took out loans from bond-fnancing banks with the 1,699 borrowers that
took out loans from deposit-fnancing banks.
Looking at the top panel, it is apparent that, compared to banks that rely on deposit
funding, banks that use bond fnancing are larger, hold less liquidity and operate with a
lower deposit-to-asset ratio. Bond fnancing banks appear to be less proftable, as they
have a lower ROA. Further, these banks may be riskier. Their ROA is less volatile, but
they have a lower capital-to-asset ratio (CAP IT AL).
Table I.
Sample characterization
a
Variables
Bond fnancing
banks
Deposit fnancing
banks Difference p-value
Differences among banks
ASSETS 467.990 85.985 382.006 0.000
ROA 0.126 0.138 ?0.012 0.000
ROAVOL 0.001 0.002 ?0.001 0.000
CHARGEOFFS 0.107 0.110 ?0.003 0.244
LIQUIDITY 19.795 26.386 ?6.591 0.000
DEPOSITS 30.602 48.749 ?18.147 0.000
CAPITAL 7.386 8.151 ?0.765 0.000
Differences in the loan policies
LOANSPD 157.969 218.325 ?60.356 0.000
AMOUNT 431.923 136.613 295.309 0.000
MATURITY 4.050 3.572 0.478 0.029
SECURED 0.416 0.624 ?0.208 0.000
SENIOR 0.964 0.930 0.034 0.000
DIVIDENDREST 0.457 0.426 0.031 0.000
CORPORATEPURP 0.304 0.263 0.041 0.000
REFINANCE 0.623 0.444 0.179 0.000
WORKINGCAPITAL 0.179 0.214 ?0.035 0.000
TERMLOAN 0.391 0.352 0.039 0.000
CREDITLINE 0.578 0.593 ?0.015 0.101
RELATIONSHIP 0.635 0.509 0.126 0.000
Differences among borrowers
AGE 23.455 15.126 8.329 0.000
SALES 6854.742 1707.320 5147.422 0.000
PROMARGIN ?0.009 ?0.052 0.044 0.026
INTERESTCOV 26.890 24.171 2.719 0.641
EARNINGSVOL 45.652 19.041 26.612 0.000
LEVERAGE 0.319 0.302 0.017 0.000
TANGIBLES 0.730 0.740 ?0.010 0.167
ADVERTISING 0.011 0.009 0.002 0.000
RD 0.028 0.042 ?0.014 0.138
NWC 8.001 5.771 2.230 0.271
MKTTOBOOK 1.784 1.744 0.040 0.136
PBOND 0.596 0.329 0.267 0.000
(continued)
279
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
In the middle panel of the table, we see that bond fnancing banks charge, on average, 60
basis points less on their loans than deposit fnancing banks. This difference may arise
because bond fnancing banks extend signifcantly larger loans or because they extend a
larger fraction of their loans borrowers that are not bank dependent. Alarger portion of the
borrowers of these banks have a credit rating or issued bonds in the public bond market in
the three years prior to the loan, confrming that borrowers of these banks are less likely to
be bank dependent. That difference in interest rates may also refect a difference in the risk
of these banks’ borrowers. However, in this case, the evidence appears to be mixed. As we
can see fromthe bottompanel, compared to borrowers of deposit funding banks, borrowers
of bondfnancingbanks are older, larger, have better proft margins andhigher net working
capital and they have more growth opportunities. All of these features suggest that bond
fnancingbanks tendtoextendloans tosafer borrowers. There is alsoevidence that suggests
otherwise. For example, borrowers of bondfnancingbanks, onaverage, have lower interest
coverage and less tangible assets. Further, they have higher leverage ratios, and their
earnings are more volatile.
Last, that difference in the interest rates that these banks charge on their loans to
corporate borrowers may arise because of a difference in these banks’ funding costs.
Table I.
Variables
Bond fnancing
banks
Deposit fnancing
banks Difference p-value
PBONDIG 0.272 0.067 0.204 0.000
PBONDBG 0.196 0.143 0.053 0.000
Notes:
a
ASSETS Bank assets in $100 million dollars. ROA, returns on assets (net income divided by
assets); ROAVOL, standard deviation of the quarterly ROA computed over the past three years;
CHARGEOFFS, net charge-offs over assets; LIQUIDITY, cash plus securities over assets; DEPOSITS,
deposits over assets; DEPOSIT COST, product between the ratio of deposits over assets and the three
month LIBOR; INT EXPENSE, interest expenses on deposits alone (over deposits); numbers reported
are for only 13,009 of the 19,930 observations in the sample because it is missing for the remaining
banks; CAPITAL, equity capital over assets; LOANSPD, all-in-drawn loan spread over LIBOR at
origination; AMOUNT, loan amount; M ATURITY, loan maturity in years; SECURED, dummy
variable equal to 1 if the loan is secured; SEN IOR, dummy variable equal to 1 if the loan is senior;
DIVIDENDREST, dummy variable equal to 1 if the borrower faces dividend restrictions in connection
with the loan; CORP P URPOSES, dummy variable equal to 1 if the loan is corporate purposes; REF IN
ANCE, dummy variable equal to 1 if the loan is to refnance existing debt; WORKCAPITAL, dummy
variable equal to 1 if the loan is for working capital; T ERM LOAN, dummy variable equal to 1 if it is
a term loan; CREDIT LIN E, dummy variable equal to 1 if it is a credit lien; AGE, age of the borrower
in years; SALES, sales in millions of dollars; P ROF M ARGIN, net income over sales; IN T COV, the
interest coverage (EBITDA divided by interest expense); EARNINGS VOL, earnings volatility (the
standard deviation of the frm’s quarterly ROA over the past three years); LEV ERAGE, leverage ratio
(debt over total assets); TANGIBLES, tangible assets (inventories plus plant, property and equipment
over total assets); ADV ERT ISIN G, expenses with advertising scaled by the frm’s sales; R&D,
expenses with R&D scaled by the frm’s sales; N W C, net working capital; M KTOBOOK, market to
book value; P BON D, dummy variable equal to 1 if the borrower issued a public bond in the past three
years prior to the loan; P BON DIG, dummy variable equal to 1 if the borrower issued a public bond
which was rated investment grade in the past three years prior to the loan; P BON DBG, dummy
variable equal to 1 if the borrower issued a public bond which was rated below investment grade in the
past three years prior to the loan
JFEP
6,3
280
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Bond fnancing banks appear to be able to raise deposit funding at lower interest rates.
In addition, they can complement this funding source with bond fnancing.
2.3.1 Banks’ bond funding costs and their loan spreads. In Table II, we take a frst
look at whether there is a link between banks’ loan pricing policies and their bond
funding costs. To that end, we compare, in the frst rowof the table, the loan spreads that
banks charge their borrowers when the cost banks pay to issue in the bond market is low
and high, respectively. We proxy for this cost by the spread between triple-B and
triple-A primary yields on new bonds issued at the time the bank issued its most recent
bond prior to the loan, BBBSP D
b
. We assume the cost to issue in the bond market is low
(high) if BBBSP D
b
is in the lowest (highest) quartile of the distribution of this variable
during the sample period. When banks pay low bond yields to issue in the bond market,
theycharge, onaverage,137 bps ontheir corporate loans. Whentheypayhighyields to issue
bonds, theycharge their borrowers, onaverage, 176 bps. Thus, whenthe banks’ cost to issue
in the bond market goes up, they increase the spreads they charge on their loans by 38 bps.
Banks appear toshield, at least inpart, their relationshipborrowers fromthese cost hikes. As
we can see from the middle panel of Table II, when the cost banks pay to issue in the bond
market goes up, they increase the spreads on the loans they extend their non-relationship
borrowers by44basis points. Onthese occasions, theyincrease the spreads theycharge their
relationship borrowers by only 35 basis points.
Finally, as shown in the bottom panel of Table II, we investigate if it is important for
borrowers to have access to the bond market. We classify borrowers that issued in the
bond markets at least once in the three years prior to the loan date to have access to the
bond market. Borrowers that have never issued in the bond market or those that only
Table II.
Loan spreads and banks’
cost of bond fnancing:
univariate analysis
a
Low High
Difference p-value BK BOND COST BK BOND COST
Loan spreads: differences as bank cost to issue in the bond market changes
137.2 175.6 38.4 0.000
Differences depending on whether the borrower has a lending relationship with the bank
Relationship 125.6 161.0 35.4 0.000
No relationship 156.1 200.3 44.1 0.000
Differences depending on whether the borrower has access to the bond market
Relationship
Access 114.1 117.2 3.14 0.540
No access 132.4 192.9 60.5 0.000
No relationship
Access 136.4 151.8 15.4 0.060
No Access 162.8 219.4 56.6 0.000
Notes:
a
Computations limited to banks that rely on bond fnancing at the time of the loan; this means
the bank issued at least once in the bond market in the three years prior to the loan, and it has public debt
in its balance sheet at the time of the loan; the loan spread is the all-in-drawn loan spread over LIBORat
origination; high (low) BK BON D COST is the top (bottom) quartile of the difference between the
Moody’s indexes on the ex ante yields of triple-B- and triple-A-rated bonds at the time of the most recent
bond the bank issued prior to the loan; borrowers have a lending relationship with the bank if they
borrowed fromit at least once in the past three years prior to the loan; borrowers have access to the bond
market if they issued at least once in the bond market in the three years prior to the loan
281
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
issued more than three years ago (from the loan date) are classifed as bank dependent.
As we can see from that panel, effectively non-bank-dependent borrowers are less
exposed to the cost banks pay to raise funding in the bond market, irrespective of
whether they have a relationship with their bank. More importantly, among relationship
borrowers, those that are not dependent on banks for funding do not appear to be
exposed to shocks to the cost their banks pay to issue in the bond market. While
relationship borrowers that are bank dependent pay, on average, an additional 61 basis
points when it is more expensive for banks to issue in the bond market (meaning that
they issued when BBBSP D
b
was in the upper quartile of the distribution of this variable
as opposed to the lower quartile of that distribution), relationship borrowers that are not
bank dependent pay only an additional 3 basis points, an increase which is not
statistically different from zero.
In sum, the results of our sample characterization suggest that banks do adjust their
loan pricing policies in response to changes in the cost they pay to raise funding in the
bond market. Banks appear to shield their relationship borrowers from changes in the
cost of this source of funding, but only if they are not dependent on them for funding.
Relationship borrowers that are dependent on banks for external funding seem to be
exposed to shocks to the cost of funding of their banks. In the rest of this paper, we look
at the exposures of these cohorts of borrowers to the cost their banks pay to issue in the
bond market more closely, using multivariate analysis.
3. Do banks pass bond market shocks onto their borrowers?
We investigate, in this section, whether banks’ reliance on the bond market to fund their
activities creates a link between the cost they pay to issue bonds and the spreads they
charge on their corporate loans. In the next section, we investigate whether banks shield
their relationship borrowers from shocks to the cost they pay to raise bond fnancing.
Table III reports the results of our tests of whether banks that rely on bond fnancing
adjust their loan spreads in response to changes in the cost they pay to issue in the bond
market. We measure this cost at the time of banks’ most recent bond issue prior to any
given loan. Given that yields are missing for a large number of bonds issued by banks,
we proxy for that cost by the log of the spread between triple-B and triple-A primary
yields on new bonds issued on the day the bank issued its most recent bond prior to the
loan, LBBBBSP D
b
[20]. Model 1 investigates whether banks adjust their loan pricing
policies in response to changes in the cost they pay to access the bond market,
controlling for our set of frm-specifc characteristics, F. Models 2 expands our controls
to account for our set of loan-specifc variables, L. As we discussed in the methodology
section, we estimate our model with and without loan controls to reduce concerns with
the potential endogeneity of some of these controls. Model 3 investigates what happens
when we further expand our controls to account for our set of bank-specifc variables, B.
Models 4 and 5, in turn, investigate the robustness of our fndings when we control for
the overall level of interest rates at the time of the loan. In model 4, we control for the log
of the spread between triple-B and triple-A primary yields on new bonds issued at the
time of the loan, LBBBBSP D
l
, and in model 5, we further control for the level of Libor at
the time of the loan, LIBOR. These controls are important to assure us that the link we
identify between loan spreads and the bank’s cost of bond fnancing is not driven by
changes in the overall interest rates or in the price of risk at the time of the loan. Finally,
JFEP
6,3
282
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table III.
Shocks to bond markets
and bank loan pricing
policies
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
L
B
B
B
S
P
D
b
0
.
2
9
3
*
*
*
(
0
.
0
1
2
2
)
0
.
3
1
1
*
*
*
(
0
.
0
1
1
5
)
0
.
2
8
5
*
*
*
(
0
.
0
1
3
3
)
0
.
1
6
9
*
*
*
(
0
.
0
2
9
4
)
0
.
1
3
7
*
*
*
(
0
.
0
2
9
5
)
0
.
1
1
2
*
*
*
(
0
.
0
4
1
6
)
L
B
B
B
S
P
D
l
0
.
1
2
3
*
*
*
(
0
.
0
2
8
1
)
0
.
1
2
6
*
*
*
(
0
.
0
2
8
1
)
0
.
1
4
5
*
*
*
(
0
.
0
4
1
5
)
L
I
B
O
R
?
0
.
0
1
7
2
*
*
*
(
0
.
0
0
3
1
7
)
?
0
.
0
2
0
0
*
*
*
(
0
.
0
0
5
6
5
)
L
A
G
E
?
0
.
1
4
5
*
*
*
(
0
.
0
0
7
3
6
)
?
0
.
1
0
5
*
*
*
(
0
.
0
0
6
5
3
)
?
0
.
1
0
3
*
*
*
(
0
.
0
0
6
5
0
)
?
0
.
1
0
3
*
*
*
(
0
.
0
0
6
5
1
)
?
0
.
1
0
4
*
*
*
(
0
.
0
0
6
4
9
)
?
0
.
1
0
5
*
*
*
(
0
.
0
1
7
8
)
L
S
A
L
E
S
?
0
.
2
3
5
*
*
*
(
0
.
0
0
3
6
1
)
?
0
.
1
0
1
*
*
*
(
0
.
0
0
5
5
4
)
?
0
.
1
0
2
*
*
*
(
0
.
0
0
5
6
4
)
?
0
.
1
0
3
*
*
*
(
0
.
0
0
5
6
5
)
?
0
.
1
0
4
*
*
*
(
0
.
0
0
5
6
3
)
?
0
.
1
0
4
*
*
*
(
0
.
0
0
9
5
0
)
L
E
V
E
R
A
G
E
0
.
3
9
5
*
*
*
(
0
.
0
4
2
5
)
0
.
3
2
1
*
*
*
(
0
.
0
3
3
6
)
0
.
3
2
1
*
*
*
(
0
.
0
3
3
5
)
0
.
3
2
2
*
*
*
(
0
.
0
3
3
3
)
0
.
3
2
2
*
*
*
(
0
.
0
3
3
0
)
0
.
3
1
6
*
*
*
(
0
.
0
5
1
7
)
T
A
N
G
I
B
L
E
S
?
0
.
0
7
4
6
*
*
*
(
0
.
0
1
8
3
)
?
0
.
0
2
7
2
*
(
0
.
0
1
5
7
)
?
0
.
0
2
5
3
(
0
.
0
1
5
6
)
?
0
.
0
2
5
1
(
0
.
0
1
5
6
)
?
0
.
0
2
7
6
*
(
0
.
0
1
5
6
)
?
0
.
0
1
6
2
(
0
.
0
1
9
6
)
R
O
A
V
O
L
0
.
0
9
9
8
*
*
*
(
0
.
0
2
4
1
)
0
.
0
9
3
5
*
*
*
(
0
.
0
2
0
8
)
0
.
0
8
5
7
*
*
*
(
0
.
0
2
0
6
)
0
.
0
8
6
6
*
*
*
(
0
.
0
2
0
6
)
0
.
0
8
3
3
*
*
*
(
0
.
0
2
0
4
)
0
.
0
8
1
8
*
*
*
(
0
.
0
2
3
3
)
R
D
?
0
.
2
0
1
*
*
*
(
0
.
0
5
7
4
)
?
0
.
1
3
7
*
*
*
(
0
.
0
4
1
4
)
?
0
.
1
3
4
*
*
*
(
0
.
0
4
1
6
)
?
0
.
1
3
3
*
*
*
(
0
.
0
4
1
6
)
?
0
.
1
3
1
*
*
*
(
0
.
0
4
1
3
)
?
0
.
1
2
6
*
*
(
0
.
0
5
6
1
)
A
D
V
E
R
T
I
S
I
N
G
?
0
.
6
5
4
*
*
*
(
0
.
2
4
2
)
?
0
.
6
3
5
*
*
*
(
0
.
1
9
1
)
?
0
.
6
5
6
*
*
*
(
0
.
1
9
4
)
?
0
.
6
5
5
*
*
*
(
0
.
1
9
4
)
?
0
.
6
6
3
*
*
*
(
0
.
1
9
5
)
?
0
.
7
3
4
*
*
*
(
0
.
2
3
7
)
L
I
N
T
C
O
V
?
0
.
2
0
5
*
*
*
(
0
.
0
1
1
9
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
6
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
1
)
?
0
.
1
6
2
*
*
*
(
0
.
0
2
0
4
)
M
K
T
B
O
O
K
?
0
.
2
0
5
*
*
*
(
0
.
0
1
1
9
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
6
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
1
)
?
0
.
1
6
2
*
*
*
(
0
.
0
2
0
4
)
P
R
O
F
M
A
R
G
I
N
?
0
.
0
3
0
9
(
0
.
0
2
2
3
)
?
0
.
0
3
0
9
*
(
0
.
0
1
6
0
)
?
0
.
0
3
0
1
*
(
0
.
0
1
6
2
)
?
0
.
0
2
9
8
*
(
0
.
0
1
6
1
)
?
0
.
0
2
8
8
*
(
0
.
0
1
6
0
)
?
0
.
0
2
9
3
(
0
.
0
2
0
5
)
N
W
C
0
.
0
7
2
2
(
0
.
0
5
6
3
)
0
.
0
6
6
1
(
0
.
0
5
7
1
)
0
.
0
6
4
4
(
0
.
0
5
6
7
)
0
.
0
6
2
3
(
0
.
0
5
7
5
)
0
.
0
6
5
7
(
0
.
0
5
8
4
)
0
.
0
6
5
7
(
0
.
0
5
1
9
)
R
E
L
A
T
I
O
N
S
H
I
P
?
0
.
0
6
4
3
*
*
*
(
0
.
0
1
1
2
)
?
0
.
0
2
6
7
*
*
*
(
0
.
0
0
9
8
7
)
?
0
.
0
2
5
3
*
*
(
0
.
0
0
9
9
2
)
?
0
.
0
2
3
9
*
*
(
0
.
0
0
9
9
1
)
?
0
.
0
2
3
9
*
*
(
0
.
0
0
9
9
0
)
?
0
.
0
2
3
2
(
0
.
0
2
1
4
)
T
R
E
N
D
0
.
0
2
5
9
*
*
*
(
0
.
0
0
1
3
0
)
0
.
0
2
1
3
*
*
*
(
0
.
0
0
1
4
9
)
0
.
0
2
5
7
*
*
*
(
0
.
0
0
2
0
0
)
0
.
0
2
5
6
*
*
*
(
0
.
0
0
1
9
9
)
0
.
0
2
2
6
*
*
*
(
0
.
0
0
2
0
8
)
0
.
0
3
0
9
*
*
*
(
0
.
0
0
7
5
6
)
L
A
M
O
U
N
T
?
0
.
1
0
8
*
*
*
(
0
.
0
0
5
7
6
)
?
0
.
1
0
8
*
*
*
(
0
.
0
0
5
7
6
)
?
0
.
1
0
8
*
*
*
(
0
.
0
0
5
7
7
)
?
0
.
1
0
6
*
*
*
(
0
.
0
0
5
7
5
)
?
0
.
1
0
7
*
*
*
(
0
.
0
0
6
7
1
)
L
M
A
T
U
R
I
T
Y
0
.
0
9
5
3
*
*
*
(
0
.
0
0
7
8
6
)
0
.
0
9
2
1
*
*
*
(
0
.
0
0
7
8
6
)
0
.
0
9
4
4
*
*
*
(
0
.
0
0
7
8
6
)
0
.
0
9
5
5
*
*
*
(
0
.
0
0
7
8
9
)
0
.
0
9
6
4
*
*
*
(
0
.
0
1
7
1
)
S
E
C
U
R
E
D
0
.
5
6
0
*
*
*
(
0
.
0
1
2
8
)
0
.
5
5
4
*
*
*
(
0
.
0
1
2
7
)
0
.
5
5
4
*
*
*
(
0
.
0
1
2
7
)
0
.
5
5
3
*
*
*
(
0
.
0
1
2
7
)
0
.
5
4
1
*
*
*
(
0
.
0
3
7
6
)
C
O
R
P
P
U
R
P
O
S
E
S
?
0
.
0
1
2
3
(
0
.
0
1
1
7
)
?
0
.
0
1
9
3
*
(
0
.
0
1
1
7
)
?
0
.
0
1
7
4
(
0
.
0
1
1
7
)
?
0
.
0
2
3
0
*
(
0
.
0
1
1
7
)
?
0
.
0
2
3
9
(
0
.
0
2
0
6
)
R
E
F
I
N
A
N
C
E
?
0
.
0
7
4
0
*
*
*
(
0
.
0
1
1
6
)
?
0
.
0
7
7
1
*
*
*
(
0
.
0
1
1
6
)
?
0
.
0
7
7
4
*
*
*
(
0
.
0
1
1
6
)
?
0
.
0
8
1
5
*
*
*
(
0
.
0
1
1
5
)
?
0
.
0
7
9
6
*
*
*
(
0
.
0
1
5
1
)
W
O
R
K
C
A
P
I
T
A
L
?
0
.
0
6
3
6
*
*
*
(
0
.
0
1
3
0
)
?
0
.
0
7
2
6
*
*
*
(
0
.
0
1
3
1
)
?
0
.
0
7
1
0
*
*
*
(
0
.
0
1
3
1
)
?
0
.
0
8
0
2
*
*
*
(
0
.
0
1
3
2
)
?
0
.
0
8
0
4
*
*
*
(
0
.
0
2
1
8
)
T
E
R
M
L
O
A
N
?
0
.
4
0
1
*
*
*
(
0
.
0
3
7
1
)
?
0
.
3
9
4
*
*
*
(
0
.
0
3
6
9
)
?
0
.
3
9
6
*
*
*
(
0
.
0
3
6
8
)
?
0
.
4
0
3
*
*
*
(
0
.
0
3
6
9
)
?
0
.
4
1
5
*
*
*
(
0
.
0
5
9
9
)
C
R
E
D
I
T
L
I
N
E
?
0
.
4
8
1
*
*
*
(
0
.
0
3
7
4
)
?
0
.
4
7
3
*
*
*
(
0
.
0
3
7
2
)
?
0
.
4
7
4
*
*
*
(
0
.
0
3
7
1
)
?
0
.
4
8
0
*
*
*
(
0
.
0
3
7
2
)
?
0
.
4
9
4
*
*
*
(
0
.
0
7
3
0
)
D
I
V
I
D
E
N
D
R
E
S
T
0
.
1
2
1
*
*
*
(
0
.
0
1
1
1
)
0
.
1
3
4
*
*
*
(
0
.
0
1
1
3
)
0
.
1
3
4
*
*
*
(
0
.
0
1
1
3
)
0
.
1
3
1
*
*
*
(
0
.
0
1
1
3
)
0
.
1
6
0
*
*
*
(
0
.
0
3
4
4
)
(
c
o
n
t
i
n
u
e
d
)
283
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table III.
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
S
E
N
I
O
R
?
0
.
1
0
2
*
*
*
(
0
.
0
2
4
9
)
?
0
.
1
1
7
*
*
*
(
0
.
0
2
4
9
)
?
0
.
1
2
6
*
*
*
(
0
.
0
2
5
0
)
?
0
.
1
0
7
*
*
*
(
0
.
0
2
5
3
)
?
0
.
0
9
9
3
*
*
(
0
.
0
4
0
6
)
L
A
S
S
E
T
S
0
.
0
1
2
7
*
*
(
0
.
0
0
5
6
8
)
0
.
0
1
4
0
*
*
(
0
.
0
0
5
6
6
)
0
.
0
1
6
7
*
*
*
(
0
.
0
0
5
7
2
)
?
0
.
0
7
1
7
(
0
.
0
5
4
3
)
R
O
A
4
.
7
2
6
(
6
.
2
9
0
)
5
.
1
3
7
(
6
.
2
9
4
)
4
.
3
7
3
(
6
.
2
7
9
)
7
.
2
4
2
(
1
1
.
5
0
)
C
H
A
R
G
E
O
F
F
S
2
0
.
4
5
*
*
*
(
6
.
1
4
4
)
2
0
.
5
5
*
*
*
(
6
.
1
2
9
)
1
7
.
9
1
*
*
*
(
6
.
1
4
2
)
?
1
.
2
9
8
(
1
0
.
2
2
)
R
O
A
V
O
L
?
1
.
2
8
8
(
4
.
6
9
7
)
?
3
.
7
0
4
(
4
.
7
3
6
)
?
0
.
5
8
3
(
4
.
7
4
8
)
4
.
8
8
1
(
1
1
.
6
9
)
L
I
Q
U
I
D
I
T
Y
?
0
.
2
1
3
*
*
*
(
0
.
0
8
0
0
)
?
0
.
1
8
0
*
*
(
0
.
0
8
0
2
)
?
0
.
2
1
5
*
*
*
(
0
.
0
8
0
1
)
?
0
.
2
7
3
(
0
.
2
3
5
)
C
A
P
I
T
A
L
?
0
.
0
3
0
0
*
*
*
(
0
.
0
0
4
0
5
)
?
0
.
0
2
8
3
*
*
*
(
0
.
0
0
4
0
6
)
?
0
.
0
2
7
3
*
*
*
(
0
.
0
0
4
0
5
)
?
0
.
0
1
3
9
(
0
.
0
1
1
8
)
D
E
P
O
S
I
T
S
0
.
2
4
7
*
*
*
(
0
.
0
4
6
4
)
0
.
2
4
4
*
*
*
(
0
.
0
4
6
4
)
0
.
2
5
5
*
*
*
(
0
.
0
4
6
3
)
0
.
6
6
7
*
*
(
0
.
2
6
6
)
S
U
B
D
E
B
T
?
3
.
0
4
5
*
*
*
(
0
.
7
1
1
)
?
3
.
1
8
4
*
*
*
(
0
.
7
1
1
)
?
3
.
4
1
3
*
*
*
(
0
.
7
1
0
)
?
3
.
3
8
3
*
(
1
.
8
4
3
)
C
O
N
S
T
A
N
T
6
.
5
9
5
*
*
*
(
0
.
0
8
2
6
)
6
.
3
6
4
*
*
*
(
0
.
0
8
6
5
)
6
.
2
9
2
*
*
*
(
0
.
1
4
2
)
6
.
2
6
3
*
*
*
(
0
.
1
4
2
)
6
.
3
1
2
*
*
*
(
0
.
1
4
1
)
7
.
6
9
6
*
*
*
(
0
.
8
5
0
)
B
k
f
x
e
d
e
f
f
e
c
t
s
N
O
N
O
N
O
N
O
N
O
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
R
2
A
d
j
u
s
t
e
d
0
.
4
8
3
0
.
5
9
7
0
.
6
0
0
0
.
6
0
0
0
.
6
0
1
0
.
5
6
0
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
w
h
i
c
h
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
r
e
m
a
i
n
i
n
g
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
R
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
JFEP
6,3
284
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
model 6 re-estimates our most comprehensive model (model 5) with bank fxed effects to
reduce concerns with sample selection.
Model 1 shows that banks that rely on bond fnancing take into account the cost they
incur to raise funding in the bond market when they decide on their loan spreads. The
coeffcient on our proxy for the cost the bank pays to issue in the bond market, the
triple-B over the triple-A yield spread at the time of the bank’s most recent bond issue
(prior to the loan), LBBBSP D
b
, is highly statistically signifcant and equal to 0.29. A
1-per cent increase in the cost banks pays to issue in the bond market leads to an increase
of 29 basis points in the loans spreads they charge their borrowers.
With regards to the effects of frm controls we use in model 1, they are generally
consistent with our discussion in the methodology section. As expected, older and larger
frms, as well as frms with more interest coverage, and tangibles pay lower spreads on
their loans. Firms that have a relationship with their banks are also able to borrow at
lower interest rates. Firms with higher levels of leverage and more volatile earnings pay
higher spreads on their loans. Contrary to expectations, although, frms with more R&D
expenses, as well as those with more advertising expenses (relative to their sales), are
able to borrow at lower interest rates.
Models 2 and 3 show that the link we fnd in model 1 between loan spreads and the
cost the bank pays to issue in the bond market continues to hold when we add our loan
controls and bank controls, respectively. Adding these controls does not alter the
statistical signifcance, and it has only a minor effect on the size of the coeffcient on
LBBBSP D
b
. With respect to our loan controls, their effects are generally consistent with
our intuition. Larger loans and senior loans pay lower interest rates. In contrast, longer
maturity loans carry higher interest rates. Similarly, secured loans and loans that give
rise to dividend constraints carry higher spreads. Even though these covenants aim at
protecting lenders, they are more often present in loans to riskier borrowers, thereby
explaining why these loans carry higher spreads. Term loans and credit lines carry
lower spreads. Refnance loans and loans for working capital, on the other hand, pay
higher spreads.
With regards to bank controls, our results confrmthat banks that incur larger losses
charge higher spreads on their loans. Banks with higher capital-to-asset ratios, those
with more liquidity, as well as banks that rely more heavily on subdebt charge lower
spreads. In contrast, banks that depend more heavily on deposit funding, usually
smaller banks, tend to charge higher spreads on their loans. Finally, bank size, as
measured by assets, has a positive sign, which may be contrary to expectations, but the
evidence on scale economies in banking is mixed.
The results of models 1-3 suggest that when the cost banks pay to raise funding in the
bond market goes up, banks respond by increasing the spreads they charge on their
corporate loans. A potential concern with this fnding is that the increase in the loan
spread is not actually driven by the cost banks pay to issue in the bond market but rather
arises as a result of an overall increase in the cost of funding. To alleviate this concern,
in model 4, we expand our set of controls to account for the cost to raise funding in the
bond market at the time of the loans, LBBBSP D
l
. In model 5, we expand our controls
further to include the Libor also measured at the time of the loan, LIBOR. Adding these
controls cuts in half the estimated loan spread elasticity with respect to the cost the bank
pays to raise funding in the bond market, suggesting that some of the effect in models
1-3 is indeed driven by an overall increase in the cost of funds. However, the positive and
285
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
statistically signifcant loan spread effect of the bank’s cost to issue in the bond market
remains. In other words, we fnd that, conditional on overall cost of funding at the time
of the loan, when banks incur higher costs to issue in the bond market, they tend to
charge higher loan spreads on their corporate loans.
Finally, in model 6, we show that this fnding continues to hold when we estimate
our model of loan spreads with bank fxed effects. Adding bank fxed effects further
reduces our loan spread elasticity to 0.11, but does not affect its statistical
signifcance. In the remainder of the paper, we refer to model 6 of Table III as our
benchmark specifcation.
3.1 Identifying the loan spread effect of banks’ bond fnancing costs
As we pointed out above, one concern with our fndings is that they may refect an
overall increase in the cost of credit rather than the cost banks pay to issue in the bond
market. Even though the cost of borrowing in the bond market and the loan spreads are
measured at different times, there is a high correlation, 0.93, between the bond spread at
the time of the bank’s bond issue and the bond spread at the time of the loan.
In this section, we attempt to reduce this concern by presenting the results of four
tests we developed for this purpose. The frst test investigates whether our result
changes over time. If our result is driven by changes in the overall cost of credit in the
economy, then it should be independent fromthe sample period we consider. The second
test is a falsifcation test which uses, as a control group, the banks that do not use bond
fnancing but extended loans at the same time as bond-fnancing banks. The next two
tests use two alternative measures of the cost of bond issue by bank – actual spread on
the bank’s most recent bond issue, which limits substantially our sample, but has only
0.21 correlation with the bond spread at the time of the loan and binary variables to
indicate whether the last bond prior to each loan was issued at a time of very high or very
low spreads in the bond market.
3.1.1 Loan spread effect of banks’ bond fnancing costs over time. As we noted earlier,
our frst test builds on the idea that if the link unveiled between loan spreads and the
bank’s cost of bond funding were driven by the overall cost of credit in the economy at
the time of the loan, then this effect should hold throughout the sample period. In
contrast, if that link is indeed driven by the cost banks pay to raise funding in the bond
market, then we should fnd stronger evidence of it in the most recent portion of the
sample period, as the importance of bond fnancing for banks has grown over the years.
To test this, we estimate our model separately for the frst half of our sample period
(1988-1997) and for the second half of the sample period (1998-2007). The results of these
tests are reported in Table IV. We use in this test, as well as in all the subsequent
robustness tests, our most general model, model 6 of Table III, which includes all the
control variables and estimated with bank fxed effects. In the interest of space, we do
not report the coeffcients on the control variables. Models 1 and 2 test for the effect of the
cost in the bond market at the time the bank issued its most recent bond prior to the loan,
LBBBSP D
b
. As we can see from these models, we fnd evidence of an effect of the cost
banks pay to raise funding in the bond market on their loan spreads in the second half of
the sample period (model 2), but not in the frst half of the sample period (model 1).
Models 3 and 4 expand the set of controls in the previous models to account for the
overall cost of credit in the economy at the time of the loan as measured by the cost to
issue in the bond market at the time of the loan, LBBBSP D
l
. This variable is only
JFEP
6,3
286
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table IV.
Bank cost of bond
fnancing and loan
spreads over time
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
S
a
m
p
l
e
p
e
r
i
o
d
1
9
8
8
-
9
7
1
9
9
8
-
0
7
1
9
8
8
-
9
7
1
9
9
8
-
0
7
1
9
8
8
-
9
7
1
9
9
8
-
0
7
L
B
B
B
S
P
D
b
0
.
0
1
4
1
(
0
.
0
5
5
4
)
0
.
2
2
1
*
*
*
(
0
.
0
3
7
0
)
?
0
.
0
2
1
3
(
0
.
0
6
9
7
)
0
.
0
8
1
4
*
*
(
0
.
0
3
6
4
)
?
0
.
0
4
6
4
(
0
.
0
6
4
4
)
0
.
0
7
5
2
*
*
(
0
.
0
3
3
6
)
L
B
B
B
S
P
D
l
0
.
0
5
4
2
(
0
.
0
6
1
1
)
0
.
1
6
7
*
*
*
(
0
.
0
4
2
4
)
0
.
2
0
3
*
*
(
0
.
0
8
0
0
)
0
.
1
7
5
*
*
(
0
.
0
8
4
8
)
D
E
P
O
S
I
T
S
I
N
S
U
R
E
D
?
L
B
B
B
S
P
D
b
?
0
.
4
6
2
*
*
(
0
.
1
8
2
)
?
0
.
0
3
6
6
(
0
.
2
7
7
)
D
E
P
O
S
I
T
S
I
N
S
U
R
E
D
0
.
3
2
6
(
0
.
3
5
9
)
0
.
5
7
8
(
0
.
3
8
5
)
D
E
P
O
S
I
T
S
0
.
2
7
6
(
0
.
3
5
1
)
0
.
6
3
9
*
(
0
.
3
5
2
)
0
.
2
4
9
(
0
.
3
5
5
)
0
.
6
3
9
*
(
0
.
3
4
5
)
L
I
B
O
R
?
0
.
0
3
3
7
*
*
*
(
0
.
0
1
0
7
)
?
0
.
0
2
4
7
*
*
*
(
0
.
0
0
6
7
9
)
?
0
.
0
3
4
2
*
*
*
(
0
.
0
1
0
7
)
?
0
.
0
2
3
5
*
*
*
(
0
.
0
0
6
6
8
)
?
0
.
0
3
1
9
*
*
*
(
0
.
0
1
1
3
)
?
0
.
0
2
5
3
*
*
*
(
0
.
0
0
6
5
1
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
4
,
5
4
6
1
1
,
5
2
1
4
,
5
4
6
1
1
,
5
2
1
4
,
5
2
4
1
1
,
5
2
1
R
2
A
d
j
u
s
t
e
d
0
.
5
7
4
0
.
5
7
0
0
.
5
7
4
0
.
5
7
1
0
.
5
7
6
0
.
5
7
1
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
D
E
P
O
S
I
T
S
I
N
S
U
R
E
D
,
b
a
n
k
’
s
i
n
s
u
r
e
d
d
e
p
o
s
i
t
s
o
v
e
r
a
s
s
e
t
s
.
D
E
P
O
S
I
T
S
,
b
a
n
k
’
s
t
o
t
a
l
d
e
p
o
s
i
t
s
o
v
e
r
a
s
s
e
t
s
.
A
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
287
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
signifcant in the second half of the sample, and while adding it reduces the size of the
coeffcient on LBBBSP D
b
, we continue to have the striking difference between the two
parts of our sample period. This difference suggests that the link we fnd between the
cost banks pay to raise funding in the bond market and the spreads they charge on their
corporate loans cannot be entirely driven by changes in the overall cost of credit at the
time of the loan.
The statistical signifcance of LBBBSP D
l
in the second half of the sample period
(models 3 and 4) may appear surprising in light of Berlin and Mester’s (1999) fnding that
banks shield their borrowers fromchanges in the overall cost of credit. Recall that Berlin
and Mester showthat banks that use relatively more core deposits offer more protection
to their borrowers from shocks to the current cost of credit which they proxy by
LBBBSP D
l
. To get closer to their specifcation, we modify the previous models and
control for the bank’s use of insured deposits (scaled by its assets) and the interaction of
this variable withLBBBSPD
l
[21]. The results of this exercise, whichare reportedinmodels
5 and 6 of Table IV, showthat the effect of LBBBSP D
l
is positive and signifcant in the two
parts of our sample period, while the effect of DEPOSITS INSURED ? LBBBSP D
l
is
negative, but signifcant only in the frst part. In other words, the fnding that Berlin and
Mester uncovered over the period of 1977-1989 that banks with more core deposits offered
more protection to their borrowers fromshocks to the current cost of credit, persisted in the
decade that followed their study (model 5), but weakened afterwards (model 6). In contrast,
the effect of our proxyfor the cost the bankpaidwhenit issuedthe last bondprior tothe loan,
LBBBSPD
b
, is onlysignifcant inthe secondhalf of the sample period. Thus, the differential
effect we haduncoveredfor LBBBSPD
b
duringour sample periodcontinues to holdinthese
models, addingsupport to our assertionthat the effect of the bank’s cost of bondfnancingis
not drivenbychanges inthe overall cost of credit at the time of the loan. The fact that we fnd
evidence of the link only in the most recent part of our sample was to be expected, as the
importance of bond fnancing for banks grew over time.
3.1.2 Falsifcation test. So far, we focused our attention on banks that rely on bond
fnancing, meaning that they issued at least once in the bond market in the three years
prior to the loan. We also have data on loans extended by banks that do not rely on this
source of funding, meaning that they issued bonds only more than three years prior to
the loan or have not issued any bonds since 1970 (this is the frst year we have
information on banks’ bond issuance. Recall that our sample of loans starts in 1987).
If our earlier fnding is driven by an increase in the overall cost of credit, then we
should detect a similar effect on the loan spreads of the latter set of loans. If, on the other
hand, our fnding is driven by the cost banks pay to issue in the bond market, then the
spreads on the latter loans should be unaffected by bond market conditions prior to the
loan. Thus, we construct the following matched sample. For each of the loans that had a
bond issued by the bank within the past three years, we identify loans that were
extended on the same day but did not have a bond issued within three years prior. We
fnd such a match for 6,659 of our loans extended by banks that have accessed the bond
market (treatment group), with 2,896 loans extended by banks that have not accessed
the bond market (control group). For the loans issued by the treatment group banks, we
continue using LBBBSP D
b
as the measure of the cost of bank funding through the bond
market, measured on the day the last bond prior to any given loan was issued. For the
control group we do not have that measure, by construction, because they did not issue
a bond. Thus, for the control group we construct a synthetic measure, which is equal to
JFEP
6,3
288
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
the average LBBBSP D
b
of the treatment loans issued on the same day. We re-estimate
our main model for this matched sample, the total of 9,555 loans, including the
interaction of the treatment indicator, that the banks accessed bond market, BK
ACCESS, with LBBBSPD
b
. We expect the main effect (coeffcient on LBBBSPD
b
, which
corresponds to the effect on the control group) to be zero, while the coeffcient on the
interaction term (which, when summed up with the coeffcient on LBBBSP D
b
,
corresponds to the effect on the treatment group) to be positive.
The results of this test are reported in Table V. As in the previous test, we use our
most comprehensive model of loan spreads, model 6 of Table III, and estimate the
models with all controls and bank fxed effects. The results of Table V vividly
demonstrate the importance of controlling for the cost of borrowing at the time of the
loans, LBBBSP D
l
, and LIBOR. In the frst two columns, the frst regression does not
control for either measure of the borrowing cost and the second does not control for
Libor. In these regressions, we fnd that the effect of the cost of the access to the bond
market is positive and statistically signifcant for the control group, which is indicative
of the spurious correlation – through correlation of the cost of borrowing over time. As
the third regressions show, once we properly control for the cost of borrowing at the time
of the loan, past cost of borrowing no longer has an effect on the pricing of the loans by
banks that did not access the bond market. In all three regressions, we fnd solid
evidence of the effect of the bond market conditions at the time of the banks’ bond issue
on loan spreads they charge their corporate borrowers – the coeffcient on the interaction
term is positive and statistically signifcant. The sum of the main effect and the
coeffcient on the interaction term is equal to 0.18 and is highly statistically signifcant
(F-statistic is 17.6), indicating that for the treatment group of loans, we continue to
observe the effect of cost of borrowing on the bond market.
Table V.
Falsifcation test
a
Variables (1) (2) (3)
LBBBSP D
b
0.237*** (0.0384) 0.146** (0.0598) 0.103 (0.0640)
BK ACCESS 0.0686* (0.0382) 0.0699* (0.0388) 0.0534 (0.0389)
BK ACCESS ?LBBBSP D
b
0.0844* (0.0462) 0.0885* (0.0463) 0.0776* (0.0457)
LBBBSP D
l
0.0922** (0.0393) 0.115*** (0.0413)
LIBOR ?0.0211*** (0.00775)
Bk fxed effects YES YES YES
Observations 9,555 9,555 9,555
R
2
Adjusted 0.499 0.499 0.501
Notes:
a
Dependent variable is LLOAN SP D which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; LBBBSP D
b
, natural log of the difference between the Moody’s indexes on
the ex ante yields of triple-B- and triple-A-rated bonds at the time of the bank’s most recent bond issue
prior to the loan; LBBBSP D
I
, natural log of the difference between the Moody’s indexes on the ex ante
yields of triple-B- and triple-A-rated bonds at the time of the loan; LIBOR, three-month-level Libor at the
time of the loan. BK ACCESS, dummy variable equal to one for the banks in the sample that issued
bonds; sample limited to the set of banks that issued bonds during the sample period and those
banks that did not issue, but match the former banks on observable characteristics; all models
include the controls used in model 6 of Table III; see defnitions of controls in Table I; robust
standard errors clustered by bank in parentheses; *signifcant at 10 per cent; **signifcant at 5 per
cent; ***signifcant at 1 per cent
289
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
There remains a possibility that borrowers that are more risky in a way that we cannot
observe or control for pushbanks toissue debt duringthe times whenthe costs of borrowing
are high. If this is the case, we will observe more expensive loans made by these banks not
because they are passing on the costs of their bond issues but because this subset of
borrowers is riskier. While we cannot test the differences in unobserved characteristics[22],
we can, nevertheless, see that loans extended by banks whose last public bond was issued
duringthe periodof highborrowingcosts are extendedto borrowers that have similar share
of tangible assets, interest coverage ratio, market-to-bookratio andnet workingcapital ratio
comparedtoborrowers of the banks that either didnot issue at all (fromour matchedsample)
or issuedinthe periodof lowborrowingcosts. Moreover, borrowers of the banks that issued
their previous public bond during high borrowing costs period, have higher proft margin
than other borrowers[23].
3.1.3 Using actual cost of bond fnancing. The results we reported thus far assume
that the cost of bond fnancing for banks is correlated with the spread between Moody’s
indexes of ex ante yields of triple-B- and triple-A-rated bonds on the date of the most recent
bondthe bankissuedprior tothe loan. As we explainedabove, we chose torelyonthese yield
indexes because primary yields are missing for many of the bonds issued by banks in our
sample. The spread between these indexes is likely a good proxy for the cost of bond
fnancingfor our banks because theyare all ratedinvestment grade. Nonetheless, ideallyone
would like to control for a measure of the cost of bond fnancing which is specifc to each
bond issue to account for idiosyncrasies across banks and over time. For this reason, we
re-estimated our model of loan spreads, but this time using the log of the yield on the most
recent bond the bank issued prior to the loan over the one-year Treasury on that date, LBK
BONDYIELDSP D, to measure the cost of bond fnancing for the bank. When we use this
measure, our sample of loans drops from 16,067 to 3,595.
The results of this test are reported in Table VI. Model 1 tests whether the bank’s cost
of bond fnancing, as determined by the yield it paid on its most recent bond issue prior
to the loan, affects the spreads on its corporate loans. Model 2 accounts for the cost to
access the bond market at the time of the loan. As before, we measure this cost by the log
of the spread between Moody’s indexes of ex ante yields of triple-B- and triple-A-rated
Table VI.
Controlling for the yields
on bank bonds
a
Variables (1) (2) (3)
LBK BON D Y IELD 0.0680*** (0.00944) 0.0569*** (0.00882) 0.0480*** (0.0120)
LBBBSP D
l
0.177*** (0.0497) 0.168*** (0.0513)
LIBOR ?0.0127 (0.0100)
Bk fxed effects YES YES YES
Observations 3,629 3,629 3,629
R
2
Adjusted 0.569 0.574 0.575
Notes:
a
Dependent variable is LLOAN SP D which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; LBBBSP D
b
, natural log of the difference between the Moody’s indexes on
the ex ante yields of triple-B- and triple-A-rated bonds at the time of the bank’s most recent bond issue
prior to the loan; LBBBSP D
l
, natural log of the difference between the Moody’s indexes on the ex ante
yields of triple-B- and triple-A-rated bonds at the time of the loan; LIBOR, three-month-level Libor at the
time of the loan; all models include the controls used in model 6 of Table III; see defnitions of controls
in Table I; robust standard errors clustered by bank in parentheses; *signifcant at 10 per cent;
**signifcant at 5 per cent; ***signifcant at 1 per cent
JFEP
6,3
290
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
bonds at the time of the loan, LBBBSP D
l
. Model 3 further controls for the level of Libor
at the time of the loan, LIBOR. As with the previous test, we use our most comprehensive
model of loan spreads, model 6 of Table III, which accounts for our sets of bank-, frm-
and loan-specifc controls, B, F and L, respectively, as well as bank fxed effects. In the
interest of space, however, we do not report the coeffcients on these control variables.
A quick look at Table VI reveals that this test confrms our earlier fnding that when
a bank pays a higher cost to issue in the bond market, it increases the spreads on its
corporate loans. According to the estimates of model 1, when the ex ante yield spread on
a bank’s bond doubles, the bank increases its spreads on the loans it extend
subsequently by 7 per cent (which, given the mean of loan spread of these banks of 158
basis points, corresponds to an increase of about 11 basis points).
The results of this test should lay to rest any concerns that may exist with our use of
the ex ante yield spread between Moody’s indexes of triple-B- and triple-A-rated bonds
to measure the cost of banks’ bond fnancing. These results are important for yet another
potential concern with our use of this measure of the bank’s cost of bond fnancing: the
correlation between the yield spread measured at the time of the bank’s most recent
bond prior to the loan, LBBBSPD
b
, and this same spread at the time of the loan, LBBBSP
D
l
. In our base model, we account for this correlation by controlling for the triple-B
spread at the time of loan, LBBBSP D
l
. Because the correlation between BK BOND
YIELD and LBBBSP D
l
is much lower than the correlation between LBBBSP D
b
and
LBBBSP D
l
, the result of this robustness test further confrms that the effect of LBBBSP
D
b
on loan spreads is attributable to an increase in the cost of the bank’s bond fnancing
and is not the result of an increase in the overall interest rates around the time of the loan.
3.1.4 Loan spread effect when banks issue during good periods and crises periods.
Another way to isolate the effect of the triple-B ex ante yield spread at the time of the
bank’s most recent bond issue prior to the loan, while controlling for the value of this
spread at the time of the loan, is to focus on periods when there was a crisis a bond
market and/or periods when the cost to issue in the bond was extraordinary low. We can
then test whether banks that issued bonds during these periods charged different
spreads on the loans they extended afterward when compared to the other banks that
also rely on bond fnancing but did not issue during these periods or to these same banks
on the loans not immediately following these periods.
To test this hypothesis, we identify the “crises” in the bond market during our sample
period, defned as extended periods of time when the ex ante triple-B over triple-A yield
spread was above one. This criteria left us with fve bond market crises during the sample
period (1987-2007)[24]. We excluded from our sample all loans taken prior to the frst crisis.
Next, we identifed loans for which the most recent public bond was issued by the lender
duringthe crisis as opposedto other loans for whichthe most recent public bondwas issued
by the lender during periods between crises[25]. This allows us to test whether loans that
followed banks’ bond issues during crisis times carried higher spreads when compared to
the loans extended after bank’s bond issues occurred during non-crisis times.
The results of this test are reported in the top panel of Table VII. Models in this panel
are similar to those in Table VI, but now the key variable is BOND CRISIS, a dummy
variable that is equal to one if the bank issued the last bond prior to each loan during the
period of high triple-B spread. As in previous tables, we do not report coeffcients on all
the control variables to save on space. Model 1 investigates whether banks that issued
bonds in the period where the spreads in the bond market were elevated charged higher
291
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
rates on the loans they extended following these bond issues, controlling for our sets of
frm-, loan- and bank-specifc controls as well as bank fxed effects. Model 2 expands this
set of controls to account for the cost to access the bond market at the time of the loan as
determined by the spread between Moody’s indexes of ex ante yields of triple-B- and
triple-A-rated bonds at the time of the loan, LBBBSP D
l
. Model 3 further expands our set
of controls to account for the level of Libor at the time of the loan, LIBOR.
The results of this test also confrm our earlier fndings. In all the models, loans that
followed banks’ bond issues placed during the periods of tight bond market conditions
carried higher spreads than loans that followed bonds issued during a tranquil period.
As before, controlling for the conditions in the bond market at the time of the loan (model
2) and additionally for the level of Libor at the time of the loan (model 3) reduces the
loan-spread elasticity vis-a`-vis our measure of the bank’s cost of funding.
Next, we expand the previous test to investigate whether banks pass onto their
borrowers any savings they enjoy when they issue bonds in periods of unusually low
cost to issue in the bond market. We defne these periods as periods during which the ex
ante triple-B over triple-A yield spread was in the lowest 25 per cent of its
distribution[26]. We follow the approach used in the previous test and defne a dummy
variable, BOND GOOD T IM ES, to isolate the loans that followed banks’ issues of
bonds during these “good times”[27]. We add this variable to the preceding regressions.
Table VII.
Controlling for bonds
banks issue in good and
crises times in the bond
market
a
Variables (1) (2) (3)
Bonds banks issue during crises times in the bond market
BON D CRISIS 0.188*** (0.0321) 0.0413* (0.0234) 0.0587** (0.0244)
LBBBSP D
l
0.228*** (0.0397) 0.169*** (0.0345)
LIBOR ?0.0293*** (0.00651)
Bk fxed effects YES YES YES
Observations 15,200 15,200 15,200
R
2
Adjusted 0.562 0.566 0.568
Bonds banks issue during good and crises times in the bond market
BD CRISIS 0.137*** (0.0296) 0.0446** (0.0222) 0.0603** (0.0238)
BD GOOD T IM ES ?0.128*** (0.0168) ?0.0771*** (0.0210) ?0.0623*** (0.0213)
LBBBSP D
l
0.175*** (0.0442) 0.130*** (0.0387)
LIBOR ?0.0276*** (0.00708)
Bk fxed effects YES YES YES
Observations 15,200 15,200 15,200
R
2
Adjusted 0.565 0.567 0.568
Notes:
a
Dependent variable is LLOAN SP D, which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; BON D CRISIS, dummy variable equal to one if the bank issued its most
recent bond during a crisis period in the bond market; see footnote (24) for the crises in the bond market
during the sample period; BDGOODTIMES, dummy variable equal to one if the bank issued its most
recent bond during a period of extraordinarily low rates in the bond market; see footnote (26) for the
good time periods in the bond market during the sample period. LBBBSPD
l
natural log of the difference
between the Moody’s indexes on the ex ante yields of triple-B- and triple-A-rated bonds at the time of the
loan; LIBOR, three-month-level Libor at the time of the loan; all models include the controls used in
model 6 of Table III; see defnitions of controls in Table I; robust standard errors clustered by bank in
parentheses; *signifcant at 10 per cent; **signifcant at 5 per cent; ***signifcant at 1 per cent
JFEP
6,3
292
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
The results of this test, which are reported in the bottompanel of Table VII, showthat
there is symmetry in the bond market effect on loan spreads. When banks fund
themselves in the bond market at a high cost, they pass a portion of this cost onto their
borrowers; when they are able to fund themselves in the bond market at a very lowcost,
they pass a portion of their “savings” onto their borrowers as well. The coeffcient on
BON D CRISIS is slightly higher than the coeffcient on BON D GOOD T IM ES in
models 1 and 2, and the opposite in model 3, but the difference between these coeffcients
in all models is not statistically signifcant. Once again, these results show that the
effects of the cost banks pay to raise funding in the bond market continue to hold when
we control for the conditions in the bond market at the time of the loan (model 2) and
additionally for the level of Libor at the time of the loan (model 3), further confrming
that they are driven by the cost banks pay to fund themselves in the bond market.
3.2 Other robustness tests
We undertake some additional robustness tests to make sure our results are not driven
by omitted variables. As in the previous section, we use model 6 of Table III, which is our
comprehensive model estimated with bank fxed effects, to do these tests. We do not
report them in the interest of space, but they are available upon request. All of them
confrm that our results continue to hold when we include additional controls.
Despite the large set of factors we account for with our controls, a concern with our
fndings is that we do not account for a potentially important determinant of a bank’s
cost of funds – the cost it incurs to raise deposit funding. We do not consider this cost
because there is no bank-level information on the interest rates banks pay on deposits.
We tried to alleviate these concerns by controlling for the three-month Libor at the time
of the loan, which is the most commonly used proxy for banks’ cost of funding. However,
if this cost is strongly correlated with our proxy for the banks’ cost to issue in the bond
market, this could explain our fndings. To investigate this possibility, we expand our
set of bank controls to account for the cost a bank incurs to raise deposit funding by
interacting with the ratio of deposits to assets with the three-month Libor. This control
variable does not enter the regression signifcantly and does not affect our results.
A more accurate proxy for the cost of deposit funding is the interest expenses on
deposits reported by each bank, but this variable is missing in the Call Reports for 35 per
cent of the observations in our sample. Nonetheless, we used this variable to create two
alternative proxies for the cost of deposit funding. In one case we complemented the
interest expenses on deposits reported by banks with the above proxy; in the other case,
we complemented that variable with total interest expense reported by banks in the Call
Reports[28]. In both cases, our results remain unchanged.
The absence of controls for loan securitization and loan sales could also be a source of
concerns, as banks could use these activities to manage their funding sources. Controlling
properly for loan sales and securitization activity by banks is not an easy task because there
is no bank-level informationonthese activities for most of the sample period[29]. We triedto
account for the effect of loan sales and securitization by controlling for the bank’s
outstanding balance of commercial and industrial (C&I) loans sold and securities (scaled by
the bank’s assets). This variable is never statistically signifcant, probably because it is
available for less thanhalf of the observations inour sample, as Call Reports beganincluding
information on banks’ securitization activity only in 2001:Q3. Controlling for this variable
doesnot affect our fndingontheeffect thebank’scost of bondfnancinghasonloanspreads.
293
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
We also tried to account for these activities by using the Call Report variable “loans held for
sale”. This variable goes back to 1991, but it does not contain separate information on C&I
loans, and it reports information on the loans the bank intends to sell, not on the loans it
effectively sells. Again, controlling for this variable (scaled by the bank’s assets) does not
affect our key fnding. Finally, as securitization activities are relatively more important for
the large banks, we test whether these banks drive our key fnding. Dropping the largest
three banks from the sample does not affect our key fnding, even though these banks
account for about 23 per cent of loans in our sample.
We have attributed the change in banks’ loan pricing policies when the spreads in the
bond market at the time of banks’ bond issuance go up to changes in the cost of banks’
bond fnancing. Could that change in banks’ loan pricing policies instead be driven by an
overall increase in the “price” of risk? The results of our falsifcation test suggest this is
not the case. In addition, our models control for the three-month Libor rate and the
triple-Bbond spread at the time of the loan, which tend to vary with the overall economic
conditions. To further reduce concerns with this hypothesis, we add the gross domestic
product growth rate and the slope of the Treasury yield curve, one at a time, as
additional proxies for the state of the economy and for potential changes in the overall
risk premium. Again, we fnd that these additional controls do not enter signifcantly
and do not affect our results.
Standard errors in our models are clustered by bank. Because many frms took multiple
loans throughout the sample period, the error term in our regression could be correlated
across loans not just for a given bank but also for a given frm. To address this issue, we
follow Petersen (2009) and rerun our core regressions with clustering by frm as well as by
bank[30]. The results of this test showonlyanegligible increase (less thanone per cent) inthe
standard errors, suggesting that clustering by bank only is, in fact, appropriate.
Yet another concern with our fndings is whether they could be driven by unobserved
heterogeneity across borrowers that is correlated with their lenders’ access to the bond
market. For instance, when there is a shock to the supply of bank loans, bank borrowers
with access to the bond market may increase their use of bond fnancing. Similarly,
when there are shocks to the bond market, borrowers with access to this source of
funding could increase their use of bank funding, possibly crowding out lending that
banks would otherwise extend to bank-dependent borrowers. To reduce concerns that
heterogeneity across borrowers drives our key fndings, we re-estimate our loan pricing
model with frm fxed effects as well as with bank-frm pair fxed effects. Our key
fndings remain unchanged.
3.3 How large are the costs to borrowers?
Now that we established that there is a statistically signifcant effect of the cost banks
pay to issue on the bond market on the interest rates they charge on their corporate
loans, we want to assess the economic signifcance of this effect.
Using the results of our benchmark model, model 6 in Table III, we can see that the
elasticity of loan spreads with respect to the bond spread at a time of the bank’s last bond
issue is 0.11. According to the estimates of this model, when the triple-B spread in the
bond market doubles, banks that issue bonds at that time increase spreads on the loans
they extend subsequently by 11 per cent[31]. This change, given the average loan spread
of these banks of 158 basis points, corresponds to an increase of about 18 basis points.
JFEP
6,3
294
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
To get a better intuition for the magnitude of this effect, consider a median-size loan
among those extended by banks that use bond fnancing, a facility of 175 million dollars.
According to the estimates of model 6 in Table III, a frm that borrows from a bank that
issued a bond during a crisis in the bond market, if that frmtakes out a loan in the three
years that follow the bank’s last bond issue, it will pay on average ?$300,000 more per
year than a frm that borrows from a bank that also uses bond fnancing but did not
issue a bond during the crisis. These effects indicate that the link we identifed between
the cost banks pay to issue in the bond market and the interest rates they charge
corporate borrowers is not only statistically signifcant but also economically
meaningful. While the magnitude of this effect is not very large, one has to keep in mind
that all borrowers in our sample are publicly traded. There are many reasons to believe,
including our fndings in the subsequent section of this paper, that these effects would
be larger for smaller private frms.
In sum, the fndings we reported thus far show that banks pass on to their corporate
borrowers a portion of the cost they incur to issue in the bond market. Consequently,
shocks to the cost of banks’ access to the bond market transmit to the cost of bank
lending. According to our estimates, the effect of these shocks on loan spreads is
economically signifcant.
In the next section, we investigate whether all borrowers are equally exposed to these
shocks. We are particularly interested in fnding out whether banks shield their
relationship borrowers from these shocks or whether they instead build on their
informational advantage over these borrowers to pass on to them the bulk of the cost
increase they face to raise funding in the bond market.
4. Do banks pass bond market shocks to all of their borrowers?
The tests we reported in the previous section showthat banks adjust their loan prices in
response to changes in the cost they pay to issue in the bond market. Those tests,
however, do not distinguish between different categories of borrowers. In particular,
they do not distinguish borrowers that have a lending relationship with their bank from
those that do not have such a relationship. As we explained in the introduction, this is
important because banks’ expectation of future business with their relationship
borrowers may lead them to shield these borrowers from the shocks to their funding
costs that arise with their use of the bond market. Alternatively, as banks are likely to
have an information advantage over relationship borrowers, it will be easier for them to
pass any shocks to their funding costs onto relationships borrowers.
To test which of these two effects dominates, we estimate our model (2) of loan
spreads, which extends model (1) to include the interaction of our relationship variable
with our proxy for the cost of banks’ bond fnancing, RELATIONSHIP ?LBBBSP D
b
.
The results of these tests are reported in Table VIII. Model 1 controls for frm-, loan- and
bank-specifc variables, as well as bank fxed effects. We do not report coeffcients on
these controls to save space. Model 2 expands this set of controls to account for the cost
to access the bond market at the time of the loan, LBBBSP D
l
, and model 3 accounts for
the level of Libor at the time of the loan, LIBOR.
We continue to fnd that banks adjust their loan pricing policies in response to changes
in the cost they incur to issue in the bond market. In all three models, the coeffcient on
LBBBSP D
b
, our measure of the bank’s cost to issue in the bond market, is positive and
highly statistically signifcant, indicating that banks increase the loan spreads on their
295
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
borrowersthat donot havearelationshipwiththemwhenthebanks’ cost toissueinthebond
market goes up. The coeffcient on RELAT ION SHIP ?LBBBSP D
b
is always negative,
suggesting that banks may shield their relationship borrowers fromshocks to their funding
costs. This coeffcient, however, is never statistically different fromzero. In other words, our
results show that banks may shield their relationship borrowers (when compared to
nonpartisanshipborrowers) fromshocks to their fundingcosts, but byanamount that is not
statistically different from zero.
4.1 Relationship lending and bank dependency
The evidence we just presented indicates that we cannot reject the hypothesis that
banks do not give “special” treatment to their relationship borrowers vis-a`-vis their
nonpartisanship borrowers when they need to raise loan spreads to make up for the
additional cost they incur when it becomes more expensive to raise bond fnancing. Our
results, however, also showthat banks do not pass the entirety of the shocks to their cost
of bond fnancing onto their borrowers. This combination of results poses an interesting
question, as banks do not offer special treatment to their relationship borrowers, why
they do not use the informational advantage they are likely to have over these borrowers
to pass onto to them the bulk of the shock to their cost of bond fnancing?
This apparent puzzle could arise because, among the relationship borrowers, some
are dependent on banks for funding while others are not. Banks’ may have more
incentives to shield relationship borrowers that are dependent on them for external
funding more than relationship borrowers that have access to other sources of funding,
such as the bond market. This is because banks are more likely to recover the “subsidy”
from future business with the former borrowers. On the other hand, banks will fnd it
easier to pass on the additional cost to they face in the bond market onto the
bank-dependent relationship borrowers because relationship borrowers that do not
Table VIII.
Relationship borrowers
and bank bond costs
a
Variables (1) (2) (3)
LBBBSP D
b
0.317*** (0.0315) 0.186*** (0.0425) 0.144*** (0.0459)
RELAT ION SHIP ?0.0300 (0.0258) ?0.0289 (0.0258) ?0.0302 (0.0257)
RELAT ION SHIP ?LBBBSP D
b
?0.0526 (0.0440) ?0.0556 (0.0431) ?0.0520 (0.0433)
LBBBSP D
l
0.142*** (0.0423) 0.146*** (0.0411)
LIBOR
?0.0198***
(0.00572)
Bk fxed effects YES YES YES
Observations 16067 16067 16067
R
2
Adjusted 0.559 0.560 0.561
Notes:
a
Dependent variable is LLOAN SP D, which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; LBBBSP D
b
, natural log of the difference between the Moody’s indexes on
the ex ante yields of triple-B- and triple-A-rated bonds at the time of the bank’s most recent bond issue
prior to the loan; LBBBSP D
l
, natural log of the difference between the Moody’s indexes on the ex ante
yields of triple-B- and triple-A-rated bonds at the time of the loan; LIBOR, three-month-level Libor at the
time of the loan. RELAT ION SHIP refers to three-year horizon in columns (1)-(3) and one-year horizon
in columns (4)-(6); all models include the controls used in model 6 of Table III; see defnitions of controls
in Table I; robust standard errors clustered by bank in parentheses; *signifcant at 10 per cent;
**signifcant at 5 per cent; ***signifcant at 1 per cent
JFEP
6,3
296
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
depend on their bank for funding may respond to any attempt of an increase in their loan
rates by looking for funding elsewhere.
To test which of these predictions is borne out in the data, we estimate our loan
pricing model separately for relationship borrowers and borrowers with no lending
relationship with their bank. Further, we modify this model to distinguish whether the
borrower is bank dependent or not, and interact this dummy variable with our proxy for
the cost the bank incurred the last time it issue in the bond market prior to the loan. We
identify borrowers as bank dependent if they never issued in the public bond market or
issued a bond more than three years prior to the loan[32]. In addition, as according to
Rajan (1992), the holdup problemis more pronounced for risky frms than for safe frms,
we also consider a specifcation in which we distinguish among the frms that have
access to the bond market those that are rated investment grade from those that are
rated below investment grade.
The results of these tests for the loans of relationship borrowers are reported in
Table IX. Model 1 investigates the effect of LBBBSP D
b
on loan spreads controlling for
our sets of frm-, loan- and bank-specifc factors, as well as bank fxed effects. Model 2
adds to this model the cost to access the bond market at the time of the loan, LBBBSSP
D
l
, while model 3 further adds to our controls the LIBOR at the time of the loan. Models
4 through 6, in turn, investigate whether relationship borrowers that have access to the
bond market are less exposed to their banks’ cost of bond fnancing. As with the
previous set of models, model 4 investigates this hypothesis controlling for our sets of
frm-, loan- and bank-specifc factors; model 5 adds to these controls the cost to access
the bond market at the time of the loan, and model 6 further adds the Libor at the time of
the loan. Finally, models 7 through 9 follow the same pattern, but distinguish among
relationship borrowers that have access to the bond market those that are rated
investment grade from those that are rated below grade. For completeness, Table X
reports the results of these same tests but for borrowers that do not have a lending
relationship with their bank. In the interest of space, we report in these two tables only
coeffcients on variables that are key to our hypotheses, leaving out the coeffcients on
our sets of frm, loan and bank controls.
Focusing on Table IX, we see that models 1-3 confrm the fndings we reported in
Table VIII – when the bank’s cost to issue in the bond market goes up, the bank passes
a portion of this cost hike onto the borrowers it has a lending relationship with. Models
4-6 show that not all relationship borrowers are exposed to the cost of the bank’s bond
fnancing. Among loans of relationship borrowers, model 4 shows the loan-spread
elasticity to the cost of the bank’s bond fnancing is 0.30 for loans of bank-dependent
borrowers and only 0.21 for loans of borrowers with access to the bond market. These
elasticities decline as we expand our controls to account for the cost to access the bond
market at the time of the loan (model 5) and to account for the level of Libor at the time
of the loan (model 6). According to model 6, the aforementioned elasticities are 0.14 and
0.04, respectively. Importantly, the former elasticity is statistically different from zero,
but we cannot reject the hypothesis that the latter elasticity is equal to zero, according to
the F-test (p-value is 0.47). In other words, when banks’ cost of bond fnancing goes up,
they passes a portion of this cost increase onto their relationship borrowers that do not
have access to the bond market. On these occasions, they also raise the spreads on their
loans to those relationship borrowers that have access to the bond market, but by an
amount that is smaller and is not statistically different from zero.
297
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table IX.
Bank cost of bond
fnancing and loan
spreads the bank charges
relationship borrowers
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
(
7
)
(
8
)
(
9
)
$
L
B
B
B
S
P
D
_
b
$
0
.
2
6
4
*
*
*
(
0
.
0
5
0
9
)
0
.
1
4
9
*
*
(
0
.
0
6
3
2
)
0
.
1
0
8
*
(
0
.
0
6
3
0
)
0
.
3
0
3
*
*
*
(
0
.
0
4
9
0
)
0
.
1
8
3
*
*
*
(
0
.
0
6
3
1
)
0
.
1
4
0
*
*
(
0
.
0
6
3
9
)
0
.
2
9
6
*
*
*
(
0
.
0
5
0
1
)
0
.
1
7
0
*
*
(
0
.
0
6
5
0
)
0
.
1
3
1
*
*
(
0
.
0
6
5
9
)
$
P
B
O
N
D
$
?
0
.
0
8
8
5
*
*
*
(
0
.
0
1
6
2
)
?
0
.
0
8
9
3
*
*
*
(
0
.
0
1
6
2
)
?
0
.
0
9
4
0
*
*
*
(
0
.
0
1
6
6
)
$
P
B
O
N
D
$
?
$
L
B
B
B
S
P
D
_
b
$
?
0
.
0
9
0
9
*
*
*
(
0
.
0
2
5
1
)
?
0
.
0
9
2
4
*
*
*
(
0
.
0
2
5
2
)
?
0
.
0
9
7
5
*
*
*
(
0
.
0
2
4
6
)
$
P
B
O
N
D
\
;
I
G
$
?
0
.
2
1
8
*
*
*
(
0
.
0
2
1
8
)
?
0
.
2
1
9
*
*
*
(
0
.
0
2
1
8
)
?
0
.
2
2
2
*
*
*
(
0
.
0
2
1
3
)
$
P
B
O
N
D
\
;
B
G
$
0
.
2
0
1
*
*
*
(
0
.
0
2
7
7
)
0
.
2
0
1
*
*
*
(
0
.
0
2
7
8
)
0
.
1
9
4
*
*
*
(
0
.
0
2
7
3
)
$
P
B
O
N
D
\
;
I
G
$
?
$
L
B
B
B
S
O
D
_
b
$
?
0
.
0
8
4
5
*
*
*
(
0
.
0
2
3
2
)
?
0
.
0
8
7
9
*
*
*
(
0
.
0
2
3
2
)
?
0
.
0
9
2
5
*
*
*
(
0
.
0
2
2
6
)
$
P
B
O
N
D
\
;
B
G
$
?
$
L
B
B
B
S
O
D
_
b
$
0
.
0
3
3
1
(
0
.
0
6
6
7
)
0
.
0
3
6
9
(
0
.
0
6
7
7
)
0
.
0
3
0
6
(
0
.
0
6
6
7
)
$
L
B
B
B
S
P
D
_
l
$
0
.
1
2
1
*
*
(
0
.
0
5
2
2
)
0
.
1
2
3
*
*
(
0
.
0
4
8
6
)
0
.
1
2
6
*
*
(
0
.
0
5
0
7
)
0
.
1
2
9
*
*
*
(
0
.
0
4
6
6
)
0
.
1
3
3
*
*
*
(
0
.
0
5
0
0
)
0
.
1
3
6
*
*
*
(
0
.
0
4
6
5
)
L
I
B
O
R
?
0
.
0
2
2
6
*
*
*
(
0
.
0
0
5
5
2
)
?
0
.
0
2
4
4
*
*
*
(
0
.
0
0
5
3
5
)
?
0
.
0
2
2
1
*
*
*
(
0
.
0
0
5
3
6
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
R
$
^
2
$
A
d
j
u
s
t
e
d
0
.
5
7
8
0
.
5
7
8
0
.
5
8
0
0
.
5
8
0
0
.
5
8
0
0
.
5
8
2
0
.
5
9
4
0
.
5
9
5
0
.
5
9
6
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
P
B
O
N
D
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
I
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
B
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
b
e
l
o
w
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
E
L
A
T
I
O
N
S
H
I
P
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
;
S
a
m
p
l
e
l
i
m
i
t
e
d
t
o
b
o
r
r
o
w
e
r
s
w
i
t
h
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
i
r
b
a
n
k
(
R
E
L
A
T
I
O
N
S
H
I
P
?
1
.
)
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
JFEP
6,3
298
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table X.
Bank cost of bond
fnancing and loan
spreads the bank charges
non-relationship
borrowers
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
(
7
)
(
8
)
(
9
)
L
B
B
B
S
P
D
b
0
.
3
1
4
*
*
*
(
0
.
0
3
4
9
)
0
.
1
6
0
*
*
(
0
.
0
6
7
9
)
0
.
1
2
0
*
(
0
.
0
7
3
4
)
0
.
3
1
2
*
*
*
(
0
.
0
4
2
7
)
0
.
1
6
0
*
*
(
0
.
0
6
6
8
)
0
.
1
1
9
(
0
.
0
7
2
2
)
0
.
3
1
3
*
*
*
(
0
.
0
4
2
9
)
0
.
1
4
1
*
*
(
0
.
0
6
2
2
)
0
.
1
0
3
(
0
.
0
6
7
2
)
P
B
O
N
D
?
0
.
0
6
1
8
*
*
(
0
.
0
2
6
9
)
?
0
.
0
6
1
7
*
*
(
0
.
0
2
7
0
)
?
0
.
0
6
6
8
*
*
(
0
.
0
2
7
9
)
P
B
O
N
D
?
L
B
B
B
S
P
D
b
0
.
0
1
1
8
(
0
.
0
4
8
9
)
0
.
0
0
9
7
0
(
0
.
0
4
9
8
)
0
.
0
0
1
1
7
(
0
.
0
5
0
3
)
P
B
O
N
D
I
G
?
0
.
1
8
9
*
*
*
(
0
.
0
3
0
9
)
?
0
.
1
9
0
*
*
*
(
0
.
0
3
0
9
)
?
0
.
1
9
4
*
*
*
(
0
.
0
3
1
7
)
P
B
O
N
D
B
G
0
.
1
9
9
*
*
*
(
0
.
0
3
1
3
)
0
.
2
0
3
*
*
*
(
0
.
0
3
2
2
)
0
.
1
9
7
*
*
*
(
0
.
0
3
2
8
)
P
B
O
N
D
I
G
?
L
B
B
B
S
O
D
b
0
.
0
2
5
9
(
0
.
0
5
8
5
)
0
.
0
2
3
3
(
0
.
0
5
9
8
)
0
.
0
1
5
0
(
0
.
0
6
0
9
)
P
B
O
N
D
B
G
?
L
B
B
B
S
O
D
b
0
.
0
9
4
2
(
0
.
0
7
4
3
)
0
.
0
9
3
7
(
0
.
0
7
5
1
)
0
.
0
8
5
8
(
0
.
0
7
4
9
)
L
B
B
B
S
P
D
l
0
.
1
6
7
*
*
(
0
.
0
7
1
8
)
0
.
1
7
4
*
*
(
0
.
0
7
2
9
)
0
.
1
6
6
*
*
(
0
.
0
7
2
1
)
0
.
1
7
4
*
*
(
0
.
0
7
3
4
)
0
.
1
8
6
*
*
*
(
0
.
0
6
6
9
)
0
.
1
9
4
*
*
*
(
0
.
0
6
8
0
)
L
I
B
O
R
?
0
.
0
1
7
3
*
*
*
(
0
.
0
0
5
0
0
)
?
0
.
0
1
8
6
*
*
*
(
0
.
0
0
5
0
6
)
?
0
.
0
1
7
3
*
*
*
(
0
.
0
0
5
0
1
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
R
2
A
d
j
u
s
t
e
d
0
.
5
3
4
0
.
5
3
5
0
.
5
3
6
0
.
5
3
5
0
.
5
3
6
0
.
5
3
6
0
.
5
4
7
0
.
5
4
8
0
.
5
4
9
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
P
B
O
N
D
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
I
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
B
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
b
e
l
o
w
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
E
L
A
T
I
O
N
S
H
I
P
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
;
S
a
m
p
l
e
l
i
m
i
t
e
d
t
o
b
o
r
r
o
w
e
r
s
w
i
t
h
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
i
r
b
a
n
k
(
R
E
L
A
T
I
O
N
S
H
I
P
?
0
.
)
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
299
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
In models 7-9, we further split relationship borrowers with access to the bond market
into those that are rated investment grade and those that are rated below grade
(according to the rating of their most recent bond issue prior to the loan). Consistent with
the insights of Rajan (1992) we fnd that risky borrowers that have access to the bond
market are more exposed to the cost of banks’ bond fnancing than safe borrowers that
have access to the bond market. Among relationship borrowers, those that have access
to the bond market but are risky, pay a premium on their loans which is similar to that
paid by borrowers that do not have access to the bond market in response to a shock to
their bank’s cost of bond fnancing. This could be because on these occasions, risky
borrowers lose their access to the bond market and, in essence, become dependent on
banks for external funds. In contrast, relationship borrowers that have access to the
bond market and are safe, pay a lower premium than relationship borrowers that are
dependent on banks for funding. In fact, in the case of models 8 and 9, we cannot reject
the hypothesis that safe borrowers do not pay any premium.
Turning our attention to Table X, which reports the same tests but for the sample of
borrowers that do not have a lending relationship with their bank, we see one important
difference with respect to the results we just discussed for relationship borrowers.
Among the non-relationship borrowers, we do not fnd any evidence that banks charge
different premiums associated with their cost of bond fnancing to those borrowers that
do not have access to the bond market and those that have access to the bond market,
regardless of the rating of their bond. This is reassuring, as it shows that it is not critical
for non-relationship borrowers to have access to the bond market, possibly because
banks do not have an informational advantage over them.
Although not as apparent, because of the way we organize our results, there is one other
piece of evidence which suggests that access to the bond market is more important for
relationship borrowers than non-relationship borrowers. If one compares the coeffcient
on LBBBSPD
b
for benchmark, bank-dependent, category of relationship borrowers (line
1 column 9 of Table IX) and of non-relationship borrowers (line 1 column 9 of Table X),
we can see that the former are more sensitive than the latter borrowers. True, this
difference is not very large and is not statistically signifcant, but it points in the
direction that among relationship borrowers, those that do not have access to the bond
market, are more exposed to shocks to banks’ cost of bond funding than
non-relationships borrowers that also do not have access to the bond market.
The results of Tables IX and X suggest that banks pass the largest portion of shocks
to their cost of bond fnancing onto their relationships borrowers that do not have access
to the bond market, and they “protect” the most their relationship borrowers that have
access to the bond market, particularly those that are safe. These fndings are important
in that they do not support the hypothesis that banks take into account the prospects of
future business with their relationship borrowers and smooth the interest rates they
charge them over time, as, in this case, we would expect relationship borrowers that are
bank dependent to receive the largest level of protection. In contrast, our fndings
suggest that the market power resulting from the informational advantage that banks
have over their borrowers drives their loan pricing policies.
The next two tests aim at further confrming these fndings. The frst test furthers
our investigation of how banks pass the shocks to their cost of bond fnancing onto
different categories of borrowers. We report the results of this test in Table XI. The
left-hand columns compare the portion of the shocks to the cost of bond fnancing
JFEP
6,3
300
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table XI.
Bank cost of bond
fnancing and loan
spreads of relationship
borrowers
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
L
B
B
B
S
P
D
b
0
.
3
0
7
*
*
*
(
0
.
0
3
5
6
)
0
.
1
7
1
*
*
*
(
0
.
0
3
6
9
)
0
.
1
2
7
*
*
*
(
0
.
0
4
1
8
)
0
.
2
6
6
*
*
*
(
0
.
0
3
4
9
)
0
.
1
3
0
*
*
*
(
0
.
0
3
4
6
)
0
.
0
8
6
3
*
*
(
0
.
0
3
7
8
)
R
L
P
B
O
N
D
I
G
?
0
.
2
6
8
*
*
*
(
0
.
0
2
3
1
)
?
0
.
2
6
8
*
*
*
(
0
.
0
2
2
9
)
?
0
.
2
7
1
*
*
*
(
0
.
0
2
3
2
)
R
L
P
B
O
N
D
I
G
?
L
B
B
B
S
P
D
b
?
0
.
0
9
9
1
*
*
*
(
0
.
0
2
8
3
)
?
0
.
1
0
4
*
*
*
(
0
.
0
2
7
6
)
?
0
.
1
0
5
*
*
*
(
0
.
0
2
7
4
)
R
L
N
O
A
C
C
E
S
S
0
.
0
5
5
0
*
*
*
(
0
.
0
0
9
7
3
)
0
.
0
5
6
1
*
*
*
(
0
.
0
0
9
6
0
)
0
.
0
5
7
8
*
*
*
(
0
.
0
0
9
1
5
)
R
L
N
O
A
C
C
E
S
S
?
L
B
B
B
S
P
D
b
0
.
0
4
6
6
*
(
0
.
0
2
5
4
)
0
.
0
4
5
8
*
*
(
0
.
0
2
4
9
)
0
.
0
5
0
7
*
*
(
0
.
0
2
5
0
)
L
B
B
B
S
P
D
l
0
.
1
4
6
*
*
*
(
0
.
0
4
0
9
)
0
.
1
5
2
*
*
*
(
0
.
0
3
9
5
)
0
.
1
4
5
*
*
*
(
0
.
0
4
0
6
)
0
.
1
5
0
*
*
*
(
0
.
0
3
9
7
)
L
I
B
O
R
?
0
.
0
2
1
4
*
*
*
(
0
.
0
0
5
2
4
)
?
0
.
0
2
0
3
*
*
*
(
0
.
0
0
5
7
3
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
R
2
A
d
j
u
s
t
e
d
0
.
5
7
0
0
.
5
7
0
0
.
5
7
2
0
.
5
5
9
0
.
5
6
0
0
.
5
6
1
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
R
L
P
B
O
N
D
I
G
D
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
a
n
d
i
t
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
l
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
L
N
O
A
C
C
E
S
S
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
,
a
n
d
i
t
h
a
s
n
e
v
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
o
r
d
i
d
i
t
o
n
l
y
m
o
r
e
t
h
a
n
t
h
r
e
e
y
e
a
r
s
a
g
o
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
301
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
that banks pass onto their relationship borrowers that have access to the bond
market and are safe vis-a`-vis the portion of the shocks that they pass onto their
remaining borrowers. The right-hand columns, in turn, compare the portion of these
shocks that banks pass onto their relationship borrowers that are dependent on
them for external funding vis-a`-vis the portion of the shocks that they pass onto their
remaining borrowers. Consistent with our earlier fndings, the differences in
elasticity of loan spreads with respect to the cost banks pay to issue in the bond
market (that is, the coeffcients on the interaction terms) are negative and signifcant
in left-hand side models, but positive and signifcant in right-hand side models. In
other words, banks do not pass the shocks to their cost of bond fnancing to their
relationship borrowers that have access to the bond market, but they do it to their
relationship borrowers that depend on them for funding.
According to our results, banks do not pass the entirety of the shock to their cost of
funding onto their relationship borrowers that do not have access to the bond market.
This could be because these borrowers are not fully dependent on banks. Because these
borrowers are publicly listed, they are likely to be less dependent on banks than
privately held borrowers that do not have access to the bond market because there will
be even less information available on the latter borrowers. We do not include privately
held borrowers in our sample because Compustat, our source of frm-level data, does not
include information for these borrowers.
Our fnal test takes this analysis another step further by comparing the costs and the
savings banks pass onto dependent and non-dependent relationship borrowers when
they issue in periods of crisis and in periods of low spreads in the bond market,
respectively. To that end, we replace our proxy for the cost the bank paid when it issued
its last bond prior to the loan, LBBBSP D
b
, with the two dummy variables we defned in
section 3.1.4 to identify the loans banks extended after issuing bonds in periods of crisis
in the bond market, BOND CRISIS, and those they extended after issuing bonds in
periods of lowspreads in the bond market, BONDGOODTIMES, respectively. Models
1-3 compare how banks pass the costs and the savings from their bond issues on these
two occasions onto their relationship borrowers that have access to the bond market and
are safe vis-a`-vis the remaining borrowers. Models 4-6 repeat this analysis, but, in this
case, we isolate borrowers that have a lending relationship with the bank and do not
have access to the bond market.
The difference inthe treatment that banks offer to these two sets of borrowers is striking.
Bothsets of borrowers enjoythe same savings as everyone else whenbanks lower loanrates
following their bond issues at low cost[33]. In contrast, when banks raise bond fnancing at
a high cost, they do not raise loan rates on their relationship borrowers that are not
dependent on them, but do raise loan rates on their relationship borrowers that are
dependent on them for external funding, by an amount that is statistically different from
zero[34]. In other words, dependent borrowers are more exposed to shocks to banks
funding costs than nondependent borrowers. Further, our evidence shows that the
former borrowers are more exposed to shocks that raise the cost of banks’ bond
fnancing than to shocks that lower the cost of this funding source for banks.
According to model 6 of Table XII, when banks raise loan spreads on their
dependent relationship borrowers following a bond issue during crisis times, they
raise these rates on average by 10 per cent. When banks lower loan spreads
following a bond issue during good times, they lower these rates on average by 5.3
JFEP
6,3
302
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table XII.
Bank cost of bond
fnancing and loan
spreads of relationship
borrowers: Further
analysis
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
B
O
N
D
C
R
I
S
I
S
0
.
1
5
4
*
*
*
(
0
.
0
2
5
1
)
0
.
0
6
2
9
*
*
*
(
0
.
0
1
9
4
)
0
.
0
7
8
4
*
*
*
(
0
.
0
1
9
7
)
0
.
1
1
6
*
*
*
(
0
.
0
2
6
4
)
0
.
0
2
2
3
(
0
.
0
2
5
4
)
0
.
0
3
6
3
(
0
.
0
2
5
4
)
B
O
N
D
G
O
O
D
T
I
M
E
S
?
0
.
1
3
3
*
*
*
(
0
.
0
1
6
0
)
?
0
.
0
8
2
9
*
*
*
(
0
.
0
2
1
3
)
?
0
.
0
6
7
6
*
*
*
(
0
.
0
2
1
9
)
?
0
.
1
3
1
*
*
*
(
0
.
0
2
0
3
)
?
0
.
0
7
9
0
*
*
*
(
0
.
0
2
3
7
)
?
0
.
0
6
4
0
*
*
*
(
0
.
0
2
3
7
)
R
L
P
B
O
N
D
I
G
?
0
.
2
3
5
*
*
*
(
0
.
0
2
6
5
)
?
0
.
2
3
3
*
*
*
(
0
.
0
2
6
1
)
?
0
.
2
3
6
*
*
*
(
0
.
0
2
7
7
)
R
L
P
B
O
N
D
I
G
?
C
R
I
S
I
S
?
0
.
0
5
7
5
*
*
(
0
.
0
2
5
0
)
?
0
.
0
6
1
1
*
*
(
0
.
0
2
5
5
)
?
0
.
0
5
8
4
*
*
(
0
.
0
2
4
1
)
R
L
P
B
O
N
D
I
G
?
G
O
O
D
T
I
M
E
S
0
.
0
0
9
4
3
(
0
.
0
4
2
6
)
0
.
0
1
1
0
(
0
.
0
4
2
8
)
0
.
0
0
9
5
9
(
0
.
0
4
2
2
)
R
L
N
O
A
C
C
E
S
S
0
.
0
1
8
4
(
0
.
0
2
1
8
)
0
.
0
1
9
2
(
0
.
0
2
1
1
)
0
.
0
1
7
8
(
0
.
0
2
2
3
)
R
L
N
O
A
C
C
E
S
S
?
C
R
I
S
I
S
0
.
0
5
9
3
*
(
0
.
0
3
3
6
)
0
.
0
5
9
1
*
(
0
.
0
3
2
5
)
0
.
0
6
3
5
*
(
0
.
0
3
3
6
)
R
L
N
O
A
C
C
E
S
S
?
G
O
O
D
T
I
M
E
S
0
.
0
1
2
9
(
0
.
0
2
2
3
)
0
.
0
1
1
1
(
0
.
0
2
1
9
)
0
.
0
1
1
0
(
0
.
0
2
2
9
)
L
B
B
B
B
l
0
.
1
7
3
*
*
*
(
0
.
0
4
3
1
)
0
.
1
2
6
*
*
*
(
0
.
0
3
7
6
)
0
.
1
7
7
*
*
*
(
0
.
0
4
3
1
)
0
.
1
3
2
*
*
*
(
0
.
0
3
7
0
)
L
I
B
O
R
?
0
.
0
2
8
1
*
*
*
(
0
.
0
0
6
7
8
)
?
0
.
0
2
7
6
*
*
*
(
0
.
0
0
7
1
6
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
R
2
A
d
j
u
s
t
e
d
0
.
5
7
6
0
.
5
7
7
0
.
5
7
9
0
.
5
6
6
0
.
5
6
7
0
.
5
6
9
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
B
O
N
D
C
R
I
S
I
S
a
n
d
B
O
N
D
G
O
O
D
T
I
M
E
S
a
r
e
d
e
f
n
e
d
a
s
i
n
T
a
b
l
e
V
I
I
.
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
R
L
P
B
O
N
D
I
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
,
a
n
d
i
t
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
l
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
L
N
O
A
C
C
E
S
S
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
,
a
n
d
i
t
h
a
s
n
e
v
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
o
r
d
i
d
i
t
o
n
l
y
m
o
r
e
t
h
a
n
t
h
r
e
e
y
e
a
r
s
a
g
o
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
303
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
for relationship dependent[35]. These fndings confrm that banks’ loan pricing
policies are driven by informational advantage more than by the prospects of future
business with borrowers.
In sum, the results we unveiled in this section confrm that banks adjust their loan
rates in response to shocks to the cost they pay on their bond issues. The results also
show that establishing a relationship lending with a bank does not guarantee a special
treatment of the borrower. When banks are able to issue bonds at very low rates, they
pass some of their cost savings to their relationship borrowers regardless of whether
they are dependent on them. However, when banks raise bond fnancing at high rates,
they do not pass any of these costs to the relationship borrowers that are not dependent
on them, but they do so to their relationship borrowers that depend on themfor funding.
These results, therefore, show that the market power resulting from the informational
advantage that banks have over borrowers more than the prospects of future business
drives their loan pricing policies.
4.1.1 Robustness tests. We undertook a set of robustness tests similar to those we
report in the previous section to investigate the robustness of our fndings on the
differential response of banks vis-a`-vis their relationship borrowers and their
non-relationship borrowers. In the interest of space, we do not report these results, but
describe them briefy.
First, we consider two subsamples as before, the frst half of our sample time
period, from 1988 to 1997, and the second half, from 1998 to 2007. For the full set of
borrowers, when we allow the effect of the cost of bond funding by banks on their
loan pricing to vary for relationship and non-relationship borrowers (as in
Table VIII), we fnd that in the early half of the sample, relationship borrowers paid
on average higher interest on their loans, and their loans were more sensitive to the
interest rate the banks paid on their bonds, but the overall effect of bank bond
issuance cost on the loan pricing is not statistically different for these two sets of
borrowers[36]. The results for the late half of our sample are almost identical to
those in Table VIII, indicating that it is the second half of the sample that drives our
main results. When we limit the sample to relationship borrowers, as in Table IX, we
lose the precision of our estimates in both subsamples, but qualitatively the story
remains unchanged. The results for non-relationship borrowers (Table X) are the
sample for the full, early and late sample periods.
Next, we re-estimates the regressions reported in Tables VIII–X with the actual
spread on the last bond issued by the bank prior to any given loan, as we did in Table VI.
Even though using these spreads leaves out a lot of observations, we continue to fnd the
results of Table VIII unchanged with this modifcation. Once we split the sample,
however, as we do in Tables IX and X, we do not fnd statistically signifcant difference
in the response of loan prices to the change in bank’s bond spreads depending on
whether the borrower has access to the bond market and the rating of their most recent
bond. This is not surprising, given that the number of observations falls from 10,212 to
2,336 and from5,855 to 1,293 in Tables IXand X, respectively, with only a small portion
of these observations accounted for by borrowers that had issued an investment grade
bond in the three years prior to the loan (676 loans of relationship borrowers and 241
loans of non-relationship borrowers).
Next, we repeat the analysis with various controls for the amount of the bond issue,
cost of deposits, and the general economic conditions. In this case, all of our results in
JFEP
6,3
304
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Tables VIII-X remain almost identical to the ones in the regressions we reported. The
only exception is, when we control for bank’s outstanding balance of C&I loans sold and
securities, which dramatically limits our sample, we lose statistical signifcance of the
effect of the interaction term of borrower issuing ING bond in the regression for
relationship borrowers (as in Table IX). We recover all of our results, however, when
instead we control for loans sold, which allows us to retain a larger sample. Neither of
these controls enters signifcantly in Table IX regressions, while loan sales enter
signifcantly (with a positive sign) in the regression of Table X, which limits the sample
to non-relationship borrowers. In this case, however, including either of the two controls
does not affect the coeffcients of interest.
Finally, we attempt different specifcations – with clustering standard errors both on
bank and frm as well as with frm fxed effect. All of our results are robust to these
modifcations. Aminor difference is that the main effect in Table Xbecomes statistically
signifcant at the 10 per cent level (but the same in magnitude) when we cluster standard
errors on both bank and frm.
5. Final remarks
Our fndings show that banks’ use of bond fnancing creates a link between the
conditions in the bond market and their loan pricing policy. The evidence we uncovered
on this link further shows that banks do not offer special protection to their relationship
borrowers. On the contrary, banks expose their relationships borrowers that depend on
them for funding the most to the shocks to their cost of bond fnancing. Banks protect
from the bond market shocks only their relationship borrowers that have access to the
bond market and are safe, possibly because they cannot hold themup for higher interest
rates. The fact that banks pass some savings onto the latter borrowers but not to the
former borrowers when they raise funding in the bond market at extraordinarily low
cost adds further support to our conclusion that market power drives banks loan pricing
policies more than relationship aspects of lending.
These fndings are novel and have important implications. Because the number of
banks that rely on bond fnancing continues to grow, our fndings indicate that fnancial
intermediation through banks will become increasingly interlinked with the
intermediation performed through fnancial markets. In addition, corporate borrowers,
in particular those that are dependent on banks for external funding, will become
increasingly exposed to adverse shocks to the bond market. Moreover, a policy push
toward longer-term bank fnancing is likely to further increase banks’ reliance on the
bond market, leading to unintended consequences of increasing the exposure of
bank-dependent borrowers to the bond market shocks.
These fnding suggest some potential ideas for future research. For instance, a
common view in the fnancial architecture literature is that banks and debt markets
operate independently fromeach other[37]. Holmstromand Tirole (1997), Allen and Gale
(2000) and Song and Thakor (2010) develop models in which banks and fnancial
markets complement each other, but none of themconsider the complementarity that we
identify in this paper. Because banks rely increasingly on market funding, including
bond fnancing, commercial paper funding and repo funding, it would be interesting to
investigate the effect of these changes in the funding structure of fnancial
intermediaries on the roles they perform.
305
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Notes
1. Following Boot et al. (1993), who show that banks with low capital are more likely to exploit
borrowers, sacrifcing reputational capital to preserve fnancial capital. One could also
hypothesize that the increase in the cost of bond funding will lead banks to renege on implicit
guarantees they have given their borrowers, including the guarantee not to explore their
informational monopoly, and thereby raise the loan rates on them. For evidence in support of
the hypothesis that banks price their informational monopoly, see Santos and Winton (2008),
Hale and Santos (2009) and Schenone (2010).
2. Publicly listed borrowers that do not have access to the bond market, for example, are not
likely to experience the same level of bank dependency than privately held borrowers that do
not have access to the bond market, but the absence of accounting information on the latter
borrowers precludes us from including them in our investigation.
3. Our paper is also related to the bank lending channel literature, including Kashyap et al.
(1993), Peek and Rosengren (1997), Kashyap and Stein (2000), Paravisini (2008) and Khwaja
and Mian (2008). This literature focuses on the effects of shocks to bank liquidity on the
volume of bank lending. Our focus instead is on the loan pricing effects of shocks to the bond
market. In this regard, our paper is related to Santos (2011) who investigate the effect of bank
losses during the subprime crisis on their loan pricing policies.
4. Our loan data source is Dealscan. This database has some non-syndicated loans, but is only
comprehensive for loans which banks syndicate. Berlin and Mester rely instead on the Survey
of Terms of Bank Lending to Business. This database reports information on every business
loan but only for a stratifed sample of about 340 banks and for the loans banks made on a
particular day (or number of days).
5. Petersen and Rajan (1995) show that provided banks have some monopoly power in the loan
market, they are able to do intertemporal interest rate smoothing to their relationship
borrowers. In their setting, banks are solely funded with deposits. If they used bond fnancing
in addition, it is easy to see that shocks to their cost of bond fnancing would hinder their
ability to smooth interest rates, notwithstanding their monopoly power in the loan market.
See Boot (2000) for a review of the benefts of relationship lending.
6. Recent papers by Acharya et al. (2012) and by Bord and Santos (2013) showthat bank’s access
to liquidity affect the price of the loans and liquidity to corporate borrowers. These results are
complementary to ours.
7. See Allen and Gale (1997, 1999), Bhattacharya and Chiesa (1995), Dewatripont and Maskin
(1995) and Boot and Thakor (1997).
8. We use the logof the loanspread, as opposedto the spreaditself because the logof the loanspread
has distribution which is closer to the normal, and because this allows us to interpret the
coeffcients on the log of other spreads we use on the left-hand side of our model as elasticities. In
anycase, usingspreads insteadof the logof spreads inour models does not affect our keyfndings.
9. Including natural logarithm of this variable, instead of the level, does not change our results.
10. Firms are requiredto report expenses withadvertisingonlywhentheyexceeda certainvalue. For
this reason, this variable is sometimes missing in Compustat. The same is true of expenses with
research and development. In either case, when the variable is missing, we set it equal to zero.
11. For frms with no debt, this variable is set equal to the difference between current assets and
current liabilities.
JFEP
6,3
306
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
12. Using a time horizon of one year to determine if the borrower has a lending relationship with
the bank, yields similar results.
13. For evidence on the endogeneity of loan covenants see Demiroglu and James (2010) and
Murfn (2012).
14. We use the volatility of ROA rather than stock return because a large number of the banks in
the sample do not have publicly traded shares.
15. Our results do not change if instead we control for insured deposits only.
16. Bharath et al. (2011) fnd that the impact of a relationship on spreads is negative; however,
Santos and Winton (2008) fnd that this effect is reversed in recessions, when information
monopolies are likely to be stronger and maintaining relationships is likely to be less
attractive to lenders.
17. We do not count privately placed bonds as a measure of public bond market access. We
believe private placements are very different from public issues, reaching a smaller set of
investors and thus not increasing informed competition as much as a public issue does. As a
practical matter, there is far less information on private placements because the Securities and
Exchange Commission fling rules on public issues do not apply to private issues. This makes
it hard to control for frms’ private placements. This is consistent with earlier work that
considers private placements to be closer to syndicated bank loans than to public bonds.
18. The process we used to link LPC, SDC, and Compustat can be summarized as follows. The
CRSP data was frst used to obtain, through name-matching procedure, CUSIPs for the
companies in LPC for which this information was missing. With a CUSIP, LPC could then be
linked to both SDC and Compustat, which are CUSIP-based data sets. We proceed by using
the PERMCO variable from CRSP to group companies across CUSIPs, as that variable tracks
the same company across CUSIPs and ticker changes.
19. The number of banks adds up to more than the total number because some banks switch from
using bond fnancing to deposit fnancing alone (or vice versa) over the sample period.
20. In the Robustness section, we test whether our results continue to hold when we measure
banks’ cost of bond fnancing by the actual ex ante yields they pay on their new bond issues.
21. Berlin and Mester use core deposits, while we rely on insured deposits, but the two concepts
share the idea that this funding source is less prone to be affected by changes in the overall
cost of credit in the economy.
22. We control for time-invariant unobserved borrower characteristics by including borrower
fxed effects, as discussed below.
23. In the interest of space, we do not report these numbers, but they are available from the
authors upon request.
24. The dates of these “crises” are as follows: August 15, 1990 through March 4, 1992; September 30,
98 through December 9, 1999; April 11, 2000 through November 24, 2003; May 2, 2005 through
May 30, 2005 and November 7, 2007 through end of sample period December 31, 2007).
25. According to this defnition, about 38 percent of loans in our sample were issued following a
bond placed during crisis times.
26. More specifcally, we defne the beginning of the episode when the spread falls below0.62 and
only include episodes during which the spread dipped below 0.6 for at least one period. We
identify “good times” as follows: July 30, 1993 through October 1, 1993; January 26, 1994
307
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
through February 17, 1994; February 8, 1995 through July 30, 1998; January 27, 2006 through
May 30, 2006; and January 18, 2007 through June 20, 2007.
27. According to this defnition, about 40 percent of loans in our sample were issued following a
bond placed during good times.
28. To assure smooth pasting of our proxy into missing observations, we frst regressed interest
expenses on deposits for the observations we had on other proxies and then we constructed
out-of-sample linear predictions based on these regressions.
29. Call Reports have informationonloansales andsecuritization, but it goes backonlyto2001, covers
onlysale andsecuritizationactivities that the bankretains some servicing, credit enhancements or
there is recourse, and is about the stock not the fow of activity in each quarter.
30. We omit bank fxed effects to conduct this test.
31. Historically, during crises in the US bond market the triple-B yield spread has more than
doubled (see footnote [24] for further details).
32. Limiting the defnition of bank dependency to borrowers that never issued in the bond market
yields similar results.
33. Note that the coeffcient on BONDGOODTIMESis negative and signifcant in all models, but
neither RL P BONDIG?BDGOODTIMES nor RL NOACCESS ?BDGOODTIMES have
effects that are statistically different from zero.
34. Note that the coeffcient on BOND CRISIS is positive and signifcant in all models, and while
the effect of RL P BOND IG ?BOND CRISIS is negative and signifcant and the sum of the
two coeffcients is not different from zero, the effect of RL N OACCESS ?BOND CRISIS is
positive and signifcant.
35. In terms of basis points, given that spreads on loans, on average, during times of no crisis and no
good times are 165 basis points (170 basis points on average overall), these would correspond to
16.5 basis points increase following crises and 8 basis points decline following good times.
36. The main effect of bond spreads is negative and not statistically signifcant. The main effect of
relationship borrower indicator is positive and highly statistically signifcant. The effect of the
interaction of these two coeffcients is positive and statistically signifcant, but the total effect of
bond spreads for relationship borrowers is very close to and not statistically different from zero.
37. See Allen and Gale (1997, 1999), Bhattacharya and Chiesa (1995), Dewatripont and Maskin
(1995) and Boot and Thakor (1997).
References
Acharya, V.V., Afonso, G. and Kovner, A. (2012), “How do global banks scramble for liquidity?
Evidence from the asset-backed commercial paper freeze of 2007”, mimeo, available on SSRN.
Allen, F. and Gale, D. (1997), “Financial markets, intermediaries, and intertemporal smoothing”,
Journal of Political Economy, Vol. 105 No. 3, pp. 523-546.
Allen, F. and Gale, D. (1999), “Diversity of opinion and fnancing of new technologies”, Journal of
Financial Intermediation, Vol. 8 Nos 1/2, pp. 68-89.
Allen, F. and Gale, D. (2000), Comparing Financial Systems, The MIT Press, MA.
Berger, A.N. and Udell, G.F. (1990), “Collateral, loan quality, and bank risk”, Journal of Monetary
Economics, Vol. 25 No. 1, pp. 21-42.
Berger, A.N. and Udell, G.F. (1992), “Some evidence on empirical signifcance of credit rationing”,
Journal of Political Economy, Vol. 100 No. 5, pp. 1047-1077.
JFEP
6,3
308
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Berlin, M. and Mester, L. (1999), “Deposits and relationship lending”, Review of Financial Studies,
Vol. 12 No. 3, pp. 579-607.
Bharath, S., Dahiya, S., Saunders, A. and Srinivasan, A. (2011), “Lending relationships and loan
contract terms”, Review of Financial Studies, Vol. 24 No. 4, pp. 1141-1203.
Bhattacharya, S. and Chiesa, G. (1995), “Proprietary information, fnancial inter-mediation, and
research incentives”, Journal of Financial Intermediation, Vol. 4 No. 4, pp. 328-357.
Boot, A.W.A., (2000), “Relationship banking: what do we know?”, Journal of Financial
Intermediation, Vol. 9 No. 1, p. 725.
Boot, A.W.A. and Thakor, A.V. (1997), “Financial system architecture”, Review of Financial
Studies, Vol. 10 No. 3, pp. 693-733.
Boot, A.W.A., Greenbaum, S.I. and Thakor, A.V. (1993), “Reputation and discretion in fnancial
contracting”, American Economic Review, Vol. 83 No. 5, pp. 1165-1183.
Bord, V.M. and Santos, J.A.C. (2013), “Banks liquidity and the cost of liquidity to corporations”,
Journal of Money Credit and Banking, Vol. 46 No. s1, pp. 13-45.
Demiroglu, C. and James, C.M. (2010), “The information content of bank loan covenants”, Review
of Financial Studies, Vol. 23 No. 10, pp. 3700-3737.
Dewatripont, M. and Maskin, E. (1995), “Credit effciency in centralized and decentralized
economies”, Review of Economic Studies, Vol. 62 No. 4, pp. 541-555.
Holmstrom, B. and Tirole, J. (1997), “Financial intermediation, loanable funds, and the real sector”,
Quarterly Journal of Economics, Vol. 112 No. 3, pp. 663-691.
Kashyap, A.K. and Stein, J.C. (2000), “What do a million observations on banks say about the
transmission of monetary policy?”, American Economic Review, Vol. 90 No. 3, pp. 407-428.
Kashyap, A.K., Stein, J.C. and Wilcox, D.W. (1993), “Monetary policy and credit conditions:
evidence from the composition of external fnance”, American Economic Review, Vol. 83
No. 1, pp. 78-98.
Khwaja, A. and Mian, A. (2008), “Tracing the impact of bank liquidity shocks: evidence from an
emerging market”, American Economic Review, Vol. 98 No. 4, pp. 141-342.
Murfn, J. (2012), “The supply-side determinants of loan contract strictness”, The Journal of
Finance, Vol. 67, pp. 1565-1601, doi: 10.1111/j.1540-6261.2012.01767.x
Paravisini, D. (2008), “Local bank fnancial constraints and frm access to external fnance”,
Journal of Finance, Vol. 63 No. 5, pp. 2161-2193.
Peek, J. and Rosengreen, E.S. (1997), “The international transmission of fnancial shocks: the case
of Japan”, American Economic Review, Vol. 87 No. 4, pp. 495-505.
Petersen, M.A. (2009), “Estimating standard errors in fnance panel data sets: comparing
approaches”, Review of Financial Studies, Vol. 22 No. 1, pp. 435-480.
Petersen, M.A. and Rajan, R.G. (1995), “The effect of credit market competition on lending
relationships”, Quarterly Journal of Economics, Vol. 110 No. 2, pp. 407-443.
Rajan, R.G. (1992), “Insiders and outsiders: the choice between informed and arm’s length debt”,
Journal of Finance, Vol. 47 No. 4, pp. 1367-1400.
Santos, J.A.C. (2011), “Bank corporate loan pricing policy following the subprime crisis”, Reviewof
Financial Studies, Vol. 24 No. 6, pp. 1916-1943.
Santos, J.A.C. and Winton, A. (2008), “Bank loans, bonds, and informational monopolies across the
business cycle”, Journal of Finance, Vol. 63 No. 3, pp. 1315-1359.
Schenone, C. (2010), “Lending relationships and information rents: do banks exploit their
information advantages?”, Review of Financial Studies, Vol. 23 No. 3, pp. 1149-1199.
309
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Sharpe, S.A. (1990), “Asymmetric information, bank lending, and implicit contracts: a stylized
model of customer relationships”, Journal of Finance, Vol. 45 No. 4, pp. 1069-1087.
Song, F. and Thakor, A.V. (2010), “Financial system architecture and the co-evolution of banks
and capital markets”, The Economic Journal, 120: 1021–1055. doi: 10.1111/
j.1468-0297.2009.02345.x
Further reading
Hale, G.B. and João A.C. Santos, (2009), “Do banks price their informational monopoly?”, Journal
of Financial Economics, Vol. 93, pp. 185-206.
Corresponding author
Galina Hale can be contacted at: [email protected]
To purchase reprints of this article please e-mail: [email protected]
Or visit our web site for further details: www.emeraldinsight.com/reprints
JFEP
6,3
310
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_694057984.pdf
This paper aims to analyze how banks transmit shocks that hit the debt market to their
borrowers. Recent financial crisis demonstrated that the banking system can be a pathway for shock
transmission.
Journal of Financial Economic Policy
Do banks propagate debt market shocks?
Galina Hale J oão A.C. Santos
Article information:
To cite this document:
Galina Hale J oão A.C. Santos , (2014),"Do banks propagate debt market shocks?", J ournal of Financial
Economic Policy, Vol. 6 Iss 3 pp. 270 - 310
Permanent link to this document:http://dx.doi.org/10.1108/J FEP-03-2014-0023
Downloaded on: 24 January 2016, At: 21:49 (PT)
References: this document contains references to 31 other documents.
To copy this document: [email protected]
The fulltext of this document has been downloaded 119 times since 2014*
Users who downloaded this article also downloaded:
Faten Ben Bouheni, (2014),"Banking regulation and supervision: can it enhance stability in Europe?",
J ournal of Financial Economic Policy, Vol. 6 Iss 3 pp. 244-269http://dx.doi.org/10.1108/J FEP-11-2013-0059
Tong Li, (2014),"Shadow banking in China: expanding scale, evolving structure", J ournal of Financial
Economic Policy, Vol. 6 Iss 3 pp. 198-211http://dx.doi.org/10.1108/J FEP-11-2013-0061
Norbert Gaillard, (2014),"Assessing sovereign risk: the case of rich countries", J ournal of Financial
Economic Policy, Vol. 6 Iss 3 pp. 212-225http://dx.doi.org/10.1108/J FEP-03-2014-0017
Access to this document was granted through an Emerald subscription provided by emerald-srm:115632 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.
*Related content and download information correct at time of download.
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Do banks propagate debt market
shocks?
Galina Hale
Federal Reserve Bank of San Francisco, San Francisco,
California, USA, and
João A.C. Santos
Federal Reserve Bank of New York, New York, NY, USA and Nova School
of Business and Economics, Lisbon, Portugal
Abstract
Purpose – This paper aims to analyze how banks transmit shocks that hit the debt market to their
borrowers. Recent fnancial crisis demonstrated that the banking system can be a pathway for shock
transmission.
Design/methodology/approach – Bank-level panel regressions.
Findings – This paper shows that when banks experience a shock to the cost of their bond fnancing,
they pass a portion of their extra costs or savings to their corporate borrowers. While banks do not offer
special protection from bond market shocks to their relationship borrowers, they also do not treat all of
themequally. Relationship borrowers that are not bank-dependent are the least exposed to bond market
shocks via their bank loans. In contrast, banks pass the highest portion of the increase in their cost of
bond fnancing to their relationship borrowers that rely exclusively on banks for external funding.
Research limitations/implications – These fndings show that banks put more weight on the
informational advantage they have over their relationship borrowers than on the prospects of future
business with these borrowers. They also show a potential side effect of the recent proposals to require
banks to use CoCos or other long-term funding.
Originality/value – The fndings are timely, given the ongoing debates on the proposals to introduce
bail-in programs and proposals to require banks to use CoCos or other long-term funding.
Keywords Banks, Debt, Credit, Bank subordinate debt, Bond spreads, Lending channel, Loan
spreads
Paper type Research paper
1. Introduction
Recent fnancial crisis demonstrated that the banking system can be a pathway for
shock transmission. In this paper, we analyze how banks transmit shocks that hit the
JEL classifcation – E51, G21, G32
The authors thank David Marqués, Evren Damar, Filipa Sá, Julio Rotemberg, Mark Flannery,
and seminar participants at Tilburg University, Paris School of Economics, Federal Reserve Bank
of San Francisco, Bank of Brazil, Bank of Canada workshop on Financial Institution Behaviors
and Regulations, ECB conference “The bank lending channel in the euro area: New models and
empirical analysis,” the 2010 University of Cambridge Conference on Networks, and the 2009
Gersenzee summer workshop for useful comments. The views stated herein are those of the
authors and are not necessarily the views of the Federal Reserve Banks of San Francisco or New
York, or the Federal Reserve System.
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1757-6385.htm
JFEP
6,3
270
Journal of Financial Economic Policy
Vol. 6 No. 3, 2014
pp. 270-310
© Emerald Group Publishing Limited
1757-6385
DOI 10.1108/JFEP-03-2014-0023
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
debt market to their borrowers. Traditionally, banks have funded their business with
deposits. Because deposits are largely inelastic with respect to the interest rate, this
source of funding gave banks an opportunity to shield their corporate borrowers from
economy-wide shocks. Berger and Udell (1992), for example, document that bank loan
rates move in a smoother fashion than the market interest rate, and Berlin and Mester
(1999) fnd that banks with more core deposits smooth loan interest rates in response to
adverse economic shocks.
In recent years, banks have been increasingly relying on the bond market to fnance
their business. At the end of 1988, the ratio of bond fnancing to deposit funding was 3.5
per cent among the top 100 US banks. By the end of 2007, this ratio had gone up to 9 per
cent. This change in banks’ funding choices is important because it is likely to make it
more diffcult for banks to shield corporate borrowers from economy-wide shocks. In
addition, it is likely to create an additional indirect link between the bond market and the
corporate sector, as shocks to the bond market may nowget propagated to the corporate
sector via banks’ loan pricing policies. We test for evidence of this link and identify
which corporate borrowers have become more exposed to bond market shocks as a
result of banks’ growing reliance on bond fnancing.
We consider two hypotheses. Our frst hypothesis links the interest rates banks
charge on their corporate loans to banks’ costs of issuing in the bond market.
Specifcally, we hypothesize that shocks to the costs banks pay to issue in the bond
market are transmitted to the corporate sector via banks’ loan pricing policies. The
alternative to this hypothesis is that banks perfectly shield their corporate borrowers
from the bond market shocks, absorbing all fuctuations in the cost of bond issuance.
Our second hypothesis is about the borrowers who are more likely to be affected by
shocks to banks’ cost of bond fnancing. Following Berlin and Mester (1999), we
hypothesize that the prospects of future business with their relationship borrowers
leads banks to protect these borrowers from shocks to banks’ cost of bond fnancing,
possibly passing these costs onto relationship borrowers over a longer period. Under
this condition, we expect banks to pass a smaller portion of the shocks to their bond
funding costs onto their relationship borrowers than onto their non-relationship
borrowers. An alternative hypothesis is following the hold-up theories of Sharpe (1990)
and Rajan (1992). It states that banks will pass a larger portion of the shocks to their
bond funding costs onto their relationship borrowers than onto their non-relationship
borrowers, as banks are more likely to have an informational advantage over their
relationship borrowers[1].
The effects of bond market shocks on loan rates under our second hypothesis or
under its alternative are likely to be more pronounced for relationship borrowers that do
not have access to the bond market. On the one hand, these borrowers will likely be more
dependent on banks for external funding, increasing banks’ prospects of future business
with them. On the other hand, banks will likely have a bigger informational advantage
over these borrowers, as these borrowers do not beneft fromthe information that comes
with the issues in the bond market. We use these differences to refne our second
hypothesis, by hypothesizing that if banks shield their relationship borrowers from
bond market shocks, then relationship borrowers that depend on themfor funding stand
to be the least affected by shocks to banks’ cost of bond funding. In contrast, if banks
take advantage of their informational monopoly over their borrowers, then these
bank-dependent relationship borrowers stand to be affected the most by such shocks.
271
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
To test our hypotheses, we start by investigating whether banks’ loan pricing varies
with the cost they paid to raise funding in the bond market the last time they issued a
bond prior to any given loan. We next test whether this link is stronger or weaker for the
borrowers that have a relationship with their bank. Finally, we investigate whether this
link is stronger for relationship borrowers that are more dependent on banks because
they do not have access to the bond market. To help disentangle our hypotheses, we
compare how banks’ loan pricing policies vary across borrowers when the bank
experiences a shock that increases its cost of bond fnancing versus when it benefts
from a shock that lowers its cost of bond fnancing.
We fnd, consistent with our frst hypothesis, that when banks experience a shock to
the cost they pay to issue in the bond market, they respond by passing a portion of the
shock onto their corporate borrowers. Importantly, we fnd that on these occasions,
banks do not offer special protection to their relationship borrowers: relationship
borrowers are less exposed to shocks to the cost of their banks’ bond fnancing than
non-relationship borrowers, but the difference is not statistically signifcant. On closer
inspection, we fnd that relationship borrowers that are not dependent on banks are less
exposed to shocks to the cost their banks pay to raise bond fnancing compared to
relationships borrowers that are dependent on banks for funding. Moreover, our results
show that when banks are able to raise bond fnancing at very low cost, they pass a
small portion of the resulting savings to all of their relationship borrowers. In contrast,
when banks experience a shock that substantially increases their cost of bond fnancing,
they pass the bulk of this cost onto their relationship borrowers that are dependent on
them, while fully protecting the relationship borrowers that have access to the bond
market. In other words, dependent borrowers are more exposed to shocks to banks
funding costs than non-dependent borrowers. Further, our evidence shows that
dependent borrowers are more exposed to shocks that raise the cost of banks’ bond
fnancing than to shocks that lower the cost of this funding source for banks.
These fndings do not support the hypothesis that the prospects of future business with
relationship borrowers leads banks to smooth, over time, the interest rates they charge on
loans to these borrowers. In contrast, our fndings are consistent with the idea that banks
take into account the informational advantage they have over their relationship borrowers
when they decide on their loan rates. Even though banks do not pass the entirety of the
shocks to their cost of bond fnancing onto their relationship borrowers that do not have
access to the bond market, possibly because these borrowers are not fully dependent on
them, banks do expose these borrowers the most to these shocks while protecting their
relationship borrowers that do have access to the bond market[2].
Our paper is most closely related to that of Berlin and Mester (1999) and complements
their work in at least three important respects[3]. Berlin and Mester focus on a period
when banks funded themselves almost entirely with deposits, the 1970s and 1980s.
Their key fnding is that banks’ use of core deposits makes it possible for them to shield
borrowers from economy-wide shocks, possibly because these deposits are
interest-inelastic. Our focus is on the two decades that follow their sample period
(1988-2007), and while we fnd evidence similar to theirs in the frst half of our sample
period, we also fnd that the effect of core deposit fnancing weakens in the second half
of our sample period. More importantly, we fnd that after controlling for the role of
deposit funding, banks still adjust their loan pricing policies in response to shocks they
experience when raising funding in the bond market. Consistent with increasing
JFEP
6,3
272
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
importance of bond fnancing for banks, this bond effect is only evident in the second
half of our sample period. It appears, therefore, that the absorbing effect of deposit
funding weakened over time, while the bond-funding effect grew in importance.
Our paper also extends Berlin and Mester’s analysis by investigating how the
propagation of the bond market shocks by banks differs across different sets of
borrowers, depending on whether they have a lending relationship with the bank and
whether they are dependent on the bank for external funding. In doing so, we also
investigate how banks pass the savings or the extra costs they incur when raising
funding in the bond market onto these sets of borrowers.
Finally, like Berlin and Mester, we have detailed information about bank lenders and
their loans[4]. In contrast to them, however, we also have information on the identity of
borrowers, which gives us the opportunity to control for frm-specifc factors known to
explain loan interest rates and to distinguish whether the borrower is a relationship
borrower and whether it is likely to be bank-dependent or not.
Our fndings are timely, given the ongoing debates on the proposals to introduce bail-in
programs and proposals to require banks to use CoCos or other long-term funding. These
debates have focused on the effects these programs would have on banks’ risk-taking
incentives and their contribution to solve problems of fnancial distress and have paid little
attention to the effects these programs might have on corporate borrowers relying on these
banks for funding. Because these proposals will require banks to access the bond market
periodically, our fndings suggest that they will further expose corporate borrowers, in
particular those that are dependent on banks, to the conditions in the bond market.
Our fndings are also important because they showthat as banks increasingly rely on
bond fnancing, they will fnd it more diffcult to promote relationship lending, which
remains a distinctive feature of banks[5]. Finally, our fndings show a new mechanism
that interlinks the fnancial intermediation done through banks with the intermediation
done through the debt market[6]. A common view in the fnancial architecture literature
is that banks and debt markets operate independently from each other[7]. Holmstrom
and Tirole (1997), Allen and Gale (2000) and Song and Thakor (2010) develop models in
which banks and fnancial markets complement each other, but none of them consider
the complementarity that we identify in this paper. In the study by Holmstrom and
Tirole (1997), the complementarity arises because access to bank funding allows some
borrowers to tap debt markets for additional funding. In the study by Allen and Gale
(2000), intermediaries provide individuals with insurance against unforeseen
contingencies in some states of nature, thereby eliminating the need for individuals to
acquire costly information. Analysis in Song and Thakor (2010) is the closest to the
complementarity we identify, but in their setting, banks rely on the equity market, not
the bond market, to raise the equity capital they need for regulatory reasons.
The remainder of our paper is organized as follows. The next section presents our
methodology and data, and characterizes our sample. Section 3 investigates whether the
interest rates banks charge their borrowers vary with the cost banks pay to issue in the
bond market. Section 4 investigates whether banks pass on the shocks to the bond
market equally to all corporate borrowers. This section also investigates whether banks
adjust their loan pricing policies differently when they experience shocks that
substantially increase their cost of bond fnancing as opposed to shocks that allowthem
to raise bond fnancing at a very low cost. Section 5 concludes the paper.
273
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
2. Methodology, data and sample characterization
2.1 Methodology
Our methodology has two parts corresponding to our two main hypotheses. Part I
investigates whether the spreads banks charge on their corporate loans vary with the
cost they pay to issue in the bond market. Part II investigates whether banks pass the
cost they pay to raise bond fnancing to a greater or lesser extent to borrowers with
whom they have a lending relationship.
2.1.1 Loan spreads and cost of banks’ access to the bond market. We start by
investigating whether the spreads banks charge on their corporate loans vary with the
cost they pay to raise funding in the bond market. To this end, we estimate the following
model of loan spreads:
LLOANSPD
b, f, l, t
? c ? ? LBK BOND COST
b
? ? LBBBSPD
t
? ?LIBOR
t
?
?
i?1
I
?
i
B
i, b, t?1
?
?
j?1
J
?
j
F
j, b, t?1
?
?
k?1
k
?
k
L
k, l
? ?
f, t
where, LLOAN SP D
b,f,l,t
is the natural log of the all-in-drawn spread over Libor of loan l
extended by bank b to frmf at date t[8]. According to Dealscan, our source of loan data, the
all-in-drawnspreadis ameasure of the overall cost of the loan, expressedas aspreadover the
benchmark Libor because it takes into account both one-time and recurring fees associated
with the loan; LBKBONDCOST
b
is the natural log of our measure of the cost bank b paid
the last time it issued in the bond market. The coeffcient on this variable, ?, measures the
elasticityof loanspreads withrespect to the cost banks paidto issue inthe bondmarket, and
so we wouldexpect it to be positive. Because bonds are fxedrate securities, we measure this
cost at the time of the bank’s most recent public bond issue (prior to the loan).
We consider the cost to issue in the bond market only if the bank issued bonds within
three years prior to the loan. If the bank issued bonds a long time ago, the cost it incurred
back then is less likely to affect its current loan pricing policy. To isolate the effect of the
bank cost of bond fnancing from the effect of a change in the overall interest rates or in
the overall “price” of risk at the time of the loan, we add the following two controls to our
model of loan spreads. We control for the cost to issue in the bond market at the time of
the loan by including in our model the log of the spread between triple-B and triple-A
primary yields on new bonds issued at the time of the loan, LBBBSP D
t
. Because this
spread tends to be correlated with the overall price of risk, we expect ? ?0. In addition,
we control for the level of interest rates at the time of the loan, byincludingLIBOR
t
inour
model. Because loan spreads are computed over Libor, the two variables tend to move in
opposite directions and so we expect ? ?0.[9]
We complement these controls for the overall interest rates in the economy, with
three sets of bank-, frm- and loan-specifc controls, B, F and L, which we describe next.
We frst discuss the set of frm-specifc variables that we use. A subset of these
variables, which includes LAGE, the log of the frm’s age in years (we compute the frm’s
age by subtracting the date the frm frst appeared in Compustat from the date of each
observation in the sample), and LSALES, the log of the frm’s sales in hundreds of
millions of dollars, control for the frm’s overall risk. Older frms are typically better
established and so less risky. Similarly, larger frms are usually better diversifed across
customers, suppliers and regions.
JFEP
6,3
274
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Asubset of frmvariables controls for the risk of the frm’s debt. It includes the frm’s
proft margin, P ROF M ARGIN (net income divided by sales); interest coverage, IN T
EREST COV (EBITDA divided by interest expense); the leverage ratio, LEVERAGE
(debt over assets); and its earnings volatility, EARNINGS VOL (the standard deviation
of the frm’s quarterly return on assets over the past three years). More proftable frms,
as well as frms with higher interest coverage, have a greater cushion for servicing debt
and so should pay lower spreads on their loans. In contrast, frms with higher
leverage and those with higher earnings volatility will likely have a higher probability
of default and so should pay higher spreads on their loans.
The next set of frm variables attempts to control for another aspect of credit risk – the
losses that debt holders incur in the event of default. To capture this, we consider several
variables that measure the size andqualityof the asset base that debt holders candrawonin
default, including the frm’s tangible assets, TANGIBLES (inventories plus plant, property
andequipment over assets); its advertisingexpenses, ADVERTISING(advertisingexpense
divided by sales); and its expenses with research and development, R&D (research and
development expense dividedbysales)[10]. Tangible assets lose less of their value indefault
than do intangible assets, so we expect this variable to have a negative effect on spreads. In
contrast, advertising expenses and R&Dexpenses, which proxy for the frm’s brand equity
and intellectual capital, respectively, are intangible, and so we also expect them to have a
positive effect onspreads. We also control for the value the frmis expectedto gainbyfuture
growth, MKTTOBOOK (frm’s market to book ratio), and the frm’s net working capital,
NWC (current assets less current liabilities over debt)[11]. Although growth opportunities
are vulnerable tofnancial distress, we alreadycontrol for the portionof the frm’s assets that
are tangible. Thus, this variable could have a negative effect on spreads if it represents
additional value (over and above the book value) that debt holders can, in part, access in the
event of default. Withregards tothe frm’s net workingcapital, as the frm’s liquidasset base
is less likely to lose value in default, we expect this variable to have a negative effect on
spreads.
We complement this set of frm controls with RELATIONSHIP, which is a dummy
variable equal to one if the frmborrowed fromthe same lead arranger in the three years
prior to the current loan[12]. In addition, we include dummy variables for single-digit
Standard Industrial Classifcation industry groups, as each industry may face
additional risk factors that are not captured by our controls, and include a time trend,
TREN D, to account for a potential secular trend in loan interest rates.
The next set of variables controls for aspects related to the loan that are likely to
affect loan spreads. It includes the log of loan amount in dollars, LAM OU N T; and the
log of the loan maturity in years, LM AT U RIT Y. Larger loans may represent more
credit risk, but they may also allow economies of scale in processing and monitoring
the loan. Similarly, loans with longer maturities may face greater credit risk, but they are
more likely to be granted to frms that are thought to be more creditworthy. So, the
effects of these variables on the spread are ambiguous. This set also includes dummy
variables equal to one if the loan has restrictions on paying dividends (DIV IDEN D
REST), is senior (SEN IOR) or is secured (SECU RED). All else equal, any of these
features should make the loan safer, decreasing the spread, but it is well known that
lenders are more likely to require these features if they think the frm is riskier (Berger
and Udell, 1990), so the relationship may be reversed. Because the purpose of the loan is
likely to affect its credit spread, we include dummy variables for loans taken out for
275
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
corporate purposes (CORP P U RP OSES), to refnance a loan (REF IN AN CE) and for
working capital purposes (WORKCAP ITAL). Similarly, we include dummy variables
to account for the type of the loan, in particular for lines of credit (CREDIT LIN E) and
for term loans (T ERM LOAN). Some of the loan controls are likely to be jointly
determined with loan spreads[13]. Because we do not have instruments for the various
loan controls, we address the concerns that may arise with this endogeneity by
estimating our models both with and without the set of loan controls. Using either
approach does not affect our key fndings.
Finally, we control for a set of bank-specifc variables. These variables aim at
controlling for aspects related to banks that are likely to play a role in their loan
pricing policies. In addition to potentially having a direct effect on the pricing of the
loan, these variables are also meant to proxy for unobserved characteristics of
banks, including their management quality that may affect their loan pricing
policies at a given time. LASSET S, the log of the bank’s total assets in hundreds of
millions of dollars, controls for bank size. Larger banks are likely to be better
diversifed or to have access to funding under better terms giving them the
opportunity to charge lower loan spreads. If safer banks are able to access funding
under better terms, then we also expect other measures of bank risk, such as the
return on assets, ROA; the volatility of return on assets, ROAVOL; and net loan
charge-offs as a fraction of assets, CHARGEOF F S0 to be correlated with the
interest rates banks charge on their corporate loans[14]. For the same reason we
expect the bank’s capital-to-assets ratio, CAPITAL, to be negatively related to loan
interest rates. This relationship may also arise because, according to Boot et al.
(1993), banks with low capital are more willing to consume reputational capital to
build up fnancial capital and thus are more likely to renege on implicit guarantees,
including the guarantee not to explore their informational monopoly.
We include the bank’s holdings of cash and marketable securities as a fraction of total
assets, LIQUIDITY, because banks with more liquid assets may fnd it easier to fund loans
onthe margin, leadingto lower loanspreads. We include the fractionof subdebt over assets,
SUBDEBT, to control for the importance of bond fnancing for the bank. Again, banks that
rely on this funding source are likely to fnd it easier to fund loans on the margin and so we
should expect them to charge lower spreads. This variable is also likely to have a negative
effect on loan spreads because bond fnancing is predominantly used by larger banks.
Finally, we include the fraction of total deposits over assets, DEPOSITS, to control for the
importance of deposit funding for the bank. Insured deposits are believed to be the least
expensive source of fundingfor banks. Our control for the importance of this fundingsource,
however, also includes uninsured deposits which tend to be more expensive[15].
2.1.2 Are relationship borrowers less exposed to the cost of banks’ bond funding? We
next investigate whether banks pass the shocks to their funding costs that arise with
their use of bond fnancing to all of their borrowers equally. Specifcally, we are
interested in learning whether borrowers that have a lending relationship with
banks are exposed to these shocks to a lesser or greater extent. If banks expect to
continue doing business with their relationship borrowers, they may be willing to
shield them from bond market shocks, thereby smoothing the interest rates they
charge their relationship borrowers over time. On the other hand, the pressure to
maintain their fnancial performance may lead banks to renege on implicit
guarantees, including the guarantee not to explore their informational monopoly
JFEP
6,3
276
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
over their relationship borrowers, and pass onto these frms a larger portion of the
cost increase they face in raising funding in the bond market[16].
Thus, we investigate whether banks shield their relationship borrowers from the
shocks they face to their bond funding costs or whether they exploit their informational
advantage over such borrowers. To that end, we estimate the following model:
LLOANSPD
b, f, l, t
? c ? ? LBK BOND COST
b
? ? RELATIONSHIP
f, t?1
? ? RELATIONSHIP
f, t?1
LBK BOND COST
b
? ?LBBBSPD
t
? ? LIBOR
t
?
?
i?1
I
?
i
B
i, b, t?1
?
?
j?1
J
?
j
F
j, f, t?1
?
?
k?1
K
?
k
L
k, l
? ?
f, t
.
where, all the variables are as defned in Part I. If banks shield their relationship
borrowers fromthe shocks to their bond fnancing costs, then we would expect ??0. If,
on the other hand, banks pass these shocks onto their relationship borrowers to a greater
extent than on other borrowers, then we would expect ? ?0.
These effects are likely to vary within the sample of relationship borrowers
depending on how dependent the borrower is on the bank for funding. For example,
banks will likely fnd it diffcult to pass shocks to their cost of funding onto borrowers
that have alternative funding sources, as these borrowers will respond by ending their
relationship with the bank. Banks will fnd it easier to pass these shocks onto borrowers
that depend on them for their funding, but the greater prospects of future business with
these borrowers may lead banks to shield them form shocks to their funding costs.
To investigate these possibilities, we need to distinguish borrowers that are
dependent on banks fromthose that are not. We assume that borrowers that have access
to the bond market are not bank dependent. Besides having access to an alternative
source of funding, there will also be more information available on these borrowers
coming from the opinions of bond analysts and ratings of rating agencies and the
spreads on their bonds. This additional information will reduce banks ability to hold
these borrowers up for higher interest rates Sharpe (1990) and Rajan (1992) when their
cost of funding goes up.
Our loan pricing model (2) distinguishes between borrowers that have and those that
do not have a lending relationship with their bank, and it compares howthese borrowers
are exposed to changes in the cost of bond fnancing of their banks. To avoid adding a
third level of interaction terms that would distinguish borrowers that have access to the
bond market among those that have relationship with their bank, we opted for
estimating model (2) separately for borrowers with a lending relationship with their
bank and those borrowers without such a relationship. Within these subsamples of
loans, we distinguish between borrowers with and without access to the bond market. In
other words, we replace RELAT ION SHIP
f,t?1
, in model (2) with M ACCESS
f,t?1
, our
proxy for borrowers’ access to the bond market, and estimate the model separately for
relationship borrowers and non-relationship borrowers.
We assume that borrowers have access to the bond market if they issued a public
bond in the recent past. For the purpose of our tests, we defne the recent past as the
three-year period prior to the loan. We get similar results if we assume that any borrower
that has issued at least one bond in the past has access to the bond market[17]. Further,
277
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
as according to Rajan (1992), the holdup problem is more acute for risky frms than safe
frms, among the relationship borrowers with access to the bond market, we would
expect those that are rated investment grade to be the least exposed to the bond market
shocks that affect their bank’s cost of funding. For this reason, we also distinguish
among the frms that have access to the bond market those that are rated investment
grade from those that are rated below investment grade. We use the rating of the
borrower’s most recent public bond (prior to a given loan) to determine whether it is
rated investment grade or below investment grade.
2.2 Data
The data for this project come from several data sources, including the Loan Pricing
Corporation’s Dealscan database (LPC), the Securities Data Corporation’s Domestic New
Bond Issuances database (SDC), the Center for Research on Securities Prices’s stock
prices database (CRSP), the Salomon Brother’s bond yields indices, Compustat and from
the Federal Reserve’s Call Reports.
We use LPC’s Dealscan database of business loans to identify frms that borrowed
from banks and to gather information on their loans. This database goes as far back as
the beginning of the 1980s, but in the frst part of that decade, it has a somewhat limited
number of entries. It is for this reason that we begin our sample in 1987. Our sample ends
in December 2007 to avoid any effects arising from the subprime fnancial crisis.
We rely on SDC’s Domestic New Bond Issuances database to identify which frms in
our sample issued bonds prior to borrowing in the syndicated loan market and to gather
information on banks’ bond issuance activity. We also rely on this database to gather
information on the bond issuance activity of banks.
We use Compustat to get information on frms’ balance sheets. Even though LPC
contains loans from both privately held and publicly listed frms, given that Compustat
is dominated by the latter, we have to exclude from our sample the loans borrowed by
privately held frms.
We rely on the CRSP database to link companies and subsidiaries that are part of the
same frm, and to link companies over time that went through mergers, acquisitions or
name changes[18]. We then use these links to merge the LPC, SDC and Compustat
databases to fnd out the fnancial condition of the frm at the time it borrowed from
banks and if, by that date, the frm had already issued bonds.
We use Salomon Brothers’ indices on the yields of new industrial long-term triple-A
and triple-B-rated bonds to control for the conditions in the bond market at the time
frms take out loans from banks.
Finally, we use the Reports of Condition and Income (Call Reports) compiled by the
FDIC, the Comptroller of the Currency, and the Federal Reserve System to obtain
bank-level data for the lead bank(s) in each loan syndicate.
2.3 Sample characterization
Table I characterizes our sample of 19,930 loans. These loans are extended by 381 banks
over the years 1987-2007 to 4,222 borrowers. The top panel compares the 150 banks that
had access to the bond market at the time of their loans with the 335 banks that relied on
deposit funding[19]. We classify a bank as having access to bond fnancing, if it issued
at least once in the bond market in the three years prior to the loan and it still had public
debt on its balance sheet at the time of the loan. Otherwise, we assume the bank funds its
JFEP
6,3
278
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
operations mainly with deposits. The middle panel compares the 16,212 loans in the
sample that “bond-fnancing” banks extended with the 3,718 loans extended by banks
that do not use bond fnancing. Finally, the bottompanel of the table compares the 3,529
borrowers that took out loans from bond-fnancing banks with the 1,699 borrowers that
took out loans from deposit-fnancing banks.
Looking at the top panel, it is apparent that, compared to banks that rely on deposit
funding, banks that use bond fnancing are larger, hold less liquidity and operate with a
lower deposit-to-asset ratio. Bond fnancing banks appear to be less proftable, as they
have a lower ROA. Further, these banks may be riskier. Their ROA is less volatile, but
they have a lower capital-to-asset ratio (CAP IT AL).
Table I.
Sample characterization
a
Variables
Bond fnancing
banks
Deposit fnancing
banks Difference p-value
Differences among banks
ASSETS 467.990 85.985 382.006 0.000
ROA 0.126 0.138 ?0.012 0.000
ROAVOL 0.001 0.002 ?0.001 0.000
CHARGEOFFS 0.107 0.110 ?0.003 0.244
LIQUIDITY 19.795 26.386 ?6.591 0.000
DEPOSITS 30.602 48.749 ?18.147 0.000
CAPITAL 7.386 8.151 ?0.765 0.000
Differences in the loan policies
LOANSPD 157.969 218.325 ?60.356 0.000
AMOUNT 431.923 136.613 295.309 0.000
MATURITY 4.050 3.572 0.478 0.029
SECURED 0.416 0.624 ?0.208 0.000
SENIOR 0.964 0.930 0.034 0.000
DIVIDENDREST 0.457 0.426 0.031 0.000
CORPORATEPURP 0.304 0.263 0.041 0.000
REFINANCE 0.623 0.444 0.179 0.000
WORKINGCAPITAL 0.179 0.214 ?0.035 0.000
TERMLOAN 0.391 0.352 0.039 0.000
CREDITLINE 0.578 0.593 ?0.015 0.101
RELATIONSHIP 0.635 0.509 0.126 0.000
Differences among borrowers
AGE 23.455 15.126 8.329 0.000
SALES 6854.742 1707.320 5147.422 0.000
PROMARGIN ?0.009 ?0.052 0.044 0.026
INTERESTCOV 26.890 24.171 2.719 0.641
EARNINGSVOL 45.652 19.041 26.612 0.000
LEVERAGE 0.319 0.302 0.017 0.000
TANGIBLES 0.730 0.740 ?0.010 0.167
ADVERTISING 0.011 0.009 0.002 0.000
RD 0.028 0.042 ?0.014 0.138
NWC 8.001 5.771 2.230 0.271
MKTTOBOOK 1.784 1.744 0.040 0.136
PBOND 0.596 0.329 0.267 0.000
(continued)
279
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
In the middle panel of the table, we see that bond fnancing banks charge, on average, 60
basis points less on their loans than deposit fnancing banks. This difference may arise
because bond fnancing banks extend signifcantly larger loans or because they extend a
larger fraction of their loans borrowers that are not bank dependent. Alarger portion of the
borrowers of these banks have a credit rating or issued bonds in the public bond market in
the three years prior to the loan, confrming that borrowers of these banks are less likely to
be bank dependent. That difference in interest rates may also refect a difference in the risk
of these banks’ borrowers. However, in this case, the evidence appears to be mixed. As we
can see fromthe bottompanel, compared to borrowers of deposit funding banks, borrowers
of bondfnancingbanks are older, larger, have better proft margins andhigher net working
capital and they have more growth opportunities. All of these features suggest that bond
fnancingbanks tendtoextendloans tosafer borrowers. There is alsoevidence that suggests
otherwise. For example, borrowers of bondfnancingbanks, onaverage, have lower interest
coverage and less tangible assets. Further, they have higher leverage ratios, and their
earnings are more volatile.
Last, that difference in the interest rates that these banks charge on their loans to
corporate borrowers may arise because of a difference in these banks’ funding costs.
Table I.
Variables
Bond fnancing
banks
Deposit fnancing
banks Difference p-value
PBONDIG 0.272 0.067 0.204 0.000
PBONDBG 0.196 0.143 0.053 0.000
Notes:
a
ASSETS Bank assets in $100 million dollars. ROA, returns on assets (net income divided by
assets); ROAVOL, standard deviation of the quarterly ROA computed over the past three years;
CHARGEOFFS, net charge-offs over assets; LIQUIDITY, cash plus securities over assets; DEPOSITS,
deposits over assets; DEPOSIT COST, product between the ratio of deposits over assets and the three
month LIBOR; INT EXPENSE, interest expenses on deposits alone (over deposits); numbers reported
are for only 13,009 of the 19,930 observations in the sample because it is missing for the remaining
banks; CAPITAL, equity capital over assets; LOANSPD, all-in-drawn loan spread over LIBOR at
origination; AMOUNT, loan amount; M ATURITY, loan maturity in years; SECURED, dummy
variable equal to 1 if the loan is secured; SEN IOR, dummy variable equal to 1 if the loan is senior;
DIVIDENDREST, dummy variable equal to 1 if the borrower faces dividend restrictions in connection
with the loan; CORP P URPOSES, dummy variable equal to 1 if the loan is corporate purposes; REF IN
ANCE, dummy variable equal to 1 if the loan is to refnance existing debt; WORKCAPITAL, dummy
variable equal to 1 if the loan is for working capital; T ERM LOAN, dummy variable equal to 1 if it is
a term loan; CREDIT LIN E, dummy variable equal to 1 if it is a credit lien; AGE, age of the borrower
in years; SALES, sales in millions of dollars; P ROF M ARGIN, net income over sales; IN T COV, the
interest coverage (EBITDA divided by interest expense); EARNINGS VOL, earnings volatility (the
standard deviation of the frm’s quarterly ROA over the past three years); LEV ERAGE, leverage ratio
(debt over total assets); TANGIBLES, tangible assets (inventories plus plant, property and equipment
over total assets); ADV ERT ISIN G, expenses with advertising scaled by the frm’s sales; R&D,
expenses with R&D scaled by the frm’s sales; N W C, net working capital; M KTOBOOK, market to
book value; P BON D, dummy variable equal to 1 if the borrower issued a public bond in the past three
years prior to the loan; P BON DIG, dummy variable equal to 1 if the borrower issued a public bond
which was rated investment grade in the past three years prior to the loan; P BON DBG, dummy
variable equal to 1 if the borrower issued a public bond which was rated below investment grade in the
past three years prior to the loan
JFEP
6,3
280
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Bond fnancing banks appear to be able to raise deposit funding at lower interest rates.
In addition, they can complement this funding source with bond fnancing.
2.3.1 Banks’ bond funding costs and their loan spreads. In Table II, we take a frst
look at whether there is a link between banks’ loan pricing policies and their bond
funding costs. To that end, we compare, in the frst rowof the table, the loan spreads that
banks charge their borrowers when the cost banks pay to issue in the bond market is low
and high, respectively. We proxy for this cost by the spread between triple-B and
triple-A primary yields on new bonds issued at the time the bank issued its most recent
bond prior to the loan, BBBSP D
b
. We assume the cost to issue in the bond market is low
(high) if BBBSP D
b
is in the lowest (highest) quartile of the distribution of this variable
during the sample period. When banks pay low bond yields to issue in the bond market,
theycharge, onaverage,137 bps ontheir corporate loans. Whentheypayhighyields to issue
bonds, theycharge their borrowers, onaverage, 176 bps. Thus, whenthe banks’ cost to issue
in the bond market goes up, they increase the spreads they charge on their loans by 38 bps.
Banks appear toshield, at least inpart, their relationshipborrowers fromthese cost hikes. As
we can see from the middle panel of Table II, when the cost banks pay to issue in the bond
market goes up, they increase the spreads on the loans they extend their non-relationship
borrowers by44basis points. Onthese occasions, theyincrease the spreads theycharge their
relationship borrowers by only 35 basis points.
Finally, as shown in the bottom panel of Table II, we investigate if it is important for
borrowers to have access to the bond market. We classify borrowers that issued in the
bond markets at least once in the three years prior to the loan date to have access to the
bond market. Borrowers that have never issued in the bond market or those that only
Table II.
Loan spreads and banks’
cost of bond fnancing:
univariate analysis
a
Low High
Difference p-value BK BOND COST BK BOND COST
Loan spreads: differences as bank cost to issue in the bond market changes
137.2 175.6 38.4 0.000
Differences depending on whether the borrower has a lending relationship with the bank
Relationship 125.6 161.0 35.4 0.000
No relationship 156.1 200.3 44.1 0.000
Differences depending on whether the borrower has access to the bond market
Relationship
Access 114.1 117.2 3.14 0.540
No access 132.4 192.9 60.5 0.000
No relationship
Access 136.4 151.8 15.4 0.060
No Access 162.8 219.4 56.6 0.000
Notes:
a
Computations limited to banks that rely on bond fnancing at the time of the loan; this means
the bank issued at least once in the bond market in the three years prior to the loan, and it has public debt
in its balance sheet at the time of the loan; the loan spread is the all-in-drawn loan spread over LIBORat
origination; high (low) BK BON D COST is the top (bottom) quartile of the difference between the
Moody’s indexes on the ex ante yields of triple-B- and triple-A-rated bonds at the time of the most recent
bond the bank issued prior to the loan; borrowers have a lending relationship with the bank if they
borrowed fromit at least once in the past three years prior to the loan; borrowers have access to the bond
market if they issued at least once in the bond market in the three years prior to the loan
281
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
issued more than three years ago (from the loan date) are classifed as bank dependent.
As we can see from that panel, effectively non-bank-dependent borrowers are less
exposed to the cost banks pay to raise funding in the bond market, irrespective of
whether they have a relationship with their bank. More importantly, among relationship
borrowers, those that are not dependent on banks for funding do not appear to be
exposed to shocks to the cost their banks pay to issue in the bond market. While
relationship borrowers that are bank dependent pay, on average, an additional 61 basis
points when it is more expensive for banks to issue in the bond market (meaning that
they issued when BBBSP D
b
was in the upper quartile of the distribution of this variable
as opposed to the lower quartile of that distribution), relationship borrowers that are not
bank dependent pay only an additional 3 basis points, an increase which is not
statistically different from zero.
In sum, the results of our sample characterization suggest that banks do adjust their
loan pricing policies in response to changes in the cost they pay to raise funding in the
bond market. Banks appear to shield their relationship borrowers from changes in the
cost of this source of funding, but only if they are not dependent on them for funding.
Relationship borrowers that are dependent on banks for external funding seem to be
exposed to shocks to the cost of funding of their banks. In the rest of this paper, we look
at the exposures of these cohorts of borrowers to the cost their banks pay to issue in the
bond market more closely, using multivariate analysis.
3. Do banks pass bond market shocks onto their borrowers?
We investigate, in this section, whether banks’ reliance on the bond market to fund their
activities creates a link between the cost they pay to issue bonds and the spreads they
charge on their corporate loans. In the next section, we investigate whether banks shield
their relationship borrowers from shocks to the cost they pay to raise bond fnancing.
Table III reports the results of our tests of whether banks that rely on bond fnancing
adjust their loan spreads in response to changes in the cost they pay to issue in the bond
market. We measure this cost at the time of banks’ most recent bond issue prior to any
given loan. Given that yields are missing for a large number of bonds issued by banks,
we proxy for that cost by the log of the spread between triple-B and triple-A primary
yields on new bonds issued on the day the bank issued its most recent bond prior to the
loan, LBBBBSP D
b
[20]. Model 1 investigates whether banks adjust their loan pricing
policies in response to changes in the cost they pay to access the bond market,
controlling for our set of frm-specifc characteristics, F. Models 2 expands our controls
to account for our set of loan-specifc variables, L. As we discussed in the methodology
section, we estimate our model with and without loan controls to reduce concerns with
the potential endogeneity of some of these controls. Model 3 investigates what happens
when we further expand our controls to account for our set of bank-specifc variables, B.
Models 4 and 5, in turn, investigate the robustness of our fndings when we control for
the overall level of interest rates at the time of the loan. In model 4, we control for the log
of the spread between triple-B and triple-A primary yields on new bonds issued at the
time of the loan, LBBBBSP D
l
, and in model 5, we further control for the level of Libor at
the time of the loan, LIBOR. These controls are important to assure us that the link we
identify between loan spreads and the bank’s cost of bond fnancing is not driven by
changes in the overall interest rates or in the price of risk at the time of the loan. Finally,
JFEP
6,3
282
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table III.
Shocks to bond markets
and bank loan pricing
policies
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
L
B
B
B
S
P
D
b
0
.
2
9
3
*
*
*
(
0
.
0
1
2
2
)
0
.
3
1
1
*
*
*
(
0
.
0
1
1
5
)
0
.
2
8
5
*
*
*
(
0
.
0
1
3
3
)
0
.
1
6
9
*
*
*
(
0
.
0
2
9
4
)
0
.
1
3
7
*
*
*
(
0
.
0
2
9
5
)
0
.
1
1
2
*
*
*
(
0
.
0
4
1
6
)
L
B
B
B
S
P
D
l
0
.
1
2
3
*
*
*
(
0
.
0
2
8
1
)
0
.
1
2
6
*
*
*
(
0
.
0
2
8
1
)
0
.
1
4
5
*
*
*
(
0
.
0
4
1
5
)
L
I
B
O
R
?
0
.
0
1
7
2
*
*
*
(
0
.
0
0
3
1
7
)
?
0
.
0
2
0
0
*
*
*
(
0
.
0
0
5
6
5
)
L
A
G
E
?
0
.
1
4
5
*
*
*
(
0
.
0
0
7
3
6
)
?
0
.
1
0
5
*
*
*
(
0
.
0
0
6
5
3
)
?
0
.
1
0
3
*
*
*
(
0
.
0
0
6
5
0
)
?
0
.
1
0
3
*
*
*
(
0
.
0
0
6
5
1
)
?
0
.
1
0
4
*
*
*
(
0
.
0
0
6
4
9
)
?
0
.
1
0
5
*
*
*
(
0
.
0
1
7
8
)
L
S
A
L
E
S
?
0
.
2
3
5
*
*
*
(
0
.
0
0
3
6
1
)
?
0
.
1
0
1
*
*
*
(
0
.
0
0
5
5
4
)
?
0
.
1
0
2
*
*
*
(
0
.
0
0
5
6
4
)
?
0
.
1
0
3
*
*
*
(
0
.
0
0
5
6
5
)
?
0
.
1
0
4
*
*
*
(
0
.
0
0
5
6
3
)
?
0
.
1
0
4
*
*
*
(
0
.
0
0
9
5
0
)
L
E
V
E
R
A
G
E
0
.
3
9
5
*
*
*
(
0
.
0
4
2
5
)
0
.
3
2
1
*
*
*
(
0
.
0
3
3
6
)
0
.
3
2
1
*
*
*
(
0
.
0
3
3
5
)
0
.
3
2
2
*
*
*
(
0
.
0
3
3
3
)
0
.
3
2
2
*
*
*
(
0
.
0
3
3
0
)
0
.
3
1
6
*
*
*
(
0
.
0
5
1
7
)
T
A
N
G
I
B
L
E
S
?
0
.
0
7
4
6
*
*
*
(
0
.
0
1
8
3
)
?
0
.
0
2
7
2
*
(
0
.
0
1
5
7
)
?
0
.
0
2
5
3
(
0
.
0
1
5
6
)
?
0
.
0
2
5
1
(
0
.
0
1
5
6
)
?
0
.
0
2
7
6
*
(
0
.
0
1
5
6
)
?
0
.
0
1
6
2
(
0
.
0
1
9
6
)
R
O
A
V
O
L
0
.
0
9
9
8
*
*
*
(
0
.
0
2
4
1
)
0
.
0
9
3
5
*
*
*
(
0
.
0
2
0
8
)
0
.
0
8
5
7
*
*
*
(
0
.
0
2
0
6
)
0
.
0
8
6
6
*
*
*
(
0
.
0
2
0
6
)
0
.
0
8
3
3
*
*
*
(
0
.
0
2
0
4
)
0
.
0
8
1
8
*
*
*
(
0
.
0
2
3
3
)
R
D
?
0
.
2
0
1
*
*
*
(
0
.
0
5
7
4
)
?
0
.
1
3
7
*
*
*
(
0
.
0
4
1
4
)
?
0
.
1
3
4
*
*
*
(
0
.
0
4
1
6
)
?
0
.
1
3
3
*
*
*
(
0
.
0
4
1
6
)
?
0
.
1
3
1
*
*
*
(
0
.
0
4
1
3
)
?
0
.
1
2
6
*
*
(
0
.
0
5
6
1
)
A
D
V
E
R
T
I
S
I
N
G
?
0
.
6
5
4
*
*
*
(
0
.
2
4
2
)
?
0
.
6
3
5
*
*
*
(
0
.
1
9
1
)
?
0
.
6
5
6
*
*
*
(
0
.
1
9
4
)
?
0
.
6
5
5
*
*
*
(
0
.
1
9
4
)
?
0
.
6
6
3
*
*
*
(
0
.
1
9
5
)
?
0
.
7
3
4
*
*
*
(
0
.
2
3
7
)
L
I
N
T
C
O
V
?
0
.
2
0
5
*
*
*
(
0
.
0
1
1
9
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
6
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
1
)
?
0
.
1
6
2
*
*
*
(
0
.
0
2
0
4
)
M
K
T
B
O
O
K
?
0
.
2
0
5
*
*
*
(
0
.
0
1
1
9
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
4
*
*
*
(
0
.
0
0
9
3
3
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
6
)
?
0
.
1
6
3
*
*
*
(
0
.
0
0
9
3
1
)
?
0
.
1
6
2
*
*
*
(
0
.
0
2
0
4
)
P
R
O
F
M
A
R
G
I
N
?
0
.
0
3
0
9
(
0
.
0
2
2
3
)
?
0
.
0
3
0
9
*
(
0
.
0
1
6
0
)
?
0
.
0
3
0
1
*
(
0
.
0
1
6
2
)
?
0
.
0
2
9
8
*
(
0
.
0
1
6
1
)
?
0
.
0
2
8
8
*
(
0
.
0
1
6
0
)
?
0
.
0
2
9
3
(
0
.
0
2
0
5
)
N
W
C
0
.
0
7
2
2
(
0
.
0
5
6
3
)
0
.
0
6
6
1
(
0
.
0
5
7
1
)
0
.
0
6
4
4
(
0
.
0
5
6
7
)
0
.
0
6
2
3
(
0
.
0
5
7
5
)
0
.
0
6
5
7
(
0
.
0
5
8
4
)
0
.
0
6
5
7
(
0
.
0
5
1
9
)
R
E
L
A
T
I
O
N
S
H
I
P
?
0
.
0
6
4
3
*
*
*
(
0
.
0
1
1
2
)
?
0
.
0
2
6
7
*
*
*
(
0
.
0
0
9
8
7
)
?
0
.
0
2
5
3
*
*
(
0
.
0
0
9
9
2
)
?
0
.
0
2
3
9
*
*
(
0
.
0
0
9
9
1
)
?
0
.
0
2
3
9
*
*
(
0
.
0
0
9
9
0
)
?
0
.
0
2
3
2
(
0
.
0
2
1
4
)
T
R
E
N
D
0
.
0
2
5
9
*
*
*
(
0
.
0
0
1
3
0
)
0
.
0
2
1
3
*
*
*
(
0
.
0
0
1
4
9
)
0
.
0
2
5
7
*
*
*
(
0
.
0
0
2
0
0
)
0
.
0
2
5
6
*
*
*
(
0
.
0
0
1
9
9
)
0
.
0
2
2
6
*
*
*
(
0
.
0
0
2
0
8
)
0
.
0
3
0
9
*
*
*
(
0
.
0
0
7
5
6
)
L
A
M
O
U
N
T
?
0
.
1
0
8
*
*
*
(
0
.
0
0
5
7
6
)
?
0
.
1
0
8
*
*
*
(
0
.
0
0
5
7
6
)
?
0
.
1
0
8
*
*
*
(
0
.
0
0
5
7
7
)
?
0
.
1
0
6
*
*
*
(
0
.
0
0
5
7
5
)
?
0
.
1
0
7
*
*
*
(
0
.
0
0
6
7
1
)
L
M
A
T
U
R
I
T
Y
0
.
0
9
5
3
*
*
*
(
0
.
0
0
7
8
6
)
0
.
0
9
2
1
*
*
*
(
0
.
0
0
7
8
6
)
0
.
0
9
4
4
*
*
*
(
0
.
0
0
7
8
6
)
0
.
0
9
5
5
*
*
*
(
0
.
0
0
7
8
9
)
0
.
0
9
6
4
*
*
*
(
0
.
0
1
7
1
)
S
E
C
U
R
E
D
0
.
5
6
0
*
*
*
(
0
.
0
1
2
8
)
0
.
5
5
4
*
*
*
(
0
.
0
1
2
7
)
0
.
5
5
4
*
*
*
(
0
.
0
1
2
7
)
0
.
5
5
3
*
*
*
(
0
.
0
1
2
7
)
0
.
5
4
1
*
*
*
(
0
.
0
3
7
6
)
C
O
R
P
P
U
R
P
O
S
E
S
?
0
.
0
1
2
3
(
0
.
0
1
1
7
)
?
0
.
0
1
9
3
*
(
0
.
0
1
1
7
)
?
0
.
0
1
7
4
(
0
.
0
1
1
7
)
?
0
.
0
2
3
0
*
(
0
.
0
1
1
7
)
?
0
.
0
2
3
9
(
0
.
0
2
0
6
)
R
E
F
I
N
A
N
C
E
?
0
.
0
7
4
0
*
*
*
(
0
.
0
1
1
6
)
?
0
.
0
7
7
1
*
*
*
(
0
.
0
1
1
6
)
?
0
.
0
7
7
4
*
*
*
(
0
.
0
1
1
6
)
?
0
.
0
8
1
5
*
*
*
(
0
.
0
1
1
5
)
?
0
.
0
7
9
6
*
*
*
(
0
.
0
1
5
1
)
W
O
R
K
C
A
P
I
T
A
L
?
0
.
0
6
3
6
*
*
*
(
0
.
0
1
3
0
)
?
0
.
0
7
2
6
*
*
*
(
0
.
0
1
3
1
)
?
0
.
0
7
1
0
*
*
*
(
0
.
0
1
3
1
)
?
0
.
0
8
0
2
*
*
*
(
0
.
0
1
3
2
)
?
0
.
0
8
0
4
*
*
*
(
0
.
0
2
1
8
)
T
E
R
M
L
O
A
N
?
0
.
4
0
1
*
*
*
(
0
.
0
3
7
1
)
?
0
.
3
9
4
*
*
*
(
0
.
0
3
6
9
)
?
0
.
3
9
6
*
*
*
(
0
.
0
3
6
8
)
?
0
.
4
0
3
*
*
*
(
0
.
0
3
6
9
)
?
0
.
4
1
5
*
*
*
(
0
.
0
5
9
9
)
C
R
E
D
I
T
L
I
N
E
?
0
.
4
8
1
*
*
*
(
0
.
0
3
7
4
)
?
0
.
4
7
3
*
*
*
(
0
.
0
3
7
2
)
?
0
.
4
7
4
*
*
*
(
0
.
0
3
7
1
)
?
0
.
4
8
0
*
*
*
(
0
.
0
3
7
2
)
?
0
.
4
9
4
*
*
*
(
0
.
0
7
3
0
)
D
I
V
I
D
E
N
D
R
E
S
T
0
.
1
2
1
*
*
*
(
0
.
0
1
1
1
)
0
.
1
3
4
*
*
*
(
0
.
0
1
1
3
)
0
.
1
3
4
*
*
*
(
0
.
0
1
1
3
)
0
.
1
3
1
*
*
*
(
0
.
0
1
1
3
)
0
.
1
6
0
*
*
*
(
0
.
0
3
4
4
)
(
c
o
n
t
i
n
u
e
d
)
283
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table III.
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
S
E
N
I
O
R
?
0
.
1
0
2
*
*
*
(
0
.
0
2
4
9
)
?
0
.
1
1
7
*
*
*
(
0
.
0
2
4
9
)
?
0
.
1
2
6
*
*
*
(
0
.
0
2
5
0
)
?
0
.
1
0
7
*
*
*
(
0
.
0
2
5
3
)
?
0
.
0
9
9
3
*
*
(
0
.
0
4
0
6
)
L
A
S
S
E
T
S
0
.
0
1
2
7
*
*
(
0
.
0
0
5
6
8
)
0
.
0
1
4
0
*
*
(
0
.
0
0
5
6
6
)
0
.
0
1
6
7
*
*
*
(
0
.
0
0
5
7
2
)
?
0
.
0
7
1
7
(
0
.
0
5
4
3
)
R
O
A
4
.
7
2
6
(
6
.
2
9
0
)
5
.
1
3
7
(
6
.
2
9
4
)
4
.
3
7
3
(
6
.
2
7
9
)
7
.
2
4
2
(
1
1
.
5
0
)
C
H
A
R
G
E
O
F
F
S
2
0
.
4
5
*
*
*
(
6
.
1
4
4
)
2
0
.
5
5
*
*
*
(
6
.
1
2
9
)
1
7
.
9
1
*
*
*
(
6
.
1
4
2
)
?
1
.
2
9
8
(
1
0
.
2
2
)
R
O
A
V
O
L
?
1
.
2
8
8
(
4
.
6
9
7
)
?
3
.
7
0
4
(
4
.
7
3
6
)
?
0
.
5
8
3
(
4
.
7
4
8
)
4
.
8
8
1
(
1
1
.
6
9
)
L
I
Q
U
I
D
I
T
Y
?
0
.
2
1
3
*
*
*
(
0
.
0
8
0
0
)
?
0
.
1
8
0
*
*
(
0
.
0
8
0
2
)
?
0
.
2
1
5
*
*
*
(
0
.
0
8
0
1
)
?
0
.
2
7
3
(
0
.
2
3
5
)
C
A
P
I
T
A
L
?
0
.
0
3
0
0
*
*
*
(
0
.
0
0
4
0
5
)
?
0
.
0
2
8
3
*
*
*
(
0
.
0
0
4
0
6
)
?
0
.
0
2
7
3
*
*
*
(
0
.
0
0
4
0
5
)
?
0
.
0
1
3
9
(
0
.
0
1
1
8
)
D
E
P
O
S
I
T
S
0
.
2
4
7
*
*
*
(
0
.
0
4
6
4
)
0
.
2
4
4
*
*
*
(
0
.
0
4
6
4
)
0
.
2
5
5
*
*
*
(
0
.
0
4
6
3
)
0
.
6
6
7
*
*
(
0
.
2
6
6
)
S
U
B
D
E
B
T
?
3
.
0
4
5
*
*
*
(
0
.
7
1
1
)
?
3
.
1
8
4
*
*
*
(
0
.
7
1
1
)
?
3
.
4
1
3
*
*
*
(
0
.
7
1
0
)
?
3
.
3
8
3
*
(
1
.
8
4
3
)
C
O
N
S
T
A
N
T
6
.
5
9
5
*
*
*
(
0
.
0
8
2
6
)
6
.
3
6
4
*
*
*
(
0
.
0
8
6
5
)
6
.
2
9
2
*
*
*
(
0
.
1
4
2
)
6
.
2
6
3
*
*
*
(
0
.
1
4
2
)
6
.
3
1
2
*
*
*
(
0
.
1
4
1
)
7
.
6
9
6
*
*
*
(
0
.
8
5
0
)
B
k
f
x
e
d
e
f
f
e
c
t
s
N
O
N
O
N
O
N
O
N
O
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
R
2
A
d
j
u
s
t
e
d
0
.
4
8
3
0
.
5
9
7
0
.
6
0
0
0
.
6
0
0
0
.
6
0
1
0
.
5
6
0
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
w
h
i
c
h
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
r
e
m
a
i
n
i
n
g
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
R
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
JFEP
6,3
284
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
model 6 re-estimates our most comprehensive model (model 5) with bank fxed effects to
reduce concerns with sample selection.
Model 1 shows that banks that rely on bond fnancing take into account the cost they
incur to raise funding in the bond market when they decide on their loan spreads. The
coeffcient on our proxy for the cost the bank pays to issue in the bond market, the
triple-B over the triple-A yield spread at the time of the bank’s most recent bond issue
(prior to the loan), LBBBSP D
b
, is highly statistically signifcant and equal to 0.29. A
1-per cent increase in the cost banks pays to issue in the bond market leads to an increase
of 29 basis points in the loans spreads they charge their borrowers.
With regards to the effects of frm controls we use in model 1, they are generally
consistent with our discussion in the methodology section. As expected, older and larger
frms, as well as frms with more interest coverage, and tangibles pay lower spreads on
their loans. Firms that have a relationship with their banks are also able to borrow at
lower interest rates. Firms with higher levels of leverage and more volatile earnings pay
higher spreads on their loans. Contrary to expectations, although, frms with more R&D
expenses, as well as those with more advertising expenses (relative to their sales), are
able to borrow at lower interest rates.
Models 2 and 3 show that the link we fnd in model 1 between loan spreads and the
cost the bank pays to issue in the bond market continues to hold when we add our loan
controls and bank controls, respectively. Adding these controls does not alter the
statistical signifcance, and it has only a minor effect on the size of the coeffcient on
LBBBSP D
b
. With respect to our loan controls, their effects are generally consistent with
our intuition. Larger loans and senior loans pay lower interest rates. In contrast, longer
maturity loans carry higher interest rates. Similarly, secured loans and loans that give
rise to dividend constraints carry higher spreads. Even though these covenants aim at
protecting lenders, they are more often present in loans to riskier borrowers, thereby
explaining why these loans carry higher spreads. Term loans and credit lines carry
lower spreads. Refnance loans and loans for working capital, on the other hand, pay
higher spreads.
With regards to bank controls, our results confrmthat banks that incur larger losses
charge higher spreads on their loans. Banks with higher capital-to-asset ratios, those
with more liquidity, as well as banks that rely more heavily on subdebt charge lower
spreads. In contrast, banks that depend more heavily on deposit funding, usually
smaller banks, tend to charge higher spreads on their loans. Finally, bank size, as
measured by assets, has a positive sign, which may be contrary to expectations, but the
evidence on scale economies in banking is mixed.
The results of models 1-3 suggest that when the cost banks pay to raise funding in the
bond market goes up, banks respond by increasing the spreads they charge on their
corporate loans. A potential concern with this fnding is that the increase in the loan
spread is not actually driven by the cost banks pay to issue in the bond market but rather
arises as a result of an overall increase in the cost of funding. To alleviate this concern,
in model 4, we expand our set of controls to account for the cost to raise funding in the
bond market at the time of the loans, LBBBSP D
l
. In model 5, we expand our controls
further to include the Libor also measured at the time of the loan, LIBOR. Adding these
controls cuts in half the estimated loan spread elasticity with respect to the cost the bank
pays to raise funding in the bond market, suggesting that some of the effect in models
1-3 is indeed driven by an overall increase in the cost of funds. However, the positive and
285
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
statistically signifcant loan spread effect of the bank’s cost to issue in the bond market
remains. In other words, we fnd that, conditional on overall cost of funding at the time
of the loan, when banks incur higher costs to issue in the bond market, they tend to
charge higher loan spreads on their corporate loans.
Finally, in model 6, we show that this fnding continues to hold when we estimate
our model of loan spreads with bank fxed effects. Adding bank fxed effects further
reduces our loan spread elasticity to 0.11, but does not affect its statistical
signifcance. In the remainder of the paper, we refer to model 6 of Table III as our
benchmark specifcation.
3.1 Identifying the loan spread effect of banks’ bond fnancing costs
As we pointed out above, one concern with our fndings is that they may refect an
overall increase in the cost of credit rather than the cost banks pay to issue in the bond
market. Even though the cost of borrowing in the bond market and the loan spreads are
measured at different times, there is a high correlation, 0.93, between the bond spread at
the time of the bank’s bond issue and the bond spread at the time of the loan.
In this section, we attempt to reduce this concern by presenting the results of four
tests we developed for this purpose. The frst test investigates whether our result
changes over time. If our result is driven by changes in the overall cost of credit in the
economy, then it should be independent fromthe sample period we consider. The second
test is a falsifcation test which uses, as a control group, the banks that do not use bond
fnancing but extended loans at the same time as bond-fnancing banks. The next two
tests use two alternative measures of the cost of bond issue by bank – actual spread on
the bank’s most recent bond issue, which limits substantially our sample, but has only
0.21 correlation with the bond spread at the time of the loan and binary variables to
indicate whether the last bond prior to each loan was issued at a time of very high or very
low spreads in the bond market.
3.1.1 Loan spread effect of banks’ bond fnancing costs over time. As we noted earlier,
our frst test builds on the idea that if the link unveiled between loan spreads and the
bank’s cost of bond funding were driven by the overall cost of credit in the economy at
the time of the loan, then this effect should hold throughout the sample period. In
contrast, if that link is indeed driven by the cost banks pay to raise funding in the bond
market, then we should fnd stronger evidence of it in the most recent portion of the
sample period, as the importance of bond fnancing for banks has grown over the years.
To test this, we estimate our model separately for the frst half of our sample period
(1988-1997) and for the second half of the sample period (1998-2007). The results of these
tests are reported in Table IV. We use in this test, as well as in all the subsequent
robustness tests, our most general model, model 6 of Table III, which includes all the
control variables and estimated with bank fxed effects. In the interest of space, we do
not report the coeffcients on the control variables. Models 1 and 2 test for the effect of the
cost in the bond market at the time the bank issued its most recent bond prior to the loan,
LBBBSP D
b
. As we can see from these models, we fnd evidence of an effect of the cost
banks pay to raise funding in the bond market on their loan spreads in the second half of
the sample period (model 2), but not in the frst half of the sample period (model 1).
Models 3 and 4 expand the set of controls in the previous models to account for the
overall cost of credit in the economy at the time of the loan as measured by the cost to
issue in the bond market at the time of the loan, LBBBSP D
l
. This variable is only
JFEP
6,3
286
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table IV.
Bank cost of bond
fnancing and loan
spreads over time
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
S
a
m
p
l
e
p
e
r
i
o
d
1
9
8
8
-
9
7
1
9
9
8
-
0
7
1
9
8
8
-
9
7
1
9
9
8
-
0
7
1
9
8
8
-
9
7
1
9
9
8
-
0
7
L
B
B
B
S
P
D
b
0
.
0
1
4
1
(
0
.
0
5
5
4
)
0
.
2
2
1
*
*
*
(
0
.
0
3
7
0
)
?
0
.
0
2
1
3
(
0
.
0
6
9
7
)
0
.
0
8
1
4
*
*
(
0
.
0
3
6
4
)
?
0
.
0
4
6
4
(
0
.
0
6
4
4
)
0
.
0
7
5
2
*
*
(
0
.
0
3
3
6
)
L
B
B
B
S
P
D
l
0
.
0
5
4
2
(
0
.
0
6
1
1
)
0
.
1
6
7
*
*
*
(
0
.
0
4
2
4
)
0
.
2
0
3
*
*
(
0
.
0
8
0
0
)
0
.
1
7
5
*
*
(
0
.
0
8
4
8
)
D
E
P
O
S
I
T
S
I
N
S
U
R
E
D
?
L
B
B
B
S
P
D
b
?
0
.
4
6
2
*
*
(
0
.
1
8
2
)
?
0
.
0
3
6
6
(
0
.
2
7
7
)
D
E
P
O
S
I
T
S
I
N
S
U
R
E
D
0
.
3
2
6
(
0
.
3
5
9
)
0
.
5
7
8
(
0
.
3
8
5
)
D
E
P
O
S
I
T
S
0
.
2
7
6
(
0
.
3
5
1
)
0
.
6
3
9
*
(
0
.
3
5
2
)
0
.
2
4
9
(
0
.
3
5
5
)
0
.
6
3
9
*
(
0
.
3
4
5
)
L
I
B
O
R
?
0
.
0
3
3
7
*
*
*
(
0
.
0
1
0
7
)
?
0
.
0
2
4
7
*
*
*
(
0
.
0
0
6
7
9
)
?
0
.
0
3
4
2
*
*
*
(
0
.
0
1
0
7
)
?
0
.
0
2
3
5
*
*
*
(
0
.
0
0
6
6
8
)
?
0
.
0
3
1
9
*
*
*
(
0
.
0
1
1
3
)
?
0
.
0
2
5
3
*
*
*
(
0
.
0
0
6
5
1
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
4
,
5
4
6
1
1
,
5
2
1
4
,
5
4
6
1
1
,
5
2
1
4
,
5
2
4
1
1
,
5
2
1
R
2
A
d
j
u
s
t
e
d
0
.
5
7
4
0
.
5
7
0
0
.
5
7
4
0
.
5
7
1
0
.
5
7
6
0
.
5
7
1
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
D
E
P
O
S
I
T
S
I
N
S
U
R
E
D
,
b
a
n
k
’
s
i
n
s
u
r
e
d
d
e
p
o
s
i
t
s
o
v
e
r
a
s
s
e
t
s
.
D
E
P
O
S
I
T
S
,
b
a
n
k
’
s
t
o
t
a
l
d
e
p
o
s
i
t
s
o
v
e
r
a
s
s
e
t
s
.
A
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
287
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
signifcant in the second half of the sample, and while adding it reduces the size of the
coeffcient on LBBBSP D
b
, we continue to have the striking difference between the two
parts of our sample period. This difference suggests that the link we fnd between the
cost banks pay to raise funding in the bond market and the spreads they charge on their
corporate loans cannot be entirely driven by changes in the overall cost of credit at the
time of the loan.
The statistical signifcance of LBBBSP D
l
in the second half of the sample period
(models 3 and 4) may appear surprising in light of Berlin and Mester’s (1999) fnding that
banks shield their borrowers fromchanges in the overall cost of credit. Recall that Berlin
and Mester showthat banks that use relatively more core deposits offer more protection
to their borrowers from shocks to the current cost of credit which they proxy by
LBBBSP D
l
. To get closer to their specifcation, we modify the previous models and
control for the bank’s use of insured deposits (scaled by its assets) and the interaction of
this variable withLBBBSPD
l
[21]. The results of this exercise, whichare reportedinmodels
5 and 6 of Table IV, showthat the effect of LBBBSP D
l
is positive and signifcant in the two
parts of our sample period, while the effect of DEPOSITS INSURED ? LBBBSP D
l
is
negative, but signifcant only in the frst part. In other words, the fnding that Berlin and
Mester uncovered over the period of 1977-1989 that banks with more core deposits offered
more protection to their borrowers fromshocks to the current cost of credit, persisted in the
decade that followed their study (model 5), but weakened afterwards (model 6). In contrast,
the effect of our proxyfor the cost the bankpaidwhenit issuedthe last bondprior tothe loan,
LBBBSPD
b
, is onlysignifcant inthe secondhalf of the sample period. Thus, the differential
effect we haduncoveredfor LBBBSPD
b
duringour sample periodcontinues to holdinthese
models, addingsupport to our assertionthat the effect of the bank’s cost of bondfnancingis
not drivenbychanges inthe overall cost of credit at the time of the loan. The fact that we fnd
evidence of the link only in the most recent part of our sample was to be expected, as the
importance of bond fnancing for banks grew over time.
3.1.2 Falsifcation test. So far, we focused our attention on banks that rely on bond
fnancing, meaning that they issued at least once in the bond market in the three years
prior to the loan. We also have data on loans extended by banks that do not rely on this
source of funding, meaning that they issued bonds only more than three years prior to
the loan or have not issued any bonds since 1970 (this is the frst year we have
information on banks’ bond issuance. Recall that our sample of loans starts in 1987).
If our earlier fnding is driven by an increase in the overall cost of credit, then we
should detect a similar effect on the loan spreads of the latter set of loans. If, on the other
hand, our fnding is driven by the cost banks pay to issue in the bond market, then the
spreads on the latter loans should be unaffected by bond market conditions prior to the
loan. Thus, we construct the following matched sample. For each of the loans that had a
bond issued by the bank within the past three years, we identify loans that were
extended on the same day but did not have a bond issued within three years prior. We
fnd such a match for 6,659 of our loans extended by banks that have accessed the bond
market (treatment group), with 2,896 loans extended by banks that have not accessed
the bond market (control group). For the loans issued by the treatment group banks, we
continue using LBBBSP D
b
as the measure of the cost of bank funding through the bond
market, measured on the day the last bond prior to any given loan was issued. For the
control group we do not have that measure, by construction, because they did not issue
a bond. Thus, for the control group we construct a synthetic measure, which is equal to
JFEP
6,3
288
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
the average LBBBSP D
b
of the treatment loans issued on the same day. We re-estimate
our main model for this matched sample, the total of 9,555 loans, including the
interaction of the treatment indicator, that the banks accessed bond market, BK
ACCESS, with LBBBSPD
b
. We expect the main effect (coeffcient on LBBBSPD
b
, which
corresponds to the effect on the control group) to be zero, while the coeffcient on the
interaction term (which, when summed up with the coeffcient on LBBBSP D
b
,
corresponds to the effect on the treatment group) to be positive.
The results of this test are reported in Table V. As in the previous test, we use our
most comprehensive model of loan spreads, model 6 of Table III, and estimate the
models with all controls and bank fxed effects. The results of Table V vividly
demonstrate the importance of controlling for the cost of borrowing at the time of the
loans, LBBBSP D
l
, and LIBOR. In the frst two columns, the frst regression does not
control for either measure of the borrowing cost and the second does not control for
Libor. In these regressions, we fnd that the effect of the cost of the access to the bond
market is positive and statistically signifcant for the control group, which is indicative
of the spurious correlation – through correlation of the cost of borrowing over time. As
the third regressions show, once we properly control for the cost of borrowing at the time
of the loan, past cost of borrowing no longer has an effect on the pricing of the loans by
banks that did not access the bond market. In all three regressions, we fnd solid
evidence of the effect of the bond market conditions at the time of the banks’ bond issue
on loan spreads they charge their corporate borrowers – the coeffcient on the interaction
term is positive and statistically signifcant. The sum of the main effect and the
coeffcient on the interaction term is equal to 0.18 and is highly statistically signifcant
(F-statistic is 17.6), indicating that for the treatment group of loans, we continue to
observe the effect of cost of borrowing on the bond market.
Table V.
Falsifcation test
a
Variables (1) (2) (3)
LBBBSP D
b
0.237*** (0.0384) 0.146** (0.0598) 0.103 (0.0640)
BK ACCESS 0.0686* (0.0382) 0.0699* (0.0388) 0.0534 (0.0389)
BK ACCESS ?LBBBSP D
b
0.0844* (0.0462) 0.0885* (0.0463) 0.0776* (0.0457)
LBBBSP D
l
0.0922** (0.0393) 0.115*** (0.0413)
LIBOR ?0.0211*** (0.00775)
Bk fxed effects YES YES YES
Observations 9,555 9,555 9,555
R
2
Adjusted 0.499 0.499 0.501
Notes:
a
Dependent variable is LLOAN SP D which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; LBBBSP D
b
, natural log of the difference between the Moody’s indexes on
the ex ante yields of triple-B- and triple-A-rated bonds at the time of the bank’s most recent bond issue
prior to the loan; LBBBSP D
I
, natural log of the difference between the Moody’s indexes on the ex ante
yields of triple-B- and triple-A-rated bonds at the time of the loan; LIBOR, three-month-level Libor at the
time of the loan. BK ACCESS, dummy variable equal to one for the banks in the sample that issued
bonds; sample limited to the set of banks that issued bonds during the sample period and those
banks that did not issue, but match the former banks on observable characteristics; all models
include the controls used in model 6 of Table III; see defnitions of controls in Table I; robust
standard errors clustered by bank in parentheses; *signifcant at 10 per cent; **signifcant at 5 per
cent; ***signifcant at 1 per cent
289
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
There remains a possibility that borrowers that are more risky in a way that we cannot
observe or control for pushbanks toissue debt duringthe times whenthe costs of borrowing
are high. If this is the case, we will observe more expensive loans made by these banks not
because they are passing on the costs of their bond issues but because this subset of
borrowers is riskier. While we cannot test the differences in unobserved characteristics[22],
we can, nevertheless, see that loans extended by banks whose last public bond was issued
duringthe periodof highborrowingcosts are extendedto borrowers that have similar share
of tangible assets, interest coverage ratio, market-to-bookratio andnet workingcapital ratio
comparedtoborrowers of the banks that either didnot issue at all (fromour matchedsample)
or issuedinthe periodof lowborrowingcosts. Moreover, borrowers of the banks that issued
their previous public bond during high borrowing costs period, have higher proft margin
than other borrowers[23].
3.1.3 Using actual cost of bond fnancing. The results we reported thus far assume
that the cost of bond fnancing for banks is correlated with the spread between Moody’s
indexes of ex ante yields of triple-B- and triple-A-rated bonds on the date of the most recent
bondthe bankissuedprior tothe loan. As we explainedabove, we chose torelyonthese yield
indexes because primary yields are missing for many of the bonds issued by banks in our
sample. The spread between these indexes is likely a good proxy for the cost of bond
fnancingfor our banks because theyare all ratedinvestment grade. Nonetheless, ideallyone
would like to control for a measure of the cost of bond fnancing which is specifc to each
bond issue to account for idiosyncrasies across banks and over time. For this reason, we
re-estimated our model of loan spreads, but this time using the log of the yield on the most
recent bond the bank issued prior to the loan over the one-year Treasury on that date, LBK
BONDYIELDSP D, to measure the cost of bond fnancing for the bank. When we use this
measure, our sample of loans drops from 16,067 to 3,595.
The results of this test are reported in Table VI. Model 1 tests whether the bank’s cost
of bond fnancing, as determined by the yield it paid on its most recent bond issue prior
to the loan, affects the spreads on its corporate loans. Model 2 accounts for the cost to
access the bond market at the time of the loan. As before, we measure this cost by the log
of the spread between Moody’s indexes of ex ante yields of triple-B- and triple-A-rated
Table VI.
Controlling for the yields
on bank bonds
a
Variables (1) (2) (3)
LBK BON D Y IELD 0.0680*** (0.00944) 0.0569*** (0.00882) 0.0480*** (0.0120)
LBBBSP D
l
0.177*** (0.0497) 0.168*** (0.0513)
LIBOR ?0.0127 (0.0100)
Bk fxed effects YES YES YES
Observations 3,629 3,629 3,629
R
2
Adjusted 0.569 0.574 0.575
Notes:
a
Dependent variable is LLOAN SP D which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; LBBBSP D
b
, natural log of the difference between the Moody’s indexes on
the ex ante yields of triple-B- and triple-A-rated bonds at the time of the bank’s most recent bond issue
prior to the loan; LBBBSP D
l
, natural log of the difference between the Moody’s indexes on the ex ante
yields of triple-B- and triple-A-rated bonds at the time of the loan; LIBOR, three-month-level Libor at the
time of the loan; all models include the controls used in model 6 of Table III; see defnitions of controls
in Table I; robust standard errors clustered by bank in parentheses; *signifcant at 10 per cent;
**signifcant at 5 per cent; ***signifcant at 1 per cent
JFEP
6,3
290
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
bonds at the time of the loan, LBBBSP D
l
. Model 3 further controls for the level of Libor
at the time of the loan, LIBOR. As with the previous test, we use our most comprehensive
model of loan spreads, model 6 of Table III, which accounts for our sets of bank-, frm-
and loan-specifc controls, B, F and L, respectively, as well as bank fxed effects. In the
interest of space, however, we do not report the coeffcients on these control variables.
A quick look at Table VI reveals that this test confrms our earlier fnding that when
a bank pays a higher cost to issue in the bond market, it increases the spreads on its
corporate loans. According to the estimates of model 1, when the ex ante yield spread on
a bank’s bond doubles, the bank increases its spreads on the loans it extend
subsequently by 7 per cent (which, given the mean of loan spread of these banks of 158
basis points, corresponds to an increase of about 11 basis points).
The results of this test should lay to rest any concerns that may exist with our use of
the ex ante yield spread between Moody’s indexes of triple-B- and triple-A-rated bonds
to measure the cost of banks’ bond fnancing. These results are important for yet another
potential concern with our use of this measure of the bank’s cost of bond fnancing: the
correlation between the yield spread measured at the time of the bank’s most recent
bond prior to the loan, LBBBSPD
b
, and this same spread at the time of the loan, LBBBSP
D
l
. In our base model, we account for this correlation by controlling for the triple-B
spread at the time of loan, LBBBSP D
l
. Because the correlation between BK BOND
YIELD and LBBBSP D
l
is much lower than the correlation between LBBBSP D
b
and
LBBBSP D
l
, the result of this robustness test further confrms that the effect of LBBBSP
D
b
on loan spreads is attributable to an increase in the cost of the bank’s bond fnancing
and is not the result of an increase in the overall interest rates around the time of the loan.
3.1.4 Loan spread effect when banks issue during good periods and crises periods.
Another way to isolate the effect of the triple-B ex ante yield spread at the time of the
bank’s most recent bond issue prior to the loan, while controlling for the value of this
spread at the time of the loan, is to focus on periods when there was a crisis a bond
market and/or periods when the cost to issue in the bond was extraordinary low. We can
then test whether banks that issued bonds during these periods charged different
spreads on the loans they extended afterward when compared to the other banks that
also rely on bond fnancing but did not issue during these periods or to these same banks
on the loans not immediately following these periods.
To test this hypothesis, we identify the “crises” in the bond market during our sample
period, defned as extended periods of time when the ex ante triple-B over triple-A yield
spread was above one. This criteria left us with fve bond market crises during the sample
period (1987-2007)[24]. We excluded from our sample all loans taken prior to the frst crisis.
Next, we identifed loans for which the most recent public bond was issued by the lender
duringthe crisis as opposedto other loans for whichthe most recent public bondwas issued
by the lender during periods between crises[25]. This allows us to test whether loans that
followed banks’ bond issues during crisis times carried higher spreads when compared to
the loans extended after bank’s bond issues occurred during non-crisis times.
The results of this test are reported in the top panel of Table VII. Models in this panel
are similar to those in Table VI, but now the key variable is BOND CRISIS, a dummy
variable that is equal to one if the bank issued the last bond prior to each loan during the
period of high triple-B spread. As in previous tables, we do not report coeffcients on all
the control variables to save on space. Model 1 investigates whether banks that issued
bonds in the period where the spreads in the bond market were elevated charged higher
291
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
rates on the loans they extended following these bond issues, controlling for our sets of
frm-, loan- and bank-specifc controls as well as bank fxed effects. Model 2 expands this
set of controls to account for the cost to access the bond market at the time of the loan as
determined by the spread between Moody’s indexes of ex ante yields of triple-B- and
triple-A-rated bonds at the time of the loan, LBBBSP D
l
. Model 3 further expands our set
of controls to account for the level of Libor at the time of the loan, LIBOR.
The results of this test also confrm our earlier fndings. In all the models, loans that
followed banks’ bond issues placed during the periods of tight bond market conditions
carried higher spreads than loans that followed bonds issued during a tranquil period.
As before, controlling for the conditions in the bond market at the time of the loan (model
2) and additionally for the level of Libor at the time of the loan (model 3) reduces the
loan-spread elasticity vis-a`-vis our measure of the bank’s cost of funding.
Next, we expand the previous test to investigate whether banks pass onto their
borrowers any savings they enjoy when they issue bonds in periods of unusually low
cost to issue in the bond market. We defne these periods as periods during which the ex
ante triple-B over triple-A yield spread was in the lowest 25 per cent of its
distribution[26]. We follow the approach used in the previous test and defne a dummy
variable, BOND GOOD T IM ES, to isolate the loans that followed banks’ issues of
bonds during these “good times”[27]. We add this variable to the preceding regressions.
Table VII.
Controlling for bonds
banks issue in good and
crises times in the bond
market
a
Variables (1) (2) (3)
Bonds banks issue during crises times in the bond market
BON D CRISIS 0.188*** (0.0321) 0.0413* (0.0234) 0.0587** (0.0244)
LBBBSP D
l
0.228*** (0.0397) 0.169*** (0.0345)
LIBOR ?0.0293*** (0.00651)
Bk fxed effects YES YES YES
Observations 15,200 15,200 15,200
R
2
Adjusted 0.562 0.566 0.568
Bonds banks issue during good and crises times in the bond market
BD CRISIS 0.137*** (0.0296) 0.0446** (0.0222) 0.0603** (0.0238)
BD GOOD T IM ES ?0.128*** (0.0168) ?0.0771*** (0.0210) ?0.0623*** (0.0213)
LBBBSP D
l
0.175*** (0.0442) 0.130*** (0.0387)
LIBOR ?0.0276*** (0.00708)
Bk fxed effects YES YES YES
Observations 15,200 15,200 15,200
R
2
Adjusted 0.565 0.567 0.568
Notes:
a
Dependent variable is LLOAN SP D, which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; BON D CRISIS, dummy variable equal to one if the bank issued its most
recent bond during a crisis period in the bond market; see footnote (24) for the crises in the bond market
during the sample period; BDGOODTIMES, dummy variable equal to one if the bank issued its most
recent bond during a period of extraordinarily low rates in the bond market; see footnote (26) for the
good time periods in the bond market during the sample period. LBBBSPD
l
natural log of the difference
between the Moody’s indexes on the ex ante yields of triple-B- and triple-A-rated bonds at the time of the
loan; LIBOR, three-month-level Libor at the time of the loan; all models include the controls used in
model 6 of Table III; see defnitions of controls in Table I; robust standard errors clustered by bank in
parentheses; *signifcant at 10 per cent; **signifcant at 5 per cent; ***signifcant at 1 per cent
JFEP
6,3
292
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
The results of this test, which are reported in the bottompanel of Table VII, showthat
there is symmetry in the bond market effect on loan spreads. When banks fund
themselves in the bond market at a high cost, they pass a portion of this cost onto their
borrowers; when they are able to fund themselves in the bond market at a very lowcost,
they pass a portion of their “savings” onto their borrowers as well. The coeffcient on
BON D CRISIS is slightly higher than the coeffcient on BON D GOOD T IM ES in
models 1 and 2, and the opposite in model 3, but the difference between these coeffcients
in all models is not statistically signifcant. Once again, these results show that the
effects of the cost banks pay to raise funding in the bond market continue to hold when
we control for the conditions in the bond market at the time of the loan (model 2) and
additionally for the level of Libor at the time of the loan (model 3), further confrming
that they are driven by the cost banks pay to fund themselves in the bond market.
3.2 Other robustness tests
We undertake some additional robustness tests to make sure our results are not driven
by omitted variables. As in the previous section, we use model 6 of Table III, which is our
comprehensive model estimated with bank fxed effects, to do these tests. We do not
report them in the interest of space, but they are available upon request. All of them
confrm that our results continue to hold when we include additional controls.
Despite the large set of factors we account for with our controls, a concern with our
fndings is that we do not account for a potentially important determinant of a bank’s
cost of funds – the cost it incurs to raise deposit funding. We do not consider this cost
because there is no bank-level information on the interest rates banks pay on deposits.
We tried to alleviate these concerns by controlling for the three-month Libor at the time
of the loan, which is the most commonly used proxy for banks’ cost of funding. However,
if this cost is strongly correlated with our proxy for the banks’ cost to issue in the bond
market, this could explain our fndings. To investigate this possibility, we expand our
set of bank controls to account for the cost a bank incurs to raise deposit funding by
interacting with the ratio of deposits to assets with the three-month Libor. This control
variable does not enter the regression signifcantly and does not affect our results.
A more accurate proxy for the cost of deposit funding is the interest expenses on
deposits reported by each bank, but this variable is missing in the Call Reports for 35 per
cent of the observations in our sample. Nonetheless, we used this variable to create two
alternative proxies for the cost of deposit funding. In one case we complemented the
interest expenses on deposits reported by banks with the above proxy; in the other case,
we complemented that variable with total interest expense reported by banks in the Call
Reports[28]. In both cases, our results remain unchanged.
The absence of controls for loan securitization and loan sales could also be a source of
concerns, as banks could use these activities to manage their funding sources. Controlling
properly for loan sales and securitization activity by banks is not an easy task because there
is no bank-level informationonthese activities for most of the sample period[29]. We triedto
account for the effect of loan sales and securitization by controlling for the bank’s
outstanding balance of commercial and industrial (C&I) loans sold and securities (scaled by
the bank’s assets). This variable is never statistically signifcant, probably because it is
available for less thanhalf of the observations inour sample, as Call Reports beganincluding
information on banks’ securitization activity only in 2001:Q3. Controlling for this variable
doesnot affect our fndingontheeffect thebank’scost of bondfnancinghasonloanspreads.
293
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
We also tried to account for these activities by using the Call Report variable “loans held for
sale”. This variable goes back to 1991, but it does not contain separate information on C&I
loans, and it reports information on the loans the bank intends to sell, not on the loans it
effectively sells. Again, controlling for this variable (scaled by the bank’s assets) does not
affect our key fnding. Finally, as securitization activities are relatively more important for
the large banks, we test whether these banks drive our key fnding. Dropping the largest
three banks from the sample does not affect our key fnding, even though these banks
account for about 23 per cent of loans in our sample.
We have attributed the change in banks’ loan pricing policies when the spreads in the
bond market at the time of banks’ bond issuance go up to changes in the cost of banks’
bond fnancing. Could that change in banks’ loan pricing policies instead be driven by an
overall increase in the “price” of risk? The results of our falsifcation test suggest this is
not the case. In addition, our models control for the three-month Libor rate and the
triple-Bbond spread at the time of the loan, which tend to vary with the overall economic
conditions. To further reduce concerns with this hypothesis, we add the gross domestic
product growth rate and the slope of the Treasury yield curve, one at a time, as
additional proxies for the state of the economy and for potential changes in the overall
risk premium. Again, we fnd that these additional controls do not enter signifcantly
and do not affect our results.
Standard errors in our models are clustered by bank. Because many frms took multiple
loans throughout the sample period, the error term in our regression could be correlated
across loans not just for a given bank but also for a given frm. To address this issue, we
follow Petersen (2009) and rerun our core regressions with clustering by frm as well as by
bank[30]. The results of this test showonlyanegligible increase (less thanone per cent) inthe
standard errors, suggesting that clustering by bank only is, in fact, appropriate.
Yet another concern with our fndings is whether they could be driven by unobserved
heterogeneity across borrowers that is correlated with their lenders’ access to the bond
market. For instance, when there is a shock to the supply of bank loans, bank borrowers
with access to the bond market may increase their use of bond fnancing. Similarly,
when there are shocks to the bond market, borrowers with access to this source of
funding could increase their use of bank funding, possibly crowding out lending that
banks would otherwise extend to bank-dependent borrowers. To reduce concerns that
heterogeneity across borrowers drives our key fndings, we re-estimate our loan pricing
model with frm fxed effects as well as with bank-frm pair fxed effects. Our key
fndings remain unchanged.
3.3 How large are the costs to borrowers?
Now that we established that there is a statistically signifcant effect of the cost banks
pay to issue on the bond market on the interest rates they charge on their corporate
loans, we want to assess the economic signifcance of this effect.
Using the results of our benchmark model, model 6 in Table III, we can see that the
elasticity of loan spreads with respect to the bond spread at a time of the bank’s last bond
issue is 0.11. According to the estimates of this model, when the triple-B spread in the
bond market doubles, banks that issue bonds at that time increase spreads on the loans
they extend subsequently by 11 per cent[31]. This change, given the average loan spread
of these banks of 158 basis points, corresponds to an increase of about 18 basis points.
JFEP
6,3
294
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
To get a better intuition for the magnitude of this effect, consider a median-size loan
among those extended by banks that use bond fnancing, a facility of 175 million dollars.
According to the estimates of model 6 in Table III, a frm that borrows from a bank that
issued a bond during a crisis in the bond market, if that frmtakes out a loan in the three
years that follow the bank’s last bond issue, it will pay on average ?$300,000 more per
year than a frm that borrows from a bank that also uses bond fnancing but did not
issue a bond during the crisis. These effects indicate that the link we identifed between
the cost banks pay to issue in the bond market and the interest rates they charge
corporate borrowers is not only statistically signifcant but also economically
meaningful. While the magnitude of this effect is not very large, one has to keep in mind
that all borrowers in our sample are publicly traded. There are many reasons to believe,
including our fndings in the subsequent section of this paper, that these effects would
be larger for smaller private frms.
In sum, the fndings we reported thus far show that banks pass on to their corporate
borrowers a portion of the cost they incur to issue in the bond market. Consequently,
shocks to the cost of banks’ access to the bond market transmit to the cost of bank
lending. According to our estimates, the effect of these shocks on loan spreads is
economically signifcant.
In the next section, we investigate whether all borrowers are equally exposed to these
shocks. We are particularly interested in fnding out whether banks shield their
relationship borrowers from these shocks or whether they instead build on their
informational advantage over these borrowers to pass on to them the bulk of the cost
increase they face to raise funding in the bond market.
4. Do banks pass bond market shocks to all of their borrowers?
The tests we reported in the previous section showthat banks adjust their loan prices in
response to changes in the cost they pay to issue in the bond market. Those tests,
however, do not distinguish between different categories of borrowers. In particular,
they do not distinguish borrowers that have a lending relationship with their bank from
those that do not have such a relationship. As we explained in the introduction, this is
important because banks’ expectation of future business with their relationship
borrowers may lead them to shield these borrowers from the shocks to their funding
costs that arise with their use of the bond market. Alternatively, as banks are likely to
have an information advantage over relationship borrowers, it will be easier for them to
pass any shocks to their funding costs onto relationships borrowers.
To test which of these two effects dominates, we estimate our model (2) of loan
spreads, which extends model (1) to include the interaction of our relationship variable
with our proxy for the cost of banks’ bond fnancing, RELATIONSHIP ?LBBBSP D
b
.
The results of these tests are reported in Table VIII. Model 1 controls for frm-, loan- and
bank-specifc variables, as well as bank fxed effects. We do not report coeffcients on
these controls to save space. Model 2 expands this set of controls to account for the cost
to access the bond market at the time of the loan, LBBBSP D
l
, and model 3 accounts for
the level of Libor at the time of the loan, LIBOR.
We continue to fnd that banks adjust their loan pricing policies in response to changes
in the cost they incur to issue in the bond market. In all three models, the coeffcient on
LBBBSP D
b
, our measure of the bank’s cost to issue in the bond market, is positive and
highly statistically signifcant, indicating that banks increase the loan spreads on their
295
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
borrowersthat donot havearelationshipwiththemwhenthebanks’ cost toissueinthebond
market goes up. The coeffcient on RELAT ION SHIP ?LBBBSP D
b
is always negative,
suggesting that banks may shield their relationship borrowers fromshocks to their funding
costs. This coeffcient, however, is never statistically different fromzero. In other words, our
results show that banks may shield their relationship borrowers (when compared to
nonpartisanshipborrowers) fromshocks to their fundingcosts, but byanamount that is not
statistically different from zero.
4.1 Relationship lending and bank dependency
The evidence we just presented indicates that we cannot reject the hypothesis that
banks do not give “special” treatment to their relationship borrowers vis-a`-vis their
nonpartisanship borrowers when they need to raise loan spreads to make up for the
additional cost they incur when it becomes more expensive to raise bond fnancing. Our
results, however, also showthat banks do not pass the entirety of the shocks to their cost
of bond fnancing onto their borrowers. This combination of results poses an interesting
question, as banks do not offer special treatment to their relationship borrowers, why
they do not use the informational advantage they are likely to have over these borrowers
to pass onto to them the bulk of the shock to their cost of bond fnancing?
This apparent puzzle could arise because, among the relationship borrowers, some
are dependent on banks for funding while others are not. Banks’ may have more
incentives to shield relationship borrowers that are dependent on them for external
funding more than relationship borrowers that have access to other sources of funding,
such as the bond market. This is because banks are more likely to recover the “subsidy”
from future business with the former borrowers. On the other hand, banks will fnd it
easier to pass on the additional cost to they face in the bond market onto the
bank-dependent relationship borrowers because relationship borrowers that do not
Table VIII.
Relationship borrowers
and bank bond costs
a
Variables (1) (2) (3)
LBBBSP D
b
0.317*** (0.0315) 0.186*** (0.0425) 0.144*** (0.0459)
RELAT ION SHIP ?0.0300 (0.0258) ?0.0289 (0.0258) ?0.0302 (0.0257)
RELAT ION SHIP ?LBBBSP D
b
?0.0526 (0.0440) ?0.0556 (0.0431) ?0.0520 (0.0433)
LBBBSP D
l
0.142*** (0.0423) 0.146*** (0.0411)
LIBOR
?0.0198***
(0.00572)
Bk fxed effects YES YES YES
Observations 16067 16067 16067
R
2
Adjusted 0.559 0.560 0.561
Notes:
a
Dependent variable is LLOAN SP D, which is the natural log of the all-in-drawn loan spread
over LIBOR at origination; LBBBSP D
b
, natural log of the difference between the Moody’s indexes on
the ex ante yields of triple-B- and triple-A-rated bonds at the time of the bank’s most recent bond issue
prior to the loan; LBBBSP D
l
, natural log of the difference between the Moody’s indexes on the ex ante
yields of triple-B- and triple-A-rated bonds at the time of the loan; LIBOR, three-month-level Libor at the
time of the loan. RELAT ION SHIP refers to three-year horizon in columns (1)-(3) and one-year horizon
in columns (4)-(6); all models include the controls used in model 6 of Table III; see defnitions of controls
in Table I; robust standard errors clustered by bank in parentheses; *signifcant at 10 per cent;
**signifcant at 5 per cent; ***signifcant at 1 per cent
JFEP
6,3
296
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
depend on their bank for funding may respond to any attempt of an increase in their loan
rates by looking for funding elsewhere.
To test which of these predictions is borne out in the data, we estimate our loan
pricing model separately for relationship borrowers and borrowers with no lending
relationship with their bank. Further, we modify this model to distinguish whether the
borrower is bank dependent or not, and interact this dummy variable with our proxy for
the cost the bank incurred the last time it issue in the bond market prior to the loan. We
identify borrowers as bank dependent if they never issued in the public bond market or
issued a bond more than three years prior to the loan[32]. In addition, as according to
Rajan (1992), the holdup problemis more pronounced for risky frms than for safe frms,
we also consider a specifcation in which we distinguish among the frms that have
access to the bond market those that are rated investment grade from those that are
rated below investment grade.
The results of these tests for the loans of relationship borrowers are reported in
Table IX. Model 1 investigates the effect of LBBBSP D
b
on loan spreads controlling for
our sets of frm-, loan- and bank-specifc factors, as well as bank fxed effects. Model 2
adds to this model the cost to access the bond market at the time of the loan, LBBBSSP
D
l
, while model 3 further adds to our controls the LIBOR at the time of the loan. Models
4 through 6, in turn, investigate whether relationship borrowers that have access to the
bond market are less exposed to their banks’ cost of bond fnancing. As with the
previous set of models, model 4 investigates this hypothesis controlling for our sets of
frm-, loan- and bank-specifc factors; model 5 adds to these controls the cost to access
the bond market at the time of the loan, and model 6 further adds the Libor at the time of
the loan. Finally, models 7 through 9 follow the same pattern, but distinguish among
relationship borrowers that have access to the bond market those that are rated
investment grade from those that are rated below grade. For completeness, Table X
reports the results of these same tests but for borrowers that do not have a lending
relationship with their bank. In the interest of space, we report in these two tables only
coeffcients on variables that are key to our hypotheses, leaving out the coeffcients on
our sets of frm, loan and bank controls.
Focusing on Table IX, we see that models 1-3 confrm the fndings we reported in
Table VIII – when the bank’s cost to issue in the bond market goes up, the bank passes
a portion of this cost hike onto the borrowers it has a lending relationship with. Models
4-6 show that not all relationship borrowers are exposed to the cost of the bank’s bond
fnancing. Among loans of relationship borrowers, model 4 shows the loan-spread
elasticity to the cost of the bank’s bond fnancing is 0.30 for loans of bank-dependent
borrowers and only 0.21 for loans of borrowers with access to the bond market. These
elasticities decline as we expand our controls to account for the cost to access the bond
market at the time of the loan (model 5) and to account for the level of Libor at the time
of the loan (model 6). According to model 6, the aforementioned elasticities are 0.14 and
0.04, respectively. Importantly, the former elasticity is statistically different from zero,
but we cannot reject the hypothesis that the latter elasticity is equal to zero, according to
the F-test (p-value is 0.47). In other words, when banks’ cost of bond fnancing goes up,
they passes a portion of this cost increase onto their relationship borrowers that do not
have access to the bond market. On these occasions, they also raise the spreads on their
loans to those relationship borrowers that have access to the bond market, but by an
amount that is smaller and is not statistically different from zero.
297
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table IX.
Bank cost of bond
fnancing and loan
spreads the bank charges
relationship borrowers
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
(
7
)
(
8
)
(
9
)
$
L
B
B
B
S
P
D
_
b
$
0
.
2
6
4
*
*
*
(
0
.
0
5
0
9
)
0
.
1
4
9
*
*
(
0
.
0
6
3
2
)
0
.
1
0
8
*
(
0
.
0
6
3
0
)
0
.
3
0
3
*
*
*
(
0
.
0
4
9
0
)
0
.
1
8
3
*
*
*
(
0
.
0
6
3
1
)
0
.
1
4
0
*
*
(
0
.
0
6
3
9
)
0
.
2
9
6
*
*
*
(
0
.
0
5
0
1
)
0
.
1
7
0
*
*
(
0
.
0
6
5
0
)
0
.
1
3
1
*
*
(
0
.
0
6
5
9
)
$
P
B
O
N
D
$
?
0
.
0
8
8
5
*
*
*
(
0
.
0
1
6
2
)
?
0
.
0
8
9
3
*
*
*
(
0
.
0
1
6
2
)
?
0
.
0
9
4
0
*
*
*
(
0
.
0
1
6
6
)
$
P
B
O
N
D
$
?
$
L
B
B
B
S
P
D
_
b
$
?
0
.
0
9
0
9
*
*
*
(
0
.
0
2
5
1
)
?
0
.
0
9
2
4
*
*
*
(
0
.
0
2
5
2
)
?
0
.
0
9
7
5
*
*
*
(
0
.
0
2
4
6
)
$
P
B
O
N
D
\
;
I
G
$
?
0
.
2
1
8
*
*
*
(
0
.
0
2
1
8
)
?
0
.
2
1
9
*
*
*
(
0
.
0
2
1
8
)
?
0
.
2
2
2
*
*
*
(
0
.
0
2
1
3
)
$
P
B
O
N
D
\
;
B
G
$
0
.
2
0
1
*
*
*
(
0
.
0
2
7
7
)
0
.
2
0
1
*
*
*
(
0
.
0
2
7
8
)
0
.
1
9
4
*
*
*
(
0
.
0
2
7
3
)
$
P
B
O
N
D
\
;
I
G
$
?
$
L
B
B
B
S
O
D
_
b
$
?
0
.
0
8
4
5
*
*
*
(
0
.
0
2
3
2
)
?
0
.
0
8
7
9
*
*
*
(
0
.
0
2
3
2
)
?
0
.
0
9
2
5
*
*
*
(
0
.
0
2
2
6
)
$
P
B
O
N
D
\
;
B
G
$
?
$
L
B
B
B
S
O
D
_
b
$
0
.
0
3
3
1
(
0
.
0
6
6
7
)
0
.
0
3
6
9
(
0
.
0
6
7
7
)
0
.
0
3
0
6
(
0
.
0
6
6
7
)
$
L
B
B
B
S
P
D
_
l
$
0
.
1
2
1
*
*
(
0
.
0
5
2
2
)
0
.
1
2
3
*
*
(
0
.
0
4
8
6
)
0
.
1
2
6
*
*
(
0
.
0
5
0
7
)
0
.
1
2
9
*
*
*
(
0
.
0
4
6
6
)
0
.
1
3
3
*
*
*
(
0
.
0
5
0
0
)
0
.
1
3
6
*
*
*
(
0
.
0
4
6
5
)
L
I
B
O
R
?
0
.
0
2
2
6
*
*
*
(
0
.
0
0
5
5
2
)
?
0
.
0
2
4
4
*
*
*
(
0
.
0
0
5
3
5
)
?
0
.
0
2
2
1
*
*
*
(
0
.
0
0
5
3
6
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
1
0
,
2
1
2
R
$
^
2
$
A
d
j
u
s
t
e
d
0
.
5
7
8
0
.
5
7
8
0
.
5
8
0
0
.
5
8
0
0
.
5
8
0
0
.
5
8
2
0
.
5
9
4
0
.
5
9
5
0
.
5
9
6
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
P
B
O
N
D
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
I
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
B
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
b
e
l
o
w
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
E
L
A
T
I
O
N
S
H
I
P
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
;
S
a
m
p
l
e
l
i
m
i
t
e
d
t
o
b
o
r
r
o
w
e
r
s
w
i
t
h
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
i
r
b
a
n
k
(
R
E
L
A
T
I
O
N
S
H
I
P
?
1
.
)
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
JFEP
6,3
298
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table X.
Bank cost of bond
fnancing and loan
spreads the bank charges
non-relationship
borrowers
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
(
7
)
(
8
)
(
9
)
L
B
B
B
S
P
D
b
0
.
3
1
4
*
*
*
(
0
.
0
3
4
9
)
0
.
1
6
0
*
*
(
0
.
0
6
7
9
)
0
.
1
2
0
*
(
0
.
0
7
3
4
)
0
.
3
1
2
*
*
*
(
0
.
0
4
2
7
)
0
.
1
6
0
*
*
(
0
.
0
6
6
8
)
0
.
1
1
9
(
0
.
0
7
2
2
)
0
.
3
1
3
*
*
*
(
0
.
0
4
2
9
)
0
.
1
4
1
*
*
(
0
.
0
6
2
2
)
0
.
1
0
3
(
0
.
0
6
7
2
)
P
B
O
N
D
?
0
.
0
6
1
8
*
*
(
0
.
0
2
6
9
)
?
0
.
0
6
1
7
*
*
(
0
.
0
2
7
0
)
?
0
.
0
6
6
8
*
*
(
0
.
0
2
7
9
)
P
B
O
N
D
?
L
B
B
B
S
P
D
b
0
.
0
1
1
8
(
0
.
0
4
8
9
)
0
.
0
0
9
7
0
(
0
.
0
4
9
8
)
0
.
0
0
1
1
7
(
0
.
0
5
0
3
)
P
B
O
N
D
I
G
?
0
.
1
8
9
*
*
*
(
0
.
0
3
0
9
)
?
0
.
1
9
0
*
*
*
(
0
.
0
3
0
9
)
?
0
.
1
9
4
*
*
*
(
0
.
0
3
1
7
)
P
B
O
N
D
B
G
0
.
1
9
9
*
*
*
(
0
.
0
3
1
3
)
0
.
2
0
3
*
*
*
(
0
.
0
3
2
2
)
0
.
1
9
7
*
*
*
(
0
.
0
3
2
8
)
P
B
O
N
D
I
G
?
L
B
B
B
S
O
D
b
0
.
0
2
5
9
(
0
.
0
5
8
5
)
0
.
0
2
3
3
(
0
.
0
5
9
8
)
0
.
0
1
5
0
(
0
.
0
6
0
9
)
P
B
O
N
D
B
G
?
L
B
B
B
S
O
D
b
0
.
0
9
4
2
(
0
.
0
7
4
3
)
0
.
0
9
3
7
(
0
.
0
7
5
1
)
0
.
0
8
5
8
(
0
.
0
7
4
9
)
L
B
B
B
S
P
D
l
0
.
1
6
7
*
*
(
0
.
0
7
1
8
)
0
.
1
7
4
*
*
(
0
.
0
7
2
9
)
0
.
1
6
6
*
*
(
0
.
0
7
2
1
)
0
.
1
7
4
*
*
(
0
.
0
7
3
4
)
0
.
1
8
6
*
*
*
(
0
.
0
6
6
9
)
0
.
1
9
4
*
*
*
(
0
.
0
6
8
0
)
L
I
B
O
R
?
0
.
0
1
7
3
*
*
*
(
0
.
0
0
5
0
0
)
?
0
.
0
1
8
6
*
*
*
(
0
.
0
0
5
0
6
)
?
0
.
0
1
7
3
*
*
*
(
0
.
0
0
5
0
1
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
5
8
5
5
R
2
A
d
j
u
s
t
e
d
0
.
5
3
4
0
.
5
3
5
0
.
5
3
6
0
.
5
3
5
0
.
5
3
6
0
.
5
3
6
0
.
5
4
7
0
.
5
4
8
0
.
5
4
9
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
P
B
O
N
D
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
I
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
P
B
O
N
D
B
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
b
e
l
o
w
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
p
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
E
L
A
T
I
O
N
S
H
I
P
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
;
S
a
m
p
l
e
l
i
m
i
t
e
d
t
o
b
o
r
r
o
w
e
r
s
w
i
t
h
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
i
r
b
a
n
k
(
R
E
L
A
T
I
O
N
S
H
I
P
?
0
.
)
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
299
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
In models 7-9, we further split relationship borrowers with access to the bond market
into those that are rated investment grade and those that are rated below grade
(according to the rating of their most recent bond issue prior to the loan). Consistent with
the insights of Rajan (1992) we fnd that risky borrowers that have access to the bond
market are more exposed to the cost of banks’ bond fnancing than safe borrowers that
have access to the bond market. Among relationship borrowers, those that have access
to the bond market but are risky, pay a premium on their loans which is similar to that
paid by borrowers that do not have access to the bond market in response to a shock to
their bank’s cost of bond fnancing. This could be because on these occasions, risky
borrowers lose their access to the bond market and, in essence, become dependent on
banks for external funds. In contrast, relationship borrowers that have access to the
bond market and are safe, pay a lower premium than relationship borrowers that are
dependent on banks for funding. In fact, in the case of models 8 and 9, we cannot reject
the hypothesis that safe borrowers do not pay any premium.
Turning our attention to Table X, which reports the same tests but for the sample of
borrowers that do not have a lending relationship with their bank, we see one important
difference with respect to the results we just discussed for relationship borrowers.
Among the non-relationship borrowers, we do not fnd any evidence that banks charge
different premiums associated with their cost of bond fnancing to those borrowers that
do not have access to the bond market and those that have access to the bond market,
regardless of the rating of their bond. This is reassuring, as it shows that it is not critical
for non-relationship borrowers to have access to the bond market, possibly because
banks do not have an informational advantage over them.
Although not as apparent, because of the way we organize our results, there is one other
piece of evidence which suggests that access to the bond market is more important for
relationship borrowers than non-relationship borrowers. If one compares the coeffcient
on LBBBSPD
b
for benchmark, bank-dependent, category of relationship borrowers (line
1 column 9 of Table IX) and of non-relationship borrowers (line 1 column 9 of Table X),
we can see that the former are more sensitive than the latter borrowers. True, this
difference is not very large and is not statistically signifcant, but it points in the
direction that among relationship borrowers, those that do not have access to the bond
market, are more exposed to shocks to banks’ cost of bond funding than
non-relationships borrowers that also do not have access to the bond market.
The results of Tables IX and X suggest that banks pass the largest portion of shocks
to their cost of bond fnancing onto their relationships borrowers that do not have access
to the bond market, and they “protect” the most their relationship borrowers that have
access to the bond market, particularly those that are safe. These fndings are important
in that they do not support the hypothesis that banks take into account the prospects of
future business with their relationship borrowers and smooth the interest rates they
charge them over time, as, in this case, we would expect relationship borrowers that are
bank dependent to receive the largest level of protection. In contrast, our fndings
suggest that the market power resulting from the informational advantage that banks
have over their borrowers drives their loan pricing policies.
The next two tests aim at further confrming these fndings. The frst test furthers
our investigation of how banks pass the shocks to their cost of bond fnancing onto
different categories of borrowers. We report the results of this test in Table XI. The
left-hand columns compare the portion of the shocks to the cost of bond fnancing
JFEP
6,3
300
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table XI.
Bank cost of bond
fnancing and loan
spreads of relationship
borrowers
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
L
B
B
B
S
P
D
b
0
.
3
0
7
*
*
*
(
0
.
0
3
5
6
)
0
.
1
7
1
*
*
*
(
0
.
0
3
6
9
)
0
.
1
2
7
*
*
*
(
0
.
0
4
1
8
)
0
.
2
6
6
*
*
*
(
0
.
0
3
4
9
)
0
.
1
3
0
*
*
*
(
0
.
0
3
4
6
)
0
.
0
8
6
3
*
*
(
0
.
0
3
7
8
)
R
L
P
B
O
N
D
I
G
?
0
.
2
6
8
*
*
*
(
0
.
0
2
3
1
)
?
0
.
2
6
8
*
*
*
(
0
.
0
2
2
9
)
?
0
.
2
7
1
*
*
*
(
0
.
0
2
3
2
)
R
L
P
B
O
N
D
I
G
?
L
B
B
B
S
P
D
b
?
0
.
0
9
9
1
*
*
*
(
0
.
0
2
8
3
)
?
0
.
1
0
4
*
*
*
(
0
.
0
2
7
6
)
?
0
.
1
0
5
*
*
*
(
0
.
0
2
7
4
)
R
L
N
O
A
C
C
E
S
S
0
.
0
5
5
0
*
*
*
(
0
.
0
0
9
7
3
)
0
.
0
5
6
1
*
*
*
(
0
.
0
0
9
6
0
)
0
.
0
5
7
8
*
*
*
(
0
.
0
0
9
1
5
)
R
L
N
O
A
C
C
E
S
S
?
L
B
B
B
S
P
D
b
0
.
0
4
6
6
*
(
0
.
0
2
5
4
)
0
.
0
4
5
8
*
*
(
0
.
0
2
4
9
)
0
.
0
5
0
7
*
*
(
0
.
0
2
5
0
)
L
B
B
B
S
P
D
l
0
.
1
4
6
*
*
*
(
0
.
0
4
0
9
)
0
.
1
5
2
*
*
*
(
0
.
0
3
9
5
)
0
.
1
4
5
*
*
*
(
0
.
0
4
0
6
)
0
.
1
5
0
*
*
*
(
0
.
0
3
9
7
)
L
I
B
O
R
?
0
.
0
2
1
4
*
*
*
(
0
.
0
0
5
2
4
)
?
0
.
0
2
0
3
*
*
*
(
0
.
0
0
5
7
3
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
1
6
,
0
6
7
R
2
A
d
j
u
s
t
e
d
0
.
5
7
0
0
.
5
7
0
0
.
5
7
2
0
.
5
5
9
0
.
5
6
0
0
.
5
6
1
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
R
L
P
B
O
N
D
I
G
D
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
a
n
d
i
t
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
l
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
L
N
O
A
C
C
E
S
S
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
,
a
n
d
i
t
h
a
s
n
e
v
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
o
r
d
i
d
i
t
o
n
l
y
m
o
r
e
t
h
a
n
t
h
r
e
e
y
e
a
r
s
a
g
o
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
301
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
that banks pass onto their relationship borrowers that have access to the bond
market and are safe vis-a`-vis the portion of the shocks that they pass onto their
remaining borrowers. The right-hand columns, in turn, compare the portion of these
shocks that banks pass onto their relationship borrowers that are dependent on
them for external funding vis-a`-vis the portion of the shocks that they pass onto their
remaining borrowers. Consistent with our earlier fndings, the differences in
elasticity of loan spreads with respect to the cost banks pay to issue in the bond
market (that is, the coeffcients on the interaction terms) are negative and signifcant
in left-hand side models, but positive and signifcant in right-hand side models. In
other words, banks do not pass the shocks to their cost of bond fnancing to their
relationship borrowers that have access to the bond market, but they do it to their
relationship borrowers that depend on them for funding.
According to our results, banks do not pass the entirety of the shock to their cost of
funding onto their relationship borrowers that do not have access to the bond market.
This could be because these borrowers are not fully dependent on banks. Because these
borrowers are publicly listed, they are likely to be less dependent on banks than
privately held borrowers that do not have access to the bond market because there will
be even less information available on the latter borrowers. We do not include privately
held borrowers in our sample because Compustat, our source of frm-level data, does not
include information for these borrowers.
Our fnal test takes this analysis another step further by comparing the costs and the
savings banks pass onto dependent and non-dependent relationship borrowers when
they issue in periods of crisis and in periods of low spreads in the bond market,
respectively. To that end, we replace our proxy for the cost the bank paid when it issued
its last bond prior to the loan, LBBBSP D
b
, with the two dummy variables we defned in
section 3.1.4 to identify the loans banks extended after issuing bonds in periods of crisis
in the bond market, BOND CRISIS, and those they extended after issuing bonds in
periods of lowspreads in the bond market, BONDGOODTIMES, respectively. Models
1-3 compare how banks pass the costs and the savings from their bond issues on these
two occasions onto their relationship borrowers that have access to the bond market and
are safe vis-a`-vis the remaining borrowers. Models 4-6 repeat this analysis, but, in this
case, we isolate borrowers that have a lending relationship with the bank and do not
have access to the bond market.
The difference inthe treatment that banks offer to these two sets of borrowers is striking.
Bothsets of borrowers enjoythe same savings as everyone else whenbanks lower loanrates
following their bond issues at low cost[33]. In contrast, when banks raise bond fnancing at
a high cost, they do not raise loan rates on their relationship borrowers that are not
dependent on them, but do raise loan rates on their relationship borrowers that are
dependent on them for external funding, by an amount that is statistically different from
zero[34]. In other words, dependent borrowers are more exposed to shocks to banks
funding costs than nondependent borrowers. Further, our evidence shows that the
former borrowers are more exposed to shocks that raise the cost of banks’ bond
fnancing than to shocks that lower the cost of this funding source for banks.
According to model 6 of Table XII, when banks raise loan spreads on their
dependent relationship borrowers following a bond issue during crisis times, they
raise these rates on average by 10 per cent. When banks lower loan spreads
following a bond issue during good times, they lower these rates on average by 5.3
JFEP
6,3
302
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Table XII.
Bank cost of bond
fnancing and loan
spreads of relationship
borrowers: Further
analysis
a
V
a
r
i
a
b
l
e
s
(
1
)
(
2
)
(
3
)
(
4
)
(
5
)
(
6
)
B
O
N
D
C
R
I
S
I
S
0
.
1
5
4
*
*
*
(
0
.
0
2
5
1
)
0
.
0
6
2
9
*
*
*
(
0
.
0
1
9
4
)
0
.
0
7
8
4
*
*
*
(
0
.
0
1
9
7
)
0
.
1
1
6
*
*
*
(
0
.
0
2
6
4
)
0
.
0
2
2
3
(
0
.
0
2
5
4
)
0
.
0
3
6
3
(
0
.
0
2
5
4
)
B
O
N
D
G
O
O
D
T
I
M
E
S
?
0
.
1
3
3
*
*
*
(
0
.
0
1
6
0
)
?
0
.
0
8
2
9
*
*
*
(
0
.
0
2
1
3
)
?
0
.
0
6
7
6
*
*
*
(
0
.
0
2
1
9
)
?
0
.
1
3
1
*
*
*
(
0
.
0
2
0
3
)
?
0
.
0
7
9
0
*
*
*
(
0
.
0
2
3
7
)
?
0
.
0
6
4
0
*
*
*
(
0
.
0
2
3
7
)
R
L
P
B
O
N
D
I
G
?
0
.
2
3
5
*
*
*
(
0
.
0
2
6
5
)
?
0
.
2
3
3
*
*
*
(
0
.
0
2
6
1
)
?
0
.
2
3
6
*
*
*
(
0
.
0
2
7
7
)
R
L
P
B
O
N
D
I
G
?
C
R
I
S
I
S
?
0
.
0
5
7
5
*
*
(
0
.
0
2
5
0
)
?
0
.
0
6
1
1
*
*
(
0
.
0
2
5
5
)
?
0
.
0
5
8
4
*
*
(
0
.
0
2
4
1
)
R
L
P
B
O
N
D
I
G
?
G
O
O
D
T
I
M
E
S
0
.
0
0
9
4
3
(
0
.
0
4
2
6
)
0
.
0
1
1
0
(
0
.
0
4
2
8
)
0
.
0
0
9
5
9
(
0
.
0
4
2
2
)
R
L
N
O
A
C
C
E
S
S
0
.
0
1
8
4
(
0
.
0
2
1
8
)
0
.
0
1
9
2
(
0
.
0
2
1
1
)
0
.
0
1
7
8
(
0
.
0
2
2
3
)
R
L
N
O
A
C
C
E
S
S
?
C
R
I
S
I
S
0
.
0
5
9
3
*
(
0
.
0
3
3
6
)
0
.
0
5
9
1
*
(
0
.
0
3
2
5
)
0
.
0
6
3
5
*
(
0
.
0
3
3
6
)
R
L
N
O
A
C
C
E
S
S
?
G
O
O
D
T
I
M
E
S
0
.
0
1
2
9
(
0
.
0
2
2
3
)
0
.
0
1
1
1
(
0
.
0
2
1
9
)
0
.
0
1
1
0
(
0
.
0
2
2
9
)
L
B
B
B
B
l
0
.
1
7
3
*
*
*
(
0
.
0
4
3
1
)
0
.
1
2
6
*
*
*
(
0
.
0
3
7
6
)
0
.
1
7
7
*
*
*
(
0
.
0
4
3
1
)
0
.
1
3
2
*
*
*
(
0
.
0
3
7
0
)
L
I
B
O
R
?
0
.
0
2
8
1
*
*
*
(
0
.
0
0
6
7
8
)
?
0
.
0
2
7
6
*
*
*
(
0
.
0
0
7
1
6
)
B
k
f
x
e
d
e
f
f
e
c
t
s
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
Y
E
S
O
b
s
e
r
v
a
t
i
o
n
s
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
1
5
,
2
0
0
R
2
A
d
j
u
s
t
e
d
0
.
5
7
6
0
.
5
7
7
0
.
5
7
9
0
.
5
6
6
0
.
5
6
7
0
.
5
6
9
N
o
t
e
s
:
a
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
L
L
O
A
N
S
P
D
i
s
t
h
e
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
a
l
l
-
i
n
-
d
r
a
w
n
l
o
a
n
s
p
r
e
a
d
o
v
e
r
L
I
B
O
R
a
t
o
r
i
g
i
n
a
t
i
o
n
;
B
O
N
D
C
R
I
S
I
S
a
n
d
B
O
N
D
G
O
O
D
T
I
M
E
S
a
r
e
d
e
f
n
e
d
a
s
i
n
T
a
b
l
e
V
I
I
.
L
B
B
B
S
P
D
b
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
b
a
n
k
’
s
m
o
s
t
r
e
c
e
n
t
b
o
n
d
i
s
s
u
e
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
L
B
B
B
S
P
D
l
,
n
a
t
u
r
a
l
l
o
g
o
f
t
h
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
e
e
n
t
h
e
M
o
o
d
y
’
s
i
n
d
e
x
e
s
o
n
t
h
e
e
x
a
n
t
e
y
i
e
l
d
s
o
f
t
r
i
p
l
e
-
B
-
a
n
d
t
r
i
p
l
e
-
A
-
r
a
t
e
d
b
o
n
d
s
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
;
L
I
B
O
R
,
t
h
r
e
e
-
m
o
n
t
h
-
l
e
v
e
l
L
i
b
o
r
a
t
t
h
e
t
i
m
e
o
f
t
h
e
l
o
a
n
.
R
L
P
B
O
N
D
I
G
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
,
a
n
d
i
t
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
w
h
i
c
h
w
a
s
r
a
t
e
d
i
n
v
e
s
t
m
e
n
t
g
r
a
d
e
i
n
t
h
e
l
a
s
t
t
h
r
e
e
y
e
a
r
s
p
r
i
o
r
t
o
t
h
e
l
o
a
n
;
R
L
N
O
A
C
C
E
S
S
,
d
u
m
m
y
v
a
r
i
a
b
l
e
e
q
u
a
l
t
o
1
i
f
t
h
e
b
o
r
r
o
w
e
r
h
a
s
a
l
e
n
d
i
n
g
r
e
l
a
t
i
o
n
s
h
i
p
w
i
t
h
t
h
e
l
e
n
d
e
r
(
t
h
e
b
o
r
r
o
w
e
r
t
o
o
k
o
u
t
a
t
l
e
a
s
t
o
n
e
l
o
a
n
f
r
o
m
t
h
e
l
e
n
d
e
r
o
f
t
h
e
c
u
r
r
e
n
t
l
o
a
n
i
n
t
h
e
p
r
e
v
i
o
u
s
t
h
r
e
e
y
e
a
r
s
)
,
a
n
d
i
t
h
a
s
n
e
v
e
r
i
s
s
u
e
d
a
p
u
b
l
i
c
b
o
n
d
o
r
d
i
d
i
t
o
n
l
y
m
o
r
e
t
h
a
n
t
h
r
e
e
y
e
a
r
s
a
g
o
;
a
l
l
m
o
d
e
l
s
i
n
c
l
u
d
e
t
h
e
c
o
n
t
r
o
l
s
u
s
e
d
i
n
m
o
d
e
l
6
o
f
T
a
b
l
e
I
I
I
;
s
e
e
d
e
f
n
i
t
i
o
n
s
o
f
c
o
n
t
r
o
l
s
i
n
T
a
b
l
e
I
;
r
o
b
u
s
t
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
b
y
b
a
n
k
i
n
p
a
r
e
n
t
h
e
s
e
s
;
*
s
i
g
n
i
f
c
a
n
t
a
t
1
0
p
e
r
c
e
n
t
;
*
*
s
i
g
n
i
f
c
a
n
t
a
t
5
p
e
r
c
e
n
t
;
*
*
*
s
i
g
n
i
f
c
a
n
t
a
t
1
p
e
r
c
e
n
t
303
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
for relationship dependent[35]. These fndings confrm that banks’ loan pricing
policies are driven by informational advantage more than by the prospects of future
business with borrowers.
In sum, the results we unveiled in this section confrm that banks adjust their loan
rates in response to shocks to the cost they pay on their bond issues. The results also
show that establishing a relationship lending with a bank does not guarantee a special
treatment of the borrower. When banks are able to issue bonds at very low rates, they
pass some of their cost savings to their relationship borrowers regardless of whether
they are dependent on them. However, when banks raise bond fnancing at high rates,
they do not pass any of these costs to the relationship borrowers that are not dependent
on them, but they do so to their relationship borrowers that depend on themfor funding.
These results, therefore, show that the market power resulting from the informational
advantage that banks have over borrowers more than the prospects of future business
drives their loan pricing policies.
4.1.1 Robustness tests. We undertook a set of robustness tests similar to those we
report in the previous section to investigate the robustness of our fndings on the
differential response of banks vis-a`-vis their relationship borrowers and their
non-relationship borrowers. In the interest of space, we do not report these results, but
describe them briefy.
First, we consider two subsamples as before, the frst half of our sample time
period, from 1988 to 1997, and the second half, from 1998 to 2007. For the full set of
borrowers, when we allow the effect of the cost of bond funding by banks on their
loan pricing to vary for relationship and non-relationship borrowers (as in
Table VIII), we fnd that in the early half of the sample, relationship borrowers paid
on average higher interest on their loans, and their loans were more sensitive to the
interest rate the banks paid on their bonds, but the overall effect of bank bond
issuance cost on the loan pricing is not statistically different for these two sets of
borrowers[36]. The results for the late half of our sample are almost identical to
those in Table VIII, indicating that it is the second half of the sample that drives our
main results. When we limit the sample to relationship borrowers, as in Table IX, we
lose the precision of our estimates in both subsamples, but qualitatively the story
remains unchanged. The results for non-relationship borrowers (Table X) are the
sample for the full, early and late sample periods.
Next, we re-estimates the regressions reported in Tables VIII–X with the actual
spread on the last bond issued by the bank prior to any given loan, as we did in Table VI.
Even though using these spreads leaves out a lot of observations, we continue to fnd the
results of Table VIII unchanged with this modifcation. Once we split the sample,
however, as we do in Tables IX and X, we do not fnd statistically signifcant difference
in the response of loan prices to the change in bank’s bond spreads depending on
whether the borrower has access to the bond market and the rating of their most recent
bond. This is not surprising, given that the number of observations falls from 10,212 to
2,336 and from5,855 to 1,293 in Tables IXand X, respectively, with only a small portion
of these observations accounted for by borrowers that had issued an investment grade
bond in the three years prior to the loan (676 loans of relationship borrowers and 241
loans of non-relationship borrowers).
Next, we repeat the analysis with various controls for the amount of the bond issue,
cost of deposits, and the general economic conditions. In this case, all of our results in
JFEP
6,3
304
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Tables VIII-X remain almost identical to the ones in the regressions we reported. The
only exception is, when we control for bank’s outstanding balance of C&I loans sold and
securities, which dramatically limits our sample, we lose statistical signifcance of the
effect of the interaction term of borrower issuing ING bond in the regression for
relationship borrowers (as in Table IX). We recover all of our results, however, when
instead we control for loans sold, which allows us to retain a larger sample. Neither of
these controls enters signifcantly in Table IX regressions, while loan sales enter
signifcantly (with a positive sign) in the regression of Table X, which limits the sample
to non-relationship borrowers. In this case, however, including either of the two controls
does not affect the coeffcients of interest.
Finally, we attempt different specifcations – with clustering standard errors both on
bank and frm as well as with frm fxed effect. All of our results are robust to these
modifcations. Aminor difference is that the main effect in Table Xbecomes statistically
signifcant at the 10 per cent level (but the same in magnitude) when we cluster standard
errors on both bank and frm.
5. Final remarks
Our fndings show that banks’ use of bond fnancing creates a link between the
conditions in the bond market and their loan pricing policy. The evidence we uncovered
on this link further shows that banks do not offer special protection to their relationship
borrowers. On the contrary, banks expose their relationships borrowers that depend on
them for funding the most to the shocks to their cost of bond fnancing. Banks protect
from the bond market shocks only their relationship borrowers that have access to the
bond market and are safe, possibly because they cannot hold themup for higher interest
rates. The fact that banks pass some savings onto the latter borrowers but not to the
former borrowers when they raise funding in the bond market at extraordinarily low
cost adds further support to our conclusion that market power drives banks loan pricing
policies more than relationship aspects of lending.
These fndings are novel and have important implications. Because the number of
banks that rely on bond fnancing continues to grow, our fndings indicate that fnancial
intermediation through banks will become increasingly interlinked with the
intermediation performed through fnancial markets. In addition, corporate borrowers,
in particular those that are dependent on banks for external funding, will become
increasingly exposed to adverse shocks to the bond market. Moreover, a policy push
toward longer-term bank fnancing is likely to further increase banks’ reliance on the
bond market, leading to unintended consequences of increasing the exposure of
bank-dependent borrowers to the bond market shocks.
These fnding suggest some potential ideas for future research. For instance, a
common view in the fnancial architecture literature is that banks and debt markets
operate independently fromeach other[37]. Holmstromand Tirole (1997), Allen and Gale
(2000) and Song and Thakor (2010) develop models in which banks and fnancial
markets complement each other, but none of themconsider the complementarity that we
identify in this paper. Because banks rely increasingly on market funding, including
bond fnancing, commercial paper funding and repo funding, it would be interesting to
investigate the effect of these changes in the funding structure of fnancial
intermediaries on the roles they perform.
305
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Notes
1. Following Boot et al. (1993), who show that banks with low capital are more likely to exploit
borrowers, sacrifcing reputational capital to preserve fnancial capital. One could also
hypothesize that the increase in the cost of bond funding will lead banks to renege on implicit
guarantees they have given their borrowers, including the guarantee not to explore their
informational monopoly, and thereby raise the loan rates on them. For evidence in support of
the hypothesis that banks price their informational monopoly, see Santos and Winton (2008),
Hale and Santos (2009) and Schenone (2010).
2. Publicly listed borrowers that do not have access to the bond market, for example, are not
likely to experience the same level of bank dependency than privately held borrowers that do
not have access to the bond market, but the absence of accounting information on the latter
borrowers precludes us from including them in our investigation.
3. Our paper is also related to the bank lending channel literature, including Kashyap et al.
(1993), Peek and Rosengren (1997), Kashyap and Stein (2000), Paravisini (2008) and Khwaja
and Mian (2008). This literature focuses on the effects of shocks to bank liquidity on the
volume of bank lending. Our focus instead is on the loan pricing effects of shocks to the bond
market. In this regard, our paper is related to Santos (2011) who investigate the effect of bank
losses during the subprime crisis on their loan pricing policies.
4. Our loan data source is Dealscan. This database has some non-syndicated loans, but is only
comprehensive for loans which banks syndicate. Berlin and Mester rely instead on the Survey
of Terms of Bank Lending to Business. This database reports information on every business
loan but only for a stratifed sample of about 340 banks and for the loans banks made on a
particular day (or number of days).
5. Petersen and Rajan (1995) show that provided banks have some monopoly power in the loan
market, they are able to do intertemporal interest rate smoothing to their relationship
borrowers. In their setting, banks are solely funded with deposits. If they used bond fnancing
in addition, it is easy to see that shocks to their cost of bond fnancing would hinder their
ability to smooth interest rates, notwithstanding their monopoly power in the loan market.
See Boot (2000) for a review of the benefts of relationship lending.
6. Recent papers by Acharya et al. (2012) and by Bord and Santos (2013) showthat bank’s access
to liquidity affect the price of the loans and liquidity to corporate borrowers. These results are
complementary to ours.
7. See Allen and Gale (1997, 1999), Bhattacharya and Chiesa (1995), Dewatripont and Maskin
(1995) and Boot and Thakor (1997).
8. We use the logof the loanspread, as opposedto the spreaditself because the logof the loanspread
has distribution which is closer to the normal, and because this allows us to interpret the
coeffcients on the log of other spreads we use on the left-hand side of our model as elasticities. In
anycase, usingspreads insteadof the logof spreads inour models does not affect our keyfndings.
9. Including natural logarithm of this variable, instead of the level, does not change our results.
10. Firms are requiredto report expenses withadvertisingonlywhentheyexceeda certainvalue. For
this reason, this variable is sometimes missing in Compustat. The same is true of expenses with
research and development. In either case, when the variable is missing, we set it equal to zero.
11. For frms with no debt, this variable is set equal to the difference between current assets and
current liabilities.
JFEP
6,3
306
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
12. Using a time horizon of one year to determine if the borrower has a lending relationship with
the bank, yields similar results.
13. For evidence on the endogeneity of loan covenants see Demiroglu and James (2010) and
Murfn (2012).
14. We use the volatility of ROA rather than stock return because a large number of the banks in
the sample do not have publicly traded shares.
15. Our results do not change if instead we control for insured deposits only.
16. Bharath et al. (2011) fnd that the impact of a relationship on spreads is negative; however,
Santos and Winton (2008) fnd that this effect is reversed in recessions, when information
monopolies are likely to be stronger and maintaining relationships is likely to be less
attractive to lenders.
17. We do not count privately placed bonds as a measure of public bond market access. We
believe private placements are very different from public issues, reaching a smaller set of
investors and thus not increasing informed competition as much as a public issue does. As a
practical matter, there is far less information on private placements because the Securities and
Exchange Commission fling rules on public issues do not apply to private issues. This makes
it hard to control for frms’ private placements. This is consistent with earlier work that
considers private placements to be closer to syndicated bank loans than to public bonds.
18. The process we used to link LPC, SDC, and Compustat can be summarized as follows. The
CRSP data was frst used to obtain, through name-matching procedure, CUSIPs for the
companies in LPC for which this information was missing. With a CUSIP, LPC could then be
linked to both SDC and Compustat, which are CUSIP-based data sets. We proceed by using
the PERMCO variable from CRSP to group companies across CUSIPs, as that variable tracks
the same company across CUSIPs and ticker changes.
19. The number of banks adds up to more than the total number because some banks switch from
using bond fnancing to deposit fnancing alone (or vice versa) over the sample period.
20. In the Robustness section, we test whether our results continue to hold when we measure
banks’ cost of bond fnancing by the actual ex ante yields they pay on their new bond issues.
21. Berlin and Mester use core deposits, while we rely on insured deposits, but the two concepts
share the idea that this funding source is less prone to be affected by changes in the overall
cost of credit in the economy.
22. We control for time-invariant unobserved borrower characteristics by including borrower
fxed effects, as discussed below.
23. In the interest of space, we do not report these numbers, but they are available from the
authors upon request.
24. The dates of these “crises” are as follows: August 15, 1990 through March 4, 1992; September 30,
98 through December 9, 1999; April 11, 2000 through November 24, 2003; May 2, 2005 through
May 30, 2005 and November 7, 2007 through end of sample period December 31, 2007).
25. According to this defnition, about 38 percent of loans in our sample were issued following a
bond placed during crisis times.
26. More specifcally, we defne the beginning of the episode when the spread falls below0.62 and
only include episodes during which the spread dipped below 0.6 for at least one period. We
identify “good times” as follows: July 30, 1993 through October 1, 1993; January 26, 1994
307
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
through February 17, 1994; February 8, 1995 through July 30, 1998; January 27, 2006 through
May 30, 2006; and January 18, 2007 through June 20, 2007.
27. According to this defnition, about 40 percent of loans in our sample were issued following a
bond placed during good times.
28. To assure smooth pasting of our proxy into missing observations, we frst regressed interest
expenses on deposits for the observations we had on other proxies and then we constructed
out-of-sample linear predictions based on these regressions.
29. Call Reports have informationonloansales andsecuritization, but it goes backonlyto2001, covers
onlysale andsecuritizationactivities that the bankretains some servicing, credit enhancements or
there is recourse, and is about the stock not the fow of activity in each quarter.
30. We omit bank fxed effects to conduct this test.
31. Historically, during crises in the US bond market the triple-B yield spread has more than
doubled (see footnote [24] for further details).
32. Limiting the defnition of bank dependency to borrowers that never issued in the bond market
yields similar results.
33. Note that the coeffcient on BONDGOODTIMESis negative and signifcant in all models, but
neither RL P BONDIG?BDGOODTIMES nor RL NOACCESS ?BDGOODTIMES have
effects that are statistically different from zero.
34. Note that the coeffcient on BOND CRISIS is positive and signifcant in all models, and while
the effect of RL P BOND IG ?BOND CRISIS is negative and signifcant and the sum of the
two coeffcients is not different from zero, the effect of RL N OACCESS ?BOND CRISIS is
positive and signifcant.
35. In terms of basis points, given that spreads on loans, on average, during times of no crisis and no
good times are 165 basis points (170 basis points on average overall), these would correspond to
16.5 basis points increase following crises and 8 basis points decline following good times.
36. The main effect of bond spreads is negative and not statistically signifcant. The main effect of
relationship borrower indicator is positive and highly statistically signifcant. The effect of the
interaction of these two coeffcients is positive and statistically signifcant, but the total effect of
bond spreads for relationship borrowers is very close to and not statistically different from zero.
37. See Allen and Gale (1997, 1999), Bhattacharya and Chiesa (1995), Dewatripont and Maskin
(1995) and Boot and Thakor (1997).
References
Acharya, V.V., Afonso, G. and Kovner, A. (2012), “How do global banks scramble for liquidity?
Evidence from the asset-backed commercial paper freeze of 2007”, mimeo, available on SSRN.
Allen, F. and Gale, D. (1997), “Financial markets, intermediaries, and intertemporal smoothing”,
Journal of Political Economy, Vol. 105 No. 3, pp. 523-546.
Allen, F. and Gale, D. (1999), “Diversity of opinion and fnancing of new technologies”, Journal of
Financial Intermediation, Vol. 8 Nos 1/2, pp. 68-89.
Allen, F. and Gale, D. (2000), Comparing Financial Systems, The MIT Press, MA.
Berger, A.N. and Udell, G.F. (1990), “Collateral, loan quality, and bank risk”, Journal of Monetary
Economics, Vol. 25 No. 1, pp. 21-42.
Berger, A.N. and Udell, G.F. (1992), “Some evidence on empirical signifcance of credit rationing”,
Journal of Political Economy, Vol. 100 No. 5, pp. 1047-1077.
JFEP
6,3
308
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Berlin, M. and Mester, L. (1999), “Deposits and relationship lending”, Review of Financial Studies,
Vol. 12 No. 3, pp. 579-607.
Bharath, S., Dahiya, S., Saunders, A. and Srinivasan, A. (2011), “Lending relationships and loan
contract terms”, Review of Financial Studies, Vol. 24 No. 4, pp. 1141-1203.
Bhattacharya, S. and Chiesa, G. (1995), “Proprietary information, fnancial inter-mediation, and
research incentives”, Journal of Financial Intermediation, Vol. 4 No. 4, pp. 328-357.
Boot, A.W.A., (2000), “Relationship banking: what do we know?”, Journal of Financial
Intermediation, Vol. 9 No. 1, p. 725.
Boot, A.W.A. and Thakor, A.V. (1997), “Financial system architecture”, Review of Financial
Studies, Vol. 10 No. 3, pp. 693-733.
Boot, A.W.A., Greenbaum, S.I. and Thakor, A.V. (1993), “Reputation and discretion in fnancial
contracting”, American Economic Review, Vol. 83 No. 5, pp. 1165-1183.
Bord, V.M. and Santos, J.A.C. (2013), “Banks liquidity and the cost of liquidity to corporations”,
Journal of Money Credit and Banking, Vol. 46 No. s1, pp. 13-45.
Demiroglu, C. and James, C.M. (2010), “The information content of bank loan covenants”, Review
of Financial Studies, Vol. 23 No. 10, pp. 3700-3737.
Dewatripont, M. and Maskin, E. (1995), “Credit effciency in centralized and decentralized
economies”, Review of Economic Studies, Vol. 62 No. 4, pp. 541-555.
Holmstrom, B. and Tirole, J. (1997), “Financial intermediation, loanable funds, and the real sector”,
Quarterly Journal of Economics, Vol. 112 No. 3, pp. 663-691.
Kashyap, A.K. and Stein, J.C. (2000), “What do a million observations on banks say about the
transmission of monetary policy?”, American Economic Review, Vol. 90 No. 3, pp. 407-428.
Kashyap, A.K., Stein, J.C. and Wilcox, D.W. (1993), “Monetary policy and credit conditions:
evidence from the composition of external fnance”, American Economic Review, Vol. 83
No. 1, pp. 78-98.
Khwaja, A. and Mian, A. (2008), “Tracing the impact of bank liquidity shocks: evidence from an
emerging market”, American Economic Review, Vol. 98 No. 4, pp. 141-342.
Murfn, J. (2012), “The supply-side determinants of loan contract strictness”, The Journal of
Finance, Vol. 67, pp. 1565-1601, doi: 10.1111/j.1540-6261.2012.01767.x
Paravisini, D. (2008), “Local bank fnancial constraints and frm access to external fnance”,
Journal of Finance, Vol. 63 No. 5, pp. 2161-2193.
Peek, J. and Rosengreen, E.S. (1997), “The international transmission of fnancial shocks: the case
of Japan”, American Economic Review, Vol. 87 No. 4, pp. 495-505.
Petersen, M.A. (2009), “Estimating standard errors in fnance panel data sets: comparing
approaches”, Review of Financial Studies, Vol. 22 No. 1, pp. 435-480.
Petersen, M.A. and Rajan, R.G. (1995), “The effect of credit market competition on lending
relationships”, Quarterly Journal of Economics, Vol. 110 No. 2, pp. 407-443.
Rajan, R.G. (1992), “Insiders and outsiders: the choice between informed and arm’s length debt”,
Journal of Finance, Vol. 47 No. 4, pp. 1367-1400.
Santos, J.A.C. (2011), “Bank corporate loan pricing policy following the subprime crisis”, Reviewof
Financial Studies, Vol. 24 No. 6, pp. 1916-1943.
Santos, J.A.C. and Winton, A. (2008), “Bank loans, bonds, and informational monopolies across the
business cycle”, Journal of Finance, Vol. 63 No. 3, pp. 1315-1359.
Schenone, C. (2010), “Lending relationships and information rents: do banks exploit their
information advantages?”, Review of Financial Studies, Vol. 23 No. 3, pp. 1149-1199.
309
Do banks
propagate debt
market shocks?
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
Sharpe, S.A. (1990), “Asymmetric information, bank lending, and implicit contracts: a stylized
model of customer relationships”, Journal of Finance, Vol. 45 No. 4, pp. 1069-1087.
Song, F. and Thakor, A.V. (2010), “Financial system architecture and the co-evolution of banks
and capital markets”, The Economic Journal, 120: 1021–1055. doi: 10.1111/
j.1468-0297.2009.02345.x
Further reading
Hale, G.B. and João A.C. Santos, (2009), “Do banks price their informational monopoly?”, Journal
of Financial Economics, Vol. 93, pp. 185-206.
Corresponding author
Galina Hale can be contacted at: [email protected]
To purchase reprints of this article please e-mail: [email protected]
Or visit our web site for further details: www.emeraldinsight.com/reprints
JFEP
6,3
310
D
o
w
n
l
o
a
d
e
d
b
y
P
O
N
D
I
C
H
E
R
R
Y
U
N
I
V
E
R
S
I
T
Y
A
t
2
1
:
4
9
2
4
J
a
n
u
a
r
y
2
0
1
6
(
P
T
)
doc_694057984.pdf